]>
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> | |
f67539c2 | 26 | #ifndef _WIN32 |
7c673cae | 27 | #include <sys/utsname.h> |
f67539c2 | 28 | #endif |
7c673cae FG |
29 | #include <sys/uio.h> |
30 | ||
31 | #include <boost/lexical_cast.hpp> | |
32 | #include <boost/fusion/include/std_pair.hpp> | |
33 | ||
f67539c2 TL |
34 | #include "common/async/waiter.h" |
35 | ||
36 | #if defined(__FreeBSD__) || defined(_WIN32) | |
7c673cae FG |
37 | #define XATTR_CREATE 0x1 |
38 | #define XATTR_REPLACE 0x2 | |
39 | #else | |
40 | #include <sys/xattr.h> | |
41 | #endif | |
42 | ||
43 | #if defined(__linux__) | |
44 | #include <linux/falloc.h> | |
45 | #endif | |
46 | ||
47 | #include <sys/statvfs.h> | |
48 | ||
49 | #include "common/config.h" | |
50 | #include "common/version.h" | |
f67539c2 | 51 | #include "common/async/blocked_completion.h" |
7c673cae | 52 | |
11fdf7f2 TL |
53 | #include "mon/MonClient.h" |
54 | ||
55 | #include "messages/MClientCaps.h" | |
56 | #include "messages/MClientLease.h" | |
57 | #include "messages/MClientQuota.h" | |
58 | #include "messages/MClientReclaim.h" | |
59 | #include "messages/MClientReclaimReply.h" | |
7c673cae | 60 | #include "messages/MClientReconnect.h" |
11fdf7f2 | 61 | #include "messages/MClientReply.h" |
7c673cae FG |
62 | #include "messages/MClientRequest.h" |
63 | #include "messages/MClientRequestForward.h" | |
11fdf7f2 | 64 | #include "messages/MClientSession.h" |
7c673cae | 65 | #include "messages/MClientSnap.h" |
f67539c2 | 66 | #include "messages/MClientMetrics.h" |
7c673cae | 67 | #include "messages/MCommandReply.h" |
7c673cae FG |
68 | #include "messages/MFSMap.h" |
69 | #include "messages/MFSMapUser.h" | |
11fdf7f2 TL |
70 | #include "messages/MMDSMap.h" |
71 | #include "messages/MOSDMap.h" | |
7c673cae FG |
72 | |
73 | #include "mds/flock.h" | |
11fdf7f2 | 74 | #include "mds/cephfs_features.h" |
7c673cae FG |
75 | #include "osd/OSDMap.h" |
76 | #include "osdc/Filer.h" | |
77 | ||
78 | #include "common/Cond.h" | |
7c673cae FG |
79 | #include "common/perf_counters.h" |
80 | #include "common/admin_socket.h" | |
81 | #include "common/errno.h" | |
82 | #include "include/str_list.h" | |
83 | ||
84 | #define dout_subsys ceph_subsys_client | |
85 | ||
86 | #include "include/lru.h" | |
87 | #include "include/compat.h" | |
88 | #include "include/stringify.h" | |
f67539c2 | 89 | #include "include/random.h" |
7c673cae FG |
90 | |
91 | #include "Client.h" | |
92 | #include "Inode.h" | |
93 | #include "Dentry.h" | |
b32b8144 | 94 | #include "Delegation.h" |
7c673cae FG |
95 | #include "Dir.h" |
96 | #include "ClientSnapRealm.h" | |
97 | #include "Fh.h" | |
98 | #include "MetaSession.h" | |
99 | #include "MetaRequest.h" | |
100 | #include "ObjecterWriteback.h" | |
101 | #include "posix_acl.h" | |
102 | ||
11fdf7f2 | 103 | #include "include/ceph_assert.h" |
7c673cae FG |
104 | #include "include/stat.h" |
105 | ||
e306af50 | 106 | #include "include/cephfs/ceph_ll_client.h" |
7c673cae FG |
107 | |
108 | #if HAVE_GETGROUPLIST | |
109 | #include <grp.h> | |
110 | #include <pwd.h> | |
111 | #include <unistd.h> | |
112 | #endif | |
113 | ||
114 | #undef dout_prefix | |
115 | #define dout_prefix *_dout << "client." << whoami << " " | |
116 | ||
117 | #define tout(cct) if (!cct->_conf->client_trace.empty()) traceout | |
118 | ||
119 | // FreeBSD fails to define this | |
120 | #ifndef O_DSYNC | |
121 | #define O_DSYNC 0x0 | |
122 | #endif | |
123 | // Darwin fails to define this | |
124 | #ifndef O_RSYNC | |
125 | #define O_RSYNC 0x0 | |
126 | #endif | |
127 | ||
128 | #ifndef O_DIRECT | |
129 | #define O_DIRECT 0x0 | |
130 | #endif | |
131 | ||
f67539c2 TL |
132 | // Windows doesn't define those values. While the Posix compatibilty layer |
133 | // doesn't support those values, the Windows native functions do provide | |
134 | // similar flags. Special care should be taken if we're going to use those | |
135 | // flags in ceph-dokan. The current values are no-ops, while propagating | |
136 | // them to the rest of the code might cause the Windows functions to reject | |
137 | // them as invalid. | |
138 | #ifndef O_NOFOLLOW | |
139 | #define O_NOFOLLOW 0x0 | |
140 | #endif | |
141 | ||
142 | #ifndef O_SYNC | |
143 | #define O_SYNC 0x0 | |
144 | #endif | |
145 | ||
7c673cae FG |
146 | #define DEBUG_GETATTR_CAPS (CEPH_CAP_XATTR_SHARED) |
147 | ||
b3b6e05e TL |
148 | #ifndef S_IXUGO |
149 | #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH) | |
150 | #endif | |
151 | ||
20effc67 TL |
152 | using std::dec; |
153 | using std::hex; | |
154 | using std::list; | |
155 | using std::oct; | |
156 | using std::pair; | |
157 | using std::string; | |
158 | using std::vector; | |
159 | ||
adb31ebb TL |
160 | using namespace TOPNSPC::common; |
161 | ||
f67539c2 TL |
162 | namespace bs = boost::system; |
163 | namespace ca = ceph::async; | |
164 | ||
7c673cae FG |
165 | void client_flush_set_callback(void *p, ObjectCacher::ObjectSet *oset) |
166 | { | |
167 | Client *client = static_cast<Client*>(p); | |
168 | client->flush_set_callback(oset); | |
169 | } | |
170 | ||
b3b6e05e TL |
171 | bool Client::is_reserved_vino(vinodeno_t &vino) { |
172 | if (MDS_IS_PRIVATE_INO(vino.ino)) { | |
173 | ldout(cct, -1) << __func__ << " attempt to access reserved inode number " << vino << dendl; | |
174 | return true; | |
175 | } | |
176 | return false; | |
177 | } | |
178 | ||
7c673cae FG |
179 | |
180 | // ------------- | |
181 | ||
182 | Client::CommandHook::CommandHook(Client *client) : | |
183 | m_client(client) | |
184 | { | |
185 | } | |
186 | ||
9f95a23c TL |
187 | int Client::CommandHook::call( |
188 | std::string_view command, | |
189 | const cmdmap_t& cmdmap, | |
190 | Formatter *f, | |
191 | std::ostream& errss, | |
192 | bufferlist& out) | |
7c673cae | 193 | { |
7c673cae | 194 | f->open_object_section("result"); |
9f95a23c | 195 | { |
f67539c2 | 196 | std::scoped_lock l{m_client->client_lock}; |
9f95a23c TL |
197 | if (command == "mds_requests") |
198 | m_client->dump_mds_requests(f); | |
adb31ebb TL |
199 | else if (command == "mds_sessions") { |
200 | bool cap_dump = false; | |
201 | cmd_getval(cmdmap, "cap_dump", cap_dump); | |
202 | m_client->dump_mds_sessions(f, cap_dump); | |
203 | } else if (command == "dump_cache") | |
9f95a23c TL |
204 | m_client->dump_cache(f); |
205 | else if (command == "kick_stale_sessions") | |
206 | m_client->_kick_stale_sessions(); | |
207 | else if (command == "status") | |
208 | m_client->dump_status(f); | |
209 | else | |
210 | ceph_abort_msg("bad command registered"); | |
211 | } | |
7c673cae | 212 | f->close_section(); |
9f95a23c | 213 | return 0; |
7c673cae FG |
214 | } |
215 | ||
216 | ||
217 | // ------------- | |
218 | ||
b3b6e05e TL |
219 | int Client::get_fd_inode(int fd, InodeRef *in) { |
220 | int r = 0; | |
221 | if (fd == CEPHFS_AT_FDCWD) { | |
222 | *in = cwd; | |
223 | } else { | |
224 | Fh *f = get_filehandle(fd); | |
225 | if (!f) { | |
226 | r = -CEPHFS_EBADF; | |
227 | } else { | |
228 | *in = f->inode; | |
229 | } | |
230 | } | |
231 | return r; | |
232 | } | |
233 | ||
7c673cae FG |
234 | dir_result_t::dir_result_t(Inode *in, const UserPerm& perms) |
235 | : inode(in), offset(0), next_offset(2), | |
236 | release_count(0), ordered_count(0), cache_index(0), start_shared_gen(0), | |
237 | perms(perms) | |
238 | { } | |
239 | ||
240 | void Client::_reset_faked_inos() | |
241 | { | |
242 | ino_t start = 1024; | |
243 | free_faked_inos.clear(); | |
244 | free_faked_inos.insert(start, (uint32_t)-1 - start + 1); | |
245 | last_used_faked_ino = 0; | |
11fdf7f2 | 246 | last_used_faked_root = 0; |
f67539c2 TL |
247 | #ifdef _WIN32 |
248 | // On Windows, sizeof(ino_t) is just 2. Despite that, most "native" | |
249 | // Windows structures, including Dokan ones, are using 64B identifiers. | |
250 | _use_faked_inos = false; | |
251 | #else | |
7c673cae | 252 | _use_faked_inos = sizeof(ino_t) < 8 || cct->_conf->client_use_faked_inos; |
f67539c2 | 253 | #endif |
7c673cae FG |
254 | } |
255 | ||
256 | void Client::_assign_faked_ino(Inode *in) | |
257 | { | |
11fdf7f2 TL |
258 | if (0 == last_used_faked_ino) |
259 | last_used_faked_ino = last_used_faked_ino + 2048; // start(1024)~2048 reserved for _assign_faked_root | |
7c673cae FG |
260 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
261 | if (it == free_faked_inos.end() && last_used_faked_ino > 0) { | |
11fdf7f2 | 262 | last_used_faked_ino = 2048; |
7c673cae FG |
263 | it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
264 | } | |
11fdf7f2 | 265 | ceph_assert(it != free_faked_inos.end()); |
7c673cae | 266 | if (last_used_faked_ino < it.get_start()) { |
11fdf7f2 | 267 | ceph_assert(it.get_len() > 0); |
7c673cae FG |
268 | last_used_faked_ino = it.get_start(); |
269 | } else { | |
270 | ++last_used_faked_ino; | |
11fdf7f2 | 271 | ceph_assert(it.get_start() + it.get_len() > last_used_faked_ino); |
7c673cae FG |
272 | } |
273 | in->faked_ino = last_used_faked_ino; | |
274 | free_faked_inos.erase(in->faked_ino); | |
275 | faked_ino_map[in->faked_ino] = in->vino(); | |
276 | } | |
277 | ||
11fdf7f2 TL |
278 | /* |
279 | * In the faked mode, if you export multiple subdirectories, | |
280 | * you will see that the inode numbers of the exported subdirectories | |
281 | * are the same. so we distinguish the mount point by reserving | |
282 | * the "fake ids" between "1024~2048" and combining the last | |
283 | * 10bits(0x3ff) of the "root inodes". | |
284 | */ | |
285 | void Client::_assign_faked_root(Inode *in) | |
286 | { | |
287 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
288 | if (it == free_faked_inos.end() && last_used_faked_root > 0) { | |
289 | last_used_faked_root = 0; | |
290 | it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
291 | } | |
20effc67 | 292 | ceph_assert(it != free_faked_inos.end()); |
11fdf7f2 TL |
293 | vinodeno_t inode_info = in->vino(); |
294 | uint64_t inode_num = (uint64_t)inode_info.ino; | |
295 | ldout(cct, 10) << "inode_num " << inode_num << "inode_num & 0x3ff=" << (inode_num & 0x3ff)<< dendl; | |
296 | last_used_faked_root = it.get_start() + (inode_num & 0x3ff); // 0x3ff mask and get_start will not exceed 2048 | |
20effc67 | 297 | ceph_assert(it.get_start() + it.get_len() > last_used_faked_root); |
11fdf7f2 TL |
298 | |
299 | in->faked_ino = last_used_faked_root; | |
300 | free_faked_inos.erase(in->faked_ino); | |
301 | faked_ino_map[in->faked_ino] = in->vino(); | |
302 | } | |
303 | ||
7c673cae FG |
304 | void Client::_release_faked_ino(Inode *in) |
305 | { | |
306 | free_faked_inos.insert(in->faked_ino); | |
307 | faked_ino_map.erase(in->faked_ino); | |
308 | } | |
309 | ||
310 | vinodeno_t Client::_map_faked_ino(ino_t ino) | |
311 | { | |
312 | vinodeno_t vino; | |
313 | if (ino == 1) | |
314 | vino = root->vino(); | |
315 | else if (faked_ino_map.count(ino)) | |
316 | vino = faked_ino_map[ino]; | |
317 | else | |
318 | vino = vinodeno_t(0, CEPH_NOSNAP); | |
11fdf7f2 | 319 | ldout(cct, 10) << __func__ << " " << ino << " -> " << vino << dendl; |
7c673cae FG |
320 | return vino; |
321 | } | |
322 | ||
323 | vinodeno_t Client::map_faked_ino(ino_t ino) | |
324 | { | |
f67539c2 | 325 | std::scoped_lock lock(client_lock); |
7c673cae FG |
326 | return _map_faked_ino(ino); |
327 | } | |
328 | ||
329 | // cons/des | |
330 | ||
331 | Client::Client(Messenger *m, MonClient *mc, Objecter *objecter_) | |
f67539c2 TL |
332 | : Dispatcher(m->cct->get()), |
333 | timer(m->cct, timer_lock, false), | |
11fdf7f2 TL |
334 | messenger(m), |
335 | monclient(mc), | |
336 | objecter(objecter_), | |
337 | whoami(mc->get_global_id()), | |
f67539c2 TL |
338 | mount_state(CLIENT_UNMOUNTED, "Client::mountstate_lock"), |
339 | initialize_state(CLIENT_NEW, "Client::initstate_lock"), | |
340 | cct_deleter{m->cct, [](CephContext *p) {p->put();}}, | |
7c673cae FG |
341 | async_ino_invalidator(m->cct), |
342 | async_dentry_invalidator(m->cct), | |
343 | interrupt_finisher(m->cct), | |
344 | remount_finisher(m->cct), | |
e306af50 | 345 | async_ino_releasor(m->cct), |
7c673cae | 346 | objecter_finisher(m->cct), |
11fdf7f2 TL |
347 | m_command_hook(this), |
348 | fscid(0) | |
7c673cae FG |
349 | { |
350 | _reset_faked_inos(); | |
7c673cae | 351 | |
7c673cae FG |
352 | user_id = cct->_conf->client_mount_uid; |
353 | group_id = cct->_conf->client_mount_gid; | |
92f5a8d4 TL |
354 | fuse_default_permissions = cct->_conf.get_val<bool>( |
355 | "fuse_default_permissions"); | |
7c673cae | 356 | |
33c7a0ef TL |
357 | _collect_and_send_global_metrics = cct->_conf.get_val<bool>( |
358 | "client_collect_and_send_global_metrics"); | |
359 | ||
7c673cae FG |
360 | if (cct->_conf->client_acl_type == "posix_acl") |
361 | acl_type = POSIX_ACL; | |
362 | ||
7c673cae FG |
363 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
364 | ||
365 | // file handles | |
366 | free_fd_set.insert(10, 1<<30); | |
367 | ||
368 | mdsmap.reset(new MDSMap); | |
369 | ||
370 | // osd interfaces | |
371 | writeback_handler.reset(new ObjecterWriteback(objecter, &objecter_finisher, | |
372 | &client_lock)); | |
373 | objectcacher.reset(new ObjectCacher(cct, "libcephfs", *writeback_handler, client_lock, | |
374 | client_flush_set_callback, // all commit callback | |
375 | (void*)this, | |
376 | cct->_conf->client_oc_size, | |
377 | cct->_conf->client_oc_max_objects, | |
378 | cct->_conf->client_oc_max_dirty, | |
379 | cct->_conf->client_oc_target_dirty, | |
380 | cct->_conf->client_oc_max_dirty_age, | |
381 | true)); | |
7c673cae FG |
382 | } |
383 | ||
384 | ||
385 | Client::~Client() | |
386 | { | |
9f95a23c | 387 | ceph_assert(ceph_mutex_is_not_locked(client_lock)); |
7c673cae | 388 | |
f67539c2 TL |
389 | // If the task is crashed or aborted and doesn't |
390 | // get any chance to run the umount and shutdow. | |
391 | { | |
392 | std::scoped_lock l{client_lock}; | |
393 | tick_thread_stopped = true; | |
394 | upkeep_cond.notify_one(); | |
395 | } | |
396 | ||
397 | if (upkeeper.joinable()) | |
398 | upkeeper.join(); | |
399 | ||
31f18b77 FG |
400 | // It is necessary to hold client_lock, because any inode destruction |
401 | // may call into ObjectCacher, which asserts that it's lock (which is | |
402 | // client_lock) is held. | |
f67539c2 | 403 | std::scoped_lock l{client_lock}; |
7c673cae FG |
404 | tear_down_cache(); |
405 | } | |
406 | ||
407 | void Client::tear_down_cache() | |
408 | { | |
409 | // fd's | |
f67539c2 TL |
410 | for (auto &[fd, fh] : fd_map) { |
411 | ldout(cct, 1) << __func__ << " forcing close of fh " << fd << " ino " << fh->inode->ino << dendl; | |
7c673cae FG |
412 | _release_fh(fh); |
413 | } | |
414 | fd_map.clear(); | |
415 | ||
416 | while (!opened_dirs.empty()) { | |
417 | dir_result_t *dirp = *opened_dirs.begin(); | |
11fdf7f2 | 418 | ldout(cct, 1) << __func__ << " forcing close of dir " << dirp << " ino " << dirp->inode->ino << dendl; |
7c673cae FG |
419 | _closedir(dirp); |
420 | } | |
421 | ||
422 | // caps! | |
423 | // *** FIXME *** | |
424 | ||
425 | // empty lru | |
7c673cae | 426 | trim_cache(); |
11fdf7f2 | 427 | ceph_assert(lru.lru_get_size() == 0); |
7c673cae FG |
428 | |
429 | // close root ino | |
11fdf7f2 | 430 | ceph_assert(inode_map.size() <= 1 + root_parents.size()); |
7c673cae | 431 | if (root && inode_map.size() == 1 + root_parents.size()) { |
b3b6e05e | 432 | root.reset(); |
7c673cae FG |
433 | } |
434 | ||
11fdf7f2 | 435 | ceph_assert(inode_map.empty()); |
7c673cae FG |
436 | } |
437 | ||
438 | inodeno_t Client::get_root_ino() | |
439 | { | |
f67539c2 | 440 | std::scoped_lock l(client_lock); |
7c673cae FG |
441 | if (use_faked_inos()) |
442 | return root->faked_ino; | |
443 | else | |
444 | return root->ino; | |
445 | } | |
446 | ||
447 | Inode *Client::get_root() | |
448 | { | |
f67539c2 | 449 | std::scoped_lock l(client_lock); |
7c673cae | 450 | root->ll_get(); |
b3b6e05e | 451 | return root.get(); |
7c673cae FG |
452 | } |
453 | ||
454 | ||
455 | // debug crapola | |
456 | ||
457 | void Client::dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected) | |
458 | { | |
459 | filepath path; | |
460 | in->make_long_path(path); | |
461 | ldout(cct, 1) << "dump_inode: " | |
462 | << (disconnected ? "DISCONNECTED ":"") | |
463 | << "inode " << in->ino | |
464 | << " " << path | |
b3b6e05e | 465 | << " ref " << in->get_nref() |
f67539c2 | 466 | << " " << *in << dendl; |
7c673cae FG |
467 | |
468 | if (f) { | |
469 | f->open_object_section("inode"); | |
470 | f->dump_stream("path") << path; | |
471 | if (disconnected) | |
472 | f->dump_int("disconnected", 1); | |
473 | in->dump(f); | |
474 | f->close_section(); | |
475 | } | |
476 | ||
477 | did.insert(in); | |
478 | if (in->dir) { | |
479 | ldout(cct, 1) << " dir " << in->dir << " size " << in->dir->dentries.size() << dendl; | |
480 | for (ceph::unordered_map<string, Dentry*>::iterator it = in->dir->dentries.begin(); | |
481 | it != in->dir->dentries.end(); | |
482 | ++it) { | |
483 | ldout(cct, 1) << " " << in->ino << " dn " << it->first << " " << it->second << " ref " << it->second->ref << dendl; | |
484 | if (f) { | |
485 | f->open_object_section("dentry"); | |
486 | it->second->dump(f); | |
487 | f->close_section(); | |
488 | } | |
489 | if (it->second->inode) | |
490 | dump_inode(f, it->second->inode.get(), did, false); | |
491 | } | |
492 | } | |
493 | } | |
494 | ||
495 | void Client::dump_cache(Formatter *f) | |
496 | { | |
497 | set<Inode*> did; | |
498 | ||
11fdf7f2 | 499 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
500 | |
501 | if (f) | |
502 | f->open_array_section("cache"); | |
503 | ||
504 | if (root) | |
b3b6e05e | 505 | dump_inode(f, root.get(), did, true); |
7c673cae FG |
506 | |
507 | // make a second pass to catch anything disconnected | |
508 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
509 | it != inode_map.end(); | |
510 | ++it) { | |
511 | if (did.count(it->second)) | |
512 | continue; | |
513 | dump_inode(f, it->second, did, true); | |
514 | } | |
515 | ||
516 | if (f) | |
517 | f->close_section(); | |
518 | } | |
519 | ||
520 | void Client::dump_status(Formatter *f) | |
521 | { | |
9f95a23c | 522 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
523 | |
524 | ldout(cct, 1) << __func__ << dendl; | |
525 | ||
526 | const epoch_t osd_epoch | |
527 | = objecter->with_osdmap(std::mem_fn(&OSDMap::get_epoch)); | |
528 | ||
529 | if (f) { | |
530 | f->open_object_section("metadata"); | |
531 | for (const auto& kv : metadata) | |
532 | f->dump_string(kv.first.c_str(), kv.second); | |
533 | f->close_section(); | |
534 | ||
535 | f->dump_int("dentry_count", lru.lru_get_size()); | |
536 | f->dump_int("dentry_pinned_count", lru.lru_get_num_pinned()); | |
537 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 538 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 | 539 | f->dump_object("inst", inst); |
11fdf7f2 TL |
540 | f->dump_object("addr", inst.addr); |
541 | f->dump_stream("inst_str") << inst.name << " " << inst.addr.get_legacy_str(); | |
542 | f->dump_string("addr_str", inst.addr.get_legacy_str()); | |
7c673cae FG |
543 | f->dump_int("inode_count", inode_map.size()); |
544 | f->dump_int("mds_epoch", mdsmap->get_epoch()); | |
545 | f->dump_int("osd_epoch", osd_epoch); | |
546 | f->dump_int("osd_epoch_barrier", cap_epoch_barrier); | |
f67539c2 | 547 | f->dump_bool("blocklisted", blocklisted); |
adb31ebb | 548 | f->dump_string("fs_name", mdsmap->get_fs_name()); |
7c673cae FG |
549 | } |
550 | } | |
551 | ||
e306af50 | 552 | void Client::_pre_init() |
7c673cae FG |
553 | { |
554 | timer.init(); | |
e306af50 TL |
555 | |
556 | objecter_finisher.start(); | |
557 | filer.reset(new Filer(objecter, &objecter_finisher)); | |
e306af50 | 558 | |
7c673cae | 559 | objectcacher->start(); |
e306af50 TL |
560 | } |
561 | ||
562 | int Client::init() | |
563 | { | |
f67539c2 TL |
564 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
565 | ceph_assert(iref_writer.is_first_writer()); | |
566 | ||
e306af50 | 567 | _pre_init(); |
9f95a23c | 568 | { |
f67539c2 | 569 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
570 | messenger->add_dispatcher_tail(this); |
571 | } | |
7c673cae | 572 | _finish_init(); |
f67539c2 | 573 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
574 | return 0; |
575 | } | |
576 | ||
577 | void Client::_finish_init() | |
578 | { | |
9f95a23c | 579 | { |
f67539c2 | 580 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
581 | // logger |
582 | PerfCountersBuilder plb(cct, "client", l_c_first, l_c_last); | |
583 | plb.add_time_avg(l_c_reply, "reply", "Latency of receiving a reply on metadata request"); | |
584 | plb.add_time_avg(l_c_lat, "lat", "Latency of processing a metadata request"); | |
585 | plb.add_time_avg(l_c_wrlat, "wrlat", "Latency of a file data write operation"); | |
586 | plb.add_time_avg(l_c_read, "rdlat", "Latency of a file data read operation"); | |
587 | plb.add_time_avg(l_c_fsync, "fsync", "Latency of a file sync operation"); | |
588 | logger.reset(plb.create_perf_counters()); | |
589 | cct->get_perfcounters_collection()->add(logger.get()); | |
590 | } | |
7c673cae | 591 | |
11fdf7f2 | 592 | cct->_conf.add_observer(this); |
7c673cae FG |
593 | |
594 | AdminSocket* admin_socket = cct->get_admin_socket(); | |
595 | int ret = admin_socket->register_command("mds_requests", | |
7c673cae FG |
596 | &m_command_hook, |
597 | "show in-progress mds requests"); | |
598 | if (ret < 0) { | |
599 | lderr(cct) << "error registering admin socket command: " | |
600 | << cpp_strerror(-ret) << dendl; | |
601 | } | |
adb31ebb TL |
602 | ret = admin_socket->register_command("mds_sessions " |
603 | "name=cap_dump,type=CephBool,req=false", | |
7c673cae FG |
604 | &m_command_hook, |
605 | "show mds session state"); | |
606 | if (ret < 0) { | |
607 | lderr(cct) << "error registering admin socket command: " | |
608 | << cpp_strerror(-ret) << dendl; | |
609 | } | |
610 | ret = admin_socket->register_command("dump_cache", | |
7c673cae FG |
611 | &m_command_hook, |
612 | "show in-memory metadata cache contents"); | |
613 | if (ret < 0) { | |
614 | lderr(cct) << "error registering admin socket command: " | |
615 | << cpp_strerror(-ret) << dendl; | |
616 | } | |
617 | ret = admin_socket->register_command("kick_stale_sessions", | |
7c673cae FG |
618 | &m_command_hook, |
619 | "kick sessions that were remote reset"); | |
620 | if (ret < 0) { | |
621 | lderr(cct) << "error registering admin socket command: " | |
622 | << cpp_strerror(-ret) << dendl; | |
623 | } | |
624 | ret = admin_socket->register_command("status", | |
7c673cae FG |
625 | &m_command_hook, |
626 | "show overall client status"); | |
627 | if (ret < 0) { | |
628 | lderr(cct) << "error registering admin socket command: " | |
629 | << cpp_strerror(-ret) << dendl; | |
630 | } | |
7c673cae FG |
631 | } |
632 | ||
633 | void Client::shutdown() | |
634 | { | |
11fdf7f2 | 635 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
636 | |
637 | // If we were not mounted, but were being used for sending | |
638 | // MDS commands, we may have sessions that need closing. | |
9f95a23c | 639 | { |
f67539c2 TL |
640 | std::scoped_lock l{client_lock}; |
641 | ||
642 | // To make sure the tick thread will be stoppped before | |
643 | // destructing the Client, just in case like the _mount() | |
644 | // failed but didn't not get a chance to stop the tick | |
645 | // thread | |
646 | tick_thread_stopped = true; | |
647 | upkeep_cond.notify_one(); | |
648 | ||
9f95a23c TL |
649 | _close_sessions(); |
650 | } | |
11fdf7f2 | 651 | cct->_conf.remove_observer(this); |
7c673cae | 652 | |
11fdf7f2 | 653 | cct->get_admin_socket()->unregister_commands(&m_command_hook); |
7c673cae FG |
654 | |
655 | if (ino_invalidate_cb) { | |
656 | ldout(cct, 10) << "shutdown stopping cache invalidator finisher" << dendl; | |
657 | async_ino_invalidator.wait_for_empty(); | |
658 | async_ino_invalidator.stop(); | |
659 | } | |
660 | ||
661 | if (dentry_invalidate_cb) { | |
662 | ldout(cct, 10) << "shutdown stopping dentry invalidator finisher" << dendl; | |
663 | async_dentry_invalidator.wait_for_empty(); | |
664 | async_dentry_invalidator.stop(); | |
665 | } | |
666 | ||
667 | if (switch_interrupt_cb) { | |
668 | ldout(cct, 10) << "shutdown stopping interrupt finisher" << dendl; | |
669 | interrupt_finisher.wait_for_empty(); | |
670 | interrupt_finisher.stop(); | |
671 | } | |
672 | ||
673 | if (remount_cb) { | |
674 | ldout(cct, 10) << "shutdown stopping remount finisher" << dendl; | |
675 | remount_finisher.wait_for_empty(); | |
676 | remount_finisher.stop(); | |
677 | } | |
678 | ||
e306af50 TL |
679 | if (ino_release_cb) { |
680 | ldout(cct, 10) << "shutdown stopping inode release finisher" << dendl; | |
681 | async_ino_releasor.wait_for_empty(); | |
682 | async_ino_releasor.stop(); | |
683 | } | |
684 | ||
7c673cae | 685 | objectcacher->stop(); // outside of client_lock! this does a join. |
f67539c2 TL |
686 | |
687 | /* | |
688 | * We are shuting down the client. | |
689 | * | |
690 | * Just declare the state to CLIENT_NEW to block and fail any | |
691 | * new comming "reader" and then try to wait all the in-flight | |
692 | * "readers" to finish. | |
693 | */ | |
694 | RWRef_t iref_writer(initialize_state, CLIENT_NEW, false); | |
695 | if (!iref_writer.is_first_writer()) | |
696 | return; | |
697 | iref_writer.wait_readers_done(); | |
698 | ||
9f95a23c | 699 | { |
f67539c2 | 700 | std::scoped_lock l(timer_lock); |
9f95a23c TL |
701 | timer.shutdown(); |
702 | } | |
f67539c2 | 703 | |
7c673cae FG |
704 | objecter_finisher.wait_for_empty(); |
705 | objecter_finisher.stop(); | |
706 | ||
707 | if (logger) { | |
708 | cct->get_perfcounters_collection()->remove(logger.get()); | |
709 | logger.reset(); | |
710 | } | |
711 | } | |
712 | ||
713 | ||
714 | // =================== | |
715 | // metadata cache stuff | |
716 | ||
717 | void Client::trim_cache(bool trim_kernel_dcache) | |
718 | { | |
181888fb FG |
719 | uint64_t max = cct->_conf->client_cache_size; |
720 | ldout(cct, 20) << "trim_cache size " << lru.lru_get_size() << " max " << max << dendl; | |
7c673cae FG |
721 | unsigned last = 0; |
722 | while (lru.lru_get_size() != last) { | |
723 | last = lru.lru_get_size(); | |
724 | ||
f67539c2 | 725 | if (!is_unmounting() && lru.lru_get_size() <= max) break; |
7c673cae FG |
726 | |
727 | // trim! | |
31f18b77 | 728 | Dentry *dn = static_cast<Dentry*>(lru.lru_get_next_expire()); |
7c673cae FG |
729 | if (!dn) |
730 | break; // done | |
f67539c2 | 731 | |
7c673cae FG |
732 | trim_dentry(dn); |
733 | } | |
734 | ||
181888fb | 735 | if (trim_kernel_dcache && lru.lru_get_size() > max) |
7c673cae FG |
736 | _invalidate_kernel_dcache(); |
737 | ||
738 | // hose root? | |
b3b6e05e | 739 | if (lru.lru_get_size() == 0 && root && root->get_nref() == 1 && inode_map.size() == 1 + root_parents.size()) { |
7c673cae | 740 | ldout(cct, 15) << "trim_cache trimmed root " << root << dendl; |
b3b6e05e | 741 | root.reset(); |
7c673cae FG |
742 | } |
743 | } | |
744 | ||
745 | void Client::trim_cache_for_reconnect(MetaSession *s) | |
746 | { | |
747 | mds_rank_t mds = s->mds_num; | |
11fdf7f2 | 748 | ldout(cct, 20) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
749 | |
750 | int trimmed = 0; | |
751 | list<Dentry*> skipped; | |
752 | while (lru.lru_get_size() > 0) { | |
753 | Dentry *dn = static_cast<Dentry*>(lru.lru_expire()); | |
754 | if (!dn) | |
755 | break; | |
756 | ||
757 | if ((dn->inode && dn->inode->caps.count(mds)) || | |
758 | dn->dir->parent_inode->caps.count(mds)) { | |
759 | trim_dentry(dn); | |
760 | trimmed++; | |
761 | } else | |
762 | skipped.push_back(dn); | |
763 | } | |
764 | ||
765 | for(list<Dentry*>::iterator p = skipped.begin(); p != skipped.end(); ++p) | |
766 | lru.lru_insert_mid(*p); | |
767 | ||
11fdf7f2 | 768 | ldout(cct, 20) << __func__ << " mds." << mds |
7c673cae FG |
769 | << " trimmed " << trimmed << " dentries" << dendl; |
770 | ||
771 | if (s->caps.size() > 0) | |
772 | _invalidate_kernel_dcache(); | |
773 | } | |
774 | ||
775 | void Client::trim_dentry(Dentry *dn) | |
776 | { | |
777 | ldout(cct, 15) << "trim_dentry unlinking dn " << dn->name | |
11fdf7f2 TL |
778 | << " in dir " |
779 | << std::hex << dn->dir->parent_inode->ino << std::dec | |
7c673cae FG |
780 | << dendl; |
781 | if (dn->inode) { | |
782 | Inode *diri = dn->dir->parent_inode; | |
7c673cae FG |
783 | clear_dir_complete_and_ordered(diri, true); |
784 | } | |
785 | unlink(dn, false, false); // drop dir, drop dentry | |
786 | } | |
787 | ||
788 | ||
1adf2230 AA |
789 | void Client::update_inode_file_size(Inode *in, int issued, uint64_t size, |
790 | uint64_t truncate_seq, uint64_t truncate_size) | |
7c673cae | 791 | { |
7c673cae FG |
792 | uint64_t prior_size = in->size; |
793 | ||
7c673cae FG |
794 | if (truncate_seq > in->truncate_seq || |
795 | (truncate_seq == in->truncate_seq && size > in->size)) { | |
796 | ldout(cct, 10) << "size " << in->size << " -> " << size << dendl; | |
797 | in->size = size; | |
798 | in->reported_size = size; | |
799 | if (truncate_seq != in->truncate_seq) { | |
800 | ldout(cct, 10) << "truncate_seq " << in->truncate_seq << " -> " | |
801 | << truncate_seq << dendl; | |
802 | in->truncate_seq = truncate_seq; | |
803 | in->oset.truncate_seq = truncate_seq; | |
804 | ||
805 | // truncate cached file data | |
806 | if (prior_size > size) { | |
807 | _invalidate_inode_cache(in, truncate_size, prior_size - truncate_size); | |
808 | } | |
809 | } | |
810 | ||
811 | // truncate inline data | |
812 | if (in->inline_version < CEPH_INLINE_NONE) { | |
813 | uint32_t len = in->inline_data.length(); | |
814 | if (size < len) | |
815 | in->inline_data.splice(size, len - size); | |
816 | } | |
817 | } | |
818 | if (truncate_seq >= in->truncate_seq && | |
819 | in->truncate_size != truncate_size) { | |
820 | if (in->is_file()) { | |
821 | ldout(cct, 10) << "truncate_size " << in->truncate_size << " -> " | |
822 | << truncate_size << dendl; | |
823 | in->truncate_size = truncate_size; | |
824 | in->oset.truncate_size = truncate_size; | |
825 | } else { | |
826 | ldout(cct, 0) << "Hmmm, truncate_seq && truncate_size changed on non-file inode!" << dendl; | |
827 | } | |
828 | } | |
1adf2230 AA |
829 | } |
830 | ||
831 | void Client::update_inode_file_time(Inode *in, int issued, uint64_t time_warp_seq, | |
832 | utime_t ctime, utime_t mtime, utime_t atime) | |
833 | { | |
834 | ldout(cct, 10) << __func__ << " " << *in << " " << ccap_string(issued) | |
835 | << " ctime " << ctime << " mtime " << mtime << dendl; | |
836 | ||
837 | if (time_warp_seq > in->time_warp_seq) | |
838 | ldout(cct, 10) << " mds time_warp_seq " << time_warp_seq | |
839 | << " is higher than local time_warp_seq " | |
840 | << in->time_warp_seq << dendl; | |
841 | ||
842 | int warn = false; | |
7c673cae FG |
843 | // be careful with size, mtime, atime |
844 | if (issued & (CEPH_CAP_FILE_EXCL| | |
845 | CEPH_CAP_FILE_WR| | |
846 | CEPH_CAP_FILE_BUFFER| | |
847 | CEPH_CAP_AUTH_EXCL| | |
848 | CEPH_CAP_XATTR_EXCL)) { | |
849 | ldout(cct, 30) << "Yay have enough caps to look at our times" << dendl; | |
850 | if (ctime > in->ctime) | |
851 | in->ctime = ctime; | |
852 | if (time_warp_seq > in->time_warp_seq) { | |
7c673cae FG |
853 | //the mds updated times, so take those! |
854 | in->mtime = mtime; | |
855 | in->atime = atime; | |
856 | in->time_warp_seq = time_warp_seq; | |
857 | } else if (time_warp_seq == in->time_warp_seq) { | |
858 | //take max times | |
859 | if (mtime > in->mtime) | |
860 | in->mtime = mtime; | |
861 | if (atime > in->atime) | |
862 | in->atime = atime; | |
863 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
864 | //ignore mds values as we have a higher seq | |
865 | } else warn = true; | |
866 | } else { | |
867 | ldout(cct, 30) << "Don't have enough caps, just taking mds' time values" << dendl; | |
868 | if (time_warp_seq >= in->time_warp_seq) { | |
869 | in->ctime = ctime; | |
870 | in->mtime = mtime; | |
871 | in->atime = atime; | |
872 | in->time_warp_seq = time_warp_seq; | |
873 | } else warn = true; | |
874 | } | |
875 | if (warn) { | |
876 | ldout(cct, 0) << "WARNING: " << *in << " mds time_warp_seq " | |
877 | << time_warp_seq << " is lower than local time_warp_seq " | |
878 | << in->time_warp_seq | |
879 | << dendl; | |
880 | } | |
881 | } | |
882 | ||
883 | void Client::_fragmap_remove_non_leaves(Inode *in) | |
884 | { | |
885 | for (map<frag_t,int>::iterator p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
886 | if (!in->dirfragtree.is_leaf(p->first)) | |
887 | in->fragmap.erase(p++); | |
888 | else | |
889 | ++p; | |
890 | } | |
891 | ||
892 | void Client::_fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds) | |
893 | { | |
894 | for (auto p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
895 | if (p->second == mds) | |
896 | in->fragmap.erase(p++); | |
897 | else | |
898 | ++p; | |
899 | } | |
900 | ||
901 | Inode * Client::add_update_inode(InodeStat *st, utime_t from, | |
902 | MetaSession *session, | |
903 | const UserPerm& request_perms) | |
904 | { | |
905 | Inode *in; | |
906 | bool was_new = false; | |
907 | if (inode_map.count(st->vino)) { | |
908 | in = inode_map[st->vino]; | |
11fdf7f2 | 909 | ldout(cct, 12) << __func__ << " had " << *in << " caps " << ccap_string(st->cap.caps) << dendl; |
7c673cae FG |
910 | } else { |
911 | in = new Inode(this, st->vino, &st->layout); | |
912 | inode_map[st->vino] = in; | |
913 | ||
914 | if (use_faked_inos()) | |
915 | _assign_faked_ino(in); | |
916 | ||
917 | if (!root) { | |
918 | root = in; | |
11fdf7f2 | 919 | if (use_faked_inos()) |
b3b6e05e | 920 | _assign_faked_root(root.get()); |
7c673cae FG |
921 | root_ancestor = in; |
922 | cwd = root; | |
f67539c2 | 923 | } else if (is_mounting()) { |
7c673cae FG |
924 | root_parents[root_ancestor] = in; |
925 | root_ancestor = in; | |
926 | } | |
927 | ||
928 | // immutable bits | |
929 | in->ino = st->vino.ino; | |
930 | in->snapid = st->vino.snapid; | |
931 | in->mode = st->mode & S_IFMT; | |
932 | was_new = true; | |
933 | } | |
934 | ||
935 | in->rdev = st->rdev; | |
936 | if (in->is_symlink()) | |
937 | in->symlink = st->symlink; | |
938 | ||
7c673cae | 939 | // only update inode if mds info is strictly newer, or it is the same and projected (odd). |
1adf2230 AA |
940 | bool new_version = false; |
941 | if (in->version == 0 || | |
942 | ((st->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
943 | (in->version & ~1) < st->version)) | |
944 | new_version = true; | |
7c673cae | 945 | |
1adf2230 AA |
946 | int issued; |
947 | in->caps_issued(&issued); | |
948 | issued |= in->caps_dirty(); | |
949 | int new_issued = ~issued & (int)st->cap.caps; | |
7c673cae | 950 | |
1adf2230 AA |
951 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
952 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
953 | in->mode = st->mode; | |
954 | in->uid = st->uid; | |
955 | in->gid = st->gid; | |
956 | in->btime = st->btime; | |
81eedcae | 957 | in->snap_btime = st->snap_btime; |
f67539c2 | 958 | in->snap_metadata = st->snap_metadata; |
1adf2230 | 959 | } |
7c673cae | 960 | |
1adf2230 AA |
961 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
962 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
963 | in->nlink = st->nlink; | |
964 | } | |
7c673cae | 965 | |
1adf2230 AA |
966 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
967 | update_inode_file_time(in, issued, st->time_warp_seq, | |
968 | st->ctime, st->mtime, st->atime); | |
969 | } | |
7c673cae | 970 | |
1adf2230 AA |
971 | if (new_version || |
972 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7c673cae | 973 | in->layout = st->layout; |
1adf2230 AA |
974 | update_inode_file_size(in, issued, st->size, st->truncate_seq, st->truncate_size); |
975 | } | |
7c673cae | 976 | |
1adf2230 AA |
977 | if (in->is_dir()) { |
978 | if (new_version || (new_issued & CEPH_CAP_FILE_SHARED)) { | |
979 | in->dirstat = st->dirstat; | |
980 | } | |
981 | // dir_layout/rstat/quota are not tracked by capability, update them only if | |
982 | // the inode stat is from auth mds | |
983 | if (new_version || (st->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
7c673cae FG |
984 | in->dir_layout = st->dir_layout; |
985 | ldout(cct, 20) << " dir hash is " << (int)in->dir_layout.dl_dir_hash << dendl; | |
1adf2230 AA |
986 | in->rstat = st->rstat; |
987 | in->quota = st->quota; | |
11fdf7f2 | 988 | in->dir_pin = st->dir_pin; |
1adf2230 AA |
989 | } |
990 | // move me if/when version reflects fragtree changes. | |
991 | if (in->dirfragtree != st->dirfragtree) { | |
992 | in->dirfragtree = st->dirfragtree; | |
993 | _fragmap_remove_non_leaves(in); | |
7c673cae | 994 | } |
7c673cae FG |
995 | } |
996 | ||
997 | if ((in->xattr_version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && | |
998 | st->xattrbl.length() && | |
999 | st->xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
1000 | auto p = st->xattrbl.cbegin(); |
1001 | decode(in->xattrs, p); | |
7c673cae FG |
1002 | in->xattr_version = st->xattr_version; |
1003 | } | |
1004 | ||
1adf2230 AA |
1005 | if (st->inline_version > in->inline_version) { |
1006 | in->inline_data = st->inline_data; | |
1007 | in->inline_version = st->inline_version; | |
7c673cae FG |
1008 | } |
1009 | ||
1adf2230 AA |
1010 | /* always take a newer change attr */ |
1011 | if (st->change_attr > in->change_attr) | |
1012 | in->change_attr = st->change_attr; | |
1013 | ||
1014 | if (st->version > in->version) | |
1015 | in->version = st->version; | |
1016 | ||
1017 | if (was_new) | |
1018 | ldout(cct, 12) << __func__ << " adding " << *in << " caps " << ccap_string(st->cap.caps) << dendl; | |
1019 | ||
1020 | if (!st->cap.caps) | |
1021 | return in; // as with readdir returning indoes in different snaprealms (no caps!) | |
1022 | ||
7c673cae | 1023 | if (in->snapid == CEPH_NOSNAP) { |
a8e16298 TL |
1024 | add_update_cap(in, session, st->cap.cap_id, st->cap.caps, st->cap.wanted, |
1025 | st->cap.seq, st->cap.mseq, inodeno_t(st->cap.realm), | |
1026 | st->cap.flags, request_perms); | |
28e407b8 | 1027 | if (in->auth_cap && in->auth_cap->session == session) { |
7c673cae | 1028 | in->max_size = st->max_size; |
28e407b8 AA |
1029 | in->rstat = st->rstat; |
1030 | } | |
7c673cae | 1031 | |
1adf2230 AA |
1032 | // setting I_COMPLETE needs to happen after adding the cap |
1033 | if (in->is_dir() && | |
1034 | (st->cap.caps & CEPH_CAP_FILE_SHARED) && | |
1035 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
1036 | in->dirstat.nfiles == 0 && | |
1037 | in->dirstat.nsubdirs == 0) { | |
1038 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on empty dir " << *in << dendl; | |
1039 | in->flags |= I_COMPLETE | I_DIR_ORDERED; | |
1040 | if (in->dir) { | |
1041 | ldout(cct, 10) << " dir is open on empty dir " << in->ino << " with " | |
1042 | << in->dir->dentries.size() << " entries, marking all dentries null" << dendl; | |
1043 | in->dir->readdir_cache.clear(); | |
1044 | for (const auto& p : in->dir->dentries) { | |
1045 | unlink(p.second, true, true); // keep dir, keep dentry | |
1046 | } | |
1047 | if (in->dir->dentries.empty()) | |
1048 | close_dir(in->dir); | |
7c673cae | 1049 | } |
7c673cae | 1050 | } |
1adf2230 AA |
1051 | } else { |
1052 | in->snap_caps |= st->cap.caps; | |
7c673cae FG |
1053 | } |
1054 | ||
f67539c2 | 1055 | in->fscrypt = st->fscrypt; |
7c673cae FG |
1056 | return in; |
1057 | } | |
1058 | ||
1059 | ||
1060 | /* | |
1061 | * insert_dentry_inode - insert + link a single dentry + inode into the metadata cache. | |
1062 | */ | |
1063 | Dentry *Client::insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease, | |
1064 | Inode *in, utime_t from, MetaSession *session, | |
1065 | Dentry *old_dentry) | |
1066 | { | |
1067 | Dentry *dn = NULL; | |
1068 | if (dir->dentries.count(dname)) | |
1069 | dn = dir->dentries[dname]; | |
1070 | ||
11fdf7f2 | 1071 | ldout(cct, 12) << __func__ << " '" << dname << "' vino " << in->vino() |
7c673cae FG |
1072 | << " in dir " << dir->parent_inode->vino() << " dn " << dn |
1073 | << dendl; | |
1074 | ||
1075 | if (dn && dn->inode) { | |
1076 | if (dn->inode->vino() == in->vino()) { | |
1077 | touch_dn(dn); | |
1078 | ldout(cct, 12) << " had dentry " << dname | |
1079 | << " with correct vino " << dn->inode->vino() | |
1080 | << dendl; | |
1081 | } else { | |
1082 | ldout(cct, 12) << " had dentry " << dname | |
1083 | << " with WRONG vino " << dn->inode->vino() | |
1084 | << dendl; | |
1085 | unlink(dn, true, true); // keep dir, keep dentry | |
1086 | } | |
1087 | } | |
1088 | ||
1089 | if (!dn || !dn->inode) { | |
1090 | InodeRef tmp_ref(in); | |
1091 | if (old_dentry) { | |
1092 | if (old_dentry->dir != dir) { | |
1093 | Inode *old_diri = old_dentry->dir->parent_inode; | |
7c673cae FG |
1094 | clear_dir_complete_and_ordered(old_diri, false); |
1095 | } | |
1096 | unlink(old_dentry, dir == old_dentry->dir, false); // drop dentry, keep dir open if its the same dir | |
1097 | } | |
1098 | Inode *diri = dir->parent_inode; | |
7c673cae FG |
1099 | clear_dir_complete_and_ordered(diri, false); |
1100 | dn = link(dir, dname, in, dn); | |
1101 | } | |
1102 | ||
1103 | update_dentry_lease(dn, dlease, from, session); | |
1104 | return dn; | |
1105 | } | |
1106 | ||
1107 | void Client::update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session) | |
1108 | { | |
1109 | utime_t dttl = from; | |
1110 | dttl += (float)dlease->duration_ms / 1000.0; | |
f67539c2 TL |
1111 | |
1112 | ldout(cct, 15) << __func__ << " " << *dn << " " << *dlease << " from " << from << dendl; | |
7c673cae | 1113 | |
11fdf7f2 | 1114 | ceph_assert(dn); |
7c673cae | 1115 | |
9f95a23c | 1116 | if (dlease->mask & CEPH_LEASE_VALID) { |
7c673cae FG |
1117 | if (dttl > dn->lease_ttl) { |
1118 | ldout(cct, 10) << "got dentry lease on " << dn->name | |
1119 | << " dur " << dlease->duration_ms << "ms ttl " << dttl << dendl; | |
1120 | dn->lease_ttl = dttl; | |
1121 | dn->lease_mds = session->mds_num; | |
1122 | dn->lease_seq = dlease->seq; | |
1123 | dn->lease_gen = session->cap_gen; | |
1124 | } | |
1125 | } | |
1126 | dn->cap_shared_gen = dn->dir->parent_inode->shared_gen; | |
f91f0fd5 TL |
1127 | if (dlease->mask & CEPH_LEASE_PRIMARY_LINK) |
1128 | dn->mark_primary(); | |
f67539c2 | 1129 | dn->alternate_name = std::move(dlease->alternate_name); |
7c673cae FG |
1130 | } |
1131 | ||
1132 | ||
1133 | /* | |
1134 | * update MDS location cache for a single inode | |
1135 | */ | |
522d829b | 1136 | void Client::update_dir_dist(Inode *in, DirStat *dst, mds_rank_t from) |
7c673cae FG |
1137 | { |
1138 | // auth | |
1139 | ldout(cct, 20) << "got dirfrag map for " << in->ino << " frag " << dst->frag << " to mds " << dst->auth << dendl; | |
1140 | if (dst->auth >= 0) { | |
1141 | in->fragmap[dst->frag] = dst->auth; | |
1142 | } else { | |
1143 | in->fragmap.erase(dst->frag); | |
1144 | } | |
1145 | if (!in->dirfragtree.is_leaf(dst->frag)) { | |
1146 | in->dirfragtree.force_to_leaf(cct, dst->frag); | |
1147 | _fragmap_remove_non_leaves(in); | |
1148 | } | |
1149 | ||
522d829b TL |
1150 | // replicated, only update from auth mds reply |
1151 | if (from == dst->auth) { | |
1152 | in->dir_replicated = !dst->dist.empty(); | |
1153 | if (!dst->dist.empty()) | |
1154 | in->frag_repmap[dst->frag].assign(dst->dist.begin(), dst->dist.end()) ; | |
1155 | else | |
1156 | in->frag_repmap.erase(dst->frag); | |
1157 | } | |
7c673cae FG |
1158 | } |
1159 | ||
1160 | void Client::clear_dir_complete_and_ordered(Inode *diri, bool complete) | |
1161 | { | |
f91f0fd5 TL |
1162 | if (complete) |
1163 | diri->dir_release_count++; | |
1164 | else | |
1165 | diri->dir_ordered_count++; | |
7c673cae FG |
1166 | if (diri->flags & I_COMPLETE) { |
1167 | if (complete) { | |
1168 | ldout(cct, 10) << " clearing (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
1169 | diri->flags &= ~(I_COMPLETE | I_DIR_ORDERED); | |
1170 | } else { | |
1171 | if (diri->flags & I_DIR_ORDERED) { | |
1172 | ldout(cct, 10) << " clearing I_DIR_ORDERED on " << *diri << dendl; | |
1173 | diri->flags &= ~I_DIR_ORDERED; | |
1174 | } | |
1175 | } | |
1176 | if (diri->dir) | |
1177 | diri->dir->readdir_cache.clear(); | |
1178 | } | |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * insert results from readdir or lssnap into the metadata cache. | |
1183 | */ | |
1184 | void Client::insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri) { | |
1185 | ||
11fdf7f2 | 1186 | auto& reply = request->reply; |
7c673cae | 1187 | ConnectionRef con = request->reply->get_connection(); |
11fdf7f2 TL |
1188 | uint64_t features; |
1189 | if(session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1190 | features = (uint64_t)-1; | |
1191 | } | |
1192 | else { | |
1193 | features = con->get_features(); | |
1194 | } | |
7c673cae FG |
1195 | |
1196 | dir_result_t *dirp = request->dirp; | |
11fdf7f2 | 1197 | ceph_assert(dirp); |
7c673cae FG |
1198 | |
1199 | // the extra buffer list is only set for readdir and lssnap replies | |
11fdf7f2 | 1200 | auto p = reply->get_extra_bl().cbegin(); |
7c673cae FG |
1201 | if (!p.end()) { |
1202 | // snapdir? | |
1203 | if (request->head.op == CEPH_MDS_OP_LSSNAP) { | |
11fdf7f2 | 1204 | ceph_assert(diri); |
7c673cae FG |
1205 | diri = open_snapdir(diri); |
1206 | } | |
1207 | ||
1208 | // only open dir if we're actually adding stuff to it! | |
1209 | Dir *dir = diri->open_dir(); | |
11fdf7f2 | 1210 | ceph_assert(dir); |
7c673cae FG |
1211 | |
1212 | // dirstat | |
11fdf7f2 | 1213 | DirStat dst(p, features); |
7c673cae FG |
1214 | __u32 numdn; |
1215 | __u16 flags; | |
11fdf7f2 TL |
1216 | decode(numdn, p); |
1217 | decode(flags, p); | |
7c673cae FG |
1218 | |
1219 | bool end = ((unsigned)flags & CEPH_READDIR_FRAG_END); | |
1220 | bool hash_order = ((unsigned)flags & CEPH_READDIR_HASH_ORDER); | |
1221 | ||
1222 | frag_t fg = (unsigned)request->head.args.readdir.frag; | |
1223 | unsigned readdir_offset = dirp->next_offset; | |
1224 | string readdir_start = dirp->last_name; | |
11fdf7f2 | 1225 | ceph_assert(!readdir_start.empty() || readdir_offset == 2); |
7c673cae FG |
1226 | |
1227 | unsigned last_hash = 0; | |
1228 | if (hash_order) { | |
1229 | if (!readdir_start.empty()) { | |
1230 | last_hash = ceph_frag_value(diri->hash_dentry_name(readdir_start)); | |
1231 | } else if (flags & CEPH_READDIR_OFFSET_HASH) { | |
1232 | /* mds understands offset_hash */ | |
1233 | last_hash = (unsigned)request->head.args.readdir.offset_hash; | |
1234 | } | |
1235 | } | |
1236 | ||
1237 | if (fg != dst.frag) { | |
1238 | ldout(cct, 10) << "insert_trace got new frag " << fg << " -> " << dst.frag << dendl; | |
1239 | fg = dst.frag; | |
1240 | if (!hash_order) { | |
1241 | readdir_offset = 2; | |
1242 | readdir_start.clear(); | |
1243 | dirp->offset = dir_result_t::make_fpos(fg, readdir_offset, false); | |
1244 | } | |
1245 | } | |
1246 | ||
1247 | ldout(cct, 10) << __func__ << " " << numdn << " readdir items, end=" << end | |
1248 | << ", hash_order=" << hash_order | |
1249 | << ", readdir_start " << readdir_start | |
1250 | << ", last_hash " << last_hash | |
1251 | << ", next_offset " << readdir_offset << dendl; | |
1252 | ||
1253 | if (diri->snapid != CEPH_SNAPDIR && | |
1254 | fg.is_leftmost() && readdir_offset == 2 && | |
1255 | !(hash_order && last_hash)) { | |
1256 | dirp->release_count = diri->dir_release_count; | |
1257 | dirp->ordered_count = diri->dir_ordered_count; | |
1258 | dirp->start_shared_gen = diri->shared_gen; | |
1259 | dirp->cache_index = 0; | |
1260 | } | |
1261 | ||
1262 | dirp->buffer_frag = fg; | |
1263 | ||
1264 | _readdir_drop_dirp_buffer(dirp); | |
1265 | dirp->buffer.reserve(numdn); | |
1266 | ||
1267 | string dname; | |
1268 | LeaseStat dlease; | |
1269 | for (unsigned i=0; i<numdn; i++) { | |
11fdf7f2 TL |
1270 | decode(dname, p); |
1271 | dlease.decode(p, features); | |
7c673cae FG |
1272 | InodeStat ist(p, features); |
1273 | ||
1274 | ldout(cct, 15) << "" << i << ": '" << dname << "'" << dendl; | |
1275 | ||
1276 | Inode *in = add_update_inode(&ist, request->sent_stamp, session, | |
1277 | request->perms); | |
1278 | Dentry *dn; | |
1279 | if (diri->dir->dentries.count(dname)) { | |
1280 | Dentry *olddn = diri->dir->dentries[dname]; | |
1281 | if (olddn->inode != in) { | |
1282 | // replace incorrect dentry | |
1283 | unlink(olddn, true, true); // keep dir, dentry | |
1284 | dn = link(dir, dname, in, olddn); | |
11fdf7f2 | 1285 | ceph_assert(dn == olddn); |
7c673cae FG |
1286 | } else { |
1287 | // keep existing dn | |
1288 | dn = olddn; | |
1289 | touch_dn(dn); | |
1290 | } | |
1291 | } else { | |
1292 | // new dn | |
1293 | dn = link(dir, dname, in, NULL); | |
1294 | } | |
f67539c2 | 1295 | dn->alternate_name = std::move(dlease.alternate_name); |
7c673cae FG |
1296 | |
1297 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1298 | if (hash_order) { | |
1299 | unsigned hash = ceph_frag_value(diri->hash_dentry_name(dname)); | |
1300 | if (hash != last_hash) | |
1301 | readdir_offset = 2; | |
1302 | last_hash = hash; | |
1303 | dn->offset = dir_result_t::make_fpos(hash, readdir_offset++, true); | |
1304 | } else { | |
1305 | dn->offset = dir_result_t::make_fpos(fg, readdir_offset++, false); | |
1306 | } | |
1307 | // add to readdir cache | |
1308 | if (dirp->release_count == diri->dir_release_count && | |
1309 | dirp->ordered_count == diri->dir_ordered_count && | |
1310 | dirp->start_shared_gen == diri->shared_gen) { | |
1311 | if (dirp->cache_index == dir->readdir_cache.size()) { | |
1312 | if (i == 0) { | |
11fdf7f2 | 1313 | ceph_assert(!dirp->inode->is_complete_and_ordered()); |
7c673cae FG |
1314 | dir->readdir_cache.reserve(dirp->cache_index + numdn); |
1315 | } | |
1316 | dir->readdir_cache.push_back(dn); | |
1317 | } else if (dirp->cache_index < dir->readdir_cache.size()) { | |
1318 | if (dirp->inode->is_complete_and_ordered()) | |
11fdf7f2 | 1319 | ceph_assert(dir->readdir_cache[dirp->cache_index] == dn); |
7c673cae FG |
1320 | else |
1321 | dir->readdir_cache[dirp->cache_index] = dn; | |
1322 | } else { | |
11fdf7f2 | 1323 | ceph_abort_msg("unexpected readdir buffer idx"); |
7c673cae FG |
1324 | } |
1325 | dirp->cache_index++; | |
1326 | } | |
1327 | // add to cached result list | |
f67539c2 | 1328 | dirp->buffer.push_back(dir_result_t::dentry(dn->offset, dname, dn->alternate_name, in)); |
7c673cae FG |
1329 | ldout(cct, 15) << __func__ << " " << hex << dn->offset << dec << ": '" << dname << "' -> " << in->ino << dendl; |
1330 | } | |
1331 | ||
1332 | if (numdn > 0) | |
1333 | dirp->last_name = dname; | |
1334 | if (end) | |
1335 | dirp->next_offset = 2; | |
1336 | else | |
1337 | dirp->next_offset = readdir_offset; | |
1338 | ||
1339 | if (dir->is_empty()) | |
1340 | close_dir(dir); | |
1341 | } | |
1342 | } | |
1343 | ||
1344 | /** insert_trace | |
1345 | * | |
1346 | * insert a trace from a MDS reply into the cache. | |
1347 | */ | |
1348 | Inode* Client::insert_trace(MetaRequest *request, MetaSession *session) | |
1349 | { | |
11fdf7f2 | 1350 | auto& reply = request->reply; |
7c673cae FG |
1351 | int op = request->get_op(); |
1352 | ||
1353 | ldout(cct, 10) << "insert_trace from " << request->sent_stamp << " mds." << session->mds_num | |
1354 | << " is_target=" << (int)reply->head.is_target | |
1355 | << " is_dentry=" << (int)reply->head.is_dentry | |
1356 | << dendl; | |
1357 | ||
11fdf7f2 | 1358 | auto p = reply->get_trace_bl().cbegin(); |
7c673cae FG |
1359 | if (request->got_unsafe) { |
1360 | ldout(cct, 10) << "insert_trace -- already got unsafe; ignoring" << dendl; | |
11fdf7f2 | 1361 | ceph_assert(p.end()); |
7c673cae FG |
1362 | return NULL; |
1363 | } | |
1364 | ||
1365 | if (p.end()) { | |
1366 | ldout(cct, 10) << "insert_trace -- no trace" << dendl; | |
1367 | ||
1368 | Dentry *d = request->dentry(); | |
1369 | if (d) { | |
1370 | Inode *diri = d->dir->parent_inode; | |
7c673cae FG |
1371 | clear_dir_complete_and_ordered(diri, true); |
1372 | } | |
1373 | ||
1374 | if (d && reply->get_result() == 0) { | |
1375 | if (op == CEPH_MDS_OP_RENAME) { | |
1376 | // rename | |
1377 | Dentry *od = request->old_dentry(); | |
1378 | ldout(cct, 10) << " unlinking rename src dn " << od << " for traceless reply" << dendl; | |
11fdf7f2 | 1379 | ceph_assert(od); |
7c673cae FG |
1380 | unlink(od, true, true); // keep dir, dentry |
1381 | } else if (op == CEPH_MDS_OP_RMDIR || | |
1382 | op == CEPH_MDS_OP_UNLINK) { | |
1383 | // unlink, rmdir | |
1384 | ldout(cct, 10) << " unlinking unlink/rmdir dn " << d << " for traceless reply" << dendl; | |
1385 | unlink(d, true, true); // keep dir, dentry | |
1386 | } | |
1387 | } | |
1388 | return NULL; | |
1389 | } | |
1390 | ||
1391 | ConnectionRef con = request->reply->get_connection(); | |
11fdf7f2 TL |
1392 | uint64_t features; |
1393 | if (session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1394 | features = (uint64_t)-1; | |
1395 | } | |
1396 | else { | |
1397 | features = con->get_features(); | |
1398 | } | |
7c673cae FG |
1399 | ldout(cct, 10) << " features 0x" << hex << features << dec << dendl; |
1400 | ||
1401 | // snap trace | |
1402 | SnapRealm *realm = NULL; | |
1403 | if (reply->snapbl.length()) | |
1404 | update_snap_trace(reply->snapbl, &realm); | |
1405 | ||
1406 | ldout(cct, 10) << " hrm " | |
1407 | << " is_target=" << (int)reply->head.is_target | |
1408 | << " is_dentry=" << (int)reply->head.is_dentry | |
1409 | << dendl; | |
1410 | ||
1411 | InodeStat dirst; | |
1412 | DirStat dst; | |
1413 | string dname; | |
1414 | LeaseStat dlease; | |
1415 | InodeStat ist; | |
1416 | ||
1417 | if (reply->head.is_dentry) { | |
1418 | dirst.decode(p, features); | |
11fdf7f2 TL |
1419 | dst.decode(p, features); |
1420 | decode(dname, p); | |
1421 | dlease.decode(p, features); | |
7c673cae FG |
1422 | } |
1423 | ||
1424 | Inode *in = 0; | |
1425 | if (reply->head.is_target) { | |
1426 | ist.decode(p, features); | |
1427 | if (cct->_conf->client_debug_getattr_caps) { | |
1428 | unsigned wanted = 0; | |
1429 | if (op == CEPH_MDS_OP_GETATTR || op == CEPH_MDS_OP_LOOKUP) | |
1430 | wanted = request->head.args.getattr.mask; | |
1431 | else if (op == CEPH_MDS_OP_OPEN || op == CEPH_MDS_OP_CREATE) | |
1432 | wanted = request->head.args.open.mask; | |
1433 | ||
1434 | if ((wanted & CEPH_CAP_XATTR_SHARED) && | |
1435 | !(ist.xattr_version > 0 && ist.xattrbl.length() > 0)) | |
11fdf7f2 | 1436 | ceph_abort_msg("MDS reply does not contain xattrs"); |
7c673cae FG |
1437 | } |
1438 | ||
1439 | in = add_update_inode(&ist, request->sent_stamp, session, | |
1440 | request->perms); | |
1441 | } | |
1442 | ||
1443 | Inode *diri = NULL; | |
1444 | if (reply->head.is_dentry) { | |
1445 | diri = add_update_inode(&dirst, request->sent_stamp, session, | |
1446 | request->perms); | |
522d829b TL |
1447 | mds_rank_t from_mds = mds_rank_t(reply->get_source().num()); |
1448 | update_dir_dist(diri, &dst, from_mds); // dir stat info is attached to .. | |
7c673cae FG |
1449 | |
1450 | if (in) { | |
1451 | Dir *dir = diri->open_dir(); | |
1452 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session, | |
1453 | (op == CEPH_MDS_OP_RENAME) ? request->old_dentry() : NULL); | |
1454 | } else { | |
1455 | Dentry *dn = NULL; | |
1456 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1457 | dn = diri->dir->dentries[dname]; | |
1458 | if (dn->inode) { | |
7c673cae FG |
1459 | clear_dir_complete_and_ordered(diri, false); |
1460 | unlink(dn, true, true); // keep dir, dentry | |
1461 | } | |
1462 | } | |
1463 | if (dlease.duration_ms > 0) { | |
1464 | if (!dn) { | |
1465 | Dir *dir = diri->open_dir(); | |
1466 | dn = link(dir, dname, NULL, NULL); | |
1467 | } | |
1468 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1469 | } | |
1470 | } | |
1471 | } else if (op == CEPH_MDS_OP_LOOKUPSNAP || | |
1472 | op == CEPH_MDS_OP_MKSNAP) { | |
1473 | ldout(cct, 10) << " faking snap lookup weirdness" << dendl; | |
1474 | // fake it for snap lookup | |
1475 | vinodeno_t vino = ist.vino; | |
1476 | vino.snapid = CEPH_SNAPDIR; | |
11fdf7f2 | 1477 | ceph_assert(inode_map.count(vino)); |
7c673cae FG |
1478 | diri = inode_map[vino]; |
1479 | ||
1480 | string dname = request->path.last_dentry(); | |
1481 | ||
1482 | LeaseStat dlease; | |
1483 | dlease.duration_ms = 0; | |
1484 | ||
1485 | if (in) { | |
1486 | Dir *dir = diri->open_dir(); | |
1487 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session); | |
1488 | } else { | |
1489 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1490 | Dentry *dn = diri->dir->dentries[dname]; | |
1491 | if (dn->inode) | |
1492 | unlink(dn, true, true); // keep dir, dentry | |
1493 | } | |
1494 | } | |
1495 | } | |
1496 | ||
1497 | if (in) { | |
1498 | if (op == CEPH_MDS_OP_READDIR || | |
1499 | op == CEPH_MDS_OP_LSSNAP) { | |
1500 | insert_readdir_results(request, session, in); | |
1501 | } else if (op == CEPH_MDS_OP_LOOKUPNAME) { | |
1502 | // hack: return parent inode instead | |
1503 | in = diri; | |
1504 | } | |
1505 | ||
1506 | if (request->dentry() == NULL && in != request->inode()) { | |
1507 | // pin the target inode if its parent dentry is not pinned | |
1508 | request->set_other_inode(in); | |
1509 | } | |
1510 | } | |
1511 | ||
1512 | if (realm) | |
1513 | put_snap_realm(realm); | |
1514 | ||
1515 | request->target = in; | |
1516 | return in; | |
1517 | } | |
1518 | ||
1519 | // ------- | |
1520 | ||
1521 | mds_rank_t Client::choose_target_mds(MetaRequest *req, Inode** phash_diri) | |
1522 | { | |
1523 | mds_rank_t mds = MDS_RANK_NONE; | |
1524 | __u32 hash = 0; | |
1525 | bool is_hash = false; | |
1526 | ||
1527 | Inode *in = NULL; | |
1528 | Dentry *de = NULL; | |
7c673cae FG |
1529 | |
1530 | if (req->resend_mds >= 0) { | |
1531 | mds = req->resend_mds; | |
1532 | req->resend_mds = -1; | |
11fdf7f2 | 1533 | ldout(cct, 10) << __func__ << " resend_mds specified as mds." << mds << dendl; |
7c673cae FG |
1534 | goto out; |
1535 | } | |
1536 | ||
1537 | if (cct->_conf->client_use_random_mds) | |
1538 | goto random_mds; | |
1539 | ||
1540 | in = req->inode(); | |
1541 | de = req->dentry(); | |
1542 | if (in) { | |
11fdf7f2 | 1543 | ldout(cct, 20) << __func__ << " starting with req->inode " << *in << dendl; |
7c673cae FG |
1544 | if (req->path.depth()) { |
1545 | hash = in->hash_dentry_name(req->path[0]); | |
11fdf7f2 | 1546 | ldout(cct, 20) << __func__ << " inode dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1547 | << " on " << req->path[0] |
1548 | << " => " << hash << dendl; | |
1549 | is_hash = true; | |
1550 | } | |
1551 | } else if (de) { | |
1552 | if (de->inode) { | |
1553 | in = de->inode.get(); | |
11fdf7f2 | 1554 | ldout(cct, 20) << __func__ << " starting with req->dentry inode " << *in << dendl; |
7c673cae FG |
1555 | } else { |
1556 | in = de->dir->parent_inode; | |
1557 | hash = in->hash_dentry_name(de->name); | |
11fdf7f2 | 1558 | ldout(cct, 20) << __func__ << " dentry dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1559 | << " on " << de->name |
1560 | << " => " << hash << dendl; | |
1561 | is_hash = true; | |
1562 | } | |
1563 | } | |
1564 | if (in) { | |
1565 | if (in->snapid != CEPH_NOSNAP) { | |
11fdf7f2 | 1566 | ldout(cct, 10) << __func__ << " " << *in << " is snapped, using nonsnap parent" << dendl; |
7c673cae FG |
1567 | while (in->snapid != CEPH_NOSNAP) { |
1568 | if (in->snapid == CEPH_SNAPDIR) | |
1569 | in = in->snapdir_parent.get(); | |
11fdf7f2 | 1570 | else if (!in->dentries.empty()) |
7c673cae FG |
1571 | /* In most cases there will only be one dentry, so getting it |
1572 | * will be the correct action. If there are multiple hard links, | |
1573 | * I think the MDS should be able to redirect as needed*/ | |
1574 | in = in->get_first_parent()->dir->parent_inode; | |
1575 | else { | |
1576 | ldout(cct, 10) << "got unlinked inode, can't look at parent" << dendl; | |
1577 | break; | |
1578 | } | |
1579 | } | |
1580 | is_hash = false; | |
1581 | } | |
1582 | ||
11fdf7f2 | 1583 | ldout(cct, 20) << __func__ << " " << *in << " is_hash=" << is_hash |
7c673cae FG |
1584 | << " hash=" << hash << dendl; |
1585 | ||
f67539c2 | 1586 | if (is_hash && S_ISDIR(in->mode) && (!in->fragmap.empty() || !in->frag_repmap.empty())) { |
7c673cae | 1587 | frag_t fg = in->dirfragtree[hash]; |
f67539c2 TL |
1588 | if (!req->auth_is_best()) { |
1589 | auto repmapit = in->frag_repmap.find(fg); | |
1590 | if (repmapit != in->frag_repmap.end()) { | |
1591 | auto& repmap = repmapit->second; | |
1592 | auto r = ceph::util::generate_random_number<uint64_t>(0, repmap.size()-1); | |
1593 | mds = repmap.at(r); | |
1594 | } | |
1595 | } else if (in->fragmap.count(fg)) { | |
7c673cae FG |
1596 | mds = in->fragmap[fg]; |
1597 | if (phash_diri) | |
1598 | *phash_diri = in; | |
91327a77 | 1599 | } else if (in->auth_cap) { |
f67539c2 | 1600 | req->send_to_auth = true; |
91327a77 AA |
1601 | mds = in->auth_cap->session->mds_num; |
1602 | } | |
1603 | if (mds >= 0) { | |
11fdf7f2 | 1604 | ldout(cct, 10) << __func__ << " from dirfragtree hash" << dendl; |
7c673cae FG |
1605 | goto out; |
1606 | } | |
1607 | } | |
1608 | ||
11fdf7f2 TL |
1609 | if (in->auth_cap && req->auth_is_best()) { |
1610 | mds = in->auth_cap->session->mds_num; | |
1611 | } else if (!in->caps.empty()) { | |
1612 | mds = in->caps.begin()->second.session->mds_num; | |
1613 | } else { | |
7c673cae | 1614 | goto random_mds; |
11fdf7f2 TL |
1615 | } |
1616 | ldout(cct, 10) << __func__ << " from caps on inode " << *in << dendl; | |
7c673cae FG |
1617 | |
1618 | goto out; | |
1619 | } | |
1620 | ||
1621 | random_mds: | |
1622 | if (mds < 0) { | |
1623 | mds = _get_random_up_mds(); | |
1624 | ldout(cct, 10) << "did not get mds through better means, so chose random mds " << mds << dendl; | |
1625 | } | |
1626 | ||
1627 | out: | |
1628 | ldout(cct, 20) << "mds is " << mds << dendl; | |
1629 | return mds; | |
1630 | } | |
1631 | ||
7c673cae FG |
1632 | void Client::connect_mds_targets(mds_rank_t mds) |
1633 | { | |
11fdf7f2 TL |
1634 | ldout(cct, 10) << __func__ << " for mds." << mds << dendl; |
1635 | ceph_assert(mds_sessions.count(mds)); | |
7c673cae | 1636 | const MDSMap::mds_info_t& info = mdsmap->get_mds_info(mds); |
f67539c2 TL |
1637 | for (const auto &rank : info.export_targets) { |
1638 | if (mds_sessions.count(rank) == 0 && | |
1639 | mdsmap->is_clientreplay_or_active_or_stopping(rank)) { | |
7c673cae | 1640 | ldout(cct, 10) << "check_mds_sessions opening mds." << mds |
f67539c2 TL |
1641 | << " export target mds." << rank << dendl; |
1642 | _open_mds_session(rank); | |
7c673cae FG |
1643 | } |
1644 | } | |
1645 | } | |
1646 | ||
adb31ebb | 1647 | void Client::dump_mds_sessions(Formatter *f, bool cap_dump) |
7c673cae FG |
1648 | { |
1649 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 1650 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 AA |
1651 | f->dump_object("inst", inst); |
1652 | f->dump_stream("inst_str") << inst; | |
1653 | f->dump_stream("addr_str") << inst.addr; | |
7c673cae | 1654 | f->open_array_section("sessions"); |
11fdf7f2 | 1655 | for (const auto &p : mds_sessions) { |
7c673cae | 1656 | f->open_object_section("session"); |
20effc67 | 1657 | p.second->dump(f, cap_dump); |
7c673cae FG |
1658 | f->close_section(); |
1659 | } | |
1660 | f->close_section(); | |
1661 | f->dump_int("mdsmap_epoch", mdsmap->get_epoch()); | |
1662 | } | |
f67539c2 | 1663 | |
7c673cae FG |
1664 | void Client::dump_mds_requests(Formatter *f) |
1665 | { | |
1666 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
1667 | p != mds_requests.end(); | |
1668 | ++p) { | |
1669 | f->open_object_section("request"); | |
1670 | p->second->dump(f); | |
1671 | f->close_section(); | |
1672 | } | |
1673 | } | |
1674 | ||
9f95a23c | 1675 | int Client::verify_reply_trace(int r, MetaSession *session, |
11fdf7f2 | 1676 | MetaRequest *request, const MConstRef<MClientReply>& reply, |
7c673cae FG |
1677 | InodeRef *ptarget, bool *pcreated, |
1678 | const UserPerm& perms) | |
1679 | { | |
1680 | // check whether this request actually did the create, and set created flag | |
1681 | bufferlist extra_bl; | |
1682 | inodeno_t created_ino; | |
1683 | bool got_created_ino = false; | |
1684 | ceph::unordered_map<vinodeno_t, Inode*>::iterator p; | |
1685 | ||
11fdf7f2 | 1686 | extra_bl = reply->get_extra_bl(); |
7c673cae | 1687 | if (extra_bl.length() >= 8) { |
9f95a23c TL |
1688 | if (session->mds_features.test(CEPHFS_FEATURE_DELEG_INO)) { |
1689 | struct openc_response_t ocres; | |
1690 | ||
1691 | decode(ocres, extra_bl); | |
1692 | created_ino = ocres.created_ino; | |
1693 | /* | |
1694 | * The userland cephfs client doesn't have a way to do an async create | |
1695 | * (yet), so just discard delegated_inos for now. Eventually we should | |
1696 | * store them and use them in create calls, even if they are synchronous, | |
1697 | * if only for testing purposes. | |
1698 | */ | |
1699 | ldout(cct, 10) << "delegated_inos: " << ocres.delegated_inos << dendl; | |
1700 | } else { | |
1701 | // u64 containing number of created ino | |
1702 | decode(created_ino, extra_bl); | |
1703 | } | |
7c673cae | 1704 | ldout(cct, 10) << "make_request created ino " << created_ino << dendl; |
9f95a23c | 1705 | got_created_ino = true; |
7c673cae FG |
1706 | } |
1707 | ||
1708 | if (pcreated) | |
1709 | *pcreated = got_created_ino; | |
1710 | ||
1711 | if (request->target) { | |
1712 | *ptarget = request->target; | |
1713 | ldout(cct, 20) << "make_request target is " << *ptarget->get() << dendl; | |
1714 | } else { | |
1715 | if (got_created_ino && (p = inode_map.find(vinodeno_t(created_ino, CEPH_NOSNAP))) != inode_map.end()) { | |
1716 | (*ptarget) = p->second; | |
1717 | ldout(cct, 20) << "make_request created, target is " << *ptarget->get() << dendl; | |
1718 | } else { | |
1719 | // we got a traceless reply, and need to look up what we just | |
1720 | // created. for now, do this by name. someday, do this by the | |
1721 | // ino... which we know! FIXME. | |
1722 | InodeRef target; | |
1723 | Dentry *d = request->dentry(); | |
1724 | if (d) { | |
1725 | if (d->dir) { | |
1726 | ldout(cct, 10) << "make_request got traceless reply, looking up #" | |
1727 | << d->dir->parent_inode->ino << "/" << d->name | |
1728 | << " got_ino " << got_created_ino | |
1729 | << " ino " << created_ino | |
1730 | << dendl; | |
1731 | r = _do_lookup(d->dir->parent_inode, d->name, request->regetattr_mask, | |
1732 | &target, perms); | |
1733 | } else { | |
1734 | // if the dentry is not linked, just do our best. see #5021. | |
11fdf7f2 | 1735 | ceph_abort_msg("how did this happen? i want logs!"); |
7c673cae FG |
1736 | } |
1737 | } else { | |
1738 | Inode *in = request->inode(); | |
1739 | ldout(cct, 10) << "make_request got traceless reply, forcing getattr on #" | |
1740 | << in->ino << dendl; | |
1741 | r = _getattr(in, request->regetattr_mask, perms, true); | |
1742 | target = in; | |
1743 | } | |
1744 | if (r >= 0) { | |
1745 | // verify ino returned in reply and trace_dist are the same | |
1746 | if (got_created_ino && | |
1747 | created_ino.val != target->ino.val) { | |
1748 | ldout(cct, 5) << "create got ino " << created_ino << " but then failed on lookup; EINTR?" << dendl; | |
f67539c2 | 1749 | r = -CEPHFS_EINTR; |
7c673cae FG |
1750 | } |
1751 | if (ptarget) | |
1752 | ptarget->swap(target); | |
1753 | } | |
1754 | } | |
1755 | } | |
1756 | ||
1757 | return r; | |
1758 | } | |
1759 | ||
1760 | ||
1761 | /** | |
1762 | * make a request | |
1763 | * | |
1764 | * Blocking helper to make an MDS request. | |
1765 | * | |
1766 | * If the ptarget flag is set, behavior changes slightly: the caller | |
1767 | * expects to get a pointer to the inode we are creating or operating | |
1768 | * on. As a result, we will follow up any traceless mutation reply | |
1769 | * with a getattr or lookup to transparently handle a traceless reply | |
1770 | * from the MDS (as when the MDS restarts and the client has to replay | |
1771 | * a request). | |
1772 | * | |
1773 | * @param request the MetaRequest to execute | |
1774 | * @param perms The user uid/gid to execute as (eventually, full group lists?) | |
1775 | * @param ptarget [optional] address to store a pointer to the target inode we want to create or operate on | |
1776 | * @param pcreated [optional; required if ptarget] where to store a bool of whether our create atomically created a file | |
1777 | * @param use_mds [optional] prefer a specific mds (-1 for default) | |
1778 | * @param pdirbl [optional; disallowed if ptarget] where to pass extra reply payload to the caller | |
1779 | */ | |
1780 | int Client::make_request(MetaRequest *request, | |
1781 | const UserPerm& perms, | |
1782 | InodeRef *ptarget, bool *pcreated, | |
1783 | mds_rank_t use_mds, | |
1784 | bufferlist *pdirbl) | |
1785 | { | |
1786 | int r = 0; | |
1787 | ||
1788 | // assign a unique tid | |
1789 | ceph_tid_t tid = ++last_tid; | |
1790 | request->set_tid(tid); | |
1791 | ||
1792 | // and timestamp | |
1793 | request->op_stamp = ceph_clock_now(); | |
1794 | ||
1795 | // make note | |
1796 | mds_requests[tid] = request->get(); | |
1797 | if (oldest_tid == 0 && request->get_op() != CEPH_MDS_OP_SETFILELOCK) | |
1798 | oldest_tid = tid; | |
1799 | ||
1800 | request->set_caller_perms(perms); | |
1801 | ||
1802 | if (cct->_conf->client_inject_fixed_oldest_tid) { | |
1803 | ldout(cct, 20) << __func__ << " injecting fixed oldest_client_tid(1)" << dendl; | |
1804 | request->set_oldest_client_tid(1); | |
1805 | } else { | |
1806 | request->set_oldest_client_tid(oldest_tid); | |
1807 | } | |
1808 | ||
1809 | // hack target mds? | |
1810 | if (use_mds >= 0) | |
1811 | request->resend_mds = use_mds; | |
1812 | ||
20effc67 | 1813 | MetaSessionRef session = NULL; |
7c673cae FG |
1814 | while (1) { |
1815 | if (request->aborted()) | |
1816 | break; | |
1817 | ||
f67539c2 TL |
1818 | if (blocklisted) { |
1819 | request->abort(-CEPHFS_EBLOCKLISTED); | |
31f18b77 FG |
1820 | break; |
1821 | } | |
1822 | ||
7c673cae | 1823 | // set up wait cond |
9f95a23c | 1824 | ceph::condition_variable caller_cond; |
7c673cae FG |
1825 | request->caller_cond = &caller_cond; |
1826 | ||
1827 | // choose mds | |
1828 | Inode *hash_diri = NULL; | |
1829 | mds_rank_t mds = choose_target_mds(request, &hash_diri); | |
1830 | int mds_state = (mds == MDS_RANK_NONE) ? MDSMap::STATE_NULL : mdsmap->get_state(mds); | |
1831 | if (mds_state != MDSMap::STATE_ACTIVE && mds_state != MDSMap::STATE_STOPPING) { | |
1832 | if (mds_state == MDSMap::STATE_NULL && mds >= mdsmap->get_max_mds()) { | |
1833 | if (hash_diri) { | |
1834 | ldout(cct, 10) << " target mds." << mds << " has stopped, remove it from fragmap" << dendl; | |
1835 | _fragmap_remove_stopped_mds(hash_diri, mds); | |
1836 | } else { | |
1837 | ldout(cct, 10) << " target mds." << mds << " has stopped, trying a random mds" << dendl; | |
1838 | request->resend_mds = _get_random_up_mds(); | |
1839 | } | |
1840 | } else { | |
1841 | ldout(cct, 10) << " target mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
1842 | wait_on_list(waiting_for_mdsmap); | |
1843 | } | |
1844 | continue; | |
1845 | } | |
1846 | ||
1847 | // open a session? | |
7c673cae FG |
1848 | if (!have_open_session(mds)) { |
1849 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 1850 | if (session->state == MetaSession::STATE_REJECTED) { |
f67539c2 | 1851 | request->abort(-CEPHFS_EPERM); |
f6b5b4d7 TL |
1852 | break; |
1853 | } | |
7c673cae FG |
1854 | // wait |
1855 | if (session->state == MetaSession::STATE_OPENING) { | |
1856 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
1857 | wait_on_context_list(session->waiting_for_open); | |
7c673cae FG |
1858 | continue; |
1859 | } | |
1860 | ||
1861 | if (!have_open_session(mds)) | |
1862 | continue; | |
1863 | } else { | |
20effc67 | 1864 | session = mds_sessions.at(mds); |
7c673cae FG |
1865 | } |
1866 | ||
1867 | // send request. | |
20effc67 | 1868 | send_request(request, session.get()); |
7c673cae FG |
1869 | |
1870 | // wait for signal | |
1871 | ldout(cct, 20) << "awaiting reply|forward|kick on " << &caller_cond << dendl; | |
1872 | request->kick = false; | |
9f95a23c TL |
1873 | std::unique_lock l{client_lock, std::adopt_lock}; |
1874 | caller_cond.wait(l, [request] { | |
1875 | return (request->reply || // reply | |
1876 | request->resend_mds >= 0 || // forward | |
1877 | request->kick); | |
1878 | }); | |
1879 | l.release(); | |
1880 | request->caller_cond = nullptr; | |
7c673cae FG |
1881 | |
1882 | // did we get a reply? | |
1883 | if (request->reply) | |
1884 | break; | |
1885 | } | |
1886 | ||
1887 | if (!request->reply) { | |
11fdf7f2 TL |
1888 | ceph_assert(request->aborted()); |
1889 | ceph_assert(!request->got_unsafe); | |
7c673cae FG |
1890 | r = request->get_abort_code(); |
1891 | request->item.remove_myself(); | |
1892 | unregister_request(request); | |
11fdf7f2 | 1893 | put_request(request); |
7c673cae FG |
1894 | return r; |
1895 | } | |
1896 | ||
1897 | // got it! | |
11fdf7f2 | 1898 | auto reply = std::move(request->reply); |
7c673cae FG |
1899 | r = reply->get_result(); |
1900 | if (r >= 0) | |
1901 | request->success = true; | |
1902 | ||
1903 | // kick dispatcher (we've got it!) | |
11fdf7f2 | 1904 | ceph_assert(request->dispatch_cond); |
9f95a23c | 1905 | request->dispatch_cond->notify_all(); |
7c673cae FG |
1906 | ldout(cct, 20) << "sendrecv kickback on tid " << tid << " " << request->dispatch_cond << dendl; |
1907 | request->dispatch_cond = 0; | |
1908 | ||
1909 | if (r >= 0 && ptarget) | |
20effc67 | 1910 | r = verify_reply_trace(r, session.get(), request, reply, ptarget, pcreated, perms); |
7c673cae FG |
1911 | |
1912 | if (pdirbl) | |
11fdf7f2 | 1913 | *pdirbl = reply->get_extra_bl(); |
7c673cae FG |
1914 | |
1915 | // -- log times -- | |
1916 | utime_t lat = ceph_clock_now(); | |
1917 | lat -= request->sent_stamp; | |
1918 | ldout(cct, 20) << "lat " << lat << dendl; | |
1919 | logger->tinc(l_c_lat, lat); | |
1920 | logger->tinc(l_c_reply, lat); | |
1921 | ||
1922 | put_request(request); | |
7c673cae FG |
1923 | return r; |
1924 | } | |
1925 | ||
1926 | void Client::unregister_request(MetaRequest *req) | |
1927 | { | |
1928 | mds_requests.erase(req->tid); | |
1929 | if (req->tid == oldest_tid) { | |
1930 | map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.upper_bound(oldest_tid); | |
1931 | while (true) { | |
1932 | if (p == mds_requests.end()) { | |
1933 | oldest_tid = 0; | |
1934 | break; | |
1935 | } | |
1936 | if (p->second->get_op() != CEPH_MDS_OP_SETFILELOCK) { | |
1937 | oldest_tid = p->first; | |
1938 | break; | |
1939 | } | |
1940 | ++p; | |
1941 | } | |
1942 | } | |
1943 | put_request(req); | |
1944 | } | |
1945 | ||
1946 | void Client::put_request(MetaRequest *request) | |
1947 | { | |
1948 | if (request->_put()) { | |
1949 | int op = -1; | |
1950 | if (request->success) | |
1951 | op = request->get_op(); | |
1952 | InodeRef other_in; | |
1953 | request->take_other_inode(&other_in); | |
1954 | delete request; | |
1955 | ||
1956 | if (other_in && | |
1957 | (op == CEPH_MDS_OP_RMDIR || | |
1958 | op == CEPH_MDS_OP_RENAME || | |
1959 | op == CEPH_MDS_OP_RMSNAP)) { | |
1960 | _try_to_trim_inode(other_in.get(), false); | |
1961 | } | |
1962 | } | |
1963 | } | |
1964 | ||
1965 | int Client::encode_inode_release(Inode *in, MetaRequest *req, | |
1966 | mds_rank_t mds, int drop, | |
1967 | int unless, int force) | |
1968 | { | |
11fdf7f2 | 1969 | ldout(cct, 20) << __func__ << " enter(in:" << *in << ", req:" << req |
f67539c2 | 1970 | << " mds:" << mds << ", drop:" << ccap_string(drop) << ", unless:" << ccap_string(unless) |
1911f103 | 1971 | << ", force:" << force << ")" << dendl; |
7c673cae | 1972 | int released = 0; |
11fdf7f2 TL |
1973 | auto it = in->caps.find(mds); |
1974 | if (it != in->caps.end()) { | |
1975 | Cap &cap = it->second; | |
7c673cae | 1976 | drop &= ~(in->dirty_caps | get_caps_used(in)); |
11fdf7f2 TL |
1977 | if ((drop & cap.issued) && |
1978 | !(unless & cap.issued)) { | |
1911f103 | 1979 | ldout(cct, 25) << "dropping caps " << ccap_string(drop) << dendl; |
11fdf7f2 TL |
1980 | cap.issued &= ~drop; |
1981 | cap.implemented &= ~drop; | |
7c673cae | 1982 | released = 1; |
7c673cae FG |
1983 | } else { |
1984 | released = force; | |
1985 | } | |
1986 | if (released) { | |
1911f103 TL |
1987 | cap.wanted = in->caps_wanted(); |
1988 | if (&cap == in->auth_cap && | |
1989 | !(cap.wanted & CEPH_CAP_ANY_FILE_WR)) { | |
1990 | in->requested_max_size = 0; | |
1991 | ldout(cct, 25) << "reset requested_max_size due to not wanting any file write cap" << dendl; | |
1992 | } | |
7c673cae FG |
1993 | ceph_mds_request_release rel; |
1994 | rel.ino = in->ino; | |
11fdf7f2 TL |
1995 | rel.cap_id = cap.cap_id; |
1996 | rel.seq = cap.seq; | |
1997 | rel.issue_seq = cap.issue_seq; | |
1998 | rel.mseq = cap.mseq; | |
1999 | rel.caps = cap.implemented; | |
2000 | rel.wanted = cap.wanted; | |
7c673cae FG |
2001 | rel.dname_len = 0; |
2002 | rel.dname_seq = 0; | |
2003 | req->cap_releases.push_back(MClientRequest::Release(rel,"")); | |
2004 | } | |
2005 | } | |
11fdf7f2 | 2006 | ldout(cct, 25) << __func__ << " exit(in:" << *in << ") released:" |
7c673cae FG |
2007 | << released << dendl; |
2008 | return released; | |
2009 | } | |
2010 | ||
2011 | void Client::encode_dentry_release(Dentry *dn, MetaRequest *req, | |
2012 | mds_rank_t mds, int drop, int unless) | |
2013 | { | |
11fdf7f2 | 2014 | ldout(cct, 20) << __func__ << " enter(dn:" |
7c673cae FG |
2015 | << dn << ")" << dendl; |
2016 | int released = 0; | |
2017 | if (dn->dir) | |
2018 | released = encode_inode_release(dn->dir->parent_inode, req, | |
2019 | mds, drop, unless, 1); | |
2020 | if (released && dn->lease_mds == mds) { | |
2021 | ldout(cct, 25) << "preemptively releasing dn to mds" << dendl; | |
11fdf7f2 | 2022 | auto& rel = req->cap_releases.back(); |
7c673cae FG |
2023 | rel.item.dname_len = dn->name.length(); |
2024 | rel.item.dname_seq = dn->lease_seq; | |
2025 | rel.dname = dn->name; | |
adb31ebb | 2026 | dn->lease_mds = -1; |
7c673cae | 2027 | } |
11fdf7f2 | 2028 | ldout(cct, 25) << __func__ << " exit(dn:" |
7c673cae FG |
2029 | << dn << ")" << dendl; |
2030 | } | |
2031 | ||
2032 | ||
2033 | /* | |
2034 | * This requires the MClientRequest *request member to be set. | |
2035 | * It will error out horribly without one. | |
2036 | * Additionally, if you set any *drop member, you'd better have | |
2037 | * set the corresponding dentry! | |
2038 | */ | |
2039 | void Client::encode_cap_releases(MetaRequest *req, mds_rank_t mds) | |
2040 | { | |
11fdf7f2 | 2041 | ldout(cct, 20) << __func__ << " enter (req: " |
7c673cae FG |
2042 | << req << ", mds: " << mds << ")" << dendl; |
2043 | if (req->inode_drop && req->inode()) | |
2044 | encode_inode_release(req->inode(), req, | |
2045 | mds, req->inode_drop, | |
2046 | req->inode_unless); | |
2047 | ||
2048 | if (req->old_inode_drop && req->old_inode()) | |
2049 | encode_inode_release(req->old_inode(), req, | |
2050 | mds, req->old_inode_drop, | |
2051 | req->old_inode_unless); | |
2052 | if (req->other_inode_drop && req->other_inode()) | |
2053 | encode_inode_release(req->other_inode(), req, | |
2054 | mds, req->other_inode_drop, | |
2055 | req->other_inode_unless); | |
2056 | ||
2057 | if (req->dentry_drop && req->dentry()) | |
2058 | encode_dentry_release(req->dentry(), req, | |
2059 | mds, req->dentry_drop, | |
2060 | req->dentry_unless); | |
2061 | ||
2062 | if (req->old_dentry_drop && req->old_dentry()) | |
2063 | encode_dentry_release(req->old_dentry(), req, | |
2064 | mds, req->old_dentry_drop, | |
2065 | req->old_dentry_unless); | |
11fdf7f2 | 2066 | ldout(cct, 25) << __func__ << " exit (req: " |
7c673cae FG |
2067 | << req << ", mds " << mds <<dendl; |
2068 | } | |
2069 | ||
2070 | bool Client::have_open_session(mds_rank_t mds) | |
2071 | { | |
11fdf7f2 TL |
2072 | const auto &it = mds_sessions.find(mds); |
2073 | return it != mds_sessions.end() && | |
20effc67 TL |
2074 | (it->second->state == MetaSession::STATE_OPEN || |
2075 | it->second->state == MetaSession::STATE_STALE); | |
7c673cae FG |
2076 | } |
2077 | ||
20effc67 | 2078 | MetaSessionRef Client::_get_mds_session(mds_rank_t mds, Connection *con) |
7c673cae | 2079 | { |
11fdf7f2 | 2080 | const auto &it = mds_sessions.find(mds); |
20effc67 | 2081 | if (it == mds_sessions.end() || it->second->con != con) { |
7c673cae | 2082 | return NULL; |
11fdf7f2 | 2083 | } else { |
20effc67 | 2084 | return it->second; |
11fdf7f2 | 2085 | } |
7c673cae FG |
2086 | } |
2087 | ||
20effc67 | 2088 | MetaSessionRef Client::_get_or_open_mds_session(mds_rank_t mds) |
7c673cae | 2089 | { |
11fdf7f2 | 2090 | auto it = mds_sessions.find(mds); |
20effc67 | 2091 | return it == mds_sessions.end() ? _open_mds_session(mds) : it->second; |
7c673cae FG |
2092 | } |
2093 | ||
2094 | /** | |
2095 | * Populate a map of strings with client-identifying metadata, | |
2096 | * such as the hostname. Call this once at initialization. | |
2097 | */ | |
2098 | void Client::populate_metadata(const std::string &mount_root) | |
2099 | { | |
2100 | // Hostname | |
f67539c2 TL |
2101 | #ifdef _WIN32 |
2102 | // TODO: move this to compat.h | |
2103 | char hostname[64]; | |
2104 | DWORD hostname_sz = 64; | |
2105 | GetComputerNameA(hostname, &hostname_sz); | |
2106 | metadata["hostname"] = hostname; | |
2107 | #else | |
7c673cae FG |
2108 | struct utsname u; |
2109 | int r = uname(&u); | |
2110 | if (r >= 0) { | |
2111 | metadata["hostname"] = u.nodename; | |
2112 | ldout(cct, 20) << __func__ << " read hostname '" << u.nodename << "'" << dendl; | |
2113 | } else { | |
2114 | ldout(cct, 1) << __func__ << " failed to read hostname (" << cpp_strerror(r) << ")" << dendl; | |
2115 | } | |
f67539c2 | 2116 | #endif |
7c673cae FG |
2117 | |
2118 | metadata["pid"] = stringify(getpid()); | |
2119 | ||
2120 | // Ceph entity id (the '0' in "client.0") | |
2121 | metadata["entity_id"] = cct->_conf->name.get_id(); | |
2122 | ||
2123 | // Our mount position | |
2124 | if (!mount_root.empty()) { | |
2125 | metadata["root"] = mount_root; | |
2126 | } | |
2127 | ||
2128 | // Ceph version | |
2129 | metadata["ceph_version"] = pretty_version_to_str(); | |
2130 | metadata["ceph_sha1"] = git_version_to_str(); | |
2131 | ||
2132 | // Apply any metadata from the user's configured overrides | |
2133 | std::vector<std::string> tokens; | |
2134 | get_str_vec(cct->_conf->client_metadata, ",", tokens); | |
2135 | for (const auto &i : tokens) { | |
2136 | auto eqpos = i.find("="); | |
2137 | // Throw out anything that isn't of the form "<str>=<str>" | |
2138 | if (eqpos == 0 || eqpos == std::string::npos || eqpos == i.size()) { | |
2139 | lderr(cct) << "Invalid metadata keyval pair: '" << i << "'" << dendl; | |
2140 | continue; | |
2141 | } | |
2142 | metadata[i.substr(0, eqpos)] = i.substr(eqpos + 1); | |
2143 | } | |
2144 | } | |
2145 | ||
2146 | /** | |
2147 | * Optionally add or override client metadata fields. | |
2148 | */ | |
2149 | void Client::update_metadata(std::string const &k, std::string const &v) | |
2150 | { | |
f67539c2 TL |
2151 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2152 | ceph_assert(iref_reader.is_state_satisfied()); | |
2153 | ||
2154 | std::scoped_lock l(client_lock); | |
7c673cae | 2155 | |
11fdf7f2 TL |
2156 | auto it = metadata.find(k); |
2157 | if (it != metadata.end()) { | |
7c673cae | 2158 | ldout(cct, 1) << __func__ << " warning, overriding metadata field '" << k |
11fdf7f2 | 2159 | << "' from '" << it->second << "' to '" << v << "'" << dendl; |
7c673cae FG |
2160 | } |
2161 | ||
2162 | metadata[k] = v; | |
2163 | } | |
2164 | ||
20effc67 | 2165 | MetaSessionRef Client::_open_mds_session(mds_rank_t mds) |
7c673cae | 2166 | { |
11fdf7f2 TL |
2167 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
2168 | auto addrs = mdsmap->get_addrs(mds); | |
2169 | auto em = mds_sessions.emplace(std::piecewise_construct, | |
2170 | std::forward_as_tuple(mds), | |
20effc67 | 2171 | std::forward_as_tuple(new MetaSession(mds, messenger->connect_to_mds(addrs), addrs))); |
11fdf7f2 | 2172 | ceph_assert(em.second); /* not already present */ |
20effc67 | 2173 | auto session = em.first->second; |
7c673cae | 2174 | |
9f95a23c | 2175 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_OPEN); |
11fdf7f2 TL |
2176 | m->metadata = metadata; |
2177 | m->supported_features = feature_bitset_t(CEPHFS_FEATURES_CLIENT_SUPPORTED); | |
f67539c2 | 2178 | m->metric_spec = feature_bitset_t(CEPHFS_METRIC_FEATURES_ALL); |
11fdf7f2 | 2179 | session->con->send_message2(std::move(m)); |
7c673cae FG |
2180 | return session; |
2181 | } | |
2182 | ||
2183 | void Client::_close_mds_session(MetaSession *s) | |
2184 | { | |
11fdf7f2 | 2185 | ldout(cct, 2) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
7c673cae | 2186 | s->state = MetaSession::STATE_CLOSING; |
9f95a23c | 2187 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
2188 | } |
2189 | ||
f6b5b4d7 | 2190 | void Client::_closed_mds_session(MetaSession *s, int err, bool rejected) |
7c673cae | 2191 | { |
11fdf7f2 | 2192 | ldout(cct, 5) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
f6b5b4d7 TL |
2193 | if (rejected && s->state != MetaSession::STATE_CLOSING) |
2194 | s->state = MetaSession::STATE_REJECTED; | |
2195 | else | |
2196 | s->state = MetaSession::STATE_CLOSED; | |
7c673cae FG |
2197 | s->con->mark_down(); |
2198 | signal_context_list(s->waiting_for_open); | |
9f95a23c | 2199 | mount_cond.notify_all(); |
f6b5b4d7 | 2200 | remove_session_caps(s, err); |
7c673cae | 2201 | kick_requests_closed(s); |
f6b5b4d7 TL |
2202 | mds_ranks_closing.erase(s->mds_num); |
2203 | if (s->state == MetaSession::STATE_CLOSED) | |
2204 | mds_sessions.erase(s->mds_num); | |
7c673cae FG |
2205 | } |
2206 | ||
11fdf7f2 | 2207 | void Client::handle_client_session(const MConstRef<MClientSession>& m) |
7c673cae FG |
2208 | { |
2209 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
11fdf7f2 | 2210 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << from << dendl; |
7c673cae | 2211 | |
f67539c2 | 2212 | std::scoped_lock cl(client_lock); |
20effc67 | 2213 | auto session = _get_mds_session(from, m->get_connection().get()); |
7c673cae FG |
2214 | if (!session) { |
2215 | ldout(cct, 10) << " discarding session message from sessionless mds " << m->get_source_inst() << dendl; | |
7c673cae FG |
2216 | return; |
2217 | } | |
2218 | ||
2219 | switch (m->get_op()) { | |
2220 | case CEPH_SESSION_OPEN: | |
11fdf7f2 TL |
2221 | { |
2222 | feature_bitset_t missing_features(CEPHFS_FEATURES_CLIENT_REQUIRED); | |
2223 | missing_features -= m->supported_features; | |
2224 | if (!missing_features.empty()) { | |
2225 | lderr(cct) << "mds." << from << " lacks required features '" | |
2226 | << missing_features << "', closing session " << dendl; | |
20effc67 TL |
2227 | _close_mds_session(session.get()); |
2228 | _closed_mds_session(session.get(), -CEPHFS_EPERM, true); | |
11fdf7f2 TL |
2229 | break; |
2230 | } | |
2231 | session->mds_features = std::move(m->supported_features); | |
33c7a0ef | 2232 | session->mds_metric_flags = std::move(m->metric_spec.metric_flags); |
11fdf7f2 | 2233 | |
20effc67 | 2234 | renew_caps(session.get()); |
11fdf7f2 | 2235 | session->state = MetaSession::STATE_OPEN; |
f67539c2 | 2236 | if (is_unmounting()) |
9f95a23c | 2237 | mount_cond.notify_all(); |
11fdf7f2 TL |
2238 | else |
2239 | connect_mds_targets(from); | |
2240 | signal_context_list(session->waiting_for_open); | |
2241 | break; | |
2242 | } | |
7c673cae FG |
2243 | |
2244 | case CEPH_SESSION_CLOSE: | |
20effc67 | 2245 | _closed_mds_session(session.get()); |
7c673cae FG |
2246 | break; |
2247 | ||
2248 | case CEPH_SESSION_RENEWCAPS: | |
2249 | if (session->cap_renew_seq == m->get_seq()) { | |
a8e16298 | 2250 | bool was_stale = ceph_clock_now() >= session->cap_ttl; |
7c673cae FG |
2251 | session->cap_ttl = |
2252 | session->last_cap_renew_request + mdsmap->get_session_timeout(); | |
a8e16298 | 2253 | if (was_stale) |
20effc67 | 2254 | wake_up_session_caps(session.get(), false); |
7c673cae FG |
2255 | } |
2256 | break; | |
2257 | ||
2258 | case CEPH_SESSION_STALE: | |
28e407b8 AA |
2259 | // invalidate session caps/leases |
2260 | session->cap_gen++; | |
2261 | session->cap_ttl = ceph_clock_now(); | |
2262 | session->cap_ttl -= 1; | |
20effc67 | 2263 | renew_caps(session.get()); |
7c673cae FG |
2264 | break; |
2265 | ||
2266 | case CEPH_SESSION_RECALL_STATE: | |
f67539c2 TL |
2267 | /* |
2268 | * Call the renew caps and flush cap releases just before | |
2269 | * triming the caps in case the tick() won't get a chance | |
2270 | * to run them, which could cause the client to be blocklisted | |
2271 | * and MDS daemons trying to recall the caps again and | |
2272 | * again. | |
2273 | * | |
2274 | * In most cases it will do nothing, and the new cap releases | |
2275 | * added by trim_caps() followed will be deferred flushing | |
2276 | * by tick(). | |
2277 | */ | |
2278 | renew_and_flush_cap_releases(); | |
20effc67 | 2279 | trim_caps(session.get(), m->get_max_caps()); |
7c673cae FG |
2280 | break; |
2281 | ||
2282 | case CEPH_SESSION_FLUSHMSG: | |
a8e16298 | 2283 | /* flush cap release */ |
11fdf7f2 TL |
2284 | if (auto& m = session->release; m) { |
2285 | session->con->send_message2(std::move(m)); | |
a8e16298 | 2286 | } |
9f95a23c | 2287 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_FLUSHMSG_ACK, m->get_seq())); |
7c673cae FG |
2288 | break; |
2289 | ||
2290 | case CEPH_SESSION_FORCE_RO: | |
20effc67 | 2291 | force_session_readonly(session.get()); |
7c673cae FG |
2292 | break; |
2293 | ||
2294 | case CEPH_SESSION_REJECT: | |
11fdf7f2 TL |
2295 | { |
2296 | std::string_view error_str; | |
2297 | auto it = m->metadata.find("error_string"); | |
2298 | if (it != m->metadata.end()) | |
2299 | error_str = it->second; | |
2300 | else | |
2301 | error_str = "unknown error"; | |
2302 | lderr(cct) << "mds." << from << " rejected us (" << error_str << ")" << dendl; | |
7c673cae | 2303 | |
20effc67 | 2304 | _closed_mds_session(session.get(), -CEPHFS_EPERM, true); |
11fdf7f2 | 2305 | } |
7c673cae FG |
2306 | break; |
2307 | ||
2308 | default: | |
2309 | ceph_abort(); | |
2310 | } | |
7c673cae FG |
2311 | } |
2312 | ||
2313 | bool Client::_any_stale_sessions() const | |
2314 | { | |
9f95a23c | 2315 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae | 2316 | |
11fdf7f2 | 2317 | for (const auto &p : mds_sessions) { |
20effc67 | 2318 | if (p.second->state == MetaSession::STATE_STALE) { |
7c673cae FG |
2319 | return true; |
2320 | } | |
2321 | } | |
2322 | ||
2323 | return false; | |
2324 | } | |
2325 | ||
2326 | void Client::_kick_stale_sessions() | |
2327 | { | |
11fdf7f2 | 2328 | ldout(cct, 1) << __func__ << dendl; |
7c673cae | 2329 | |
11fdf7f2 | 2330 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
20effc67 TL |
2331 | auto s = it->second; |
2332 | if (s->state == MetaSession::STATE_REJECTED) { | |
2333 | mds_sessions.erase(it->first); | |
f6b5b4d7 TL |
2334 | continue; |
2335 | } | |
20effc67 TL |
2336 | if (s->state == MetaSession::STATE_STALE) |
2337 | _closed_mds_session(s.get()); | |
7c673cae FG |
2338 | } |
2339 | } | |
2340 | ||
2341 | void Client::send_request(MetaRequest *request, MetaSession *session, | |
2342 | bool drop_cap_releases) | |
2343 | { | |
2344 | // make the request | |
2345 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2346 | ldout(cct, 10) << __func__ << " rebuilding request " << request->get_tid() |
7c673cae | 2347 | << " for mds." << mds << dendl; |
11fdf7f2 | 2348 | auto r = build_client_request(request); |
7c673cae FG |
2349 | if (request->dentry()) { |
2350 | r->set_dentry_wanted(); | |
2351 | } | |
2352 | if (request->got_unsafe) { | |
2353 | r->set_replayed_op(); | |
2354 | if (request->target) | |
2355 | r->head.ino = request->target->ino; | |
2356 | } else { | |
2357 | encode_cap_releases(request, mds); | |
2358 | if (drop_cap_releases) // we haven't send cap reconnect yet, drop cap releases | |
2359 | request->cap_releases.clear(); | |
2360 | else | |
2361 | r->releases.swap(request->cap_releases); | |
2362 | } | |
2363 | r->set_mdsmap_epoch(mdsmap->get_epoch()); | |
2364 | if (r->head.op == CEPH_MDS_OP_SETXATTR) { | |
2365 | objecter->with_osdmap([r](const OSDMap& o) { | |
2366 | r->set_osdmap_epoch(o.get_epoch()); | |
2367 | }); | |
2368 | } | |
2369 | ||
2370 | if (request->mds == -1) { | |
2371 | request->sent_stamp = ceph_clock_now(); | |
11fdf7f2 | 2372 | ldout(cct, 20) << __func__ << " set sent_stamp to " << request->sent_stamp << dendl; |
7c673cae FG |
2373 | } |
2374 | request->mds = mds; | |
2375 | ||
2376 | Inode *in = request->inode(); | |
11fdf7f2 TL |
2377 | if (in) { |
2378 | auto it = in->caps.find(mds); | |
2379 | if (it != in->caps.end()) { | |
2380 | request->sent_on_mseq = it->second.mseq; | |
2381 | } | |
2382 | } | |
7c673cae FG |
2383 | |
2384 | session->requests.push_back(&request->item); | |
2385 | ||
11fdf7f2 TL |
2386 | ldout(cct, 10) << __func__ << " " << *r << " to mds." << mds << dendl; |
2387 | session->con->send_message2(std::move(r)); | |
7c673cae FG |
2388 | } |
2389 | ||
9f95a23c | 2390 | ref_t<MClientRequest> Client::build_client_request(MetaRequest *request) |
7c673cae | 2391 | { |
9f95a23c | 2392 | auto req = make_message<MClientRequest>(request->get_op()); |
7c673cae FG |
2393 | req->set_tid(request->tid); |
2394 | req->set_stamp(request->op_stamp); | |
2395 | memcpy(&req->head, &request->head, sizeof(ceph_mds_request_head)); | |
2396 | ||
2397 | // if the filepath's haven't been set, set them! | |
2398 | if (request->path.empty()) { | |
2399 | Inode *in = request->inode(); | |
2400 | Dentry *de = request->dentry(); | |
2401 | if (in) | |
2402 | in->make_nosnap_relative_path(request->path); | |
2403 | else if (de) { | |
2404 | if (de->inode) | |
2405 | de->inode->make_nosnap_relative_path(request->path); | |
2406 | else if (de->dir) { | |
2407 | de->dir->parent_inode->make_nosnap_relative_path(request->path); | |
2408 | request->path.push_dentry(de->name); | |
2409 | } | |
2410 | else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2411 | << " No path, inode, or appropriately-endowed dentry given!" | |
2412 | << dendl; | |
2413 | } else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2414 | << " No path, inode, or dentry given!" | |
2415 | << dendl; | |
2416 | } | |
2417 | req->set_filepath(request->get_filepath()); | |
2418 | req->set_filepath2(request->get_filepath2()); | |
f67539c2 | 2419 | req->set_alternate_name(request->alternate_name); |
7c673cae FG |
2420 | req->set_data(request->data); |
2421 | req->set_retry_attempt(request->retry_attempt++); | |
2422 | req->head.num_fwd = request->num_fwd; | |
2423 | const gid_t *_gids; | |
2424 | int gid_count = request->perms.get_gids(&_gids); | |
2425 | req->set_gid_list(gid_count, _gids); | |
2426 | return req; | |
2427 | } | |
2428 | ||
2429 | ||
2430 | ||
11fdf7f2 | 2431 | void Client::handle_client_request_forward(const MConstRef<MClientRequestForward>& fwd) |
7c673cae FG |
2432 | { |
2433 | mds_rank_t mds = mds_rank_t(fwd->get_source().num()); | |
f67539c2 TL |
2434 | |
2435 | std::scoped_lock cl(client_lock); | |
20effc67 | 2436 | auto session = _get_mds_session(mds, fwd->get_connection().get()); |
7c673cae | 2437 | if (!session) { |
7c673cae FG |
2438 | return; |
2439 | } | |
2440 | ceph_tid_t tid = fwd->get_tid(); | |
2441 | ||
2442 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2443 | ldout(cct, 10) << __func__ << " no pending request on tid " << tid << dendl; |
7c673cae FG |
2444 | return; |
2445 | } | |
2446 | ||
2447 | MetaRequest *request = mds_requests[tid]; | |
11fdf7f2 | 2448 | ceph_assert(request); |
7c673cae | 2449 | |
33c7a0ef TL |
2450 | /* |
2451 | * The type of 'num_fwd' in ceph 'MClientRequestForward' | |
2452 | * is 'int32_t', while in 'ceph_mds_request_head' the | |
2453 | * type is '__u8'. So in case the request bounces between | |
2454 | * MDSes exceeding 256 times, the client will get stuck. | |
2455 | * | |
2456 | * In this case it's ususally a bug in MDS and continue | |
2457 | * bouncing the request makes no sense. | |
2458 | * | |
2459 | * In future this could be fixed in ceph code, so avoid | |
2460 | * using the hardcode here. | |
2461 | */ | |
2462 | int max_fwd = sizeof(((struct ceph_mds_request_head*)0)->num_fwd); | |
2463 | max_fwd = 1 << (max_fwd * CHAR_BIT) - 1; | |
2464 | auto num_fwd = fwd->get_num_fwd(); | |
2465 | if (num_fwd <= request->num_fwd || num_fwd >= max_fwd) { | |
2466 | if (request->num_fwd >= max_fwd || num_fwd >= max_fwd) { | |
2467 | request->abort(-EMULTIHOP); | |
2468 | request->caller_cond->notify_all(); | |
2469 | ldout(cct, 1) << __func__ << " tid " << tid << " seq overflow" | |
2470 | << ", abort it" << dendl; | |
2471 | } else { | |
2472 | ldout(cct, 10) << __func__ << " tid " << tid | |
2473 | << " old fwd seq " << fwd->get_num_fwd() | |
2474 | << " <= req fwd " << request->num_fwd | |
2475 | << ", ignore it" << dendl; | |
2476 | } | |
2477 | return; | |
2478 | } | |
2479 | ||
7c673cae FG |
2480 | // reset retry counter |
2481 | request->retry_attempt = 0; | |
2482 | ||
2483 | // request not forwarded, or dest mds has no session. | |
2484 | // resend. | |
11fdf7f2 | 2485 | ldout(cct, 10) << __func__ << " tid " << tid |
7c673cae FG |
2486 | << " fwd " << fwd->get_num_fwd() |
2487 | << " to mds." << fwd->get_dest_mds() | |
2488 | << ", resending to " << fwd->get_dest_mds() | |
2489 | << dendl; | |
2490 | ||
2491 | request->mds = -1; | |
2492 | request->item.remove_myself(); | |
33c7a0ef | 2493 | request->num_fwd = num_fwd; |
7c673cae | 2494 | request->resend_mds = fwd->get_dest_mds(); |
9f95a23c | 2495 | request->caller_cond->notify_all(); |
7c673cae FG |
2496 | } |
2497 | ||
2498 | bool Client::is_dir_operation(MetaRequest *req) | |
2499 | { | |
2500 | int op = req->get_op(); | |
2501 | if (op == CEPH_MDS_OP_MKNOD || op == CEPH_MDS_OP_LINK || | |
2502 | op == CEPH_MDS_OP_UNLINK || op == CEPH_MDS_OP_RENAME || | |
2503 | op == CEPH_MDS_OP_MKDIR || op == CEPH_MDS_OP_RMDIR || | |
2504 | op == CEPH_MDS_OP_SYMLINK || op == CEPH_MDS_OP_CREATE) | |
2505 | return true; | |
2506 | return false; | |
2507 | } | |
2508 | ||
11fdf7f2 | 2509 | void Client::handle_client_reply(const MConstRef<MClientReply>& reply) |
7c673cae FG |
2510 | { |
2511 | mds_rank_t mds_num = mds_rank_t(reply->get_source().num()); | |
f67539c2 TL |
2512 | |
2513 | std::scoped_lock cl(client_lock); | |
20effc67 | 2514 | auto session = _get_mds_session(mds_num, reply->get_connection().get()); |
7c673cae | 2515 | if (!session) { |
7c673cae FG |
2516 | return; |
2517 | } | |
2518 | ||
2519 | ceph_tid_t tid = reply->get_tid(); | |
2520 | bool is_safe = reply->is_safe(); | |
2521 | ||
2522 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2523 | lderr(cct) << __func__ << " no pending request on tid " << tid |
7c673cae | 2524 | << " safe is:" << is_safe << dendl; |
7c673cae FG |
2525 | return; |
2526 | } | |
2527 | MetaRequest *request = mds_requests.at(tid); | |
2528 | ||
11fdf7f2 | 2529 | ldout(cct, 20) << __func__ << " got a reply. Safe:" << is_safe |
7c673cae FG |
2530 | << " tid " << tid << dendl; |
2531 | ||
2532 | if (request->got_unsafe && !is_safe) { | |
2533 | //duplicate response | |
2534 | ldout(cct, 0) << "got a duplicate reply on tid " << tid << " from mds " | |
2535 | << mds_num << " safe:" << is_safe << dendl; | |
7c673cae FG |
2536 | return; |
2537 | } | |
2538 | ||
f67539c2 | 2539 | if (-CEPHFS_ESTALE == reply->get_result()) { // see if we can get to proper MDS |
7c673cae FG |
2540 | ldout(cct, 20) << "got ESTALE on tid " << request->tid |
2541 | << " from mds." << request->mds << dendl; | |
2542 | request->send_to_auth = true; | |
2543 | request->resend_mds = choose_target_mds(request); | |
2544 | Inode *in = request->inode(); | |
11fdf7f2 | 2545 | std::map<mds_rank_t, Cap>::const_iterator it; |
7c673cae FG |
2546 | if (request->resend_mds >= 0 && |
2547 | request->resend_mds == request->mds && | |
2548 | (in == NULL || | |
11fdf7f2 TL |
2549 | (it = in->caps.find(request->resend_mds)) != in->caps.end() || |
2550 | request->sent_on_mseq == it->second.mseq)) { | |
2551 | ldout(cct, 20) << "have to return ESTALE" << dendl; | |
7c673cae | 2552 | } else { |
9f95a23c | 2553 | request->caller_cond->notify_all(); |
7c673cae FG |
2554 | return; |
2555 | } | |
7c673cae FG |
2556 | } |
2557 | ||
11fdf7f2 | 2558 | ceph_assert(!request->reply); |
7c673cae | 2559 | request->reply = reply; |
20effc67 | 2560 | insert_trace(request, session.get()); |
7c673cae FG |
2561 | |
2562 | // Handle unsafe reply | |
2563 | if (!is_safe) { | |
2564 | request->got_unsafe = true; | |
2565 | session->unsafe_requests.push_back(&request->unsafe_item); | |
2566 | if (is_dir_operation(request)) { | |
2567 | Inode *dir = request->inode(); | |
11fdf7f2 | 2568 | ceph_assert(dir); |
7c673cae FG |
2569 | dir->unsafe_ops.push_back(&request->unsafe_dir_item); |
2570 | } | |
2571 | if (request->target) { | |
2572 | InodeRef &in = request->target; | |
2573 | in->unsafe_ops.push_back(&request->unsafe_target_item); | |
2574 | } | |
2575 | } | |
2576 | ||
2577 | // Only signal the caller once (on the first reply): | |
2578 | // Either its an unsafe reply, or its a safe reply and no unsafe reply was sent. | |
2579 | if (!is_safe || !request->got_unsafe) { | |
9f95a23c | 2580 | ceph::condition_variable cond; |
7c673cae FG |
2581 | request->dispatch_cond = &cond; |
2582 | ||
2583 | // wake up waiter | |
11fdf7f2 | 2584 | ldout(cct, 20) << __func__ << " signalling caller " << (void*)request->caller_cond << dendl; |
9f95a23c | 2585 | request->caller_cond->notify_all(); |
7c673cae FG |
2586 | |
2587 | // wake for kick back | |
9f95a23c TL |
2588 | std::unique_lock l{client_lock, std::adopt_lock}; |
2589 | cond.wait(l, [tid, request, &cond, this] { | |
2590 | if (request->dispatch_cond) { | |
2591 | ldout(cct, 20) << "handle_client_reply awaiting kickback on tid " | |
2592 | << tid << " " << &cond << dendl; | |
2593 | } | |
2594 | return !request->dispatch_cond; | |
2595 | }); | |
2596 | l.release(); | |
7c673cae FG |
2597 | } |
2598 | ||
2599 | if (is_safe) { | |
2600 | // the filesystem change is committed to disk | |
2601 | // we're done, clean up | |
2602 | if (request->got_unsafe) { | |
2603 | request->unsafe_item.remove_myself(); | |
2604 | request->unsafe_dir_item.remove_myself(); | |
2605 | request->unsafe_target_item.remove_myself(); | |
2606 | signal_cond_list(request->waitfor_safe); | |
2607 | } | |
2608 | request->item.remove_myself(); | |
2609 | unregister_request(request); | |
2610 | } | |
f67539c2 | 2611 | if (is_unmounting()) |
9f95a23c | 2612 | mount_cond.notify_all(); |
7c673cae FG |
2613 | } |
2614 | ||
2615 | void Client::_handle_full_flag(int64_t pool) | |
2616 | { | |
2617 | ldout(cct, 1) << __func__ << ": FULL: cancelling outstanding operations " | |
2618 | << "on " << pool << dendl; | |
f67539c2 | 2619 | // Cancel all outstanding ops in this pool with -CEPHFS_ENOSPC: it is necessary |
7c673cae FG |
2620 | // to do this rather than blocking, because otherwise when we fill up we |
2621 | // potentially lock caps forever on files with dirty pages, and we need | |
2622 | // to be able to release those caps to the MDS so that it can delete files | |
2623 | // and free up space. | |
f67539c2 | 2624 | epoch_t cancelled_epoch = objecter->op_cancel_writes(-CEPHFS_ENOSPC, pool); |
7c673cae FG |
2625 | |
2626 | // For all inodes with layouts in this pool and a pending flush write op | |
2627 | // (i.e. one of the ones we will cancel), we've got to purge_set their data | |
2628 | // from ObjectCacher so that it doesn't re-issue the write in response to | |
2629 | // the ENOSPC error. | |
2630 | // Fortunately since we're cancelling everything in a given pool, we don't | |
2631 | // need to know which ops belong to which ObjectSet, we can just blow all | |
2632 | // the un-flushed cached data away and mark any dirty inodes' async_err | |
f67539c2 | 2633 | // field with -CEPHFS_ENOSPC as long as we're sure all the ops we cancelled were |
7c673cae FG |
2634 | // affecting this pool, and all the objectsets we're purging were also |
2635 | // in this pool. | |
2636 | for (unordered_map<vinodeno_t,Inode*>::iterator i = inode_map.begin(); | |
2637 | i != inode_map.end(); ++i) | |
2638 | { | |
2639 | Inode *inode = i->second; | |
2640 | if (inode->oset.dirty_or_tx | |
2641 | && (pool == -1 || inode->layout.pool_id == pool)) { | |
2642 | ldout(cct, 4) << __func__ << ": FULL: inode 0x" << std::hex << i->first << std::dec | |
2643 | << " has dirty objects, purging and setting ENOSPC" << dendl; | |
2644 | objectcacher->purge_set(&inode->oset); | |
f67539c2 | 2645 | inode->set_async_err(-CEPHFS_ENOSPC); |
7c673cae FG |
2646 | } |
2647 | } | |
2648 | ||
2649 | if (cancelled_epoch != (epoch_t)-1) { | |
2650 | set_cap_epoch_barrier(cancelled_epoch); | |
2651 | } | |
2652 | } | |
2653 | ||
11fdf7f2 | 2654 | void Client::handle_osd_map(const MConstRef<MOSDMap>& m) |
7c673cae | 2655 | { |
f67539c2 | 2656 | std::scoped_lock cl(client_lock); |
31f18b77 | 2657 | |
11fdf7f2 | 2658 | const auto myaddrs = messenger->get_myaddrs(); |
33c7a0ef | 2659 | bool new_blocklist = objecter->with_osdmap( |
11fdf7f2 | 2660 | [&](const OSDMap& o) { |
33c7a0ef | 2661 | return o.is_blocklisted(myaddrs); |
11fdf7f2 | 2662 | }); |
33c7a0ef TL |
2663 | |
2664 | if (new_blocklist && !blocklisted) { | |
31f18b77 FG |
2665 | auto epoch = objecter->with_osdmap([](const OSDMap &o){ |
2666 | return o.get_epoch(); | |
2667 | }); | |
f67539c2 TL |
2668 | lderr(cct) << "I was blocklisted at osd epoch " << epoch << dendl; |
2669 | blocklisted = true; | |
31f18b77 | 2670 | |
f67539c2 | 2671 | _abort_mds_sessions(-CEPHFS_EBLOCKLISTED); |
31f18b77 FG |
2672 | |
2673 | // Since we know all our OSD ops will fail, cancel them all preemtively, | |
2674 | // so that on an unhealthy cluster we can umount promptly even if e.g. | |
2675 | // some PGs were inaccessible. | |
f67539c2 TL |
2676 | objecter->op_cancel_writes(-CEPHFS_EBLOCKLISTED); |
2677 | ||
2678 | } | |
31f18b77 | 2679 | |
f67539c2 TL |
2680 | if (blocklisted) { |
2681 | // Handle case where we were blocklisted but no longer are | |
2682 | blocklisted = objecter->with_osdmap([myaddrs](const OSDMap &o){ | |
2683 | return o.is_blocklisted(myaddrs);}); | |
31f18b77 FG |
2684 | } |
2685 | ||
f67539c2 TL |
2686 | // Always subscribe to next osdmap for blocklisted client |
2687 | // until this client is not blocklisted. | |
2688 | if (blocklisted) { | |
f64942e4 AA |
2689 | objecter->maybe_request_map(); |
2690 | } | |
2691 | ||
7c673cae FG |
2692 | if (objecter->osdmap_full_flag()) { |
2693 | _handle_full_flag(-1); | |
2694 | } else { | |
2695 | // Accumulate local list of full pools so that I can drop | |
2696 | // the objecter lock before re-entering objecter in | |
2697 | // cancel_writes | |
2698 | std::vector<int64_t> full_pools; | |
2699 | ||
2700 | objecter->with_osdmap([&full_pools](const OSDMap &o) { | |
2701 | for (const auto& kv : o.get_pools()) { | |
2702 | if (kv.second.has_flag(pg_pool_t::FLAG_FULL)) { | |
2703 | full_pools.push_back(kv.first); | |
2704 | } | |
2705 | } | |
2706 | }); | |
2707 | ||
2708 | for (auto p : full_pools) | |
2709 | _handle_full_flag(p); | |
2710 | ||
2711 | // Subscribe to subsequent maps to watch for the full flag going | |
2712 | // away. For the global full flag objecter does this for us, but | |
2713 | // it pays no attention to the per-pool full flag so in this branch | |
2714 | // we do it ourselves. | |
2715 | if (!full_pools.empty()) { | |
2716 | objecter->maybe_request_map(); | |
2717 | } | |
2718 | } | |
7c673cae FG |
2719 | } |
2720 | ||
2721 | ||
2722 | // ------------------------ | |
2723 | // incoming messages | |
2724 | ||
2725 | ||
11fdf7f2 | 2726 | bool Client::ms_dispatch2(const MessageRef &m) |
7c673cae | 2727 | { |
f67539c2 TL |
2728 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2729 | if (!iref_reader.is_state_satisfied()) { | |
7c673cae | 2730 | ldout(cct, 10) << "inactive, discarding " << *m << dendl; |
7c673cae FG |
2731 | return true; |
2732 | } | |
2733 | ||
2734 | switch (m->get_type()) { | |
2735 | // mounting and mds sessions | |
2736 | case CEPH_MSG_MDS_MAP: | |
9f95a23c | 2737 | handle_mds_map(ref_cast<MMDSMap>(m)); |
7c673cae FG |
2738 | break; |
2739 | case CEPH_MSG_FS_MAP: | |
9f95a23c | 2740 | handle_fs_map(ref_cast<MFSMap>(m)); |
7c673cae FG |
2741 | break; |
2742 | case CEPH_MSG_FS_MAP_USER: | |
9f95a23c | 2743 | handle_fs_map_user(ref_cast<MFSMapUser>(m)); |
7c673cae FG |
2744 | break; |
2745 | case CEPH_MSG_CLIENT_SESSION: | |
9f95a23c | 2746 | handle_client_session(ref_cast<MClientSession>(m)); |
7c673cae FG |
2747 | break; |
2748 | ||
2749 | case CEPH_MSG_OSD_MAP: | |
9f95a23c | 2750 | handle_osd_map(ref_cast<MOSDMap>(m)); |
7c673cae FG |
2751 | break; |
2752 | ||
2753 | // requests | |
2754 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: | |
9f95a23c | 2755 | handle_client_request_forward(ref_cast<MClientRequestForward>(m)); |
7c673cae FG |
2756 | break; |
2757 | case CEPH_MSG_CLIENT_REPLY: | |
9f95a23c | 2758 | handle_client_reply(ref_cast<MClientReply>(m)); |
11fdf7f2 TL |
2759 | break; |
2760 | ||
2761 | // reclaim reply | |
2762 | case CEPH_MSG_CLIENT_RECLAIM_REPLY: | |
9f95a23c | 2763 | handle_client_reclaim_reply(ref_cast<MClientReclaimReply>(m)); |
7c673cae FG |
2764 | break; |
2765 | ||
2766 | case CEPH_MSG_CLIENT_SNAP: | |
9f95a23c | 2767 | handle_snap(ref_cast<MClientSnap>(m)); |
7c673cae FG |
2768 | break; |
2769 | case CEPH_MSG_CLIENT_CAPS: | |
9f95a23c | 2770 | handle_caps(ref_cast<MClientCaps>(m)); |
7c673cae FG |
2771 | break; |
2772 | case CEPH_MSG_CLIENT_LEASE: | |
9f95a23c | 2773 | handle_lease(ref_cast<MClientLease>(m)); |
7c673cae FG |
2774 | break; |
2775 | case MSG_COMMAND_REPLY: | |
2776 | if (m->get_source().type() == CEPH_ENTITY_TYPE_MDS) { | |
9f95a23c | 2777 | handle_command_reply(ref_cast<MCommandReply>(m)); |
7c673cae FG |
2778 | } else { |
2779 | return false; | |
2780 | } | |
2781 | break; | |
2782 | case CEPH_MSG_CLIENT_QUOTA: | |
9f95a23c | 2783 | handle_quota(ref_cast<MClientQuota>(m)); |
7c673cae FG |
2784 | break; |
2785 | ||
2786 | default: | |
2787 | return false; | |
2788 | } | |
2789 | ||
2790 | // unmounting? | |
f67539c2 TL |
2791 | std::scoped_lock cl(client_lock); |
2792 | if (is_unmounting()) { | |
7c673cae FG |
2793 | ldout(cct, 10) << "unmounting: trim pass, size was " << lru.lru_get_size() |
2794 | << "+" << inode_map.size() << dendl; | |
f67539c2 | 2795 | uint64_t size = lru.lru_get_size() + inode_map.size(); |
7c673cae | 2796 | trim_cache(); |
f67539c2 | 2797 | if (size > lru.lru_get_size() + inode_map.size()) { |
7c673cae | 2798 | ldout(cct, 10) << "unmounting: trim pass, cache shrank, poking unmount()" << dendl; |
9f95a23c | 2799 | mount_cond.notify_all(); |
7c673cae FG |
2800 | } else { |
2801 | ldout(cct, 10) << "unmounting: trim pass, size still " << lru.lru_get_size() | |
2802 | << "+" << inode_map.size() << dendl; | |
2803 | } | |
2804 | } | |
2805 | ||
2806 | return true; | |
2807 | } | |
2808 | ||
11fdf7f2 | 2809 | void Client::handle_fs_map(const MConstRef<MFSMap>& m) |
7c673cae | 2810 | { |
f67539c2 | 2811 | std::scoped_lock cl(client_lock); |
7c673cae | 2812 | fsmap.reset(new FSMap(m->get_fsmap())); |
7c673cae FG |
2813 | |
2814 | signal_cond_list(waiting_for_fsmap); | |
2815 | ||
2816 | monclient->sub_got("fsmap", fsmap->get_epoch()); | |
2817 | } | |
2818 | ||
11fdf7f2 | 2819 | void Client::handle_fs_map_user(const MConstRef<MFSMapUser>& m) |
7c673cae | 2820 | { |
f67539c2 | 2821 | std::scoped_lock cl(client_lock); |
7c673cae FG |
2822 | fsmap_user.reset(new FSMapUser); |
2823 | *fsmap_user = m->get_fsmap(); | |
7c673cae FG |
2824 | |
2825 | monclient->sub_got("fsmap.user", fsmap_user->get_epoch()); | |
2826 | signal_cond_list(waiting_for_fsmap); | |
2827 | } | |
2828 | ||
f67539c2 TL |
2829 | // Cancel all the commands for missing or laggy GIDs |
2830 | void Client::cancel_commands(const MDSMap& newmap) | |
7c673cae | 2831 | { |
f67539c2 | 2832 | std::vector<ceph_tid_t> cancel_ops; |
7c673cae | 2833 | |
f67539c2 | 2834 | std::scoped_lock cmd_lock(command_lock); |
7c673cae | 2835 | auto &commands = command_table.get_commands(); |
f67539c2 | 2836 | for (const auto &[tid, op] : commands) { |
7c673cae | 2837 | const mds_gid_t op_mds_gid = op.mds_gid; |
f67539c2 TL |
2838 | if (newmap.is_dne_gid(op_mds_gid) || newmap.is_laggy_gid(op_mds_gid)) { |
2839 | ldout(cct, 1) << __func__ << ": cancelling command op " << tid << dendl; | |
2840 | cancel_ops.push_back(tid); | |
7c673cae FG |
2841 | if (op.outs) { |
2842 | std::ostringstream ss; | |
2843 | ss << "MDS " << op_mds_gid << " went away"; | |
2844 | *(op.outs) = ss.str(); | |
2845 | } | |
f67539c2 TL |
2846 | /* |
2847 | * No need to make the con->mark_down under | |
2848 | * client_lock here, because the con will | |
2849 | * has its own lock. | |
2850 | */ | |
7c673cae | 2851 | op.con->mark_down(); |
f67539c2 TL |
2852 | if (op.on_finish) |
2853 | op.on_finish->complete(-CEPHFS_ETIMEDOUT); | |
7c673cae FG |
2854 | } |
2855 | } | |
2856 | ||
f67539c2 TL |
2857 | for (const auto &tid : cancel_ops) |
2858 | command_table.erase(tid); | |
2859 | } | |
2860 | ||
2861 | void Client::handle_mds_map(const MConstRef<MMDSMap>& m) | |
2862 | { | |
2863 | std::unique_lock cl(client_lock); | |
2864 | if (m->get_epoch() <= mdsmap->get_epoch()) { | |
2865 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() | |
2866 | << " is identical to or older than our " | |
2867 | << mdsmap->get_epoch() << dendl; | |
2868 | return; | |
7c673cae FG |
2869 | } |
2870 | ||
f67539c2 TL |
2871 | cl.unlock(); |
2872 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() << dendl; | |
2873 | std::unique_ptr<MDSMap> _mdsmap(new MDSMap); | |
2874 | _mdsmap->decode(m->get_encoded()); | |
2875 | cancel_commands(*_mdsmap.get()); | |
2876 | cl.lock(); | |
2877 | ||
2878 | _mdsmap.swap(mdsmap); | |
2879 | ||
7c673cae | 2880 | // reset session |
11fdf7f2 | 2881 | for (auto p = mds_sessions.begin(); p != mds_sessions.end(); ) { |
7c673cae | 2882 | mds_rank_t mds = p->first; |
20effc67 | 2883 | MetaSessionRef session = p->second; |
7c673cae FG |
2884 | ++p; |
2885 | ||
f67539c2 | 2886 | int oldstate = _mdsmap->get_state(mds); |
7c673cae FG |
2887 | int newstate = mdsmap->get_state(mds); |
2888 | if (!mdsmap->is_up(mds)) { | |
2889 | session->con->mark_down(); | |
11fdf7f2 | 2890 | } else if (mdsmap->get_addrs(mds) != session->addrs) { |
f67539c2 TL |
2891 | auto old_inc = _mdsmap->get_incarnation(mds); |
2892 | auto new_inc = mdsmap->get_incarnation(mds); | |
f64942e4 AA |
2893 | if (old_inc != new_inc) { |
2894 | ldout(cct, 1) << "mds incarnation changed from " | |
2895 | << old_inc << " to " << new_inc << dendl; | |
2896 | oldstate = MDSMap::STATE_NULL; | |
2897 | } | |
7c673cae | 2898 | session->con->mark_down(); |
11fdf7f2 | 2899 | session->addrs = mdsmap->get_addrs(mds); |
7c673cae FG |
2900 | // When new MDS starts to take over, notify kernel to trim unused entries |
2901 | // in its dcache/icache. Hopefully, the kernel will release some unused | |
2902 | // inodes before the new MDS enters reconnect state. | |
20effc67 | 2903 | trim_cache_for_reconnect(session.get()); |
7c673cae FG |
2904 | } else if (oldstate == newstate) |
2905 | continue; // no change | |
f67539c2 | 2906 | |
7c673cae FG |
2907 | session->mds_state = newstate; |
2908 | if (newstate == MDSMap::STATE_RECONNECT) { | |
11fdf7f2 | 2909 | session->con = messenger->connect_to_mds(session->addrs); |
20effc67 | 2910 | send_reconnect(session.get()); |
81eedcae TL |
2911 | } else if (newstate > MDSMap::STATE_RECONNECT) { |
2912 | if (oldstate < MDSMap::STATE_RECONNECT) { | |
2913 | ldout(cct, 1) << "we may miss the MDSMap::RECONNECT, close mds session ... " << dendl; | |
20effc67 | 2914 | _closed_mds_session(session.get()); |
81eedcae TL |
2915 | continue; |
2916 | } | |
2917 | if (newstate >= MDSMap::STATE_ACTIVE) { | |
2918 | if (oldstate < MDSMap::STATE_ACTIVE) { | |
2919 | // kick new requests | |
20effc67 TL |
2920 | kick_requests(session.get()); |
2921 | kick_flushing_caps(session.get()); | |
81eedcae | 2922 | signal_context_list(session->waiting_for_open); |
20effc67 | 2923 | wake_up_session_caps(session.get(), true); |
81eedcae TL |
2924 | } |
2925 | connect_mds_targets(mds); | |
7c673cae | 2926 | } |
7c673cae FG |
2927 | } else if (newstate == MDSMap::STATE_NULL && |
2928 | mds >= mdsmap->get_max_mds()) { | |
20effc67 | 2929 | _closed_mds_session(session.get()); |
7c673cae FG |
2930 | } |
2931 | } | |
2932 | ||
2933 | // kick any waiting threads | |
2934 | signal_cond_list(waiting_for_mdsmap); | |
2935 | ||
7c673cae FG |
2936 | monclient->sub_got("mdsmap", mdsmap->get_epoch()); |
2937 | } | |
2938 | ||
2939 | void Client::send_reconnect(MetaSession *session) | |
2940 | { | |
2941 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2942 | ldout(cct, 10) << __func__ << " to mds." << mds << dendl; |
7c673cae FG |
2943 | |
2944 | // trim unused caps to reduce MDS's cache rejoin time | |
2945 | trim_cache_for_reconnect(session); | |
2946 | ||
2947 | session->readonly = false; | |
2948 | ||
11fdf7f2 | 2949 | session->release.reset(); |
7c673cae FG |
2950 | |
2951 | // reset my cap seq number | |
2952 | session->seq = 0; | |
2953 | //connect to the mds' offload targets | |
2954 | connect_mds_targets(mds); | |
2955 | //make sure unsafe requests get saved | |
2956 | resend_unsafe_requests(session); | |
2957 | ||
11fdf7f2 TL |
2958 | early_kick_flushing_caps(session); |
2959 | ||
9f95a23c | 2960 | auto m = make_message<MClientReconnect>(); |
11fdf7f2 | 2961 | bool allow_multi = session->mds_features.test(CEPHFS_FEATURE_MULTI_RECONNECT); |
7c673cae FG |
2962 | |
2963 | // i have an open session. | |
2964 | ceph::unordered_set<inodeno_t> did_snaprealm; | |
2965 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator p = inode_map.begin(); | |
2966 | p != inode_map.end(); | |
2967 | ++p) { | |
2968 | Inode *in = p->second; | |
11fdf7f2 TL |
2969 | auto it = in->caps.find(mds); |
2970 | if (it != in->caps.end()) { | |
2971 | if (allow_multi && | |
9f95a23c TL |
2972 | m->get_approx_size() >= |
2973 | static_cast<size_t>((std::numeric_limits<int>::max() >> 1))) { | |
11fdf7f2 TL |
2974 | m->mark_more(); |
2975 | session->con->send_message2(std::move(m)); | |
2976 | ||
9f95a23c | 2977 | m = make_message<MClientReconnect>(); |
11fdf7f2 TL |
2978 | } |
2979 | ||
2980 | Cap &cap = it->second; | |
7c673cae | 2981 | ldout(cct, 10) << " caps on " << p->first |
11fdf7f2 | 2982 | << " " << ccap_string(cap.issued) |
7c673cae FG |
2983 | << " wants " << ccap_string(in->caps_wanted()) |
2984 | << dendl; | |
2985 | filepath path; | |
f91f0fd5 | 2986 | in->make_short_path(path); |
7c673cae FG |
2987 | ldout(cct, 10) << " path " << path << dendl; |
2988 | ||
2989 | bufferlist flockbl; | |
2990 | _encode_filelocks(in, flockbl); | |
2991 | ||
11fdf7f2 TL |
2992 | cap.seq = 0; // reset seq. |
2993 | cap.issue_seq = 0; // reset seq. | |
2994 | cap.mseq = 0; // reset seq. | |
2995 | // cap gen should catch up with session cap_gen | |
2996 | if (cap.gen < session->cap_gen) { | |
2997 | cap.gen = session->cap_gen; | |
2998 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
2999 | } else { | |
3000 | cap.issued = cap.implemented; | |
3001 | } | |
7c673cae FG |
3002 | snapid_t snap_follows = 0; |
3003 | if (!in->cap_snaps.empty()) | |
3004 | snap_follows = in->cap_snaps.begin()->first; | |
3005 | ||
3006 | m->add_cap(p->first.ino, | |
11fdf7f2 | 3007 | cap.cap_id, |
7c673cae FG |
3008 | path.get_ino(), path.get_path(), // ino |
3009 | in->caps_wanted(), // wanted | |
11fdf7f2 | 3010 | cap.issued, // issued |
7c673cae FG |
3011 | in->snaprealm->ino, |
3012 | snap_follows, | |
3013 | flockbl); | |
3014 | ||
3015 | if (did_snaprealm.count(in->snaprealm->ino) == 0) { | |
3016 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
3017 | m->add_snaprealm(in->snaprealm->ino, in->snaprealm->seq, in->snaprealm->parent); | |
3018 | did_snaprealm.insert(in->snaprealm->ino); | |
3019 | } | |
3020 | } | |
3021 | } | |
3022 | ||
11fdf7f2 TL |
3023 | if (!allow_multi) |
3024 | m->set_encoding_version(0); // use connection features to choose encoding | |
3025 | session->con->send_message2(std::move(m)); | |
7c673cae | 3026 | |
9f95a23c | 3027 | mount_cond.notify_all(); |
11fdf7f2 TL |
3028 | |
3029 | if (session->reclaim_state == MetaSession::RECLAIMING) | |
3030 | signal_cond_list(waiting_for_reclaim); | |
7c673cae FG |
3031 | } |
3032 | ||
3033 | ||
3034 | void Client::kick_requests(MetaSession *session) | |
3035 | { | |
11fdf7f2 | 3036 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3037 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3038 | p != mds_requests.end(); | |
3039 | ++p) { | |
31f18b77 FG |
3040 | MetaRequest *req = p->second; |
3041 | if (req->got_unsafe) | |
3042 | continue; | |
3043 | if (req->aborted()) { | |
3044 | if (req->caller_cond) { | |
3045 | req->kick = true; | |
9f95a23c | 3046 | req->caller_cond->notify_all(); |
31f18b77 | 3047 | } |
7c673cae | 3048 | continue; |
31f18b77 FG |
3049 | } |
3050 | if (req->retry_attempt > 0) | |
7c673cae | 3051 | continue; // new requests only |
31f18b77 | 3052 | if (req->mds == session->mds_num) { |
7c673cae FG |
3053 | send_request(p->second, session); |
3054 | } | |
3055 | } | |
3056 | } | |
3057 | ||
3058 | void Client::resend_unsafe_requests(MetaSession *session) | |
3059 | { | |
3060 | for (xlist<MetaRequest*>::iterator iter = session->unsafe_requests.begin(); | |
3061 | !iter.end(); | |
3062 | ++iter) | |
3063 | send_request(*iter, session); | |
3064 | ||
3065 | // also re-send old requests when MDS enters reconnect stage. So that MDS can | |
3066 | // process completed requests in clientreplay stage. | |
3067 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
3068 | p != mds_requests.end(); | |
3069 | ++p) { | |
3070 | MetaRequest *req = p->second; | |
3071 | if (req->got_unsafe) | |
3072 | continue; | |
31f18b77 FG |
3073 | if (req->aborted()) |
3074 | continue; | |
7c673cae FG |
3075 | if (req->retry_attempt == 0) |
3076 | continue; // old requests only | |
3077 | if (req->mds == session->mds_num) | |
3078 | send_request(req, session, true); | |
3079 | } | |
3080 | } | |
3081 | ||
3082 | void Client::wait_unsafe_requests() | |
3083 | { | |
3084 | list<MetaRequest*> last_unsafe_reqs; | |
11fdf7f2 | 3085 | for (const auto &p : mds_sessions) { |
20effc67 TL |
3086 | const auto s = p.second; |
3087 | if (!s->unsafe_requests.empty()) { | |
3088 | MetaRequest *req = s->unsafe_requests.back(); | |
7c673cae FG |
3089 | req->get(); |
3090 | last_unsafe_reqs.push_back(req); | |
3091 | } | |
3092 | } | |
3093 | ||
3094 | for (list<MetaRequest*>::iterator p = last_unsafe_reqs.begin(); | |
3095 | p != last_unsafe_reqs.end(); | |
3096 | ++p) { | |
3097 | MetaRequest *req = *p; | |
3098 | if (req->unsafe_item.is_on_list()) | |
3099 | wait_on_list(req->waitfor_safe); | |
3100 | put_request(req); | |
3101 | } | |
3102 | } | |
3103 | ||
3104 | void Client::kick_requests_closed(MetaSession *session) | |
3105 | { | |
11fdf7f2 | 3106 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3107 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3108 | p != mds_requests.end(); ) { | |
3109 | MetaRequest *req = p->second; | |
3110 | ++p; | |
3111 | if (req->mds == session->mds_num) { | |
3112 | if (req->caller_cond) { | |
3113 | req->kick = true; | |
9f95a23c | 3114 | req->caller_cond->notify_all(); |
7c673cae FG |
3115 | } |
3116 | req->item.remove_myself(); | |
3117 | if (req->got_unsafe) { | |
11fdf7f2 | 3118 | lderr(cct) << __func__ << " removing unsafe request " << req->get_tid() << dendl; |
7c673cae | 3119 | req->unsafe_item.remove_myself(); |
eafe8130 TL |
3120 | if (is_dir_operation(req)) { |
3121 | Inode *dir = req->inode(); | |
20effc67 | 3122 | ceph_assert(dir); |
f67539c2 | 3123 | dir->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3124 | lderr(cct) << "kick_requests_closed drop req of inode(dir) : " |
3125 | << dir->ino << " " << req->get_tid() << dendl; | |
3126 | req->unsafe_dir_item.remove_myself(); | |
3127 | } | |
3128 | if (req->target) { | |
3129 | InodeRef &in = req->target; | |
f67539c2 | 3130 | in->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3131 | lderr(cct) << "kick_requests_closed drop req of inode : " |
3132 | << in->ino << " " << req->get_tid() << dendl; | |
3133 | req->unsafe_target_item.remove_myself(); | |
3134 | } | |
7c673cae FG |
3135 | signal_cond_list(req->waitfor_safe); |
3136 | unregister_request(req); | |
3137 | } | |
3138 | } | |
3139 | } | |
11fdf7f2 TL |
3140 | ceph_assert(session->requests.empty()); |
3141 | ceph_assert(session->unsafe_requests.empty()); | |
7c673cae FG |
3142 | } |
3143 | ||
3144 | ||
3145 | ||
3146 | ||
3147 | /************ | |
3148 | * leases | |
3149 | */ | |
3150 | ||
3151 | void Client::got_mds_push(MetaSession *s) | |
3152 | { | |
3153 | s->seq++; | |
3154 | ldout(cct, 10) << " mds." << s->mds_num << " seq now " << s->seq << dendl; | |
3155 | if (s->state == MetaSession::STATE_CLOSING) { | |
9f95a23c | 3156 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
3157 | } |
3158 | } | |
3159 | ||
11fdf7f2 | 3160 | void Client::handle_lease(const MConstRef<MClientLease>& m) |
7c673cae | 3161 | { |
11fdf7f2 | 3162 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 3163 | |
11fdf7f2 | 3164 | ceph_assert(m->get_action() == CEPH_MDS_LEASE_REVOKE); |
7c673cae | 3165 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
3166 | |
3167 | std::scoped_lock cl(client_lock); | |
20effc67 | 3168 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 3169 | if (!session) { |
7c673cae FG |
3170 | return; |
3171 | } | |
3172 | ||
20effc67 | 3173 | got_mds_push(session.get()); |
7c673cae FG |
3174 | |
3175 | ceph_seq_t seq = m->get_seq(); | |
3176 | ||
3177 | Inode *in; | |
3178 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); | |
3179 | if (inode_map.count(vino) == 0) { | |
3180 | ldout(cct, 10) << " don't have vino " << vino << dendl; | |
3181 | goto revoke; | |
3182 | } | |
3183 | in = inode_map[vino]; | |
3184 | ||
9f95a23c | 3185 | if (m->get_mask() & CEPH_LEASE_VALID) { |
7c673cae FG |
3186 | if (!in->dir || in->dir->dentries.count(m->dname) == 0) { |
3187 | ldout(cct, 10) << " don't have dir|dentry " << m->get_ino() << "/" << m->dname <<dendl; | |
3188 | goto revoke; | |
3189 | } | |
3190 | Dentry *dn = in->dir->dentries[m->dname]; | |
3191 | ldout(cct, 10) << " revoked DN lease on " << dn << dendl; | |
3192 | dn->lease_mds = -1; | |
3193 | } | |
3194 | ||
3195 | revoke: | |
11fdf7f2 | 3196 | { |
9f95a23c TL |
3197 | auto reply = make_message<MClientLease>(CEPH_MDS_LEASE_RELEASE, seq, |
3198 | m->get_mask(), m->get_ino(), | |
3199 | m->get_first(), m->get_last(), m->dname); | |
11fdf7f2 TL |
3200 | m->get_connection()->send_message2(std::move(reply)); |
3201 | } | |
7c673cae FG |
3202 | } |
3203 | ||
f67539c2 | 3204 | void Client::_put_inode(Inode *in, int n) |
7c673cae | 3205 | { |
f67539c2 TL |
3206 | ldout(cct, 10) << __func__ << " on " << *in << " n = " << n << dendl; |
3207 | ||
b3b6e05e TL |
3208 | int left = in->get_nref(); |
3209 | ceph_assert(left >= n + 1); | |
3210 | in->iput(n); | |
3211 | left -= n; | |
3212 | if (left == 1) { // the last one will be held by the inode_map | |
7c673cae FG |
3213 | // release any caps |
3214 | remove_all_caps(in); | |
3215 | ||
11fdf7f2 | 3216 | ldout(cct, 10) << __func__ << " deleting " << *in << dendl; |
7c673cae | 3217 | bool unclean = objectcacher->release_set(&in->oset); |
11fdf7f2 | 3218 | ceph_assert(!unclean); |
7c673cae FG |
3219 | inode_map.erase(in->vino()); |
3220 | if (use_faked_inos()) | |
3221 | _release_faked_ino(in); | |
3222 | ||
b3b6e05e | 3223 | if (root == nullptr) { |
7c673cae FG |
3224 | root_ancestor = 0; |
3225 | while (!root_parents.empty()) | |
3226 | root_parents.erase(root_parents.begin()); | |
3227 | } | |
3228 | ||
b3b6e05e | 3229 | in->iput(); |
7c673cae FG |
3230 | } |
3231 | } | |
3232 | ||
f67539c2 TL |
3233 | void Client::delay_put_inodes(bool wakeup) |
3234 | { | |
3235 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
3236 | ||
3237 | std::map<Inode*,int> release; | |
3238 | { | |
3239 | std::scoped_lock dl(delay_i_lock); | |
3240 | release.swap(delay_i_release); | |
3241 | } | |
3242 | ||
3243 | if (release.empty()) | |
3244 | return; | |
3245 | ||
3246 | for (auto &[in, cnt] : release) | |
3247 | _put_inode(in, cnt); | |
3248 | ||
3249 | if (wakeup) | |
3250 | mount_cond.notify_all(); | |
3251 | } | |
3252 | ||
3253 | void Client::put_inode(Inode *in, int n) | |
3254 | { | |
3255 | ldout(cct, 20) << __func__ << " on " << *in << " n = " << n << dendl; | |
3256 | ||
3257 | std::scoped_lock dl(delay_i_lock); | |
3258 | delay_i_release[in] += n; | |
3259 | } | |
3260 | ||
7c673cae FG |
3261 | void Client::close_dir(Dir *dir) |
3262 | { | |
3263 | Inode *in = dir->parent_inode; | |
11fdf7f2 TL |
3264 | ldout(cct, 15) << __func__ << " dir " << dir << " on " << in << dendl; |
3265 | ceph_assert(dir->is_empty()); | |
3266 | ceph_assert(in->dir == dir); | |
3267 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked | |
3268 | if (!in->dentries.empty()) | |
7c673cae FG |
3269 | in->get_first_parent()->put(); // unpin dentry |
3270 | ||
3271 | delete in->dir; | |
3272 | in->dir = 0; | |
3273 | put_inode(in); // unpin inode | |
3274 | } | |
3275 | ||
3276 | /** | |
3277 | * Don't call this with in==NULL, use get_or_create for that | |
3278 | * leave dn set to default NULL unless you're trying to add | |
3279 | * a new inode to a pre-created Dentry | |
3280 | */ | |
3281 | Dentry* Client::link(Dir *dir, const string& name, Inode *in, Dentry *dn) | |
3282 | { | |
3283 | if (!dn) { | |
3284 | // create a new Dentry | |
11fdf7f2 TL |
3285 | dn = new Dentry(dir, name); |
3286 | ||
7c673cae FG |
3287 | lru.lru_insert_mid(dn); // mid or top? |
3288 | ||
3289 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in | |
3290 | << " dn " << dn << " (new dn)" << dendl; | |
3291 | } else { | |
11fdf7f2 | 3292 | ceph_assert(!dn->inode); |
7c673cae FG |
3293 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in |
3294 | << " dn " << dn << " (old dn)" << dendl; | |
3295 | } | |
3296 | ||
3297 | if (in) { // link to inode | |
11fdf7f2 | 3298 | InodeRef tmp_ref; |
7c673cae | 3299 | // only one parent for directories! |
11fdf7f2 TL |
3300 | if (in->is_dir() && !in->dentries.empty()) { |
3301 | tmp_ref = in; // prevent unlink below from freeing the inode. | |
7c673cae | 3302 | Dentry *olddn = in->get_first_parent(); |
11fdf7f2 | 3303 | ceph_assert(olddn->dir != dir || olddn->name != name); |
7c673cae | 3304 | Inode *old_diri = olddn->dir->parent_inode; |
7c673cae FG |
3305 | clear_dir_complete_and_ordered(old_diri, true); |
3306 | unlink(olddn, true, true); // keep dir, dentry | |
3307 | } | |
3308 | ||
11fdf7f2 | 3309 | dn->link(in); |
f67539c2 | 3310 | inc_dentry_nr(); |
11fdf7f2 | 3311 | ldout(cct, 20) << "link inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3312 | } |
3313 | ||
3314 | return dn; | |
3315 | } | |
3316 | ||
3317 | void Client::unlink(Dentry *dn, bool keepdir, bool keepdentry) | |
3318 | { | |
11fdf7f2 | 3319 | InodeRef in(dn->inode); |
7c673cae FG |
3320 | ldout(cct, 15) << "unlink dir " << dn->dir->parent_inode << " '" << dn->name << "' dn " << dn |
3321 | << " inode " << dn->inode << dendl; | |
3322 | ||
3323 | // unlink from inode | |
11fdf7f2 TL |
3324 | if (dn->inode) { |
3325 | dn->unlink(); | |
f67539c2 | 3326 | dec_dentry_nr(); |
11fdf7f2 | 3327 | ldout(cct, 20) << "unlink inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3328 | } |
3329 | ||
3330 | if (keepdentry) { | |
3331 | dn->lease_mds = -1; | |
3332 | } else { | |
3333 | ldout(cct, 15) << "unlink removing '" << dn->name << "' dn " << dn << dendl; | |
3334 | ||
3335 | // unlink from dir | |
11fdf7f2 TL |
3336 | Dir *dir = dn->dir; |
3337 | dn->detach(); | |
7c673cae FG |
3338 | |
3339 | // delete den | |
3340 | lru.lru_remove(dn); | |
3341 | dn->put(); | |
11fdf7f2 TL |
3342 | |
3343 | if (dir->is_empty() && !keepdir) | |
3344 | close_dir(dir); | |
7c673cae FG |
3345 | } |
3346 | } | |
3347 | ||
3348 | /** | |
3349 | * For asynchronous flushes, check for errors from the IO and | |
3350 | * update the inode if necessary | |
3351 | */ | |
3352 | class C_Client_FlushComplete : public Context { | |
3353 | private: | |
3354 | Client *client; | |
3355 | InodeRef inode; | |
3356 | public: | |
3357 | C_Client_FlushComplete(Client *c, Inode *in) : client(c), inode(in) { } | |
3358 | void finish(int r) override { | |
9f95a23c | 3359 | ceph_assert(ceph_mutex_is_locked_by_me(client->client_lock)); |
7c673cae FG |
3360 | if (r != 0) { |
3361 | client_t const whoami = client->whoami; // For the benefit of ldout prefix | |
3362 | ldout(client->cct, 1) << "I/O error from flush on inode " << inode | |
3363 | << " 0x" << std::hex << inode->ino << std::dec | |
3364 | << ": " << r << "(" << cpp_strerror(r) << ")" << dendl; | |
3365 | inode->set_async_err(r); | |
3366 | } | |
3367 | } | |
3368 | }; | |
3369 | ||
3370 | ||
3371 | /**** | |
3372 | * caps | |
3373 | */ | |
3374 | ||
3375 | void Client::get_cap_ref(Inode *in, int cap) | |
3376 | { | |
3377 | if ((cap & CEPH_CAP_FILE_BUFFER) && | |
3378 | in->cap_refs[CEPH_CAP_FILE_BUFFER] == 0) { | |
11fdf7f2 | 3379 | ldout(cct, 5) << __func__ << " got first FILE_BUFFER ref on " << *in << dendl; |
b3b6e05e | 3380 | in->iget(); |
7c673cae FG |
3381 | } |
3382 | if ((cap & CEPH_CAP_FILE_CACHE) && | |
3383 | in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
11fdf7f2 | 3384 | ldout(cct, 5) << __func__ << " got first FILE_CACHE ref on " << *in << dendl; |
b3b6e05e | 3385 | in->iget(); |
7c673cae FG |
3386 | } |
3387 | in->get_cap_ref(cap); | |
3388 | } | |
3389 | ||
3390 | void Client::put_cap_ref(Inode *in, int cap) | |
3391 | { | |
3392 | int last = in->put_cap_ref(cap); | |
3393 | if (last) { | |
3394 | int put_nref = 0; | |
3395 | int drop = last & ~in->caps_issued(); | |
3396 | if (in->snapid == CEPH_NOSNAP) { | |
f67539c2 | 3397 | if ((last & (CEPH_CAP_FILE_WR | CEPH_CAP_FILE_BUFFER)) && |
7c673cae FG |
3398 | !in->cap_snaps.empty() && |
3399 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3400 | ldout(cct, 10) << __func__ << " finishing pending cap_snap on " << *in << dendl; |
7c673cae FG |
3401 | in->cap_snaps.rbegin()->second.writing = 0; |
3402 | finish_cap_snap(in, in->cap_snaps.rbegin()->second, get_caps_used(in)); | |
3403 | signal_cond_list(in->waitfor_caps); // wake up blocked sync writers | |
3404 | } | |
3405 | if (last & CEPH_CAP_FILE_BUFFER) { | |
3406 | for (auto &p : in->cap_snaps) | |
3407 | p.second.dirty_data = 0; | |
3408 | signal_cond_list(in->waitfor_commit); | |
11fdf7f2 | 3409 | ldout(cct, 5) << __func__ << " dropped last FILE_BUFFER ref on " << *in << dendl; |
7c673cae FG |
3410 | ++put_nref; |
3411 | } | |
3412 | } | |
3413 | if (last & CEPH_CAP_FILE_CACHE) { | |
11fdf7f2 | 3414 | ldout(cct, 5) << __func__ << " dropped last FILE_CACHE ref on " << *in << dendl; |
7c673cae FG |
3415 | ++put_nref; |
3416 | } | |
3417 | if (drop) | |
3418 | check_caps(in, 0); | |
3419 | if (put_nref) | |
3420 | put_inode(in, put_nref); | |
3421 | } | |
3422 | } | |
3423 | ||
f67539c2 TL |
3424 | // get caps for a given file handle -- the inode should have @need caps |
3425 | // issued by the mds and @want caps not revoked (or not under revocation). | |
3426 | // this routine blocks till the cap requirement is satisfied. also account | |
3427 | // (track) for capability hit when required (when cap requirement succeedes). | |
f6b5b4d7 | 3428 | int Client::get_caps(Fh *fh, int need, int want, int *phave, loff_t endoff) |
7c673cae | 3429 | { |
f6b5b4d7 TL |
3430 | Inode *in = fh->inode.get(); |
3431 | ||
7c673cae FG |
3432 | int r = check_pool_perm(in, need); |
3433 | if (r < 0) | |
3434 | return r; | |
3435 | ||
3436 | while (1) { | |
3437 | int file_wanted = in->caps_file_wanted(); | |
3438 | if ((file_wanted & need) != need) { | |
3439 | ldout(cct, 10) << "get_caps " << *in << " need " << ccap_string(need) | |
3440 | << " file_wanted " << ccap_string(file_wanted) << ", EBADF " | |
3441 | << dendl; | |
f67539c2 | 3442 | return -CEPHFS_EBADF; |
7c673cae FG |
3443 | } |
3444 | ||
f6b5b4d7 | 3445 | if ((fh->mode & CEPH_FILE_MODE_WR) && fh->gen != fd_gen) |
f67539c2 | 3446 | return -CEPHFS_EBADF; |
f6b5b4d7 TL |
3447 | |
3448 | if ((in->flags & I_ERROR_FILELOCK) && fh->has_any_filelocks()) | |
f67539c2 | 3449 | return -CEPHFS_EIO; |
f6b5b4d7 | 3450 | |
7c673cae FG |
3451 | int implemented; |
3452 | int have = in->caps_issued(&implemented); | |
3453 | ||
3454 | bool waitfor_caps = false; | |
3455 | bool waitfor_commit = false; | |
3456 | ||
3457 | if (have & need & CEPH_CAP_FILE_WR) { | |
1911f103 TL |
3458 | if (endoff > 0) { |
3459 | if ((endoff >= (loff_t)in->max_size || | |
3460 | endoff > (loff_t)(in->size << 1)) && | |
3461 | endoff > (loff_t)in->wanted_max_size) { | |
3462 | ldout(cct, 10) << "wanted_max_size " << in->wanted_max_size << " -> " << endoff << dendl; | |
3463 | in->wanted_max_size = endoff; | |
3464 | } | |
3465 | if (in->wanted_max_size > in->max_size && | |
3466 | in->wanted_max_size > in->requested_max_size) | |
3467 | check_caps(in, 0); | |
7c673cae FG |
3468 | } |
3469 | ||
3470 | if (endoff >= 0 && endoff > (loff_t)in->max_size) { | |
3471 | ldout(cct, 10) << "waiting on max_size, endoff " << endoff << " max_size " << in->max_size << " on " << *in << dendl; | |
3472 | waitfor_caps = true; | |
3473 | } | |
3474 | if (!in->cap_snaps.empty()) { | |
3475 | if (in->cap_snaps.rbegin()->second.writing) { | |
3476 | ldout(cct, 10) << "waiting on cap_snap write to complete" << dendl; | |
3477 | waitfor_caps = true; | |
3478 | } | |
3479 | for (auto &p : in->cap_snaps) { | |
3480 | if (p.second.dirty_data) { | |
3481 | waitfor_commit = true; | |
3482 | break; | |
3483 | } | |
3484 | } | |
3485 | if (waitfor_commit) { | |
3486 | _flush(in, new C_Client_FlushComplete(this, in)); | |
3487 | ldout(cct, 10) << "waiting for WRBUFFER to get dropped" << dendl; | |
3488 | } | |
3489 | } | |
3490 | } | |
3491 | ||
3492 | if (!waitfor_caps && !waitfor_commit) { | |
3493 | if ((have & need) == need) { | |
7c673cae FG |
3494 | int revoking = implemented & ~have; |
3495 | ldout(cct, 10) << "get_caps " << *in << " have " << ccap_string(have) | |
3496 | << " need " << ccap_string(need) << " want " << ccap_string(want) | |
c07f9fc5 | 3497 | << " revoking " << ccap_string(revoking) |
7c673cae | 3498 | << dendl; |
c07f9fc5 | 3499 | if ((revoking & want) == 0) { |
7c673cae FG |
3500 | *phave = need | (have & want); |
3501 | in->get_cap_ref(need); | |
f67539c2 | 3502 | cap_hit(); |
7c673cae FG |
3503 | return 0; |
3504 | } | |
3505 | } | |
3506 | ldout(cct, 10) << "waiting for caps " << *in << " need " << ccap_string(need) << " want " << ccap_string(want) << dendl; | |
3507 | waitfor_caps = true; | |
3508 | } | |
3509 | ||
3510 | if ((need & CEPH_CAP_FILE_WR) && in->auth_cap && | |
3511 | in->auth_cap->session->readonly) | |
f67539c2 | 3512 | return -CEPHFS_EROFS; |
7c673cae FG |
3513 | |
3514 | if (in->flags & I_CAP_DROPPED) { | |
3515 | int mds_wanted = in->caps_mds_wanted(); | |
3516 | if ((mds_wanted & need) != need) { | |
3517 | int ret = _renew_caps(in); | |
3518 | if (ret < 0) | |
3519 | return ret; | |
3520 | continue; | |
3521 | } | |
a8e16298 | 3522 | if (!(file_wanted & ~mds_wanted)) |
7c673cae | 3523 | in->flags &= ~I_CAP_DROPPED; |
7c673cae FG |
3524 | } |
3525 | ||
3526 | if (waitfor_caps) | |
3527 | wait_on_list(in->waitfor_caps); | |
3528 | else if (waitfor_commit) | |
3529 | wait_on_list(in->waitfor_commit); | |
3530 | } | |
3531 | } | |
3532 | ||
3533 | int Client::get_caps_used(Inode *in) | |
3534 | { | |
3535 | unsigned used = in->caps_used(); | |
3536 | if (!(used & CEPH_CAP_FILE_CACHE) && | |
3537 | !objectcacher->set_is_empty(&in->oset)) | |
3538 | used |= CEPH_CAP_FILE_CACHE; | |
3539 | return used; | |
3540 | } | |
3541 | ||
3542 | void Client::cap_delay_requeue(Inode *in) | |
3543 | { | |
11fdf7f2 | 3544 | ldout(cct, 10) << __func__ << " on " << *in << dendl; |
7c673cae FG |
3545 | in->hold_caps_until = ceph_clock_now(); |
3546 | in->hold_caps_until += cct->_conf->client_caps_release_delay; | |
28e407b8 | 3547 | delayed_list.push_back(&in->delay_cap_item); |
7c673cae FG |
3548 | } |
3549 | ||
3550 | void Client::send_cap(Inode *in, MetaSession *session, Cap *cap, | |
eafe8130 | 3551 | int flags, int used, int want, int retain, |
7c673cae FG |
3552 | int flush, ceph_tid_t flush_tid) |
3553 | { | |
3554 | int held = cap->issued | cap->implemented; | |
3555 | int revoking = cap->implemented & ~cap->issued; | |
3556 | retain &= ~revoking; | |
3557 | int dropping = cap->issued & ~retain; | |
3558 | int op = CEPH_CAP_OP_UPDATE; | |
3559 | ||
11fdf7f2 | 3560 | ldout(cct, 10) << __func__ << " " << *in |
7c673cae | 3561 | << " mds." << session->mds_num << " seq " << cap->seq |
7c673cae FG |
3562 | << " used " << ccap_string(used) |
3563 | << " want " << ccap_string(want) | |
3564 | << " flush " << ccap_string(flush) | |
3565 | << " retain " << ccap_string(retain) | |
3566 | << " held "<< ccap_string(held) | |
3567 | << " revoking " << ccap_string(revoking) | |
3568 | << " dropping " << ccap_string(dropping) | |
3569 | << dendl; | |
3570 | ||
3571 | if (cct->_conf->client_inject_release_failure && revoking) { | |
3572 | const int would_have_issued = cap->issued & retain; | |
3573 | const int would_have_implemented = cap->implemented & (cap->issued | used); | |
3574 | // Simulated bug: | |
3575 | // - tell the server we think issued is whatever they issued plus whatever we implemented | |
3576 | // - leave what we have implemented in place | |
3577 | ldout(cct, 20) << __func__ << " injecting failure to release caps" << dendl; | |
3578 | cap->issued = cap->issued | cap->implemented; | |
3579 | ||
3580 | // Make an exception for revoking xattr caps: we are injecting | |
3581 | // failure to release other caps, but allow xattr because client | |
3582 | // will block on xattr ops if it can't release these to MDS (#9800) | |
3583 | const int xattr_mask = CEPH_CAP_XATTR_SHARED | CEPH_CAP_XATTR_EXCL; | |
3584 | cap->issued ^= xattr_mask & revoking; | |
3585 | cap->implemented ^= xattr_mask & revoking; | |
3586 | ||
3587 | ldout(cct, 20) << __func__ << " issued " << ccap_string(cap->issued) << " vs " << ccap_string(would_have_issued) << dendl; | |
3588 | ldout(cct, 20) << __func__ << " implemented " << ccap_string(cap->implemented) << " vs " << ccap_string(would_have_implemented) << dendl; | |
3589 | } else { | |
3590 | // Normal behaviour | |
3591 | cap->issued &= retain; | |
3592 | cap->implemented &= cap->issued | used; | |
3593 | } | |
3594 | ||
3595 | snapid_t follows = 0; | |
3596 | ||
3597 | if (flush) | |
3598 | follows = in->snaprealm->get_snap_context().seq; | |
20effc67 | 3599 | |
9f95a23c | 3600 | auto m = make_message<MClientCaps>(op, |
7c673cae FG |
3601 | in->ino, |
3602 | 0, | |
3603 | cap->cap_id, cap->seq, | |
3604 | cap->implemented, | |
3605 | want, | |
3606 | flush, | |
3607 | cap->mseq, | |
3608 | cap_epoch_barrier); | |
3609 | m->caller_uid = in->cap_dirtier_uid; | |
3610 | m->caller_gid = in->cap_dirtier_gid; | |
3611 | ||
3612 | m->head.issue_seq = cap->issue_seq; | |
3613 | m->set_tid(flush_tid); | |
3614 | ||
3615 | m->head.uid = in->uid; | |
3616 | m->head.gid = in->gid; | |
3617 | m->head.mode = in->mode; | |
20effc67 | 3618 | |
7c673cae | 3619 | m->head.nlink = in->nlink; |
20effc67 | 3620 | |
7c673cae | 3621 | if (flush & CEPH_CAP_XATTR_EXCL) { |
11fdf7f2 | 3622 | encode(in->xattrs, m->xattrbl); |
7c673cae FG |
3623 | m->head.xattr_version = in->xattr_version; |
3624 | } | |
20effc67 | 3625 | |
7c673cae FG |
3626 | m->size = in->size; |
3627 | m->max_size = in->max_size; | |
3628 | m->truncate_seq = in->truncate_seq; | |
3629 | m->truncate_size = in->truncate_size; | |
3630 | m->mtime = in->mtime; | |
3631 | m->atime = in->atime; | |
3632 | m->ctime = in->ctime; | |
3633 | m->btime = in->btime; | |
3634 | m->time_warp_seq = in->time_warp_seq; | |
3635 | m->change_attr = in->change_attr; | |
eafe8130 TL |
3636 | |
3637 | if (!(flags & MClientCaps::FLAG_PENDING_CAPSNAP) && | |
3638 | !in->cap_snaps.empty() && | |
3639 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3640 | flags |= MClientCaps::FLAG_PENDING_CAPSNAP; | |
3641 | m->flags = flags; | |
3642 | ||
7c673cae FG |
3643 | if (flush & CEPH_CAP_FILE_WR) { |
3644 | m->inline_version = in->inline_version; | |
3645 | m->inline_data = in->inline_data; | |
3646 | } | |
3647 | ||
3648 | in->reported_size = in->size; | |
3649 | m->set_snap_follows(follows); | |
3650 | cap->wanted = want; | |
3651 | if (cap == in->auth_cap) { | |
1911f103 TL |
3652 | if (want & CEPH_CAP_ANY_FILE_WR) { |
3653 | m->set_max_size(in->wanted_max_size); | |
3654 | in->requested_max_size = in->wanted_max_size; | |
3655 | ldout(cct, 15) << "auth cap, requesting max_size " << in->requested_max_size << dendl; | |
3656 | } else { | |
3657 | in->requested_max_size = 0; | |
3658 | ldout(cct, 15) << "auth cap, reset requested_max_size due to not wanting any file write cap" << dendl; | |
3659 | } | |
7c673cae FG |
3660 | } |
3661 | ||
3662 | if (!session->flushing_caps_tids.empty()) | |
3663 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3664 | ||
11fdf7f2 | 3665 | session->con->send_message2(std::move(m)); |
7c673cae FG |
3666 | } |
3667 | ||
31f18b77 FG |
3668 | static bool is_max_size_approaching(Inode *in) |
3669 | { | |
3670 | /* mds will adjust max size according to the reported size */ | |
3671 | if (in->flushing_caps & CEPH_CAP_FILE_WR) | |
3672 | return false; | |
3673 | if (in->size >= in->max_size) | |
3674 | return true; | |
3675 | /* half of previous max_size increment has been used */ | |
3676 | if (in->max_size > in->reported_size && | |
3677 | (in->size << 1) >= in->max_size + in->reported_size) | |
3678 | return true; | |
3679 | return false; | |
3680 | } | |
7c673cae | 3681 | |
11fdf7f2 TL |
3682 | static int adjust_caps_used_for_lazyio(int used, int issued, int implemented) |
3683 | { | |
3684 | if (!(used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER))) | |
3685 | return used; | |
3686 | if (!(implemented & CEPH_CAP_FILE_LAZYIO)) | |
3687 | return used; | |
3688 | ||
3689 | if (issued & CEPH_CAP_FILE_LAZYIO) { | |
3690 | if (!(issued & CEPH_CAP_FILE_CACHE)) { | |
3691 | used &= ~CEPH_CAP_FILE_CACHE; | |
3692 | used |= CEPH_CAP_FILE_LAZYIO; | |
3693 | } | |
3694 | if (!(issued & CEPH_CAP_FILE_BUFFER)) { | |
3695 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3696 | used |= CEPH_CAP_FILE_LAZYIO; | |
3697 | } | |
3698 | } else { | |
3699 | if (!(implemented & CEPH_CAP_FILE_CACHE)) { | |
3700 | used &= ~CEPH_CAP_FILE_CACHE; | |
3701 | used |= CEPH_CAP_FILE_LAZYIO; | |
3702 | } | |
3703 | if (!(implemented & CEPH_CAP_FILE_BUFFER)) { | |
3704 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3705 | used |= CEPH_CAP_FILE_LAZYIO; | |
3706 | } | |
3707 | } | |
3708 | return used; | |
3709 | } | |
3710 | ||
7c673cae FG |
3711 | /** |
3712 | * check_caps | |
3713 | * | |
3714 | * Examine currently used and wanted versus held caps. Release, flush or ack | |
3715 | * revoked caps to the MDS as appropriate. | |
3716 | * | |
3717 | * @param in the inode to check | |
3718 | * @param flags flags to apply to cap check | |
3719 | */ | |
3720 | void Client::check_caps(Inode *in, unsigned flags) | |
3721 | { | |
3722 | unsigned wanted = in->caps_wanted(); | |
3723 | unsigned used = get_caps_used(in); | |
3724 | unsigned cap_used; | |
3725 | ||
7c673cae FG |
3726 | int implemented; |
3727 | int issued = in->caps_issued(&implemented); | |
3728 | int revoking = implemented & ~issued; | |
3729 | ||
11fdf7f2 TL |
3730 | int orig_used = used; |
3731 | used = adjust_caps_used_for_lazyio(used, issued, implemented); | |
3732 | ||
7c673cae | 3733 | int retain = wanted | used | CEPH_CAP_PIN; |
f67539c2 | 3734 | if (!is_unmounting() && in->nlink > 0) { |
a8e16298 | 3735 | if (wanted) { |
7c673cae | 3736 | retain |= CEPH_CAP_ANY; |
a8e16298 TL |
3737 | } else if (in->is_dir() && |
3738 | (issued & CEPH_CAP_FILE_SHARED) && | |
3739 | (in->flags & I_COMPLETE)) { | |
3740 | // we do this here because we don't want to drop to Fs (and then | |
3741 | // drop the Fs if we do a create!) if that alone makes us send lookups | |
3742 | // to the MDS. Doing it in in->caps_wanted() has knock-on effects elsewhere | |
3743 | wanted = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; | |
3744 | retain |= wanted; | |
3745 | } else { | |
7c673cae | 3746 | retain |= CEPH_CAP_ANY_SHARED; |
a8e16298 TL |
3747 | // keep RD only if we didn't have the file open RW, |
3748 | // because then the mds would revoke it anyway to | |
3749 | // journal max_size=0. | |
3750 | if (in->max_size == 0) | |
3751 | retain |= CEPH_CAP_ANY_RD; | |
3752 | } | |
7c673cae FG |
3753 | } |
3754 | ||
11fdf7f2 | 3755 | ldout(cct, 10) << __func__ << " on " << *in |
7c673cae FG |
3756 | << " wanted " << ccap_string(wanted) |
3757 | << " used " << ccap_string(used) | |
3758 | << " issued " << ccap_string(issued) | |
3759 | << " revoking " << ccap_string(revoking) | |
3760 | << " flags=" << flags | |
3761 | << dendl; | |
3762 | ||
3763 | if (in->snapid != CEPH_NOSNAP) | |
3764 | return; //snap caps last forever, can't write | |
3765 | ||
3766 | if (in->caps.empty()) | |
3767 | return; // guard if at end of func | |
3768 | ||
11fdf7f2 TL |
3769 | if (!(orig_used & CEPH_CAP_FILE_BUFFER) && |
3770 | (revoking & used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
94b18763 | 3771 | if (_release(in)) |
11fdf7f2 | 3772 | used &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
94b18763 | 3773 | } |
7c673cae | 3774 | |
20effc67 TL |
3775 | for (auto &[mds, cap] : in->caps) { |
3776 | auto session = mds_sessions.at(mds); | |
7c673cae FG |
3777 | |
3778 | cap_used = used; | |
11fdf7f2 | 3779 | if (in->auth_cap && &cap != in->auth_cap) |
7c673cae FG |
3780 | cap_used &= ~in->auth_cap->issued; |
3781 | ||
11fdf7f2 | 3782 | revoking = cap.implemented & ~cap.issued; |
20effc67 | 3783 | |
7c673cae | 3784 | ldout(cct, 10) << " cap mds." << mds |
11fdf7f2 TL |
3785 | << " issued " << ccap_string(cap.issued) |
3786 | << " implemented " << ccap_string(cap.implemented) | |
7c673cae FG |
3787 | << " revoking " << ccap_string(revoking) << dendl; |
3788 | ||
3789 | if (in->wanted_max_size > in->max_size && | |
3790 | in->wanted_max_size > in->requested_max_size && | |
11fdf7f2 | 3791 | &cap == in->auth_cap) |
7c673cae FG |
3792 | goto ack; |
3793 | ||
3794 | /* approaching file_max? */ | |
11fdf7f2 TL |
3795 | if ((cap.issued & CEPH_CAP_FILE_WR) && |
3796 | &cap == in->auth_cap && | |
31f18b77 | 3797 | is_max_size_approaching(in)) { |
7c673cae | 3798 | ldout(cct, 10) << "size " << in->size << " approaching max_size " << in->max_size |
31f18b77 | 3799 | << ", reported " << in->reported_size << dendl; |
7c673cae FG |
3800 | goto ack; |
3801 | } | |
3802 | ||
3803 | /* completed revocation? */ | |
3804 | if (revoking && (revoking & cap_used) == 0) { | |
11fdf7f2 | 3805 | ldout(cct, 10) << "completed revocation of " << ccap_string(cap.implemented & ~cap.issued) << dendl; |
7c673cae FG |
3806 | goto ack; |
3807 | } | |
3808 | ||
3809 | /* want more caps from mds? */ | |
11fdf7f2 | 3810 | if (wanted & ~(cap.wanted | cap.issued)) |
7c673cae FG |
3811 | goto ack; |
3812 | ||
f67539c2 | 3813 | if (!revoking && is_unmounting() && (cap_used == 0)) |
7c673cae FG |
3814 | goto ack; |
3815 | ||
11fdf7f2 | 3816 | if ((cap.issued & ~retain) == 0 && // and we don't have anything we wouldn't like |
a8e16298 | 3817 | !in->dirty_caps) // and we have no dirty caps |
7c673cae FG |
3818 | continue; |
3819 | ||
11fdf7f2 | 3820 | if (!(flags & CHECK_CAPS_NODELAY)) { |
7c673cae | 3821 | ldout(cct, 10) << "delaying cap release" << dendl; |
11fdf7f2 | 3822 | cap_delay_requeue(in); |
7c673cae FG |
3823 | continue; |
3824 | } | |
3825 | ||
3826 | ack: | |
eafe8130 TL |
3827 | if (&cap == in->auth_cap) { |
3828 | if (in->flags & I_KICK_FLUSH) { | |
3829 | ldout(cct, 20) << " reflushing caps (check_caps) on " << *in | |
3830 | << " to mds." << mds << dendl; | |
20effc67 | 3831 | kick_flushing_caps(in, session.get()); |
eafe8130 TL |
3832 | } |
3833 | if (!in->cap_snaps.empty() && | |
3834 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3835 | flush_snaps(in); | |
7c673cae FG |
3836 | } |
3837 | ||
3838 | int flushing; | |
e306af50 | 3839 | int msg_flags = 0; |
7c673cae | 3840 | ceph_tid_t flush_tid; |
11fdf7f2 | 3841 | if (in->auth_cap == &cap && in->dirty_caps) { |
7c673cae | 3842 | flushing = mark_caps_flushing(in, &flush_tid); |
e306af50 TL |
3843 | if (flags & CHECK_CAPS_SYNCHRONOUS) |
3844 | msg_flags |= MClientCaps::FLAG_SYNC; | |
7c673cae FG |
3845 | } else { |
3846 | flushing = 0; | |
3847 | flush_tid = 0; | |
3848 | } | |
3849 | ||
20effc67 TL |
3850 | in->delay_cap_item.remove_myself(); |
3851 | send_cap(in, session.get(), &cap, msg_flags, cap_used, wanted, retain, | |
eafe8130 | 3852 | flushing, flush_tid); |
7c673cae FG |
3853 | } |
3854 | } | |
3855 | ||
3856 | ||
3857 | void Client::queue_cap_snap(Inode *in, SnapContext& old_snapc) | |
3858 | { | |
3859 | int used = get_caps_used(in); | |
3860 | int dirty = in->caps_dirty(); | |
11fdf7f2 | 3861 | ldout(cct, 10) << __func__ << " " << *in << " snapc " << old_snapc << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3862 | |
3863 | if (in->cap_snaps.size() && | |
3864 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3865 | ldout(cct, 10) << __func__ << " already have pending cap_snap on " << *in << dendl; |
7c673cae FG |
3866 | return; |
3867 | } else if (in->caps_dirty() || | |
3868 | (used & CEPH_CAP_FILE_WR) || | |
3869 | (dirty & CEPH_CAP_ANY_WR)) { | |
3870 | const auto &capsnapem = in->cap_snaps.emplace(std::piecewise_construct, std::make_tuple(old_snapc.seq), std::make_tuple(in)); | |
11fdf7f2 | 3871 | ceph_assert(capsnapem.second); /* element inserted */ |
7c673cae FG |
3872 | CapSnap &capsnap = capsnapem.first->second; |
3873 | capsnap.context = old_snapc; | |
3874 | capsnap.issued = in->caps_issued(); | |
3875 | capsnap.dirty = in->caps_dirty(); | |
f67539c2 | 3876 | |
7c673cae | 3877 | capsnap.dirty_data = (used & CEPH_CAP_FILE_BUFFER); |
f67539c2 | 3878 | |
7c673cae FG |
3879 | capsnap.uid = in->uid; |
3880 | capsnap.gid = in->gid; | |
3881 | capsnap.mode = in->mode; | |
3882 | capsnap.btime = in->btime; | |
3883 | capsnap.xattrs = in->xattrs; | |
3884 | capsnap.xattr_version = in->xattr_version; | |
11fdf7f2 TL |
3885 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; |
3886 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
f67539c2 | 3887 | |
7c673cae | 3888 | if (used & CEPH_CAP_FILE_WR) { |
11fdf7f2 | 3889 | ldout(cct, 10) << __func__ << " WR used on " << *in << dendl; |
7c673cae FG |
3890 | capsnap.writing = 1; |
3891 | } else { | |
3892 | finish_cap_snap(in, capsnap, used); | |
3893 | } | |
3894 | } else { | |
11fdf7f2 | 3895 | ldout(cct, 10) << __func__ << " not dirty|writing on " << *in << dendl; |
7c673cae FG |
3896 | } |
3897 | } | |
3898 | ||
3899 | void Client::finish_cap_snap(Inode *in, CapSnap &capsnap, int used) | |
3900 | { | |
11fdf7f2 | 3901 | ldout(cct, 10) << __func__ << " " << *in << " capsnap " << (void *)&capsnap << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3902 | capsnap.size = in->size; |
3903 | capsnap.mtime = in->mtime; | |
3904 | capsnap.atime = in->atime; | |
3905 | capsnap.ctime = in->ctime; | |
3906 | capsnap.time_warp_seq = in->time_warp_seq; | |
3907 | capsnap.change_attr = in->change_attr; | |
7c673cae FG |
3908 | capsnap.dirty |= in->caps_dirty(); |
3909 | ||
11fdf7f2 TL |
3910 | /* Only reset it if it wasn't set before */ |
3911 | if (capsnap.cap_dirtier_uid == -1) { | |
3912 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; | |
3913 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
3914 | } | |
3915 | ||
7c673cae FG |
3916 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { |
3917 | capsnap.inline_data = in->inline_data; | |
3918 | capsnap.inline_version = in->inline_version; | |
3919 | } | |
3920 | ||
3921 | if (used & CEPH_CAP_FILE_BUFFER) { | |
f67539c2 | 3922 | capsnap.writing = 1; |
11fdf7f2 | 3923 | ldout(cct, 10) << __func__ << " " << *in << " cap_snap " << &capsnap << " used " << used |
7c673cae FG |
3924 | << " WRBUFFER, delaying" << dendl; |
3925 | } else { | |
3926 | capsnap.dirty_data = 0; | |
3927 | flush_snaps(in); | |
3928 | } | |
3929 | } | |
3930 | ||
eafe8130 TL |
3931 | void Client::send_flush_snap(Inode *in, MetaSession *session, |
3932 | snapid_t follows, CapSnap& capsnap) | |
3933 | { | |
9f95a23c TL |
3934 | auto m = make_message<MClientCaps>(CEPH_CAP_OP_FLUSHSNAP, |
3935 | in->ino, in->snaprealm->ino, 0, | |
3936 | in->auth_cap->mseq, cap_epoch_barrier); | |
eafe8130 TL |
3937 | m->caller_uid = capsnap.cap_dirtier_uid; |
3938 | m->caller_gid = capsnap.cap_dirtier_gid; | |
3939 | ||
3940 | m->set_client_tid(capsnap.flush_tid); | |
3941 | m->head.snap_follows = follows; | |
3942 | ||
3943 | m->head.caps = capsnap.issued; | |
3944 | m->head.dirty = capsnap.dirty; | |
3945 | ||
3946 | m->head.uid = capsnap.uid; | |
3947 | m->head.gid = capsnap.gid; | |
3948 | m->head.mode = capsnap.mode; | |
3949 | m->btime = capsnap.btime; | |
3950 | ||
3951 | m->size = capsnap.size; | |
3952 | ||
3953 | m->head.xattr_version = capsnap.xattr_version; | |
3954 | encode(capsnap.xattrs, m->xattrbl); | |
3955 | ||
3956 | m->ctime = capsnap.ctime; | |
3957 | m->btime = capsnap.btime; | |
3958 | m->mtime = capsnap.mtime; | |
3959 | m->atime = capsnap.atime; | |
3960 | m->time_warp_seq = capsnap.time_warp_seq; | |
3961 | m->change_attr = capsnap.change_attr; | |
3962 | ||
3963 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { | |
3964 | m->inline_version = in->inline_version; | |
3965 | m->inline_data = in->inline_data; | |
3966 | } | |
3967 | ||
3968 | ceph_assert(!session->flushing_caps_tids.empty()); | |
3969 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3970 | ||
3971 | session->con->send_message2(std::move(m)); | |
3972 | } | |
3973 | ||
3974 | void Client::flush_snaps(Inode *in) | |
7c673cae | 3975 | { |
eafe8130 | 3976 | ldout(cct, 10) << "flush_snaps on " << *in << dendl; |
11fdf7f2 | 3977 | ceph_assert(in->cap_snaps.size()); |
7c673cae FG |
3978 | |
3979 | // pick auth mds | |
11fdf7f2 | 3980 | ceph_assert(in->auth_cap); |
7c673cae | 3981 | MetaSession *session = in->auth_cap->session; |
7c673cae FG |
3982 | |
3983 | for (auto &p : in->cap_snaps) { | |
3984 | CapSnap &capsnap = p.second; | |
eafe8130 TL |
3985 | // only do new flush |
3986 | if (capsnap.flush_tid > 0) | |
3987 | continue; | |
7c673cae FG |
3988 | |
3989 | ldout(cct, 10) << "flush_snaps mds." << session->mds_num | |
3990 | << " follows " << p.first | |
3991 | << " size " << capsnap.size | |
3992 | << " mtime " << capsnap.mtime | |
3993 | << " dirty_data=" << capsnap.dirty_data | |
3994 | << " writing=" << capsnap.writing | |
3995 | << " on " << *in << dendl; | |
3996 | if (capsnap.dirty_data || capsnap.writing) | |
eafe8130 | 3997 | break; |
f67539c2 | 3998 | |
eafe8130 TL |
3999 | capsnap.flush_tid = ++last_flush_tid; |
4000 | session->flushing_caps_tids.insert(capsnap.flush_tid); | |
4001 | in->flushing_cap_tids[capsnap.flush_tid] = 0; | |
4002 | if (!in->flushing_cap_item.is_on_list()) | |
4003 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
7c673cae | 4004 | |
eafe8130 | 4005 | send_flush_snap(in, session, p.first, capsnap); |
7c673cae FG |
4006 | } |
4007 | } | |
4008 | ||
9f95a23c | 4009 | void Client::wait_on_list(list<ceph::condition_variable*>& ls) |
7c673cae | 4010 | { |
9f95a23c | 4011 | ceph::condition_variable cond; |
7c673cae | 4012 | ls.push_back(&cond); |
9f95a23c TL |
4013 | std::unique_lock l{client_lock, std::adopt_lock}; |
4014 | cond.wait(l); | |
4015 | l.release(); | |
7c673cae FG |
4016 | ls.remove(&cond); |
4017 | } | |
4018 | ||
9f95a23c | 4019 | void Client::signal_cond_list(list<ceph::condition_variable*>& ls) |
7c673cae | 4020 | { |
9f95a23c TL |
4021 | for (auto cond : ls) { |
4022 | cond->notify_all(); | |
4023 | } | |
7c673cae FG |
4024 | } |
4025 | ||
4026 | void Client::wait_on_context_list(list<Context*>& ls) | |
4027 | { | |
9f95a23c | 4028 | ceph::condition_variable cond; |
7c673cae FG |
4029 | bool done = false; |
4030 | int r; | |
9f95a23c TL |
4031 | ls.push_back(new C_Cond(cond, &done, &r)); |
4032 | std::unique_lock l{client_lock, std::adopt_lock}; | |
4033 | cond.wait(l, [&done] { return done;}); | |
4034 | l.release(); | |
7c673cae FG |
4035 | } |
4036 | ||
4037 | void Client::signal_context_list(list<Context*>& ls) | |
4038 | { | |
4039 | while (!ls.empty()) { | |
4040 | ls.front()->complete(0); | |
4041 | ls.pop_front(); | |
4042 | } | |
4043 | } | |
4044 | ||
a8e16298 | 4045 | void Client::wake_up_session_caps(MetaSession *s, bool reconnect) |
7c673cae | 4046 | { |
11fdf7f2 TL |
4047 | for (const auto &cap : s->caps) { |
4048 | auto &in = cap->inode; | |
a8e16298 | 4049 | if (reconnect) { |
11fdf7f2 TL |
4050 | in.requested_max_size = 0; |
4051 | in.wanted_max_size = 0; | |
a8e16298 TL |
4052 | } else { |
4053 | if (cap->gen < s->cap_gen) { | |
4054 | // mds did not re-issue stale cap. | |
4055 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
4056 | // make sure mds knows what we want. | |
11fdf7f2 TL |
4057 | if (in.caps_file_wanted() & ~cap->wanted) |
4058 | in.flags |= I_CAP_DROPPED; | |
a8e16298 TL |
4059 | } |
4060 | } | |
11fdf7f2 | 4061 | signal_cond_list(in.waitfor_caps); |
7c673cae FG |
4062 | } |
4063 | } | |
4064 | ||
4065 | ||
4066 | // flush dirty data (from objectcache) | |
4067 | ||
4068 | class C_Client_CacheInvalidate : public Context { | |
4069 | private: | |
4070 | Client *client; | |
4071 | vinodeno_t ino; | |
4072 | int64_t offset, length; | |
4073 | public: | |
4074 | C_Client_CacheInvalidate(Client *c, Inode *in, int64_t off, int64_t len) : | |
4075 | client(c), offset(off), length(len) { | |
4076 | if (client->use_faked_inos()) | |
4077 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4078 | else | |
4079 | ino = in->vino(); | |
4080 | } | |
4081 | void finish(int r) override { | |
4082 | // _async_invalidate takes the lock when it needs to, call this back from outside of lock. | |
9f95a23c | 4083 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
4084 | client->_async_invalidate(ino, offset, length); |
4085 | } | |
4086 | }; | |
4087 | ||
4088 | void Client::_async_invalidate(vinodeno_t ino, int64_t off, int64_t len) | |
4089 | { | |
f67539c2 TL |
4090 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4091 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4092 | return; |
f67539c2 | 4093 | |
11fdf7f2 | 4094 | ldout(cct, 10) << __func__ << " " << ino << " " << off << "~" << len << dendl; |
7c673cae FG |
4095 | ino_invalidate_cb(callback_handle, ino, off, len); |
4096 | } | |
4097 | ||
4098 | void Client::_schedule_invalidate_callback(Inode *in, int64_t off, int64_t len) { | |
4099 | ||
4100 | if (ino_invalidate_cb) | |
4101 | // we queue the invalidate, which calls the callback and decrements the ref | |
4102 | async_ino_invalidator.queue(new C_Client_CacheInvalidate(this, in, off, len)); | |
4103 | } | |
4104 | ||
4105 | void Client::_invalidate_inode_cache(Inode *in) | |
4106 | { | |
11fdf7f2 | 4107 | ldout(cct, 10) << __func__ << " " << *in << dendl; |
7c673cae FG |
4108 | |
4109 | // invalidate our userspace inode cache | |
94b18763 | 4110 | if (cct->_conf->client_oc) { |
7c673cae | 4111 | objectcacher->release_set(&in->oset); |
94b18763 FG |
4112 | if (!objectcacher->set_is_empty(&in->oset)) |
4113 | lderr(cct) << "failed to invalidate cache for " << *in << dendl; | |
4114 | } | |
7c673cae FG |
4115 | |
4116 | _schedule_invalidate_callback(in, 0, 0); | |
4117 | } | |
4118 | ||
4119 | void Client::_invalidate_inode_cache(Inode *in, int64_t off, int64_t len) | |
4120 | { | |
11fdf7f2 | 4121 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
4122 | |
4123 | // invalidate our userspace inode cache | |
4124 | if (cct->_conf->client_oc) { | |
4125 | vector<ObjectExtent> ls; | |
4126 | Striper::file_to_extents(cct, in->ino, &in->layout, off, len, in->truncate_size, ls); | |
28e407b8 | 4127 | objectcacher->discard_writeback(&in->oset, ls, nullptr); |
7c673cae FG |
4128 | } |
4129 | ||
4130 | _schedule_invalidate_callback(in, off, len); | |
4131 | } | |
4132 | ||
4133 | bool Client::_release(Inode *in) | |
4134 | { | |
4135 | ldout(cct, 20) << "_release " << *in << dendl; | |
4136 | if (in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
4137 | _invalidate_inode_cache(in); | |
4138 | return true; | |
4139 | } | |
4140 | return false; | |
4141 | } | |
4142 | ||
4143 | bool Client::_flush(Inode *in, Context *onfinish) | |
4144 | { | |
4145 | ldout(cct, 10) << "_flush " << *in << dendl; | |
4146 | ||
4147 | if (!in->oset.dirty_or_tx) { | |
4148 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4149 | onfinish->complete(0); | |
4150 | return true; | |
4151 | } | |
4152 | ||
4153 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
1adf2230 | 4154 | ldout(cct, 8) << __func__ << ": FULL, purging for ENOSPC" << dendl; |
7c673cae FG |
4155 | objectcacher->purge_set(&in->oset); |
4156 | if (onfinish) { | |
f67539c2 | 4157 | onfinish->complete(-CEPHFS_ENOSPC); |
7c673cae FG |
4158 | } |
4159 | return true; | |
4160 | } | |
4161 | ||
4162 | return objectcacher->flush_set(&in->oset, onfinish); | |
4163 | } | |
4164 | ||
4165 | void Client::_flush_range(Inode *in, int64_t offset, uint64_t size) | |
4166 | { | |
f67539c2 | 4167 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
4168 | if (!in->oset.dirty_or_tx) { |
4169 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4170 | return; | |
4171 | } | |
4172 | ||
11fdf7f2 | 4173 | C_SaferCond onflush("Client::_flush_range flock"); |
7c673cae | 4174 | bool ret = objectcacher->file_flush(&in->oset, &in->layout, in->snaprealm->get_snap_context(), |
11fdf7f2 | 4175 | offset, size, &onflush); |
7c673cae FG |
4176 | if (!ret) { |
4177 | // wait for flush | |
9f95a23c | 4178 | client_lock.unlock(); |
11fdf7f2 | 4179 | onflush.wait(); |
9f95a23c | 4180 | client_lock.lock(); |
7c673cae FG |
4181 | } |
4182 | } | |
4183 | ||
4184 | void Client::flush_set_callback(ObjectCacher::ObjectSet *oset) | |
4185 | { | |
f67539c2 TL |
4186 | // std::scoped_lock l(client_lock); |
4187 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); // will be called via dispatch() -> objecter -> ... | |
7c673cae | 4188 | Inode *in = static_cast<Inode *>(oset->parent); |
11fdf7f2 | 4189 | ceph_assert(in); |
7c673cae FG |
4190 | _flushed(in); |
4191 | } | |
4192 | ||
4193 | void Client::_flushed(Inode *in) | |
4194 | { | |
4195 | ldout(cct, 10) << "_flushed " << *in << dendl; | |
4196 | ||
4197 | put_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
4198 | } | |
4199 | ||
4200 | ||
4201 | ||
4202 | // checks common to add_update_cap, handle_cap_grant | |
11fdf7f2 | 4203 | void Client::check_cap_issue(Inode *in, unsigned issued) |
7c673cae FG |
4204 | { |
4205 | unsigned had = in->caps_issued(); | |
4206 | ||
4207 | if ((issued & CEPH_CAP_FILE_CACHE) && | |
4208 | !(had & CEPH_CAP_FILE_CACHE)) | |
4209 | in->cache_gen++; | |
4210 | ||
f91f0fd5 TL |
4211 | if ((issued & CEPH_CAP_FILE_SHARED) != |
4212 | (had & CEPH_CAP_FILE_SHARED)) { | |
4213 | if (issued & CEPH_CAP_FILE_SHARED) | |
4214 | in->shared_gen++; | |
7c673cae FG |
4215 | if (in->is_dir()) |
4216 | clear_dir_complete_and_ordered(in, true); | |
4217 | } | |
4218 | } | |
4219 | ||
4220 | void Client::add_update_cap(Inode *in, MetaSession *mds_session, uint64_t cap_id, | |
a8e16298 TL |
4221 | unsigned issued, unsigned wanted, unsigned seq, unsigned mseq, |
4222 | inodeno_t realm, int flags, const UserPerm& cap_perms) | |
7c673cae | 4223 | { |
11fdf7f2 TL |
4224 | if (!in->is_any_caps()) { |
4225 | ceph_assert(in->snaprealm == 0); | |
4226 | in->snaprealm = get_snap_realm(realm); | |
4227 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4228 | ldout(cct, 15) << __func__ << " first one, opened snaprealm " << in->snaprealm << dendl; | |
4229 | } else { | |
4230 | ceph_assert(in->snaprealm); | |
4231 | if ((flags & CEPH_CAP_FLAG_AUTH) && | |
4232 | realm != inodeno_t(-1) && in->snaprealm->ino != realm) { | |
4233 | in->snaprealm_item.remove_myself(); | |
4234 | auto oldrealm = in->snaprealm; | |
4235 | in->snaprealm = get_snap_realm(realm); | |
4236 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4237 | put_snap_realm(oldrealm); | |
4238 | } | |
4239 | } | |
4240 | ||
7c673cae | 4241 | mds_rank_t mds = mds_session->mds_num; |
11fdf7f2 TL |
4242 | const auto &capem = in->caps.emplace(std::piecewise_construct, std::forward_as_tuple(mds), std::forward_as_tuple(*in, mds_session)); |
4243 | Cap &cap = capem.first->second; | |
4244 | if (!capem.second) { | |
4245 | if (cap.gen < mds_session->cap_gen) | |
4246 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
7c673cae FG |
4247 | |
4248 | /* | |
4249 | * auth mds of the inode changed. we received the cap export | |
4250 | * message, but still haven't received the cap import message. | |
4251 | * handle_cap_export() updated the new auth MDS' cap. | |
4252 | * | |
4253 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
4254 | * a message that was send before the cap import message. So | |
4255 | * don't remove caps. | |
4256 | */ | |
11fdf7f2 | 4257 | if (ceph_seq_cmp(seq, cap.seq) <= 0) { |
92f5a8d4 TL |
4258 | if (&cap != in->auth_cap) |
4259 | ldout(cct, 0) << "WARNING: " << "inode " << *in << " caps on mds." << mds << " != auth_cap." << dendl; | |
4260 | ||
11fdf7f2 TL |
4261 | ceph_assert(cap.cap_id == cap_id); |
4262 | seq = cap.seq; | |
4263 | mseq = cap.mseq; | |
4264 | issued |= cap.issued; | |
7c673cae FG |
4265 | flags |= CEPH_CAP_FLAG_AUTH; |
4266 | } | |
f67539c2 TL |
4267 | } else { |
4268 | inc_pinned_icaps(); | |
7c673cae FG |
4269 | } |
4270 | ||
11fdf7f2 | 4271 | check_cap_issue(in, issued); |
7c673cae FG |
4272 | |
4273 | if (flags & CEPH_CAP_FLAG_AUTH) { | |
11fdf7f2 | 4274 | if (in->auth_cap != &cap && |
7c673cae FG |
4275 | (!in->auth_cap || ceph_seq_cmp(in->auth_cap->mseq, mseq) < 0)) { |
4276 | if (in->auth_cap && in->flushing_cap_item.is_on_list()) { | |
11fdf7f2 | 4277 | ldout(cct, 10) << __func__ << " changing auth cap: " |
7c673cae FG |
4278 | << "add myself to new auth MDS' flushing caps list" << dendl; |
4279 | adjust_session_flushing_caps(in, in->auth_cap->session, mds_session); | |
4280 | } | |
11fdf7f2 | 4281 | in->auth_cap = ∩ |
7c673cae FG |
4282 | } |
4283 | } | |
4284 | ||
11fdf7f2 TL |
4285 | unsigned old_caps = cap.issued; |
4286 | cap.cap_id = cap_id; | |
4287 | cap.issued = issued; | |
4288 | cap.implemented |= issued; | |
4289 | if (ceph_seq_cmp(mseq, cap.mseq) > 0) | |
4290 | cap.wanted = wanted; | |
a8e16298 | 4291 | else |
11fdf7f2 TL |
4292 | cap.wanted |= wanted; |
4293 | cap.seq = seq; | |
4294 | cap.issue_seq = seq; | |
4295 | cap.mseq = mseq; | |
4296 | cap.gen = mds_session->cap_gen; | |
4297 | cap.latest_perms = cap_perms; | |
4298 | ldout(cct, 10) << __func__ << " issued " << ccap_string(old_caps) << " -> " << ccap_string(cap.issued) | |
4299 | << " from mds." << mds | |
4300 | << " on " << *in | |
4301 | << dendl; | |
4302 | ||
4303 | if ((issued & ~old_caps) && in->auth_cap == &cap) { | |
7c673cae | 4304 | // non-auth MDS is revoking the newly grant caps ? |
11fdf7f2 TL |
4305 | for (auto &p : in->caps) { |
4306 | if (&p.second == &cap) | |
7c673cae | 4307 | continue; |
11fdf7f2 | 4308 | if (p.second.implemented & ~p.second.issued & issued) { |
7c673cae FG |
4309 | check_caps(in, CHECK_CAPS_NODELAY); |
4310 | break; | |
4311 | } | |
4312 | } | |
4313 | } | |
4314 | ||
4315 | if (issued & ~old_caps) | |
4316 | signal_cond_list(in->waitfor_caps); | |
4317 | } | |
4318 | ||
4319 | void Client::remove_cap(Cap *cap, bool queue_release) | |
4320 | { | |
11fdf7f2 | 4321 | auto &in = cap->inode; |
7c673cae FG |
4322 | MetaSession *session = cap->session; |
4323 | mds_rank_t mds = cap->session->mds_num; | |
4324 | ||
11fdf7f2 | 4325 | ldout(cct, 10) << __func__ << " mds." << mds << " on " << in << dendl; |
7c673cae FG |
4326 | |
4327 | if (queue_release) { | |
4328 | session->enqueue_cap_release( | |
11fdf7f2 | 4329 | in.ino, |
7c673cae FG |
4330 | cap->cap_id, |
4331 | cap->issue_seq, | |
4332 | cap->mseq, | |
4333 | cap_epoch_barrier); | |
f67539c2 TL |
4334 | } else { |
4335 | dec_pinned_icaps(); | |
7c673cae FG |
4336 | } |
4337 | ||
f67539c2 | 4338 | |
11fdf7f2 TL |
4339 | if (in.auth_cap == cap) { |
4340 | if (in.flushing_cap_item.is_on_list()) { | |
7c673cae | 4341 | ldout(cct, 10) << " removing myself from flushing_cap list" << dendl; |
11fdf7f2 | 4342 | in.flushing_cap_item.remove_myself(); |
7c673cae | 4343 | } |
11fdf7f2 | 4344 | in.auth_cap = NULL; |
7c673cae | 4345 | } |
11fdf7f2 TL |
4346 | size_t n = in.caps.erase(mds); |
4347 | ceph_assert(n == 1); | |
7c673cae FG |
4348 | cap = nullptr; |
4349 | ||
11fdf7f2 TL |
4350 | if (!in.is_any_caps()) { |
4351 | ldout(cct, 15) << __func__ << " last one, closing snaprealm " << in.snaprealm << dendl; | |
4352 | in.snaprealm_item.remove_myself(); | |
4353 | put_snap_realm(in.snaprealm); | |
4354 | in.snaprealm = 0; | |
7c673cae FG |
4355 | } |
4356 | } | |
4357 | ||
4358 | void Client::remove_all_caps(Inode *in) | |
4359 | { | |
4360 | while (!in->caps.empty()) | |
11fdf7f2 | 4361 | remove_cap(&in->caps.begin()->second, true); |
7c673cae FG |
4362 | } |
4363 | ||
f6b5b4d7 | 4364 | void Client::remove_session_caps(MetaSession *s, int err) |
7c673cae | 4365 | { |
11fdf7f2 | 4366 | ldout(cct, 10) << __func__ << " mds." << s->mds_num << dendl; |
7c673cae FG |
4367 | |
4368 | while (s->caps.size()) { | |
4369 | Cap *cap = *s->caps.begin(); | |
11fdf7f2 | 4370 | InodeRef in(&cap->inode); |
eafe8130 | 4371 | bool dirty_caps = false; |
7c673cae | 4372 | if (in->auth_cap == cap) { |
7c673cae FG |
4373 | dirty_caps = in->dirty_caps | in->flushing_caps; |
4374 | in->wanted_max_size = 0; | |
4375 | in->requested_max_size = 0; | |
f6b5b4d7 TL |
4376 | if (in->has_any_filelocks()) |
4377 | in->flags |= I_ERROR_FILELOCK; | |
7c673cae | 4378 | } |
f6b5b4d7 | 4379 | auto caps = cap->implemented; |
a8e16298 TL |
4380 | if (cap->wanted | cap->issued) |
4381 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 4382 | remove_cap(cap, false); |
eafe8130 | 4383 | in->cap_snaps.clear(); |
7c673cae | 4384 | if (dirty_caps) { |
11fdf7f2 | 4385 | lderr(cct) << __func__ << " still has dirty|flushing caps on " << *in << dendl; |
7c673cae FG |
4386 | if (in->flushing_caps) { |
4387 | num_flushing_caps--; | |
4388 | in->flushing_cap_tids.clear(); | |
4389 | } | |
4390 | in->flushing_caps = 0; | |
28e407b8 | 4391 | in->mark_caps_clean(); |
11fdf7f2 | 4392 | put_inode(in.get()); |
7c673cae | 4393 | } |
f6b5b4d7 TL |
4394 | caps &= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER; |
4395 | if (caps && !in->caps_issued_mask(caps, true)) { | |
f67539c2 | 4396 | if (err == -CEPHFS_EBLOCKLISTED) { |
f6b5b4d7 TL |
4397 | if (in->oset.dirty_or_tx) { |
4398 | lderr(cct) << __func__ << " still has dirty data on " << *in << dendl; | |
4399 | in->set_async_err(err); | |
4400 | } | |
4401 | objectcacher->purge_set(&in->oset); | |
4402 | } else { | |
4403 | objectcacher->release_set(&in->oset); | |
4404 | } | |
4405 | _schedule_invalidate_callback(in.get(), 0, 0); | |
4406 | } | |
4407 | ||
a8e16298 | 4408 | signal_cond_list(in->waitfor_caps); |
7c673cae FG |
4409 | } |
4410 | s->flushing_caps_tids.clear(); | |
9f95a23c | 4411 | sync_cond.notify_all(); |
7c673cae FG |
4412 | } |
4413 | ||
1d09f67e | 4414 | std::pair<int, bool> Client::_do_remount(bool retry_on_error) |
b32b8144 | 4415 | { |
adb31ebb | 4416 | uint64_t max_retries = cct->_conf.get_val<uint64_t>("mds_max_retries_on_remount_failure"); |
1d09f67e | 4417 | bool abort_on_failure = false; |
91327a77 | 4418 | |
b32b8144 FG |
4419 | errno = 0; |
4420 | int r = remount_cb(callback_handle); | |
91327a77 AA |
4421 | if (r == 0) { |
4422 | retries_on_invalidate = 0; | |
4423 | } else { | |
b32b8144 FG |
4424 | int e = errno; |
4425 | client_t whoami = get_nodeid(); | |
4426 | if (r == -1) { | |
4427 | lderr(cct) << | |
4428 | "failed to remount (to trim kernel dentries): " | |
4429 | "errno = " << e << " (" << strerror(e) << ")" << dendl; | |
4430 | } else { | |
4431 | lderr(cct) << | |
4432 | "failed to remount (to trim kernel dentries): " | |
4433 | "return code = " << r << dendl; | |
4434 | } | |
91327a77 | 4435 | bool should_abort = |
11fdf7f2 TL |
4436 | (cct->_conf.get_val<bool>("client_die_on_failed_remount") || |
4437 | cct->_conf.get_val<bool>("client_die_on_failed_dentry_invalidate")) && | |
91327a77 | 4438 | !(retry_on_error && (++retries_on_invalidate < max_retries)); |
f67539c2 | 4439 | if (should_abort && !is_unmounting()) { |
b32b8144 | 4440 | lderr(cct) << "failed to remount for kernel dentry trimming; quitting!" << dendl; |
1d09f67e | 4441 | abort_on_failure = true; |
b32b8144 FG |
4442 | } |
4443 | } | |
1d09f67e | 4444 | return std::make_pair(r, abort_on_failure); |
b32b8144 FG |
4445 | } |
4446 | ||
7c673cae FG |
4447 | class C_Client_Remount : public Context { |
4448 | private: | |
4449 | Client *client; | |
4450 | public: | |
4451 | explicit C_Client_Remount(Client *c) : client(c) {} | |
4452 | void finish(int r) override { | |
11fdf7f2 | 4453 | ceph_assert(r == 0); |
91327a77 | 4454 | client->_do_remount(true); |
7c673cae FG |
4455 | } |
4456 | }; | |
4457 | ||
4458 | void Client::_invalidate_kernel_dcache() | |
4459 | { | |
f67539c2 TL |
4460 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4461 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4462 | return; |
f67539c2 | 4463 | |
94b18763 FG |
4464 | if (can_invalidate_dentries) { |
4465 | if (dentry_invalidate_cb && root->dir) { | |
4466 | for (ceph::unordered_map<string, Dentry*>::iterator p = root->dir->dentries.begin(); | |
4467 | p != root->dir->dentries.end(); | |
4468 | ++p) { | |
4469 | if (p->second->inode) | |
4470 | _schedule_invalidate_dentry_callback(p->second, false); | |
4471 | } | |
7c673cae FG |
4472 | } |
4473 | } else if (remount_cb) { | |
4474 | // Hacky: | |
4475 | // when remounting a file system, linux kernel trims all unused dentries in the fs | |
4476 | remount_finisher.queue(new C_Client_Remount(this)); | |
4477 | } | |
4478 | } | |
4479 | ||
91327a77 AA |
4480 | void Client::_trim_negative_child_dentries(InodeRef& in) |
4481 | { | |
4482 | if (!in->is_dir()) | |
4483 | return; | |
4484 | ||
4485 | Dir* dir = in->dir; | |
4486 | if (dir && dir->dentries.size() == dir->num_null_dentries) { | |
4487 | for (auto p = dir->dentries.begin(); p != dir->dentries.end(); ) { | |
4488 | Dentry *dn = p->second; | |
4489 | ++p; | |
11fdf7f2 | 4490 | ceph_assert(!dn->inode); |
91327a77 AA |
4491 | if (dn->lru_is_expireable()) |
4492 | unlink(dn, true, false); // keep dir, drop dentry | |
4493 | } | |
4494 | if (dir->dentries.empty()) { | |
4495 | close_dir(dir); | |
4496 | } | |
4497 | } | |
4498 | ||
4499 | if (in->flags & I_SNAPDIR_OPEN) { | |
4500 | InodeRef snapdir = open_snapdir(in.get()); | |
4501 | _trim_negative_child_dentries(snapdir); | |
4502 | } | |
4503 | } | |
4504 | ||
e306af50 TL |
4505 | class C_Client_CacheRelease : public Context { |
4506 | private: | |
4507 | Client *client; | |
4508 | vinodeno_t ino; | |
4509 | public: | |
4510 | C_Client_CacheRelease(Client *c, Inode *in) : | |
4511 | client(c) { | |
4512 | if (client->use_faked_inos()) | |
4513 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4514 | else | |
4515 | ino = in->vino(); | |
4516 | } | |
4517 | void finish(int r) override { | |
4518 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); | |
4519 | client->_async_inode_release(ino); | |
4520 | } | |
4521 | }; | |
4522 | ||
4523 | void Client::_async_inode_release(vinodeno_t ino) | |
4524 | { | |
f67539c2 TL |
4525 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4526 | if (!mref_reader.is_state_satisfied()) | |
e306af50 | 4527 | return; |
f67539c2 | 4528 | |
e306af50 TL |
4529 | ldout(cct, 10) << __func__ << " " << ino << dendl; |
4530 | ino_release_cb(callback_handle, ino); | |
4531 | } | |
4532 | ||
4533 | void Client::_schedule_ino_release_callback(Inode *in) { | |
4534 | ||
4535 | if (ino_release_cb) | |
4536 | // we queue the invalidate, which calls the callback and decrements the ref | |
4537 | async_ino_releasor.queue(new C_Client_CacheRelease(this, in)); | |
4538 | } | |
4539 | ||
28e407b8 | 4540 | void Client::trim_caps(MetaSession *s, uint64_t max) |
7c673cae FG |
4541 | { |
4542 | mds_rank_t mds = s->mds_num; | |
28e407b8 | 4543 | size_t caps_size = s->caps.size(); |
11fdf7f2 | 4544 | ldout(cct, 10) << __func__ << " mds." << mds << " max " << max |
7c673cae FG |
4545 | << " caps " << caps_size << dendl; |
4546 | ||
28e407b8 AA |
4547 | uint64_t trimmed = 0; |
4548 | auto p = s->caps.begin(); | |
4549 | std::set<Dentry *> to_trim; /* this avoids caps other than the one we're | |
4550 | * looking at from getting deleted during traversal. */ | |
7c673cae FG |
4551 | while ((caps_size - trimmed) > max && !p.end()) { |
4552 | Cap *cap = *p; | |
11fdf7f2 | 4553 | InodeRef in(&cap->inode); |
7c673cae FG |
4554 | |
4555 | // Increment p early because it will be invalidated if cap | |
4556 | // is deleted inside remove_cap | |
4557 | ++p; | |
4558 | ||
4559 | if (in->caps.size() > 1 && cap != in->auth_cap) { | |
4560 | int mine = cap->issued | cap->implemented; | |
4561 | int oissued = in->auth_cap ? in->auth_cap->issued : 0; | |
4562 | // disposable non-auth cap | |
b32b8144 | 4563 | if (!(get_caps_used(in.get()) & ~oissued & mine)) { |
7c673cae | 4564 | ldout(cct, 20) << " removing unused, unneeded non-auth cap on " << *in << dendl; |
28e407b8 | 4565 | cap = (remove_cap(cap, true), nullptr); |
7c673cae FG |
4566 | trimmed++; |
4567 | } | |
4568 | } else { | |
4569 | ldout(cct, 20) << " trying to trim dentries for " << *in << dendl; | |
91327a77 | 4570 | _trim_negative_child_dentries(in); |
7c673cae | 4571 | bool all = true; |
11fdf7f2 TL |
4572 | auto q = in->dentries.begin(); |
4573 | while (q != in->dentries.end()) { | |
4574 | Dentry *dn = *q; | |
4575 | ++q; | |
7c673cae FG |
4576 | if (dn->lru_is_expireable()) { |
4577 | if (can_invalidate_dentries && | |
b3b6e05e | 4578 | dn->dir->parent_inode->ino == CEPH_INO_ROOT) { |
7c673cae FG |
4579 | // Only issue one of these per DN for inodes in root: handle |
4580 | // others more efficiently by calling for root-child DNs at | |
4581 | // the end of this function. | |
4582 | _schedule_invalidate_dentry_callback(dn, true); | |
4583 | } | |
28e407b8 AA |
4584 | ldout(cct, 20) << " queueing dentry for trimming: " << dn->name << dendl; |
4585 | to_trim.insert(dn); | |
7c673cae FG |
4586 | } else { |
4587 | ldout(cct, 20) << " not expirable: " << dn->name << dendl; | |
4588 | all = false; | |
4589 | } | |
4590 | } | |
b3b6e05e | 4591 | if (in->ll_ref == 1 && in->ino != CEPH_INO_ROOT) { |
f91f0fd5 TL |
4592 | _schedule_ino_release_callback(in.get()); |
4593 | } | |
b3b6e05e | 4594 | if (all && in->ino != CEPH_INO_ROOT) { |
7c673cae FG |
4595 | ldout(cct, 20) << __func__ << " counting as trimmed: " << *in << dendl; |
4596 | trimmed++; | |
4597 | } | |
4598 | } | |
4599 | } | |
28e407b8 AA |
4600 | ldout(cct, 20) << " trimming queued dentries: " << dendl; |
4601 | for (const auto &dn : to_trim) { | |
4602 | trim_dentry(dn); | |
4603 | } | |
4604 | to_trim.clear(); | |
7c673cae | 4605 | |
b32b8144 | 4606 | caps_size = s->caps.size(); |
11fdf7f2 | 4607 | if (caps_size > (size_t)max) |
7c673cae FG |
4608 | _invalidate_kernel_dcache(); |
4609 | } | |
4610 | ||
4611 | void Client::force_session_readonly(MetaSession *s) | |
4612 | { | |
4613 | s->readonly = true; | |
4614 | for (xlist<Cap*>::iterator p = s->caps.begin(); !p.end(); ++p) { | |
11fdf7f2 TL |
4615 | auto &in = (*p)->inode; |
4616 | if (in.caps_wanted() & CEPH_CAP_FILE_WR) | |
4617 | signal_cond_list(in.waitfor_caps); | |
7c673cae FG |
4618 | } |
4619 | } | |
4620 | ||
7c673cae FG |
4621 | int Client::mark_caps_flushing(Inode *in, ceph_tid_t* ptid) |
4622 | { | |
4623 | MetaSession *session = in->auth_cap->session; | |
4624 | ||
4625 | int flushing = in->dirty_caps; | |
11fdf7f2 | 4626 | ceph_assert(flushing); |
7c673cae FG |
4627 | |
4628 | ceph_tid_t flush_tid = ++last_flush_tid; | |
4629 | in->flushing_cap_tids[flush_tid] = flushing; | |
4630 | ||
4631 | if (!in->flushing_caps) { | |
11fdf7f2 | 4632 | ldout(cct, 10) << __func__ << " " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4633 | num_flushing_caps++; |
4634 | } else { | |
11fdf7f2 | 4635 | ldout(cct, 10) << __func__ << " (more) " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4636 | } |
4637 | ||
4638 | in->flushing_caps |= flushing; | |
28e407b8 | 4639 | in->mark_caps_clean(); |
7c673cae FG |
4640 | |
4641 | if (!in->flushing_cap_item.is_on_list()) | |
4642 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
4643 | session->flushing_caps_tids.insert(flush_tid); | |
4644 | ||
4645 | *ptid = flush_tid; | |
4646 | return flushing; | |
4647 | } | |
4648 | ||
4649 | void Client::adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s) | |
4650 | { | |
4651 | for (auto &p : in->cap_snaps) { | |
4652 | CapSnap &capsnap = p.second; | |
4653 | if (capsnap.flush_tid > 0) { | |
4654 | old_s->flushing_caps_tids.erase(capsnap.flush_tid); | |
4655 | new_s->flushing_caps_tids.insert(capsnap.flush_tid); | |
4656 | } | |
4657 | } | |
4658 | for (map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4659 | it != in->flushing_cap_tids.end(); | |
4660 | ++it) { | |
4661 | old_s->flushing_caps_tids.erase(it->first); | |
4662 | new_s->flushing_caps_tids.insert(it->first); | |
4663 | } | |
4664 | new_s->flushing_caps.push_back(&in->flushing_cap_item); | |
4665 | } | |
4666 | ||
4667 | /* | |
20effc67 TL |
4668 | * Flush all the dirty caps back to the MDS. Because the callers |
4669 | * generally wait on the result of this function (syncfs and umount | |
4670 | * cases), we set CHECK_CAPS_SYNCHRONOUS on the last check_caps call. | |
7c673cae FG |
4671 | */ |
4672 | void Client::flush_caps_sync() | |
4673 | { | |
4674 | ldout(cct, 10) << __func__ << dendl; | |
20effc67 TL |
4675 | for (auto &q : mds_sessions) { |
4676 | auto s = q.second; | |
4677 | xlist<Inode*>::iterator p = s->dirty_list.begin(); | |
4678 | while (!p.end()) { | |
4679 | unsigned flags = CHECK_CAPS_NODELAY; | |
4680 | Inode *in = *p; | |
7c673cae | 4681 | |
20effc67 TL |
4682 | ++p; |
4683 | if (p.end()) | |
4684 | flags |= CHECK_CAPS_SYNCHRONOUS; | |
4685 | check_caps(in, flags); | |
4686 | } | |
7c673cae FG |
4687 | } |
4688 | } | |
4689 | ||
7c673cae FG |
4690 | void Client::wait_sync_caps(Inode *in, ceph_tid_t want) |
4691 | { | |
4692 | while (in->flushing_caps) { | |
4693 | map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
11fdf7f2 | 4694 | ceph_assert(it != in->flushing_cap_tids.end()); |
7c673cae FG |
4695 | if (it->first > want) |
4696 | break; | |
11fdf7f2 | 4697 | ldout(cct, 10) << __func__ << " on " << *in << " flushing " |
7c673cae FG |
4698 | << ccap_string(it->second) << " want " << want |
4699 | << " last " << it->first << dendl; | |
4700 | wait_on_list(in->waitfor_caps); | |
4701 | } | |
4702 | } | |
4703 | ||
4704 | void Client::wait_sync_caps(ceph_tid_t want) | |
4705 | { | |
4706 | retry: | |
11fdf7f2 | 4707 | ldout(cct, 10) << __func__ << " want " << want << " (last is " << last_flush_tid << ", " |
7c673cae | 4708 | << num_flushing_caps << " total flushing)" << dendl; |
11fdf7f2 | 4709 | for (auto &p : mds_sessions) { |
20effc67 | 4710 | auto s = p.second; |
7c673cae FG |
4711 | if (s->flushing_caps_tids.empty()) |
4712 | continue; | |
4713 | ceph_tid_t oldest_tid = *s->flushing_caps_tids.begin(); | |
4714 | if (oldest_tid <= want) { | |
11fdf7f2 | 4715 | ldout(cct, 10) << " waiting on mds." << p.first << " tid " << oldest_tid |
7c673cae | 4716 | << " (want " << want << ")" << dendl; |
9f95a23c TL |
4717 | std::unique_lock l{client_lock, std::adopt_lock}; |
4718 | sync_cond.wait(l); | |
4719 | l.release(); | |
7c673cae FG |
4720 | goto retry; |
4721 | } | |
4722 | } | |
4723 | } | |
4724 | ||
eafe8130 TL |
4725 | void Client::kick_flushing_caps(Inode *in, MetaSession *session) |
4726 | { | |
4727 | in->flags &= ~I_KICK_FLUSH; | |
4728 | ||
4729 | Cap *cap = in->auth_cap; | |
4730 | ceph_assert(cap->session == session); | |
4731 | ||
4732 | ceph_tid_t last_snap_flush = 0; | |
4733 | for (auto p = in->flushing_cap_tids.rbegin(); | |
4734 | p != in->flushing_cap_tids.rend(); | |
4735 | ++p) { | |
4736 | if (!p->second) { | |
4737 | last_snap_flush = p->first; | |
4738 | break; | |
4739 | } | |
4740 | } | |
4741 | ||
4742 | int wanted = in->caps_wanted(); | |
4743 | int used = get_caps_used(in) | in->caps_dirty(); | |
4744 | auto it = in->cap_snaps.begin(); | |
4745 | for (auto& p : in->flushing_cap_tids) { | |
4746 | if (p.second) { | |
4747 | int msg_flags = p.first < last_snap_flush ? MClientCaps::FLAG_PENDING_CAPSNAP : 0; | |
4748 | send_cap(in, session, cap, msg_flags, used, wanted, (cap->issued | cap->implemented), | |
4749 | p.second, p.first); | |
4750 | } else { | |
4751 | ceph_assert(it != in->cap_snaps.end()); | |
4752 | ceph_assert(it->second.flush_tid == p.first); | |
4753 | send_flush_snap(in, session, it->first, it->second); | |
4754 | ++it; | |
4755 | } | |
4756 | } | |
4757 | } | |
4758 | ||
7c673cae FG |
4759 | void Client::kick_flushing_caps(MetaSession *session) |
4760 | { | |
4761 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 4762 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
4763 | |
4764 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { | |
4765 | Inode *in = *p; | |
eafe8130 TL |
4766 | if (in->flags & I_KICK_FLUSH) { |
4767 | ldout(cct, 20) << " reflushing caps on " << *in << " to mds." << mds << dendl; | |
4768 | kick_flushing_caps(in, session); | |
4769 | } | |
7c673cae | 4770 | } |
7c673cae FG |
4771 | } |
4772 | ||
4773 | void Client::early_kick_flushing_caps(MetaSession *session) | |
4774 | { | |
7c673cae FG |
4775 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { |
4776 | Inode *in = *p; | |
11fdf7f2 TL |
4777 | Cap *cap = in->auth_cap; |
4778 | ceph_assert(cap); | |
7c673cae FG |
4779 | |
4780 | // if flushing caps were revoked, we re-send the cap flush in client reconnect | |
4781 | // stage. This guarantees that MDS processes the cap flush message before issuing | |
4782 | // the flushing caps to other client. | |
eafe8130 TL |
4783 | if ((in->flushing_caps & in->auth_cap->issued) == in->flushing_caps) { |
4784 | in->flags |= I_KICK_FLUSH; | |
7c673cae | 4785 | continue; |
eafe8130 | 4786 | } |
7c673cae FG |
4787 | |
4788 | ldout(cct, 20) << " reflushing caps (early_kick) on " << *in | |
4789 | << " to mds." << session->mds_num << dendl; | |
11fdf7f2 TL |
4790 | // send_reconnect() also will reset these sequence numbers. make sure |
4791 | // sequence numbers in cap flush message match later reconnect message. | |
4792 | cap->seq = 0; | |
4793 | cap->issue_seq = 0; | |
4794 | cap->mseq = 0; | |
4795 | cap->issued = cap->implemented; | |
4796 | ||
eafe8130 | 4797 | kick_flushing_caps(in, session); |
7c673cae FG |
4798 | } |
4799 | } | |
4800 | ||
7c673cae FG |
4801 | void Client::invalidate_snaprealm_and_children(SnapRealm *realm) |
4802 | { | |
4803 | list<SnapRealm*> q; | |
4804 | q.push_back(realm); | |
4805 | ||
4806 | while (!q.empty()) { | |
4807 | realm = q.front(); | |
4808 | q.pop_front(); | |
4809 | ||
11fdf7f2 | 4810 | ldout(cct, 10) << __func__ << " " << *realm << dendl; |
7c673cae FG |
4811 | realm->invalidate_cache(); |
4812 | ||
4813 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4814 | p != realm->pchildren.end(); | |
4815 | ++p) | |
4816 | q.push_back(*p); | |
4817 | } | |
4818 | } | |
4819 | ||
4820 | SnapRealm *Client::get_snap_realm(inodeno_t r) | |
4821 | { | |
4822 | SnapRealm *realm = snap_realms[r]; | |
4823 | if (!realm) | |
4824 | snap_realms[r] = realm = new SnapRealm(r); | |
11fdf7f2 | 4825 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4826 | realm->nref++; |
4827 | return realm; | |
4828 | } | |
4829 | ||
4830 | SnapRealm *Client::get_snap_realm_maybe(inodeno_t r) | |
4831 | { | |
4832 | if (snap_realms.count(r) == 0) { | |
11fdf7f2 | 4833 | ldout(cct, 20) << __func__ << " " << r << " fail" << dendl; |
7c673cae FG |
4834 | return NULL; |
4835 | } | |
4836 | SnapRealm *realm = snap_realms[r]; | |
11fdf7f2 | 4837 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4838 | realm->nref++; |
4839 | return realm; | |
4840 | } | |
4841 | ||
4842 | void Client::put_snap_realm(SnapRealm *realm) | |
4843 | { | |
11fdf7f2 | 4844 | ldout(cct, 20) << __func__ << " " << realm->ino << " " << realm |
7c673cae FG |
4845 | << " " << realm->nref << " -> " << (realm->nref - 1) << dendl; |
4846 | if (--realm->nref == 0) { | |
4847 | snap_realms.erase(realm->ino); | |
4848 | if (realm->pparent) { | |
4849 | realm->pparent->pchildren.erase(realm); | |
4850 | put_snap_realm(realm->pparent); | |
4851 | } | |
4852 | delete realm; | |
4853 | } | |
4854 | } | |
4855 | ||
4856 | bool Client::adjust_realm_parent(SnapRealm *realm, inodeno_t parent) | |
4857 | { | |
4858 | if (realm->parent != parent) { | |
11fdf7f2 | 4859 | ldout(cct, 10) << __func__ << " " << *realm |
7c673cae FG |
4860 | << " " << realm->parent << " -> " << parent << dendl; |
4861 | realm->parent = parent; | |
4862 | if (realm->pparent) { | |
4863 | realm->pparent->pchildren.erase(realm); | |
4864 | put_snap_realm(realm->pparent); | |
4865 | } | |
4866 | realm->pparent = get_snap_realm(parent); | |
4867 | realm->pparent->pchildren.insert(realm); | |
4868 | return true; | |
4869 | } | |
4870 | return false; | |
4871 | } | |
4872 | ||
4873 | static bool has_new_snaps(const SnapContext& old_snapc, | |
4874 | const SnapContext& new_snapc) | |
4875 | { | |
4876 | return !new_snapc.snaps.empty() && new_snapc.snaps[0] > old_snapc.seq; | |
4877 | } | |
4878 | ||
4879 | ||
11fdf7f2 | 4880 | void Client::update_snap_trace(const bufferlist& bl, SnapRealm **realm_ret, bool flush) |
7c673cae FG |
4881 | { |
4882 | SnapRealm *first_realm = NULL; | |
11fdf7f2 | 4883 | ldout(cct, 10) << __func__ << " len " << bl.length() << dendl; |
7c673cae FG |
4884 | |
4885 | map<SnapRealm*, SnapContext> dirty_realms; | |
4886 | ||
11fdf7f2 | 4887 | auto p = bl.cbegin(); |
7c673cae FG |
4888 | while (!p.end()) { |
4889 | SnapRealmInfo info; | |
11fdf7f2 | 4890 | decode(info, p); |
7c673cae FG |
4891 | SnapRealm *realm = get_snap_realm(info.ino()); |
4892 | ||
4893 | bool invalidate = false; | |
4894 | ||
4895 | if (info.seq() > realm->seq) { | |
11fdf7f2 | 4896 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() << " > " << realm->seq |
7c673cae FG |
4897 | << dendl; |
4898 | ||
4899 | if (flush) { | |
4900 | // writeback any dirty caps _before_ updating snap list (i.e. with old snap info) | |
4901 | // flush me + children | |
4902 | list<SnapRealm*> q; | |
4903 | q.push_back(realm); | |
4904 | while (!q.empty()) { | |
4905 | SnapRealm *realm = q.front(); | |
4906 | q.pop_front(); | |
4907 | ||
4908 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4909 | p != realm->pchildren.end(); | |
4910 | ++p) | |
4911 | q.push_back(*p); | |
4912 | ||
4913 | if (dirty_realms.count(realm) == 0) { | |
4914 | realm->nref++; | |
4915 | dirty_realms[realm] = realm->get_snap_context(); | |
4916 | } | |
4917 | } | |
4918 | } | |
4919 | ||
4920 | // update | |
4921 | realm->seq = info.seq(); | |
4922 | realm->created = info.created(); | |
4923 | realm->parent_since = info.parent_since(); | |
4924 | realm->prior_parent_snaps = info.prior_parent_snaps; | |
4925 | realm->my_snaps = info.my_snaps; | |
4926 | invalidate = true; | |
4927 | } | |
4928 | ||
4929 | // _always_ verify parent | |
4930 | if (adjust_realm_parent(realm, info.parent())) | |
4931 | invalidate = true; | |
4932 | ||
4933 | if (invalidate) { | |
4934 | invalidate_snaprealm_and_children(realm); | |
11fdf7f2 | 4935 | ldout(cct, 15) << __func__ << " " << *realm << " self|parent updated" << dendl; |
7c673cae FG |
4936 | ldout(cct, 15) << " snapc " << realm->get_snap_context() << dendl; |
4937 | } else { | |
11fdf7f2 | 4938 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() |
7c673cae FG |
4939 | << " <= " << realm->seq << " and same parent, SKIPPING" << dendl; |
4940 | } | |
f67539c2 | 4941 | |
7c673cae FG |
4942 | if (!first_realm) |
4943 | first_realm = realm; | |
4944 | else | |
4945 | put_snap_realm(realm); | |
4946 | } | |
4947 | ||
f67539c2 | 4948 | for (auto &[realm, snapc] : dirty_realms) { |
7c673cae | 4949 | // if there are new snaps ? |
f67539c2 | 4950 | if (has_new_snaps(snapc, realm->get_snap_context())) { |
7c673cae | 4951 | ldout(cct, 10) << " flushing caps on " << *realm << dendl; |
f67539c2 TL |
4952 | for (auto&& in : realm->inodes_with_caps) { |
4953 | queue_cap_snap(in, snapc); | |
7c673cae FG |
4954 | } |
4955 | } else { | |
4956 | ldout(cct, 10) << " no new snap on " << *realm << dendl; | |
4957 | } | |
4958 | put_snap_realm(realm); | |
4959 | } | |
4960 | ||
4961 | if (realm_ret) | |
4962 | *realm_ret = first_realm; | |
4963 | else | |
4964 | put_snap_realm(first_realm); | |
4965 | } | |
4966 | ||
11fdf7f2 | 4967 | void Client::handle_snap(const MConstRef<MClientSnap>& m) |
7c673cae | 4968 | { |
11fdf7f2 | 4969 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 4970 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
4971 | |
4972 | std::scoped_lock cl(client_lock); | |
20effc67 | 4973 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 4974 | if (!session) { |
7c673cae FG |
4975 | return; |
4976 | } | |
4977 | ||
20effc67 | 4978 | got_mds_push(session.get()); |
7c673cae FG |
4979 | |
4980 | map<Inode*, SnapContext> to_move; | |
4981 | SnapRealm *realm = 0; | |
4982 | ||
4983 | if (m->head.op == CEPH_SNAP_OP_SPLIT) { | |
11fdf7f2 | 4984 | ceph_assert(m->head.split); |
7c673cae | 4985 | SnapRealmInfo info; |
11fdf7f2 TL |
4986 | auto p = m->bl.cbegin(); |
4987 | decode(info, p); | |
4988 | ceph_assert(info.ino() == m->head.split); | |
7c673cae FG |
4989 | |
4990 | // flush, then move, ino's. | |
4991 | realm = get_snap_realm(info.ino()); | |
4992 | ldout(cct, 10) << " splitting off " << *realm << dendl; | |
11fdf7f2 TL |
4993 | for (auto& ino : m->split_inos) { |
4994 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
7c673cae FG |
4995 | if (inode_map.count(vino)) { |
4996 | Inode *in = inode_map[vino]; | |
4997 | if (!in->snaprealm || in->snaprealm == realm) | |
4998 | continue; | |
4999 | if (in->snaprealm->created > info.created()) { | |
5000 | ldout(cct, 10) << " NOT moving " << *in << " from _newer_ realm " | |
5001 | << *in->snaprealm << dendl; | |
5002 | continue; | |
5003 | } | |
5004 | ldout(cct, 10) << " moving " << *in << " from " << *in->snaprealm << dendl; | |
5005 | ||
5006 | ||
5007 | in->snaprealm_item.remove_myself(); | |
5008 | to_move[in] = in->snaprealm->get_snap_context(); | |
5009 | put_snap_realm(in->snaprealm); | |
5010 | } | |
5011 | } | |
5012 | ||
5013 | // move child snaprealms, too | |
11fdf7f2 TL |
5014 | for (auto& child_realm : m->split_realms) { |
5015 | ldout(cct, 10) << "adjusting snaprealm " << child_realm << " parent" << dendl; | |
5016 | SnapRealm *child = get_snap_realm_maybe(child_realm); | |
7c673cae FG |
5017 | if (!child) |
5018 | continue; | |
5019 | adjust_realm_parent(child, realm->ino); | |
5020 | put_snap_realm(child); | |
5021 | } | |
5022 | } | |
5023 | ||
5024 | update_snap_trace(m->bl, NULL, m->head.op != CEPH_SNAP_OP_DESTROY); | |
5025 | ||
5026 | if (realm) { | |
5027 | for (auto p = to_move.begin(); p != to_move.end(); ++p) { | |
5028 | Inode *in = p->first; | |
5029 | in->snaprealm = realm; | |
5030 | realm->inodes_with_caps.push_back(&in->snaprealm_item); | |
5031 | realm->nref++; | |
5032 | // queue for snap writeback | |
5033 | if (has_new_snaps(p->second, realm->get_snap_context())) | |
5034 | queue_cap_snap(in, p->second); | |
5035 | } | |
5036 | put_snap_realm(realm); | |
5037 | } | |
7c673cae FG |
5038 | } |
5039 | ||
11fdf7f2 | 5040 | void Client::handle_quota(const MConstRef<MClientQuota>& m) |
7c673cae FG |
5041 | { |
5042 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5043 | |
5044 | std::scoped_lock cl(client_lock); | |
20effc67 | 5045 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 5046 | if (!session) { |
7c673cae FG |
5047 | return; |
5048 | } | |
5049 | ||
20effc67 | 5050 | got_mds_push(session.get()); |
7c673cae | 5051 | |
11fdf7f2 | 5052 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << mds << dendl; |
7c673cae FG |
5053 | |
5054 | vinodeno_t vino(m->ino, CEPH_NOSNAP); | |
5055 | if (inode_map.count(vino)) { | |
5056 | Inode *in = NULL; | |
5057 | in = inode_map[vino]; | |
5058 | ||
5059 | if (in) { | |
5060 | in->quota = m->quota; | |
5061 | in->rstat = m->rstat; | |
5062 | } | |
5063 | } | |
7c673cae FG |
5064 | } |
5065 | ||
11fdf7f2 | 5066 | void Client::handle_caps(const MConstRef<MClientCaps>& m) |
7c673cae FG |
5067 | { |
5068 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5069 | |
5070 | std::scoped_lock cl(client_lock); | |
20effc67 | 5071 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 5072 | if (!session) { |
7c673cae FG |
5073 | return; |
5074 | } | |
5075 | ||
5076 | if (m->osd_epoch_barrier && !objecter->have_map(m->osd_epoch_barrier)) { | |
5077 | // Pause RADOS operations until we see the required epoch | |
5078 | objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
5079 | } | |
5080 | ||
5081 | if (m->osd_epoch_barrier > cap_epoch_barrier) { | |
5082 | // Record the barrier so that we will transmit it to MDS when releasing | |
5083 | set_cap_epoch_barrier(m->osd_epoch_barrier); | |
5084 | } | |
5085 | ||
20effc67 | 5086 | got_mds_push(session.get()); |
7c673cae | 5087 | |
11fdf7f2 | 5088 | Inode *in; |
7c673cae | 5089 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); |
11fdf7f2 TL |
5090 | if (auto it = inode_map.find(vino); it != inode_map.end()) { |
5091 | in = it->second; | |
5092 | } else { | |
7c673cae | 5093 | if (m->get_op() == CEPH_CAP_OP_IMPORT) { |
11fdf7f2 | 5094 | ldout(cct, 5) << __func__ << " don't have vino " << vino << " on IMPORT, immediately releasing" << dendl; |
7c673cae FG |
5095 | session->enqueue_cap_release( |
5096 | m->get_ino(), | |
5097 | m->get_cap_id(), | |
5098 | m->get_seq(), | |
5099 | m->get_mseq(), | |
5100 | cap_epoch_barrier); | |
5101 | } else { | |
11fdf7f2 | 5102 | ldout(cct, 5) << __func__ << " don't have vino " << vino << ", dropping" << dendl; |
7c673cae | 5103 | } |
7c673cae FG |
5104 | |
5105 | // in case the mds is waiting on e.g. a revocation | |
5106 | flush_cap_releases(); | |
5107 | return; | |
5108 | } | |
5109 | ||
5110 | switch (m->get_op()) { | |
20effc67 TL |
5111 | case CEPH_CAP_OP_EXPORT: return handle_cap_export(session.get(), in, m); |
5112 | case CEPH_CAP_OP_FLUSHSNAP_ACK: return handle_cap_flushsnap_ack(session.get(), in, m); | |
5113 | case CEPH_CAP_OP_IMPORT: /* no return */ handle_cap_import(session.get(), in, m); | |
7c673cae FG |
5114 | } |
5115 | ||
11fdf7f2 TL |
5116 | if (auto it = in->caps.find(mds); it != in->caps.end()) { |
5117 | Cap &cap = in->caps.at(mds); | |
7c673cae | 5118 | |
11fdf7f2 | 5119 | switch (m->get_op()) { |
20effc67 | 5120 | case CEPH_CAP_OP_TRUNC: return handle_cap_trunc(session.get(), in, m); |
11fdf7f2 TL |
5121 | case CEPH_CAP_OP_IMPORT: |
5122 | case CEPH_CAP_OP_REVOKE: | |
20effc67 TL |
5123 | case CEPH_CAP_OP_GRANT: return handle_cap_grant(session.get(), in, &cap, m); |
5124 | case CEPH_CAP_OP_FLUSH_ACK: return handle_cap_flush_ack(session.get(), in, &cap, m); | |
11fdf7f2 TL |
5125 | } |
5126 | } else { | |
5127 | ldout(cct, 5) << __func__ << " don't have " << *in << " cap on mds." << mds << dendl; | |
5128 | return; | |
7c673cae FG |
5129 | } |
5130 | } | |
5131 | ||
11fdf7f2 | 5132 | void Client::handle_cap_import(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5133 | { |
5134 | mds_rank_t mds = session->mds_num; | |
5135 | ||
11fdf7f2 | 5136 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5137 | << " IMPORT from mds." << mds << dendl; |
5138 | ||
5139 | const mds_rank_t peer_mds = mds_rank_t(m->peer.mds); | |
5140 | Cap *cap = NULL; | |
5141 | UserPerm cap_perms; | |
11fdf7f2 TL |
5142 | if (auto it = in->caps.find(peer_mds); m->peer.cap_id && it != in->caps.end()) { |
5143 | cap = &it->second; | |
5144 | cap_perms = cap->latest_perms; | |
7c673cae FG |
5145 | } |
5146 | ||
5147 | // add/update it | |
5148 | SnapRealm *realm = NULL; | |
5149 | update_snap_trace(m->snapbl, &realm); | |
5150 | ||
1911f103 TL |
5151 | int issued = m->get_caps(); |
5152 | int wanted = m->get_wanted(); | |
7c673cae | 5153 | add_update_cap(in, session, m->get_cap_id(), |
1911f103 | 5154 | issued, wanted, m->get_seq(), m->get_mseq(), |
a8e16298 | 5155 | m->get_realm(), CEPH_CAP_FLAG_AUTH, cap_perms); |
7c673cae FG |
5156 | |
5157 | if (cap && cap->cap_id == m->peer.cap_id) { | |
5158 | remove_cap(cap, (m->peer.flags & CEPH_CAP_FLAG_RELEASE)); | |
5159 | } | |
5160 | ||
5161 | if (realm) | |
5162 | put_snap_realm(realm); | |
5163 | ||
eafe8130 | 5164 | if (in->auth_cap && in->auth_cap->session == session) { |
1911f103 TL |
5165 | if (!(wanted & CEPH_CAP_ANY_FILE_WR) || |
5166 | in->requested_max_size > m->get_max_size()) { | |
5167 | in->requested_max_size = 0; | |
5168 | ldout(cct, 15) << "reset requested_max_size after cap import" << dendl; | |
5169 | } | |
7c673cae | 5170 | // reflush any/all caps (if we are now the auth_cap) |
eafe8130 | 5171 | kick_flushing_caps(in, session); |
7c673cae FG |
5172 | } |
5173 | } | |
5174 | ||
11fdf7f2 | 5175 | void Client::handle_cap_export(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5176 | { |
5177 | mds_rank_t mds = session->mds_num; | |
5178 | ||
11fdf7f2 | 5179 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5180 | << " EXPORT from mds." << mds << dendl; |
5181 | ||
11fdf7f2 TL |
5182 | auto it = in->caps.find(mds); |
5183 | if (it != in->caps.end()) { | |
5184 | Cap &cap = it->second; | |
5185 | if (cap.cap_id == m->get_cap_id()) { | |
5186 | if (m->peer.cap_id) { | |
5187 | const auto peer_mds = mds_rank_t(m->peer.mds); | |
20effc67 | 5188 | auto tsession = _get_or_open_mds_session(peer_mds); |
11fdf7f2 TL |
5189 | auto it = in->caps.find(peer_mds); |
5190 | if (it != in->caps.end()) { | |
5191 | Cap &tcap = it->second; | |
5192 | if (tcap.cap_id == m->peer.cap_id && | |
5193 | ceph_seq_cmp(tcap.seq, m->peer.seq) < 0) { | |
5194 | tcap.cap_id = m->peer.cap_id; | |
5195 | tcap.seq = m->peer.seq - 1; | |
5196 | tcap.issue_seq = tcap.seq; | |
5197 | tcap.issued |= cap.issued; | |
5198 | tcap.implemented |= cap.issued; | |
5199 | if (&cap == in->auth_cap) | |
5200 | in->auth_cap = &tcap; | |
5201 | if (in->auth_cap == &tcap && in->flushing_cap_item.is_on_list()) | |
20effc67 | 5202 | adjust_session_flushing_caps(in, session, tsession.get()); |
11fdf7f2 TL |
5203 | } |
5204 | } else { | |
20effc67 | 5205 | add_update_cap(in, tsession.get(), m->peer.cap_id, cap.issued, 0, |
11fdf7f2 TL |
5206 | m->peer.seq - 1, m->peer.mseq, (uint64_t)-1, |
5207 | &cap == in->auth_cap ? CEPH_CAP_FLAG_AUTH : 0, | |
5208 | cap.latest_perms); | |
5209 | } | |
7c673cae | 5210 | } else { |
11fdf7f2 TL |
5211 | if (cap.wanted | cap.issued) |
5212 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 5213 | } |
7c673cae | 5214 | |
11fdf7f2 TL |
5215 | remove_cap(&cap, false); |
5216 | } | |
7c673cae | 5217 | } |
7c673cae FG |
5218 | } |
5219 | ||
11fdf7f2 | 5220 | void Client::handle_cap_trunc(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5221 | { |
5222 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 5223 | ceph_assert(in->caps.count(mds)); |
7c673cae | 5224 | |
11fdf7f2 | 5225 | ldout(cct, 10) << __func__ << " on ino " << *in |
7c673cae FG |
5226 | << " size " << in->size << " -> " << m->get_size() |
5227 | << dendl; | |
5228 | ||
1adf2230 AA |
5229 | int issued; |
5230 | in->caps_issued(&issued); | |
5231 | issued |= in->caps_dirty(); | |
5232 | update_inode_file_size(in, issued, m->get_size(), | |
5233 | m->get_truncate_seq(), m->get_truncate_size()); | |
7c673cae FG |
5234 | } |
5235 | ||
11fdf7f2 | 5236 | void Client::handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5237 | { |
5238 | ceph_tid_t flush_ack_tid = m->get_client_tid(); | |
5239 | int dirty = m->get_dirty(); | |
5240 | int cleaned = 0; | |
5241 | int flushed = 0; | |
5242 | ||
11fdf7f2 TL |
5243 | auto it = in->flushing_cap_tids.begin(); |
5244 | if (it->first < flush_ack_tid) { | |
5245 | ldout(cct, 0) << __func__ << " mds." << session->mds_num | |
5246 | << " got unexpected flush ack tid " << flush_ack_tid | |
5247 | << " expected is " << it->first << dendl; | |
5248 | } | |
5249 | for (; it != in->flushing_cap_tids.end(); ) { | |
eafe8130 TL |
5250 | if (!it->second) { |
5251 | // cap snap | |
5252 | ++it; | |
5253 | continue; | |
5254 | } | |
7c673cae FG |
5255 | if (it->first == flush_ack_tid) |
5256 | cleaned = it->second; | |
5257 | if (it->first <= flush_ack_tid) { | |
5258 | session->flushing_caps_tids.erase(it->first); | |
5259 | in->flushing_cap_tids.erase(it++); | |
5260 | ++flushed; | |
5261 | continue; | |
5262 | } | |
5263 | cleaned &= ~it->second; | |
5264 | if (!cleaned) | |
5265 | break; | |
5266 | ++it; | |
5267 | } | |
5268 | ||
11fdf7f2 | 5269 | ldout(cct, 5) << __func__ << " mds." << session->mds_num |
7c673cae FG |
5270 | << " cleaned " << ccap_string(cleaned) << " on " << *in |
5271 | << " with " << ccap_string(dirty) << dendl; | |
5272 | ||
5273 | if (flushed) { | |
5274 | signal_cond_list(in->waitfor_caps); | |
5275 | if (session->flushing_caps_tids.empty() || | |
5276 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5277 | sync_cond.notify_all(); |
7c673cae FG |
5278 | } |
5279 | ||
5280 | if (!dirty) { | |
5281 | in->cap_dirtier_uid = -1; | |
5282 | in->cap_dirtier_gid = -1; | |
5283 | } | |
5284 | ||
5285 | if (!cleaned) { | |
5286 | ldout(cct, 10) << " tid " << m->get_client_tid() << " != any cap bit tids" << dendl; | |
5287 | } else { | |
5288 | if (in->flushing_caps) { | |
5289 | ldout(cct, 5) << " flushing_caps " << ccap_string(in->flushing_caps) | |
5290 | << " -> " << ccap_string(in->flushing_caps & ~cleaned) << dendl; | |
5291 | in->flushing_caps &= ~cleaned; | |
5292 | if (in->flushing_caps == 0) { | |
5293 | ldout(cct, 10) << " " << *in << " !flushing" << dendl; | |
5294 | num_flushing_caps--; | |
eafe8130 | 5295 | if (in->flushing_cap_tids.empty()) |
7c673cae FG |
5296 | in->flushing_cap_item.remove_myself(); |
5297 | } | |
5298 | if (!in->caps_dirty()) | |
5299 | put_inode(in); | |
5300 | } | |
5301 | } | |
7c673cae FG |
5302 | } |
5303 | ||
5304 | ||
11fdf7f2 | 5305 | void Client::handle_cap_flushsnap_ack(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae | 5306 | { |
eafe8130 | 5307 | ceph_tid_t flush_ack_tid = m->get_client_tid(); |
7c673cae | 5308 | mds_rank_t mds = session->mds_num; |
11fdf7f2 | 5309 | ceph_assert(in->caps.count(mds)); |
7c673cae FG |
5310 | snapid_t follows = m->get_snap_follows(); |
5311 | ||
11fdf7f2 TL |
5312 | if (auto it = in->cap_snaps.find(follows); it != in->cap_snaps.end()) { |
5313 | auto& capsnap = it->second; | |
eafe8130 TL |
5314 | if (flush_ack_tid != capsnap.flush_tid) { |
5315 | ldout(cct, 10) << " tid " << flush_ack_tid << " != " << capsnap.flush_tid << dendl; | |
7c673cae | 5316 | } else { |
eafe8130 | 5317 | InodeRef tmp_ref(in); |
11fdf7f2 | 5318 | ldout(cct, 5) << __func__ << " mds." << mds << " flushed snap follows " << follows |
7c673cae | 5319 | << " on " << *in << dendl; |
7c673cae | 5320 | session->flushing_caps_tids.erase(capsnap.flush_tid); |
eafe8130 TL |
5321 | in->flushing_cap_tids.erase(capsnap.flush_tid); |
5322 | if (in->flushing_caps == 0 && in->flushing_cap_tids.empty()) | |
5323 | in->flushing_cap_item.remove_myself(); | |
11fdf7f2 | 5324 | in->cap_snaps.erase(it); |
eafe8130 TL |
5325 | |
5326 | signal_cond_list(in->waitfor_caps); | |
5327 | if (session->flushing_caps_tids.empty() || | |
5328 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5329 | sync_cond.notify_all(); |
7c673cae FG |
5330 | } |
5331 | } else { | |
11fdf7f2 | 5332 | ldout(cct, 5) << __func__ << " DUP(?) mds." << mds << " flushed snap follows " << follows |
7c673cae FG |
5333 | << " on " << *in << dendl; |
5334 | // we may not have it if we send multiple FLUSHSNAP requests and (got multiple FLUSHEDSNAPs back) | |
5335 | } | |
7c673cae FG |
5336 | } |
5337 | ||
5338 | class C_Client_DentryInvalidate : public Context { | |
5339 | private: | |
5340 | Client *client; | |
5341 | vinodeno_t dirino; | |
5342 | vinodeno_t ino; | |
5343 | string name; | |
5344 | public: | |
5345 | C_Client_DentryInvalidate(Client *c, Dentry *dn, bool del) : | |
5346 | client(c), name(dn->name) { | |
5347 | if (client->use_faked_inos()) { | |
5348 | dirino.ino = dn->dir->parent_inode->faked_ino; | |
5349 | if (del) | |
5350 | ino.ino = dn->inode->faked_ino; | |
5351 | } else { | |
5352 | dirino = dn->dir->parent_inode->vino(); | |
5353 | if (del) | |
5354 | ino = dn->inode->vino(); | |
5355 | } | |
5356 | if (!del) | |
5357 | ino.ino = inodeno_t(); | |
5358 | } | |
5359 | void finish(int r) override { | |
5360 | // _async_dentry_invalidate is responsible for its own locking | |
9f95a23c | 5361 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
5362 | client->_async_dentry_invalidate(dirino, ino, name); |
5363 | } | |
5364 | }; | |
5365 | ||
5366 | void Client::_async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name) | |
5367 | { | |
f67539c2 TL |
5368 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
5369 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 5370 | return; |
f67539c2 | 5371 | |
11fdf7f2 | 5372 | ldout(cct, 10) << __func__ << " '" << name << "' ino " << ino |
7c673cae | 5373 | << " in dir " << dirino << dendl; |
e306af50 | 5374 | dentry_invalidate_cb(callback_handle, dirino, ino, name.c_str(), name.length()); |
7c673cae FG |
5375 | } |
5376 | ||
5377 | void Client::_schedule_invalidate_dentry_callback(Dentry *dn, bool del) | |
5378 | { | |
5379 | if (dentry_invalidate_cb && dn->inode->ll_ref > 0) | |
5380 | async_dentry_invalidator.queue(new C_Client_DentryInvalidate(this, dn, del)); | |
5381 | } | |
5382 | ||
5383 | void Client::_try_to_trim_inode(Inode *in, bool sched_inval) | |
5384 | { | |
b3b6e05e | 5385 | int ref = in->get_nref(); |
494da23a | 5386 | ldout(cct, 5) << __func__ << " in " << *in <<dendl; |
7c673cae FG |
5387 | |
5388 | if (in->dir && !in->dir->dentries.empty()) { | |
5389 | for (auto p = in->dir->dentries.begin(); | |
5390 | p != in->dir->dentries.end(); ) { | |
5391 | Dentry *dn = p->second; | |
5392 | ++p; | |
5393 | /* rmsnap removes whole subtree, need trim inodes recursively. | |
5394 | * we don't need to invalidate dentries recursively. because | |
5395 | * invalidating a directory dentry effectively invalidate | |
5396 | * whole subtree */ | |
5397 | if (in->snapid != CEPH_NOSNAP && dn->inode && dn->inode->is_dir()) | |
5398 | _try_to_trim_inode(dn->inode.get(), false); | |
5399 | ||
5400 | if (dn->lru_is_expireable()) | |
5401 | unlink(dn, true, false); // keep dir, drop dentry | |
5402 | } | |
5403 | if (in->dir->dentries.empty()) { | |
5404 | close_dir(in->dir); | |
5405 | --ref; | |
5406 | } | |
5407 | } | |
5408 | ||
b3b6e05e | 5409 | if (ref > 1 && (in->flags & I_SNAPDIR_OPEN)) { |
7c673cae FG |
5410 | InodeRef snapdir = open_snapdir(in); |
5411 | _try_to_trim_inode(snapdir.get(), false); | |
5412 | --ref; | |
5413 | } | |
5414 | ||
b3b6e05e | 5415 | if (ref > 1) { |
11fdf7f2 TL |
5416 | auto q = in->dentries.begin(); |
5417 | while (q != in->dentries.end()) { | |
5418 | Dentry *dn = *q; | |
5419 | ++q; | |
494da23a TL |
5420 | if( in->ll_ref > 0 && sched_inval) { |
5421 | // FIXME: we play lots of unlink/link tricks when handling MDS replies, | |
5422 | // so in->dentries doesn't always reflect the state of kernel's dcache. | |
5423 | _schedule_invalidate_dentry_callback(dn, true); | |
5424 | } | |
7c673cae FG |
5425 | unlink(dn, true, true); |
5426 | } | |
5427 | } | |
5428 | } | |
5429 | ||
11fdf7f2 | 5430 | void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5431 | { |
5432 | mds_rank_t mds = session->mds_num; | |
5433 | int used = get_caps_used(in); | |
5434 | int wanted = in->caps_wanted(); | |
a4b75251 | 5435 | int flags = 0; |
7c673cae | 5436 | |
a8e16298 TL |
5437 | const unsigned new_caps = m->get_caps(); |
5438 | const bool was_stale = session->cap_gen > cap->gen; | |
11fdf7f2 | 5439 | ldout(cct, 5) << __func__ << " on in " << m->get_ino() |
7c673cae FG |
5440 | << " mds." << mds << " seq " << m->get_seq() |
5441 | << " caps now " << ccap_string(new_caps) | |
a8e16298 | 5442 | << " was " << ccap_string(cap->issued) |
92f5a8d4 | 5443 | << (was_stale ? " (stale)" : "") << dendl; |
a8e16298 TL |
5444 | |
5445 | if (was_stale) | |
5446 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
7c673cae | 5447 | cap->seq = m->get_seq(); |
28e407b8 | 5448 | cap->gen = session->cap_gen; |
7c673cae | 5449 | |
11fdf7f2 | 5450 | check_cap_issue(in, new_caps); |
a8e16298 | 5451 | |
7c673cae | 5452 | // update inode |
1adf2230 AA |
5453 | int issued; |
5454 | in->caps_issued(&issued); | |
5455 | issued |= in->caps_dirty(); | |
7c673cae | 5456 | |
1adf2230 AA |
5457 | if ((new_caps & CEPH_CAP_AUTH_SHARED) && |
5458 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
5459 | in->mode = m->head.mode; |
5460 | in->uid = m->head.uid; | |
5461 | in->gid = m->head.gid; | |
5462 | in->btime = m->btime; | |
5463 | } | |
5464 | bool deleted_inode = false; | |
1adf2230 AA |
5465 | if ((new_caps & CEPH_CAP_LINK_SHARED) && |
5466 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
7c673cae | 5467 | in->nlink = m->head.nlink; |
20effc67 | 5468 | if (in->nlink == 0) |
7c673cae FG |
5469 | deleted_inode = true; |
5470 | } | |
1adf2230 | 5471 | if (!(issued & CEPH_CAP_XATTR_EXCL) && |
7c673cae FG |
5472 | m->xattrbl.length() && |
5473 | m->head.xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
5474 | auto p = m->xattrbl.cbegin(); |
5475 | decode(in->xattrs, p); | |
7c673cae FG |
5476 | in->xattr_version = m->head.xattr_version; |
5477 | } | |
28e407b8 AA |
5478 | |
5479 | if ((new_caps & CEPH_CAP_FILE_SHARED) && m->dirstat_is_valid()) { | |
5480 | in->dirstat.nfiles = m->get_nfiles(); | |
5481 | in->dirstat.nsubdirs = m->get_nsubdirs(); | |
5482 | } | |
5483 | ||
1adf2230 AA |
5484 | if (new_caps & CEPH_CAP_ANY_RD) { |
5485 | update_inode_file_time(in, issued, m->get_time_warp_seq(), | |
5486 | m->get_ctime(), m->get_mtime(), m->get_atime()); | |
5487 | } | |
5488 | ||
5489 | if (new_caps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { | |
5490 | in->layout = m->get_layout(); | |
5491 | update_inode_file_size(in, issued, m->get_size(), | |
5492 | m->get_truncate_seq(), m->get_truncate_size()); | |
5493 | } | |
5494 | ||
5495 | if (m->inline_version > in->inline_version) { | |
5496 | in->inline_data = m->inline_data; | |
5497 | in->inline_version = m->inline_version; | |
5498 | } | |
5499 | ||
5500 | /* always take a newer change attr */ | |
5501 | if (m->get_change_attr() > in->change_attr) | |
5502 | in->change_attr = m->get_change_attr(); | |
7c673cae FG |
5503 | |
5504 | // max_size | |
5505 | if (cap == in->auth_cap && | |
1adf2230 AA |
5506 | (new_caps & CEPH_CAP_ANY_FILE_WR) && |
5507 | (m->get_max_size() != in->max_size)) { | |
7c673cae FG |
5508 | ldout(cct, 10) << "max_size " << in->max_size << " -> " << m->get_max_size() << dendl; |
5509 | in->max_size = m->get_max_size(); | |
5510 | if (in->max_size > in->wanted_max_size) { | |
5511 | in->wanted_max_size = 0; | |
5512 | in->requested_max_size = 0; | |
5513 | } | |
5514 | } | |
5515 | ||
5516 | bool check = false; | |
a8e16298 TL |
5517 | if ((was_stale || m->get_op() == CEPH_CAP_OP_IMPORT) && |
5518 | (wanted & ~(cap->wanted | new_caps))) { | |
5519 | // If mds is importing cap, prior cap messages that update 'wanted' | |
5520 | // may get dropped by mds (migrate seq mismatch). | |
5521 | // | |
5522 | // We don't send cap message to update 'wanted' if what we want are | |
5523 | // already issued. If mds revokes caps, cap message that releases caps | |
5524 | // also tells mds what we want. But if caps got revoked by mds forcedly | |
5525 | // (session stale). We may haven't told mds what we want. | |
7c673cae | 5526 | check = true; |
a8e16298 | 5527 | } |
7c673cae | 5528 | |
7c673cae FG |
5529 | |
5530 | // update caps | |
a8e16298 | 5531 | auto revoked = cap->issued & ~new_caps; |
b32b8144 FG |
5532 | if (revoked) { |
5533 | ldout(cct, 10) << " revocation of " << ccap_string(revoked) << dendl; | |
7c673cae FG |
5534 | cap->issued = new_caps; |
5535 | cap->implemented |= new_caps; | |
5536 | ||
b32b8144 FG |
5537 | // recall delegations if we're losing caps necessary for them |
5538 | if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_RD)) | |
5539 | in->recall_deleg(false); | |
5540 | else if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_WR)) | |
5541 | in->recall_deleg(true); | |
5542 | ||
11fdf7f2 TL |
5543 | used = adjust_caps_used_for_lazyio(used, cap->issued, cap->implemented); |
5544 | if ((used & revoked & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO)) && | |
28e407b8 | 5545 | !_flush(in, new C_Client_FlushComplete(this, in))) { |
7c673cae | 5546 | // waitin' for flush |
11fdf7f2 | 5547 | } else if (used & revoked & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) { |
a4b75251 TL |
5548 | if (_release(in)) { |
5549 | check = true; | |
5550 | flags = CHECK_CAPS_NODELAY; | |
5551 | } | |
7c673cae FG |
5552 | } else { |
5553 | cap->wanted = 0; // don't let check_caps skip sending a response to MDS | |
5554 | check = true; | |
a4b75251 | 5555 | flags = CHECK_CAPS_NODELAY; |
7c673cae | 5556 | } |
a8e16298 TL |
5557 | } else if (cap->issued == new_caps) { |
5558 | ldout(cct, 10) << " caps unchanged at " << ccap_string(cap->issued) << dendl; | |
7c673cae | 5559 | } else { |
a8e16298 | 5560 | ldout(cct, 10) << " grant, new caps are " << ccap_string(new_caps & ~cap->issued) << dendl; |
7c673cae FG |
5561 | cap->issued = new_caps; |
5562 | cap->implemented |= new_caps; | |
5563 | ||
5564 | if (cap == in->auth_cap) { | |
5565 | // non-auth MDS is revoking the newly grant caps ? | |
11fdf7f2 TL |
5566 | for (const auto &p : in->caps) { |
5567 | if (&p.second == cap) | |
7c673cae | 5568 | continue; |
11fdf7f2 | 5569 | if (p.second.implemented & ~p.second.issued & new_caps) { |
7c673cae FG |
5570 | check = true; |
5571 | break; | |
5572 | } | |
5573 | } | |
5574 | } | |
5575 | } | |
5576 | ||
5577 | if (check) | |
a4b75251 | 5578 | check_caps(in, flags); |
7c673cae FG |
5579 | |
5580 | // wake up waiters | |
5581 | if (new_caps) | |
5582 | signal_cond_list(in->waitfor_caps); | |
5583 | ||
5584 | // may drop inode's last ref | |
5585 | if (deleted_inode) | |
5586 | _try_to_trim_inode(in, true); | |
7c673cae FG |
5587 | } |
5588 | ||
7c673cae FG |
5589 | int Client::inode_permission(Inode *in, const UserPerm& perms, unsigned want) |
5590 | { | |
b3b6e05e TL |
5591 | if (perms.uid() == 0) { |
5592 | // Executable are overridable when there is at least one exec bit set | |
5593 | if((want & MAY_EXEC) && !(in->mode & S_IXUGO)) | |
5594 | return -CEPHFS_EACCES; | |
7c673cae | 5595 | return 0; |
b3b6e05e | 5596 | } |
7c673cae FG |
5597 | |
5598 | if (perms.uid() != in->uid && (in->mode & S_IRWXG)) { | |
5599 | int ret = _posix_acl_permission(in, perms, want); | |
f67539c2 | 5600 | if (ret != -CEPHFS_EAGAIN) |
7c673cae FG |
5601 | return ret; |
5602 | } | |
5603 | ||
5604 | // check permissions before doing anything else | |
5605 | if (!in->check_mode(perms, want)) | |
f67539c2 | 5606 | return -CEPHFS_EACCES; |
7c673cae FG |
5607 | return 0; |
5608 | } | |
5609 | ||
5610 | int Client::xattr_permission(Inode *in, const char *name, unsigned want, | |
5611 | const UserPerm& perms) | |
5612 | { | |
5613 | int r = _getattr_for_perm(in, perms); | |
5614 | if (r < 0) | |
5615 | goto out; | |
5616 | ||
5617 | r = 0; | |
5618 | if (strncmp(name, "system.", 7) == 0) { | |
5619 | if ((want & MAY_WRITE) && (perms.uid() != 0 && perms.uid() != in->uid)) | |
f67539c2 | 5620 | r = -CEPHFS_EPERM; |
7c673cae FG |
5621 | } else { |
5622 | r = inode_permission(in, perms, want); | |
5623 | } | |
5624 | out: | |
1adf2230 | 5625 | ldout(cct, 5) << __func__ << " " << in << " = " << r << dendl; |
7c673cae FG |
5626 | return r; |
5627 | } | |
5628 | ||
20effc67 | 5629 | std::ostream& operator<<(std::ostream &out, const UserPerm& perm) { |
7c673cae FG |
5630 | out << "UserPerm(uid: " << perm.uid() << ", gid: " << perm.gid() << ")"; |
5631 | return out; | |
5632 | } | |
5633 | ||
5634 | int Client::may_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
5635 | const UserPerm& perms) | |
5636 | { | |
181888fb | 5637 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5638 | int r = _getattr_for_perm(in, perms); |
5639 | if (r < 0) | |
5640 | goto out; | |
5641 | ||
5642 | if (mask & CEPH_SETATTR_SIZE) { | |
5643 | r = inode_permission(in, perms, MAY_WRITE); | |
5644 | if (r < 0) | |
5645 | goto out; | |
5646 | } | |
5647 | ||
f67539c2 | 5648 | r = -CEPHFS_EPERM; |
7c673cae FG |
5649 | if (mask & CEPH_SETATTR_UID) { |
5650 | if (perms.uid() != 0 && (perms.uid() != in->uid || stx->stx_uid != in->uid)) | |
5651 | goto out; | |
5652 | } | |
5653 | if (mask & CEPH_SETATTR_GID) { | |
5654 | if (perms.uid() != 0 && (perms.uid() != in->uid || | |
5655 | (!perms.gid_in_groups(stx->stx_gid) && stx->stx_gid != in->gid))) | |
5656 | goto out; | |
5657 | } | |
5658 | ||
5659 | if (mask & CEPH_SETATTR_MODE) { | |
5660 | if (perms.uid() != 0 && perms.uid() != in->uid) | |
5661 | goto out; | |
5662 | ||
5663 | gid_t i_gid = (mask & CEPH_SETATTR_GID) ? stx->stx_gid : in->gid; | |
5664 | if (perms.uid() != 0 && !perms.gid_in_groups(i_gid)) | |
5665 | stx->stx_mode &= ~S_ISGID; | |
5666 | } | |
5667 | ||
5668 | if (mask & (CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME | | |
5669 | CEPH_SETATTR_MTIME | CEPH_SETATTR_ATIME)) { | |
5670 | if (perms.uid() != 0 && perms.uid() != in->uid) { | |
5671 | int check_mask = CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME; | |
5672 | if (!(mask & CEPH_SETATTR_MTIME_NOW)) | |
5673 | check_mask |= CEPH_SETATTR_MTIME; | |
5674 | if (!(mask & CEPH_SETATTR_ATIME_NOW)) | |
5675 | check_mask |= CEPH_SETATTR_ATIME; | |
5676 | if (check_mask & mask) { | |
5677 | goto out; | |
5678 | } else { | |
5679 | r = inode_permission(in, perms, MAY_WRITE); | |
5680 | if (r < 0) | |
5681 | goto out; | |
5682 | } | |
5683 | } | |
5684 | } | |
5685 | r = 0; | |
5686 | out: | |
5687 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5688 | return r; | |
5689 | } | |
5690 | ||
5691 | int Client::may_open(Inode *in, int flags, const UserPerm& perms) | |
5692 | { | |
181888fb | 5693 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5694 | unsigned want = 0; |
5695 | ||
5696 | if ((flags & O_ACCMODE) == O_WRONLY) | |
5697 | want = MAY_WRITE; | |
5698 | else if ((flags & O_ACCMODE) == O_RDWR) | |
5699 | want = MAY_READ | MAY_WRITE; | |
5700 | else if ((flags & O_ACCMODE) == O_RDONLY) | |
5701 | want = MAY_READ; | |
5702 | if (flags & O_TRUNC) | |
5703 | want |= MAY_WRITE; | |
5704 | ||
5705 | int r = 0; | |
5706 | switch (in->mode & S_IFMT) { | |
5707 | case S_IFLNK: | |
f67539c2 | 5708 | r = -CEPHFS_ELOOP; |
7c673cae FG |
5709 | goto out; |
5710 | case S_IFDIR: | |
5711 | if (want & MAY_WRITE) { | |
f67539c2 | 5712 | r = -CEPHFS_EISDIR; |
7c673cae FG |
5713 | goto out; |
5714 | } | |
5715 | break; | |
5716 | } | |
5717 | ||
5718 | r = _getattr_for_perm(in, perms); | |
5719 | if (r < 0) | |
5720 | goto out; | |
5721 | ||
5722 | r = inode_permission(in, perms, want); | |
5723 | out: | |
5724 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5725 | return r; | |
5726 | } | |
5727 | ||
5728 | int Client::may_lookup(Inode *dir, const UserPerm& perms) | |
5729 | { | |
181888fb | 5730 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5731 | int r = _getattr_for_perm(dir, perms); |
5732 | if (r < 0) | |
5733 | goto out; | |
5734 | ||
5735 | r = inode_permission(dir, perms, MAY_EXEC); | |
5736 | out: | |
5737 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5738 | return r; | |
5739 | } | |
5740 | ||
5741 | int Client::may_create(Inode *dir, const UserPerm& perms) | |
5742 | { | |
181888fb | 5743 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5744 | int r = _getattr_for_perm(dir, perms); |
5745 | if (r < 0) | |
5746 | goto out; | |
5747 | ||
5748 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5749 | out: | |
5750 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5751 | return r; | |
5752 | } | |
5753 | ||
5754 | int Client::may_delete(Inode *dir, const char *name, const UserPerm& perms) | |
5755 | { | |
181888fb | 5756 | ldout(cct, 20) << __func__ << " " << *dir << "; " << "; name " << name << "; " << perms << dendl; |
7c673cae FG |
5757 | int r = _getattr_for_perm(dir, perms); |
5758 | if (r < 0) | |
5759 | goto out; | |
5760 | ||
5761 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5762 | if (r < 0) | |
5763 | goto out; | |
5764 | ||
f67539c2 | 5765 | /* 'name == NULL' means rmsnap w/o permission checks */ |
7c673cae FG |
5766 | if (perms.uid() != 0 && name && (dir->mode & S_ISVTX)) { |
5767 | InodeRef otherin; | |
5768 | r = _lookup(dir, name, CEPH_CAP_AUTH_SHARED, &otherin, perms); | |
5769 | if (r < 0) | |
5770 | goto out; | |
5771 | if (dir->uid != perms.uid() && otherin->uid != perms.uid()) | |
f67539c2 | 5772 | r = -CEPHFS_EPERM; |
7c673cae FG |
5773 | } |
5774 | out: | |
5775 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5776 | return r; | |
5777 | } | |
5778 | ||
f67539c2 TL |
5779 | int Client::may_delete(const char *relpath, const UserPerm& perms) { |
5780 | ldout(cct, 20) << __func__ << " " << relpath << "; " << perms << dendl; | |
5781 | ||
5782 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
5783 | if (!mref_reader.is_state_satisfied()) | |
5784 | return -ENOTCONN; | |
5785 | ||
5786 | filepath path(relpath); | |
5787 | string name = path.last_dentry(); | |
5788 | path.pop_dentry(); | |
5789 | InodeRef dir; | |
5790 | ||
5791 | std::scoped_lock lock(client_lock); | |
5792 | int r = path_walk(path, &dir, perms); | |
5793 | if (r < 0) | |
5794 | return r; | |
5795 | if (cct->_conf->client_permissions) { | |
5796 | int r = may_delete(dir.get(), name.c_str(), perms); | |
5797 | if (r < 0) | |
5798 | return r; | |
5799 | } | |
5800 | ||
5801 | return 0; | |
5802 | } | |
5803 | ||
7c673cae FG |
5804 | int Client::may_hardlink(Inode *in, const UserPerm& perms) |
5805 | { | |
181888fb | 5806 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5807 | int r = _getattr_for_perm(in, perms); |
5808 | if (r < 0) | |
5809 | goto out; | |
5810 | ||
5811 | if (perms.uid() == 0 || perms.uid() == in->uid) { | |
5812 | r = 0; | |
5813 | goto out; | |
5814 | } | |
5815 | ||
f67539c2 | 5816 | r = -CEPHFS_EPERM; |
7c673cae FG |
5817 | if (!S_ISREG(in->mode)) |
5818 | goto out; | |
5819 | ||
5820 | if (in->mode & S_ISUID) | |
5821 | goto out; | |
5822 | ||
5823 | if ((in->mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
5824 | goto out; | |
5825 | ||
5826 | r = inode_permission(in, perms, MAY_READ | MAY_WRITE); | |
5827 | out: | |
5828 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5829 | return r; | |
5830 | } | |
5831 | ||
5832 | int Client::_getattr_for_perm(Inode *in, const UserPerm& perms) | |
5833 | { | |
5834 | int mask = CEPH_STAT_CAP_MODE; | |
5835 | bool force = false; | |
5836 | if (acl_type != NO_ACL) { | |
5837 | mask |= CEPH_STAT_CAP_XATTR; | |
5838 | force = in->xattr_version == 0; | |
5839 | } | |
5840 | return _getattr(in, mask, perms, force); | |
5841 | } | |
5842 | ||
5843 | vinodeno_t Client::_get_vino(Inode *in) | |
5844 | { | |
5845 | /* The caller must hold the client lock */ | |
5846 | return vinodeno_t(in->ino, in->snapid); | |
5847 | } | |
5848 | ||
7c673cae FG |
5849 | /** |
5850 | * Resolve an MDS spec to a list of MDS daemon GIDs. | |
5851 | * | |
5852 | * The spec is a string representing a GID, rank, filesystem:rank, or name/id. | |
5853 | * It may be '*' in which case it matches all GIDs. | |
5854 | * | |
5855 | * If no error is returned, the `targets` vector will be populated with at least | |
5856 | * one MDS. | |
5857 | */ | |
5858 | int Client::resolve_mds( | |
5859 | const std::string &mds_spec, | |
5860 | std::vector<mds_gid_t> *targets) | |
5861 | { | |
11fdf7f2 TL |
5862 | ceph_assert(fsmap); |
5863 | ceph_assert(targets != nullptr); | |
7c673cae FG |
5864 | |
5865 | mds_role_t role; | |
f67539c2 TL |
5866 | CachedStackStringStream css; |
5867 | int role_r = fsmap->parse_role(mds_spec, &role, *css); | |
7c673cae FG |
5868 | if (role_r == 0) { |
5869 | // We got a role, resolve it to a GID | |
f67539c2 TL |
5870 | auto& info = fsmap->get_filesystem(role.fscid)->mds_map.get_info(role.rank); |
5871 | ldout(cct, 10) << __func__ << ": resolved " << mds_spec << " to role '" | |
5872 | << role << "' aka " << info.human_name() << dendl; | |
5873 | targets->push_back(info.global_id); | |
7c673cae FG |
5874 | return 0; |
5875 | } | |
5876 | ||
5877 | std::string strtol_err; | |
5878 | long long rank_or_gid = strict_strtoll(mds_spec.c_str(), 10, &strtol_err); | |
5879 | if (strtol_err.empty()) { | |
5880 | // It is a possible GID | |
5881 | const mds_gid_t mds_gid = mds_gid_t(rank_or_gid); | |
5882 | if (fsmap->gid_exists(mds_gid)) { | |
f67539c2 TL |
5883 | auto& info = fsmap->get_info_gid(mds_gid); |
5884 | ldout(cct, 10) << __func__ << ": validated gid " << mds_gid << " aka " | |
5885 | << info.human_name() << dendl; | |
7c673cae | 5886 | targets->push_back(mds_gid); |
f67539c2 | 5887 | return 0; |
7c673cae | 5888 | } else { |
f67539c2 | 5889 | lderr(cct) << __func__ << ": gid " << mds_gid << " not in MDS map" |
7c673cae | 5890 | << dendl; |
f67539c2 TL |
5891 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
5892 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5893 | } |
5894 | } else if (mds_spec == "*") { | |
5895 | // It is a wildcard: use all MDSs | |
f67539c2 | 5896 | const auto& mds_info = fsmap->get_mds_info(); |
7c673cae | 5897 | |
f67539c2 | 5898 | ldout(cct, 10) << __func__ << ": resolving `*' to all MDS daemons" << dendl; |
7c673cae | 5899 | if (mds_info.empty()) { |
f67539c2 TL |
5900 | lderr(cct) << __func__ << ": no MDS daemons found" << dendl; |
5901 | lderr(cct) << "FSMap: " << *fsmap << dendl; | |
5902 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5903 | } |
5904 | ||
f67539c2 TL |
5905 | for (const auto& [gid, info] : mds_info) { |
5906 | ldout(cct, 10) << __func__ << ": appending " << info.human_name() << " to targets" << dendl; | |
5907 | targets->push_back(gid); | |
7c673cae | 5908 | } |
f67539c2 | 5909 | return 0; |
7c673cae FG |
5910 | } else { |
5911 | // It did not parse as an integer, it is not a wildcard, it must be a name | |
5912 | const mds_gid_t mds_gid = fsmap->find_mds_gid_by_name(mds_spec); | |
5913 | if (mds_gid == 0) { | |
f67539c2 | 5914 | lderr(cct) << __func__ << ": no MDS daemons found by name `" << mds_spec << "'" << dendl; |
7c673cae | 5915 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
f67539c2 | 5916 | return -CEPHFS_ENOENT; |
7c673cae | 5917 | } else { |
f67539c2 TL |
5918 | auto& info = fsmap->get_info_gid(mds_gid); |
5919 | ldout(cct, 10) << __func__ << ": resolved name '" << mds_spec | |
5920 | << "' to " << info.human_name() << dendl; | |
7c673cae FG |
5921 | targets->push_back(mds_gid); |
5922 | } | |
f67539c2 | 5923 | return 0; |
7c673cae | 5924 | } |
7c673cae FG |
5925 | } |
5926 | ||
5927 | ||
5928 | /** | |
5929 | * Authenticate with mon and establish global ID | |
5930 | */ | |
5931 | int Client::authenticate() | |
5932 | { | |
9f95a23c | 5933 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
5934 | |
5935 | if (monclient->is_authenticated()) { | |
5936 | return 0; | |
5937 | } | |
5938 | ||
9f95a23c | 5939 | client_lock.unlock(); |
7c673cae | 5940 | int r = monclient->authenticate(cct->_conf->client_mount_timeout); |
9f95a23c | 5941 | client_lock.lock(); |
7c673cae FG |
5942 | if (r < 0) { |
5943 | return r; | |
5944 | } | |
5945 | ||
5946 | whoami = monclient->get_global_id(); | |
5947 | messenger->set_myname(entity_name_t::CLIENT(whoami.v)); | |
5948 | ||
5949 | return 0; | |
5950 | } | |
5951 | ||
5952 | int Client::fetch_fsmap(bool user) | |
5953 | { | |
f67539c2 TL |
5954 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
5955 | ||
7c673cae FG |
5956 | // Retrieve FSMap to enable looking up daemon addresses. We need FSMap |
5957 | // rather than MDSMap because no one MDSMap contains all the daemons, and | |
5958 | // a `tell` can address any daemon. | |
5959 | version_t fsmap_latest; | |
f67539c2 | 5960 | bs::error_code ec; |
7c673cae | 5961 | do { |
9f95a23c | 5962 | client_lock.unlock(); |
f67539c2 TL |
5963 | std::tie(fsmap_latest, std::ignore) = |
5964 | monclient->get_version("fsmap", ca::use_blocked[ec]); | |
9f95a23c | 5965 | client_lock.lock(); |
f67539c2 | 5966 | } while (ec == bs::errc::resource_unavailable_try_again); |
7c673cae | 5967 | |
f67539c2 TL |
5968 | if (ec) { |
5969 | lderr(cct) << "Failed to learn FSMap version: " << ec << dendl; | |
5970 | return ceph::from_error_code(ec); | |
7c673cae FG |
5971 | } |
5972 | ||
5973 | ldout(cct, 10) << __func__ << " learned FSMap version " << fsmap_latest << dendl; | |
5974 | ||
5975 | if (user) { | |
5976 | if (!fsmap_user || fsmap_user->get_epoch() < fsmap_latest) { | |
5977 | monclient->sub_want("fsmap.user", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5978 | monclient->renew_subs(); | |
5979 | wait_on_list(waiting_for_fsmap); | |
5980 | } | |
11fdf7f2 TL |
5981 | ceph_assert(fsmap_user); |
5982 | ceph_assert(fsmap_user->get_epoch() >= fsmap_latest); | |
7c673cae FG |
5983 | } else { |
5984 | if (!fsmap || fsmap->get_epoch() < fsmap_latest) { | |
5985 | monclient->sub_want("fsmap", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5986 | monclient->renew_subs(); | |
5987 | wait_on_list(waiting_for_fsmap); | |
5988 | } | |
11fdf7f2 TL |
5989 | ceph_assert(fsmap); |
5990 | ceph_assert(fsmap->get_epoch() >= fsmap_latest); | |
7c673cae FG |
5991 | } |
5992 | ldout(cct, 10) << __func__ << " finished waiting for FSMap version " | |
5993 | << fsmap_latest << dendl; | |
5994 | return 0; | |
5995 | } | |
5996 | ||
5997 | /** | |
5998 | * | |
5999 | * @mds_spec one of ID, rank, GID, "*" | |
6000 | * | |
6001 | */ | |
6002 | int Client::mds_command( | |
6003 | const std::string &mds_spec, | |
6004 | const vector<string>& cmd, | |
6005 | const bufferlist& inbl, | |
6006 | bufferlist *outbl, | |
6007 | string *outs, | |
6008 | Context *onfinish) | |
6009 | { | |
f67539c2 TL |
6010 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
6011 | if (!iref_reader.is_state_satisfied()) | |
6012 | return -CEPHFS_ENOTCONN; | |
7c673cae | 6013 | |
f67539c2 | 6014 | std::unique_lock cl(client_lock); |
7c673cae FG |
6015 | |
6016 | int r; | |
6017 | r = authenticate(); | |
6018 | if (r < 0) { | |
6019 | return r; | |
6020 | } | |
6021 | ||
6022 | r = fetch_fsmap(false); | |
6023 | if (r < 0) { | |
6024 | return r; | |
6025 | } | |
6026 | ||
6027 | // Look up MDS target(s) of the command | |
6028 | std::vector<mds_gid_t> targets; | |
6029 | r = resolve_mds(mds_spec, &targets); | |
6030 | if (r < 0) { | |
6031 | return r; | |
6032 | } | |
6033 | ||
6034 | // If daemons are laggy, we won't send them commands. If all | |
6035 | // are laggy then we fail. | |
6036 | std::vector<mds_gid_t> non_laggy; | |
f67539c2 | 6037 | for (const auto& gid : targets) { |
7c673cae FG |
6038 | const auto info = fsmap->get_info_gid(gid); |
6039 | if (!info.laggy()) { | |
6040 | non_laggy.push_back(gid); | |
6041 | } | |
6042 | } | |
6043 | if (non_laggy.size() == 0) { | |
6044 | *outs = "All targeted MDS daemons are laggy"; | |
f67539c2 | 6045 | return -CEPHFS_ENOENT; |
7c673cae FG |
6046 | } |
6047 | ||
6048 | if (metadata.empty()) { | |
6049 | // We are called on an unmounted client, so metadata | |
6050 | // won't be initialized yet. | |
6051 | populate_metadata(""); | |
6052 | } | |
6053 | ||
6054 | // Send commands to targets | |
6055 | C_GatherBuilder gather(cct, onfinish); | |
f67539c2 | 6056 | for (const auto& target_gid : non_laggy) { |
7c673cae FG |
6057 | const auto info = fsmap->get_info_gid(target_gid); |
6058 | ||
6059 | // Open a connection to the target MDS | |
11fdf7f2 | 6060 | ConnectionRef conn = messenger->connect_to_mds(info.get_addrs()); |
7c673cae | 6061 | |
f67539c2 TL |
6062 | cl.unlock(); |
6063 | { | |
6064 | std::scoped_lock cmd_lock(command_lock); | |
6065 | // Generate MDSCommandOp state | |
6066 | auto &op = command_table.start_command(); | |
7c673cae | 6067 | |
f67539c2 TL |
6068 | op.on_finish = gather.new_sub(); |
6069 | op.cmd = cmd; | |
6070 | op.outbl = outbl; | |
6071 | op.outs = outs; | |
6072 | op.inbl = inbl; | |
6073 | op.mds_gid = target_gid; | |
6074 | op.con = conn; | |
7c673cae | 6075 | |
f67539c2 TL |
6076 | ldout(cct, 4) << __func__ << ": new command op to " << target_gid |
6077 | << " tid=" << op.tid << cmd << dendl; | |
7c673cae | 6078 | |
f67539c2 TL |
6079 | // Construct and send MCommand |
6080 | MessageRef m = op.get_message(monclient->get_fsid()); | |
6081 | conn->send_message2(std::move(m)); | |
6082 | } | |
6083 | cl.lock(); | |
7c673cae FG |
6084 | } |
6085 | gather.activate(); | |
6086 | ||
6087 | return 0; | |
6088 | } | |
6089 | ||
11fdf7f2 | 6090 | void Client::handle_command_reply(const MConstRef<MCommandReply>& m) |
7c673cae FG |
6091 | { |
6092 | ceph_tid_t const tid = m->get_tid(); | |
6093 | ||
6094 | ldout(cct, 10) << __func__ << ": tid=" << m->get_tid() << dendl; | |
6095 | ||
f67539c2 | 6096 | std::scoped_lock cmd_lock(command_lock); |
7c673cae FG |
6097 | if (!command_table.exists(tid)) { |
6098 | ldout(cct, 1) << __func__ << ": unknown tid " << tid << ", dropping" << dendl; | |
7c673cae FG |
6099 | return; |
6100 | } | |
6101 | ||
6102 | auto &op = command_table.get_command(tid); | |
6103 | if (op.outbl) { | |
11fdf7f2 | 6104 | *op.outbl = m->get_data(); |
7c673cae FG |
6105 | } |
6106 | if (op.outs) { | |
6107 | *op.outs = m->rs; | |
6108 | } | |
6109 | ||
6110 | if (op.on_finish) { | |
6111 | op.on_finish->complete(m->r); | |
6112 | } | |
6113 | ||
6114 | command_table.erase(tid); | |
7c673cae FG |
6115 | } |
6116 | ||
6117 | // ------------------- | |
6118 | // MOUNT | |
6119 | ||
11fdf7f2 | 6120 | int Client::subscribe_mdsmap(const std::string &fs_name) |
7c673cae | 6121 | { |
7c673cae FG |
6122 | int r = authenticate(); |
6123 | if (r < 0) { | |
6124 | lderr(cct) << "authentication failed: " << cpp_strerror(r) << dendl; | |
6125 | return r; | |
6126 | } | |
6127 | ||
11fdf7f2 TL |
6128 | std::string resolved_fs_name; |
6129 | if (fs_name.empty()) { | |
9f95a23c TL |
6130 | resolved_fs_name = cct->_conf.get_val<std::string>("client_fs"); |
6131 | if (resolved_fs_name.empty()) | |
6132 | // Try the backwards compatibility fs name option | |
6133 | resolved_fs_name = cct->_conf.get_val<std::string>("client_mds_namespace"); | |
11fdf7f2 TL |
6134 | } else { |
6135 | resolved_fs_name = fs_name; | |
6136 | } | |
6137 | ||
7c673cae | 6138 | std::string want = "mdsmap"; |
11fdf7f2 | 6139 | if (!resolved_fs_name.empty()) { |
7c673cae FG |
6140 | r = fetch_fsmap(true); |
6141 | if (r < 0) | |
6142 | return r; | |
11fdf7f2 TL |
6143 | fscid = fsmap_user->get_fs_cid(resolved_fs_name); |
6144 | if (fscid == FS_CLUSTER_ID_NONE) { | |
f67539c2 | 6145 | return -CEPHFS_ENOENT; |
11fdf7f2 | 6146 | } |
7c673cae FG |
6147 | |
6148 | std::ostringstream oss; | |
11fdf7f2 | 6149 | oss << want << "." << fscid; |
7c673cae FG |
6150 | want = oss.str(); |
6151 | } | |
6152 | ldout(cct, 10) << "Subscribing to map '" << want << "'" << dendl; | |
6153 | ||
6154 | monclient->sub_want(want, 0, 0); | |
6155 | monclient->renew_subs(); | |
6156 | ||
11fdf7f2 TL |
6157 | return 0; |
6158 | } | |
6159 | ||
6160 | int Client::mount(const std::string &mount_root, const UserPerm& perms, | |
6161 | bool require_mds, const std::string &fs_name) | |
6162 | { | |
f67539c2 | 6163 | ceph_assert(is_initialized()); |
11fdf7f2 | 6164 | |
f67539c2 TL |
6165 | /* |
6166 | * To make sure that the _unmount() must wait until the mount() | |
6167 | * is done. | |
6168 | */ | |
6169 | RWRef_t mref_writer(mount_state, CLIENT_MOUNTING, false); | |
6170 | if (!mref_writer.is_first_writer()) // already mounting or mounted | |
11fdf7f2 | 6171 | return 0; |
11fdf7f2 | 6172 | |
f67539c2 | 6173 | std::unique_lock cl(client_lock); |
11fdf7f2 TL |
6174 | |
6175 | int r = subscribe_mdsmap(fs_name); | |
6176 | if (r < 0) { | |
6177 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
6178 | return r; | |
6179 | } | |
6180 | ||
f67539c2 TL |
6181 | start_tick_thread(); // start tick thread |
6182 | ||
7c673cae FG |
6183 | if (require_mds) { |
6184 | while (1) { | |
6185 | auto availability = mdsmap->is_cluster_available(); | |
6186 | if (availability == MDSMap::STUCK_UNAVAILABLE) { | |
6187 | // Error out | |
6188 | ldout(cct, 10) << "mds cluster unavailable: epoch=" << mdsmap->get_epoch() << dendl; | |
6189 | return CEPH_FUSE_NO_MDS_UP; | |
6190 | } else if (availability == MDSMap::AVAILABLE) { | |
6191 | // Continue to mount | |
6192 | break; | |
6193 | } else if (availability == MDSMap::TRANSIENT_UNAVAILABLE) { | |
6194 | // Else, wait. MDSMonitor will update the map to bring | |
6195 | // us to a conclusion eventually. | |
6196 | wait_on_list(waiting_for_mdsmap); | |
6197 | } else { | |
6198 | // Unexpected value! | |
6199 | ceph_abort(); | |
6200 | } | |
6201 | } | |
6202 | } | |
6203 | ||
6204 | populate_metadata(mount_root.empty() ? "/" : mount_root); | |
6205 | ||
6206 | filepath fp(CEPH_INO_ROOT); | |
6207 | if (!mount_root.empty()) { | |
6208 | fp = filepath(mount_root.c_str()); | |
6209 | } | |
6210 | while (true) { | |
6211 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
6212 | req->set_filepath(fp); | |
6213 | req->head.args.getattr.mask = CEPH_STAT_CAP_INODE_ALL; | |
6214 | int res = make_request(req, perms); | |
6215 | if (res < 0) { | |
f67539c2 | 6216 | if (res == -CEPHFS_EACCES && root) { |
7c673cae FG |
6217 | ldout(cct, 1) << __func__ << " EACCES on parent of mount point; quotas may not work" << dendl; |
6218 | break; | |
6219 | } | |
6220 | return res; | |
6221 | } | |
6222 | ||
6223 | if (fp.depth()) | |
6224 | fp.pop_dentry(); | |
6225 | else | |
6226 | break; | |
6227 | } | |
6228 | ||
11fdf7f2 | 6229 | ceph_assert(root); |
b3b6e05e | 6230 | _ll_get(root.get()); |
7c673cae | 6231 | |
7c673cae FG |
6232 | // trace? |
6233 | if (!cct->_conf->client_trace.empty()) { | |
6234 | traceout.open(cct->_conf->client_trace.c_str()); | |
6235 | if (traceout.is_open()) { | |
6236 | ldout(cct, 1) << "opened trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6237 | } else { | |
6238 | ldout(cct, 1) << "FAILED to open trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6239 | } | |
6240 | } | |
6241 | ||
6242 | /* | |
6243 | ldout(cct, 3) << "op: // client trace data structs" << dendl; | |
6244 | ldout(cct, 3) << "op: struct stat st;" << dendl; | |
6245 | ldout(cct, 3) << "op: struct utimbuf utim;" << dendl; | |
6246 | ldout(cct, 3) << "op: int readlinkbuf_len = 1000;" << dendl; | |
6247 | ldout(cct, 3) << "op: char readlinkbuf[readlinkbuf_len];" << dendl; | |
6248 | ldout(cct, 3) << "op: map<string, inode_t*> dir_contents;" << dendl; | |
6249 | ldout(cct, 3) << "op: map<int, int> open_files;" << dendl; | |
6250 | ldout(cct, 3) << "op: int fd;" << dendl; | |
6251 | */ | |
f67539c2 TL |
6252 | |
6253 | mref_writer.update_state(CLIENT_MOUNTED); | |
7c673cae FG |
6254 | return 0; |
6255 | } | |
6256 | ||
6257 | // UNMOUNT | |
6258 | ||
6259 | void Client::_close_sessions() | |
6260 | { | |
f6b5b4d7 | 6261 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
20effc67 | 6262 | if (it->second->state == MetaSession::STATE_REJECTED) |
f6b5b4d7 TL |
6263 | mds_sessions.erase(it++); |
6264 | else | |
6265 | ++it; | |
6266 | } | |
6267 | ||
7c673cae FG |
6268 | while (!mds_sessions.empty()) { |
6269 | // send session closes! | |
11fdf7f2 | 6270 | for (auto &p : mds_sessions) { |
20effc67 TL |
6271 | if (p.second->state != MetaSession::STATE_CLOSING) { |
6272 | _close_mds_session(p.second.get()); | |
f6b5b4d7 | 6273 | mds_ranks_closing.insert(p.first); |
7c673cae FG |
6274 | } |
6275 | } | |
6276 | ||
6277 | // wait for sessions to close | |
f6b5b4d7 TL |
6278 | double timo = cct->_conf.get_val<std::chrono::seconds>("client_shutdown_timeout").count(); |
6279 | ldout(cct, 2) << "waiting for " << mds_ranks_closing.size() << " mds session(s) to close (timeout: " | |
6280 | << timo << "s)" << dendl; | |
9f95a23c | 6281 | std::unique_lock l{client_lock, std::adopt_lock}; |
f6b5b4d7 TL |
6282 | if (!timo) { |
6283 | mount_cond.wait(l); | |
6284 | } else if (!mount_cond.wait_for(l, ceph::make_timespan(timo), [this] { return mds_ranks_closing.empty(); })) { | |
6285 | ldout(cct, 1) << mds_ranks_closing.size() << " mds(s) did not respond to session close -- timing out." << dendl; | |
6286 | while (!mds_ranks_closing.empty()) { | |
6287 | auto session = mds_sessions.at(*mds_ranks_closing.begin()); | |
6288 | // this prunes entry from mds_sessions and mds_ranks_closing | |
20effc67 | 6289 | _closed_mds_session(session.get(), -CEPHFS_ETIMEDOUT); |
f6b5b4d7 TL |
6290 | } |
6291 | } | |
6292 | ||
6293 | mds_ranks_closing.clear(); | |
9f95a23c | 6294 | l.release(); |
7c673cae FG |
6295 | } |
6296 | } | |
6297 | ||
522d829b TL |
6298 | void Client::flush_mdlog_sync(Inode *in) |
6299 | { | |
6300 | if (in->unsafe_ops.empty()) { | |
6301 | return; | |
6302 | } | |
6303 | ||
6304 | std::set<mds_rank_t> anchor; | |
6305 | for (auto &&p : in->unsafe_ops) { | |
6306 | anchor.emplace(p->mds); | |
6307 | } | |
6308 | if (in->auth_cap) { | |
6309 | anchor.emplace(in->auth_cap->session->mds_num); | |
6310 | } | |
6311 | ||
6312 | for (auto &rank : anchor) { | |
6313 | auto session = &mds_sessions.at(rank); | |
20effc67 | 6314 | flush_mdlog(session->get()); |
522d829b TL |
6315 | } |
6316 | } | |
6317 | ||
31f18b77 FG |
6318 | void Client::flush_mdlog_sync() |
6319 | { | |
522d829b | 6320 | if (mds_requests.empty()) |
31f18b77 | 6321 | return; |
11fdf7f2 | 6322 | for (auto &p : mds_sessions) { |
20effc67 | 6323 | flush_mdlog(p.second.get()); |
31f18b77 FG |
6324 | } |
6325 | } | |
6326 | ||
6327 | void Client::flush_mdlog(MetaSession *session) | |
6328 | { | |
6329 | // Only send this to Luminous or newer MDS daemons, older daemons | |
6330 | // will crash if they see an unknown CEPH_SESSION_* value in this msg. | |
6331 | const uint64_t features = session->con->get_features(); | |
6332 | if (HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
9f95a23c | 6333 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_FLUSH_MDLOG); |
11fdf7f2 | 6334 | session->con->send_message2(std::move(m)); |
31f18b77 FG |
6335 | } |
6336 | } | |
6337 | ||
6338 | ||
11fdf7f2 TL |
6339 | void Client::_abort_mds_sessions(int err) |
6340 | { | |
6341 | for (auto p = mds_requests.begin(); p != mds_requests.end(); ) { | |
6342 | auto req = p->second; | |
6343 | ++p; | |
6344 | // unsafe requests will be removed during close session below. | |
6345 | if (req->got_unsafe) | |
6346 | continue; | |
6347 | ||
6348 | req->abort(err); | |
6349 | if (req->caller_cond) { | |
6350 | req->kick = true; | |
9f95a23c | 6351 | req->caller_cond->notify_all(); |
11fdf7f2 TL |
6352 | } |
6353 | } | |
6354 | ||
6355 | // Process aborts on any requests that were on this waitlist. | |
6356 | // Any requests that were on a waiting_for_open session waitlist | |
6357 | // will get kicked during close session below. | |
6358 | signal_cond_list(waiting_for_mdsmap); | |
6359 | ||
6360 | // Force-close all sessions | |
6361 | while(!mds_sessions.empty()) { | |
20effc67 TL |
6362 | auto session = mds_sessions.begin()->second; |
6363 | _closed_mds_session(session.get(), err); | |
11fdf7f2 TL |
6364 | } |
6365 | } | |
6366 | ||
6367 | void Client::_unmount(bool abort) | |
7c673cae | 6368 | { |
f67539c2 TL |
6369 | /* |
6370 | * We are unmounting the client. | |
6371 | * | |
6372 | * Just declare the state to STATE_UNMOUNTING to block and fail | |
6373 | * any new comming "reader" and then try to wait all the in-flight | |
6374 | * "readers" to finish. | |
6375 | */ | |
6376 | RWRef_t mref_writer(mount_state, CLIENT_UNMOUNTING, false); | |
6377 | if (!mref_writer.is_first_writer()) | |
181888fb | 6378 | return; |
f67539c2 | 6379 | mref_writer.wait_readers_done(); |
7c673cae | 6380 | |
f67539c2 TL |
6381 | std::unique_lock lock{client_lock}; |
6382 | ||
6383 | if (abort || blocklisted) { | |
6384 | ldout(cct, 2) << "unmounting (" << (abort ? "abort)" : "blocklisted)") << dendl; | |
11fdf7f2 TL |
6385 | } else { |
6386 | ldout(cct, 2) << "unmounting" << dendl; | |
6387 | } | |
7c673cae | 6388 | |
b32b8144 FG |
6389 | deleg_timeout = 0; |
6390 | ||
11fdf7f2 | 6391 | if (abort) { |
f67539c2 | 6392 | mount_aborted = true; |
11fdf7f2 | 6393 | // Abort all mds sessions |
f67539c2 | 6394 | _abort_mds_sessions(-CEPHFS_ENOTCONN); |
11fdf7f2 | 6395 | |
f67539c2 | 6396 | objecter->op_cancel_writes(-CEPHFS_ENOTCONN); |
11fdf7f2 TL |
6397 | } else { |
6398 | // flush the mdlog for pending requests, if any | |
6399 | flush_mdlog_sync(); | |
6400 | } | |
6401 | ||
9f95a23c TL |
6402 | mount_cond.wait(lock, [this] { |
6403 | if (!mds_requests.empty()) { | |
6404 | ldout(cct, 10) << "waiting on " << mds_requests.size() << " requests" | |
6405 | << dendl; | |
6406 | } | |
6407 | return mds_requests.empty(); | |
6408 | }); | |
7c673cae FG |
6409 | |
6410 | cwd.reset(); | |
b3b6e05e | 6411 | root.reset(); |
7c673cae FG |
6412 | |
6413 | // clean up any unclosed files | |
6414 | while (!fd_map.empty()) { | |
6415 | Fh *fh = fd_map.begin()->second; | |
6416 | fd_map.erase(fd_map.begin()); | |
6417 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *fh->inode << dendl; | |
6418 | _release_fh(fh); | |
6419 | } | |
6420 | ||
6421 | while (!ll_unclosed_fh_set.empty()) { | |
6422 | set<Fh*>::iterator it = ll_unclosed_fh_set.begin(); | |
6423 | Fh *fh = *it; | |
6424 | ll_unclosed_fh_set.erase(fh); | |
6425 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *(fh->inode) << dendl; | |
6426 | _release_fh(fh); | |
6427 | } | |
6428 | ||
6429 | while (!opened_dirs.empty()) { | |
6430 | dir_result_t *dirp = *opened_dirs.begin(); | |
6431 | ldout(cct, 0) << " destroyed lost open dir " << dirp << " on " << *dirp->inode << dendl; | |
6432 | _closedir(dirp); | |
6433 | } | |
6434 | ||
6435 | _ll_drop_pins(); | |
6436 | ||
7c673cae FG |
6437 | if (cct->_conf->client_oc) { |
6438 | // flush/release all buffered data | |
11fdf7f2 TL |
6439 | std::list<InodeRef> anchor; |
6440 | for (auto& p : inode_map) { | |
6441 | Inode *in = p.second; | |
7c673cae | 6442 | if (!in) { |
11fdf7f2 TL |
6443 | ldout(cct, 0) << "null inode_map entry ino " << p.first << dendl; |
6444 | ceph_assert(in); | |
7c673cae | 6445 | } |
11fdf7f2 TL |
6446 | |
6447 | // prevent inode from getting freed | |
6448 | anchor.emplace_back(in); | |
6449 | ||
f67539c2 | 6450 | if (abort || blocklisted) { |
11fdf7f2 TL |
6451 | objectcacher->purge_set(&in->oset); |
6452 | } else if (!in->caps.empty()) { | |
7c673cae FG |
6453 | _release(in); |
6454 | _flush(in, new C_Client_FlushComplete(this, in)); | |
6455 | } | |
6456 | } | |
6457 | } | |
6458 | ||
f67539c2 | 6459 | if (abort || blocklisted) { |
20effc67 TL |
6460 | for (auto &q : mds_sessions) { |
6461 | auto s = q.second; | |
6462 | for (auto p = s->dirty_list.begin(); !p.end(); ) { | |
6463 | Inode *in = *p; | |
6464 | ++p; | |
6465 | if (in->dirty_caps) { | |
6466 | ldout(cct, 0) << " drop dirty caps on " << *in << dendl; | |
6467 | in->mark_caps_clean(); | |
6468 | put_inode(in); | |
6469 | } | |
11fdf7f2 TL |
6470 | } |
6471 | } | |
6472 | } else { | |
6473 | flush_caps_sync(); | |
6474 | wait_sync_caps(last_flush_tid); | |
6475 | } | |
7c673cae FG |
6476 | |
6477 | // empty lru cache | |
7c673cae FG |
6478 | trim_cache(); |
6479 | ||
f67539c2 TL |
6480 | delay_put_inodes(); |
6481 | ||
7c673cae FG |
6482 | while (lru.lru_get_size() > 0 || |
6483 | !inode_map.empty()) { | |
6484 | ldout(cct, 2) << "cache still has " << lru.lru_get_size() | |
6485 | << "+" << inode_map.size() << " items" | |
6486 | << ", waiting (for caps to release?)" | |
6487 | << dendl; | |
f67539c2 | 6488 | |
9f95a23c TL |
6489 | if (auto r = mount_cond.wait_for(lock, ceph::make_timespan(5)); |
6490 | r == std::cv_status::timeout) { | |
7c673cae FG |
6491 | dump_cache(NULL); |
6492 | } | |
6493 | } | |
11fdf7f2 TL |
6494 | ceph_assert(lru.lru_get_size() == 0); |
6495 | ceph_assert(inode_map.empty()); | |
7c673cae FG |
6496 | |
6497 | // stop tracing | |
6498 | if (!cct->_conf->client_trace.empty()) { | |
6499 | ldout(cct, 1) << "closing trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6500 | traceout.close(); | |
6501 | } | |
6502 | ||
f67539c2 TL |
6503 | // stop the tick thread |
6504 | tick_thread_stopped = true; | |
6505 | upkeep_cond.notify_one(); | |
6506 | ||
7c673cae FG |
6507 | _close_sessions(); |
6508 | ||
f67539c2 | 6509 | mref_writer.update_state(CLIENT_UNMOUNTED); |
7c673cae FG |
6510 | |
6511 | ldout(cct, 2) << "unmounted." << dendl; | |
6512 | } | |
6513 | ||
b32b8144 FG |
6514 | void Client::unmount() |
6515 | { | |
11fdf7f2 TL |
6516 | _unmount(false); |
6517 | } | |
6518 | ||
6519 | void Client::abort_conn() | |
6520 | { | |
11fdf7f2 | 6521 | _unmount(true); |
b32b8144 FG |
6522 | } |
6523 | ||
7c673cae FG |
6524 | void Client::flush_cap_releases() |
6525 | { | |
f67539c2 TL |
6526 | uint64_t nr_caps = 0; |
6527 | ||
7c673cae | 6528 | // send any cap releases |
11fdf7f2 | 6529 | for (auto &p : mds_sessions) { |
20effc67 TL |
6530 | auto session = p.second; |
6531 | if (session->release && mdsmap->is_clientreplay_or_active_or_stopping( | |
11fdf7f2 | 6532 | p.first)) { |
20effc67 | 6533 | nr_caps += session->release->caps.size(); |
7c673cae FG |
6534 | if (cct->_conf->client_inject_release_failure) { |
6535 | ldout(cct, 20) << __func__ << " injecting failure to send cap release message" << dendl; | |
7c673cae | 6536 | } else { |
20effc67 | 6537 | session->con->send_message2(std::move(session->release)); |
7c673cae | 6538 | } |
20effc67 | 6539 | session->release.reset(); |
7c673cae FG |
6540 | } |
6541 | } | |
f67539c2 TL |
6542 | |
6543 | if (nr_caps > 0) { | |
6544 | dec_pinned_icaps(nr_caps); | |
6545 | } | |
7c673cae FG |
6546 | } |
6547 | ||
f67539c2 | 6548 | void Client::renew_and_flush_cap_releases() |
7c673cae | 6549 | { |
f67539c2 TL |
6550 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
6551 | ||
6552 | if (!mount_aborted && mdsmap->get_epoch()) { | |
6553 | // renew caps? | |
6554 | utime_t el = ceph_clock_now() - last_cap_renew; | |
6555 | if (unlikely(el > mdsmap->get_session_timeout() / 3.0)) | |
6556 | renew_caps(); | |
6557 | ||
6558 | flush_cap_releases(); | |
7c673cae | 6559 | } |
f67539c2 TL |
6560 | } |
6561 | ||
6562 | void Client::tick() | |
6563 | { | |
6564 | ldout(cct, 20) << "tick" << dendl; | |
7c673cae | 6565 | |
7c673cae FG |
6566 | utime_t now = ceph_clock_now(); |
6567 | ||
f67539c2 TL |
6568 | /* |
6569 | * If the mount() is not finished | |
6570 | */ | |
6571 | if (is_mounting() && !mds_requests.empty()) { | |
7c673cae | 6572 | MetaRequest *req = mds_requests.begin()->second; |
f67539c2 | 6573 | |
7c673cae | 6574 | if (req->op_stamp + cct->_conf->client_mount_timeout < now) { |
f67539c2 | 6575 | req->abort(-CEPHFS_ETIMEDOUT); |
7c673cae | 6576 | if (req->caller_cond) { |
f67539c2 TL |
6577 | req->kick = true; |
6578 | req->caller_cond->notify_all(); | |
7c673cae FG |
6579 | } |
6580 | signal_cond_list(waiting_for_mdsmap); | |
11fdf7f2 | 6581 | for (auto &p : mds_sessions) { |
20effc67 | 6582 | signal_context_list(p.second->waiting_for_open); |
11fdf7f2 | 6583 | } |
7c673cae FG |
6584 | } |
6585 | } | |
6586 | ||
f67539c2 | 6587 | renew_and_flush_cap_releases(); |
7c673cae FG |
6588 | |
6589 | // delayed caps | |
28e407b8 | 6590 | xlist<Inode*>::iterator p = delayed_list.begin(); |
7c673cae FG |
6591 | while (!p.end()) { |
6592 | Inode *in = *p; | |
6593 | ++p; | |
f67539c2 | 6594 | if (!mount_aborted && in->hold_caps_until > now) |
7c673cae | 6595 | break; |
28e407b8 | 6596 | delayed_list.pop_front(); |
f67539c2 TL |
6597 | if (!mount_aborted) |
6598 | check_caps(in, CHECK_CAPS_NODELAY); | |
7c673cae FG |
6599 | } |
6600 | ||
f67539c2 TL |
6601 | if (!mount_aborted) |
6602 | collect_and_send_metrics(); | |
6603 | ||
6604 | delay_put_inodes(is_unmounting()); | |
7c673cae | 6605 | trim_cache(true); |
f6b5b4d7 | 6606 | |
f67539c2 | 6607 | if (blocklisted && (is_mounted() || is_unmounting()) && |
f6b5b4d7 TL |
6608 | last_auto_reconnect + 30 * 60 < now && |
6609 | cct->_conf.get_val<bool>("client_reconnect_stale")) { | |
6610 | messenger->client_reset(); | |
6611 | fd_gen++; // invalidate open files | |
f67539c2 | 6612 | blocklisted = false; |
f6b5b4d7 TL |
6613 | _kick_stale_sessions(); |
6614 | last_auto_reconnect = now; | |
6615 | } | |
7c673cae FG |
6616 | } |
6617 | ||
f67539c2 TL |
6618 | void Client::start_tick_thread() |
6619 | { | |
6620 | upkeeper = std::thread([this]() { | |
6621 | using time = ceph::coarse_mono_time; | |
6622 | using sec = std::chrono::seconds; | |
6623 | ||
6624 | auto last_tick = time::min(); | |
6625 | ||
6626 | std::unique_lock cl(client_lock); | |
6627 | while (!tick_thread_stopped) { | |
6628 | auto now = clock::now(); | |
6629 | auto since = now - last_tick; | |
6630 | ||
6631 | auto t_interval = clock::duration(cct->_conf.get_val<sec>("client_tick_interval")); | |
6632 | auto d_interval = clock::duration(cct->_conf.get_val<sec>("client_debug_inject_tick_delay")); | |
6633 | ||
6634 | auto interval = std::max(t_interval, d_interval); | |
6635 | if (likely(since >= interval*.90)) { | |
6636 | tick(); | |
6637 | last_tick = clock::now(); | |
6638 | } else { | |
6639 | interval -= since; | |
6640 | } | |
6641 | ||
6642 | ldout(cct, 20) << "upkeep thread waiting interval " << interval << dendl; | |
6643 | if (!tick_thread_stopped) | |
6644 | upkeep_cond.wait_for(cl, interval); | |
6645 | } | |
6646 | }); | |
6647 | } | |
6648 | ||
6649 | void Client::collect_and_send_metrics() { | |
6650 | ldout(cct, 20) << __func__ << dendl; | |
6651 | ||
6652 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6653 | ||
6654 | // right now, we only track and send global metrics. its sufficient | |
6655 | // to send these metrics to MDS rank0. | |
6656 | collect_and_send_global_metrics(); | |
6657 | } | |
6658 | ||
6659 | void Client::collect_and_send_global_metrics() { | |
6660 | ldout(cct, 20) << __func__ << dendl; | |
6661 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6662 | ||
6663 | if (!have_open_session((mds_rank_t)0)) { | |
6664 | ldout(cct, 5) << __func__ << ": no session with rank=0 -- not sending metric" | |
6665 | << dendl; | |
6666 | return; | |
6667 | } | |
6668 | auto session = _get_or_open_mds_session((mds_rank_t)0); | |
6669 | if (!session->mds_features.test(CEPHFS_FEATURE_METRIC_COLLECT)) { | |
6670 | ldout(cct, 5) << __func__ << ": rank=0 does not support metrics" << dendl; | |
6671 | return; | |
6672 | } | |
6673 | ||
6674 | ClientMetricMessage metric; | |
6675 | std::vector<ClientMetricMessage> message; | |
6676 | ||
6677 | // read latency | |
33c7a0ef TL |
6678 | if (_collect_and_send_global_metrics || |
6679 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_READ_LATENCY)) { | |
6680 | metric = ClientMetricMessage(ReadLatencyPayload(logger->tget(l_c_read))); | |
6681 | message.push_back(metric); | |
6682 | } | |
f67539c2 TL |
6683 | |
6684 | // write latency | |
33c7a0ef TL |
6685 | if (_collect_and_send_global_metrics || |
6686 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_WRITE_LATENCY)) { | |
6687 | metric = ClientMetricMessage(WriteLatencyPayload(logger->tget(l_c_wrlat))); | |
6688 | message.push_back(metric); | |
6689 | } | |
f67539c2 TL |
6690 | |
6691 | // metadata latency | |
33c7a0ef TL |
6692 | if (_collect_and_send_global_metrics || |
6693 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_METADATA_LATENCY)) { | |
6694 | metric = ClientMetricMessage(MetadataLatencyPayload(logger->tget(l_c_lat))); | |
6695 | message.push_back(metric); | |
6696 | } | |
f67539c2 TL |
6697 | |
6698 | // cap hit ratio -- nr_caps is unused right now | |
33c7a0ef TL |
6699 | if (_collect_and_send_global_metrics || |
6700 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_CAP_INFO)) { | |
6701 | auto [cap_hits, cap_misses] = get_cap_hit_rates(); | |
6702 | metric = ClientMetricMessage(CapInfoPayload(cap_hits, cap_misses, 0)); | |
6703 | message.push_back(metric); | |
6704 | } | |
f67539c2 TL |
6705 | |
6706 | // dentry lease hit ratio | |
33c7a0ef TL |
6707 | if (_collect_and_send_global_metrics || |
6708 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_DENTRY_LEASE)) { | |
6709 | auto [dlease_hits, dlease_misses, nr] = get_dlease_hit_rates(); | |
6710 | metric = ClientMetricMessage(DentryLeasePayload(dlease_hits, dlease_misses, nr)); | |
6711 | message.push_back(metric); | |
6712 | } | |
f67539c2 TL |
6713 | |
6714 | // opened files | |
33c7a0ef TL |
6715 | if (_collect_and_send_global_metrics || |
6716 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_OPENED_FILES)) { | |
f67539c2 TL |
6717 | auto [opened_files, total_inodes] = get_opened_files_rates(); |
6718 | metric = ClientMetricMessage(OpenedFilesPayload(opened_files, total_inodes)); | |
33c7a0ef | 6719 | message.push_back(metric); |
f67539c2 | 6720 | } |
f67539c2 TL |
6721 | |
6722 | // pinned i_caps | |
33c7a0ef TL |
6723 | if (_collect_and_send_global_metrics || |
6724 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_PINNED_ICAPS)) { | |
f67539c2 TL |
6725 | auto [pinned_icaps, total_inodes] = get_pinned_icaps_rates(); |
6726 | metric = ClientMetricMessage(PinnedIcapsPayload(pinned_icaps, total_inodes)); | |
33c7a0ef | 6727 | message.push_back(metric); |
f67539c2 | 6728 | } |
f67539c2 TL |
6729 | |
6730 | // opened inodes | |
33c7a0ef TL |
6731 | if (_collect_and_send_global_metrics || |
6732 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_OPENED_INODES)) { | |
f67539c2 TL |
6733 | auto [opened_inodes, total_inodes] = get_opened_inodes_rates(); |
6734 | metric = ClientMetricMessage(OpenedInodesPayload(opened_inodes, total_inodes)); | |
33c7a0ef | 6735 | message.push_back(metric); |
f67539c2 | 6736 | } |
f67539c2 | 6737 | |
a4b75251 | 6738 | // read io sizes |
33c7a0ef TL |
6739 | if (_collect_and_send_global_metrics || |
6740 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_READ_IO_SIZES)) { | |
6741 | metric = ClientMetricMessage(ReadIoSizesPayload(total_read_ops, | |
6742 | total_read_size)); | |
6743 | message.push_back(metric); | |
6744 | } | |
a4b75251 TL |
6745 | |
6746 | // write io sizes | |
33c7a0ef TL |
6747 | if (_collect_and_send_global_metrics || |
6748 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_WRITE_IO_SIZES)) { | |
6749 | metric = ClientMetricMessage(WriteIoSizesPayload(total_write_ops, | |
6750 | total_write_size)); | |
6751 | message.push_back(metric); | |
6752 | } | |
a4b75251 | 6753 | |
f67539c2 TL |
6754 | session->con->send_message2(make_message<MClientMetrics>(std::move(message))); |
6755 | } | |
6756 | ||
7c673cae FG |
6757 | void Client::renew_caps() |
6758 | { | |
6759 | ldout(cct, 10) << "renew_caps()" << dendl; | |
6760 | last_cap_renew = ceph_clock_now(); | |
6761 | ||
11fdf7f2 TL |
6762 | for (auto &p : mds_sessions) { |
6763 | ldout(cct, 15) << "renew_caps requesting from mds." << p.first << dendl; | |
6764 | if (mdsmap->get_state(p.first) >= MDSMap::STATE_REJOIN) | |
20effc67 | 6765 | renew_caps(p.second.get()); |
7c673cae FG |
6766 | } |
6767 | } | |
6768 | ||
6769 | void Client::renew_caps(MetaSession *session) | |
6770 | { | |
6771 | ldout(cct, 10) << "renew_caps mds." << session->mds_num << dendl; | |
6772 | session->last_cap_renew_request = ceph_clock_now(); | |
6773 | uint64_t seq = ++session->cap_renew_seq; | |
9f95a23c | 6774 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_RENEWCAPS, seq)); |
7c673cae FG |
6775 | } |
6776 | ||
6777 | ||
6778 | // =============================================================== | |
6779 | // high level (POSIXy) interface | |
6780 | ||
6781 | int Client::_do_lookup(Inode *dir, const string& name, int mask, | |
6782 | InodeRef *target, const UserPerm& perms) | |
6783 | { | |
6784 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
6785 | MetaRequest *req = new MetaRequest(op); | |
6786 | filepath path; | |
6787 | dir->make_nosnap_relative_path(path); | |
6788 | path.push_dentry(name); | |
6789 | req->set_filepath(path); | |
6790 | req->set_inode(dir); | |
6791 | if (cct->_conf->client_debug_getattr_caps && op == CEPH_MDS_OP_LOOKUP) | |
6792 | mask |= DEBUG_GETATTR_CAPS; | |
6793 | req->head.args.getattr.mask = mask; | |
6794 | ||
11fdf7f2 | 6795 | ldout(cct, 10) << __func__ << " on " << path << dendl; |
7c673cae FG |
6796 | |
6797 | int r = make_request(req, perms, target); | |
11fdf7f2 | 6798 | ldout(cct, 10) << __func__ << " res is " << r << dendl; |
7c673cae FG |
6799 | return r; |
6800 | } | |
6801 | ||
f67539c2 TL |
6802 | bool Client::_dentry_valid(const Dentry *dn) |
6803 | { | |
6804 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6805 | ||
6806 | // is dn lease valid? | |
6807 | utime_t now = ceph_clock_now(); | |
6808 | if (dn->lease_mds >= 0 && dn->lease_ttl > now && | |
6809 | mds_sessions.count(dn->lease_mds)) { | |
20effc67 TL |
6810 | auto s = mds_sessions.at(dn->lease_mds); |
6811 | if (s->cap_ttl > now && s->cap_gen == dn->lease_gen) { | |
f67539c2 TL |
6812 | dlease_hit(); |
6813 | return true; | |
6814 | } | |
6815 | ||
20effc67 | 6816 | ldout(cct, 20) << " bad lease, cap_ttl " << s->cap_ttl << ", cap_gen " << s->cap_gen |
f67539c2 TL |
6817 | << " vs lease_gen " << dn->lease_gen << dendl; |
6818 | } | |
6819 | ||
6820 | dlease_miss(); | |
6821 | return false; | |
6822 | } | |
6823 | ||
7c673cae | 6824 | int Client::_lookup(Inode *dir, const string& dname, int mask, InodeRef *target, |
f67539c2 | 6825 | const UserPerm& perms, std::string* alternate_name) |
7c673cae FG |
6826 | { |
6827 | int r = 0; | |
6828 | Dentry *dn = NULL; | |
f67539c2 | 6829 | bool did_lookup_request = false; |
f91f0fd5 TL |
6830 | // can only request shared caps |
6831 | mask &= CEPH_CAP_ANY_SHARED | CEPH_STAT_RSTAT; | |
7c673cae | 6832 | |
7c673cae | 6833 | if (dname == "..") { |
11fdf7f2 TL |
6834 | if (dir->dentries.empty()) { |
6835 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); | |
6836 | filepath path(dir->ino); | |
6837 | req->set_filepath(path); | |
6838 | ||
6839 | InodeRef tmptarget; | |
6840 | int r = make_request(req, perms, &tmptarget, NULL, rand() % mdsmap->get_num_in_mds()); | |
6841 | ||
6842 | if (r == 0) { | |
f91f0fd5 | 6843 | *target = std::move(tmptarget); |
11fdf7f2 TL |
6844 | ldout(cct, 8) << __func__ << " found target " << (*target)->ino << dendl; |
6845 | } else { | |
6846 | *target = dir; | |
6847 | } | |
6848 | } | |
7c673cae FG |
6849 | else |
6850 | *target = dir->get_first_parent()->dir->parent_inode; //dirs can't be hard-linked | |
6851 | goto done; | |
6852 | } | |
6853 | ||
6854 | if (dname == ".") { | |
6855 | *target = dir; | |
6856 | goto done; | |
6857 | } | |
6858 | ||
11fdf7f2 | 6859 | if (!dir->is_dir()) { |
f67539c2 | 6860 | r = -CEPHFS_ENOTDIR; |
11fdf7f2 TL |
6861 | goto done; |
6862 | } | |
6863 | ||
7c673cae | 6864 | if (dname.length() > NAME_MAX) { |
f67539c2 | 6865 | r = -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
6866 | goto done; |
6867 | } | |
6868 | ||
6869 | if (dname == cct->_conf->client_snapdir && | |
6870 | dir->snapid == CEPH_NOSNAP) { | |
6871 | *target = open_snapdir(dir); | |
6872 | goto done; | |
6873 | } | |
6874 | ||
f67539c2 | 6875 | relookup: |
7c673cae FG |
6876 | if (dir->dir && |
6877 | dir->dir->dentries.count(dname)) { | |
6878 | dn = dir->dir->dentries[dname]; | |
6879 | ||
f67539c2 TL |
6880 | ldout(cct, 20) << __func__ << " have " << *dn << " from mds." << dn->lease_mds |
6881 | << " ttl " << dn->lease_ttl << " seq " << dn->lease_seq << dendl; | |
7c673cae | 6882 | |
94b18763 | 6883 | if (!dn->inode || dn->inode->caps_issued_mask(mask, true)) { |
f67539c2 TL |
6884 | if (_dentry_valid(dn)) { |
6885 | // touch this mds's dir cap too, even though we don't _explicitly_ use it here, to | |
6886 | // make trim_caps() behave. | |
6887 | dir->try_touch_cap(dn->lease_mds); | |
6888 | goto hit_dn; | |
7c673cae | 6889 | } |
92f5a8d4 | 6890 | // dir shared caps? |
94b18763 | 6891 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 6892 | if (dn->cap_shared_gen == dir->shared_gen && |
94b18763 | 6893 | (!dn->inode || dn->inode->caps_issued_mask(mask, true))) |
7c673cae FG |
6894 | goto hit_dn; |
6895 | if (!dn->inode && (dir->flags & I_COMPLETE)) { | |
11fdf7f2 | 6896 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " |
7c673cae | 6897 | << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 6898 | return -CEPHFS_ENOENT; |
7c673cae FG |
6899 | } |
6900 | } | |
6901 | } else { | |
6902 | ldout(cct, 20) << " no cap on " << dn->inode->vino() << dendl; | |
6903 | } | |
6904 | } else { | |
6905 | // can we conclude ENOENT locally? | |
94b18763 | 6906 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true) && |
7c673cae | 6907 | (dir->flags & I_COMPLETE)) { |
11fdf7f2 | 6908 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 6909 | return -CEPHFS_ENOENT; |
7c673cae FG |
6910 | } |
6911 | } | |
6912 | ||
f67539c2 TL |
6913 | if (did_lookup_request) { |
6914 | r = 0; | |
6915 | goto done; | |
6916 | } | |
7c673cae | 6917 | r = _do_lookup(dir, dname, mask, target, perms); |
f67539c2 TL |
6918 | did_lookup_request = true; |
6919 | if (r == 0) { | |
6920 | /* complete lookup to get dentry for alternate_name */ | |
6921 | goto relookup; | |
6922 | } else { | |
6923 | goto done; | |
6924 | } | |
6925 | ||
6926 | hit_dn: | |
6927 | if (dn->inode) { | |
7c673cae | 6928 | *target = dn->inode; |
f67539c2 TL |
6929 | if (alternate_name) |
6930 | *alternate_name = dn->alternate_name; | |
7c673cae | 6931 | } else { |
f67539c2 | 6932 | r = -CEPHFS_ENOENT; |
7c673cae FG |
6933 | } |
6934 | touch_dn(dn); | |
f67539c2 | 6935 | goto done; |
7c673cae FG |
6936 | |
6937 | done: | |
6938 | if (r < 0) | |
11fdf7f2 | 6939 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << r << dendl; |
7c673cae | 6940 | else |
11fdf7f2 | 6941 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << **target << dendl; |
7c673cae FG |
6942 | return r; |
6943 | } | |
6944 | ||
6945 | int Client::get_or_create(Inode *dir, const char* name, | |
6946 | Dentry **pdn, bool expect_null) | |
6947 | { | |
6948 | // lookup | |
11fdf7f2 | 6949 | ldout(cct, 20) << __func__ << " " << *dir << " name " << name << dendl; |
7c673cae FG |
6950 | dir->open_dir(); |
6951 | if (dir->dir->dentries.count(name)) { | |
6952 | Dentry *dn = dir->dir->dentries[name]; | |
f67539c2 TL |
6953 | if (_dentry_valid(dn)) { |
6954 | if (expect_null) | |
6955 | return -CEPHFS_EEXIST; | |
7c673cae FG |
6956 | } |
6957 | *pdn = dn; | |
6958 | } else { | |
6959 | // otherwise link up a new one | |
6960 | *pdn = link(dir->dir, name, NULL, NULL); | |
6961 | } | |
6962 | ||
6963 | // success | |
6964 | return 0; | |
6965 | } | |
6966 | ||
f67539c2 TL |
6967 | int Client::walk(std::string_view path, walk_dentry_result* wdr, const UserPerm& perms, bool followsym) |
6968 | { | |
6969 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
6970 | if (!mref_reader.is_state_satisfied()) | |
6971 | return -CEPHFS_ENOTCONN; | |
6972 | ||
6973 | ldout(cct, 10) << __func__ << ": " << path << dendl; | |
6974 | ||
6975 | std::scoped_lock lock(client_lock); | |
6976 | ||
6977 | return path_walk(path, wdr, perms, followsym); | |
6978 | } | |
6979 | ||
7c673cae | 6980 | int Client::path_walk(const filepath& origpath, InodeRef *end, |
b3b6e05e | 6981 | const UserPerm& perms, bool followsym, int mask, InodeRef dirinode) |
f67539c2 TL |
6982 | { |
6983 | walk_dentry_result wdr; | |
b3b6e05e | 6984 | int rc = path_walk(origpath, &wdr, perms, followsym, mask, dirinode); |
f67539c2 TL |
6985 | *end = std::move(wdr.in); |
6986 | return rc; | |
6987 | } | |
6988 | ||
b3b6e05e TL |
6989 | int Client::path_walk(const filepath& origpath, walk_dentry_result* result, const UserPerm& perms, |
6990 | bool followsym, int mask, InodeRef dirinode) | |
7c673cae FG |
6991 | { |
6992 | filepath path = origpath; | |
6993 | InodeRef cur; | |
f67539c2 | 6994 | std::string alternate_name; |
7c673cae FG |
6995 | if (origpath.absolute()) |
6996 | cur = root; | |
b3b6e05e | 6997 | else if (!dirinode) |
7c673cae | 6998 | cur = cwd; |
b3b6e05e TL |
6999 | else { |
7000 | cur = dirinode; | |
7001 | } | |
11fdf7f2 | 7002 | ceph_assert(cur); |
7c673cae | 7003 | |
b3b6e05e | 7004 | ldout(cct, 20) << __func__ << " cur=" << *cur << dendl; |
11fdf7f2 | 7005 | ldout(cct, 10) << __func__ << " " << path << dendl; |
7c673cae FG |
7006 | |
7007 | int symlinks = 0; | |
7008 | ||
7009 | unsigned i=0; | |
7010 | while (i < path.depth() && cur) { | |
7011 | int caps = 0; | |
7012 | const string &dname = path[i]; | |
7013 | ldout(cct, 10) << " " << i << " " << *cur << " " << dname << dendl; | |
7014 | ldout(cct, 20) << " (path is " << path << ")" << dendl; | |
7015 | InodeRef next; | |
7016 | if (cct->_conf->client_permissions) { | |
7017 | int r = may_lookup(cur.get(), perms); | |
7018 | if (r < 0) | |
7019 | return r; | |
7020 | caps = CEPH_CAP_AUTH_SHARED; | |
7021 | } | |
7022 | ||
7023 | /* Get extra requested caps on the last component */ | |
7024 | if (i == (path.depth() - 1)) | |
7025 | caps |= mask; | |
f67539c2 | 7026 | int r = _lookup(cur.get(), dname, caps, &next, perms, &alternate_name); |
7c673cae FG |
7027 | if (r < 0) |
7028 | return r; | |
7029 | // only follow trailing symlink if followsym. always follow | |
7030 | // 'directory' symlinks. | |
7031 | if (next && next->is_symlink()) { | |
7032 | symlinks++; | |
7033 | ldout(cct, 20) << " symlink count " << symlinks << ", value is '" << next->symlink << "'" << dendl; | |
7034 | if (symlinks > MAXSYMLINKS) { | |
f67539c2 | 7035 | return -CEPHFS_ELOOP; |
7c673cae FG |
7036 | } |
7037 | ||
7038 | if (i < path.depth() - 1) { | |
7039 | // dir symlink | |
7040 | // replace consumed components of path with symlink dir target | |
7041 | filepath resolved(next->symlink.c_str()); | |
7042 | resolved.append(path.postfixpath(i + 1)); | |
7043 | path = resolved; | |
7044 | i = 0; | |
7045 | if (next->symlink[0] == '/') { | |
7046 | cur = root; | |
7047 | } | |
7048 | continue; | |
7049 | } else if (followsym) { | |
7050 | if (next->symlink[0] == '/') { | |
7051 | path = next->symlink.c_str(); | |
7052 | i = 0; | |
7053 | // reset position | |
7054 | cur = root; | |
7055 | } else { | |
7056 | filepath more(next->symlink.c_str()); | |
7057 | // we need to remove the symlink component from off of the path | |
7058 | // before adding the target that the symlink points to. remain | |
7059 | // at the same position in the path. | |
7060 | path.pop_dentry(); | |
7061 | path.append(more); | |
7062 | } | |
7063 | continue; | |
7064 | } | |
7065 | } | |
7066 | cur.swap(next); | |
7067 | i++; | |
7068 | } | |
7069 | if (!cur) | |
f67539c2 TL |
7070 | return -CEPHFS_ENOENT; |
7071 | if (result) { | |
7072 | result->in = std::move(cur); | |
7073 | result->alternate_name = std::move(alternate_name); | |
7074 | } | |
7c673cae FG |
7075 | return 0; |
7076 | } | |
7077 | ||
7078 | ||
7079 | // namespace ops | |
7080 | ||
f67539c2 | 7081 | int Client::link(const char *relexisting, const char *relpath, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7082 | { |
f67539c2 TL |
7083 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7084 | if (!mref_reader.is_state_satisfied()) | |
7085 | return -CEPHFS_ENOTCONN; | |
7086 | ||
7c673cae FG |
7087 | tout(cct) << "link" << std::endl; |
7088 | tout(cct) << relexisting << std::endl; | |
7089 | tout(cct) << relpath << std::endl; | |
7090 | ||
7091 | filepath existing(relexisting); | |
7092 | ||
7093 | InodeRef in, dir; | |
f67539c2 TL |
7094 | |
7095 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7096 | int r = path_walk(existing, &in, perm, true); |
7097 | if (r < 0) | |
7098 | return r; | |
7099 | if (std::string(relpath) == "/") { | |
f67539c2 | 7100 | r = -CEPHFS_EEXIST; |
7c673cae FG |
7101 | return r; |
7102 | } | |
7103 | filepath path(relpath); | |
7104 | string name = path.last_dentry(); | |
7105 | path.pop_dentry(); | |
7106 | ||
7107 | r = path_walk(path, &dir, perm, true); | |
7108 | if (r < 0) | |
7109 | return r; | |
7110 | if (cct->_conf->client_permissions) { | |
7111 | if (S_ISDIR(in->mode)) { | |
f67539c2 | 7112 | r = -CEPHFS_EPERM; |
7c673cae FG |
7113 | return r; |
7114 | } | |
7115 | r = may_hardlink(in.get(), perm); | |
7116 | if (r < 0) | |
7117 | return r; | |
7118 | r = may_create(dir.get(), perm); | |
7119 | if (r < 0) | |
7120 | return r; | |
7121 | } | |
f67539c2 | 7122 | r = _link(in.get(), dir.get(), name.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7123 | return r; |
7124 | } | |
7125 | ||
7126 | int Client::unlink(const char *relpath, const UserPerm& perm) | |
b3b6e05e TL |
7127 | { |
7128 | return unlinkat(CEPHFS_AT_FDCWD, relpath, 0, perm); | |
7129 | } | |
7130 | ||
7131 | int Client::unlinkat(int dirfd, const char *relpath, int flags, const UserPerm& perm) | |
7c673cae | 7132 | { |
f67539c2 | 7133 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7134 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7135 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7136 | } |
f67539c2 | 7137 | |
11fdf7f2 | 7138 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7139 | tout(cct) << dirfd << std::endl; |
7c673cae | 7140 | tout(cct) << relpath << std::endl; |
b3b6e05e | 7141 | tout(cct) << flags << std::endl; |
7c673cae | 7142 | |
b3b6e05e TL |
7143 | if (std::string(relpath) == "/") { |
7144 | return flags & AT_REMOVEDIR ? -CEPHFS_EBUSY : -CEPHFS_EISDIR; | |
7145 | } | |
7c673cae FG |
7146 | |
7147 | filepath path(relpath); | |
7148 | string name = path.last_dentry(); | |
7149 | path.pop_dentry(); | |
7150 | InodeRef dir; | |
f67539c2 TL |
7151 | |
7152 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7153 | |
7154 | InodeRef dirinode; | |
7155 | int r = get_fd_inode(dirfd, &dirinode); | |
7156 | if (r < 0) { | |
7157 | return r; | |
7158 | } | |
7159 | ||
7160 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7161 | if (r < 0) { | |
7c673cae | 7162 | return r; |
b3b6e05e | 7163 | } |
7c673cae FG |
7164 | if (cct->_conf->client_permissions) { |
7165 | r = may_delete(dir.get(), name.c_str(), perm); | |
b3b6e05e | 7166 | if (r < 0) { |
7c673cae | 7167 | return r; |
b3b6e05e | 7168 | } |
7c673cae | 7169 | } |
b3b6e05e TL |
7170 | if (flags & AT_REMOVEDIR) { |
7171 | r = _rmdir(dir.get(), name.c_str(), perm); | |
7172 | } else { | |
7173 | r = _unlink(dir.get(), name.c_str(), perm); | |
7174 | } | |
7175 | return r; | |
7c673cae FG |
7176 | } |
7177 | ||
f67539c2 | 7178 | int Client::rename(const char *relfrom, const char *relto, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7179 | { |
f67539c2 TL |
7180 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7181 | if (!mref_reader.is_state_satisfied()) | |
7182 | return -CEPHFS_ENOTCONN; | |
7183 | ||
11fdf7f2 | 7184 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7185 | tout(cct) << relfrom << std::endl; |
7186 | tout(cct) << relto << std::endl; | |
7187 | ||
7188 | if (std::string(relfrom) == "/" || std::string(relto) == "/") | |
f67539c2 | 7189 | return -CEPHFS_EBUSY; |
7c673cae FG |
7190 | |
7191 | filepath from(relfrom); | |
7192 | filepath to(relto); | |
7193 | string fromname = from.last_dentry(); | |
7194 | from.pop_dentry(); | |
7195 | string toname = to.last_dentry(); | |
7196 | to.pop_dentry(); | |
7197 | ||
7198 | InodeRef fromdir, todir; | |
f67539c2 TL |
7199 | |
7200 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7201 | int r = path_walk(from, &fromdir, perm); |
7202 | if (r < 0) | |
7203 | goto out; | |
7204 | r = path_walk(to, &todir, perm); | |
7205 | if (r < 0) | |
7206 | goto out; | |
7207 | ||
7208 | if (cct->_conf->client_permissions) { | |
7209 | int r = may_delete(fromdir.get(), fromname.c_str(), perm); | |
7210 | if (r < 0) | |
7211 | return r; | |
7212 | r = may_delete(todir.get(), toname.c_str(), perm); | |
f67539c2 | 7213 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
7214 | return r; |
7215 | } | |
f67539c2 | 7216 | r = _rename(fromdir.get(), fromname.c_str(), todir.get(), toname.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7217 | out: |
7218 | return r; | |
7219 | } | |
7220 | ||
7221 | // dirs | |
7222 | ||
f67539c2 | 7223 | int Client::mkdir(const char *relpath, mode_t mode, const UserPerm& perm, std::string alternate_name) |
b3b6e05e TL |
7224 | { |
7225 | return mkdirat(CEPHFS_AT_FDCWD, relpath, mode, perm, alternate_name); | |
7226 | } | |
7227 | ||
7228 | int Client::mkdirat(int dirfd, const char *relpath, mode_t mode, const UserPerm& perm, | |
7229 | std::string alternate_name) | |
7c673cae | 7230 | { |
f67539c2 TL |
7231 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7232 | if (!mref_reader.is_state_satisfied()) | |
7233 | return -CEPHFS_ENOTCONN; | |
7234 | ||
11fdf7f2 | 7235 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7236 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7237 | tout(cct) << relpath << std::endl; |
7238 | tout(cct) << mode << std::endl; | |
11fdf7f2 | 7239 | ldout(cct, 10) << __func__ << ": " << relpath << dendl; |
7c673cae | 7240 | |
b3b6e05e | 7241 | if (std::string(relpath) == "/") { |
f67539c2 | 7242 | return -CEPHFS_EEXIST; |
b3b6e05e | 7243 | } |
7c673cae FG |
7244 | |
7245 | filepath path(relpath); | |
7246 | string name = path.last_dentry(); | |
7247 | path.pop_dentry(); | |
7248 | InodeRef dir; | |
f67539c2 TL |
7249 | |
7250 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7251 | |
7252 | InodeRef dirinode; | |
7253 | int r = get_fd_inode(dirfd, &dirinode); | |
7254 | if (r < 0) { | |
7c673cae | 7255 | return r; |
b3b6e05e TL |
7256 | } |
7257 | ||
7258 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7259 | if (r < 0) { | |
7260 | return r; | |
7261 | } | |
7c673cae FG |
7262 | if (cct->_conf->client_permissions) { |
7263 | r = may_create(dir.get(), perm); | |
b3b6e05e | 7264 | if (r < 0) { |
7c673cae | 7265 | return r; |
b3b6e05e | 7266 | } |
7c673cae | 7267 | } |
f67539c2 | 7268 | return _mkdir(dir.get(), name.c_str(), mode, perm, 0, {}, std::move(alternate_name)); |
7c673cae FG |
7269 | } |
7270 | ||
7271 | int Client::mkdirs(const char *relpath, mode_t mode, const UserPerm& perms) | |
7272 | { | |
f67539c2 TL |
7273 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7274 | if (!mref_reader.is_state_satisfied()) | |
7275 | return -CEPHFS_ENOTCONN; | |
7276 | ||
7c673cae | 7277 | ldout(cct, 10) << "Client::mkdirs " << relpath << dendl; |
11fdf7f2 | 7278 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7279 | tout(cct) << relpath << std::endl; |
7280 | tout(cct) << mode << std::endl; | |
7281 | ||
7282 | //get through existing parts of path | |
7283 | filepath path(relpath); | |
7284 | unsigned int i; | |
7285 | int r = 0, caps = 0; | |
7286 | InodeRef cur, next; | |
f67539c2 TL |
7287 | |
7288 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7289 | cur = cwd; |
7290 | for (i=0; i<path.depth(); ++i) { | |
7291 | if (cct->_conf->client_permissions) { | |
7292 | r = may_lookup(cur.get(), perms); | |
7293 | if (r < 0) | |
7294 | break; | |
7295 | caps = CEPH_CAP_AUTH_SHARED; | |
7296 | } | |
7297 | r = _lookup(cur.get(), path[i].c_str(), caps, &next, perms); | |
7298 | if (r < 0) | |
7299 | break; | |
7300 | cur.swap(next); | |
7301 | } | |
f67539c2 | 7302 | if (r!=-CEPHFS_ENOENT) return r; |
11fdf7f2 | 7303 | ldout(cct, 20) << __func__ << " got through " << i << " directories on path " << relpath << dendl; |
7c673cae FG |
7304 | //make new directory at each level |
7305 | for (; i<path.depth(); ++i) { | |
7306 | if (cct->_conf->client_permissions) { | |
7307 | r = may_create(cur.get(), perms); | |
7308 | if (r < 0) | |
7309 | return r; | |
7310 | } | |
7311 | //make new dir | |
7312 | r = _mkdir(cur.get(), path[i].c_str(), mode, perms, &next); | |
c07f9fc5 | 7313 | |
7c673cae | 7314 | //check proper creation/existence |
f67539c2 | 7315 | if(-CEPHFS_EEXIST == r && i < path.depth() - 1) { |
c07f9fc5 FG |
7316 | r = _lookup(cur.get(), path[i].c_str(), CEPH_CAP_AUTH_SHARED, &next, perms); |
7317 | } | |
7318 | if (r < 0) | |
7319 | return r; | |
7c673cae FG |
7320 | //move to new dir and continue |
7321 | cur.swap(next); | |
11fdf7f2 | 7322 | ldout(cct, 20) << __func__ << ": successfully created directory " |
7c673cae FG |
7323 | << filepath(cur->ino).get_path() << dendl; |
7324 | } | |
7325 | return 0; | |
7326 | } | |
7327 | ||
7328 | int Client::rmdir(const char *relpath, const UserPerm& perms) | |
7329 | { | |
b3b6e05e | 7330 | return unlinkat(CEPHFS_AT_FDCWD, relpath, AT_REMOVEDIR, perms); |
7c673cae FG |
7331 | } |
7332 | ||
7333 | int Client::mknod(const char *relpath, mode_t mode, const UserPerm& perms, dev_t rdev) | |
f67539c2 TL |
7334 | { |
7335 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
7336 | if (!mref_reader.is_state_satisfied()) | |
7337 | return -CEPHFS_ENOTCONN; | |
7338 | ||
11fdf7f2 | 7339 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7340 | tout(cct) << relpath << std::endl; |
7341 | tout(cct) << mode << std::endl; | |
7342 | tout(cct) << rdev << std::endl; | |
7343 | ||
7344 | if (std::string(relpath) == "/") | |
f67539c2 | 7345 | return -CEPHFS_EEXIST; |
7c673cae FG |
7346 | |
7347 | filepath path(relpath); | |
7348 | string name = path.last_dentry(); | |
7349 | path.pop_dentry(); | |
7350 | InodeRef dir; | |
f67539c2 TL |
7351 | |
7352 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7353 | int r = path_walk(path, &dir, perms); |
7354 | if (r < 0) | |
7355 | return r; | |
7356 | if (cct->_conf->client_permissions) { | |
7357 | int r = may_create(dir.get(), perms); | |
7358 | if (r < 0) | |
7359 | return r; | |
7360 | } | |
7361 | return _mknod(dir.get(), name.c_str(), mode, rdev, perms); | |
7362 | } | |
7363 | ||
7364 | // symlinks | |
7365 | ||
f67539c2 | 7366 | int Client::symlink(const char *target, const char *relpath, const UserPerm& perms, std::string alternate_name) |
b3b6e05e TL |
7367 | { |
7368 | return symlinkat(target, CEPHFS_AT_FDCWD, relpath, perms, alternate_name); | |
7369 | } | |
7370 | ||
7371 | int Client::symlinkat(const char *target, int dirfd, const char *relpath, const UserPerm& perms, | |
7372 | std::string alternate_name) | |
7c673cae | 7373 | { |
f67539c2 | 7374 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7375 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7376 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7377 | } |
f67539c2 | 7378 | |
11fdf7f2 | 7379 | tout(cct) << __func__ << std::endl; |
7c673cae | 7380 | tout(cct) << target << std::endl; |
b3b6e05e | 7381 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7382 | tout(cct) << relpath << std::endl; |
7383 | ||
b3b6e05e | 7384 | if (std::string(relpath) == "/") { |
f67539c2 | 7385 | return -CEPHFS_EEXIST; |
b3b6e05e | 7386 | } |
7c673cae FG |
7387 | |
7388 | filepath path(relpath); | |
7389 | string name = path.last_dentry(); | |
7390 | path.pop_dentry(); | |
7391 | InodeRef dir; | |
f67539c2 TL |
7392 | |
7393 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7394 | |
7395 | InodeRef dirinode; | |
7396 | int r = get_fd_inode(dirfd, &dirinode); | |
7397 | if (r < 0) { | |
7c673cae | 7398 | return r; |
b3b6e05e TL |
7399 | } |
7400 | r = path_walk(path, &dir, perms, true, 0, dirinode); | |
7401 | if (r < 0) { | |
7402 | return r; | |
7403 | } | |
7c673cae FG |
7404 | if (cct->_conf->client_permissions) { |
7405 | int r = may_create(dir.get(), perms); | |
b3b6e05e | 7406 | if (r < 0) { |
7c673cae | 7407 | return r; |
b3b6e05e | 7408 | } |
7c673cae | 7409 | } |
f67539c2 | 7410 | return _symlink(dir.get(), name.c_str(), target, perms, std::move(alternate_name)); |
7c673cae FG |
7411 | } |
7412 | ||
7413 | int Client::readlink(const char *relpath, char *buf, loff_t size, const UserPerm& perms) | |
7414 | { | |
b3b6e05e TL |
7415 | return readlinkat(CEPHFS_AT_FDCWD, relpath, buf, size, perms); |
7416 | } | |
7417 | ||
7418 | int Client::readlinkat(int dirfd, const char *relpath, char *buf, loff_t size, const UserPerm& perms) { | |
f67539c2 | 7419 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7420 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7421 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7422 | } |
f67539c2 | 7423 | |
11fdf7f2 | 7424 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7425 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7426 | tout(cct) << relpath << std::endl; |
7427 | ||
b3b6e05e | 7428 | InodeRef dirinode; |
f67539c2 | 7429 | std::scoped_lock lock(client_lock); |
b3b6e05e TL |
7430 | int r = get_fd_inode(dirfd, &dirinode); |
7431 | if (r < 0) { | |
7c673cae | 7432 | return r; |
b3b6e05e TL |
7433 | } |
7434 | ||
7435 | InodeRef in; | |
7436 | filepath path(relpath); | |
7437 | r = path_walk(path, &in, perms, false, 0, dirinode); | |
7438 | if (r < 0) { | |
7439 | return r; | |
7440 | } | |
7c673cae FG |
7441 | |
7442 | return _readlink(in.get(), buf, size); | |
7443 | } | |
7444 | ||
7445 | int Client::_readlink(Inode *in, char *buf, size_t size) | |
7446 | { | |
7447 | if (!in->is_symlink()) | |
f67539c2 | 7448 | return -CEPHFS_EINVAL; |
7c673cae FG |
7449 | |
7450 | // copy into buf (at most size bytes) | |
7451 | int r = in->symlink.length(); | |
7452 | if (r > (int)size) | |
7453 | r = size; | |
7454 | memcpy(buf, in->symlink.c_str(), r); | |
7455 | return r; | |
7456 | } | |
7457 | ||
7458 | ||
7459 | // inode stuff | |
7460 | ||
7461 | int Client::_getattr(Inode *in, int mask, const UserPerm& perms, bool force) | |
7462 | { | |
94b18763 | 7463 | bool yes = in->caps_issued_mask(mask, true); |
7c673cae | 7464 | |
11fdf7f2 | 7465 | ldout(cct, 10) << __func__ << " mask " << ccap_string(mask) << " issued=" << yes << dendl; |
7c673cae FG |
7466 | if (yes && !force) |
7467 | return 0; | |
7468 | ||
7469 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
7470 | filepath path; | |
7471 | in->make_nosnap_relative_path(path); | |
7472 | req->set_filepath(path); | |
7473 | req->set_inode(in); | |
7474 | req->head.args.getattr.mask = mask; | |
7475 | ||
7476 | int res = make_request(req, perms); | |
11fdf7f2 | 7477 | ldout(cct, 10) << __func__ << " result=" << res << dendl; |
7c673cae FG |
7478 | return res; |
7479 | } | |
7480 | ||
1d09f67e TL |
7481 | int Client::_getvxattr( |
7482 | Inode *in, | |
7483 | const UserPerm& perms, | |
7484 | const char *xattr_name, | |
7485 | ssize_t size, | |
7486 | void *value, | |
7487 | mds_rank_t rank) | |
7488 | { | |
7489 | if (!xattr_name || strlen(xattr_name) <= 0 || strlen(xattr_name) > 255) { | |
7490 | return -CEPHFS_ENODATA; | |
7491 | } | |
7492 | ||
7493 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETVXATTR); | |
7494 | filepath path; | |
7495 | in->make_nosnap_relative_path(path); | |
7496 | req->set_filepath(path); | |
7497 | req->set_inode(in); | |
7498 | req->set_string2(xattr_name); | |
7499 | ||
7500 | bufferlist bl; | |
7501 | int res = make_request(req, perms, nullptr, nullptr, rank, &bl); | |
7502 | ldout(cct, 10) << __func__ << " result=" << res << dendl; | |
7503 | ||
7504 | if (res < 0) { | |
7505 | return res; | |
7506 | } | |
7507 | ||
7508 | std::string buf; | |
7509 | auto p = bl.cbegin(); | |
7510 | ||
7511 | DECODE_START(1, p); | |
7512 | decode(buf, p); | |
7513 | DECODE_FINISH(p); | |
7514 | ||
7515 | ssize_t len = buf.length(); | |
7516 | ||
7517 | res = len; // refer to man getxattr(2) for output buffer size == 0 | |
7518 | ||
7519 | if (size > 0) { | |
7520 | if (len > size) { | |
7521 | res = -CEPHFS_ERANGE; // insufficient output buffer space | |
7522 | } else { | |
7523 | memcpy(value, buf.c_str(), len); | |
7524 | } | |
7525 | } | |
7526 | return res; | |
7527 | } | |
7528 | ||
7c673cae FG |
7529 | int Client::_do_setattr(Inode *in, struct ceph_statx *stx, int mask, |
7530 | const UserPerm& perms, InodeRef *inp) | |
7531 | { | |
7532 | int issued = in->caps_issued(); | |
20effc67 TL |
7533 | union ceph_mds_request_args args; |
7534 | bool kill_sguid = false; | |
7535 | int inode_drop = 0; | |
7c673cae | 7536 | |
11fdf7f2 | 7537 | ldout(cct, 10) << __func__ << " mask " << mask << " issued " << |
7c673cae FG |
7538 | ccap_string(issued) << dendl; |
7539 | ||
7540 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 7541 | return -CEPHFS_EROFS; |
7c673cae FG |
7542 | } |
7543 | if ((mask & CEPH_SETATTR_SIZE) && | |
f67539c2 TL |
7544 | (uint64_t)stx->stx_size > in->size && |
7545 | is_quota_bytes_exceeded(in, (uint64_t)stx->stx_size - in->size, | |
7c673cae | 7546 | perms)) { |
f67539c2 | 7547 | return -CEPHFS_EDQUOT; |
7c673cae FG |
7548 | } |
7549 | ||
20effc67 TL |
7550 | memset(&args, 0, sizeof(args)); |
7551 | ||
7c673cae FG |
7552 | // make the change locally? |
7553 | if ((in->cap_dirtier_uid >= 0 && perms.uid() != in->cap_dirtier_uid) || | |
7554 | (in->cap_dirtier_gid >= 0 && perms.gid() != in->cap_dirtier_gid)) { | |
7555 | ldout(cct, 10) << __func__ << " caller " << perms.uid() << ":" << perms.gid() | |
7556 | << " != cap dirtier " << in->cap_dirtier_uid << ":" | |
7557 | << in->cap_dirtier_gid << ", forcing sync setattr" | |
7558 | << dendl; | |
7559 | /* | |
7560 | * This works because we implicitly flush the caps as part of the | |
7561 | * request, so the cap update check will happen with the writeback | |
7562 | * cap context, and then the setattr check will happen with the | |
7563 | * caller's context. | |
7564 | * | |
7565 | * In reality this pattern is likely pretty rare (different users | |
7566 | * setattr'ing the same file). If that turns out not to be the | |
7567 | * case later, we can build a more complex pipelined cap writeback | |
7568 | * infrastructure... | |
7569 | */ | |
20effc67 | 7570 | mask |= CEPH_SETATTR_CTIME; |
7c673cae FG |
7571 | } |
7572 | ||
7573 | if (!mask) { | |
7574 | // caller just needs us to bump the ctime | |
7575 | in->ctime = ceph_clock_now(); | |
7576 | in->cap_dirtier_uid = perms.uid(); | |
7577 | in->cap_dirtier_gid = perms.gid(); | |
7578 | if (issued & CEPH_CAP_AUTH_EXCL) | |
28e407b8 | 7579 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7580 | else if (issued & CEPH_CAP_FILE_EXCL) |
28e407b8 | 7581 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
7c673cae | 7582 | else if (issued & CEPH_CAP_XATTR_EXCL) |
28e407b8 | 7583 | in->mark_caps_dirty(CEPH_CAP_XATTR_EXCL); |
7c673cae FG |
7584 | else |
7585 | mask |= CEPH_SETATTR_CTIME; | |
7586 | } | |
7587 | ||
7588 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
20effc67 | 7589 | kill_sguid = mask & (CEPH_SETATTR_SIZE|CEPH_SETATTR_KILL_SGUID); |
7c673cae FG |
7590 | |
7591 | mask &= ~CEPH_SETATTR_KILL_SGUID; | |
20effc67 TL |
7592 | } else if (mask & CEPH_SETATTR_SIZE) { |
7593 | /* If we don't have Ax, then we must ask the server to clear them on truncate */ | |
7594 | mask |= CEPH_SETATTR_KILL_SGUID; | |
7595 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7596 | } | |
7597 | ||
7598 | if (mask & CEPH_SETATTR_UID) { | |
7599 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
7c673cae | 7600 | |
20effc67 | 7601 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { |
7c673cae FG |
7602 | in->ctime = ceph_clock_now(); |
7603 | in->cap_dirtier_uid = perms.uid(); | |
7604 | in->cap_dirtier_gid = perms.gid(); | |
7605 | in->uid = stx->stx_uid; | |
28e407b8 | 7606 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7607 | mask &= ~CEPH_SETATTR_UID; |
7608 | kill_sguid = true; | |
20effc67 TL |
7609 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7610 | in->uid != stx->stx_uid) { | |
7611 | args.setattr.uid = stx->stx_uid; | |
7612 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7613 | } else { | |
7614 | mask &= ~CEPH_SETATTR_UID; | |
7c673cae | 7615 | } |
20effc67 TL |
7616 | } |
7617 | ||
7618 | if (mask & CEPH_SETATTR_GID) { | |
7619 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
7620 | ||
7621 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
7622 | in->ctime = ceph_clock_now(); |
7623 | in->cap_dirtier_uid = perms.uid(); | |
7624 | in->cap_dirtier_gid = perms.gid(); | |
7625 | in->gid = stx->stx_gid; | |
28e407b8 | 7626 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7627 | mask &= ~CEPH_SETATTR_GID; |
7628 | kill_sguid = true; | |
20effc67 TL |
7629 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7630 | in->gid != stx->stx_gid) { | |
7631 | args.setattr.gid = stx->stx_gid; | |
7632 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7633 | } else { | |
7634 | mask &= ~CEPH_SETATTR_GID; | |
7c673cae | 7635 | } |
20effc67 | 7636 | } |
7c673cae | 7637 | |
20effc67 TL |
7638 | if (mask & CEPH_SETATTR_MODE) { |
7639 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
7640 | ||
7641 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
7642 | in->ctime = ceph_clock_now(); |
7643 | in->cap_dirtier_uid = perms.uid(); | |
7644 | in->cap_dirtier_gid = perms.gid(); | |
7645 | in->mode = (in->mode & ~07777) | (stx->stx_mode & 07777); | |
28e407b8 | 7646 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7647 | mask &= ~CEPH_SETATTR_MODE; |
20effc67 TL |
7648 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7649 | in->mode != stx->stx_mode) { | |
7650 | args.setattr.mode = stx->stx_mode; | |
7651 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7652 | } else { | |
7653 | mask &= ~CEPH_SETATTR_MODE; | |
7c673cae | 7654 | } |
20effc67 TL |
7655 | } else if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL) && |
7656 | kill_sguid && S_ISREG(in->mode) && | |
7657 | (in->mode & (S_IXUSR|S_IXGRP|S_IXOTH))) { | |
7658 | /* Must squash the any setuid/setgid bits with an ownership change */ | |
7659 | in->mode &= ~(S_ISUID|S_ISGID); | |
7660 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); | |
7661 | } | |
7662 | ||
7663 | if (mask & CEPH_SETATTR_BTIME) { | |
7664 | ldout(cct,10) << "changing btime to " << in->btime << dendl; | |
7c673cae | 7665 | |
20effc67 | 7666 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { |
7c673cae FG |
7667 | in->ctime = ceph_clock_now(); |
7668 | in->cap_dirtier_uid = perms.uid(); | |
7669 | in->cap_dirtier_gid = perms.gid(); | |
7670 | in->btime = utime_t(stx->stx_btime); | |
28e407b8 | 7671 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7672 | mask &= ~CEPH_SETATTR_BTIME; |
20effc67 TL |
7673 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7674 | in->btime != utime_t(stx->stx_btime)) { | |
7675 | args.setattr.btime = utime_t(stx->stx_btime); | |
7676 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7677 | } else { | |
7678 | mask &= ~CEPH_SETATTR_BTIME; | |
7679 | } | |
7680 | } | |
7681 | ||
7682 | if (mask & CEPH_SETATTR_SIZE) { | |
7683 | if ((uint64_t)stx->stx_size >= mdsmap->get_max_filesize()) { | |
7684 | //too big! | |
7685 | ldout(cct,10) << "unable to set size to " << stx->stx_size << ". Too large!" << dendl; | |
7686 | return -CEPHFS_EFBIG; | |
7687 | } | |
7688 | ||
7689 | ldout(cct,10) << "changing size to " << stx->stx_size << dendl; | |
7690 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL) && | |
7691 | !(mask & CEPH_SETATTR_KILL_SGUID) && | |
7692 | stx->stx_size >= in->size) { | |
7693 | if (stx->stx_size > in->size) { | |
7694 | in->size = in->reported_size = stx->stx_size; | |
7695 | in->cap_dirtier_uid = perms.uid(); | |
7696 | in->cap_dirtier_gid = perms.gid(); | |
7697 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); | |
7698 | mask &= ~(CEPH_SETATTR_SIZE); | |
7699 | mask |= CEPH_SETATTR_MTIME; | |
7700 | } else { | |
7701 | // ignore it when size doesn't change | |
7702 | mask &= ~(CEPH_SETATTR_SIZE); | |
7703 | } | |
7704 | } else { | |
7705 | args.setattr.size = stx->stx_size; | |
7706 | inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
7707 | CEPH_CAP_FILE_WR; | |
7708 | } | |
7709 | } | |
7710 | ||
7711 | if (mask & CEPH_SETATTR_MTIME) { | |
7712 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
7713 | in->mtime = utime_t(stx->stx_mtime); | |
7714 | in->ctime = ceph_clock_now(); | |
7715 | in->cap_dirtier_uid = perms.uid(); | |
7716 | in->cap_dirtier_gid = perms.gid(); | |
7717 | in->time_warp_seq++; | |
7718 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); | |
7719 | mask &= ~CEPH_SETATTR_MTIME; | |
7720 | } else if (in->caps_issued_mask(CEPH_CAP_FILE_WR) && | |
7721 | utime_t(stx->stx_mtime) > in->mtime) { | |
7722 | in->mtime = utime_t(stx->stx_mtime); | |
7723 | in->ctime = ceph_clock_now(); | |
7724 | in->cap_dirtier_uid = perms.uid(); | |
7725 | in->cap_dirtier_gid = perms.gid(); | |
7726 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); | |
7727 | mask &= ~CEPH_SETATTR_MTIME; | |
7728 | } else if (!in->caps_issued_mask(CEPH_CAP_FILE_SHARED) || | |
7729 | in->mtime != utime_t(stx->stx_mtime)) { | |
7730 | args.setattr.mtime = utime_t(stx->stx_mtime); | |
7731 | inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
7732 | CEPH_CAP_FILE_WR; | |
7733 | } else { | |
7734 | mask &= ~CEPH_SETATTR_MTIME; | |
7c673cae | 7735 | } |
7c673cae FG |
7736 | } |
7737 | ||
20effc67 TL |
7738 | if (mask & CEPH_SETATTR_ATIME) { |
7739 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
7740 | in->atime = utime_t(stx->stx_atime); | |
7c673cae FG |
7741 | in->ctime = ceph_clock_now(); |
7742 | in->cap_dirtier_uid = perms.uid(); | |
7743 | in->cap_dirtier_gid = perms.gid(); | |
7744 | in->time_warp_seq++; | |
28e407b8 | 7745 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
20effc67 TL |
7746 | mask &= ~CEPH_SETATTR_ATIME; |
7747 | } else if (in->caps_issued_mask(CEPH_CAP_FILE_WR) && | |
7748 | utime_t(stx->stx_atime) > in->atime) { | |
7749 | in->atime = utime_t(stx->stx_atime); | |
7750 | in->ctime = ceph_clock_now(); | |
7751 | in->cap_dirtier_uid = perms.uid(); | |
7752 | in->cap_dirtier_gid = perms.gid(); | |
7753 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); | |
7754 | mask &= ~CEPH_SETATTR_ATIME; | |
7755 | } else if (!in->caps_issued_mask(CEPH_CAP_FILE_SHARED) || | |
7756 | in->atime != utime_t(stx->stx_atime)) { | |
7757 | args.setattr.atime = utime_t(stx->stx_atime); | |
7758 | inode_drop |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
7759 | CEPH_CAP_FILE_WR; | |
7760 | } else { | |
7761 | mask &= ~CEPH_SETATTR_ATIME; | |
7c673cae FG |
7762 | } |
7763 | } | |
20effc67 | 7764 | |
7c673cae FG |
7765 | if (!mask) { |
7766 | in->change_attr++; | |
7767 | return 0; | |
7768 | } | |
7769 | ||
7c673cae FG |
7770 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETATTR); |
7771 | ||
7772 | filepath path; | |
7773 | ||
7774 | in->make_nosnap_relative_path(path); | |
7775 | req->set_filepath(path); | |
7776 | req->set_inode(in); | |
7777 | ||
20effc67 TL |
7778 | req->head.args = args; |
7779 | req->inode_drop = inode_drop; | |
7c673cae | 7780 | req->head.args.setattr.mask = mask; |
7c673cae FG |
7781 | req->regetattr_mask = mask; |
7782 | ||
7783 | int res = make_request(req, perms, inp); | |
7784 | ldout(cct, 10) << "_setattr result=" << res << dendl; | |
7785 | return res; | |
7786 | } | |
7787 | ||
7788 | /* Note that we only care about attrs that setattr cares about */ | |
7789 | void Client::stat_to_statx(struct stat *st, struct ceph_statx *stx) | |
7790 | { | |
7791 | stx->stx_size = st->st_size; | |
7792 | stx->stx_mode = st->st_mode; | |
7793 | stx->stx_uid = st->st_uid; | |
7794 | stx->stx_gid = st->st_gid; | |
11fdf7f2 TL |
7795 | #ifdef __APPLE__ |
7796 | stx->stx_mtime = st->st_mtimespec; | |
7797 | stx->stx_atime = st->st_atimespec; | |
f67539c2 TL |
7798 | #elif __WIN32 |
7799 | stx->stx_mtime.tv_sec = st->st_mtime; | |
7800 | stx->stx_atime.tv_sec = st->st_atime; | |
11fdf7f2 | 7801 | #else |
7c673cae FG |
7802 | stx->stx_mtime = st->st_mtim; |
7803 | stx->stx_atime = st->st_atim; | |
11fdf7f2 | 7804 | #endif |
7c673cae FG |
7805 | } |
7806 | ||
7807 | int Client::__setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
7808 | const UserPerm& perms, InodeRef *inp) | |
7809 | { | |
7810 | int ret = _do_setattr(in, stx, mask, perms, inp); | |
7811 | if (ret < 0) | |
7812 | return ret; | |
7813 | if (mask & CEPH_SETATTR_MODE) | |
7814 | ret = _posix_acl_chmod(in, stx->stx_mode, perms); | |
7815 | return ret; | |
7816 | } | |
7817 | ||
7818 | int Client::_setattrx(InodeRef &in, struct ceph_statx *stx, int mask, | |
7819 | const UserPerm& perms) | |
7820 | { | |
7821 | mask &= (CEPH_SETATTR_MODE | CEPH_SETATTR_UID | | |
7822 | CEPH_SETATTR_GID | CEPH_SETATTR_MTIME | | |
7823 | CEPH_SETATTR_ATIME | CEPH_SETATTR_SIZE | | |
7824 | CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME); | |
7825 | if (cct->_conf->client_permissions) { | |
7826 | int r = may_setattr(in.get(), stx, mask, perms); | |
7827 | if (r < 0) | |
7828 | return r; | |
7829 | } | |
7830 | return __setattrx(in.get(), stx, mask, perms); | |
7831 | } | |
7832 | ||
7833 | int Client::_setattr(InodeRef &in, struct stat *attr, int mask, | |
7834 | const UserPerm& perms) | |
7835 | { | |
7836 | struct ceph_statx stx; | |
7837 | ||
7838 | stat_to_statx(attr, &stx); | |
7839 | mask &= ~CEPH_SETATTR_BTIME; | |
181888fb FG |
7840 | |
7841 | if ((mask & CEPH_SETATTR_UID) && attr->st_uid == static_cast<uid_t>(-1)) { | |
7842 | mask &= ~CEPH_SETATTR_UID; | |
7843 | } | |
7844 | if ((mask & CEPH_SETATTR_GID) && attr->st_gid == static_cast<uid_t>(-1)) { | |
7845 | mask &= ~CEPH_SETATTR_GID; | |
7846 | } | |
7847 | ||
7c673cae FG |
7848 | return _setattrx(in, &stx, mask, perms); |
7849 | } | |
7850 | ||
7851 | int Client::setattr(const char *relpath, struct stat *attr, int mask, | |
7852 | const UserPerm& perms) | |
7853 | { | |
f67539c2 TL |
7854 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7855 | if (!mref_reader.is_state_satisfied()) | |
7856 | return -CEPHFS_ENOTCONN; | |
7857 | ||
11fdf7f2 | 7858 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7859 | tout(cct) << relpath << std::endl; |
7860 | tout(cct) << mask << std::endl; | |
7861 | ||
7862 | filepath path(relpath); | |
7863 | InodeRef in; | |
f67539c2 TL |
7864 | |
7865 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7866 | int r = path_walk(path, &in, perms); |
7867 | if (r < 0) | |
7868 | return r; | |
7869 | return _setattr(in, attr, mask, perms); | |
7870 | } | |
7871 | ||
7872 | int Client::setattrx(const char *relpath, struct ceph_statx *stx, int mask, | |
7873 | const UserPerm& perms, int flags) | |
7874 | { | |
f67539c2 TL |
7875 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7876 | if (!mref_reader.is_state_satisfied()) | |
7877 | return -CEPHFS_ENOTCONN; | |
7878 | ||
11fdf7f2 | 7879 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7880 | tout(cct) << relpath << std::endl; |
7881 | tout(cct) << mask << std::endl; | |
7882 | ||
7883 | filepath path(relpath); | |
7884 | InodeRef in; | |
f67539c2 TL |
7885 | |
7886 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7887 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW)); |
7888 | if (r < 0) | |
7889 | return r; | |
7890 | return _setattrx(in, stx, mask, perms); | |
7891 | } | |
7892 | ||
7893 | int Client::fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms) | |
7894 | { | |
f67539c2 TL |
7895 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7896 | if (!mref_reader.is_state_satisfied()) | |
7897 | return -CEPHFS_ENOTCONN; | |
7898 | ||
11fdf7f2 | 7899 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7900 | tout(cct) << fd << std::endl; |
7901 | tout(cct) << mask << std::endl; | |
7902 | ||
f67539c2 | 7903 | std::scoped_lock lock(client_lock); |
7c673cae FG |
7904 | Fh *f = get_filehandle(fd); |
7905 | if (!f) | |
f67539c2 | 7906 | return -CEPHFS_EBADF; |
7c673cae FG |
7907 | #if defined(__linux__) && defined(O_PATH) |
7908 | if (f->flags & O_PATH) | |
f67539c2 | 7909 | return -CEPHFS_EBADF; |
7c673cae FG |
7910 | #endif |
7911 | return _setattr(f->inode, attr, mask, perms); | |
7912 | } | |
7913 | ||
7914 | int Client::fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms) | |
7915 | { | |
f67539c2 TL |
7916 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7917 | if (!mref_reader.is_state_satisfied()) | |
7918 | return -CEPHFS_ENOTCONN; | |
7919 | ||
11fdf7f2 | 7920 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7921 | tout(cct) << fd << std::endl; |
7922 | tout(cct) << mask << std::endl; | |
7923 | ||
f67539c2 | 7924 | std::scoped_lock lock(client_lock); |
7c673cae FG |
7925 | Fh *f = get_filehandle(fd); |
7926 | if (!f) | |
f67539c2 | 7927 | return -CEPHFS_EBADF; |
7c673cae FG |
7928 | #if defined(__linux__) && defined(O_PATH) |
7929 | if (f->flags & O_PATH) | |
f67539c2 | 7930 | return -CEPHFS_EBADF; |
7c673cae FG |
7931 | #endif |
7932 | return _setattrx(f->inode, stx, mask, perms); | |
7933 | } | |
7934 | ||
7935 | int Client::stat(const char *relpath, struct stat *stbuf, const UserPerm& perms, | |
7936 | frag_info_t *dirstat, int mask) | |
7937 | { | |
f67539c2 TL |
7938 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7939 | if (!mref_reader.is_state_satisfied()) | |
7940 | return -CEPHFS_ENOTCONN; | |
7941 | ||
11fdf7f2 | 7942 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7943 | tout(cct) << "stat" << std::endl; |
7944 | tout(cct) << relpath << std::endl; | |
181888fb | 7945 | |
7c673cae FG |
7946 | filepath path(relpath); |
7947 | InodeRef in; | |
f67539c2 TL |
7948 | |
7949 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7950 | int r = path_walk(path, &in, perms, true, mask); |
7951 | if (r < 0) | |
7952 | return r; | |
7953 | r = _getattr(in, mask, perms); | |
7954 | if (r < 0) { | |
11fdf7f2 | 7955 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7956 | return r; |
7957 | } | |
7958 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 7959 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7960 | return r; |
7961 | } | |
7962 | ||
7963 | unsigned Client::statx_to_mask(unsigned int flags, unsigned int want) | |
7964 | { | |
7965 | unsigned mask = 0; | |
7966 | ||
7967 | /* if NO_ATTR_SYNC is set, then we don't need any -- just use what's in cache */ | |
7968 | if (flags & AT_NO_ATTR_SYNC) | |
7969 | goto out; | |
7970 | ||
7971 | /* Always set PIN to distinguish from AT_NO_ATTR_SYNC case */ | |
7972 | mask |= CEPH_CAP_PIN; | |
7973 | if (want & (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7974 | mask |= CEPH_CAP_AUTH_SHARED; | |
7975 | if (want & (CEPH_STATX_NLINK|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7976 | mask |= CEPH_CAP_LINK_SHARED; | |
adb31ebb | 7977 | if (want & (CEPH_STATX_NLINK|CEPH_STATX_ATIME|CEPH_STATX_MTIME|CEPH_STATX_CTIME|CEPH_STATX_SIZE|CEPH_STATX_BLOCKS|CEPH_STATX_VERSION)) |
7c673cae FG |
7978 | mask |= CEPH_CAP_FILE_SHARED; |
7979 | if (want & (CEPH_STATX_VERSION|CEPH_STATX_CTIME)) | |
7980 | mask |= CEPH_CAP_XATTR_SHARED; | |
7981 | out: | |
7982 | return mask; | |
7983 | } | |
7984 | ||
7985 | int Client::statx(const char *relpath, struct ceph_statx *stx, | |
7986 | const UserPerm& perms, | |
7987 | unsigned int want, unsigned int flags) | |
7988 | { | |
b3b6e05e | 7989 | return statxat(CEPHFS_AT_FDCWD, relpath, stx, perms, want, flags); |
7c673cae FG |
7990 | } |
7991 | ||
7992 | int Client::lstat(const char *relpath, struct stat *stbuf, | |
7993 | const UserPerm& perms, frag_info_t *dirstat, int mask) | |
7994 | { | |
f67539c2 TL |
7995 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7996 | if (!mref_reader.is_state_satisfied()) | |
7997 | return -CEPHFS_ENOTCONN; | |
7998 | ||
11fdf7f2 | 7999 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
11fdf7f2 | 8000 | tout(cct) << __func__ << std::endl; |
7c673cae | 8001 | tout(cct) << relpath << std::endl; |
181888fb | 8002 | |
7c673cae FG |
8003 | filepath path(relpath); |
8004 | InodeRef in; | |
f67539c2 TL |
8005 | |
8006 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8007 | // don't follow symlinks |
8008 | int r = path_walk(path, &in, perms, false, mask); | |
8009 | if (r < 0) | |
8010 | return r; | |
8011 | r = _getattr(in, mask, perms); | |
8012 | if (r < 0) { | |
11fdf7f2 | 8013 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
8014 | return r; |
8015 | } | |
8016 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 8017 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
8018 | return r; |
8019 | } | |
8020 | ||
8021 | int Client::fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat, nest_info_t *rstat) | |
8022 | { | |
11fdf7f2 | 8023 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
8024 | << " mode 0" << oct << in->mode << dec |
8025 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
8026 | memset(st, 0, sizeof(struct stat)); | |
8027 | if (use_faked_inos()) | |
8028 | st->st_ino = in->faked_ino; | |
8029 | else | |
8030 | st->st_ino = in->ino; | |
8031 | st->st_dev = in->snapid; | |
8032 | st->st_mode = in->mode; | |
8033 | st->st_rdev = in->rdev; | |
28e407b8 AA |
8034 | if (in->is_dir()) { |
8035 | switch (in->nlink) { | |
8036 | case 0: | |
8037 | st->st_nlink = 0; /* dir is unlinked */ | |
8038 | break; | |
8039 | case 1: | |
8040 | st->st_nlink = 1 /* parent dentry */ | |
8041 | + 1 /* <dir>/. */ | |
8042 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
8043 | break; | |
8044 | default: | |
8045 | ceph_abort(); | |
8046 | } | |
8047 | } else { | |
8048 | st->st_nlink = in->nlink; | |
8049 | } | |
7c673cae FG |
8050 | st->st_uid = in->uid; |
8051 | st->st_gid = in->gid; | |
8052 | if (in->ctime > in->mtime) { | |
8053 | stat_set_ctime_sec(st, in->ctime.sec()); | |
8054 | stat_set_ctime_nsec(st, in->ctime.nsec()); | |
8055 | } else { | |
8056 | stat_set_ctime_sec(st, in->mtime.sec()); | |
8057 | stat_set_ctime_nsec(st, in->mtime.nsec()); | |
8058 | } | |
8059 | stat_set_atime_sec(st, in->atime.sec()); | |
8060 | stat_set_atime_nsec(st, in->atime.nsec()); | |
8061 | stat_set_mtime_sec(st, in->mtime.sec()); | |
8062 | stat_set_mtime_nsec(st, in->mtime.nsec()); | |
8063 | if (in->is_dir()) { | |
8064 | if (cct->_conf->client_dirsize_rbytes) | |
8065 | st->st_size = in->rstat.rbytes; | |
8066 | else | |
8067 | st->st_size = in->dirstat.size(); | |
f67539c2 TL |
8068 | // The Windows "stat" structure provides just a subset of the fields that are |
8069 | // available on Linux. | |
8070 | #ifndef _WIN32 | |
7c673cae | 8071 | st->st_blocks = 1; |
f67539c2 | 8072 | #endif |
7c673cae FG |
8073 | } else { |
8074 | st->st_size = in->size; | |
f67539c2 | 8075 | #ifndef _WIN32 |
7c673cae | 8076 | st->st_blocks = (in->size + 511) >> 9; |
f67539c2 | 8077 | #endif |
7c673cae | 8078 | } |
f67539c2 | 8079 | #ifndef _WIN32 |
11fdf7f2 | 8080 | st->st_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
f67539c2 | 8081 | #endif |
7c673cae FG |
8082 | |
8083 | if (dirstat) | |
8084 | *dirstat = in->dirstat; | |
8085 | if (rstat) | |
8086 | *rstat = in->rstat; | |
8087 | ||
8088 | return in->caps_issued(); | |
8089 | } | |
8090 | ||
8091 | void Client::fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx) | |
8092 | { | |
11fdf7f2 | 8093 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
8094 | << " mode 0" << oct << in->mode << dec |
8095 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
8096 | memset(stx, 0, sizeof(struct ceph_statx)); | |
8097 | ||
8098 | /* | |
8099 | * If mask is 0, then the caller set AT_NO_ATTR_SYNC. Reset the mask | |
8100 | * so that all bits are set. | |
8101 | */ | |
8102 | if (!mask) | |
8103 | mask = ~0; | |
8104 | ||
8105 | /* These are always considered to be available */ | |
8106 | stx->stx_dev = in->snapid; | |
11fdf7f2 | 8107 | stx->stx_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
7c673cae FG |
8108 | |
8109 | /* Type bits are always set, even when CEPH_STATX_MODE is not */ | |
8110 | stx->stx_mode = S_IFMT & in->mode; | |
8111 | stx->stx_ino = use_faked_inos() ? in->faked_ino : (ino_t)in->ino; | |
8112 | stx->stx_rdev = in->rdev; | |
8113 | stx->stx_mask |= (CEPH_STATX_INO|CEPH_STATX_RDEV); | |
8114 | ||
8115 | if (mask & CEPH_CAP_AUTH_SHARED) { | |
8116 | stx->stx_uid = in->uid; | |
8117 | stx->stx_gid = in->gid; | |
8118 | stx->stx_mode = in->mode; | |
8119 | in->btime.to_timespec(&stx->stx_btime); | |
8120 | stx->stx_mask |= (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME); | |
8121 | } | |
8122 | ||
8123 | if (mask & CEPH_CAP_LINK_SHARED) { | |
28e407b8 AA |
8124 | if (in->is_dir()) { |
8125 | switch (in->nlink) { | |
8126 | case 0: | |
8127 | stx->stx_nlink = 0; /* dir is unlinked */ | |
8128 | break; | |
8129 | case 1: | |
8130 | stx->stx_nlink = 1 /* parent dentry */ | |
8131 | + 1 /* <dir>/. */ | |
8132 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
8133 | break; | |
8134 | default: | |
8135 | ceph_abort(); | |
8136 | } | |
8137 | } else { | |
8138 | stx->stx_nlink = in->nlink; | |
8139 | } | |
7c673cae FG |
8140 | stx->stx_mask |= CEPH_STATX_NLINK; |
8141 | } | |
8142 | ||
8143 | if (mask & CEPH_CAP_FILE_SHARED) { | |
8144 | ||
8145 | in->atime.to_timespec(&stx->stx_atime); | |
8146 | in->mtime.to_timespec(&stx->stx_mtime); | |
8147 | ||
8148 | if (in->is_dir()) { | |
8149 | if (cct->_conf->client_dirsize_rbytes) | |
8150 | stx->stx_size = in->rstat.rbytes; | |
8151 | else | |
8152 | stx->stx_size = in->dirstat.size(); | |
8153 | stx->stx_blocks = 1; | |
8154 | } else { | |
8155 | stx->stx_size = in->size; | |
8156 | stx->stx_blocks = (in->size + 511) >> 9; | |
8157 | } | |
8158 | stx->stx_mask |= (CEPH_STATX_ATIME|CEPH_STATX_MTIME| | |
8159 | CEPH_STATX_SIZE|CEPH_STATX_BLOCKS); | |
8160 | } | |
8161 | ||
8162 | /* Change time and change_attr both require all shared caps to view */ | |
8163 | if ((mask & CEPH_STAT_CAP_INODE_ALL) == CEPH_STAT_CAP_INODE_ALL) { | |
8164 | stx->stx_version = in->change_attr; | |
8165 | if (in->ctime > in->mtime) | |
8166 | in->ctime.to_timespec(&stx->stx_ctime); | |
8167 | else | |
8168 | in->mtime.to_timespec(&stx->stx_ctime); | |
8169 | stx->stx_mask |= (CEPH_STATX_CTIME|CEPH_STATX_VERSION); | |
8170 | } | |
8171 | ||
8172 | } | |
8173 | ||
8174 | void Client::touch_dn(Dentry *dn) | |
8175 | { | |
8176 | lru.lru_touch(dn); | |
8177 | } | |
8178 | ||
8179 | int Client::chmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
8180 | { | |
b3b6e05e | 8181 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, 0, perms); |
7c673cae FG |
8182 | } |
8183 | ||
8184 | int Client::fchmod(int fd, mode_t mode, const UserPerm& perms) | |
8185 | { | |
f67539c2 TL |
8186 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8187 | if (!mref_reader.is_state_satisfied()) | |
8188 | return -CEPHFS_ENOTCONN; | |
8189 | ||
11fdf7f2 | 8190 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8191 | tout(cct) << fd << std::endl; |
8192 | tout(cct) << mode << std::endl; | |
181888fb | 8193 | |
f67539c2 | 8194 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8195 | Fh *f = get_filehandle(fd); |
8196 | if (!f) | |
f67539c2 | 8197 | return -CEPHFS_EBADF; |
7c673cae FG |
8198 | #if defined(__linux__) && defined(O_PATH) |
8199 | if (f->flags & O_PATH) | |
f67539c2 | 8200 | return -CEPHFS_EBADF; |
7c673cae FG |
8201 | #endif |
8202 | struct stat attr; | |
8203 | attr.st_mode = mode; | |
8204 | return _setattr(f->inode, &attr, CEPH_SETATTR_MODE, perms); | |
8205 | } | |
8206 | ||
b3b6e05e TL |
8207 | int Client::chmodat(int dirfd, const char *relpath, mode_t mode, int flags, |
8208 | const UserPerm& perms) { | |
f67539c2 | 8209 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8210 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8211 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8212 | } |
f67539c2 | 8213 | |
11fdf7f2 | 8214 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8215 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8216 | tout(cct) << relpath << std::endl; |
8217 | tout(cct) << mode << std::endl; | |
b3b6e05e | 8218 | tout(cct) << flags << std::endl; |
181888fb | 8219 | |
7c673cae FG |
8220 | filepath path(relpath); |
8221 | InodeRef in; | |
b3b6e05e | 8222 | InodeRef dirinode; |
f67539c2 TL |
8223 | |
8224 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8225 | int r = get_fd_inode(dirfd, &dirinode); |
8226 | if (r < 0) { | |
8227 | return r; | |
8228 | } | |
8229 | ||
8230 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8231 | if (r < 0) { | |
7c673cae | 8232 | return r; |
b3b6e05e | 8233 | } |
7c673cae FG |
8234 | struct stat attr; |
8235 | attr.st_mode = mode; | |
8236 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
8237 | } | |
8238 | ||
b3b6e05e TL |
8239 | int Client::lchmod(const char *relpath, mode_t mode, const UserPerm& perms) |
8240 | { | |
8241 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, AT_SYMLINK_NOFOLLOW, perms); | |
8242 | } | |
8243 | ||
7c673cae FG |
8244 | int Client::chown(const char *relpath, uid_t new_uid, gid_t new_gid, |
8245 | const UserPerm& perms) | |
8246 | { | |
b3b6e05e | 8247 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, 0, perms); |
7c673cae FG |
8248 | } |
8249 | ||
8250 | int Client::fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms) | |
8251 | { | |
f67539c2 TL |
8252 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8253 | if (!mref_reader.is_state_satisfied()) | |
8254 | return -CEPHFS_ENOTCONN; | |
8255 | ||
11fdf7f2 | 8256 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8257 | tout(cct) << fd << std::endl; |
8258 | tout(cct) << new_uid << std::endl; | |
8259 | tout(cct) << new_gid << std::endl; | |
181888fb | 8260 | |
f67539c2 | 8261 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8262 | Fh *f = get_filehandle(fd); |
8263 | if (!f) | |
f67539c2 | 8264 | return -CEPHFS_EBADF; |
7c673cae FG |
8265 | #if defined(__linux__) && defined(O_PATH) |
8266 | if (f->flags & O_PATH) | |
f67539c2 | 8267 | return -CEPHFS_EBADF; |
7c673cae FG |
8268 | #endif |
8269 | struct stat attr; | |
8270 | attr.st_uid = new_uid; | |
8271 | attr.st_gid = new_gid; | |
8272 | int mask = 0; | |
8273 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
8274 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
8275 | return _setattr(f->inode, &attr, mask, perms); | |
8276 | } | |
8277 | ||
8278 | int Client::lchown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
8279 | const UserPerm& perms) | |
8280 | { | |
b3b6e05e TL |
8281 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, AT_SYMLINK_NOFOLLOW, perms); |
8282 | } | |
8283 | ||
8284 | int Client::chownat(int dirfd, const char *relpath, uid_t new_uid, gid_t new_gid, | |
8285 | int flags, const UserPerm& perms) { | |
f67539c2 | 8286 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8287 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8288 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8289 | } |
f67539c2 | 8290 | |
11fdf7f2 | 8291 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8292 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8293 | tout(cct) << relpath << std::endl; |
8294 | tout(cct) << new_uid << std::endl; | |
8295 | tout(cct) << new_gid << std::endl; | |
b3b6e05e | 8296 | tout(cct) << flags << std::endl; |
181888fb | 8297 | |
7c673cae FG |
8298 | filepath path(relpath); |
8299 | InodeRef in; | |
b3b6e05e | 8300 | InodeRef dirinode; |
f67539c2 TL |
8301 | |
8302 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8303 | int r = get_fd_inode(dirfd, &dirinode); |
8304 | if (r < 0) { | |
7c673cae | 8305 | return r; |
b3b6e05e TL |
8306 | } |
8307 | ||
8308 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8309 | if (r < 0) { | |
8310 | return r; | |
8311 | } | |
7c673cae FG |
8312 | struct stat attr; |
8313 | attr.st_uid = new_uid; | |
8314 | attr.st_gid = new_gid; | |
b3b6e05e | 8315 | return _setattr(in, &attr, CEPH_SETATTR_UID|CEPH_SETATTR_GID, perms); |
7c673cae FG |
8316 | } |
8317 | ||
11fdf7f2 TL |
8318 | static void attr_set_atime_and_mtime(struct stat *attr, |
8319 | const utime_t &atime, | |
8320 | const utime_t &mtime) | |
8321 | { | |
8322 | stat_set_atime_sec(attr, atime.tv.tv_sec); | |
8323 | stat_set_atime_nsec(attr, atime.tv.tv_nsec); | |
8324 | stat_set_mtime_sec(attr, mtime.tv.tv_sec); | |
8325 | stat_set_mtime_nsec(attr, mtime.tv.tv_nsec); | |
8326 | } | |
8327 | ||
8328 | // for [l]utime() invoke the timeval variant as the timespec | |
8329 | // variant are not yet implemented. for futime[s](), invoke | |
8330 | // the timespec variant. | |
7c673cae FG |
8331 | int Client::utime(const char *relpath, struct utimbuf *buf, |
8332 | const UserPerm& perms) | |
8333 | { | |
11fdf7f2 TL |
8334 | struct timeval tv[2]; |
8335 | tv[0].tv_sec = buf->actime; | |
8336 | tv[0].tv_usec = 0; | |
8337 | tv[1].tv_sec = buf->modtime; | |
8338 | tv[1].tv_usec = 0; | |
8339 | ||
8340 | return utimes(relpath, tv, perms); | |
8341 | } | |
8342 | ||
8343 | int Client::lutime(const char *relpath, struct utimbuf *buf, | |
8344 | const UserPerm& perms) | |
8345 | { | |
8346 | struct timeval tv[2]; | |
8347 | tv[0].tv_sec = buf->actime; | |
8348 | tv[0].tv_usec = 0; | |
8349 | tv[1].tv_sec = buf->modtime; | |
8350 | tv[1].tv_usec = 0; | |
8351 | ||
8352 | return lutimes(relpath, tv, perms); | |
8353 | } | |
8354 | ||
8355 | int Client::futime(int fd, struct utimbuf *buf, const UserPerm& perms) | |
8356 | { | |
8357 | struct timespec ts[2]; | |
8358 | ts[0].tv_sec = buf->actime; | |
8359 | ts[0].tv_nsec = 0; | |
8360 | ts[1].tv_sec = buf->modtime; | |
8361 | ts[1].tv_nsec = 0; | |
8362 | ||
8363 | return futimens(fd, ts, perms); | |
8364 | } | |
8365 | ||
8366 | int Client::utimes(const char *relpath, struct timeval times[2], | |
8367 | const UserPerm& perms) | |
8368 | { | |
f67539c2 TL |
8369 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8370 | if (!mref_reader.is_state_satisfied()) | |
8371 | return -CEPHFS_ENOTCONN; | |
8372 | ||
11fdf7f2 | 8373 | tout(cct) << __func__ << std::endl; |
7c673cae | 8374 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8375 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8376 | << std::endl; | |
8377 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8378 | << std::endl; | |
181888fb | 8379 | |
7c673cae FG |
8380 | filepath path(relpath); |
8381 | InodeRef in; | |
f67539c2 TL |
8382 | |
8383 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8384 | int r = path_walk(path, &in, perms); |
8385 | if (r < 0) | |
8386 | return r; | |
8387 | struct stat attr; | |
11fdf7f2 TL |
8388 | utime_t atime(times[0]); |
8389 | utime_t mtime(times[1]); | |
8390 | ||
8391 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8392 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8393 | } | |
8394 | ||
11fdf7f2 TL |
8395 | int Client::lutimes(const char *relpath, struct timeval times[2], |
8396 | const UserPerm& perms) | |
7c673cae | 8397 | { |
f67539c2 TL |
8398 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8399 | if (!mref_reader.is_state_satisfied()) | |
8400 | return -CEPHFS_ENOTCONN; | |
8401 | ||
11fdf7f2 | 8402 | tout(cct) << __func__ << std::endl; |
7c673cae | 8403 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8404 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8405 | << std::endl; | |
8406 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8407 | << std::endl; | |
181888fb | 8408 | |
7c673cae FG |
8409 | filepath path(relpath); |
8410 | InodeRef in; | |
f67539c2 TL |
8411 | |
8412 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8413 | int r = path_walk(path, &in, perms, false); |
8414 | if (r < 0) | |
8415 | return r; | |
8416 | struct stat attr; | |
11fdf7f2 TL |
8417 | utime_t atime(times[0]); |
8418 | utime_t mtime(times[1]); | |
8419 | ||
8420 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8421 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8422 | } | |
8423 | ||
11fdf7f2 TL |
8424 | int Client::futimes(int fd, struct timeval times[2], const UserPerm& perms) |
8425 | { | |
8426 | struct timespec ts[2]; | |
8427 | ts[0].tv_sec = times[0].tv_sec; | |
8428 | ts[0].tv_nsec = times[0].tv_usec * 1000; | |
8429 | ts[1].tv_sec = times[1].tv_sec; | |
8430 | ts[1].tv_nsec = times[1].tv_usec * 1000; | |
8431 | ||
8432 | return futimens(fd, ts, perms); | |
8433 | } | |
8434 | ||
8435 | int Client::futimens(int fd, struct timespec times[2], const UserPerm& perms) | |
8436 | { | |
f67539c2 TL |
8437 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8438 | if (!mref_reader.is_state_satisfied()) | |
8439 | return -CEPHFS_ENOTCONN; | |
8440 | ||
11fdf7f2 TL |
8441 | tout(cct) << __func__ << std::endl; |
8442 | tout(cct) << fd << std::endl; | |
8443 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8444 | << std::endl; | |
8445 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8446 | << std::endl; | |
8447 | ||
f67539c2 | 8448 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
8449 | Fh *f = get_filehandle(fd); |
8450 | if (!f) | |
f67539c2 | 8451 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8452 | #if defined(__linux__) && defined(O_PATH) |
8453 | if (f->flags & O_PATH) | |
f67539c2 | 8454 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8455 | #endif |
8456 | struct stat attr; | |
8457 | utime_t atime(times[0]); | |
8458 | utime_t mtime(times[1]); | |
8459 | ||
8460 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8461 | return _setattr(f->inode, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8462 | } | |
8463 | ||
b3b6e05e TL |
8464 | int Client::utimensat(int dirfd, const char *relpath, struct timespec times[2], int flags, |
8465 | const UserPerm& perms) { | |
8466 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8467 | if (!mref_reader.is_state_satisfied()) { | |
8468 | return -CEPHFS_ENOTCONN; | |
8469 | } | |
8470 | ||
8471 | tout(cct) << __func__ << std::endl; | |
8472 | tout(cct) << dirfd << std::endl; | |
8473 | tout(cct) << relpath << std::endl; | |
8474 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8475 | << std::endl; | |
8476 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8477 | << std::endl; | |
8478 | tout(cct) << flags << std::endl; | |
8479 | ||
8480 | filepath path(relpath); | |
8481 | InodeRef in; | |
8482 | InodeRef dirinode; | |
8483 | ||
8484 | std::scoped_lock lock(client_lock); | |
8485 | int r = get_fd_inode(dirfd, &dirinode); | |
8486 | if (r < 0) { | |
8487 | return r; | |
8488 | } | |
8489 | ||
8490 | #if defined(__linux__) && defined(O_PATH) | |
8491 | if (flags & O_PATH) { | |
8492 | return -CEPHFS_EBADF; | |
8493 | } | |
8494 | #endif | |
8495 | ||
8496 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8497 | if (r < 0) { | |
8498 | return r; | |
8499 | } | |
8500 | struct stat attr; | |
8501 | utime_t atime(times[0]); | |
8502 | utime_t mtime(times[1]); | |
8503 | ||
8504 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8505 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8506 | } | |
8507 | ||
7c673cae FG |
8508 | int Client::flock(int fd, int operation, uint64_t owner) |
8509 | { | |
f67539c2 TL |
8510 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8511 | if (!mref_reader.is_state_satisfied()) | |
8512 | return -CEPHFS_ENOTCONN; | |
8513 | ||
11fdf7f2 | 8514 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8515 | tout(cct) << fd << std::endl; |
8516 | tout(cct) << operation << std::endl; | |
8517 | tout(cct) << owner << std::endl; | |
181888fb | 8518 | |
f67539c2 | 8519 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8520 | Fh *f = get_filehandle(fd); |
8521 | if (!f) | |
f67539c2 | 8522 | return -CEPHFS_EBADF; |
7c673cae FG |
8523 | |
8524 | return _flock(f, operation, owner); | |
8525 | } | |
8526 | ||
8527 | int Client::opendir(const char *relpath, dir_result_t **dirpp, const UserPerm& perms) | |
8528 | { | |
f67539c2 TL |
8529 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8530 | if (!mref_reader.is_state_satisfied()) | |
8531 | return -CEPHFS_ENOTCONN; | |
8532 | ||
11fdf7f2 | 8533 | tout(cct) << __func__ << std::endl; |
7c673cae | 8534 | tout(cct) << relpath << std::endl; |
181888fb | 8535 | |
7c673cae FG |
8536 | filepath path(relpath); |
8537 | InodeRef in; | |
f67539c2 TL |
8538 | |
8539 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8540 | int r = path_walk(path, &in, perms, true); |
8541 | if (r < 0) | |
8542 | return r; | |
8543 | if (cct->_conf->client_permissions) { | |
8544 | int r = may_open(in.get(), O_RDONLY, perms); | |
8545 | if (r < 0) | |
8546 | return r; | |
8547 | } | |
8548 | r = _opendir(in.get(), dirpp, perms); | |
8549 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
f67539c2 TL |
8550 | if (r != -CEPHFS_ENOTDIR) |
8551 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
7c673cae FG |
8552 | return r; |
8553 | } | |
8554 | ||
b3b6e05e TL |
8555 | int Client::fdopendir(int dirfd, dir_result_t **dirpp, const UserPerm &perms) { |
8556 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8557 | if (!mref_reader.is_state_satisfied()) { | |
8558 | return -CEPHFS_ENOTCONN; | |
8559 | } | |
8560 | ||
8561 | tout(cct) << __func__ << std::endl; | |
8562 | tout(cct) << dirfd << std::endl; | |
8563 | ||
8564 | InodeRef dirinode; | |
8565 | std::scoped_lock locker(client_lock); | |
8566 | int r = get_fd_inode(dirfd, &dirinode); | |
8567 | if (r < 0) { | |
8568 | return r; | |
8569 | } | |
8570 | ||
8571 | if (cct->_conf->client_permissions) { | |
8572 | r = may_open(dirinode.get(), O_RDONLY, perms); | |
8573 | if (r < 0) { | |
8574 | return r; | |
8575 | } | |
8576 | } | |
8577 | r = _opendir(dirinode.get(), dirpp, perms); | |
8578 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
8579 | if (r != -CEPHFS_ENOTDIR) { | |
8580 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
8581 | } | |
8582 | return r; | |
8583 | } | |
8584 | ||
7c673cae FG |
8585 | int Client::_opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms) |
8586 | { | |
8587 | if (!in->is_dir()) | |
f67539c2 | 8588 | return -CEPHFS_ENOTDIR; |
7c673cae FG |
8589 | *dirpp = new dir_result_t(in, perms); |
8590 | opened_dirs.insert(*dirpp); | |
11fdf7f2 | 8591 | ldout(cct, 8) << __func__ << "(" << in->ino << ") = " << 0 << " (" << *dirpp << ")" << dendl; |
7c673cae FG |
8592 | return 0; |
8593 | } | |
8594 | ||
8595 | ||
8596 | int Client::closedir(dir_result_t *dir) | |
8597 | { | |
11fdf7f2 | 8598 | tout(cct) << __func__ << std::endl; |
f67539c2 | 8599 | tout(cct) << (uintptr_t)dir << std::endl; |
7c673cae | 8600 | |
11fdf7f2 | 8601 | ldout(cct, 3) << __func__ << "(" << dir << ") = 0" << dendl; |
f67539c2 | 8602 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8603 | _closedir(dir); |
8604 | return 0; | |
8605 | } | |
8606 | ||
8607 | void Client::_closedir(dir_result_t *dirp) | |
8608 | { | |
11fdf7f2 | 8609 | ldout(cct, 10) << __func__ << "(" << dirp << ")" << dendl; |
f67539c2 | 8610 | |
7c673cae | 8611 | if (dirp->inode) { |
11fdf7f2 | 8612 | ldout(cct, 10) << __func__ << " detaching inode " << dirp->inode << dendl; |
7c673cae FG |
8613 | dirp->inode.reset(); |
8614 | } | |
8615 | _readdir_drop_dirp_buffer(dirp); | |
8616 | opened_dirs.erase(dirp); | |
8617 | delete dirp; | |
8618 | } | |
8619 | ||
8620 | void Client::rewinddir(dir_result_t *dirp) | |
8621 | { | |
11fdf7f2 | 8622 | ldout(cct, 3) << __func__ << "(" << dirp << ")" << dendl; |
181888fb | 8623 | |
f67539c2 TL |
8624 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8625 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8626 | return; |
8627 | ||
f67539c2 | 8628 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8629 | dir_result_t *d = static_cast<dir_result_t*>(dirp); |
8630 | _readdir_drop_dirp_buffer(d); | |
8631 | d->reset(); | |
8632 | } | |
8633 | ||
8634 | loff_t Client::telldir(dir_result_t *dirp) | |
8635 | { | |
8636 | dir_result_t *d = static_cast<dir_result_t*>(dirp); | |
11fdf7f2 | 8637 | ldout(cct, 3) << __func__ << "(" << dirp << ") = " << d->offset << dendl; |
7c673cae FG |
8638 | return d->offset; |
8639 | } | |
8640 | ||
8641 | void Client::seekdir(dir_result_t *dirp, loff_t offset) | |
8642 | { | |
11fdf7f2 | 8643 | ldout(cct, 3) << __func__ << "(" << dirp << ", " << offset << ")" << dendl; |
7c673cae | 8644 | |
f67539c2 TL |
8645 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8646 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8647 | return; |
8648 | ||
f67539c2 TL |
8649 | std::scoped_lock lock(client_lock); |
8650 | ||
7c673cae FG |
8651 | if (offset == dirp->offset) |
8652 | return; | |
8653 | ||
8654 | if (offset > dirp->offset) | |
8655 | dirp->release_count = 0; // bump if we do a forward seek | |
8656 | else | |
8657 | dirp->ordered_count = 0; // disable filling readdir cache | |
8658 | ||
8659 | if (dirp->hash_order()) { | |
8660 | if (dirp->offset > offset) { | |
8661 | _readdir_drop_dirp_buffer(dirp); | |
8662 | dirp->reset(); | |
8663 | } | |
8664 | } else { | |
8665 | if (offset == 0 || | |
8666 | dirp->buffer_frag != frag_t(dir_result_t::fpos_high(offset)) || | |
8667 | dirp->offset_low() > dir_result_t::fpos_low(offset)) { | |
8668 | _readdir_drop_dirp_buffer(dirp); | |
8669 | dirp->reset(); | |
8670 | } | |
8671 | } | |
8672 | ||
8673 | dirp->offset = offset; | |
8674 | } | |
8675 | ||
8676 | ||
8677 | //struct dirent { | |
8678 | // ino_t d_ino; /* inode number */ | |
8679 | // off_t d_off; /* offset to the next dirent */ | |
8680 | // unsigned short d_reclen; /* length of this record */ | |
8681 | // unsigned char d_type; /* type of file */ | |
8682 | // char d_name[256]; /* filename */ | |
8683 | //}; | |
8684 | void Client::fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off) | |
8685 | { | |
8686 | strncpy(de->d_name, name, 255); | |
8687 | de->d_name[255] = '\0'; | |
f67539c2 | 8688 | #if !defined(__CYGWIN__) && !(defined(_WIN32)) |
7c673cae | 8689 | de->d_ino = ino; |
11fdf7f2 | 8690 | #if !defined(__APPLE__) && !defined(__FreeBSD__) |
7c673cae FG |
8691 | de->d_off = next_off; |
8692 | #endif | |
8693 | de->d_reclen = 1; | |
8694 | de->d_type = IFTODT(type); | |
11fdf7f2 | 8695 | ldout(cct, 10) << __func__ << " '" << de->d_name << "' -> " << inodeno_t(de->d_ino) |
7c673cae FG |
8696 | << " type " << (int)de->d_type << " w/ next_off " << hex << next_off << dec << dendl; |
8697 | #endif | |
8698 | } | |
8699 | ||
8700 | void Client::_readdir_next_frag(dir_result_t *dirp) | |
8701 | { | |
8702 | frag_t fg = dirp->buffer_frag; | |
8703 | ||
8704 | if (fg.is_rightmost()) { | |
11fdf7f2 | 8705 | ldout(cct, 10) << __func__ << " advance from " << fg << " to END" << dendl; |
7c673cae FG |
8706 | dirp->set_end(); |
8707 | return; | |
8708 | } | |
8709 | ||
8710 | // advance | |
8711 | fg = fg.next(); | |
11fdf7f2 | 8712 | ldout(cct, 10) << __func__ << " advance from " << dirp->buffer_frag << " to " << fg << dendl; |
7c673cae FG |
8713 | |
8714 | if (dirp->hash_order()) { | |
8715 | // keep last_name | |
8716 | int64_t new_offset = dir_result_t::make_fpos(fg.value(), 2, true); | |
8717 | if (dirp->offset < new_offset) // don't decrease offset | |
8718 | dirp->offset = new_offset; | |
8719 | } else { | |
8720 | dirp->last_name.clear(); | |
8721 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); | |
8722 | _readdir_rechoose_frag(dirp); | |
8723 | } | |
8724 | } | |
8725 | ||
8726 | void Client::_readdir_rechoose_frag(dir_result_t *dirp) | |
8727 | { | |
11fdf7f2 | 8728 | ceph_assert(dirp->inode); |
7c673cae FG |
8729 | |
8730 | if (dirp->hash_order()) | |
8731 | return; | |
8732 | ||
8733 | frag_t cur = frag_t(dirp->offset_high()); | |
8734 | frag_t fg = dirp->inode->dirfragtree[cur.value()]; | |
8735 | if (fg != cur) { | |
11fdf7f2 | 8736 | ldout(cct, 10) << __func__ << " frag " << cur << " maps to " << fg << dendl; |
7c673cae FG |
8737 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); |
8738 | dirp->last_name.clear(); | |
8739 | dirp->next_offset = 2; | |
8740 | } | |
8741 | } | |
8742 | ||
8743 | void Client::_readdir_drop_dirp_buffer(dir_result_t *dirp) | |
8744 | { | |
11fdf7f2 | 8745 | ldout(cct, 10) << __func__ << " " << dirp << dendl; |
7c673cae FG |
8746 | dirp->buffer.clear(); |
8747 | } | |
8748 | ||
8749 | int Client::_readdir_get_frag(dir_result_t *dirp) | |
8750 | { | |
11fdf7f2 TL |
8751 | ceph_assert(dirp); |
8752 | ceph_assert(dirp->inode); | |
7c673cae FG |
8753 | |
8754 | // get the current frag. | |
8755 | frag_t fg; | |
8756 | if (dirp->hash_order()) | |
8757 | fg = dirp->inode->dirfragtree[dirp->offset_high()]; | |
8758 | else | |
8759 | fg = frag_t(dirp->offset_high()); | |
8760 | ||
11fdf7f2 | 8761 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " fg " << fg |
7c673cae FG |
8762 | << " offset " << hex << dirp->offset << dec << dendl; |
8763 | ||
8764 | int op = CEPH_MDS_OP_READDIR; | |
8765 | if (dirp->inode && dirp->inode->snapid == CEPH_SNAPDIR) | |
8766 | op = CEPH_MDS_OP_LSSNAP; | |
8767 | ||
8768 | InodeRef& diri = dirp->inode; | |
8769 | ||
8770 | MetaRequest *req = new MetaRequest(op); | |
8771 | filepath path; | |
8772 | diri->make_nosnap_relative_path(path); | |
8773 | req->set_filepath(path); | |
8774 | req->set_inode(diri.get()); | |
8775 | req->head.args.readdir.frag = fg; | |
8776 | req->head.args.readdir.flags = CEPH_READDIR_REPLY_BITFLAGS; | |
8777 | if (dirp->last_name.length()) { | |
94b18763 | 8778 | req->path2.set_path(dirp->last_name); |
7c673cae FG |
8779 | } else if (dirp->hash_order()) { |
8780 | req->head.args.readdir.offset_hash = dirp->offset_high(); | |
8781 | } | |
8782 | req->dirp = dirp; | |
8783 | ||
8784 | bufferlist dirbl; | |
8785 | int res = make_request(req, dirp->perms, NULL, NULL, -1, &dirbl); | |
8786 | ||
f67539c2 | 8787 | if (res == -CEPHFS_EAGAIN) { |
11fdf7f2 | 8788 | ldout(cct, 10) << __func__ << " got EAGAIN, retrying" << dendl; |
7c673cae FG |
8789 | _readdir_rechoose_frag(dirp); |
8790 | return _readdir_get_frag(dirp); | |
8791 | } | |
8792 | ||
8793 | if (res == 0) { | |
11fdf7f2 | 8794 | ldout(cct, 10) << __func__ << " " << dirp << " got frag " << dirp->buffer_frag |
7c673cae FG |
8795 | << " size " << dirp->buffer.size() << dendl; |
8796 | } else { | |
11fdf7f2 | 8797 | ldout(cct, 10) << __func__ << " got error " << res << ", setting end flag" << dendl; |
7c673cae FG |
8798 | dirp->set_end(); |
8799 | } | |
8800 | ||
8801 | return res; | |
8802 | } | |
8803 | ||
8804 | struct dentry_off_lt { | |
8805 | bool operator()(const Dentry* dn, int64_t off) const { | |
8806 | return dir_result_t::fpos_cmp(dn->offset, off) < 0; | |
8807 | } | |
8808 | }; | |
8809 | ||
8810 | int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, | |
8811 | int caps, bool getref) | |
8812 | { | |
f67539c2 | 8813 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11fdf7f2 | 8814 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino |
7c673cae FG |
8815 | << " last_name " << dirp->last_name << " offset " << hex << dirp->offset << dec |
8816 | << dendl; | |
8817 | Dir *dir = dirp->inode->dir; | |
8818 | ||
8819 | if (!dir) { | |
8820 | ldout(cct, 10) << " dir is empty" << dendl; | |
8821 | dirp->set_end(); | |
8822 | return 0; | |
8823 | } | |
8824 | ||
8825 | vector<Dentry*>::iterator pd = std::lower_bound(dir->readdir_cache.begin(), | |
8826 | dir->readdir_cache.end(), | |
8827 | dirp->offset, dentry_off_lt()); | |
8828 | ||
8829 | string dn_name; | |
8830 | while (true) { | |
adb31ebb | 8831 | int mask = caps; |
7c673cae | 8832 | if (!dirp->inode->is_complete_and_ordered()) |
f67539c2 | 8833 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8834 | if (pd == dir->readdir_cache.end()) |
8835 | break; | |
8836 | Dentry *dn = *pd; | |
8837 | if (dn->inode == NULL) { | |
8838 | ldout(cct, 15) << " skipping null '" << dn->name << "'" << dendl; | |
8839 | ++pd; | |
8840 | continue; | |
8841 | } | |
8842 | if (dn->cap_shared_gen != dir->parent_inode->shared_gen) { | |
8843 | ldout(cct, 15) << " skipping mismatch shared gen '" << dn->name << "'" << dendl; | |
8844 | ++pd; | |
8845 | continue; | |
8846 | } | |
8847 | ||
92f5a8d4 | 8848 | int idx = pd - dir->readdir_cache.begin(); |
adb31ebb TL |
8849 | if (dn->inode->is_dir()) { |
8850 | mask |= CEPH_STAT_RSTAT; | |
8851 | } | |
8852 | int r = _getattr(dn->inode, mask, dirp->perms); | |
7c673cae FG |
8853 | if (r < 0) |
8854 | return r; | |
92f5a8d4 TL |
8855 | |
8856 | // the content of readdir_cache may change after _getattr(), so pd may be invalid iterator | |
8857 | pd = dir->readdir_cache.begin() + idx; | |
8858 | if (pd >= dir->readdir_cache.end() || *pd != dn) | |
f67539c2 | 8859 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8860 | |
8861 | struct ceph_statx stx; | |
8862 | struct dirent de; | |
8863 | fill_statx(dn->inode, caps, &stx); | |
8864 | ||
8865 | uint64_t next_off = dn->offset + 1; | |
eafe8130 | 8866 | fill_dirent(&de, dn->name.c_str(), stx.stx_mode, stx.stx_ino, next_off); |
7c673cae FG |
8867 | ++pd; |
8868 | if (pd == dir->readdir_cache.end()) | |
8869 | next_off = dir_result_t::END; | |
8870 | ||
8871 | Inode *in = NULL; | |
7c673cae FG |
8872 | if (getref) { |
8873 | in = dn->inode.get(); | |
8874 | _ll_get(in); | |
8875 | } | |
8876 | ||
8877 | dn_name = dn->name; // fill in name while we have lock | |
8878 | ||
9f95a23c | 8879 | client_lock.unlock(); |
7c673cae | 8880 | r = cb(p, &de, &stx, next_off, in); // _next_ offset |
9f95a23c | 8881 | client_lock.lock(); |
7c673cae FG |
8882 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << dn->offset << dec |
8883 | << " = " << r << dendl; | |
8884 | if (r < 0) { | |
8885 | return r; | |
8886 | } | |
8887 | ||
8888 | dirp->offset = next_off; | |
8889 | if (dirp->at_end()) | |
8890 | dirp->next_offset = 2; | |
8891 | else | |
8892 | dirp->next_offset = dirp->offset_low(); | |
8893 | dirp->last_name = dn_name; // we successfully returned this one; update! | |
28e407b8 | 8894 | dirp->release_count = 0; // last_name no longer match cache index |
7c673cae FG |
8895 | if (r > 0) |
8896 | return r; | |
8897 | } | |
8898 | ||
11fdf7f2 | 8899 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " at end" << dendl; |
7c673cae FG |
8900 | dirp->set_end(); |
8901 | return 0; | |
8902 | } | |
8903 | ||
8904 | int Client::readdir_r_cb(dir_result_t *d, add_dirent_cb_t cb, void *p, | |
8905 | unsigned want, unsigned flags, bool getref) | |
8906 | { | |
8907 | int caps = statx_to_mask(flags, want); | |
8908 | ||
f67539c2 TL |
8909 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8910 | if (!mref_reader.is_state_satisfied()) | |
8911 | return -CEPHFS_ENOTCONN; | |
7c673cae | 8912 | |
f67539c2 | 8913 | std::unique_lock cl(client_lock); |
181888fb | 8914 | |
7c673cae FG |
8915 | dir_result_t *dirp = static_cast<dir_result_t*>(d); |
8916 | ||
11fdf7f2 | 8917 | ldout(cct, 10) << __func__ << " " << *dirp->inode << " offset " << hex << dirp->offset |
7c673cae FG |
8918 | << dec << " at_end=" << dirp->at_end() |
8919 | << " hash_order=" << dirp->hash_order() << dendl; | |
8920 | ||
8921 | struct dirent de; | |
8922 | struct ceph_statx stx; | |
8923 | memset(&de, 0, sizeof(de)); | |
8924 | memset(&stx, 0, sizeof(stx)); | |
8925 | ||
8926 | InodeRef& diri = dirp->inode; | |
8927 | ||
8928 | if (dirp->at_end()) | |
8929 | return 0; | |
8930 | ||
8931 | if (dirp->offset == 0) { | |
8932 | ldout(cct, 15) << " including ." << dendl; | |
11fdf7f2 | 8933 | ceph_assert(diri->dentries.size() < 2); // can't have multiple hard-links to a dir |
7c673cae FG |
8934 | uint64_t next_off = 1; |
8935 | ||
8936 | int r; | |
adb31ebb | 8937 | r = _getattr(diri, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
8938 | if (r < 0) |
8939 | return r; | |
8940 | ||
8941 | fill_statx(diri, caps, &stx); | |
8942 | fill_dirent(&de, ".", S_IFDIR, stx.stx_ino, next_off); | |
8943 | ||
8944 | Inode *inode = NULL; | |
8945 | if (getref) { | |
8946 | inode = diri.get(); | |
8947 | _ll_get(inode); | |
8948 | } | |
8949 | ||
f67539c2 | 8950 | cl.unlock(); |
7c673cae | 8951 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 8952 | cl.lock(); |
7c673cae FG |
8953 | if (r < 0) |
8954 | return r; | |
8955 | ||
8956 | dirp->offset = next_off; | |
8957 | if (r > 0) | |
8958 | return r; | |
8959 | } | |
8960 | if (dirp->offset == 1) { | |
8961 | ldout(cct, 15) << " including .." << dendl; | |
8962 | uint64_t next_off = 2; | |
8963 | InodeRef in; | |
11fdf7f2 | 8964 | if (diri->dentries.empty()) |
7c673cae FG |
8965 | in = diri; |
8966 | else | |
94b18763 | 8967 | in = diri->get_first_parent()->dir->parent_inode; |
7c673cae FG |
8968 | |
8969 | int r; | |
adb31ebb | 8970 | r = _getattr(in, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
8971 | if (r < 0) |
8972 | return r; | |
8973 | ||
8974 | fill_statx(in, caps, &stx); | |
8975 | fill_dirent(&de, "..", S_IFDIR, stx.stx_ino, next_off); | |
8976 | ||
8977 | Inode *inode = NULL; | |
8978 | if (getref) { | |
8979 | inode = in.get(); | |
8980 | _ll_get(inode); | |
8981 | } | |
8982 | ||
f67539c2 | 8983 | cl.unlock(); |
7c673cae | 8984 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 8985 | cl.lock(); |
7c673cae FG |
8986 | if (r < 0) |
8987 | return r; | |
8988 | ||
8989 | dirp->offset = next_off; | |
8990 | if (r > 0) | |
8991 | return r; | |
8992 | } | |
8993 | ||
8994 | // can we read from our cache? | |
8995 | ldout(cct, 10) << "offset " << hex << dirp->offset << dec | |
8996 | << " snapid " << dirp->inode->snapid << " (complete && ordered) " | |
8997 | << dirp->inode->is_complete_and_ordered() | |
8998 | << " issued " << ccap_string(dirp->inode->caps_issued()) | |
8999 | << dendl; | |
9000 | if (dirp->inode->snapid != CEPH_SNAPDIR && | |
9001 | dirp->inode->is_complete_and_ordered() && | |
94b18763 | 9002 | dirp->inode->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 9003 | int err = _readdir_cache_cb(dirp, cb, p, caps, getref); |
f67539c2 | 9004 | if (err != -CEPHFS_EAGAIN) |
7c673cae FG |
9005 | return err; |
9006 | } | |
9007 | ||
9008 | while (1) { | |
9009 | if (dirp->at_end()) | |
9010 | return 0; | |
9011 | ||
9012 | bool check_caps = true; | |
9013 | if (!dirp->is_cached()) { | |
9014 | int r = _readdir_get_frag(dirp); | |
9015 | if (r) | |
9016 | return r; | |
9017 | // _readdir_get_frag () may updates dirp->offset if the replied dirfrag is | |
9018 | // different than the requested one. (our dirfragtree was outdated) | |
9019 | check_caps = false; | |
9020 | } | |
9021 | frag_t fg = dirp->buffer_frag; | |
9022 | ||
9023 | ldout(cct, 10) << "frag " << fg << " buffer size " << dirp->buffer.size() | |
9024 | << " offset " << hex << dirp->offset << dendl; | |
9025 | ||
9026 | for (auto it = std::lower_bound(dirp->buffer.begin(), dirp->buffer.end(), | |
9027 | dirp->offset, dir_result_t::dentry_off_lt()); | |
9028 | it != dirp->buffer.end(); | |
9029 | ++it) { | |
9030 | dir_result_t::dentry &entry = *it; | |
9031 | ||
9032 | uint64_t next_off = entry.offset + 1; | |
9033 | ||
9034 | int r; | |
9035 | if (check_caps) { | |
adb31ebb TL |
9036 | int mask = caps; |
9037 | if(entry.inode->is_dir()){ | |
9038 | mask |= CEPH_STAT_RSTAT; | |
9039 | } | |
9040 | r = _getattr(entry.inode, mask, dirp->perms); | |
7c673cae FG |
9041 | if (r < 0) |
9042 | return r; | |
9043 | } | |
9044 | ||
9045 | fill_statx(entry.inode, caps, &stx); | |
9046 | fill_dirent(&de, entry.name.c_str(), stx.stx_mode, stx.stx_ino, next_off); | |
9047 | ||
9048 | Inode *inode = NULL; | |
9049 | if (getref) { | |
9050 | inode = entry.inode.get(); | |
9051 | _ll_get(inode); | |
9052 | } | |
9053 | ||
f67539c2 | 9054 | cl.unlock(); |
7c673cae | 9055 | r = cb(p, &de, &stx, next_off, inode); // _next_ offset |
f67539c2 | 9056 | cl.lock(); |
7c673cae FG |
9057 | |
9058 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << next_off - 1 << dec | |
9059 | << " = " << r << dendl; | |
9060 | if (r < 0) | |
9061 | return r; | |
9062 | ||
9063 | dirp->offset = next_off; | |
9064 | if (r > 0) | |
9065 | return r; | |
9066 | } | |
9067 | ||
9068 | if (dirp->next_offset > 2) { | |
9069 | ldout(cct, 10) << " fetching next chunk of this frag" << dendl; | |
9070 | _readdir_drop_dirp_buffer(dirp); | |
9071 | continue; // more! | |
9072 | } | |
9073 | ||
9074 | if (!fg.is_rightmost()) { | |
9075 | // next frag! | |
9076 | _readdir_next_frag(dirp); | |
9077 | continue; | |
9078 | } | |
9079 | ||
9080 | if (diri->shared_gen == dirp->start_shared_gen && | |
9081 | diri->dir_release_count == dirp->release_count) { | |
9082 | if (diri->dir_ordered_count == dirp->ordered_count) { | |
9083 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
9084 | if (diri->dir) { | |
11fdf7f2 | 9085 | ceph_assert(diri->dir->readdir_cache.size() >= dirp->cache_index); |
7c673cae FG |
9086 | diri->dir->readdir_cache.resize(dirp->cache_index); |
9087 | } | |
9088 | diri->flags |= I_COMPLETE | I_DIR_ORDERED; | |
9089 | } else { | |
9090 | ldout(cct, 10) << " marking I_COMPLETE on " << *diri << dendl; | |
9091 | diri->flags |= I_COMPLETE; | |
9092 | } | |
9093 | } | |
9094 | ||
9095 | dirp->set_end(); | |
9096 | return 0; | |
9097 | } | |
9098 | ceph_abort(); | |
9099 | return 0; | |
9100 | } | |
9101 | ||
9102 | ||
9103 | int Client::readdir_r(dir_result_t *d, struct dirent *de) | |
9104 | { | |
9105 | return readdirplus_r(d, de, 0, 0, 0, NULL); | |
9106 | } | |
9107 | ||
9108 | /* | |
9109 | * readdirplus_r | |
9110 | * | |
9111 | * returns | |
9112 | * 1 if we got a dirent | |
9113 | * 0 for end of directory | |
9114 | * <0 on error | |
9115 | */ | |
9116 | ||
9117 | struct single_readdir { | |
9118 | struct dirent *de; | |
9119 | struct ceph_statx *stx; | |
9120 | Inode *inode; | |
9121 | bool full; | |
9122 | }; | |
9123 | ||
9124 | static int _readdir_single_dirent_cb(void *p, struct dirent *de, | |
9125 | struct ceph_statx *stx, off_t off, | |
9126 | Inode *in) | |
9127 | { | |
9128 | single_readdir *c = static_cast<single_readdir *>(p); | |
9129 | ||
9130 | if (c->full) | |
9131 | return -1; // already filled this dirent | |
9132 | ||
9133 | *c->de = *de; | |
9134 | if (c->stx) | |
9135 | *c->stx = *stx; | |
9136 | c->inode = in; | |
9137 | c->full = true; | |
9138 | return 1; | |
9139 | } | |
9140 | ||
9141 | struct dirent *Client::readdir(dir_result_t *d) | |
9142 | { | |
9143 | int ret; | |
f91f0fd5 | 9144 | auto& de = d->de; |
7c673cae FG |
9145 | single_readdir sr; |
9146 | sr.de = &de; | |
9147 | sr.stx = NULL; | |
9148 | sr.inode = NULL; | |
9149 | sr.full = false; | |
9150 | ||
9151 | // our callback fills the dirent and sets sr.full=true on first | |
9152 | // call, and returns -1 the second time around. | |
9153 | ret = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr); | |
9154 | if (ret < -1) { | |
9155 | errno = -ret; // this sucks. | |
9156 | return (dirent *) NULL; | |
9157 | } | |
9158 | if (sr.full) { | |
9159 | return &de; | |
9160 | } | |
9161 | return (dirent *) NULL; | |
9162 | } | |
9163 | ||
9164 | int Client::readdirplus_r(dir_result_t *d, struct dirent *de, | |
9165 | struct ceph_statx *stx, unsigned want, | |
9166 | unsigned flags, Inode **out) | |
9167 | { | |
9168 | single_readdir sr; | |
9169 | sr.de = de; | |
9170 | sr.stx = stx; | |
9171 | sr.inode = NULL; | |
9172 | sr.full = false; | |
9173 | ||
9174 | // our callback fills the dirent and sets sr.full=true on first | |
9175 | // call, and returns -1 the second time around. | |
9176 | int r = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr, want, flags, out); | |
9177 | if (r < -1) | |
9178 | return r; | |
9179 | if (out) | |
9180 | *out = sr.inode; | |
9181 | if (sr.full) | |
9182 | return 1; | |
9183 | return 0; | |
9184 | } | |
9185 | ||
9186 | ||
9187 | /* getdents */ | |
9188 | struct getdents_result { | |
9189 | char *buf; | |
9190 | int buflen; | |
9191 | int pos; | |
9192 | bool fullent; | |
9193 | }; | |
9194 | ||
9195 | static int _readdir_getdent_cb(void *p, struct dirent *de, | |
9196 | struct ceph_statx *stx, off_t off, Inode *in) | |
9197 | { | |
9198 | struct getdents_result *c = static_cast<getdents_result *>(p); | |
9199 | ||
9200 | int dlen; | |
9201 | if (c->fullent) | |
9202 | dlen = sizeof(*de); | |
9203 | else | |
9204 | dlen = strlen(de->d_name) + 1; | |
9205 | ||
9206 | if (c->pos + dlen > c->buflen) | |
9207 | return -1; // doesn't fit | |
9208 | ||
9209 | if (c->fullent) { | |
9210 | memcpy(c->buf + c->pos, de, sizeof(*de)); | |
9211 | } else { | |
9212 | memcpy(c->buf + c->pos, de->d_name, dlen); | |
9213 | } | |
9214 | c->pos += dlen; | |
9215 | return 0; | |
9216 | } | |
9217 | ||
9218 | int Client::_getdents(dir_result_t *dir, char *buf, int buflen, bool fullent) | |
9219 | { | |
9220 | getdents_result gr; | |
9221 | gr.buf = buf; | |
9222 | gr.buflen = buflen; | |
9223 | gr.fullent = fullent; | |
9224 | gr.pos = 0; | |
9225 | ||
9226 | int r = readdir_r_cb(dir, _readdir_getdent_cb, (void *)&gr); | |
9227 | ||
9228 | if (r < 0) { // some error | |
9229 | if (r == -1) { // buffer ran out of space | |
9230 | if (gr.pos) { // but we got some entries already! | |
9231 | return gr.pos; | |
9232 | } // or we need a larger buffer | |
f67539c2 | 9233 | return -CEPHFS_ERANGE; |
7c673cae FG |
9234 | } else { // actual error, return it |
9235 | return r; | |
9236 | } | |
9237 | } | |
9238 | return gr.pos; | |
9239 | } | |
9240 | ||
9241 | ||
9242 | /* getdir */ | |
9243 | struct getdir_result { | |
9244 | list<string> *contents; | |
9245 | int num; | |
9246 | }; | |
9247 | ||
9248 | static int _getdir_cb(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in) | |
9249 | { | |
9250 | getdir_result *r = static_cast<getdir_result *>(p); | |
9251 | ||
9252 | r->contents->push_back(de->d_name); | |
9253 | r->num++; | |
9254 | return 0; | |
9255 | } | |
9256 | ||
9257 | int Client::getdir(const char *relpath, list<string>& contents, | |
9258 | const UserPerm& perms) | |
9259 | { | |
9260 | ldout(cct, 3) << "getdir(" << relpath << ")" << dendl; | |
f67539c2 TL |
9261 | tout(cct) << "getdir" << std::endl; |
9262 | tout(cct) << relpath << std::endl; | |
7c673cae FG |
9263 | |
9264 | dir_result_t *d; | |
9265 | int r = opendir(relpath, &d, perms); | |
9266 | if (r < 0) | |
9267 | return r; | |
9268 | ||
9269 | getdir_result gr; | |
9270 | gr.contents = &contents; | |
9271 | gr.num = 0; | |
9272 | r = readdir_r_cb(d, _getdir_cb, (void *)&gr); | |
9273 | ||
9274 | closedir(d); | |
9275 | ||
9276 | if (r < 0) | |
9277 | return r; | |
9278 | return gr.num; | |
9279 | } | |
9280 | ||
9281 | ||
9282 | /****** file i/o **********/ | |
f67539c2 | 9283 | |
b3b6e05e | 9284 | // common parts for open and openat. call with client_lock locked. |
20effc67 | 9285 | int Client::create_and_open(int dirfd, const char *relpath, int flags, |
b3b6e05e TL |
9286 | const UserPerm& perms, mode_t mode, int stripe_unit, |
9287 | int stripe_count, int object_size, const char *data_pool, | |
9288 | std::string alternate_name) { | |
9289 | ceph_assert(ceph_mutex_is_locked(client_lock)); | |
f91f0fd5 | 9290 | int cflags = ceph_flags_sys2wire(flags); |
f91f0fd5 | 9291 | tout(cct) << cflags << std::endl; |
7c673cae FG |
9292 | |
9293 | Fh *fh = NULL; | |
9294 | ||
9295 | #if defined(__linux__) && defined(O_PATH) | |
9296 | /* When the O_PATH is being specified, others flags than O_DIRECTORY | |
9297 | * and O_NOFOLLOW are ignored. Please refer do_entry_open() function | |
9298 | * in kernel (fs/open.c). */ | |
9299 | if (flags & O_PATH) | |
9300 | flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; | |
9301 | #endif | |
9302 | ||
9303 | filepath path(relpath); | |
9304 | InodeRef in; | |
9305 | bool created = false; | |
9306 | /* O_CREATE with O_EXCL enforces O_NOFOLLOW. */ | |
9307 | bool followsym = !((flags & O_NOFOLLOW) || ((flags & O_CREAT) && (flags & O_EXCL))); | |
f91f0fd5 TL |
9308 | int mask = ceph_caps_for_mode(ceph_flags_to_mode(cflags)); |
9309 | ||
b3b6e05e | 9310 | InodeRef dirinode = nullptr; |
20effc67 TL |
9311 | int r = get_fd_inode(dirfd, &dirinode); |
9312 | if (r < 0) { | |
9313 | return r; | |
b3b6e05e | 9314 | } |
7c673cae | 9315 | |
20effc67 | 9316 | r = path_walk(path, &in, perms, followsym, mask, dirinode); |
7c673cae | 9317 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) |
f67539c2 | 9318 | return -CEPHFS_EEXIST; |
7c673cae FG |
9319 | |
9320 | #if defined(__linux__) && defined(O_PATH) | |
9321 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW) && !(flags & O_PATH)) | |
9322 | #else | |
b3b6e05e | 9323 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW)) |
7c673cae | 9324 | #endif |
f67539c2 | 9325 | return -CEPHFS_ELOOP; |
7c673cae | 9326 | |
f67539c2 | 9327 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
7c673cae FG |
9328 | filepath dirpath = path; |
9329 | string dname = dirpath.last_dentry(); | |
9330 | dirpath.pop_dentry(); | |
9331 | InodeRef dir; | |
9332 | r = path_walk(dirpath, &dir, perms, true, | |
b3b6e05e TL |
9333 | cct->_conf->client_permissions ? CEPH_CAP_AUTH_SHARED : 0, dirinode); |
9334 | if (r < 0) { | |
7c673cae | 9335 | goto out; |
b3b6e05e | 9336 | } |
7c673cae FG |
9337 | if (cct->_conf->client_permissions) { |
9338 | r = may_create(dir.get(), perms); | |
9339 | if (r < 0) | |
b3b6e05e | 9340 | goto out; |
7c673cae FG |
9341 | } |
9342 | r = _create(dir.get(), dname.c_str(), flags, mode, &in, &fh, stripe_unit, | |
f67539c2 TL |
9343 | stripe_count, object_size, data_pool, &created, perms, |
9344 | std::move(alternate_name)); | |
7c673cae FG |
9345 | } |
9346 | if (r < 0) | |
9347 | goto out; | |
9348 | ||
9349 | if (!created) { | |
9350 | // posix says we can only check permissions of existing files | |
9351 | if (cct->_conf->client_permissions) { | |
9352 | r = may_open(in.get(), flags, perms); | |
9353 | if (r < 0) | |
b3b6e05e | 9354 | goto out; |
7c673cae FG |
9355 | } |
9356 | } | |
9357 | ||
9358 | if (!fh) | |
9359 | r = _open(in.get(), flags, mode, &fh, perms); | |
9360 | if (r >= 0) { | |
9361 | // allocate a integer file descriptor | |
11fdf7f2 | 9362 | ceph_assert(fh); |
7c673cae | 9363 | r = get_fd(); |
11fdf7f2 | 9364 | ceph_assert(fd_map.count(r) == 0); |
7c673cae FG |
9365 | fd_map[r] = fh; |
9366 | } | |
9367 | ||
9368 | out: | |
b3b6e05e TL |
9369 | return r; |
9370 | } | |
9371 | ||
9372 | int Client::open(const char *relpath, int flags, const UserPerm& perms, | |
9373 | mode_t mode, int stripe_unit, int stripe_count, | |
9374 | int object_size, const char *data_pool, std::string alternate_name) | |
9375 | { | |
9376 | return openat(CEPHFS_AT_FDCWD, relpath, flags, perms, mode, stripe_unit, | |
9377 | stripe_count, object_size, data_pool, alternate_name); | |
9378 | } | |
9379 | ||
b3b6e05e TL |
9380 | int Client::openat(int dirfd, const char *relpath, int flags, const UserPerm& perms, |
9381 | mode_t mode, int stripe_unit, int stripe_count, int object_size, | |
9382 | const char *data_pool, std::string alternate_name) { | |
9383 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9384 | if (!mref_reader.is_state_satisfied()) { | |
9385 | return -CEPHFS_ENOTCONN; | |
9386 | } | |
9387 | ||
9388 | ldout(cct, 3) << "openat enter(" << relpath << ")" << dendl; | |
9389 | tout(cct) << dirfd << std::endl; | |
9390 | tout(cct) << relpath << std::endl; | |
9391 | tout(cct) << flags << std::endl; | |
9392 | tout(cct) << mode << std::endl; | |
9393 | ||
9394 | std::scoped_lock locker(client_lock); | |
9395 | int r = create_and_open(dirfd, relpath, flags, perms, mode, stripe_unit, stripe_count, | |
9396 | object_size, data_pool, alternate_name); | |
9397 | ||
7c673cae | 9398 | tout(cct) << r << std::endl; |
b3b6e05e | 9399 | ldout(cct, 3) << "openat exit(" << relpath << ")" << dendl; |
7c673cae FG |
9400 | return r; |
9401 | } | |
9402 | ||
7c673cae FG |
9403 | int Client::lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name, |
9404 | const UserPerm& perms) | |
9405 | { | |
11fdf7f2 | 9406 | ldout(cct, 3) << __func__ << " enter(" << ino << ", #" << dirino << "/" << name << ")" << dendl; |
7c673cae | 9407 | |
f67539c2 TL |
9408 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9409 | if (!mref_reader.is_state_satisfied()) | |
9410 | return -CEPHFS_ENOTCONN; | |
181888fb | 9411 | |
f67539c2 | 9412 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9413 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPHASH); |
9414 | filepath path(ino); | |
9415 | req->set_filepath(path); | |
9416 | ||
9417 | uint32_t h = ceph_str_hash(CEPH_STR_HASH_RJENKINS, name, strlen(name)); | |
9418 | char f[30]; | |
9419 | sprintf(f, "%u", h); | |
9420 | filepath path2(dirino); | |
9421 | path2.push_dentry(string(f)); | |
9422 | req->set_filepath2(path2); | |
9423 | ||
9424 | int r = make_request(req, perms, NULL, NULL, | |
9425 | rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9426 | ldout(cct, 3) << __func__ << " exit(" << ino << ", #" << dirino << "/" << name << ") = " << r << dendl; |
7c673cae FG |
9427 | return r; |
9428 | } | |
9429 | ||
9430 | ||
9431 | /** | |
9432 | * Load inode into local cache. | |
9433 | * | |
9434 | * If inode pointer is non-NULL, and take a reference on | |
9435 | * the resulting Inode object in one operation, so that caller | |
9436 | * can safely assume inode will still be there after return. | |
9437 | */ | |
f67539c2 | 9438 | int Client::_lookup_vino(vinodeno_t vino, const UserPerm& perms, Inode **inode) |
7c673cae | 9439 | { |
f67539c2 | 9440 | ldout(cct, 8) << __func__ << " enter(" << vino << ")" << dendl; |
7c673cae | 9441 | |
f67539c2 TL |
9442 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9443 | if (!mref_reader.is_state_satisfied()) | |
9444 | return -CEPHFS_ENOTCONN; | |
181888fb | 9445 | |
b3b6e05e TL |
9446 | if (is_reserved_vino(vino)) |
9447 | return -CEPHFS_ESTALE; | |
9448 | ||
7c673cae | 9449 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPINO); |
f67539c2 | 9450 | filepath path(vino.ino); |
7c673cae FG |
9451 | req->set_filepath(path); |
9452 | ||
f67539c2 TL |
9453 | /* |
9454 | * The MDS expects either a "real" snapid here or 0. The special value | |
9455 | * carveouts for the snapid are all at the end of the range so we can | |
9456 | * just look for any snapid below this value. | |
9457 | */ | |
9458 | if (vino.snapid < CEPH_NOSNAP) | |
9459 | req->head.args.lookupino.snapid = vino.snapid; | |
9460 | ||
7c673cae FG |
9461 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); |
9462 | if (r == 0 && inode != NULL) { | |
7c673cae | 9463 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
11fdf7f2 | 9464 | ceph_assert(p != inode_map.end()); |
7c673cae FG |
9465 | *inode = p->second; |
9466 | _ll_get(*inode); | |
9467 | } | |
f67539c2 | 9468 | ldout(cct, 8) << __func__ << " exit(" << vino << ") = " << r << dendl; |
7c673cae FG |
9469 | return r; |
9470 | } | |
9471 | ||
1adf2230 AA |
9472 | int Client::lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode) |
9473 | { | |
f67539c2 TL |
9474 | vinodeno_t vino(ino, CEPH_NOSNAP); |
9475 | std::scoped_lock lock(client_lock); | |
9476 | return _lookup_vino(vino, perms, inode); | |
1adf2230 | 9477 | } |
7c673cae FG |
9478 | |
9479 | /** | |
9480 | * Find the parent inode of `ino` and insert it into | |
9481 | * our cache. Conditionally also set `parent` to a referenced | |
9482 | * Inode* if caller provides non-NULL value. | |
9483 | */ | |
1adf2230 | 9484 | int Client::_lookup_parent(Inode *ino, const UserPerm& perms, Inode **parent) |
7c673cae | 9485 | { |
11fdf7f2 | 9486 | ldout(cct, 8) << __func__ << " enter(" << ino->ino << ")" << dendl; |
7c673cae | 9487 | |
7c673cae FG |
9488 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); |
9489 | filepath path(ino->ino); | |
9490 | req->set_filepath(path); | |
9491 | ||
9492 | InodeRef target; | |
9493 | int r = make_request(req, perms, &target, NULL, rand() % mdsmap->get_num_in_mds()); | |
9494 | // Give caller a reference to the parent ino if they provided a pointer. | |
9495 | if (parent != NULL) { | |
9496 | if (r == 0) { | |
9497 | *parent = target.get(); | |
9498 | _ll_get(*parent); | |
11fdf7f2 | 9499 | ldout(cct, 8) << __func__ << " found parent " << (*parent)->ino << dendl; |
7c673cae FG |
9500 | } else { |
9501 | *parent = NULL; | |
9502 | } | |
9503 | } | |
11fdf7f2 | 9504 | ldout(cct, 8) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9505 | return r; |
9506 | } | |
9507 | ||
7c673cae FG |
9508 | /** |
9509 | * Populate the parent dentry for `ino`, provided it is | |
9510 | * a child of `parent`. | |
9511 | */ | |
1adf2230 | 9512 | int Client::_lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
7c673cae | 9513 | { |
11fdf7f2 TL |
9514 | ceph_assert(parent->is_dir()); |
9515 | ldout(cct, 3) << __func__ << " enter(" << ino->ino << ")" << dendl; | |
7c673cae | 9516 | |
f67539c2 TL |
9517 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9518 | if (!mref_reader.is_state_satisfied()) | |
9519 | return -CEPHFS_ENOTCONN; | |
181888fb | 9520 | |
7c673cae FG |
9521 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
9522 | req->set_filepath2(filepath(parent->ino)); | |
9523 | req->set_filepath(filepath(ino->ino)); | |
9524 | req->set_inode(ino); | |
9525 | ||
9526 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9527 | ldout(cct, 3) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9528 | return r; |
9529 | } | |
9530 | ||
1adf2230 AA |
9531 | int Client::lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
9532 | { | |
f67539c2 | 9533 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
9534 | return _lookup_name(ino, parent, perms); |
9535 | } | |
7c673cae | 9536 | |
11fdf7f2 | 9537 | Fh *Client::_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms) |
7c673cae | 9538 | { |
11fdf7f2 | 9539 | ceph_assert(in); |
f6b5b4d7 | 9540 | Fh *f = new Fh(in, flags, cmode, fd_gen, perms); |
7c673cae | 9541 | |
11fdf7f2 | 9542 | ldout(cct, 10) << __func__ << " " << in->ino << " mode " << cmode << dendl; |
7c673cae FG |
9543 | |
9544 | if (in->snapid != CEPH_NOSNAP) { | |
9545 | in->snap_cap_refs++; | |
9546 | ldout(cct, 5) << "open success, fh is " << f << " combined IMMUTABLE SNAP caps " | |
9547 | << ccap_string(in->caps_issued()) << dendl; | |
9548 | } | |
9549 | ||
11fdf7f2 | 9550 | const auto& conf = cct->_conf; |
7c673cae FG |
9551 | f->readahead.set_trigger_requests(1); |
9552 | f->readahead.set_min_readahead_size(conf->client_readahead_min); | |
9553 | uint64_t max_readahead = Readahead::NO_LIMIT; | |
9554 | if (conf->client_readahead_max_bytes) { | |
11fdf7f2 | 9555 | max_readahead = std::min(max_readahead, (uint64_t)conf->client_readahead_max_bytes); |
7c673cae FG |
9556 | } |
9557 | if (conf->client_readahead_max_periods) { | |
11fdf7f2 | 9558 | max_readahead = std::min(max_readahead, in->layout.get_period()*(uint64_t)conf->client_readahead_max_periods); |
7c673cae FG |
9559 | } |
9560 | f->readahead.set_max_readahead_size(max_readahead); | |
9561 | vector<uint64_t> alignments; | |
9562 | alignments.push_back(in->layout.get_period()); | |
9563 | alignments.push_back(in->layout.stripe_unit); | |
9564 | f->readahead.set_alignments(alignments); | |
9565 | ||
9566 | return f; | |
9567 | } | |
9568 | ||
9569 | int Client::_release_fh(Fh *f) | |
9570 | { | |
9571 | //ldout(cct, 3) << "op: client->close(open_files[ " << fh << " ]);" << dendl; | |
9572 | //ldout(cct, 3) << "op: open_files.erase( " << fh << " );" << dendl; | |
9573 | Inode *in = f->inode.get(); | |
11fdf7f2 | 9574 | ldout(cct, 8) << __func__ << " " << f << " mode " << f->mode << " on " << *in << dendl; |
7c673cae | 9575 | |
b32b8144 FG |
9576 | in->unset_deleg(f); |
9577 | ||
7c673cae FG |
9578 | if (in->snapid == CEPH_NOSNAP) { |
9579 | if (in->put_open_ref(f->mode)) { | |
9580 | _flush(in, new C_Client_FlushComplete(this, in)); | |
9581 | check_caps(in, 0); | |
9582 | } | |
9583 | } else { | |
11fdf7f2 | 9584 | ceph_assert(in->snap_cap_refs > 0); |
7c673cae FG |
9585 | in->snap_cap_refs--; |
9586 | } | |
9587 | ||
9588 | _release_filelocks(f); | |
9589 | ||
9590 | // Finally, read any async err (i.e. from flushes) | |
9591 | int err = f->take_async_err(); | |
9592 | if (err != 0) { | |
11fdf7f2 | 9593 | ldout(cct, 1) << __func__ << " " << f << " on inode " << *in << " caught async_err = " |
7c673cae FG |
9594 | << cpp_strerror(err) << dendl; |
9595 | } else { | |
11fdf7f2 | 9596 | ldout(cct, 10) << __func__ << " " << f << " on inode " << *in << " no async_err state" << dendl; |
7c673cae FG |
9597 | } |
9598 | ||
9599 | _put_fh(f); | |
9600 | ||
9601 | return err; | |
9602 | } | |
9603 | ||
9604 | void Client::_put_fh(Fh *f) | |
9605 | { | |
9606 | int left = f->put(); | |
9607 | if (!left) { | |
9608 | delete f; | |
9609 | } | |
9610 | } | |
9611 | ||
9612 | int Client::_open(Inode *in, int flags, mode_t mode, Fh **fhp, | |
9613 | const UserPerm& perms) | |
9614 | { | |
9615 | if (in->snapid != CEPH_NOSNAP && | |
9616 | (flags & (O_WRONLY | O_RDWR | O_CREAT | O_TRUNC | O_APPEND))) { | |
f67539c2 | 9617 | return -CEPHFS_EROFS; |
7c673cae FG |
9618 | } |
9619 | ||
9620 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
9621 | int cflags = ceph_flags_sys2wire(flags); |
9622 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
9623 | cflags |= CEPH_O_LAZY; | |
9624 | ||
9625 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
9626 | int want = ceph_caps_for_mode(cmode); |
9627 | int result = 0; | |
9628 | ||
9629 | in->get_open_ref(cmode); // make note of pending open, since it effects _wanted_ caps. | |
9630 | ||
b32b8144 | 9631 | if ((flags & O_TRUNC) == 0 && in->caps_issued_mask(want)) { |
7c673cae FG |
9632 | // update wanted? |
9633 | check_caps(in, CHECK_CAPS_NODELAY); | |
9634 | } else { | |
b32b8144 | 9635 | |
7c673cae FG |
9636 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); |
9637 | filepath path; | |
9638 | in->make_nosnap_relative_path(path); | |
9639 | req->set_filepath(path); | |
11fdf7f2 | 9640 | req->head.args.open.flags = cflags & ~CEPH_O_CREAT; |
7c673cae FG |
9641 | req->head.args.open.mode = mode; |
9642 | req->head.args.open.pool = -1; | |
9643 | if (cct->_conf->client_debug_getattr_caps) | |
9644 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9645 | else | |
9646 | req->head.args.open.mask = 0; | |
9647 | req->head.args.open.old_size = in->size; // for O_TRUNC | |
9648 | req->set_inode(in); | |
9649 | result = make_request(req, perms); | |
b32b8144 FG |
9650 | |
9651 | /* | |
9652 | * NFS expects that delegations will be broken on a conflicting open, | |
9653 | * not just when there is actual conflicting access to the file. SMB leases | |
9654 | * and oplocks also have similar semantics. | |
9655 | * | |
9656 | * Ensure that clients that have delegations enabled will wait on minimal | |
9657 | * caps during open, just to ensure that other clients holding delegations | |
9658 | * return theirs first. | |
9659 | */ | |
9660 | if (deleg_timeout && result == 0) { | |
9661 | int need = 0, have; | |
9662 | ||
9663 | if (cmode & CEPH_FILE_MODE_WR) | |
9664 | need |= CEPH_CAP_FILE_WR; | |
9665 | if (cmode & CEPH_FILE_MODE_RD) | |
9666 | need |= CEPH_CAP_FILE_RD; | |
9667 | ||
f6b5b4d7 TL |
9668 | Fh fh(in, flags, cmode, fd_gen, perms); |
9669 | result = get_caps(&fh, need, want, &have, -1); | |
b32b8144 | 9670 | if (result < 0) { |
1adf2230 | 9671 | ldout(cct, 8) << "Unable to get caps after open of inode " << *in << |
b32b8144 FG |
9672 | " . Denying open: " << |
9673 | cpp_strerror(result) << dendl; | |
b32b8144 FG |
9674 | } else { |
9675 | put_cap_ref(in, need); | |
9676 | } | |
9677 | } | |
7c673cae FG |
9678 | } |
9679 | ||
9680 | // success? | |
9681 | if (result >= 0) { | |
9682 | if (fhp) | |
9683 | *fhp = _create_fh(in, flags, cmode, perms); | |
9684 | } else { | |
9685 | in->put_open_ref(cmode); | |
9686 | } | |
9687 | ||
9688 | trim_cache(); | |
9689 | ||
9690 | return result; | |
9691 | } | |
9692 | ||
9693 | int Client::_renew_caps(Inode *in) | |
9694 | { | |
9695 | int wanted = in->caps_file_wanted(); | |
9696 | if (in->is_any_caps() && | |
9697 | ((wanted & CEPH_CAP_ANY_WR) == 0 || in->auth_cap)) { | |
9698 | check_caps(in, CHECK_CAPS_NODELAY); | |
9699 | return 0; | |
9700 | } | |
9701 | ||
9702 | int flags = 0; | |
9703 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
9704 | flags = O_RDWR; | |
9705 | else if (wanted & CEPH_CAP_FILE_RD) | |
9706 | flags = O_RDONLY; | |
9707 | else if (wanted & CEPH_CAP_FILE_WR) | |
9708 | flags = O_WRONLY; | |
9709 | ||
9710 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); | |
9711 | filepath path; | |
9712 | in->make_nosnap_relative_path(path); | |
9713 | req->set_filepath(path); | |
9714 | req->head.args.open.flags = flags; | |
9715 | req->head.args.open.pool = -1; | |
9716 | if (cct->_conf->client_debug_getattr_caps) | |
9717 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9718 | else | |
9719 | req->head.args.open.mask = 0; | |
9720 | req->set_inode(in); | |
9721 | ||
9722 | // duplicate in case Cap goes away; not sure if that race is a concern? | |
9723 | const UserPerm *pperm = in->get_best_perms(); | |
9724 | UserPerm perms; | |
9725 | if (pperm != NULL) | |
9726 | perms = *pperm; | |
9727 | int ret = make_request(req, perms); | |
9728 | return ret; | |
9729 | } | |
9730 | ||
b3b6e05e | 9731 | int Client::_close(int fd) |
7c673cae FG |
9732 | { |
9733 | ldout(cct, 3) << "close enter(" << fd << ")" << dendl; | |
7c673cae FG |
9734 | tout(cct) << "close" << std::endl; |
9735 | tout(cct) << fd << std::endl; | |
9736 | ||
9737 | Fh *fh = get_filehandle(fd); | |
9738 | if (!fh) | |
f67539c2 | 9739 | return -CEPHFS_EBADF; |
7c673cae FG |
9740 | int err = _release_fh(fh); |
9741 | fd_map.erase(fd); | |
9742 | put_fd(fd); | |
9743 | ldout(cct, 3) << "close exit(" << fd << ")" << dendl; | |
9744 | return err; | |
9745 | } | |
9746 | ||
b3b6e05e TL |
9747 | int Client::close(int fd) { |
9748 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9749 | if (!mref_reader.is_state_satisfied()) | |
9750 | return -CEPHFS_ENOTCONN; | |
9751 | ||
9752 | std::scoped_lock lock(client_lock); | |
9753 | return _close(fd); | |
9754 | } | |
7c673cae FG |
9755 | |
9756 | // ------------ | |
9757 | // read, write | |
9758 | ||
9759 | loff_t Client::lseek(int fd, loff_t offset, int whence) | |
9760 | { | |
f67539c2 TL |
9761 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9762 | if (!mref_reader.is_state_satisfied()) | |
9763 | return -CEPHFS_ENOTCONN; | |
9764 | ||
7c673cae FG |
9765 | tout(cct) << "lseek" << std::endl; |
9766 | tout(cct) << fd << std::endl; | |
9767 | tout(cct) << offset << std::endl; | |
9768 | tout(cct) << whence << std::endl; | |
9769 | ||
f67539c2 | 9770 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9771 | Fh *f = get_filehandle(fd); |
9772 | if (!f) | |
f67539c2 | 9773 | return -CEPHFS_EBADF; |
7c673cae FG |
9774 | #if defined(__linux__) && defined(O_PATH) |
9775 | if (f->flags & O_PATH) | |
f67539c2 | 9776 | return -CEPHFS_EBADF; |
7c673cae FG |
9777 | #endif |
9778 | return _lseek(f, offset, whence); | |
9779 | } | |
9780 | ||
9781 | loff_t Client::_lseek(Fh *f, loff_t offset, int whence) | |
9782 | { | |
9783 | Inode *in = f->inode.get(); | |
9f95a23c | 9784 | bool whence_check = false; |
11fdf7f2 | 9785 | loff_t pos = -1; |
7c673cae | 9786 | |
9f95a23c TL |
9787 | switch (whence) { |
9788 | case SEEK_END: | |
9789 | whence_check = true; | |
9790 | break; | |
9791 | ||
9792 | #ifdef SEEK_DATA | |
9793 | case SEEK_DATA: | |
9794 | whence_check = true; | |
9795 | break; | |
9796 | #endif | |
9797 | ||
9798 | #ifdef SEEK_HOLE | |
9799 | case SEEK_HOLE: | |
9800 | whence_check = true; | |
9801 | break; | |
9802 | #endif | |
9803 | } | |
9804 | ||
9805 | if (whence_check) { | |
9806 | int r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
9807 | if (r < 0) | |
92f5a8d4 | 9808 | return r; |
92f5a8d4 TL |
9809 | } |
9810 | ||
7c673cae FG |
9811 | switch (whence) { |
9812 | case SEEK_SET: | |
11fdf7f2 | 9813 | pos = offset; |
7c673cae FG |
9814 | break; |
9815 | ||
9816 | case SEEK_CUR: | |
92f5a8d4 | 9817 | pos = f->pos + offset; |
7c673cae FG |
9818 | break; |
9819 | ||
9820 | case SEEK_END: | |
11fdf7f2 | 9821 | pos = in->size + offset; |
7c673cae FG |
9822 | break; |
9823 | ||
9f95a23c | 9824 | #ifdef SEEK_DATA |
92f5a8d4 | 9825 | case SEEK_DATA: |
9f95a23c | 9826 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9827 | return -CEPHFS_ENXIO; |
92f5a8d4 TL |
9828 | pos = offset; |
9829 | break; | |
9f95a23c | 9830 | #endif |
92f5a8d4 | 9831 | |
9f95a23c | 9832 | #ifdef SEEK_HOLE |
92f5a8d4 | 9833 | case SEEK_HOLE: |
9f95a23c | 9834 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9835 | return -CEPHFS_ENXIO; |
9f95a23c | 9836 | pos = in->size; |
92f5a8d4 | 9837 | break; |
9f95a23c | 9838 | #endif |
92f5a8d4 | 9839 | |
7c673cae | 9840 | default: |
92f5a8d4 | 9841 | ldout(cct, 1) << __func__ << ": invalid whence value " << whence << dendl; |
f67539c2 | 9842 | return -CEPHFS_EINVAL; |
7c673cae FG |
9843 | } |
9844 | ||
11fdf7f2 | 9845 | if (pos < 0) { |
f67539c2 | 9846 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
9847 | } else { |
9848 | f->pos = pos; | |
9849 | } | |
9850 | ||
1adf2230 | 9851 | ldout(cct, 8) << "_lseek(" << f << ", " << offset << ", " << whence << ") = " << f->pos << dendl; |
7c673cae FG |
9852 | return f->pos; |
9853 | } | |
9854 | ||
9855 | ||
9856 | void Client::lock_fh_pos(Fh *f) | |
9857 | { | |
11fdf7f2 | 9858 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae FG |
9859 | |
9860 | if (f->pos_locked || !f->pos_waiters.empty()) { | |
9f95a23c | 9861 | ceph::condition_variable cond; |
7c673cae | 9862 | f->pos_waiters.push_back(&cond); |
11fdf7f2 | 9863 | ldout(cct, 10) << __func__ << " BLOCKING on " << f << dendl; |
9f95a23c TL |
9864 | std::unique_lock l{client_lock, std::adopt_lock}; |
9865 | cond.wait(l, [f, me=&cond] { | |
9866 | return !f->pos_locked && f->pos_waiters.front() == me; | |
9867 | }); | |
9868 | l.release(); | |
11fdf7f2 TL |
9869 | ldout(cct, 10) << __func__ << " UNBLOCKING on " << f << dendl; |
9870 | ceph_assert(f->pos_waiters.front() == &cond); | |
7c673cae FG |
9871 | f->pos_waiters.pop_front(); |
9872 | } | |
9873 | ||
9874 | f->pos_locked = true; | |
9875 | } | |
9876 | ||
9877 | void Client::unlock_fh_pos(Fh *f) | |
9878 | { | |
f67539c2 TL |
9879 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9880 | ||
11fdf7f2 | 9881 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae | 9882 | f->pos_locked = false; |
f67539c2 TL |
9883 | if (!f->pos_waiters.empty()) { |
9884 | // only wake up the oldest waiter | |
9885 | auto cond = f->pos_waiters.front(); | |
9886 | cond->notify_one(); | |
9887 | } | |
7c673cae FG |
9888 | } |
9889 | ||
9890 | int Client::uninline_data(Inode *in, Context *onfinish) | |
9891 | { | |
9892 | if (!in->inline_data.length()) { | |
9893 | onfinish->complete(0); | |
9894 | return 0; | |
9895 | } | |
9896 | ||
9897 | char oid_buf[32]; | |
9898 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (long long unsigned)in->ino); | |
9899 | object_t oid = oid_buf; | |
9900 | ||
9901 | ObjectOperation create_ops; | |
9902 | create_ops.create(false); | |
9903 | ||
9904 | objecter->mutate(oid, | |
9905 | OSDMap::file_to_object_locator(in->layout), | |
9906 | create_ops, | |
9907 | in->snaprealm->get_snap_context(), | |
9908 | ceph::real_clock::now(), | |
9909 | 0, | |
9910 | NULL); | |
9911 | ||
9912 | bufferlist inline_version_bl; | |
11fdf7f2 | 9913 | encode(in->inline_version, inline_version_bl); |
7c673cae FG |
9914 | |
9915 | ObjectOperation uninline_ops; | |
9916 | uninline_ops.cmpxattr("inline_version", | |
9917 | CEPH_OSD_CMPXATTR_OP_GT, | |
9918 | CEPH_OSD_CMPXATTR_MODE_U64, | |
9919 | inline_version_bl); | |
9920 | bufferlist inline_data = in->inline_data; | |
9921 | uninline_ops.write(0, inline_data, in->truncate_size, in->truncate_seq); | |
9922 | uninline_ops.setxattr("inline_version", stringify(in->inline_version)); | |
9923 | ||
9924 | objecter->mutate(oid, | |
9925 | OSDMap::file_to_object_locator(in->layout), | |
9926 | uninline_ops, | |
9927 | in->snaprealm->get_snap_context(), | |
9928 | ceph::real_clock::now(), | |
9929 | 0, | |
9930 | onfinish); | |
9931 | ||
9932 | return 0; | |
9933 | } | |
9934 | ||
9935 | // | |
9936 | ||
9937 | // blocking osd interface | |
9938 | ||
9939 | int Client::read(int fd, char *buf, loff_t size, loff_t offset) | |
9940 | { | |
f67539c2 TL |
9941 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9942 | if (!mref_reader.is_state_satisfied()) | |
9943 | return -CEPHFS_ENOTCONN; | |
9944 | ||
7c673cae FG |
9945 | tout(cct) << "read" << std::endl; |
9946 | tout(cct) << fd << std::endl; | |
9947 | tout(cct) << size << std::endl; | |
9948 | tout(cct) << offset << std::endl; | |
9949 | ||
f67539c2 | 9950 | std::unique_lock lock(client_lock); |
7c673cae FG |
9951 | Fh *f = get_filehandle(fd); |
9952 | if (!f) | |
f67539c2 | 9953 | return -CEPHFS_EBADF; |
7c673cae FG |
9954 | #if defined(__linux__) && defined(O_PATH) |
9955 | if (f->flags & O_PATH) | |
f67539c2 | 9956 | return -CEPHFS_EBADF; |
7c673cae FG |
9957 | #endif |
9958 | bufferlist bl; | |
11fdf7f2 TL |
9959 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
9960 | size = std::min(size, (loff_t)INT_MAX); | |
7c673cae FG |
9961 | int r = _read(f, offset, size, &bl); |
9962 | ldout(cct, 3) << "read(" << fd << ", " << (void*)buf << ", " << size << ", " << offset << ") = " << r << dendl; | |
9963 | if (r >= 0) { | |
f6b5b4d7 | 9964 | lock.unlock(); |
9f95a23c | 9965 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
9966 | r = bl.length(); |
9967 | } | |
9968 | return r; | |
9969 | } | |
9970 | ||
9971 | int Client::preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset) | |
9972 | { | |
9973 | if (iovcnt < 0) | |
f67539c2 | 9974 | return -CEPHFS_EINVAL; |
7c673cae FG |
9975 | return _preadv_pwritev(fd, iov, iovcnt, offset, false); |
9976 | } | |
9977 | ||
11fdf7f2 | 9978 | int64_t Client::_read(Fh *f, int64_t offset, uint64_t size, bufferlist *bl) |
7c673cae | 9979 | { |
f67539c2 TL |
9980 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9981 | ||
11fdf7f2 TL |
9982 | int want, have = 0; |
9983 | bool movepos = false; | |
9984 | std::unique_ptr<C_SaferCond> onuninline; | |
adb31ebb | 9985 | int64_t rc = 0; |
11fdf7f2 | 9986 | const auto& conf = cct->_conf; |
7c673cae | 9987 | Inode *in = f->inode.get(); |
11fdf7f2 TL |
9988 | utime_t lat; |
9989 | utime_t start = ceph_clock_now(); | |
7c673cae FG |
9990 | |
9991 | if ((f->mode & CEPH_FILE_MODE_RD) == 0) | |
f67539c2 | 9992 | return -CEPHFS_EBADF; |
7c673cae FG |
9993 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
9994 | ||
7c673cae FG |
9995 | if (offset < 0) { |
9996 | lock_fh_pos(f); | |
9997 | offset = f->pos; | |
9998 | movepos = true; | |
9999 | } | |
10000 | loff_t start_pos = offset; | |
10001 | ||
10002 | if (in->inline_version == 0) { | |
adb31ebb | 10003 | auto r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); |
c07f9fc5 | 10004 | if (r < 0) { |
adb31ebb | 10005 | rc = r; |
11fdf7f2 | 10006 | goto done; |
c07f9fc5 | 10007 | } |
11fdf7f2 | 10008 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
10009 | } |
10010 | ||
10011 | retry: | |
11fdf7f2 TL |
10012 | if (f->mode & CEPH_FILE_MODE_LAZY) |
10013 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
10014 | else | |
10015 | want = CEPH_CAP_FILE_CACHE; | |
adb31ebb TL |
10016 | { |
10017 | auto r = get_caps(f, CEPH_CAP_FILE_RD, want, &have, -1); | |
10018 | if (r < 0) { | |
10019 | rc = r; | |
10020 | goto done; | |
10021 | } | |
c07f9fc5 | 10022 | } |
7c673cae | 10023 | if (f->flags & O_DIRECT) |
11fdf7f2 | 10024 | have &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
10025 | |
10026 | if (in->inline_version < CEPH_INLINE_NONE) { | |
10027 | if (!(have & CEPH_CAP_FILE_CACHE)) { | |
11fdf7f2 TL |
10028 | onuninline.reset(new C_SaferCond("Client::_read_uninline_data flock")); |
10029 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
10030 | } else { |
10031 | uint32_t len = in->inline_data.length(); | |
7c673cae FG |
10032 | uint64_t endoff = offset + size; |
10033 | if (endoff > in->size) | |
10034 | endoff = in->size; | |
10035 | ||
10036 | if (offset < len) { | |
10037 | if (endoff <= len) { | |
10038 | bl->substr_of(in->inline_data, offset, endoff - offset); | |
10039 | } else { | |
10040 | bl->substr_of(in->inline_data, offset, len - offset); | |
10041 | bl->append_zero(endoff - len); | |
10042 | } | |
adb31ebb | 10043 | rc = endoff - offset; |
7c673cae FG |
10044 | } else if ((uint64_t)offset < endoff) { |
10045 | bl->append_zero(endoff - offset); | |
adb31ebb | 10046 | rc = endoff - offset; |
11fdf7f2 | 10047 | } else { |
adb31ebb | 10048 | rc = 0; |
7c673cae | 10049 | } |
7c673cae FG |
10050 | goto success; |
10051 | } | |
10052 | } | |
10053 | ||
10054 | if (!conf->client_debug_force_sync_read && | |
11fdf7f2 TL |
10055 | conf->client_oc && |
10056 | (have & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
10057 | |
10058 | if (f->flags & O_RSYNC) { | |
10059 | _flush_range(in, offset, size); | |
10060 | } | |
adb31ebb TL |
10061 | rc = _read_async(f, offset, size, bl); |
10062 | if (rc < 0) | |
7c673cae FG |
10063 | goto done; |
10064 | } else { | |
10065 | if (f->flags & O_DIRECT) | |
10066 | _flush_range(in, offset, size); | |
10067 | ||
10068 | bool checkeof = false; | |
adb31ebb TL |
10069 | rc = _read_sync(f, offset, size, bl, &checkeof); |
10070 | if (rc < 0) | |
7c673cae FG |
10071 | goto done; |
10072 | if (checkeof) { | |
adb31ebb TL |
10073 | offset += rc; |
10074 | size -= rc; | |
7c673cae FG |
10075 | |
10076 | put_cap_ref(in, CEPH_CAP_FILE_RD); | |
10077 | have = 0; | |
10078 | // reverify size | |
adb31ebb TL |
10079 | { |
10080 | auto r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
10081 | if (r < 0) { | |
10082 | rc = r; | |
10083 | goto done; | |
10084 | } | |
10085 | } | |
7c673cae FG |
10086 | |
10087 | // eof? short read. | |
10088 | if ((uint64_t)offset < in->size) | |
10089 | goto retry; | |
10090 | } | |
10091 | } | |
10092 | ||
10093 | success: | |
adb31ebb | 10094 | ceph_assert(rc >= 0); |
a4b75251 | 10095 | update_read_io_size(bl->length()); |
7c673cae FG |
10096 | if (movepos) { |
10097 | // adjust fd pos | |
adb31ebb | 10098 | f->pos = start_pos + rc; |
7c673cae | 10099 | } |
11fdf7f2 TL |
10100 | |
10101 | lat = ceph_clock_now(); | |
10102 | lat -= start; | |
10103 | logger->tinc(l_c_read, lat); | |
7c673cae FG |
10104 | |
10105 | done: | |
10106 | // done! | |
11fdf7f2 | 10107 | |
7c673cae | 10108 | if (onuninline) { |
9f95a23c | 10109 | client_lock.unlock(); |
11fdf7f2 | 10110 | int ret = onuninline->wait(); |
9f95a23c | 10111 | client_lock.lock(); |
f67539c2 | 10112 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
10113 | in->inline_data.clear(); |
10114 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 10115 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10116 | check_caps(in, 0); |
10117 | } else | |
adb31ebb | 10118 | rc = ret; |
7c673cae | 10119 | } |
11fdf7f2 | 10120 | if (have) { |
7c673cae | 10121 | put_cap_ref(in, CEPH_CAP_FILE_RD); |
11fdf7f2 TL |
10122 | } |
10123 | if (movepos) { | |
10124 | unlock_fh_pos(f); | |
10125 | } | |
adb31ebb | 10126 | return rc; |
7c673cae FG |
10127 | } |
10128 | ||
10129 | Client::C_Readahead::C_Readahead(Client *c, Fh *f) : | |
10130 | client(c), f(f) { | |
10131 | f->get(); | |
10132 | f->readahead.inc_pending(); | |
10133 | } | |
10134 | ||
10135 | Client::C_Readahead::~C_Readahead() { | |
10136 | f->readahead.dec_pending(); | |
10137 | client->_put_fh(f); | |
10138 | } | |
10139 | ||
10140 | void Client::C_Readahead::finish(int r) { | |
10141 | lgeneric_subdout(client->cct, client, 20) << "client." << client->get_nodeid() << " " << "C_Readahead on " << f->inode << dendl; | |
10142 | client->put_cap_ref(f->inode.get(), CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
a4b75251 TL |
10143 | if (r > 0) { |
10144 | client->update_read_io_size(r); | |
10145 | } | |
7c673cae FG |
10146 | } |
10147 | ||
10148 | int Client::_read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl) | |
10149 | { | |
f67539c2 TL |
10150 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10151 | ||
11fdf7f2 | 10152 | const auto& conf = cct->_conf; |
7c673cae FG |
10153 | Inode *in = f->inode.get(); |
10154 | ||
11fdf7f2 | 10155 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
10156 | |
10157 | // trim read based on file size? | |
10158 | if (off >= in->size) | |
10159 | return 0; | |
10160 | if (len == 0) | |
10161 | return 0; | |
10162 | if (off + len > in->size) { | |
10163 | len = in->size - off; | |
10164 | } | |
10165 | ||
10166 | ldout(cct, 10) << " min_bytes=" << f->readahead.get_min_readahead_size() | |
10167 | << " max_bytes=" << f->readahead.get_max_readahead_size() | |
10168 | << " max_periods=" << conf->client_readahead_max_periods << dendl; | |
10169 | ||
10170 | // read (and possibly block) | |
11fdf7f2 TL |
10171 | int r = 0; |
10172 | C_SaferCond onfinish("Client::_read_async flock"); | |
7c673cae | 10173 | r = objectcacher->file_read(&in->oset, &in->layout, in->snapid, |
11fdf7f2 | 10174 | off, len, bl, 0, &onfinish); |
7c673cae FG |
10175 | if (r == 0) { |
10176 | get_cap_ref(in, CEPH_CAP_FILE_CACHE); | |
9f95a23c | 10177 | client_lock.unlock(); |
11fdf7f2 | 10178 | r = onfinish.wait(); |
9f95a23c | 10179 | client_lock.lock(); |
7c673cae | 10180 | put_cap_ref(in, CEPH_CAP_FILE_CACHE); |
a4b75251 | 10181 | update_read_io_size(bl->length()); |
7c673cae FG |
10182 | } |
10183 | ||
10184 | if(f->readahead.get_min_readahead_size() > 0) { | |
10185 | pair<uint64_t, uint64_t> readahead_extent = f->readahead.update(off, len, in->size); | |
10186 | if (readahead_extent.second > 0) { | |
10187 | ldout(cct, 20) << "readahead " << readahead_extent.first << "~" << readahead_extent.second | |
10188 | << " (caller wants " << off << "~" << len << ")" << dendl; | |
10189 | Context *onfinish2 = new C_Readahead(this, f); | |
10190 | int r2 = objectcacher->file_read(&in->oset, &in->layout, in->snapid, | |
10191 | readahead_extent.first, readahead_extent.second, | |
10192 | NULL, 0, onfinish2); | |
10193 | if (r2 == 0) { | |
10194 | ldout(cct, 20) << "readahead initiated, c " << onfinish2 << dendl; | |
10195 | get_cap_ref(in, CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
10196 | } else { | |
10197 | ldout(cct, 20) << "readahead was no-op, already cached" << dendl; | |
10198 | delete onfinish2; | |
10199 | } | |
10200 | } | |
10201 | } | |
10202 | ||
10203 | return r; | |
10204 | } | |
10205 | ||
10206 | int Client::_read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, | |
10207 | bool *checkeof) | |
10208 | { | |
f67539c2 TL |
10209 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10210 | ||
7c673cae FG |
10211 | Inode *in = f->inode.get(); |
10212 | uint64_t pos = off; | |
10213 | int left = len; | |
10214 | int read = 0; | |
10215 | ||
11fdf7f2 | 10216 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae | 10217 | |
f67539c2 TL |
10218 | // 0 success, 1 continue and < 0 error happen. |
10219 | auto wait_and_copy = [&](C_SaferCond &onfinish, bufferlist &tbl, int wanted) { | |
11fdf7f2 | 10220 | int r = onfinish.wait(); |
7c673cae FG |
10221 | |
10222 | // if we get ENOENT from OSD, assume 0 bytes returned | |
f67539c2 | 10223 | if (r == -CEPHFS_ENOENT) |
7c673cae FG |
10224 | r = 0; |
10225 | if (r < 0) | |
10226 | return r; | |
f67539c2 | 10227 | |
7c673cae FG |
10228 | if (tbl.length()) { |
10229 | r = tbl.length(); | |
10230 | ||
10231 | read += r; | |
10232 | pos += r; | |
10233 | left -= r; | |
10234 | bl->claim_append(tbl); | |
10235 | } | |
10236 | // short read? | |
10237 | if (r >= 0 && r < wanted) { | |
10238 | if (pos < in->size) { | |
10239 | // zero up to known EOF | |
10240 | int64_t some = in->size - pos; | |
10241 | if (some > left) | |
10242 | some = left; | |
11fdf7f2 TL |
10243 | auto z = buffer::ptr_node::create(some); |
10244 | z->zero(); | |
10245 | bl->push_back(std::move(z)); | |
7c673cae FG |
10246 | read += some; |
10247 | pos += some; | |
10248 | left -= some; | |
10249 | if (left == 0) | |
f67539c2 | 10250 | return 0; |
7c673cae FG |
10251 | } |
10252 | ||
10253 | *checkeof = true; | |
f67539c2 | 10254 | return 0; |
7c673cae | 10255 | } |
f67539c2 TL |
10256 | return 1; |
10257 | }; | |
7c673cae | 10258 | |
f67539c2 TL |
10259 | while (left > 0) { |
10260 | C_SaferCond onfinish("Client::_read_sync flock"); | |
10261 | bufferlist tbl; | |
7c673cae | 10262 | |
f67539c2 TL |
10263 | int wanted = left; |
10264 | filer->read_trunc(in->ino, &in->layout, in->snapid, | |
10265 | pos, left, &tbl, 0, | |
10266 | in->truncate_size, in->truncate_seq, | |
10267 | &onfinish); | |
10268 | client_lock.unlock(); | |
10269 | int r = wait_and_copy(onfinish, tbl, wanted); | |
10270 | client_lock.lock(); | |
10271 | if (!r) | |
10272 | return read; | |
10273 | if (r < 0) | |
10274 | return r; | |
7c673cae | 10275 | } |
f67539c2 | 10276 | return read; |
7c673cae FG |
10277 | } |
10278 | ||
10279 | int Client::write(int fd, const char *buf, loff_t size, loff_t offset) | |
10280 | { | |
f67539c2 TL |
10281 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10282 | if (!mref_reader.is_state_satisfied()) | |
10283 | return -CEPHFS_ENOTCONN; | |
10284 | ||
7c673cae FG |
10285 | tout(cct) << "write" << std::endl; |
10286 | tout(cct) << fd << std::endl; | |
10287 | tout(cct) << size << std::endl; | |
10288 | tout(cct) << offset << std::endl; | |
10289 | ||
f67539c2 | 10290 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10291 | Fh *fh = get_filehandle(fd); |
10292 | if (!fh) | |
f67539c2 | 10293 | return -CEPHFS_EBADF; |
7c673cae FG |
10294 | #if defined(__linux__) && defined(O_PATH) |
10295 | if (fh->flags & O_PATH) | |
f67539c2 | 10296 | return -CEPHFS_EBADF; |
7c673cae | 10297 | #endif |
11fdf7f2 TL |
10298 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
10299 | size = std::min(size, (loff_t)INT_MAX); | |
10300 | int r = _write(fh, offset, size, buf, NULL, false); | |
7c673cae FG |
10301 | ldout(cct, 3) << "write(" << fd << ", \"...\", " << size << ", " << offset << ") = " << r << dendl; |
10302 | return r; | |
10303 | } | |
10304 | ||
10305 | int Client::pwritev(int fd, const struct iovec *iov, int iovcnt, int64_t offset) | |
10306 | { | |
10307 | if (iovcnt < 0) | |
f67539c2 | 10308 | return -CEPHFS_EINVAL; |
7c673cae FG |
10309 | return _preadv_pwritev(fd, iov, iovcnt, offset, true); |
10310 | } | |
10311 | ||
11fdf7f2 | 10312 | int64_t Client::_preadv_pwritev_locked(Fh *fh, const struct iovec *iov, |
20effc67 TL |
10313 | unsigned iovcnt, int64_t offset, |
10314 | bool write, bool clamp_to_int) | |
7c673cae | 10315 | { |
20effc67 TL |
10316 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10317 | ||
7c673cae FG |
10318 | #if defined(__linux__) && defined(O_PATH) |
10319 | if (fh->flags & O_PATH) | |
f67539c2 | 10320 | return -CEPHFS_EBADF; |
7c673cae FG |
10321 | #endif |
10322 | loff_t totallen = 0; | |
10323 | for (unsigned i = 0; i < iovcnt; i++) { | |
10324 | totallen += iov[i].iov_len; | |
10325 | } | |
11fdf7f2 TL |
10326 | |
10327 | /* | |
10328 | * Some of the API functions take 64-bit size values, but only return | |
10329 | * 32-bit signed integers. Clamp the I/O sizes in those functions so that | |
10330 | * we don't do I/Os larger than the values we can return. | |
10331 | */ | |
10332 | if (clamp_to_int) { | |
10333 | totallen = std::min(totallen, (loff_t)INT_MAX); | |
10334 | } | |
7c673cae | 10335 | if (write) { |
11fdf7f2 TL |
10336 | int64_t w = _write(fh, offset, totallen, NULL, iov, iovcnt); |
10337 | ldout(cct, 3) << "pwritev(" << fh << ", \"...\", " << totallen << ", " << offset << ") = " << w << dendl; | |
7c673cae FG |
10338 | return w; |
10339 | } else { | |
10340 | bufferlist bl; | |
11fdf7f2 TL |
10341 | int64_t r = _read(fh, offset, totallen, &bl); |
10342 | ldout(cct, 3) << "preadv(" << fh << ", " << offset << ") = " << r << dendl; | |
7c673cae FG |
10343 | if (r <= 0) |
10344 | return r; | |
10345 | ||
20effc67 | 10346 | client_lock.unlock(); |
9f95a23c | 10347 | auto iter = bl.cbegin(); |
7c673cae FG |
10348 | for (unsigned j = 0, resid = r; j < iovcnt && resid > 0; j++) { |
10349 | /* | |
f67539c2 TL |
10350 | * This piece of code aims to handle the case that bufferlist |
10351 | * does not have enough data to fill in the iov | |
7c673cae | 10352 | */ |
9f95a23c TL |
10353 | const auto round_size = std::min<unsigned>(resid, iov[j].iov_len); |
10354 | iter.copy(round_size, reinterpret_cast<char*>(iov[j].iov_base)); | |
10355 | resid -= round_size; | |
10356 | /* iter is self-updating */ | |
7c673cae | 10357 | } |
20effc67 | 10358 | client_lock.lock(); |
f67539c2 | 10359 | return r; |
7c673cae FG |
10360 | } |
10361 | } | |
10362 | ||
11fdf7f2 TL |
10363 | int Client::_preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write) |
10364 | { | |
f67539c2 TL |
10365 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10366 | if (!mref_reader.is_state_satisfied()) | |
10367 | return -CEPHFS_ENOTCONN; | |
10368 | ||
11fdf7f2 TL |
10369 | tout(cct) << fd << std::endl; |
10370 | tout(cct) << offset << std::endl; | |
10371 | ||
20effc67 | 10372 | std::scoped_lock cl(client_lock); |
11fdf7f2 TL |
10373 | Fh *fh = get_filehandle(fd); |
10374 | if (!fh) | |
f67539c2 | 10375 | return -CEPHFS_EBADF; |
20effc67 | 10376 | return _preadv_pwritev_locked(fh, iov, iovcnt, offset, write, true); |
11fdf7f2 TL |
10377 | } |
10378 | ||
10379 | int64_t Client::_write(Fh *f, int64_t offset, uint64_t size, const char *buf, | |
10380 | const struct iovec *iov, int iovcnt) | |
7c673cae | 10381 | { |
f67539c2 TL |
10382 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10383 | ||
f64942e4 AA |
10384 | uint64_t fpos = 0; |
10385 | ||
7c673cae | 10386 | if ((uint64_t)(offset+size) > mdsmap->get_max_filesize()) //too large! |
f67539c2 | 10387 | return -CEPHFS_EFBIG; |
7c673cae FG |
10388 | |
10389 | //ldout(cct, 7) << "write fh " << fh << " size " << size << " offset " << offset << dendl; | |
10390 | Inode *in = f->inode.get(); | |
10391 | ||
10392 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
f67539c2 | 10393 | return -CEPHFS_ENOSPC; |
7c673cae FG |
10394 | } |
10395 | ||
11fdf7f2 | 10396 | ceph_assert(in->snapid == CEPH_NOSNAP); |
7c673cae FG |
10397 | |
10398 | // was Fh opened as writeable? | |
10399 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 10400 | return -CEPHFS_EBADF; |
7c673cae | 10401 | |
7c673cae FG |
10402 | // use/adjust fd pos? |
10403 | if (offset < 0) { | |
10404 | lock_fh_pos(f); | |
10405 | /* | |
10406 | * FIXME: this is racy in that we may block _after_ this point waiting for caps, and size may | |
10407 | * change out from under us. | |
10408 | */ | |
10409 | if (f->flags & O_APPEND) { | |
9f95a23c | 10410 | auto r = _lseek(f, 0, SEEK_END); |
7c673cae FG |
10411 | if (r < 0) { |
10412 | unlock_fh_pos(f); | |
10413 | return r; | |
10414 | } | |
10415 | } | |
10416 | offset = f->pos; | |
f64942e4 | 10417 | fpos = offset+size; |
7c673cae FG |
10418 | unlock_fh_pos(f); |
10419 | } | |
10420 | ||
11fdf7f2 TL |
10421 | // check quota |
10422 | uint64_t endoff = offset + size; | |
10423 | if (endoff > in->size && is_quota_bytes_exceeded(in, endoff - in->size, | |
10424 | f->actor_perms)) { | |
f67539c2 | 10425 | return -CEPHFS_EDQUOT; |
11fdf7f2 TL |
10426 | } |
10427 | ||
7c673cae FG |
10428 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
10429 | ||
10430 | ldout(cct, 10) << "cur file size is " << in->size << dendl; | |
10431 | ||
10432 | // time it. | |
10433 | utime_t start = ceph_clock_now(); | |
10434 | ||
10435 | if (in->inline_version == 0) { | |
10436 | int r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); | |
10437 | if (r < 0) | |
10438 | return r; | |
11fdf7f2 | 10439 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
10440 | } |
10441 | ||
10442 | // copy into fresh buffer (since our write may be resub, async) | |
10443 | bufferlist bl; | |
10444 | if (buf) { | |
10445 | if (size > 0) | |
10446 | bl.append(buf, size); | |
10447 | } else if (iov){ | |
10448 | for (int i = 0; i < iovcnt; i++) { | |
10449 | if (iov[i].iov_len > 0) { | |
10450 | bl.append((const char *)iov[i].iov_base, iov[i].iov_len); | |
10451 | } | |
10452 | } | |
10453 | } | |
10454 | ||
10455 | utime_t lat; | |
10456 | uint64_t totalwritten; | |
11fdf7f2 TL |
10457 | int want, have; |
10458 | if (f->mode & CEPH_FILE_MODE_LAZY) | |
10459 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
10460 | else | |
10461 | want = CEPH_CAP_FILE_BUFFER; | |
f6b5b4d7 | 10462 | int r = get_caps(f, CEPH_CAP_FILE_WR|CEPH_CAP_AUTH_SHARED, want, &have, endoff); |
7c673cae FG |
10463 | if (r < 0) |
10464 | return r; | |
10465 | ||
10466 | /* clear the setuid/setgid bits, if any */ | |
181888fb | 10467 | if (unlikely(in->mode & (S_ISUID|S_ISGID)) && size > 0) { |
7c673cae FG |
10468 | struct ceph_statx stx = { 0 }; |
10469 | ||
10470 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10471 | r = __setattrx(in, &stx, CEPH_SETATTR_KILL_SGUID, f->actor_perms); | |
10472 | if (r < 0) | |
10473 | return r; | |
10474 | } else { | |
10475 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10476 | } | |
10477 | ||
10478 | if (f->flags & O_DIRECT) | |
11fdf7f2 | 10479 | have &= ~(CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
10480 | |
10481 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
10482 | ||
11fdf7f2 TL |
10483 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
10484 | ||
7c673cae FG |
10485 | if (in->inline_version < CEPH_INLINE_NONE) { |
10486 | if (endoff > cct->_conf->client_max_inline_size || | |
10487 | endoff > CEPH_INLINE_MAX_SIZE || | |
10488 | !(have & CEPH_CAP_FILE_BUFFER)) { | |
11fdf7f2 TL |
10489 | onuninline.reset(new C_SaferCond("Client::_write_uninline_data flock")); |
10490 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
10491 | } else { |
10492 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10493 | ||
10494 | uint32_t len = in->inline_data.length(); | |
10495 | ||
10496 | if (endoff < len) | |
9f95a23c | 10497 | in->inline_data.begin(endoff).copy(len - endoff, bl); // XXX |
7c673cae FG |
10498 | |
10499 | if (offset < len) | |
10500 | in->inline_data.splice(offset, len - offset); | |
10501 | else if (offset > len) | |
10502 | in->inline_data.append_zero(offset - len); | |
10503 | ||
10504 | in->inline_data.append(bl); | |
10505 | in->inline_version++; | |
10506 | ||
10507 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10508 | ||
10509 | goto success; | |
10510 | } | |
10511 | } | |
10512 | ||
11fdf7f2 TL |
10513 | if (cct->_conf->client_oc && |
10514 | (have & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
10515 | // do buffered write |
10516 | if (!in->oset.dirty_or_tx) | |
10517 | get_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
10518 | ||
10519 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10520 | ||
10521 | // async, caching, non-blocking. | |
10522 | r = objectcacher->file_write(&in->oset, &in->layout, | |
10523 | in->snaprealm->get_snap_context(), | |
10524 | offset, size, bl, ceph::real_clock::now(), | |
10525 | 0); | |
10526 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10527 | ||
10528 | if (r < 0) | |
10529 | goto done; | |
10530 | ||
10531 | // flush cached write if O_SYNC is set on file fh | |
10532 | // O_DSYNC == O_SYNC on linux < 2.6.33 | |
10533 | // O_SYNC = __O_SYNC | O_DSYNC on linux >= 2.6.33 | |
10534 | if ((f->flags & O_SYNC) || (f->flags & O_DSYNC)) { | |
10535 | _flush_range(in, offset, size); | |
10536 | } | |
10537 | } else { | |
10538 | if (f->flags & O_DIRECT) | |
10539 | _flush_range(in, offset, size); | |
10540 | ||
10541 | // simple, non-atomic sync write | |
11fdf7f2 | 10542 | C_SaferCond onfinish("Client::_write flock"); |
f67539c2 | 10543 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
10544 | |
10545 | filer->write_trunc(in->ino, &in->layout, in->snaprealm->get_snap_context(), | |
10546 | offset, size, bl, ceph::real_clock::now(), 0, | |
10547 | in->truncate_size, in->truncate_seq, | |
11fdf7f2 | 10548 | &onfinish); |
9f95a23c | 10549 | client_lock.unlock(); |
f6b5b4d7 | 10550 | r = onfinish.wait(); |
9f95a23c | 10551 | client_lock.lock(); |
f67539c2 | 10552 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
f6b5b4d7 TL |
10553 | if (r < 0) |
10554 | goto done; | |
7c673cae FG |
10555 | } |
10556 | ||
10557 | // if we get here, write was successful, update client metadata | |
10558 | success: | |
a4b75251 | 10559 | update_write_io_size(size); |
7c673cae FG |
10560 | // time |
10561 | lat = ceph_clock_now(); | |
10562 | lat -= start; | |
10563 | logger->tinc(l_c_wrlat, lat); | |
10564 | ||
f64942e4 AA |
10565 | if (fpos) { |
10566 | lock_fh_pos(f); | |
10567 | f->pos = fpos; | |
10568 | unlock_fh_pos(f); | |
10569 | } | |
7c673cae | 10570 | totalwritten = size; |
11fdf7f2 | 10571 | r = (int64_t)totalwritten; |
7c673cae FG |
10572 | |
10573 | // extend file? | |
10574 | if (totalwritten + offset > in->size) { | |
10575 | in->size = totalwritten + offset; | |
28e407b8 | 10576 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 10577 | |
11fdf7f2 | 10578 | if (is_quota_bytes_approaching(in, f->actor_perms)) { |
7c673cae | 10579 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
10580 | } else if (is_max_size_approaching(in)) { |
10581 | check_caps(in, 0); | |
7c673cae FG |
10582 | } |
10583 | ||
10584 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", extending file size" << dendl; | |
10585 | } else { | |
10586 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", leaving file size at " << in->size << dendl; | |
10587 | } | |
10588 | ||
10589 | // mtime | |
91327a77 | 10590 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 10591 | in->change_attr++; |
28e407b8 | 10592 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10593 | |
10594 | done: | |
10595 | ||
11fdf7f2 | 10596 | if (nullptr != onuninline) { |
9f95a23c | 10597 | client_lock.unlock(); |
11fdf7f2 | 10598 | int uninline_ret = onuninline->wait(); |
9f95a23c | 10599 | client_lock.lock(); |
7c673cae | 10600 | |
f67539c2 | 10601 | if (uninline_ret >= 0 || uninline_ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
10602 | in->inline_data.clear(); |
10603 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 10604 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10605 | check_caps(in, 0); |
10606 | } else | |
10607 | r = uninline_ret; | |
10608 | } | |
10609 | ||
10610 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
10611 | return r; | |
10612 | } | |
10613 | ||
10614 | int Client::_flush(Fh *f) | |
10615 | { | |
10616 | Inode *in = f->inode.get(); | |
10617 | int err = f->take_async_err(); | |
10618 | if (err != 0) { | |
10619 | ldout(cct, 1) << __func__ << ": " << f << " on inode " << *in << " caught async_err = " | |
10620 | << cpp_strerror(err) << dendl; | |
10621 | } else { | |
10622 | ldout(cct, 10) << __func__ << ": " << f << " on inode " << *in << " no async_err state" << dendl; | |
10623 | } | |
10624 | ||
10625 | return err; | |
10626 | } | |
10627 | ||
10628 | int Client::truncate(const char *relpath, loff_t length, const UserPerm& perms) | |
10629 | { | |
10630 | struct ceph_statx stx; | |
10631 | stx.stx_size = length; | |
10632 | return setattrx(relpath, &stx, CEPH_SETATTR_SIZE, perms); | |
10633 | } | |
10634 | ||
10635 | int Client::ftruncate(int fd, loff_t length, const UserPerm& perms) | |
10636 | { | |
f67539c2 TL |
10637 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10638 | if (!mref_reader.is_state_satisfied()) | |
10639 | return -CEPHFS_ENOTCONN; | |
10640 | ||
11fdf7f2 | 10641 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
10642 | tout(cct) << fd << std::endl; |
10643 | tout(cct) << length << std::endl; | |
10644 | ||
f67539c2 | 10645 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10646 | Fh *f = get_filehandle(fd); |
10647 | if (!f) | |
f67539c2 | 10648 | return -CEPHFS_EBADF; |
7c673cae FG |
10649 | #if defined(__linux__) && defined(O_PATH) |
10650 | if (f->flags & O_PATH) | |
f67539c2 | 10651 | return -CEPHFS_EBADF; |
7c673cae | 10652 | #endif |
adb31ebb | 10653 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) |
f67539c2 | 10654 | return -CEPHFS_EBADF; |
7c673cae FG |
10655 | struct stat attr; |
10656 | attr.st_size = length; | |
10657 | return _setattr(f->inode, &attr, CEPH_SETATTR_SIZE, perms); | |
10658 | } | |
10659 | ||
10660 | int Client::fsync(int fd, bool syncdataonly) | |
10661 | { | |
f67539c2 TL |
10662 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10663 | if (!mref_reader.is_state_satisfied()) | |
10664 | return -CEPHFS_ENOTCONN; | |
10665 | ||
7c673cae FG |
10666 | tout(cct) << "fsync" << std::endl; |
10667 | tout(cct) << fd << std::endl; | |
10668 | tout(cct) << syncdataonly << std::endl; | |
10669 | ||
f67539c2 | 10670 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10671 | Fh *f = get_filehandle(fd); |
10672 | if (!f) | |
f67539c2 | 10673 | return -CEPHFS_EBADF; |
7c673cae FG |
10674 | #if defined(__linux__) && defined(O_PATH) |
10675 | if (f->flags & O_PATH) | |
f67539c2 | 10676 | return -CEPHFS_EBADF; |
7c673cae FG |
10677 | #endif |
10678 | int r = _fsync(f, syncdataonly); | |
10679 | if (r == 0) { | |
10680 | // The IOs in this fsync were okay, but maybe something happened | |
10681 | // in the background that we shoudl be reporting? | |
10682 | r = f->take_async_err(); | |
1adf2230 | 10683 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly |
7c673cae FG |
10684 | << ") = 0, async_err = " << r << dendl; |
10685 | } else { | |
10686 | // Assume that an error we encountered during fsync, even reported | |
10687 | // synchronously, would also have applied the error to the Fh, and we | |
10688 | // should clear it here to avoid returning the same error again on next | |
10689 | // call. | |
1adf2230 | 10690 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly << ") = " |
7c673cae FG |
10691 | << r << dendl; |
10692 | f->take_async_err(); | |
10693 | } | |
10694 | return r; | |
10695 | } | |
10696 | ||
10697 | int Client::_fsync(Inode *in, bool syncdataonly) | |
10698 | { | |
f67539c2 TL |
10699 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10700 | ||
7c673cae | 10701 | int r = 0; |
11fdf7f2 | 10702 | std::unique_ptr<C_SaferCond> object_cacher_completion = nullptr; |
7c673cae FG |
10703 | ceph_tid_t flush_tid = 0; |
10704 | InodeRef tmp_ref; | |
11fdf7f2 TL |
10705 | utime_t lat; |
10706 | utime_t start = ceph_clock_now(); | |
7c673cae | 10707 | |
1adf2230 | 10708 | ldout(cct, 8) << "_fsync on " << *in << " " << (syncdataonly ? "(dataonly)":"(data+metadata)") << dendl; |
7c673cae FG |
10709 | |
10710 | if (cct->_conf->client_oc) { | |
11fdf7f2 TL |
10711 | object_cacher_completion.reset(new C_SaferCond("Client::_fsync::lock")); |
10712 | tmp_ref = in; // take a reference; C_SaferCond doesn't and _flush won't either | |
10713 | _flush(in, object_cacher_completion.get()); | |
7c673cae FG |
10714 | ldout(cct, 15) << "using return-valued form of _fsync" << dendl; |
10715 | } | |
10716 | ||
10717 | if (!syncdataonly && in->dirty_caps) { | |
10718 | check_caps(in, CHECK_CAPS_NODELAY|CHECK_CAPS_SYNCHRONOUS); | |
10719 | if (in->flushing_caps) | |
10720 | flush_tid = last_flush_tid; | |
10721 | } else ldout(cct, 10) << "no metadata needs to commit" << dendl; | |
10722 | ||
10723 | if (!syncdataonly && !in->unsafe_ops.empty()) { | |
522d829b | 10724 | flush_mdlog_sync(in); |
28e407b8 | 10725 | |
7c673cae FG |
10726 | MetaRequest *req = in->unsafe_ops.back(); |
10727 | ldout(cct, 15) << "waiting on unsafe requests, last tid " << req->get_tid() << dendl; | |
10728 | ||
10729 | req->get(); | |
10730 | wait_on_list(req->waitfor_safe); | |
10731 | put_request(req); | |
10732 | } | |
10733 | ||
11fdf7f2 | 10734 | if (nullptr != object_cacher_completion) { // wait on a real reply instead of guessing |
9f95a23c | 10735 | client_lock.unlock(); |
7c673cae | 10736 | ldout(cct, 15) << "waiting on data to flush" << dendl; |
11fdf7f2 | 10737 | r = object_cacher_completion->wait(); |
9f95a23c | 10738 | client_lock.lock(); |
7c673cae FG |
10739 | ldout(cct, 15) << "got " << r << " from flush writeback" << dendl; |
10740 | } else { | |
10741 | // FIXME: this can starve | |
10742 | while (in->cap_refs[CEPH_CAP_FILE_BUFFER] > 0) { | |
10743 | ldout(cct, 10) << "ino " << in->ino << " has " << in->cap_refs[CEPH_CAP_FILE_BUFFER] | |
10744 | << " uncommitted, waiting" << dendl; | |
10745 | wait_on_list(in->waitfor_commit); | |
10746 | } | |
10747 | } | |
10748 | ||
10749 | if (!r) { | |
10750 | if (flush_tid > 0) | |
10751 | wait_sync_caps(in, flush_tid); | |
10752 | ||
10753 | ldout(cct, 10) << "ino " << in->ino << " has no uncommitted writes" << dendl; | |
10754 | } else { | |
1adf2230 | 10755 | ldout(cct, 8) << "ino " << in->ino << " failed to commit to disk! " |
7c673cae FG |
10756 | << cpp_strerror(-r) << dendl; |
10757 | } | |
11fdf7f2 TL |
10758 | |
10759 | lat = ceph_clock_now(); | |
10760 | lat -= start; | |
10761 | logger->tinc(l_c_fsync, lat); | |
7c673cae FG |
10762 | |
10763 | return r; | |
10764 | } | |
10765 | ||
10766 | int Client::_fsync(Fh *f, bool syncdataonly) | |
10767 | { | |
1adf2230 | 10768 | ldout(cct, 8) << "_fsync(" << f << ", " << (syncdataonly ? "dataonly)":"data+metadata)") << dendl; |
7c673cae FG |
10769 | return _fsync(f->inode.get(), syncdataonly); |
10770 | } | |
10771 | ||
10772 | int Client::fstat(int fd, struct stat *stbuf, const UserPerm& perms, int mask) | |
10773 | { | |
f67539c2 TL |
10774 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10775 | if (!mref_reader.is_state_satisfied()) | |
10776 | return -CEPHFS_ENOTCONN; | |
10777 | ||
7c673cae FG |
10778 | tout(cct) << "fstat mask " << hex << mask << dec << std::endl; |
10779 | tout(cct) << fd << std::endl; | |
10780 | ||
f67539c2 | 10781 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10782 | Fh *f = get_filehandle(fd); |
10783 | if (!f) | |
f67539c2 | 10784 | return -CEPHFS_EBADF; |
7c673cae FG |
10785 | int r = _getattr(f->inode, mask, perms); |
10786 | if (r < 0) | |
10787 | return r; | |
10788 | fill_stat(f->inode, stbuf, NULL); | |
1adf2230 | 10789 | ldout(cct, 5) << "fstat(" << fd << ", " << stbuf << ") = " << r << dendl; |
7c673cae FG |
10790 | return r; |
10791 | } | |
10792 | ||
10793 | int Client::fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms, | |
10794 | unsigned int want, unsigned int flags) | |
10795 | { | |
f67539c2 TL |
10796 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10797 | if (!mref_reader.is_state_satisfied()) | |
10798 | return -CEPHFS_ENOTCONN; | |
10799 | ||
7c673cae FG |
10800 | tout(cct) << "fstatx flags " << hex << flags << " want " << want << dec << std::endl; |
10801 | tout(cct) << fd << std::endl; | |
10802 | ||
f67539c2 | 10803 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10804 | Fh *f = get_filehandle(fd); |
10805 | if (!f) | |
f67539c2 | 10806 | return -CEPHFS_EBADF; |
7c673cae FG |
10807 | |
10808 | unsigned mask = statx_to_mask(flags, want); | |
10809 | ||
10810 | int r = 0; | |
b3b6e05e | 10811 | if (mask) { |
7c673cae FG |
10812 | r = _getattr(f->inode, mask, perms); |
10813 | if (r < 0) { | |
10814 | ldout(cct, 3) << "fstatx exit on error!" << dendl; | |
10815 | return r; | |
10816 | } | |
10817 | } | |
10818 | ||
10819 | fill_statx(f->inode, mask, stx); | |
10820 | ldout(cct, 3) << "fstatx(" << fd << ", " << stx << ") = " << r << dendl; | |
10821 | return r; | |
10822 | } | |
10823 | ||
b3b6e05e TL |
10824 | int Client::statxat(int dirfd, const char *relpath, |
10825 | struct ceph_statx *stx, const UserPerm& perms, | |
10826 | unsigned int want, unsigned int flags) { | |
10827 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
10828 | if (!mref_reader.is_state_satisfied()) { | |
10829 | return -CEPHFS_ENOTCONN; | |
10830 | } | |
10831 | ||
10832 | tout(cct) << __func__ << " flags " << hex << flags << " want " << want << dec << std::endl; | |
10833 | tout(cct) << dirfd << std::endl; | |
10834 | tout(cct) << relpath << std::endl; | |
10835 | ||
10836 | unsigned mask = statx_to_mask(flags, want); | |
10837 | ||
10838 | InodeRef dirinode; | |
10839 | std::scoped_lock lock(client_lock); | |
10840 | int r = get_fd_inode(dirfd, &dirinode); | |
10841 | if (r < 0) { | |
10842 | return r; | |
10843 | } | |
10844 | ||
10845 | InodeRef in; | |
10846 | filepath path(relpath); | |
10847 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask, dirinode); | |
10848 | if (r < 0) { | |
10849 | return r; | |
10850 | } | |
10851 | r = _getattr(in, mask, perms); | |
10852 | if (r < 0) { | |
10853 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; | |
10854 | return r; | |
10855 | } | |
10856 | ||
10857 | fill_statx(in, mask, stx); | |
10858 | ldout(cct, 3) << __func__ << " dirfd" << dirfd << ", r= " << r << dendl; | |
10859 | return r; | |
10860 | } | |
10861 | ||
7c673cae FG |
10862 | // not written yet, but i want to link! |
10863 | ||
10864 | int Client::chdir(const char *relpath, std::string &new_cwd, | |
10865 | const UserPerm& perms) | |
10866 | { | |
f67539c2 TL |
10867 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10868 | if (!mref_reader.is_state_satisfied()) | |
10869 | return -CEPHFS_ENOTCONN; | |
10870 | ||
7c673cae FG |
10871 | tout(cct) << "chdir" << std::endl; |
10872 | tout(cct) << relpath << std::endl; | |
181888fb | 10873 | |
7c673cae FG |
10874 | filepath path(relpath); |
10875 | InodeRef in; | |
f67539c2 TL |
10876 | |
10877 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
10878 | int r = path_walk(path, &in, perms); |
10879 | if (r < 0) | |
10880 | return r; | |
92f5a8d4 TL |
10881 | |
10882 | if (!(in.get()->is_dir())) | |
f67539c2 | 10883 | return -CEPHFS_ENOTDIR; |
92f5a8d4 | 10884 | |
7c673cae FG |
10885 | if (cwd != in) |
10886 | cwd.swap(in); | |
10887 | ldout(cct, 3) << "chdir(" << relpath << ") cwd now " << cwd->ino << dendl; | |
10888 | ||
b5b8bbf5 | 10889 | _getcwd(new_cwd, perms); |
7c673cae FG |
10890 | return 0; |
10891 | } | |
10892 | ||
b5b8bbf5 | 10893 | void Client::_getcwd(string& dir, const UserPerm& perms) |
7c673cae FG |
10894 | { |
10895 | filepath path; | |
11fdf7f2 | 10896 | ldout(cct, 10) << __func__ << " " << *cwd << dendl; |
7c673cae FG |
10897 | |
10898 | Inode *in = cwd.get(); | |
b3b6e05e | 10899 | while (in != root.get()) { |
11fdf7f2 | 10900 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked |
7c673cae FG |
10901 | |
10902 | // A cwd or ancester is unlinked | |
11fdf7f2 | 10903 | if (in->dentries.empty()) { |
7c673cae FG |
10904 | return; |
10905 | } | |
10906 | ||
10907 | Dentry *dn = in->get_first_parent(); | |
10908 | ||
10909 | ||
10910 | if (!dn) { | |
10911 | // look it up | |
11fdf7f2 | 10912 | ldout(cct, 10) << __func__ << " looking up parent for " << *in << dendl; |
7c673cae FG |
10913 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
10914 | filepath path(in->ino); | |
10915 | req->set_filepath(path); | |
10916 | req->set_inode(in); | |
10917 | int res = make_request(req, perms); | |
10918 | if (res < 0) | |
10919 | break; | |
10920 | ||
10921 | // start over | |
10922 | path = filepath(); | |
10923 | in = cwd.get(); | |
10924 | continue; | |
10925 | } | |
10926 | path.push_front_dentry(dn->name); | |
10927 | in = dn->dir->parent_inode; | |
10928 | } | |
10929 | dir = "/"; | |
10930 | dir += path.get_path(); | |
10931 | } | |
10932 | ||
b5b8bbf5 FG |
10933 | void Client::getcwd(string& dir, const UserPerm& perms) |
10934 | { | |
f67539c2 TL |
10935 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10936 | if (!mref_reader.is_state_satisfied()) | |
10937 | return; | |
10938 | ||
10939 | std::scoped_lock l(client_lock); | |
10940 | ||
10941 | _getcwd(dir, perms); | |
b5b8bbf5 FG |
10942 | } |
10943 | ||
7c673cae FG |
10944 | int Client::statfs(const char *path, struct statvfs *stbuf, |
10945 | const UserPerm& perms) | |
10946 | { | |
f67539c2 TL |
10947 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10948 | if (!mref_reader.is_state_satisfied()) | |
10949 | return -CEPHFS_ENOTCONN; | |
10950 | ||
11fdf7f2 | 10951 | tout(cct) << __func__ << std::endl; |
91327a77 | 10952 | unsigned long int total_files_on_fs; |
7c673cae FG |
10953 | |
10954 | ceph_statfs stats; | |
10955 | C_SaferCond cond; | |
d2e6a577 | 10956 | |
f67539c2 | 10957 | std::unique_lock lock(client_lock); |
d2e6a577 FG |
10958 | const vector<int64_t> &data_pools = mdsmap->get_data_pools(); |
10959 | if (data_pools.size() == 1) { | |
10960 | objecter->get_fs_stats(stats, data_pools[0], &cond); | |
10961 | } else { | |
20effc67 | 10962 | objecter->get_fs_stats(stats, std::optional<int64_t>(), &cond); |
d2e6a577 | 10963 | } |
7c673cae | 10964 | |
f67539c2 | 10965 | lock.unlock(); |
7c673cae | 10966 | int rval = cond.wait(); |
f67539c2 TL |
10967 | lock.lock(); |
10968 | ||
20effc67 | 10969 | ceph_assert(root); |
91327a77 | 10970 | total_files_on_fs = root->rstat.rfiles + root->rstat.rsubdirs; |
7c673cae FG |
10971 | |
10972 | if (rval < 0) { | |
10973 | ldout(cct, 1) << "underlying call to statfs returned error: " | |
10974 | << cpp_strerror(rval) | |
10975 | << dendl; | |
10976 | return rval; | |
10977 | } | |
10978 | ||
10979 | memset(stbuf, 0, sizeof(*stbuf)); | |
10980 | ||
10981 | /* | |
10982 | * we're going to set a block size of 4MB so we can represent larger | |
10983 | * FSes without overflowing. Additionally convert the space | |
10984 | * measurements from KB to bytes while making them in terms of | |
10985 | * blocks. We use 4MB only because it is big enough, and because it | |
10986 | * actually *is* the (ceph) default block size. | |
10987 | */ | |
10988 | const int CEPH_BLOCK_SHIFT = 22; | |
10989 | stbuf->f_frsize = 1 << CEPH_BLOCK_SHIFT; | |
10990 | stbuf->f_bsize = 1 << CEPH_BLOCK_SHIFT; | |
91327a77 | 10991 | stbuf->f_files = total_files_on_fs; |
f67539c2 | 10992 | stbuf->f_ffree = -1; |
7c673cae FG |
10993 | stbuf->f_favail = -1; |
10994 | stbuf->f_fsid = -1; // ?? | |
10995 | stbuf->f_flag = 0; // ?? | |
10996 | stbuf->f_namemax = NAME_MAX; | |
10997 | ||
10998 | // Usually quota_root will == root_ancestor, but if the mount root has no | |
10999 | // quota but we can see a parent of it that does have a quota, we'll | |
11000 | // respect that one instead. | |
11fdf7f2 | 11001 | ceph_assert(root != nullptr); |
b3b6e05e | 11002 | InodeRef quota_root = root->quota.is_enable() ? root : get_quota_root(root.get(), perms); |
7c673cae FG |
11003 | |
11004 | // get_quota_root should always give us something | |
11005 | // because client quotas are always enabled | |
11fdf7f2 | 11006 | ceph_assert(quota_root != nullptr); |
7c673cae FG |
11007 | |
11008 | if (quota_root && cct->_conf->client_quota_df && quota_root->quota.max_bytes) { | |
11009 | ||
11010 | // Skip the getattr if any sessions are stale, as we don't want to | |
11011 | // block `df` if this client has e.g. been evicted, or if the MDS cluster | |
11012 | // is unhealthy. | |
11013 | if (!_any_stale_sessions()) { | |
11014 | int r = _getattr(quota_root, 0, perms, true); | |
11015 | if (r != 0) { | |
11016 | // Ignore return value: error getting latest inode metadata is not a good | |
11017 | // reason to break "df". | |
11018 | lderr(cct) << "Error in getattr on quota root 0x" | |
11019 | << std::hex << quota_root->ino << std::dec | |
11020 | << " statfs result may be outdated" << dendl; | |
11021 | } | |
11022 | } | |
11023 | ||
11024 | // Special case: if there is a size quota set on the Inode acting | |
11025 | // as the root for this client mount, then report the quota status | |
11026 | // as the filesystem statistics. | |
11027 | const fsblkcnt_t total = quota_root->quota.max_bytes >> CEPH_BLOCK_SHIFT; | |
11028 | const fsblkcnt_t used = quota_root->rstat.rbytes >> CEPH_BLOCK_SHIFT; | |
31f18b77 FG |
11029 | // It is possible for a quota to be exceeded: arithmetic here must |
11030 | // handle case where used > total. | |
11031 | const fsblkcnt_t free = total > used ? total - used : 0; | |
7c673cae FG |
11032 | |
11033 | stbuf->f_blocks = total; | |
11034 | stbuf->f_bfree = free; | |
11035 | stbuf->f_bavail = free; | |
11036 | } else { | |
d2e6a577 | 11037 | // General case: report the cluster statistics returned from RADOS. Because |
7c673cae FG |
11038 | // multiple pools may be used without one filesystem namespace via |
11039 | // layouts, this is the most correct thing we can do. | |
11040 | stbuf->f_blocks = stats.kb >> (CEPH_BLOCK_SHIFT - 10); | |
11041 | stbuf->f_bfree = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
11042 | stbuf->f_bavail = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
11043 | } | |
11044 | ||
11045 | return rval; | |
11046 | } | |
11047 | ||
11048 | int Client::_do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep, | |
11049 | struct flock *fl, uint64_t owner, bool removing) | |
11050 | { | |
11fdf7f2 | 11051 | ldout(cct, 10) << __func__ << " ino " << in->ino |
7c673cae FG |
11052 | << (lock_type == CEPH_LOCK_FCNTL ? " fcntl" : " flock") |
11053 | << " type " << fl->l_type << " owner " << owner | |
11054 | << " " << fl->l_start << "~" << fl->l_len << dendl; | |
11055 | ||
f6b5b4d7 | 11056 | if (in->flags & I_ERROR_FILELOCK) |
f67539c2 | 11057 | return -CEPHFS_EIO; |
f6b5b4d7 | 11058 | |
7c673cae FG |
11059 | int lock_cmd; |
11060 | if (F_RDLCK == fl->l_type) | |
11061 | lock_cmd = CEPH_LOCK_SHARED; | |
11062 | else if (F_WRLCK == fl->l_type) | |
11063 | lock_cmd = CEPH_LOCK_EXCL; | |
11064 | else if (F_UNLCK == fl->l_type) | |
11065 | lock_cmd = CEPH_LOCK_UNLOCK; | |
11066 | else | |
f67539c2 | 11067 | return -CEPHFS_EIO; |
7c673cae FG |
11068 | |
11069 | if (op != CEPH_MDS_OP_SETFILELOCK || lock_cmd == CEPH_LOCK_UNLOCK) | |
11070 | sleep = 0; | |
11071 | ||
11072 | /* | |
11073 | * Set the most significant bit, so that MDS knows the 'owner' | |
11074 | * is sufficient to identify the owner of lock. (old code uses | |
11075 | * both 'owner' and 'pid') | |
11076 | */ | |
11077 | owner |= (1ULL << 63); | |
11078 | ||
11079 | MetaRequest *req = new MetaRequest(op); | |
11080 | filepath path; | |
11081 | in->make_nosnap_relative_path(path); | |
11082 | req->set_filepath(path); | |
11083 | req->set_inode(in); | |
11084 | ||
11085 | req->head.args.filelock_change.rule = lock_type; | |
11086 | req->head.args.filelock_change.type = lock_cmd; | |
11087 | req->head.args.filelock_change.owner = owner; | |
11088 | req->head.args.filelock_change.pid = fl->l_pid; | |
11089 | req->head.args.filelock_change.start = fl->l_start; | |
11090 | req->head.args.filelock_change.length = fl->l_len; | |
11091 | req->head.args.filelock_change.wait = sleep; | |
11092 | ||
11093 | int ret; | |
11094 | bufferlist bl; | |
11095 | ||
11096 | if (sleep && switch_interrupt_cb) { | |
11097 | // enable interrupt | |
11098 | switch_interrupt_cb(callback_handle, req->get()); | |
11099 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
7c673cae FG |
11100 | // disable interrupt |
11101 | switch_interrupt_cb(callback_handle, NULL); | |
31f18b77 FG |
11102 | if (ret == 0 && req->aborted()) { |
11103 | // effect of this lock request has been revoked by the 'lock intr' request | |
11104 | ret = req->get_abort_code(); | |
11105 | } | |
7c673cae FG |
11106 | put_request(req); |
11107 | } else { | |
11108 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
11109 | } | |
11110 | ||
11111 | if (ret == 0) { | |
11112 | if (op == CEPH_MDS_OP_GETFILELOCK) { | |
11113 | ceph_filelock filelock; | |
11fdf7f2 TL |
11114 | auto p = bl.cbegin(); |
11115 | decode(filelock, p); | |
7c673cae FG |
11116 | |
11117 | if (CEPH_LOCK_SHARED == filelock.type) | |
11118 | fl->l_type = F_RDLCK; | |
11119 | else if (CEPH_LOCK_EXCL == filelock.type) | |
11120 | fl->l_type = F_WRLCK; | |
11121 | else | |
11122 | fl->l_type = F_UNLCK; | |
11123 | ||
11124 | fl->l_whence = SEEK_SET; | |
11125 | fl->l_start = filelock.start; | |
11126 | fl->l_len = filelock.length; | |
11127 | fl->l_pid = filelock.pid; | |
11128 | } else if (op == CEPH_MDS_OP_SETFILELOCK) { | |
11129 | ceph_lock_state_t *lock_state; | |
11130 | if (lock_type == CEPH_LOCK_FCNTL) { | |
11131 | if (!in->fcntl_locks) | |
11fdf7f2 TL |
11132 | in->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
11133 | lock_state = in->fcntl_locks.get(); | |
7c673cae FG |
11134 | } else if (lock_type == CEPH_LOCK_FLOCK) { |
11135 | if (!in->flock_locks) | |
11fdf7f2 TL |
11136 | in->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
11137 | lock_state = in->flock_locks.get(); | |
7c673cae FG |
11138 | } else { |
11139 | ceph_abort(); | |
f67539c2 | 11140 | return -CEPHFS_EINVAL; |
7c673cae FG |
11141 | } |
11142 | _update_lock_state(fl, owner, lock_state); | |
11143 | ||
11144 | if (!removing) { | |
11145 | if (lock_type == CEPH_LOCK_FCNTL) { | |
11146 | if (!fh->fcntl_locks) | |
11fdf7f2 TL |
11147 | fh->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
11148 | lock_state = fh->fcntl_locks.get(); | |
7c673cae FG |
11149 | } else { |
11150 | if (!fh->flock_locks) | |
11fdf7f2 TL |
11151 | fh->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
11152 | lock_state = fh->flock_locks.get(); | |
7c673cae FG |
11153 | } |
11154 | _update_lock_state(fl, owner, lock_state); | |
11155 | } | |
11156 | } else | |
11157 | ceph_abort(); | |
11158 | } | |
11159 | return ret; | |
11160 | } | |
11161 | ||
11162 | int Client::_interrupt_filelock(MetaRequest *req) | |
11163 | { | |
31f18b77 FG |
11164 | // Set abort code, but do not kick. The abort code prevents the request |
11165 | // from being re-sent. | |
f67539c2 | 11166 | req->abort(-CEPHFS_EINTR); |
31f18b77 FG |
11167 | if (req->mds < 0) |
11168 | return 0; // haven't sent the request | |
11169 | ||
7c673cae FG |
11170 | Inode *in = req->inode(); |
11171 | ||
11172 | int lock_type; | |
11173 | if (req->head.args.filelock_change.rule == CEPH_LOCK_FLOCK) | |
11174 | lock_type = CEPH_LOCK_FLOCK_INTR; | |
11175 | else if (req->head.args.filelock_change.rule == CEPH_LOCK_FCNTL) | |
11176 | lock_type = CEPH_LOCK_FCNTL_INTR; | |
11177 | else { | |
11178 | ceph_abort(); | |
f67539c2 | 11179 | return -CEPHFS_EINVAL; |
7c673cae FG |
11180 | } |
11181 | ||
11182 | MetaRequest *intr_req = new MetaRequest(CEPH_MDS_OP_SETFILELOCK); | |
11183 | filepath path; | |
11184 | in->make_nosnap_relative_path(path); | |
11185 | intr_req->set_filepath(path); | |
11186 | intr_req->set_inode(in); | |
11187 | intr_req->head.args.filelock_change = req->head.args.filelock_change; | |
11188 | intr_req->head.args.filelock_change.rule = lock_type; | |
11189 | intr_req->head.args.filelock_change.type = CEPH_LOCK_UNLOCK; | |
11190 | ||
11191 | UserPerm perms(req->get_uid(), req->get_gid()); | |
11192 | return make_request(intr_req, perms, NULL, NULL, -1); | |
11193 | } | |
11194 | ||
11195 | void Client::_encode_filelocks(Inode *in, bufferlist& bl) | |
11196 | { | |
11197 | if (!in->fcntl_locks && !in->flock_locks) | |
11198 | return; | |
11199 | ||
11200 | unsigned nr_fcntl_locks = in->fcntl_locks ? in->fcntl_locks->held_locks.size() : 0; | |
11fdf7f2 | 11201 | encode(nr_fcntl_locks, bl); |
7c673cae | 11202 | if (nr_fcntl_locks) { |
11fdf7f2 | 11203 | auto &lock_state = in->fcntl_locks; |
20effc67 | 11204 | for(auto p = lock_state->held_locks.begin(); |
7c673cae FG |
11205 | p != lock_state->held_locks.end(); |
11206 | ++p) | |
11fdf7f2 | 11207 | encode(p->second, bl); |
7c673cae FG |
11208 | } |
11209 | ||
11210 | unsigned nr_flock_locks = in->flock_locks ? in->flock_locks->held_locks.size() : 0; | |
11fdf7f2 | 11211 | encode(nr_flock_locks, bl); |
7c673cae | 11212 | if (nr_flock_locks) { |
11fdf7f2 | 11213 | auto &lock_state = in->flock_locks; |
20effc67 | 11214 | for(auto p = lock_state->held_locks.begin(); |
7c673cae FG |
11215 | p != lock_state->held_locks.end(); |
11216 | ++p) | |
11fdf7f2 | 11217 | encode(p->second, bl); |
7c673cae FG |
11218 | } |
11219 | ||
11fdf7f2 | 11220 | ldout(cct, 10) << __func__ << " ino " << in->ino << ", " << nr_fcntl_locks |
7c673cae FG |
11221 | << " fcntl locks, " << nr_flock_locks << " flock locks" << dendl; |
11222 | } | |
11223 | ||
11224 | void Client::_release_filelocks(Fh *fh) | |
11225 | { | |
11226 | if (!fh->fcntl_locks && !fh->flock_locks) | |
11227 | return; | |
11228 | ||
11229 | Inode *in = fh->inode.get(); | |
11fdf7f2 | 11230 | ldout(cct, 10) << __func__ << " " << fh << " ino " << in->ino << dendl; |
7c673cae | 11231 | |
f6b5b4d7 TL |
11232 | list<ceph_filelock> activated_locks; |
11233 | ||
7c673cae FG |
11234 | list<pair<int, ceph_filelock> > to_release; |
11235 | ||
11236 | if (fh->fcntl_locks) { | |
11fdf7f2 | 11237 | auto &lock_state = fh->fcntl_locks; |
f6b5b4d7 TL |
11238 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11239 | auto q = p++; | |
11240 | if (in->flags & I_ERROR_FILELOCK) { | |
11241 | lock_state->remove_lock(q->second, activated_locks); | |
11242 | } else { | |
11243 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FCNTL, q->second)); | |
11244 | } | |
11245 | } | |
11fdf7f2 | 11246 | lock_state.reset(); |
7c673cae FG |
11247 | } |
11248 | if (fh->flock_locks) { | |
11fdf7f2 | 11249 | auto &lock_state = fh->flock_locks; |
f6b5b4d7 TL |
11250 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11251 | auto q = p++; | |
11252 | if (in->flags & I_ERROR_FILELOCK) { | |
11253 | lock_state->remove_lock(q->second, activated_locks); | |
11254 | } else { | |
11255 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FLOCK, q->second)); | |
11256 | } | |
11257 | } | |
11fdf7f2 | 11258 | lock_state.reset(); |
7c673cae FG |
11259 | } |
11260 | ||
f6b5b4d7 TL |
11261 | if ((in->flags & I_ERROR_FILELOCK) && !in->has_any_filelocks()) |
11262 | in->flags &= ~I_ERROR_FILELOCK; | |
7c673cae | 11263 | |
f6b5b4d7 | 11264 | if (to_release.empty()) |
11fdf7f2 TL |
11265 | return; |
11266 | ||
7c673cae FG |
11267 | struct flock fl; |
11268 | memset(&fl, 0, sizeof(fl)); | |
11269 | fl.l_whence = SEEK_SET; | |
11270 | fl.l_type = F_UNLCK; | |
11271 | ||
11272 | for (list<pair<int, ceph_filelock> >::iterator p = to_release.begin(); | |
11273 | p != to_release.end(); | |
11274 | ++p) { | |
11275 | fl.l_start = p->second.start; | |
11276 | fl.l_len = p->second.length; | |
11277 | fl.l_pid = p->second.pid; | |
11278 | _do_filelock(in, fh, p->first, CEPH_MDS_OP_SETFILELOCK, 0, &fl, | |
11279 | p->second.owner, true); | |
11280 | } | |
11281 | } | |
11282 | ||
11283 | void Client::_update_lock_state(struct flock *fl, uint64_t owner, | |
11284 | ceph_lock_state_t *lock_state) | |
11285 | { | |
11286 | int lock_cmd; | |
11287 | if (F_RDLCK == fl->l_type) | |
11288 | lock_cmd = CEPH_LOCK_SHARED; | |
11289 | else if (F_WRLCK == fl->l_type) | |
11290 | lock_cmd = CEPH_LOCK_EXCL; | |
11291 | else | |
11292 | lock_cmd = CEPH_LOCK_UNLOCK;; | |
11293 | ||
11294 | ceph_filelock filelock; | |
11295 | filelock.start = fl->l_start; | |
11296 | filelock.length = fl->l_len; | |
11297 | filelock.client = 0; | |
11298 | // see comment in _do_filelock() | |
11299 | filelock.owner = owner | (1ULL << 63); | |
11300 | filelock.pid = fl->l_pid; | |
11301 | filelock.type = lock_cmd; | |
11302 | ||
11303 | if (filelock.type == CEPH_LOCK_UNLOCK) { | |
11304 | list<ceph_filelock> activated_locks; | |
11305 | lock_state->remove_lock(filelock, activated_locks); | |
11306 | } else { | |
11307 | bool r = lock_state->add_lock(filelock, false, false, NULL); | |
11fdf7f2 | 11308 | ceph_assert(r); |
7c673cae FG |
11309 | } |
11310 | } | |
11311 | ||
11312 | int Client::_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
11313 | { | |
11314 | Inode *in = fh->inode.get(); | |
11315 | ldout(cct, 10) << "_getlk " << fh << " ino " << in->ino << dendl; | |
11316 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_GETFILELOCK, 0, fl, owner); | |
11317 | return ret; | |
11318 | } | |
11319 | ||
11320 | int Client::_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
11321 | { | |
11322 | Inode *in = fh->inode.get(); | |
11323 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << dendl; | |
11324 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_SETFILELOCK, sleep, fl, owner); | |
11325 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11326 | return ret; | |
11327 | } | |
11328 | ||
11329 | int Client::_flock(Fh *fh, int cmd, uint64_t owner) | |
11330 | { | |
11331 | Inode *in = fh->inode.get(); | |
11332 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << dendl; | |
11333 | ||
11334 | int sleep = !(cmd & LOCK_NB); | |
11335 | cmd &= ~LOCK_NB; | |
11336 | ||
11337 | int type; | |
11338 | switch (cmd) { | |
11339 | case LOCK_SH: | |
11340 | type = F_RDLCK; | |
11341 | break; | |
11342 | case LOCK_EX: | |
11343 | type = F_WRLCK; | |
11344 | break; | |
11345 | case LOCK_UN: | |
11346 | type = F_UNLCK; | |
11347 | break; | |
11348 | default: | |
f67539c2 | 11349 | return -CEPHFS_EINVAL; |
7c673cae FG |
11350 | } |
11351 | ||
11352 | struct flock fl; | |
11353 | memset(&fl, 0, sizeof(fl)); | |
11354 | fl.l_type = type; | |
11355 | fl.l_whence = SEEK_SET; | |
11356 | ||
11357 | int ret = _do_filelock(in, fh, CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK, sleep, &fl, owner); | |
11358 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11359 | return ret; | |
11360 | } | |
11361 | ||
f67539c2 TL |
11362 | int Client::get_snap_info(const char *path, const UserPerm &perms, SnapInfo *snap_info) { |
11363 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11364 | if (!mref_reader.is_state_satisfied()) { | |
11365 | return -CEPHFS_ENOTCONN; | |
11366 | } | |
11367 | ||
20effc67 | 11368 | std::scoped_lock lock(client_lock); |
f67539c2 TL |
11369 | InodeRef in; |
11370 | int r = Client::path_walk(path, &in, perms, true); | |
11371 | if (r < 0) { | |
11372 | return r; | |
11373 | } | |
11374 | ||
11375 | if (in->snapid == CEPH_NOSNAP) { | |
11376 | return -CEPHFS_EINVAL; | |
11377 | } | |
11378 | ||
11379 | snap_info->id = in->snapid; | |
11380 | snap_info->metadata = in->snap_metadata; | |
11381 | return 0; | |
11382 | } | |
11383 | ||
11384 | int Client::ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms) | |
11385 | { | |
11386 | /* Since the only thing this does is wrap a call to statfs, and | |
11387 | statfs takes a lock, it doesn't seem we have a need to split it | |
11388 | out. */ | |
7c673cae FG |
11389 | return statfs(0, stbuf, perms); |
11390 | } | |
11391 | ||
20effc67 | 11392 | void Client::_ll_register_callbacks(struct ceph_client_callback_args *args) |
7c673cae FG |
11393 | { |
11394 | if (!args) | |
11395 | return; | |
20effc67 | 11396 | |
11fdf7f2 | 11397 | ldout(cct, 10) << __func__ << " cb " << args->handle |
7c673cae FG |
11398 | << " invalidate_ino_cb " << args->ino_cb |
11399 | << " invalidate_dentry_cb " << args->dentry_cb | |
7c673cae FG |
11400 | << " switch_interrupt_cb " << args->switch_intr_cb |
11401 | << " remount_cb " << args->remount_cb | |
11402 | << dendl; | |
11403 | callback_handle = args->handle; | |
11404 | if (args->ino_cb) { | |
11405 | ino_invalidate_cb = args->ino_cb; | |
11406 | async_ino_invalidator.start(); | |
11407 | } | |
11408 | if (args->dentry_cb) { | |
11409 | dentry_invalidate_cb = args->dentry_cb; | |
11410 | async_dentry_invalidator.start(); | |
11411 | } | |
11412 | if (args->switch_intr_cb) { | |
11413 | switch_interrupt_cb = args->switch_intr_cb; | |
11414 | interrupt_finisher.start(); | |
11415 | } | |
11416 | if (args->remount_cb) { | |
11417 | remount_cb = args->remount_cb; | |
11418 | remount_finisher.start(); | |
11419 | } | |
e306af50 TL |
11420 | if (args->ino_release_cb) { |
11421 | ino_release_cb = args->ino_release_cb; | |
11422 | async_ino_releasor.start(); | |
11423 | } | |
11424 | if (args->umask_cb) | |
11425 | umask_cb = args->umask_cb; | |
7c673cae FG |
11426 | } |
11427 | ||
20effc67 TL |
11428 | // This is deprecated, use ll_register_callbacks2() instead. |
11429 | void Client::ll_register_callbacks(struct ceph_client_callback_args *args) | |
11430 | { | |
11431 | ceph_assert(!is_mounting() && !is_mounted() && !is_unmounting()); | |
11432 | ||
11433 | _ll_register_callbacks(args); | |
11434 | } | |
11435 | ||
11436 | int Client::ll_register_callbacks2(struct ceph_client_callback_args *args) | |
11437 | { | |
11438 | if (is_mounting() || is_mounted() || is_unmounting()) | |
11439 | return -CEPHFS_EBUSY; | |
11440 | ||
11441 | _ll_register_callbacks(args); | |
11442 | return 0; | |
11443 | } | |
11444 | ||
1d09f67e | 11445 | std::pair<int, bool> Client::test_dentry_handling(bool can_invalidate) |
7c673cae | 11446 | { |
1d09f67e | 11447 | std::pair <int, bool> r(0, false); |
7c673cae | 11448 | |
f67539c2 TL |
11449 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
11450 | if (!iref_reader.is_state_satisfied()) | |
1d09f67e | 11451 | return std::make_pair(-CEPHFS_ENOTCONN, false); |
f67539c2 | 11452 | |
7c673cae FG |
11453 | can_invalidate_dentries = can_invalidate; |
11454 | ||
11455 | if (can_invalidate_dentries) { | |
11fdf7f2 | 11456 | ceph_assert(dentry_invalidate_cb); |
7c673cae | 11457 | ldout(cct, 1) << "using dentry_invalidate_cb" << dendl; |
11fdf7f2 TL |
11458 | } else { |
11459 | ceph_assert(remount_cb); | |
7c673cae | 11460 | ldout(cct, 1) << "using remount_cb" << dendl; |
91327a77 | 11461 | r = _do_remount(false); |
b32b8144 | 11462 | } |
11fdf7f2 | 11463 | |
7c673cae FG |
11464 | return r; |
11465 | } | |
11466 | ||
11467 | int Client::_sync_fs() | |
11468 | { | |
f67539c2 TL |
11469 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11470 | ||
11fdf7f2 | 11471 | ldout(cct, 10) << __func__ << dendl; |
7c673cae FG |
11472 | |
11473 | // flush file data | |
11fdf7f2 TL |
11474 | std::unique_ptr<C_SaferCond> cond = nullptr; |
11475 | if (cct->_conf->client_oc) { | |
11476 | cond.reset(new C_SaferCond("Client::_sync_fs:lock")); | |
11477 | objectcacher->flush_all(cond.get()); | |
11478 | } | |
7c673cae FG |
11479 | |
11480 | // flush caps | |
11481 | flush_caps_sync(); | |
11482 | ceph_tid_t flush_tid = last_flush_tid; | |
11483 | ||
11484 | // wait for unsafe mds requests | |
11485 | wait_unsafe_requests(); | |
11486 | ||
11487 | wait_sync_caps(flush_tid); | |
11488 | ||
11fdf7f2 | 11489 | if (nullptr != cond) { |
9f95a23c | 11490 | client_lock.unlock(); |
11fdf7f2 TL |
11491 | ldout(cct, 15) << __func__ << " waiting on data to flush" << dendl; |
11492 | cond->wait(); | |
11493 | ldout(cct, 15) << __func__ << " flush finished" << dendl; | |
9f95a23c | 11494 | client_lock.lock(); |
7c673cae FG |
11495 | } |
11496 | ||
11497 | return 0; | |
11498 | } | |
11499 | ||
11500 | int Client::sync_fs() | |
11501 | { | |
f67539c2 TL |
11502 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11503 | if (!mref_reader.is_state_satisfied()) | |
11504 | return -CEPHFS_ENOTCONN; | |
181888fb | 11505 | |
f67539c2 | 11506 | std::scoped_lock l(client_lock); |
181888fb | 11507 | |
7c673cae FG |
11508 | return _sync_fs(); |
11509 | } | |
11510 | ||
11511 | int64_t Client::drop_caches() | |
11512 | { | |
f67539c2 | 11513 | std::scoped_lock l(client_lock); |
7c673cae FG |
11514 | return objectcacher->release_all(); |
11515 | } | |
11516 | ||
11fdf7f2 TL |
11517 | int Client::_lazyio(Fh *fh, int enable) |
11518 | { | |
11519 | Inode *in = fh->inode.get(); | |
11520 | ldout(cct, 20) << __func__ << " " << *in << " " << !!enable << dendl; | |
11521 | ||
11522 | if (!!(fh->mode & CEPH_FILE_MODE_LAZY) == !!enable) | |
11523 | return 0; | |
11524 | ||
11525 | int orig_mode = fh->mode; | |
11526 | if (enable) { | |
11527 | fh->mode |= CEPH_FILE_MODE_LAZY; | |
11528 | in->get_open_ref(fh->mode); | |
11529 | in->put_open_ref(orig_mode); | |
11530 | check_caps(in, CHECK_CAPS_NODELAY); | |
11531 | } else { | |
11532 | fh->mode &= ~CEPH_FILE_MODE_LAZY; | |
11533 | in->get_open_ref(fh->mode); | |
11534 | in->put_open_ref(orig_mode); | |
11535 | check_caps(in, 0); | |
11536 | } | |
11537 | ||
11538 | return 0; | |
11539 | } | |
11540 | ||
11541 | int Client::lazyio(int fd, int enable) | |
11542 | { | |
f67539c2 | 11543 | std::scoped_lock l(client_lock); |
11fdf7f2 TL |
11544 | Fh *f = get_filehandle(fd); |
11545 | if (!f) | |
f67539c2 | 11546 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
11547 | |
11548 | return _lazyio(f, enable); | |
11549 | } | |
11550 | ||
11551 | int Client::ll_lazyio(Fh *fh, int enable) | |
11552 | { | |
11fdf7f2 TL |
11553 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << !!enable << dendl; |
11554 | tout(cct) << __func__ << std::endl; | |
11555 | ||
f67539c2 | 11556 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
11557 | return _lazyio(fh, enable); |
11558 | } | |
7c673cae | 11559 | |
92f5a8d4 | 11560 | int Client::lazyio_propagate(int fd, loff_t offset, size_t count) |
7c673cae | 11561 | { |
f67539c2 | 11562 | std::scoped_lock l(client_lock); |
92f5a8d4 | 11563 | ldout(cct, 3) << "op: client->lazyio_propagate(" << fd |
7c673cae FG |
11564 | << ", " << offset << ", " << count << ")" << dendl; |
11565 | ||
11566 | Fh *f = get_filehandle(fd); | |
11567 | if (!f) | |
f67539c2 | 11568 | return -CEPHFS_EBADF; |
7c673cae FG |
11569 | |
11570 | // for now | |
11571 | _fsync(f, true); | |
11572 | ||
11573 | return 0; | |
11574 | } | |
11575 | ||
11576 | int Client::lazyio_synchronize(int fd, loff_t offset, size_t count) | |
11577 | { | |
f67539c2 | 11578 | std::scoped_lock l(client_lock); |
7c673cae FG |
11579 | ldout(cct, 3) << "op: client->lazyio_synchronize(" << fd |
11580 | << ", " << offset << ", " << count << ")" << dendl; | |
11581 | ||
11582 | Fh *f = get_filehandle(fd); | |
11583 | if (!f) | |
f67539c2 | 11584 | return -CEPHFS_EBADF; |
7c673cae FG |
11585 | Inode *in = f->inode.get(); |
11586 | ||
11587 | _fsync(f, true); | |
92f5a8d4 TL |
11588 | if (_release(in)) { |
11589 | int r =_getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
11590 | if (r < 0) | |
11591 | return r; | |
11592 | } | |
7c673cae FG |
11593 | return 0; |
11594 | } | |
11595 | ||
11596 | ||
11597 | // ============================= | |
11598 | // snaps | |
11599 | ||
f67539c2 TL |
11600 | int Client::mksnap(const char *relpath, const char *name, const UserPerm& perm, |
11601 | mode_t mode, const std::map<std::string, std::string> &metadata) | |
7c673cae | 11602 | { |
f67539c2 TL |
11603 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11604 | if (!mref_reader.is_state_satisfied()) | |
11605 | return -CEPHFS_ENOTCONN; | |
181888fb | 11606 | |
f67539c2 | 11607 | std::scoped_lock l(client_lock); |
181888fb | 11608 | |
7c673cae FG |
11609 | filepath path(relpath); |
11610 | InodeRef in; | |
11611 | int r = path_walk(path, &in, perm); | |
11612 | if (r < 0) | |
11613 | return r; | |
11614 | if (cct->_conf->client_permissions) { | |
11615 | r = may_create(in.get(), perm); | |
11616 | if (r < 0) | |
11617 | return r; | |
11618 | } | |
11619 | Inode *snapdir = open_snapdir(in.get()); | |
f67539c2 | 11620 | return _mkdir(snapdir, name, mode, perm, nullptr, metadata); |
7c673cae | 11621 | } |
181888fb | 11622 | |
f67539c2 | 11623 | int Client::rmsnap(const char *relpath, const char *name, const UserPerm& perms, bool check_perms) |
7c673cae | 11624 | { |
f67539c2 TL |
11625 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11626 | if (!mref_reader.is_state_satisfied()) | |
11627 | return -CEPHFS_ENOTCONN; | |
181888fb | 11628 | |
f67539c2 | 11629 | std::scoped_lock l(client_lock); |
181888fb | 11630 | |
7c673cae FG |
11631 | filepath path(relpath); |
11632 | InodeRef in; | |
11633 | int r = path_walk(path, &in, perms); | |
11634 | if (r < 0) | |
11635 | return r; | |
f67539c2 | 11636 | Inode *snapdir = open_snapdir(in.get()); |
7c673cae | 11637 | if (cct->_conf->client_permissions) { |
f67539c2 | 11638 | r = may_delete(snapdir, check_perms ? name : NULL, perms); |
7c673cae FG |
11639 | if (r < 0) |
11640 | return r; | |
11641 | } | |
7c673cae FG |
11642 | return _rmdir(snapdir, name, perms); |
11643 | } | |
11644 | ||
11645 | // ============================= | |
11646 | // expose caps | |
11647 | ||
f67539c2 TL |
11648 | int Client::get_caps_issued(int fd) |
11649 | { | |
11650 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11651 | if (!mref_reader.is_state_satisfied()) | |
11652 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11653 | |
f67539c2 | 11654 | std::scoped_lock lock(client_lock); |
181888fb | 11655 | |
7c673cae FG |
11656 | Fh *f = get_filehandle(fd); |
11657 | if (!f) | |
f67539c2 | 11658 | return -CEPHFS_EBADF; |
7c673cae FG |
11659 | |
11660 | return f->inode->caps_issued(); | |
11661 | } | |
11662 | ||
11663 | int Client::get_caps_issued(const char *path, const UserPerm& perms) | |
11664 | { | |
f67539c2 TL |
11665 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11666 | if (!mref_reader.is_state_satisfied()) | |
11667 | return -CEPHFS_ENOTCONN; | |
181888fb | 11668 | |
f67539c2 | 11669 | std::scoped_lock lock(client_lock); |
181888fb | 11670 | |
7c673cae FG |
11671 | filepath p(path); |
11672 | InodeRef in; | |
11673 | int r = path_walk(p, &in, perms, true); | |
11674 | if (r < 0) | |
11675 | return r; | |
11676 | return in->caps_issued(); | |
11677 | } | |
11678 | ||
11679 | // ========================================= | |
11680 | // low level | |
11681 | ||
11682 | Inode *Client::open_snapdir(Inode *diri) | |
11683 | { | |
11684 | Inode *in; | |
11685 | vinodeno_t vino(diri->ino, CEPH_SNAPDIR); | |
11686 | if (!inode_map.count(vino)) { | |
11687 | in = new Inode(this, vino, &diri->layout); | |
11688 | ||
11689 | in->ino = diri->ino; | |
11690 | in->snapid = CEPH_SNAPDIR; | |
11691 | in->mode = diri->mode; | |
11692 | in->uid = diri->uid; | |
11693 | in->gid = diri->gid; | |
494da23a | 11694 | in->nlink = 1; |
7c673cae FG |
11695 | in->mtime = diri->mtime; |
11696 | in->ctime = diri->ctime; | |
11697 | in->btime = diri->btime; | |
f6b5b4d7 | 11698 | in->atime = diri->atime; |
7c673cae FG |
11699 | in->size = diri->size; |
11700 | in->change_attr = diri->change_attr; | |
11701 | ||
11702 | in->dirfragtree.clear(); | |
11703 | in->snapdir_parent = diri; | |
11704 | diri->flags |= I_SNAPDIR_OPEN; | |
11705 | inode_map[vino] = in; | |
11706 | if (use_faked_inos()) | |
11707 | _assign_faked_ino(in); | |
11708 | ldout(cct, 10) << "open_snapdir created snapshot inode " << *in << dendl; | |
11709 | } else { | |
11710 | in = inode_map[vino]; | |
11711 | ldout(cct, 10) << "open_snapdir had snapshot inode " << *in << dendl; | |
11712 | } | |
11713 | return in; | |
11714 | } | |
11715 | ||
11716 | int Client::ll_lookup(Inode *parent, const char *name, struct stat *attr, | |
11717 | Inode **out, const UserPerm& perms) | |
11718 | { | |
f67539c2 TL |
11719 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11720 | if (!mref_reader.is_state_satisfied()) | |
11721 | return -CEPHFS_ENOTCONN; | |
11722 | ||
31f18b77 | 11723 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 TL |
11724 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
11725 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11726 | tout(cct) << name << std::endl; |
11727 | ||
f67539c2 | 11728 | std::scoped_lock lock(client_lock); |
181888fb | 11729 | |
7c673cae | 11730 | int r = 0; |
11fdf7f2 TL |
11731 | if (!fuse_default_permissions) { |
11732 | if (strcmp(name, ".") && strcmp(name, "..")) { | |
11733 | r = may_lookup(parent, perms); | |
11734 | if (r < 0) | |
11735 | return r; | |
11736 | } | |
7c673cae FG |
11737 | } |
11738 | ||
11739 | string dname(name); | |
11740 | InodeRef in; | |
11741 | ||
11742 | r = _lookup(parent, dname, CEPH_STAT_CAP_INODE_ALL, &in, perms); | |
11743 | if (r < 0) { | |
11744 | attr->st_ino = 0; | |
11745 | goto out; | |
11746 | } | |
11747 | ||
11fdf7f2 | 11748 | ceph_assert(in); |
7c673cae FG |
11749 | fill_stat(in, attr); |
11750 | _ll_get(in.get()); | |
11751 | ||
11752 | out: | |
11fdf7f2 | 11753 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11754 | << " -> " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; |
11755 | tout(cct) << attr->st_ino << std::endl; | |
11756 | *out = in.get(); | |
11757 | return r; | |
11758 | } | |
11759 | ||
f67539c2 TL |
11760 | int Client::ll_lookup_vino( |
11761 | vinodeno_t vino, | |
1adf2230 AA |
11762 | const UserPerm& perms, |
11763 | Inode **inode) | |
11764 | { | |
81eedcae | 11765 | ceph_assert(inode != NULL); |
f67539c2 TL |
11766 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11767 | if (!mref_reader.is_state_satisfied()) | |
11768 | return -CEPHFS_ENOTCONN; | |
81eedcae | 11769 | |
b3b6e05e TL |
11770 | if (is_reserved_vino(vino)) |
11771 | return -CEPHFS_ESTALE; | |
11772 | ||
f67539c2 TL |
11773 | std::scoped_lock lock(client_lock); |
11774 | ldout(cct, 3) << __func__ << " " << vino << dendl; | |
1adf2230 | 11775 | |
f67539c2 TL |
11776 | // Check the cache first |
11777 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
11778 | if (p != inode_map.end()) { | |
11779 | *inode = p->second; | |
11780 | _ll_get(*inode); | |
81eedcae TL |
11781 | return 0; |
11782 | } | |
11783 | ||
f67539c2 | 11784 | uint64_t snapid = vino.snapid; |
81eedcae | 11785 | |
f67539c2 TL |
11786 | // for snapdir, find the non-snapped dir inode |
11787 | if (snapid == CEPH_SNAPDIR) | |
11788 | vino.snapid = CEPH_NOSNAP; | |
11789 | ||
11790 | int r = _lookup_vino(vino, perms, inode); | |
11791 | if (r) | |
1adf2230 | 11792 | return r; |
f67539c2 | 11793 | ceph_assert(*inode != NULL); |
81eedcae | 11794 | |
f67539c2 TL |
11795 | if (snapid == CEPH_SNAPDIR) { |
11796 | Inode *tmp = *inode; | |
1adf2230 | 11797 | |
f67539c2 TL |
11798 | // open the snapdir and put the inode ref |
11799 | *inode = open_snapdir(tmp); | |
11800 | _ll_forget(tmp, 1); | |
11801 | _ll_get(*inode); | |
1adf2230 | 11802 | } |
1adf2230 AA |
11803 | return 0; |
11804 | } | |
11805 | ||
f67539c2 TL |
11806 | int Client::ll_lookup_inode( |
11807 | struct inodeno_t ino, | |
11808 | const UserPerm& perms, | |
11809 | Inode **inode) | |
11810 | { | |
11811 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
11812 | return ll_lookup_vino(vino, perms, inode); | |
11813 | } | |
11814 | ||
7c673cae FG |
11815 | int Client::ll_lookupx(Inode *parent, const char *name, Inode **out, |
11816 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
11817 | const UserPerm& perms) | |
11818 | { | |
f67539c2 TL |
11819 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11820 | if (!mref_reader.is_state_satisfied()) | |
11821 | return -CEPHFS_ENOTCONN; | |
11822 | ||
31f18b77 | 11823 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 | 11824 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
7c673cae FG |
11825 | tout(cct) << "ll_lookupx" << std::endl; |
11826 | tout(cct) << name << std::endl; | |
11827 | ||
f67539c2 | 11828 | std::scoped_lock lock(client_lock); |
181888fb | 11829 | |
7c673cae | 11830 | int r = 0; |
11fdf7f2 | 11831 | if (!fuse_default_permissions) { |
7c673cae FG |
11832 | r = may_lookup(parent, perms); |
11833 | if (r < 0) | |
11834 | return r; | |
11835 | } | |
11836 | ||
11837 | string dname(name); | |
11838 | InodeRef in; | |
11839 | ||
11840 | unsigned mask = statx_to_mask(flags, want); | |
11841 | r = _lookup(parent, dname, mask, &in, perms); | |
11842 | if (r < 0) { | |
11843 | stx->stx_ino = 0; | |
11844 | stx->stx_mask = 0; | |
11845 | } else { | |
11fdf7f2 | 11846 | ceph_assert(in); |
7c673cae FG |
11847 | fill_statx(in, mask, stx); |
11848 | _ll_get(in.get()); | |
11849 | } | |
11850 | ||
11fdf7f2 | 11851 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11852 | << " -> " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; |
11853 | tout(cct) << stx->stx_ino << std::endl; | |
11854 | *out = in.get(); | |
11855 | return r; | |
11856 | } | |
11857 | ||
11858 | int Client::ll_walk(const char* name, Inode **out, struct ceph_statx *stx, | |
11859 | unsigned int want, unsigned int flags, const UserPerm& perms) | |
11860 | { | |
f67539c2 TL |
11861 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11862 | if (!mref_reader.is_state_satisfied()) | |
11863 | return -CEPHFS_ENOTCONN; | |
181888fb | 11864 | |
7c673cae FG |
11865 | filepath fp(name, 0); |
11866 | InodeRef in; | |
11867 | int rc; | |
11868 | unsigned mask = statx_to_mask(flags, want); | |
11869 | ||
11fdf7f2 TL |
11870 | ldout(cct, 3) << __func__ << " " << name << dendl; |
11871 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11872 | tout(cct) << name << std::endl; |
11873 | ||
f67539c2 | 11874 | std::scoped_lock lock(client_lock); |
7c673cae FG |
11875 | rc = path_walk(fp, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); |
11876 | if (rc < 0) { | |
11877 | /* zero out mask, just in case... */ | |
11878 | stx->stx_mask = 0; | |
11879 | stx->stx_ino = 0; | |
11880 | *out = NULL; | |
11881 | return rc; | |
11882 | } else { | |
11fdf7f2 | 11883 | ceph_assert(in); |
7c673cae FG |
11884 | fill_statx(in, mask, stx); |
11885 | _ll_get(in.get()); | |
11886 | *out = in.get(); | |
11887 | return 0; | |
11888 | } | |
11889 | } | |
11890 | ||
11891 | void Client::_ll_get(Inode *in) | |
11892 | { | |
11893 | if (in->ll_ref == 0) { | |
b3b6e05e | 11894 | in->iget(); |
11fdf7f2 TL |
11895 | if (in->is_dir() && !in->dentries.empty()) { |
11896 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
11897 | in->get_first_parent()->get(); // pin dentry |
11898 | } | |
11fdf7f2 TL |
11899 | if (in->snapid != CEPH_NOSNAP) |
11900 | ll_snap_ref[in->snapid]++; | |
7c673cae FG |
11901 | } |
11902 | in->ll_get(); | |
11fdf7f2 | 11903 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " -> " << in->ll_ref << dendl; |
7c673cae FG |
11904 | } |
11905 | ||
494da23a | 11906 | int Client::_ll_put(Inode *in, uint64_t num) |
7c673cae FG |
11907 | { |
11908 | in->ll_put(num); | |
11fdf7f2 | 11909 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " " << num << " -> " << in->ll_ref << dendl; |
7c673cae | 11910 | if (in->ll_ref == 0) { |
11fdf7f2 TL |
11911 | if (in->is_dir() && !in->dentries.empty()) { |
11912 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
11913 | in->get_first_parent()->put(); // unpin dentry |
11914 | } | |
11fdf7f2 TL |
11915 | if (in->snapid != CEPH_NOSNAP) { |
11916 | auto p = ll_snap_ref.find(in->snapid); | |
11917 | ceph_assert(p != ll_snap_ref.end()); | |
11918 | ceph_assert(p->second > 0); | |
11919 | if (--p->second == 0) | |
11920 | ll_snap_ref.erase(p); | |
11921 | } | |
7c673cae FG |
11922 | put_inode(in); |
11923 | return 0; | |
11924 | } else { | |
11925 | return in->ll_ref; | |
11926 | } | |
11927 | } | |
11928 | ||
11929 | void Client::_ll_drop_pins() | |
11930 | { | |
11fdf7f2 | 11931 | ldout(cct, 10) << __func__ << dendl; |
1adf2230 | 11932 | std::set<InodeRef> to_be_put; //this set will be deconstructed item by item when exit |
7c673cae FG |
11933 | ceph::unordered_map<vinodeno_t, Inode*>::iterator next; |
11934 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
11935 | it != inode_map.end(); | |
11936 | it = next) { | |
11937 | Inode *in = it->second; | |
11938 | next = it; | |
11939 | ++next; | |
1adf2230 AA |
11940 | if (in->ll_ref){ |
11941 | to_be_put.insert(in); | |
7c673cae | 11942 | _ll_put(in, in->ll_ref); |
1adf2230 | 11943 | } |
7c673cae FG |
11944 | } |
11945 | } | |
11946 | ||
494da23a | 11947 | bool Client::_ll_forget(Inode *in, uint64_t count) |
7c673cae | 11948 | { |
11fdf7f2 | 11949 | inodeno_t ino = in->ino; |
7c673cae | 11950 | |
11fdf7f2 TL |
11951 | ldout(cct, 8) << __func__ << " " << ino << " " << count << dendl; |
11952 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11953 | tout(cct) << ino.val << std::endl; |
11954 | tout(cct) << count << std::endl; | |
11955 | ||
181888fb | 11956 | // Ignore forget if we're no longer mounted |
f67539c2 TL |
11957 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11958 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
11959 | return true; |
11960 | ||
7c673cae FG |
11961 | if (ino == 1) return true; // ignore forget on root. |
11962 | ||
11963 | bool last = false; | |
11964 | if (in->ll_ref < count) { | |
11965 | ldout(cct, 1) << "WARNING: ll_forget on " << ino << " " << count | |
11966 | << ", which only has ll_ref=" << in->ll_ref << dendl; | |
11967 | _ll_put(in, in->ll_ref); | |
11968 | last = true; | |
11969 | } else { | |
11970 | if (_ll_put(in, count) == 0) | |
11971 | last = true; | |
11972 | } | |
11973 | ||
11974 | return last; | |
11975 | } | |
11976 | ||
494da23a | 11977 | bool Client::ll_forget(Inode *in, uint64_t count) |
1adf2230 | 11978 | { |
f67539c2 | 11979 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
11980 | return _ll_forget(in, count); |
11981 | } | |
11982 | ||
7c673cae FG |
11983 | bool Client::ll_put(Inode *in) |
11984 | { | |
11985 | /* ll_forget already takes the lock */ | |
11986 | return ll_forget(in, 1); | |
11987 | } | |
11988 | ||
11fdf7f2 TL |
11989 | int Client::ll_get_snap_ref(snapid_t snap) |
11990 | { | |
f67539c2 | 11991 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
11992 | auto p = ll_snap_ref.find(snap); |
11993 | if (p != ll_snap_ref.end()) | |
11994 | return p->second; | |
11995 | return 0; | |
11996 | } | |
11997 | ||
7c673cae FG |
11998 | snapid_t Client::ll_get_snapid(Inode *in) |
11999 | { | |
f67539c2 | 12000 | std::scoped_lock lock(client_lock); |
7c673cae FG |
12001 | return in->snapid; |
12002 | } | |
12003 | ||
12004 | Inode *Client::ll_get_inode(ino_t ino) | |
12005 | { | |
f67539c2 TL |
12006 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12007 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
12008 | return NULL; |
12009 | ||
f67539c2 TL |
12010 | std::scoped_lock lock(client_lock); |
12011 | ||
7c673cae FG |
12012 | vinodeno_t vino = _map_faked_ino(ino); |
12013 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
12014 | if (p == inode_map.end()) | |
12015 | return NULL; | |
12016 | Inode *in = p->second; | |
12017 | _ll_get(in); | |
12018 | return in; | |
12019 | } | |
12020 | ||
12021 | Inode *Client::ll_get_inode(vinodeno_t vino) | |
12022 | { | |
f67539c2 TL |
12023 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12024 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
12025 | return NULL; |
12026 | ||
b3b6e05e TL |
12027 | if (is_reserved_vino(vino)) |
12028 | return NULL; | |
12029 | ||
f67539c2 TL |
12030 | std::scoped_lock lock(client_lock); |
12031 | ||
7c673cae FG |
12032 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
12033 | if (p == inode_map.end()) | |
12034 | return NULL; | |
12035 | Inode *in = p->second; | |
12036 | _ll_get(in); | |
12037 | return in; | |
12038 | } | |
12039 | ||
12040 | int Client::_ll_getattr(Inode *in, int caps, const UserPerm& perms) | |
12041 | { | |
12042 | vinodeno_t vino = _get_vino(in); | |
12043 | ||
11fdf7f2 TL |
12044 | ldout(cct, 8) << __func__ << " " << vino << dendl; |
12045 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12046 | tout(cct) << vino.ino.val << std::endl; |
12047 | ||
12048 | if (vino.snapid < CEPH_NOSNAP) | |
12049 | return 0; | |
12050 | else | |
12051 | return _getattr(in, caps, perms); | |
12052 | } | |
12053 | ||
12054 | int Client::ll_getattr(Inode *in, struct stat *attr, const UserPerm& perms) | |
12055 | { | |
f67539c2 TL |
12056 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12057 | if (!mref_reader.is_state_satisfied()) | |
12058 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12059 | |
f67539c2 | 12060 | std::scoped_lock lock(client_lock); |
181888fb | 12061 | |
7c673cae FG |
12062 | int res = _ll_getattr(in, CEPH_STAT_CAP_INODE_ALL, perms); |
12063 | ||
12064 | if (res == 0) | |
12065 | fill_stat(in, attr); | |
11fdf7f2 | 12066 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12067 | return res; |
12068 | } | |
12069 | ||
12070 | int Client::ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want, | |
12071 | unsigned int flags, const UserPerm& perms) | |
12072 | { | |
f67539c2 TL |
12073 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12074 | if (!mref_reader.is_state_satisfied()) | |
12075 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12076 | |
f67539c2 | 12077 | std::scoped_lock lock(client_lock); |
181888fb | 12078 | |
7c673cae FG |
12079 | int res = 0; |
12080 | unsigned mask = statx_to_mask(flags, want); | |
12081 | ||
94b18763 | 12082 | if (mask && !in->caps_issued_mask(mask, true)) |
7c673cae FG |
12083 | res = _ll_getattr(in, mask, perms); |
12084 | ||
12085 | if (res == 0) | |
12086 | fill_statx(in, mask, stx); | |
11fdf7f2 | 12087 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12088 | return res; |
12089 | } | |
12090 | ||
12091 | int Client::_ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
12092 | const UserPerm& perms, InodeRef *inp) | |
12093 | { | |
12094 | vinodeno_t vino = _get_vino(in); | |
12095 | ||
11fdf7f2 | 12096 | ldout(cct, 8) << __func__ << " " << vino << " mask " << hex << mask << dec |
7c673cae | 12097 | << dendl; |
11fdf7f2 | 12098 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
12099 | tout(cct) << vino.ino.val << std::endl; |
12100 | tout(cct) << stx->stx_mode << std::endl; | |
12101 | tout(cct) << stx->stx_uid << std::endl; | |
12102 | tout(cct) << stx->stx_gid << std::endl; | |
12103 | tout(cct) << stx->stx_size << std::endl; | |
12104 | tout(cct) << stx->stx_mtime << std::endl; | |
12105 | tout(cct) << stx->stx_atime << std::endl; | |
12106 | tout(cct) << stx->stx_btime << std::endl; | |
12107 | tout(cct) << mask << std::endl; | |
12108 | ||
11fdf7f2 | 12109 | if (!fuse_default_permissions) { |
7c673cae FG |
12110 | int res = may_setattr(in, stx, mask, perms); |
12111 | if (res < 0) | |
12112 | return res; | |
12113 | } | |
12114 | ||
12115 | mask &= ~(CEPH_SETATTR_MTIME_NOW | CEPH_SETATTR_ATIME_NOW); | |
12116 | ||
12117 | return __setattrx(in, stx, mask, perms, inp); | |
12118 | } | |
12119 | ||
12120 | int Client::ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
12121 | const UserPerm& perms) | |
12122 | { | |
f67539c2 TL |
12123 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12124 | if (!mref_reader.is_state_satisfied()) | |
12125 | return -CEPHFS_ENOTCONN; | |
181888fb | 12126 | |
f67539c2 | 12127 | std::scoped_lock lock(client_lock); |
181888fb | 12128 | |
7c673cae FG |
12129 | InodeRef target(in); |
12130 | int res = _ll_setattrx(in, stx, mask, perms, &target); | |
12131 | if (res == 0) { | |
11fdf7f2 | 12132 | ceph_assert(in == target.get()); |
7c673cae FG |
12133 | fill_statx(in, in->caps_issued(), stx); |
12134 | } | |
12135 | ||
11fdf7f2 | 12136 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12137 | return res; |
12138 | } | |
12139 | ||
12140 | int Client::ll_setattr(Inode *in, struct stat *attr, int mask, | |
12141 | const UserPerm& perms) | |
12142 | { | |
12143 | struct ceph_statx stx; | |
12144 | stat_to_statx(attr, &stx); | |
12145 | ||
f67539c2 TL |
12146 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12147 | if (!mref_reader.is_state_satisfied()) | |
12148 | return -CEPHFS_ENOTCONN; | |
181888fb | 12149 | |
f67539c2 | 12150 | std::scoped_lock lock(client_lock); |
181888fb | 12151 | |
7c673cae FG |
12152 | InodeRef target(in); |
12153 | int res = _ll_setattrx(in, &stx, mask, perms, &target); | |
12154 | if (res == 0) { | |
11fdf7f2 | 12155 | ceph_assert(in == target.get()); |
7c673cae FG |
12156 | fill_stat(in, attr); |
12157 | } | |
12158 | ||
11fdf7f2 | 12159 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12160 | return res; |
12161 | } | |
12162 | ||
12163 | ||
12164 | // ---------- | |
12165 | // xattrs | |
12166 | ||
12167 | int Client::getxattr(const char *path, const char *name, void *value, size_t size, | |
12168 | const UserPerm& perms) | |
12169 | { | |
f67539c2 TL |
12170 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12171 | if (!mref_reader.is_state_satisfied()) | |
12172 | return -CEPHFS_ENOTCONN; | |
181888fb | 12173 | |
f67539c2 | 12174 | std::scoped_lock lock(client_lock); |
181888fb | 12175 | |
7c673cae FG |
12176 | InodeRef in; |
12177 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12178 | if (r < 0) | |
12179 | return r; | |
12180 | return _getxattr(in, name, value, size, perms); | |
12181 | } | |
12182 | ||
12183 | int Client::lgetxattr(const char *path, const char *name, void *value, size_t size, | |
12184 | const UserPerm& perms) | |
12185 | { | |
f67539c2 TL |
12186 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12187 | if (!mref_reader.is_state_satisfied()) | |
12188 | return -CEPHFS_ENOTCONN; | |
181888fb | 12189 | |
f67539c2 | 12190 | std::scoped_lock lock(client_lock); |
181888fb | 12191 | |
7c673cae FG |
12192 | InodeRef in; |
12193 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12194 | if (r < 0) | |
12195 | return r; | |
12196 | return _getxattr(in, name, value, size, perms); | |
12197 | } | |
12198 | ||
12199 | int Client::fgetxattr(int fd, const char *name, void *value, size_t size, | |
12200 | const UserPerm& perms) | |
12201 | { | |
f67539c2 TL |
12202 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12203 | if (!mref_reader.is_state_satisfied()) | |
12204 | return -CEPHFS_ENOTCONN; | |
181888fb | 12205 | |
f67539c2 | 12206 | std::scoped_lock lock(client_lock); |
181888fb | 12207 | |
7c673cae FG |
12208 | Fh *f = get_filehandle(fd); |
12209 | if (!f) | |
f67539c2 | 12210 | return -CEPHFS_EBADF; |
7c673cae FG |
12211 | return _getxattr(f->inode, name, value, size, perms); |
12212 | } | |
12213 | ||
12214 | int Client::listxattr(const char *path, char *list, size_t size, | |
12215 | const UserPerm& perms) | |
12216 | { | |
f67539c2 TL |
12217 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12218 | if (!mref_reader.is_state_satisfied()) | |
12219 | return -CEPHFS_ENOTCONN; | |
181888fb | 12220 | |
f67539c2 | 12221 | std::scoped_lock lock(client_lock); |
181888fb | 12222 | |
7c673cae FG |
12223 | InodeRef in; |
12224 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12225 | if (r < 0) | |
12226 | return r; | |
12227 | return Client::_listxattr(in.get(), list, size, perms); | |
12228 | } | |
12229 | ||
12230 | int Client::llistxattr(const char *path, char *list, size_t size, | |
12231 | const UserPerm& perms) | |
12232 | { | |
f67539c2 TL |
12233 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12234 | if (!mref_reader.is_state_satisfied()) | |
12235 | return -CEPHFS_ENOTCONN; | |
181888fb | 12236 | |
f67539c2 | 12237 | std::scoped_lock lock(client_lock); |
181888fb | 12238 | |
7c673cae FG |
12239 | InodeRef in; |
12240 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12241 | if (r < 0) | |
12242 | return r; | |
12243 | return Client::_listxattr(in.get(), list, size, perms); | |
12244 | } | |
12245 | ||
12246 | int Client::flistxattr(int fd, char *list, size_t size, const UserPerm& perms) | |
12247 | { | |
f67539c2 TL |
12248 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12249 | if (!mref_reader.is_state_satisfied()) | |
12250 | return -CEPHFS_ENOTCONN; | |
181888fb | 12251 | |
f67539c2 | 12252 | std::scoped_lock lock(client_lock); |
181888fb | 12253 | |
7c673cae FG |
12254 | Fh *f = get_filehandle(fd); |
12255 | if (!f) | |
f67539c2 | 12256 | return -CEPHFS_EBADF; |
7c673cae FG |
12257 | return Client::_listxattr(f->inode.get(), list, size, perms); |
12258 | } | |
12259 | ||
12260 | int Client::removexattr(const char *path, const char *name, | |
12261 | const UserPerm& perms) | |
12262 | { | |
f67539c2 TL |
12263 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12264 | if (!mref_reader.is_state_satisfied()) | |
12265 | return -CEPHFS_ENOTCONN; | |
181888fb | 12266 | |
f67539c2 | 12267 | std::scoped_lock lock(client_lock); |
181888fb | 12268 | |
7c673cae FG |
12269 | InodeRef in; |
12270 | int r = Client::path_walk(path, &in, perms, true); | |
12271 | if (r < 0) | |
12272 | return r; | |
12273 | return _removexattr(in, name, perms); | |
12274 | } | |
12275 | ||
12276 | int Client::lremovexattr(const char *path, const char *name, | |
12277 | const UserPerm& perms) | |
12278 | { | |
f67539c2 TL |
12279 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12280 | if (!mref_reader.is_state_satisfied()) | |
12281 | return -CEPHFS_ENOTCONN; | |
181888fb | 12282 | |
f67539c2 | 12283 | std::scoped_lock lock(client_lock); |
181888fb | 12284 | |
7c673cae FG |
12285 | InodeRef in; |
12286 | int r = Client::path_walk(path, &in, perms, false); | |
12287 | if (r < 0) | |
12288 | return r; | |
12289 | return _removexattr(in, name, perms); | |
12290 | } | |
12291 | ||
12292 | int Client::fremovexattr(int fd, const char *name, const UserPerm& perms) | |
12293 | { | |
f67539c2 TL |
12294 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12295 | if (!mref_reader.is_state_satisfied()) | |
12296 | return -CEPHFS_ENOTCONN; | |
181888fb | 12297 | |
f67539c2 | 12298 | std::scoped_lock lock(client_lock); |
181888fb | 12299 | |
7c673cae FG |
12300 | Fh *f = get_filehandle(fd); |
12301 | if (!f) | |
f67539c2 | 12302 | return -CEPHFS_EBADF; |
7c673cae FG |
12303 | return _removexattr(f->inode, name, perms); |
12304 | } | |
12305 | ||
12306 | int Client::setxattr(const char *path, const char *name, const void *value, | |
12307 | size_t size, int flags, const UserPerm& perms) | |
12308 | { | |
f67539c2 TL |
12309 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12310 | if (!mref_reader.is_state_satisfied()) | |
12311 | return -CEPHFS_ENOTCONN; | |
12312 | ||
7c673cae FG |
12313 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
12314 | ||
f67539c2 | 12315 | std::scoped_lock lock(client_lock); |
181888fb | 12316 | |
7c673cae FG |
12317 | InodeRef in; |
12318 | int r = Client::path_walk(path, &in, perms, true); | |
12319 | if (r < 0) | |
12320 | return r; | |
12321 | return _setxattr(in, name, value, size, flags, perms); | |
12322 | } | |
12323 | ||
12324 | int Client::lsetxattr(const char *path, const char *name, const void *value, | |
12325 | size_t size, int flags, const UserPerm& perms) | |
12326 | { | |
f67539c2 TL |
12327 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12328 | if (!mref_reader.is_state_satisfied()) | |
12329 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12330 | |
f67539c2 | 12331 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12332 | |
f67539c2 | 12333 | std::scoped_lock lock(client_lock); |
181888fb | 12334 | |
7c673cae FG |
12335 | InodeRef in; |
12336 | int r = Client::path_walk(path, &in, perms, false); | |
12337 | if (r < 0) | |
12338 | return r; | |
12339 | return _setxattr(in, name, value, size, flags, perms); | |
12340 | } | |
12341 | ||
12342 | int Client::fsetxattr(int fd, const char *name, const void *value, size_t size, | |
12343 | int flags, const UserPerm& perms) | |
12344 | { | |
f67539c2 TL |
12345 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12346 | if (!mref_reader.is_state_satisfied()) | |
12347 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12348 | |
f67539c2 | 12349 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12350 | |
f67539c2 | 12351 | std::scoped_lock lock(client_lock); |
181888fb | 12352 | |
7c673cae FG |
12353 | Fh *f = get_filehandle(fd); |
12354 | if (!f) | |
f67539c2 | 12355 | return -CEPHFS_EBADF; |
7c673cae FG |
12356 | return _setxattr(f->inode, name, value, size, flags, perms); |
12357 | } | |
12358 | ||
12359 | int Client::_getxattr(Inode *in, const char *name, void *value, size_t size, | |
12360 | const UserPerm& perms) | |
12361 | { | |
12362 | int r; | |
1d09f67e | 12363 | const VXattr *vxattr = nullptr; |
7c673cae | 12364 | |
1d09f67e | 12365 | vxattr = _match_vxattr(in, name); |
7c673cae | 12366 | if (vxattr) { |
f67539c2 | 12367 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12368 | |
12369 | // Do a force getattr to get the latest quota before returning | |
12370 | // a value to userspace. | |
28e407b8 AA |
12371 | int flags = 0; |
12372 | if (vxattr->flags & VXATTR_RSTAT) { | |
12373 | flags |= CEPH_STAT_RSTAT; | |
12374 | } | |
adb31ebb TL |
12375 | if (vxattr->flags & VXATTR_DIRSTAT) { |
12376 | flags |= CEPH_CAP_FILE_SHARED; | |
12377 | } | |
f67539c2 | 12378 | r = _getattr(in, flags | CEPH_STAT_CAP_XATTR, perms, true); |
7c673cae FG |
12379 | if (r != 0) { |
12380 | // Error from getattr! | |
12381 | return r; | |
12382 | } | |
12383 | ||
12384 | // call pointer-to-member function | |
12385 | char buf[256]; | |
12386 | if (!(vxattr->exists_cb && !(this->*(vxattr->exists_cb))(in))) { | |
12387 | r = (this->*(vxattr->getxattr_cb))(in, buf, sizeof(buf)); | |
12388 | } else { | |
f67539c2 | 12389 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12390 | } |
12391 | ||
12392 | if (size != 0) { | |
12393 | if (r > (int)size) { | |
f67539c2 | 12394 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12395 | } else if (r > 0) { |
12396 | memcpy(value, buf, r); | |
12397 | } | |
12398 | } | |
12399 | goto out; | |
12400 | } | |
12401 | ||
1d09f67e TL |
12402 | if (!strncmp(name, "ceph.", 5)) { |
12403 | r = _getvxattr(in, perms, name, size, value, MDS_RANK_NONE); | |
12404 | goto out; | |
12405 | } | |
12406 | ||
7c673cae | 12407 | if (acl_type == NO_ACL && !strncmp(name, "system.", 7)) { |
f67539c2 | 12408 | r = -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12409 | goto out; |
12410 | } | |
12411 | ||
12412 | r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
12413 | if (r == 0) { | |
12414 | string n(name); | |
f67539c2 | 12415 | r = -CEPHFS_ENODATA; |
1d09f67e | 12416 | if (in->xattrs.count(n)) { |
7c673cae FG |
12417 | r = in->xattrs[n].length(); |
12418 | if (r > 0 && size != 0) { | |
12419 | if (size >= (unsigned)r) | |
12420 | memcpy(value, in->xattrs[n].c_str(), r); | |
12421 | else | |
f67539c2 | 12422 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12423 | } |
12424 | } | |
12425 | } | |
12426 | out: | |
1adf2230 | 12427 | ldout(cct, 8) << "_getxattr(" << in->ino << ", \"" << name << "\", " << size << ") = " << r << dendl; |
7c673cae FG |
12428 | return r; |
12429 | } | |
12430 | ||
12431 | int Client::_getxattr(InodeRef &in, const char *name, void *value, size_t size, | |
12432 | const UserPerm& perms) | |
12433 | { | |
12434 | if (cct->_conf->client_permissions) { | |
12435 | int r = xattr_permission(in.get(), name, MAY_READ, perms); | |
12436 | if (r < 0) | |
12437 | return r; | |
12438 | } | |
12439 | return _getxattr(in.get(), name, value, size, perms); | |
12440 | } | |
12441 | ||
12442 | int Client::ll_getxattr(Inode *in, const char *name, void *value, | |
12443 | size_t size, const UserPerm& perms) | |
12444 | { | |
f67539c2 TL |
12445 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12446 | if (!mref_reader.is_state_satisfied()) | |
12447 | return -CEPHFS_ENOTCONN; | |
181888fb | 12448 | |
7c673cae FG |
12449 | vinodeno_t vino = _get_vino(in); |
12450 | ||
11fdf7f2 TL |
12451 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12452 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12453 | tout(cct) << vino.ino.val << std::endl; |
12454 | tout(cct) << name << std::endl; | |
12455 | ||
f67539c2 | 12456 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12457 | if (!fuse_default_permissions) { |
7c673cae FG |
12458 | int r = xattr_permission(in, name, MAY_READ, perms); |
12459 | if (r < 0) | |
12460 | return r; | |
12461 | } | |
12462 | ||
12463 | return _getxattr(in, name, value, size, perms); | |
12464 | } | |
12465 | ||
12466 | int Client::_listxattr(Inode *in, char *name, size_t size, | |
12467 | const UserPerm& perms) | |
12468 | { | |
81eedcae | 12469 | bool len_only = (size == 0); |
7c673cae | 12470 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); |
81eedcae TL |
12471 | if (r != 0) { |
12472 | goto out; | |
12473 | } | |
7c673cae | 12474 | |
81eedcae | 12475 | r = 0; |
f67539c2 TL |
12476 | for ([[maybe_unused]] const auto &[xattr_name, xattr_value_bl] : in->xattrs) { |
12477 | if (xattr_name.rfind("ceph.", 0) == 0) { | |
12478 | continue; | |
12479 | } | |
12480 | ||
12481 | size_t this_len = xattr_name.length() + 1; | |
81eedcae TL |
12482 | r += this_len; |
12483 | if (len_only) | |
12484 | continue; | |
7c673cae | 12485 | |
81eedcae | 12486 | if (this_len > size) { |
f67539c2 | 12487 | r = -CEPHFS_ERANGE; |
81eedcae TL |
12488 | goto out; |
12489 | } | |
12490 | ||
f67539c2 | 12491 | memcpy(name, xattr_name.c_str(), this_len); |
81eedcae TL |
12492 | name += this_len; |
12493 | size -= this_len; | |
12494 | } | |
81eedcae | 12495 | out: |
11fdf7f2 | 12496 | ldout(cct, 8) << __func__ << "(" << in->ino << ", " << size << ") = " << r << dendl; |
7c673cae FG |
12497 | return r; |
12498 | } | |
12499 | ||
12500 | int Client::ll_listxattr(Inode *in, char *names, size_t size, | |
12501 | const UserPerm& perms) | |
12502 | { | |
f67539c2 TL |
12503 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12504 | if (!mref_reader.is_state_satisfied()) | |
12505 | return -CEPHFS_ENOTCONN; | |
181888fb | 12506 | |
7c673cae FG |
12507 | vinodeno_t vino = _get_vino(in); |
12508 | ||
11fdf7f2 TL |
12509 | ldout(cct, 3) << __func__ << " " << vino << " size " << size << dendl; |
12510 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12511 | tout(cct) << vino.ino.val << std::endl; |
12512 | tout(cct) << size << std::endl; | |
12513 | ||
f67539c2 | 12514 | std::scoped_lock lock(client_lock); |
7c673cae FG |
12515 | return _listxattr(in, names, size, perms); |
12516 | } | |
12517 | ||
12518 | int Client::_do_setxattr(Inode *in, const char *name, const void *value, | |
12519 | size_t size, int flags, const UserPerm& perms) | |
12520 | { | |
12521 | ||
12522 | int xattr_flags = 0; | |
12523 | if (!value) | |
12524 | xattr_flags |= CEPH_XATTR_REMOVE; | |
12525 | if (flags & XATTR_CREATE) | |
12526 | xattr_flags |= CEPH_XATTR_CREATE; | |
12527 | if (flags & XATTR_REPLACE) | |
12528 | xattr_flags |= CEPH_XATTR_REPLACE; | |
12529 | ||
12530 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETXATTR); | |
12531 | filepath path; | |
12532 | in->make_nosnap_relative_path(path); | |
12533 | req->set_filepath(path); | |
12534 | req->set_string2(name); | |
12535 | req->set_inode(in); | |
12536 | req->head.args.setxattr.flags = xattr_flags; | |
12537 | ||
12538 | bufferlist bl; | |
20effc67 | 12539 | ceph_assert(value || size == 0); |
7c673cae FG |
12540 | bl.append((const char*)value, size); |
12541 | req->set_data(bl); | |
12542 | ||
12543 | int res = make_request(req, perms); | |
12544 | ||
12545 | trim_cache(); | |
11fdf7f2 | 12546 | ldout(cct, 3) << __func__ << "(" << in->ino << ", \"" << name << "\") = " << |
7c673cae FG |
12547 | res << dendl; |
12548 | return res; | |
12549 | } | |
12550 | ||
12551 | int Client::_setxattr(Inode *in, const char *name, const void *value, | |
12552 | size_t size, int flags, const UserPerm& perms) | |
12553 | { | |
12554 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12555 | return -CEPHFS_EROFS; |
7c673cae FG |
12556 | } |
12557 | ||
f6b5b4d7 TL |
12558 | if (size == 0) { |
12559 | value = ""; | |
12560 | } else if (value == NULL) { | |
f67539c2 | 12561 | return -CEPHFS_EINVAL; |
f6b5b4d7 TL |
12562 | } |
12563 | ||
7c673cae FG |
12564 | bool posix_acl_xattr = false; |
12565 | if (acl_type == POSIX_ACL) | |
12566 | posix_acl_xattr = !strncmp(name, "system.", 7); | |
12567 | ||
12568 | if (strncmp(name, "user.", 5) && | |
12569 | strncmp(name, "security.", 9) && | |
12570 | strncmp(name, "trusted.", 8) && | |
12571 | strncmp(name, "ceph.", 5) && | |
12572 | !posix_acl_xattr) | |
f67539c2 | 12573 | return -CEPHFS_EOPNOTSUPP; |
7c673cae | 12574 | |
11fdf7f2 TL |
12575 | bool check_realm = false; |
12576 | ||
7c673cae FG |
12577 | if (posix_acl_xattr) { |
12578 | if (!strcmp(name, ACL_EA_ACCESS)) { | |
12579 | mode_t new_mode = in->mode; | |
12580 | if (value) { | |
12581 | int ret = posix_acl_equiv_mode(value, size, &new_mode); | |
12582 | if (ret < 0) | |
12583 | return ret; | |
12584 | if (ret == 0) { | |
12585 | value = NULL; | |
12586 | size = 0; | |
12587 | } | |
12588 | if (new_mode != in->mode) { | |
12589 | struct ceph_statx stx; | |
12590 | stx.stx_mode = new_mode; | |
12591 | ret = _do_setattr(in, &stx, CEPH_SETATTR_MODE, perms, NULL); | |
12592 | if (ret < 0) | |
12593 | return ret; | |
12594 | } | |
12595 | } | |
12596 | } else if (!strcmp(name, ACL_EA_DEFAULT)) { | |
12597 | if (value) { | |
12598 | if (!S_ISDIR(in->mode)) | |
f67539c2 | 12599 | return -CEPHFS_EACCES; |
7c673cae FG |
12600 | int ret = posix_acl_check(value, size); |
12601 | if (ret < 0) | |
f67539c2 | 12602 | return -CEPHFS_EINVAL; |
7c673cae FG |
12603 | if (ret == 0) { |
12604 | value = NULL; | |
12605 | size = 0; | |
12606 | } | |
12607 | } | |
12608 | } else { | |
f67539c2 | 12609 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12610 | } |
12611 | } else { | |
12612 | const VXattr *vxattr = _match_vxattr(in, name); | |
11fdf7f2 TL |
12613 | if (vxattr) { |
12614 | if (vxattr->readonly) | |
f67539c2 | 12615 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12616 | if (vxattr->name.compare(0, 10, "ceph.quota") == 0 && value) |
12617 | check_realm = true; | |
12618 | } | |
7c673cae FG |
12619 | } |
12620 | ||
11fdf7f2 TL |
12621 | int ret = _do_setxattr(in, name, value, size, flags, perms); |
12622 | if (ret >= 0 && check_realm) { | |
12623 | // check if snaprealm was created for quota inode | |
12624 | if (in->quota.is_enable() && | |
12625 | !(in->snaprealm && in->snaprealm->ino == in->ino)) | |
f67539c2 | 12626 | ret = -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12627 | } |
12628 | ||
12629 | return ret; | |
7c673cae FG |
12630 | } |
12631 | ||
12632 | int Client::_setxattr(InodeRef &in, const char *name, const void *value, | |
12633 | size_t size, int flags, const UserPerm& perms) | |
12634 | { | |
12635 | if (cct->_conf->client_permissions) { | |
12636 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12637 | if (r < 0) | |
12638 | return r; | |
12639 | } | |
12640 | return _setxattr(in.get(), name, value, size, flags, perms); | |
12641 | } | |
12642 | ||
12643 | int Client::_setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap) | |
12644 | { | |
12645 | string tmp; | |
12646 | if (name == "layout") { | |
12647 | string::iterator begin = value.begin(); | |
12648 | string::iterator end = value.end(); | |
12649 | keys_and_values<string::iterator> p; // create instance of parser | |
12650 | std::map<string, string> m; // map to receive results | |
12651 | if (!qi::parse(begin, end, p, m)) { // returns true if successful | |
f67539c2 | 12652 | return -CEPHFS_EINVAL; |
7c673cae FG |
12653 | } |
12654 | if (begin != end) | |
f67539c2 | 12655 | return -CEPHFS_EINVAL; |
7c673cae FG |
12656 | for (map<string,string>::iterator q = m.begin(); q != m.end(); ++q) { |
12657 | if (q->first == "pool") { | |
12658 | tmp = q->second; | |
12659 | break; | |
12660 | } | |
12661 | } | |
12662 | } else if (name == "layout.pool") { | |
12663 | tmp = value; | |
12664 | } | |
12665 | ||
12666 | if (tmp.length()) { | |
12667 | int64_t pool; | |
12668 | try { | |
12669 | pool = boost::lexical_cast<unsigned>(tmp); | |
12670 | if (!osdmap->have_pg_pool(pool)) | |
f67539c2 | 12671 | return -CEPHFS_ENOENT; |
7c673cae FG |
12672 | } catch (boost::bad_lexical_cast const&) { |
12673 | pool = osdmap->lookup_pg_pool_name(tmp); | |
12674 | if (pool < 0) { | |
f67539c2 | 12675 | return -CEPHFS_ENOENT; |
7c673cae FG |
12676 | } |
12677 | } | |
12678 | } | |
12679 | ||
12680 | return 0; | |
12681 | } | |
12682 | ||
12683 | void Client::_setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t size) | |
12684 | { | |
12685 | // For setting pool of layout, MetaRequest need osdmap epoch. | |
12686 | // There is a race which create a new data pool but client and mds both don't have. | |
12687 | // Make client got the latest osdmap which make mds quickly judge whether get newer osdmap. | |
f67539c2 | 12688 | ldout(cct, 15) << __func__ << ": name = " << name << dendl; |
7c673cae FG |
12689 | if (strcmp(name, "ceph.file.layout.pool") == 0 || strcmp(name, "ceph.dir.layout.pool") == 0 || |
12690 | strcmp(name, "ceph.file.layout") == 0 || strcmp(name, "ceph.dir.layout") == 0) { | |
12691 | string rest(strstr(name, "layout")); | |
12692 | string v((const char*)value, size); | |
12693 | int r = objecter->with_osdmap([&](const OSDMap& o) { | |
12694 | return _setxattr_check_data_pool(rest, v, &o); | |
12695 | }); | |
12696 | ||
f67539c2 TL |
12697 | if (r == -CEPHFS_ENOENT) { |
12698 | bs::error_code ec; | |
12699 | ldout(cct, 20) << __func__ << ": waiting for latest osdmap" << dendl; | |
12700 | objecter->wait_for_latest_osdmap(ca::use_blocked[ec]); | |
12701 | ldout(cct, 20) << __func__ << ": got latest osdmap: " << ec << dendl; | |
7c673cae FG |
12702 | } |
12703 | } | |
12704 | } | |
12705 | ||
12706 | int Client::ll_setxattr(Inode *in, const char *name, const void *value, | |
12707 | size_t size, int flags, const UserPerm& perms) | |
12708 | { | |
f67539c2 TL |
12709 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12710 | if (!mref_reader.is_state_satisfied()) | |
12711 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12712 | |
f67539c2 | 12713 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12714 | |
7c673cae FG |
12715 | vinodeno_t vino = _get_vino(in); |
12716 | ||
11fdf7f2 TL |
12717 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12718 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12719 | tout(cct) << vino.ino.val << std::endl; |
12720 | tout(cct) << name << std::endl; | |
12721 | ||
f67539c2 | 12722 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12723 | if (!fuse_default_permissions) { |
7c673cae FG |
12724 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12725 | if (r < 0) | |
12726 | return r; | |
12727 | } | |
12728 | return _setxattr(in, name, value, size, flags, perms); | |
12729 | } | |
12730 | ||
12731 | int Client::_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12732 | { | |
12733 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12734 | return -CEPHFS_EROFS; |
7c673cae FG |
12735 | } |
12736 | ||
12737 | // same xattrs supported by kernel client | |
12738 | if (strncmp(name, "user.", 5) && | |
12739 | strncmp(name, "system.", 7) && | |
12740 | strncmp(name, "security.", 9) && | |
12741 | strncmp(name, "trusted.", 8) && | |
12742 | strncmp(name, "ceph.", 5)) | |
f67539c2 | 12743 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12744 | |
12745 | const VXattr *vxattr = _match_vxattr(in, name); | |
12746 | if (vxattr && vxattr->readonly) | |
f67539c2 | 12747 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12748 | |
12749 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_RMXATTR); | |
12750 | filepath path; | |
12751 | in->make_nosnap_relative_path(path); | |
12752 | req->set_filepath(path); | |
12753 | req->set_filepath2(name); | |
12754 | req->set_inode(in); | |
12755 | ||
12756 | int res = make_request(req, perms); | |
12757 | ||
12758 | trim_cache(); | |
1adf2230 | 12759 | ldout(cct, 8) << "_removexattr(" << in->ino << ", \"" << name << "\") = " << res << dendl; |
7c673cae FG |
12760 | return res; |
12761 | } | |
12762 | ||
12763 | int Client::_removexattr(InodeRef &in, const char *name, const UserPerm& perms) | |
12764 | { | |
12765 | if (cct->_conf->client_permissions) { | |
12766 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12767 | if (r < 0) | |
12768 | return r; | |
12769 | } | |
12770 | return _removexattr(in.get(), name, perms); | |
12771 | } | |
12772 | ||
12773 | int Client::ll_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12774 | { | |
f67539c2 TL |
12775 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12776 | if (!mref_reader.is_state_satisfied()) | |
12777 | return -CEPHFS_ENOTCONN; | |
181888fb | 12778 | |
7c673cae FG |
12779 | vinodeno_t vino = _get_vino(in); |
12780 | ||
12781 | ldout(cct, 3) << "ll_removexattr " << vino << " " << name << dendl; | |
12782 | tout(cct) << "ll_removexattr" << std::endl; | |
12783 | tout(cct) << vino.ino.val << std::endl; | |
12784 | tout(cct) << name << std::endl; | |
12785 | ||
f67539c2 | 12786 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12787 | if (!fuse_default_permissions) { |
7c673cae FG |
12788 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12789 | if (r < 0) | |
12790 | return r; | |
12791 | } | |
12792 | ||
12793 | return _removexattr(in, name, perms); | |
12794 | } | |
12795 | ||
12796 | bool Client::_vxattrcb_quota_exists(Inode *in) | |
12797 | { | |
11fdf7f2 | 12798 | return in->quota.is_enable() && |
f6b5b4d7 TL |
12799 | (in->snapid != CEPH_NOSNAP || |
12800 | (in->snaprealm && in->snaprealm->ino == in->ino)); | |
7c673cae FG |
12801 | } |
12802 | size_t Client::_vxattrcb_quota(Inode *in, char *val, size_t size) | |
12803 | { | |
12804 | return snprintf(val, size, | |
12805 | "max_bytes=%lld max_files=%lld", | |
12806 | (long long int)in->quota.max_bytes, | |
12807 | (long long int)in->quota.max_files); | |
12808 | } | |
12809 | size_t Client::_vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size) | |
12810 | { | |
12811 | return snprintf(val, size, "%lld", (long long int)in->quota.max_bytes); | |
12812 | } | |
12813 | size_t Client::_vxattrcb_quota_max_files(Inode *in, char *val, size_t size) | |
12814 | { | |
12815 | return snprintf(val, size, "%lld", (long long int)in->quota.max_files); | |
12816 | } | |
12817 | ||
12818 | bool Client::_vxattrcb_layout_exists(Inode *in) | |
12819 | { | |
12820 | return in->layout != file_layout_t(); | |
12821 | } | |
12822 | size_t Client::_vxattrcb_layout(Inode *in, char *val, size_t size) | |
12823 | { | |
12824 | int r = snprintf(val, size, | |
11fdf7f2 | 12825 | "stripe_unit=%llu stripe_count=%llu object_size=%llu pool=", |
7c673cae FG |
12826 | (unsigned long long)in->layout.stripe_unit, |
12827 | (unsigned long long)in->layout.stripe_count, | |
12828 | (unsigned long long)in->layout.object_size); | |
12829 | objecter->with_osdmap([&](const OSDMap& o) { | |
12830 | if (o.have_pg_pool(in->layout.pool_id)) | |
12831 | r += snprintf(val + r, size - r, "%s", | |
12832 | o.get_pool_name(in->layout.pool_id).c_str()); | |
12833 | else | |
12834 | r += snprintf(val + r, size - r, "%" PRIu64, | |
12835 | (uint64_t)in->layout.pool_id); | |
12836 | }); | |
12837 | if (in->layout.pool_ns.length()) | |
12838 | r += snprintf(val + r, size - r, " pool_namespace=%s", | |
12839 | in->layout.pool_ns.c_str()); | |
12840 | return r; | |
12841 | } | |
12842 | size_t Client::_vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size) | |
12843 | { | |
11fdf7f2 | 12844 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_unit); |
7c673cae FG |
12845 | } |
12846 | size_t Client::_vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size) | |
12847 | { | |
11fdf7f2 | 12848 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_count); |
7c673cae FG |
12849 | } |
12850 | size_t Client::_vxattrcb_layout_object_size(Inode *in, char *val, size_t size) | |
12851 | { | |
11fdf7f2 | 12852 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.object_size); |
7c673cae FG |
12853 | } |
12854 | size_t Client::_vxattrcb_layout_pool(Inode *in, char *val, size_t size) | |
12855 | { | |
12856 | size_t r; | |
12857 | objecter->with_osdmap([&](const OSDMap& o) { | |
12858 | if (o.have_pg_pool(in->layout.pool_id)) | |
12859 | r = snprintf(val, size, "%s", o.get_pool_name( | |
12860 | in->layout.pool_id).c_str()); | |
12861 | else | |
12862 | r = snprintf(val, size, "%" PRIu64, (uint64_t)in->layout.pool_id); | |
12863 | }); | |
12864 | return r; | |
12865 | } | |
12866 | size_t Client::_vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size) | |
12867 | { | |
12868 | return snprintf(val, size, "%s", in->layout.pool_ns.c_str()); | |
12869 | } | |
12870 | size_t Client::_vxattrcb_dir_entries(Inode *in, char *val, size_t size) | |
12871 | { | |
11fdf7f2 | 12872 | return snprintf(val, size, "%llu", (unsigned long long)(in->dirstat.nfiles + in->dirstat.nsubdirs)); |
7c673cae FG |
12873 | } |
12874 | size_t Client::_vxattrcb_dir_files(Inode *in, char *val, size_t size) | |
12875 | { | |
11fdf7f2 | 12876 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nfiles); |
7c673cae FG |
12877 | } |
12878 | size_t Client::_vxattrcb_dir_subdirs(Inode *in, char *val, size_t size) | |
12879 | { | |
11fdf7f2 | 12880 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nsubdirs); |
7c673cae FG |
12881 | } |
12882 | size_t Client::_vxattrcb_dir_rentries(Inode *in, char *val, size_t size) | |
12883 | { | |
11fdf7f2 | 12884 | return snprintf(val, size, "%llu", (unsigned long long)(in->rstat.rfiles + in->rstat.rsubdirs)); |
7c673cae FG |
12885 | } |
12886 | size_t Client::_vxattrcb_dir_rfiles(Inode *in, char *val, size_t size) | |
12887 | { | |
11fdf7f2 | 12888 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rfiles); |
7c673cae FG |
12889 | } |
12890 | size_t Client::_vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size) | |
12891 | { | |
11fdf7f2 | 12892 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsubdirs); |
7c673cae | 12893 | } |
f67539c2 TL |
12894 | size_t Client::_vxattrcb_dir_rsnaps(Inode *in, char *val, size_t size) |
12895 | { | |
12896 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsnaps); | |
12897 | } | |
7c673cae FG |
12898 | size_t Client::_vxattrcb_dir_rbytes(Inode *in, char *val, size_t size) |
12899 | { | |
11fdf7f2 | 12900 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rbytes); |
7c673cae FG |
12901 | } |
12902 | size_t Client::_vxattrcb_dir_rctime(Inode *in, char *val, size_t size) | |
12903 | { | |
81eedcae | 12904 | return snprintf(val, size, "%ld.%09ld", (long)in->rstat.rctime.sec(), |
7c673cae FG |
12905 | (long)in->rstat.rctime.nsec()); |
12906 | } | |
11fdf7f2 TL |
12907 | bool Client::_vxattrcb_dir_pin_exists(Inode *in) |
12908 | { | |
f67539c2 | 12909 | return in->dir_pin != -CEPHFS_ENODATA; |
11fdf7f2 TL |
12910 | } |
12911 | size_t Client::_vxattrcb_dir_pin(Inode *in, char *val, size_t size) | |
12912 | { | |
12913 | return snprintf(val, size, "%ld", (long)in->dir_pin); | |
12914 | } | |
7c673cae | 12915 | |
81eedcae TL |
12916 | bool Client::_vxattrcb_snap_btime_exists(Inode *in) |
12917 | { | |
12918 | return !in->snap_btime.is_zero(); | |
12919 | } | |
12920 | ||
12921 | size_t Client::_vxattrcb_snap_btime(Inode *in, char *val, size_t size) | |
12922 | { | |
12923 | return snprintf(val, size, "%llu.%09lu", | |
12924 | (long long unsigned)in->snap_btime.sec(), | |
12925 | (long unsigned)in->snap_btime.nsec()); | |
12926 | } | |
12927 | ||
20effc67 TL |
12928 | size_t Client::_vxattrcb_caps(Inode *in, char *val, size_t size) |
12929 | { | |
12930 | int issued; | |
12931 | ||
12932 | in->caps_issued(&issued); | |
12933 | return snprintf(val, size, "%s/0x%x", ccap_string(issued).c_str(), issued); | |
12934 | } | |
12935 | ||
f67539c2 TL |
12936 | bool Client::_vxattrcb_mirror_info_exists(Inode *in) |
12937 | { | |
12938 | // checking one of the xattrs would suffice | |
12939 | return in->xattrs.count("ceph.mirror.info.cluster_id") != 0; | |
12940 | } | |
12941 | ||
12942 | size_t Client::_vxattrcb_mirror_info(Inode *in, char *val, size_t size) | |
12943 | { | |
12944 | return snprintf(val, size, "cluster_id=%.*s fs_id=%.*s", | |
12945 | in->xattrs["ceph.mirror.info.cluster_id"].length(), | |
12946 | in->xattrs["ceph.mirror.info.cluster_id"].c_str(), | |
12947 | in->xattrs["ceph.mirror.info.fs_id"].length(), | |
12948 | in->xattrs["ceph.mirror.info.fs_id"].c_str()); | |
12949 | } | |
12950 | ||
adb31ebb TL |
12951 | size_t Client::_vxattrcb_cluster_fsid(Inode *in, char *val, size_t size) |
12952 | { | |
12953 | return snprintf(val, size, "%s", monclient->get_fsid().to_string().c_str()); | |
12954 | } | |
12955 | ||
12956 | size_t Client::_vxattrcb_client_id(Inode *in, char *val, size_t size) | |
12957 | { | |
12958 | auto name = messenger->get_myname(); | |
20effc67 | 12959 | return snprintf(val, size, "%s%" PRId64, name.type_str(), name.num()); |
adb31ebb TL |
12960 | } |
12961 | ||
7c673cae FG |
12962 | #define CEPH_XATTR_NAME(_type, _name) "ceph." #_type "." #_name |
12963 | #define CEPH_XATTR_NAME2(_type, _name, _name2) "ceph." #_type "." #_name "." #_name2 | |
12964 | ||
adb31ebb | 12965 | #define XATTR_NAME_CEPH(_type, _name, _flags) \ |
28e407b8 AA |
12966 | { \ |
12967 | name: CEPH_XATTR_NAME(_type, _name), \ | |
12968 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
12969 | readonly: true, \ | |
28e407b8 AA |
12970 | exists_cb: NULL, \ |
12971 | flags: _flags, \ | |
7c673cae FG |
12972 | } |
12973 | #define XATTR_LAYOUT_FIELD(_type, _name, _field) \ | |
12974 | { \ | |
12975 | name: CEPH_XATTR_NAME2(_type, _name, _field), \ | |
12976 | getxattr_cb: &Client::_vxattrcb_ ## _name ## _ ## _field, \ | |
12977 | readonly: false, \ | |
7c673cae | 12978 | exists_cb: &Client::_vxattrcb_layout_exists, \ |
28e407b8 | 12979 | flags: 0, \ |
7c673cae FG |
12980 | } |
12981 | #define XATTR_QUOTA_FIELD(_type, _name) \ | |
12982 | { \ | |
12983 | name: CEPH_XATTR_NAME(_type, _name), \ | |
12984 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
12985 | readonly: false, \ | |
7c673cae | 12986 | exists_cb: &Client::_vxattrcb_quota_exists, \ |
28e407b8 | 12987 | flags: 0, \ |
7c673cae FG |
12988 | } |
12989 | ||
12990 | const Client::VXattr Client::_dir_vxattrs[] = { | |
12991 | { | |
12992 | name: "ceph.dir.layout", | |
12993 | getxattr_cb: &Client::_vxattrcb_layout, | |
12994 | readonly: false, | |
7c673cae | 12995 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 12996 | flags: 0, |
7c673cae | 12997 | }, |
1d09f67e TL |
12998 | // FIXME |
12999 | // Delete the following dir layout field definitions for release "S" | |
7c673cae FG |
13000 | XATTR_LAYOUT_FIELD(dir, layout, stripe_unit), |
13001 | XATTR_LAYOUT_FIELD(dir, layout, stripe_count), | |
13002 | XATTR_LAYOUT_FIELD(dir, layout, object_size), | |
13003 | XATTR_LAYOUT_FIELD(dir, layout, pool), | |
13004 | XATTR_LAYOUT_FIELD(dir, layout, pool_namespace), | |
adb31ebb TL |
13005 | XATTR_NAME_CEPH(dir, entries, VXATTR_DIRSTAT), |
13006 | XATTR_NAME_CEPH(dir, files, VXATTR_DIRSTAT), | |
13007 | XATTR_NAME_CEPH(dir, subdirs, VXATTR_DIRSTAT), | |
13008 | XATTR_NAME_CEPH(dir, rentries, VXATTR_RSTAT), | |
13009 | XATTR_NAME_CEPH(dir, rfiles, VXATTR_RSTAT), | |
13010 | XATTR_NAME_CEPH(dir, rsubdirs, VXATTR_RSTAT), | |
f67539c2 | 13011 | XATTR_NAME_CEPH(dir, rsnaps, VXATTR_RSTAT), |
adb31ebb TL |
13012 | XATTR_NAME_CEPH(dir, rbytes, VXATTR_RSTAT), |
13013 | XATTR_NAME_CEPH(dir, rctime, VXATTR_RSTAT), | |
7c673cae FG |
13014 | { |
13015 | name: "ceph.quota", | |
13016 | getxattr_cb: &Client::_vxattrcb_quota, | |
13017 | readonly: false, | |
7c673cae | 13018 | exists_cb: &Client::_vxattrcb_quota_exists, |
28e407b8 | 13019 | flags: 0, |
7c673cae FG |
13020 | }, |
13021 | XATTR_QUOTA_FIELD(quota, max_bytes), | |
13022 | XATTR_QUOTA_FIELD(quota, max_files), | |
1d09f67e TL |
13023 | // FIXME |
13024 | // Delete the following dir pin field definitions for release "S" | |
11fdf7f2 TL |
13025 | { |
13026 | name: "ceph.dir.pin", | |
13027 | getxattr_cb: &Client::_vxattrcb_dir_pin, | |
13028 | readonly: false, | |
11fdf7f2 TL |
13029 | exists_cb: &Client::_vxattrcb_dir_pin_exists, |
13030 | flags: 0, | |
13031 | }, | |
81eedcae TL |
13032 | { |
13033 | name: "ceph.snap.btime", | |
13034 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
13035 | readonly: true, | |
81eedcae TL |
13036 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
13037 | flags: 0, | |
13038 | }, | |
f67539c2 TL |
13039 | { |
13040 | name: "ceph.mirror.info", | |
13041 | getxattr_cb: &Client::_vxattrcb_mirror_info, | |
13042 | readonly: false, | |
13043 | exists_cb: &Client::_vxattrcb_mirror_info_exists, | |
13044 | flags: 0, | |
13045 | }, | |
20effc67 TL |
13046 | { |
13047 | name: "ceph.caps", | |
13048 | getxattr_cb: &Client::_vxattrcb_caps, | |
13049 | readonly: true, | |
13050 | exists_cb: NULL, | |
13051 | flags: 0, | |
13052 | }, | |
7c673cae FG |
13053 | { name: "" } /* Required table terminator */ |
13054 | }; | |
13055 | ||
13056 | const Client::VXattr Client::_file_vxattrs[] = { | |
13057 | { | |
13058 | name: "ceph.file.layout", | |
13059 | getxattr_cb: &Client::_vxattrcb_layout, | |
13060 | readonly: false, | |
7c673cae | 13061 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 13062 | flags: 0, |
7c673cae FG |
13063 | }, |
13064 | XATTR_LAYOUT_FIELD(file, layout, stripe_unit), | |
13065 | XATTR_LAYOUT_FIELD(file, layout, stripe_count), | |
13066 | XATTR_LAYOUT_FIELD(file, layout, object_size), | |
13067 | XATTR_LAYOUT_FIELD(file, layout, pool), | |
13068 | XATTR_LAYOUT_FIELD(file, layout, pool_namespace), | |
81eedcae TL |
13069 | { |
13070 | name: "ceph.snap.btime", | |
13071 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
13072 | readonly: true, | |
81eedcae TL |
13073 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
13074 | flags: 0, | |
13075 | }, | |
20effc67 TL |
13076 | { |
13077 | name: "ceph.caps", | |
13078 | getxattr_cb: &Client::_vxattrcb_caps, | |
13079 | readonly: true, | |
13080 | exists_cb: NULL, | |
13081 | flags: 0, | |
13082 | }, | |
7c673cae FG |
13083 | { name: "" } /* Required table terminator */ |
13084 | }; | |
13085 | ||
adb31ebb TL |
13086 | const Client::VXattr Client::_common_vxattrs[] = { |
13087 | { | |
13088 | name: "ceph.cluster_fsid", | |
13089 | getxattr_cb: &Client::_vxattrcb_cluster_fsid, | |
13090 | readonly: true, | |
13091 | exists_cb: nullptr, | |
13092 | flags: 0, | |
13093 | }, | |
13094 | { | |
13095 | name: "ceph.client_id", | |
13096 | getxattr_cb: &Client::_vxattrcb_client_id, | |
13097 | readonly: true, | |
13098 | exists_cb: nullptr, | |
13099 | flags: 0, | |
13100 | }, | |
13101 | { name: "" } /* Required table terminator */ | |
13102 | }; | |
13103 | ||
7c673cae FG |
13104 | const Client::VXattr *Client::_get_vxattrs(Inode *in) |
13105 | { | |
13106 | if (in->is_dir()) | |
13107 | return _dir_vxattrs; | |
13108 | else if (in->is_file()) | |
13109 | return _file_vxattrs; | |
13110 | return NULL; | |
13111 | } | |
13112 | ||
13113 | const Client::VXattr *Client::_match_vxattr(Inode *in, const char *name) | |
13114 | { | |
13115 | if (strncmp(name, "ceph.", 5) == 0) { | |
13116 | const VXattr *vxattr = _get_vxattrs(in); | |
13117 | if (vxattr) { | |
13118 | while (!vxattr->name.empty()) { | |
13119 | if (vxattr->name == name) | |
13120 | return vxattr; | |
13121 | vxattr++; | |
13122 | } | |
13123 | } | |
adb31ebb TL |
13124 | |
13125 | // for common vxattrs | |
13126 | vxattr = _common_vxattrs; | |
13127 | while (!vxattr->name.empty()) { | |
13128 | if (vxattr->name == name) | |
13129 | return vxattr; | |
13130 | vxattr++; | |
13131 | } | |
7c673cae | 13132 | } |
adb31ebb | 13133 | |
7c673cae FG |
13134 | return NULL; |
13135 | } | |
13136 | ||
7c673cae FG |
13137 | int Client::ll_readlink(Inode *in, char *buf, size_t buflen, const UserPerm& perms) |
13138 | { | |
f67539c2 TL |
13139 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13140 | if (!mref_reader.is_state_satisfied()) | |
13141 | return -CEPHFS_ENOTCONN; | |
181888fb | 13142 | |
7c673cae FG |
13143 | vinodeno_t vino = _get_vino(in); |
13144 | ||
13145 | ldout(cct, 3) << "ll_readlink " << vino << dendl; | |
13146 | tout(cct) << "ll_readlink" << std::endl; | |
13147 | tout(cct) << vino.ino.val << std::endl; | |
13148 | ||
f67539c2 | 13149 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
13150 | for (auto dn : in->dentries) { |
13151 | touch_dn(dn); | |
7c673cae FG |
13152 | } |
13153 | ||
13154 | int r = _readlink(in, buf, buflen); // FIXME: no permission checking! | |
13155 | ldout(cct, 3) << "ll_readlink " << vino << " = " << r << dendl; | |
13156 | return r; | |
13157 | } | |
13158 | ||
13159 | int Client::_mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev, | |
13160 | const UserPerm& perms, InodeRef *inp) | |
13161 | { | |
1adf2230 | 13162 | ldout(cct, 8) << "_mknod(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
13163 | << mode << dec << ", " << rdev << ", uid " << perms.uid() |
13164 | << ", gid " << perms.gid() << ")" << dendl; | |
13165 | ||
13166 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13167 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13168 | |
13169 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13170 | return -CEPHFS_EROFS; |
7c673cae FG |
13171 | } |
13172 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13173 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13174 | } |
13175 | ||
13176 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_MKNOD); | |
13177 | ||
13178 | filepath path; | |
13179 | dir->make_nosnap_relative_path(path); | |
13180 | path.push_dentry(name); | |
13181 | req->set_filepath(path); | |
13182 | req->set_inode(dir); | |
13183 | req->head.args.mknod.rdev = rdev; | |
13184 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13185 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13186 | ||
13187 | bufferlist xattrs_bl; | |
13188 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
13189 | if (res < 0) | |
13190 | goto fail; | |
13191 | req->head.args.mknod.mode = mode; | |
13192 | if (xattrs_bl.length() > 0) | |
13193 | req->set_data(xattrs_bl); | |
13194 | ||
13195 | Dentry *de; | |
13196 | res = get_or_create(dir, name, &de); | |
13197 | if (res < 0) | |
13198 | goto fail; | |
13199 | req->set_dentry(de); | |
13200 | ||
13201 | res = make_request(req, perms, inp); | |
13202 | ||
13203 | trim_cache(); | |
13204 | ||
1adf2230 | 13205 | ldout(cct, 8) << "mknod(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13206 | return res; |
13207 | ||
13208 | fail: | |
13209 | put_request(req); | |
13210 | return res; | |
13211 | } | |
13212 | ||
13213 | int Client::ll_mknod(Inode *parent, const char *name, mode_t mode, | |
13214 | dev_t rdev, struct stat *attr, Inode **out, | |
13215 | const UserPerm& perms) | |
13216 | { | |
f67539c2 TL |
13217 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13218 | if (!mref_reader.is_state_satisfied()) | |
13219 | return -CEPHFS_ENOTCONN; | |
181888fb | 13220 | |
7c673cae FG |
13221 | vinodeno_t vparent = _get_vino(parent); |
13222 | ||
13223 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name << dendl; | |
13224 | tout(cct) << "ll_mknod" << std::endl; | |
13225 | tout(cct) << vparent.ino.val << std::endl; | |
13226 | tout(cct) << name << std::endl; | |
13227 | tout(cct) << mode << std::endl; | |
13228 | tout(cct) << rdev << std::endl; | |
13229 | ||
f67539c2 | 13230 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 13231 | if (!fuse_default_permissions) { |
7c673cae FG |
13232 | int r = may_create(parent, perms); |
13233 | if (r < 0) | |
13234 | return r; | |
13235 | } | |
13236 | ||
13237 | InodeRef in; | |
13238 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13239 | if (r == 0) { | |
13240 | fill_stat(in, attr); | |
13241 | _ll_get(in.get()); | |
13242 | } | |
13243 | tout(cct) << attr->st_ino << std::endl; | |
13244 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name | |
13245 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13246 | *out = in.get(); | |
13247 | return r; | |
13248 | } | |
13249 | ||
13250 | int Client::ll_mknodx(Inode *parent, const char *name, mode_t mode, | |
13251 | dev_t rdev, Inode **out, | |
13252 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13253 | const UserPerm& perms) | |
13254 | { | |
f67539c2 TL |
13255 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13256 | if (!mref_reader.is_state_satisfied()) | |
13257 | return -CEPHFS_ENOTCONN; | |
7c673cae | 13258 | |
f67539c2 | 13259 | unsigned caps = statx_to_mask(flags, want); |
181888fb | 13260 | |
7c673cae FG |
13261 | vinodeno_t vparent = _get_vino(parent); |
13262 | ||
13263 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name << dendl; | |
13264 | tout(cct) << "ll_mknodx" << std::endl; | |
13265 | tout(cct) << vparent.ino.val << std::endl; | |
13266 | tout(cct) << name << std::endl; | |
13267 | tout(cct) << mode << std::endl; | |
13268 | tout(cct) << rdev << std::endl; | |
13269 | ||
f67539c2 TL |
13270 | std::scoped_lock lock(client_lock); |
13271 | ||
11fdf7f2 | 13272 | if (!fuse_default_permissions) { |
7c673cae FG |
13273 | int r = may_create(parent, perms); |
13274 | if (r < 0) | |
13275 | return r; | |
13276 | } | |
13277 | ||
13278 | InodeRef in; | |
13279 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13280 | if (r == 0) { | |
13281 | fill_statx(in, caps, stx); | |
13282 | _ll_get(in.get()); | |
13283 | } | |
13284 | tout(cct) << stx->stx_ino << std::endl; | |
13285 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name | |
13286 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13287 | *out = in.get(); | |
13288 | return r; | |
13289 | } | |
13290 | ||
13291 | int Client::_create(Inode *dir, const char *name, int flags, mode_t mode, | |
13292 | InodeRef *inp, Fh **fhp, int stripe_unit, int stripe_count, | |
13293 | int object_size, const char *data_pool, bool *created, | |
f67539c2 | 13294 | const UserPerm& perms, std::string alternate_name) |
7c673cae | 13295 | { |
1adf2230 | 13296 | ldout(cct, 8) << "_create(" << dir->ino << " " << name << ", 0" << oct << |
7c673cae FG |
13297 | mode << dec << ")" << dendl; |
13298 | ||
13299 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13300 | return -CEPHFS_ENAMETOOLONG; |
7c673cae | 13301 | if (dir->snapid != CEPH_NOSNAP) { |
f67539c2 | 13302 | return -CEPHFS_EROFS; |
7c673cae FG |
13303 | } |
13304 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13305 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13306 | } |
13307 | ||
13308 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
13309 | int cflags = ceph_flags_sys2wire(flags); |
13310 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
13311 | cflags |= CEPH_O_LAZY; | |
13312 | ||
13313 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
13314 | |
13315 | int64_t pool_id = -1; | |
13316 | if (data_pool && *data_pool) { | |
13317 | pool_id = objecter->with_osdmap( | |
13318 | std::mem_fn(&OSDMap::lookup_pg_pool_name), data_pool); | |
13319 | if (pool_id < 0) | |
f67539c2 | 13320 | return -CEPHFS_EINVAL; |
7c673cae | 13321 | if (pool_id > 0xffffffffll) |
f67539c2 | 13322 | return -CEPHFS_ERANGE; // bummer! |
7c673cae FG |
13323 | } |
13324 | ||
13325 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_CREATE); | |
13326 | ||
13327 | filepath path; | |
13328 | dir->make_nosnap_relative_path(path); | |
13329 | path.push_dentry(name); | |
13330 | req->set_filepath(path); | |
f67539c2 | 13331 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae | 13332 | req->set_inode(dir); |
11fdf7f2 | 13333 | req->head.args.open.flags = cflags | CEPH_O_CREAT; |
7c673cae FG |
13334 | |
13335 | req->head.args.open.stripe_unit = stripe_unit; | |
13336 | req->head.args.open.stripe_count = stripe_count; | |
13337 | req->head.args.open.object_size = object_size; | |
13338 | if (cct->_conf->client_debug_getattr_caps) | |
13339 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
13340 | else | |
13341 | req->head.args.open.mask = 0; | |
13342 | req->head.args.open.pool = pool_id; | |
13343 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13344 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13345 | ||
13346 | mode |= S_IFREG; | |
13347 | bufferlist xattrs_bl; | |
13348 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
13349 | if (res < 0) | |
13350 | goto fail; | |
13351 | req->head.args.open.mode = mode; | |
13352 | if (xattrs_bl.length() > 0) | |
13353 | req->set_data(xattrs_bl); | |
13354 | ||
13355 | Dentry *de; | |
13356 | res = get_or_create(dir, name, &de); | |
13357 | if (res < 0) | |
13358 | goto fail; | |
13359 | req->set_dentry(de); | |
13360 | ||
13361 | res = make_request(req, perms, inp, created); | |
13362 | if (res < 0) { | |
13363 | goto reply_error; | |
13364 | } | |
13365 | ||
13366 | /* If the caller passed a value in fhp, do the open */ | |
13367 | if(fhp) { | |
13368 | (*inp)->get_open_ref(cmode); | |
13369 | *fhp = _create_fh(inp->get(), flags, cmode, perms); | |
13370 | } | |
13371 | ||
13372 | reply_error: | |
13373 | trim_cache(); | |
13374 | ||
1adf2230 | 13375 | ldout(cct, 8) << "create(" << path << ", 0" << oct << mode << dec |
7c673cae FG |
13376 | << " layout " << stripe_unit |
13377 | << ' ' << stripe_count | |
13378 | << ' ' << object_size | |
13379 | <<") = " << res << dendl; | |
13380 | return res; | |
13381 | ||
13382 | fail: | |
13383 | put_request(req); | |
13384 | return res; | |
13385 | } | |
13386 | ||
7c673cae | 13387 | int Client::_mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm, |
f67539c2 TL |
13388 | InodeRef *inp, const std::map<std::string, std::string> &metadata, |
13389 | std::string alternate_name) | |
7c673cae | 13390 | { |
1adf2230 | 13391 | ldout(cct, 8) << "_mkdir(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
13392 | << mode << dec << ", uid " << perm.uid() |
13393 | << ", gid " << perm.gid() << ")" << dendl; | |
13394 | ||
13395 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13396 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13397 | |
13398 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13399 | return -CEPHFS_EROFS; |
7c673cae FG |
13400 | } |
13401 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 13402 | return -CEPHFS_EDQUOT; |
7c673cae | 13403 | } |
f67539c2 TL |
13404 | |
13405 | bool is_snap_op = dir->snapid == CEPH_SNAPDIR; | |
13406 | MetaRequest *req = new MetaRequest(is_snap_op ? | |
7c673cae FG |
13407 | CEPH_MDS_OP_MKSNAP : CEPH_MDS_OP_MKDIR); |
13408 | ||
13409 | filepath path; | |
13410 | dir->make_nosnap_relative_path(path); | |
13411 | path.push_dentry(name); | |
13412 | req->set_filepath(path); | |
13413 | req->set_inode(dir); | |
13414 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13415 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
f67539c2 | 13416 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13417 | |
13418 | mode |= S_IFDIR; | |
f67539c2 TL |
13419 | bufferlist bl; |
13420 | int res = _posix_acl_create(dir, &mode, bl, perm); | |
7c673cae FG |
13421 | if (res < 0) |
13422 | goto fail; | |
13423 | req->head.args.mkdir.mode = mode; | |
f67539c2 TL |
13424 | if (is_snap_op) { |
13425 | SnapPayload payload; | |
13426 | // clear the bufferlist that may have been populated by the call | |
13427 | // to _posix_acl_create(). MDS mksnap does not make use of it. | |
13428 | // So, reuse it to pass metadata payload. | |
13429 | bl.clear(); | |
13430 | payload.metadata = metadata; | |
13431 | encode(payload, bl); | |
13432 | } | |
13433 | if (bl.length() > 0) { | |
13434 | req->set_data(bl); | |
13435 | } | |
7c673cae FG |
13436 | |
13437 | Dentry *de; | |
13438 | res = get_or_create(dir, name, &de); | |
13439 | if (res < 0) | |
13440 | goto fail; | |
13441 | req->set_dentry(de); | |
13442 | ||
13443 | ldout(cct, 10) << "_mkdir: making request" << dendl; | |
13444 | res = make_request(req, perm, inp); | |
13445 | ldout(cct, 10) << "_mkdir result is " << res << dendl; | |
13446 | ||
13447 | trim_cache(); | |
13448 | ||
1adf2230 | 13449 | ldout(cct, 8) << "_mkdir(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13450 | return res; |
13451 | ||
13452 | fail: | |
13453 | put_request(req); | |
13454 | return res; | |
13455 | } | |
13456 | ||
13457 | int Client::ll_mkdir(Inode *parent, const char *name, mode_t mode, | |
13458 | struct stat *attr, Inode **out, const UserPerm& perm) | |
13459 | { | |
f67539c2 TL |
13460 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13461 | if (!mref_reader.is_state_satisfied()) | |
13462 | return -CEPHFS_ENOTCONN; | |
181888fb | 13463 | |
7c673cae FG |
13464 | vinodeno_t vparent = _get_vino(parent); |
13465 | ||
13466 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name << dendl; | |
13467 | tout(cct) << "ll_mkdir" << std::endl; | |
13468 | tout(cct) << vparent.ino.val << std::endl; | |
13469 | tout(cct) << name << std::endl; | |
13470 | tout(cct) << mode << std::endl; | |
13471 | ||
f67539c2 TL |
13472 | std::scoped_lock lock(client_lock); |
13473 | ||
11fdf7f2 | 13474 | if (!fuse_default_permissions) { |
7c673cae FG |
13475 | int r = may_create(parent, perm); |
13476 | if (r < 0) | |
13477 | return r; | |
13478 | } | |
13479 | ||
13480 | InodeRef in; | |
13481 | int r = _mkdir(parent, name, mode, perm, &in); | |
13482 | if (r == 0) { | |
13483 | fill_stat(in, attr); | |
13484 | _ll_get(in.get()); | |
13485 | } | |
13486 | tout(cct) << attr->st_ino << std::endl; | |
13487 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name | |
13488 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13489 | *out = in.get(); | |
13490 | return r; | |
13491 | } | |
13492 | ||
13493 | int Client::ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out, | |
13494 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13495 | const UserPerm& perms) | |
13496 | { | |
f67539c2 TL |
13497 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13498 | if (!mref_reader.is_state_satisfied()) | |
13499 | return -CEPHFS_ENOTCONN; | |
181888fb | 13500 | |
7c673cae FG |
13501 | vinodeno_t vparent = _get_vino(parent); |
13502 | ||
13503 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name << dendl; | |
13504 | tout(cct) << "ll_mkdirx" << std::endl; | |
13505 | tout(cct) << vparent.ino.val << std::endl; | |
13506 | tout(cct) << name << std::endl; | |
13507 | tout(cct) << mode << std::endl; | |
13508 | ||
f67539c2 TL |
13509 | std::scoped_lock lock(client_lock); |
13510 | ||
11fdf7f2 | 13511 | if (!fuse_default_permissions) { |
7c673cae FG |
13512 | int r = may_create(parent, perms); |
13513 | if (r < 0) | |
13514 | return r; | |
13515 | } | |
13516 | ||
13517 | InodeRef in; | |
13518 | int r = _mkdir(parent, name, mode, perms, &in); | |
13519 | if (r == 0) { | |
13520 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13521 | _ll_get(in.get()); | |
13522 | } else { | |
13523 | stx->stx_ino = 0; | |
13524 | stx->stx_mask = 0; | |
13525 | } | |
13526 | tout(cct) << stx->stx_ino << std::endl; | |
13527 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name | |
13528 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13529 | *out = in.get(); | |
13530 | return r; | |
13531 | } | |
13532 | ||
13533 | int Client::_symlink(Inode *dir, const char *name, const char *target, | |
f67539c2 | 13534 | const UserPerm& perms, std::string alternate_name, InodeRef *inp) |
7c673cae | 13535 | { |
1adf2230 | 13536 | ldout(cct, 8) << "_symlink(" << dir->ino << " " << name << ", " << target |
7c673cae FG |
13537 | << ", uid " << perms.uid() << ", gid " << perms.gid() << ")" |
13538 | << dendl; | |
13539 | ||
13540 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13541 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13542 | |
13543 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13544 | return -CEPHFS_EROFS; |
7c673cae FG |
13545 | } |
13546 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13547 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13548 | } |
13549 | ||
13550 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SYMLINK); | |
13551 | ||
13552 | filepath path; | |
13553 | dir->make_nosnap_relative_path(path); | |
13554 | path.push_dentry(name); | |
13555 | req->set_filepath(path); | |
f67539c2 | 13556 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13557 | req->set_inode(dir); |
13558 | req->set_string2(target); | |
13559 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13560 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13561 | ||
13562 | Dentry *de; | |
13563 | int res = get_or_create(dir, name, &de); | |
13564 | if (res < 0) | |
13565 | goto fail; | |
13566 | req->set_dentry(de); | |
13567 | ||
13568 | res = make_request(req, perms, inp); | |
13569 | ||
13570 | trim_cache(); | |
1adf2230 | 13571 | ldout(cct, 8) << "_symlink(\"" << path << "\", \"" << target << "\") = " << |
7c673cae FG |
13572 | res << dendl; |
13573 | return res; | |
13574 | ||
13575 | fail: | |
13576 | put_request(req); | |
13577 | return res; | |
13578 | } | |
13579 | ||
13580 | int Client::ll_symlink(Inode *parent, const char *name, const char *value, | |
13581 | struct stat *attr, Inode **out, const UserPerm& perms) | |
13582 | { | |
f67539c2 TL |
13583 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13584 | if (!mref_reader.is_state_satisfied()) | |
13585 | return -CEPHFS_ENOTCONN; | |
181888fb | 13586 | |
7c673cae FG |
13587 | vinodeno_t vparent = _get_vino(parent); |
13588 | ||
13589 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name << " -> " << value | |
13590 | << dendl; | |
13591 | tout(cct) << "ll_symlink" << std::endl; | |
13592 | tout(cct) << vparent.ino.val << std::endl; | |
13593 | tout(cct) << name << std::endl; | |
13594 | tout(cct) << value << std::endl; | |
13595 | ||
f67539c2 TL |
13596 | std::scoped_lock lock(client_lock); |
13597 | ||
11fdf7f2 | 13598 | if (!fuse_default_permissions) { |
7c673cae FG |
13599 | int r = may_create(parent, perms); |
13600 | if (r < 0) | |
13601 | return r; | |
13602 | } | |
13603 | ||
13604 | InodeRef in; | |
f67539c2 | 13605 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13606 | if (r == 0) { |
13607 | fill_stat(in, attr); | |
13608 | _ll_get(in.get()); | |
13609 | } | |
13610 | tout(cct) << attr->st_ino << std::endl; | |
13611 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name | |
13612 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13613 | *out = in.get(); | |
13614 | return r; | |
13615 | } | |
13616 | ||
13617 | int Client::ll_symlinkx(Inode *parent, const char *name, const char *value, | |
13618 | Inode **out, struct ceph_statx *stx, unsigned want, | |
13619 | unsigned flags, const UserPerm& perms) | |
13620 | { | |
f67539c2 TL |
13621 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13622 | if (!mref_reader.is_state_satisfied()) | |
13623 | return -CEPHFS_ENOTCONN; | |
181888fb | 13624 | |
7c673cae FG |
13625 | vinodeno_t vparent = _get_vino(parent); |
13626 | ||
13627 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name << " -> " << value | |
13628 | << dendl; | |
13629 | tout(cct) << "ll_symlinkx" << std::endl; | |
13630 | tout(cct) << vparent.ino.val << std::endl; | |
13631 | tout(cct) << name << std::endl; | |
13632 | tout(cct) << value << std::endl; | |
13633 | ||
f67539c2 TL |
13634 | std::scoped_lock lock(client_lock); |
13635 | ||
11fdf7f2 | 13636 | if (!fuse_default_permissions) { |
7c673cae FG |
13637 | int r = may_create(parent, perms); |
13638 | if (r < 0) | |
13639 | return r; | |
13640 | } | |
13641 | ||
13642 | InodeRef in; | |
f67539c2 | 13643 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13644 | if (r == 0) { |
13645 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13646 | _ll_get(in.get()); | |
13647 | } | |
13648 | tout(cct) << stx->stx_ino << std::endl; | |
13649 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name | |
13650 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13651 | *out = in.get(); | |
13652 | return r; | |
13653 | } | |
13654 | ||
13655 | int Client::_unlink(Inode *dir, const char *name, const UserPerm& perm) | |
13656 | { | |
1adf2230 | 13657 | ldout(cct, 8) << "_unlink(" << dir->ino << " " << name |
7c673cae FG |
13658 | << " uid " << perm.uid() << " gid " << perm.gid() |
13659 | << ")" << dendl; | |
13660 | ||
13661 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13662 | return -CEPHFS_EROFS; |
7c673cae FG |
13663 | } |
13664 | ||
13665 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_UNLINK); | |
13666 | ||
13667 | filepath path; | |
13668 | dir->make_nosnap_relative_path(path); | |
13669 | path.push_dentry(name); | |
13670 | req->set_filepath(path); | |
13671 | ||
13672 | InodeRef otherin; | |
b32b8144 | 13673 | Inode *in; |
7c673cae | 13674 | Dentry *de; |
b32b8144 | 13675 | |
7c673cae FG |
13676 | int res = get_or_create(dir, name, &de); |
13677 | if (res < 0) | |
13678 | goto fail; | |
13679 | req->set_dentry(de); | |
13680 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13681 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13682 | ||
13683 | res = _lookup(dir, name, 0, &otherin, perm); | |
13684 | if (res < 0) | |
13685 | goto fail; | |
b32b8144 FG |
13686 | |
13687 | in = otherin.get(); | |
13688 | req->set_other_inode(in); | |
13689 | in->break_all_delegs(); | |
7c673cae FG |
13690 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
13691 | ||
13692 | req->set_inode(dir); | |
13693 | ||
13694 | res = make_request(req, perm); | |
13695 | ||
13696 | trim_cache(); | |
1adf2230 | 13697 | ldout(cct, 8) << "unlink(" << path << ") = " << res << dendl; |
7c673cae FG |
13698 | return res; |
13699 | ||
13700 | fail: | |
13701 | put_request(req); | |
13702 | return res; | |
13703 | } | |
13704 | ||
13705 | int Client::ll_unlink(Inode *in, const char *name, const UserPerm& perm) | |
13706 | { | |
f67539c2 TL |
13707 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13708 | if (!mref_reader.is_state_satisfied()) | |
13709 | return -CEPHFS_ENOTCONN; | |
181888fb | 13710 | |
7c673cae FG |
13711 | vinodeno_t vino = _get_vino(in); |
13712 | ||
13713 | ldout(cct, 3) << "ll_unlink " << vino << " " << name << dendl; | |
13714 | tout(cct) << "ll_unlink" << std::endl; | |
13715 | tout(cct) << vino.ino.val << std::endl; | |
13716 | tout(cct) << name << std::endl; | |
13717 | ||
f67539c2 TL |
13718 | std::scoped_lock lock(client_lock); |
13719 | ||
11fdf7f2 | 13720 | if (!fuse_default_permissions) { |
7c673cae FG |
13721 | int r = may_delete(in, name, perm); |
13722 | if (r < 0) | |
13723 | return r; | |
13724 | } | |
13725 | return _unlink(in, name, perm); | |
13726 | } | |
13727 | ||
13728 | int Client::_rmdir(Inode *dir, const char *name, const UserPerm& perms) | |
13729 | { | |
1adf2230 | 13730 | ldout(cct, 8) << "_rmdir(" << dir->ino << " " << name << " uid " |
7c673cae FG |
13731 | << perms.uid() << " gid " << perms.gid() << ")" << dendl; |
13732 | ||
13733 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13734 | return -CEPHFS_EROFS; |
7c673cae | 13735 | } |
b32b8144 FG |
13736 | |
13737 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_RMSNAP : CEPH_MDS_OP_RMDIR; | |
13738 | MetaRequest *req = new MetaRequest(op); | |
7c673cae FG |
13739 | filepath path; |
13740 | dir->make_nosnap_relative_path(path); | |
13741 | path.push_dentry(name); | |
13742 | req->set_filepath(path); | |
11fdf7f2 | 13743 | req->set_inode(dir); |
7c673cae FG |
13744 | |
13745 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13746 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13747 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
13748 | ||
13749 | InodeRef in; | |
13750 | ||
13751 | Dentry *de; | |
13752 | int res = get_or_create(dir, name, &de); | |
13753 | if (res < 0) | |
13754 | goto fail; | |
b32b8144 FG |
13755 | if (op == CEPH_MDS_OP_RMDIR) |
13756 | req->set_dentry(de); | |
13757 | else | |
13758 | de->get(); | |
13759 | ||
7c673cae FG |
13760 | res = _lookup(dir, name, 0, &in, perms); |
13761 | if (res < 0) | |
13762 | goto fail; | |
11fdf7f2 TL |
13763 | |
13764 | if (op == CEPH_MDS_OP_RMSNAP) { | |
7c673cae | 13765 | unlink(de, true, true); |
b32b8144 | 13766 | de->put(); |
7c673cae | 13767 | } |
11fdf7f2 | 13768 | req->set_other_inode(in.get()); |
7c673cae FG |
13769 | |
13770 | res = make_request(req, perms); | |
13771 | ||
13772 | trim_cache(); | |
1adf2230 | 13773 | ldout(cct, 8) << "rmdir(" << path << ") = " << res << dendl; |
7c673cae FG |
13774 | return res; |
13775 | ||
13776 | fail: | |
13777 | put_request(req); | |
13778 | return res; | |
13779 | } | |
13780 | ||
13781 | int Client::ll_rmdir(Inode *in, const char *name, const UserPerm& perms) | |
13782 | { | |
f67539c2 TL |
13783 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13784 | if (!mref_reader.is_state_satisfied()) | |
13785 | return -CEPHFS_ENOTCONN; | |
181888fb | 13786 | |
7c673cae FG |
13787 | vinodeno_t vino = _get_vino(in); |
13788 | ||
13789 | ldout(cct, 3) << "ll_rmdir " << vino << " " << name << dendl; | |
13790 | tout(cct) << "ll_rmdir" << std::endl; | |
13791 | tout(cct) << vino.ino.val << std::endl; | |
13792 | tout(cct) << name << std::endl; | |
13793 | ||
f67539c2 TL |
13794 | std::scoped_lock lock(client_lock); |
13795 | ||
11fdf7f2 | 13796 | if (!fuse_default_permissions) { |
7c673cae FG |
13797 | int r = may_delete(in, name, perms); |
13798 | if (r < 0) | |
13799 | return r; | |
13800 | } | |
13801 | ||
13802 | return _rmdir(in, name, perms); | |
13803 | } | |
13804 | ||
f67539c2 | 13805 | int Client::_rename(Inode *fromdir, const char *fromname, Inode *todir, const char *toname, const UserPerm& perm, std::string alternate_name) |
7c673cae | 13806 | { |
1adf2230 | 13807 | ldout(cct, 8) << "_rename(" << fromdir->ino << " " << fromname << " to " |
7c673cae FG |
13808 | << todir->ino << " " << toname |
13809 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" | |
13810 | << dendl; | |
13811 | ||
13812 | if (fromdir->snapid != todir->snapid) | |
f67539c2 | 13813 | return -CEPHFS_EXDEV; |
7c673cae FG |
13814 | |
13815 | int op = CEPH_MDS_OP_RENAME; | |
13816 | if (fromdir->snapid != CEPH_NOSNAP) { | |
13817 | if (fromdir == todir && fromdir->snapid == CEPH_SNAPDIR) | |
13818 | op = CEPH_MDS_OP_RENAMESNAP; | |
13819 | else | |
f67539c2 TL |
13820 | return -CEPHFS_EROFS; |
13821 | } | |
13822 | if (fromdir != todir) { | |
13823 | Inode *fromdir_root = | |
13824 | fromdir->quota.is_enable() ? fromdir : get_quota_root(fromdir, perm); | |
13825 | Inode *todir_root = | |
13826 | todir->quota.is_enable() ? todir : get_quota_root(todir, perm); | |
13827 | if (fromdir_root != todir_root) { | |
13828 | return -CEPHFS_EXDEV; | |
13829 | } | |
7c673cae | 13830 | } |
7c673cae FG |
13831 | |
13832 | InodeRef target; | |
13833 | MetaRequest *req = new MetaRequest(op); | |
13834 | ||
13835 | filepath from; | |
13836 | fromdir->make_nosnap_relative_path(from); | |
13837 | from.push_dentry(fromname); | |
13838 | filepath to; | |
13839 | todir->make_nosnap_relative_path(to); | |
13840 | to.push_dentry(toname); | |
13841 | req->set_filepath(to); | |
13842 | req->set_filepath2(from); | |
f67539c2 | 13843 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13844 | |
13845 | Dentry *oldde; | |
13846 | int res = get_or_create(fromdir, fromname, &oldde); | |
13847 | if (res < 0) | |
13848 | goto fail; | |
13849 | Dentry *de; | |
13850 | res = get_or_create(todir, toname, &de); | |
13851 | if (res < 0) | |
13852 | goto fail; | |
13853 | ||
13854 | if (op == CEPH_MDS_OP_RENAME) { | |
13855 | req->set_old_dentry(oldde); | |
13856 | req->old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
13857 | req->old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
13858 | ||
13859 | req->set_dentry(de); | |
13860 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13861 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13862 | ||
13863 | InodeRef oldin, otherin; | |
f67539c2 | 13864 | res = _lookup(fromdir, fromname, 0, &oldin, perm); |
7c673cae FG |
13865 | if (res < 0) |
13866 | goto fail; | |
b32b8144 FG |
13867 | |
13868 | Inode *oldinode = oldin.get(); | |
13869 | oldinode->break_all_delegs(); | |
13870 | req->set_old_inode(oldinode); | |
7c673cae FG |
13871 | req->old_inode_drop = CEPH_CAP_LINK_SHARED; |
13872 | ||
13873 | res = _lookup(todir, toname, 0, &otherin, perm); | |
b32b8144 FG |
13874 | switch (res) { |
13875 | case 0: | |
13876 | { | |
13877 | Inode *in = otherin.get(); | |
13878 | req->set_other_inode(in); | |
13879 | in->break_all_delegs(); | |
13880 | } | |
7c673cae | 13881 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
b32b8144 | 13882 | break; |
f67539c2 | 13883 | case -CEPHFS_ENOENT: |
b32b8144 FG |
13884 | break; |
13885 | default: | |
13886 | goto fail; | |
7c673cae FG |
13887 | } |
13888 | ||
13889 | req->set_inode(todir); | |
13890 | } else { | |
13891 | // renamesnap reply contains no tracedn, so we need to invalidate | |
13892 | // dentry manually | |
13893 | unlink(oldde, true, true); | |
13894 | unlink(de, true, true); | |
11fdf7f2 TL |
13895 | |
13896 | req->set_inode(todir); | |
7c673cae FG |
13897 | } |
13898 | ||
13899 | res = make_request(req, perm, &target); | |
13900 | ldout(cct, 10) << "rename result is " << res << dendl; | |
13901 | ||
13902 | // renamed item from our cache | |
13903 | ||
13904 | trim_cache(); | |
1adf2230 | 13905 | ldout(cct, 8) << "_rename(" << from << ", " << to << ") = " << res << dendl; |
7c673cae FG |
13906 | return res; |
13907 | ||
13908 | fail: | |
13909 | put_request(req); | |
13910 | return res; | |
13911 | } | |
13912 | ||
13913 | int Client::ll_rename(Inode *parent, const char *name, Inode *newparent, | |
13914 | const char *newname, const UserPerm& perm) | |
13915 | { | |
f67539c2 TL |
13916 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13917 | if (!mref_reader.is_state_satisfied()) | |
13918 | return -CEPHFS_ENOTCONN; | |
181888fb | 13919 | |
7c673cae FG |
13920 | vinodeno_t vparent = _get_vino(parent); |
13921 | vinodeno_t vnewparent = _get_vino(newparent); | |
13922 | ||
13923 | ldout(cct, 3) << "ll_rename " << vparent << " " << name << " to " | |
13924 | << vnewparent << " " << newname << dendl; | |
13925 | tout(cct) << "ll_rename" << std::endl; | |
13926 | tout(cct) << vparent.ino.val << std::endl; | |
13927 | tout(cct) << name << std::endl; | |
13928 | tout(cct) << vnewparent.ino.val << std::endl; | |
13929 | tout(cct) << newname << std::endl; | |
13930 | ||
f67539c2 TL |
13931 | std::scoped_lock lock(client_lock); |
13932 | ||
11fdf7f2 | 13933 | if (!fuse_default_permissions) { |
7c673cae FG |
13934 | int r = may_delete(parent, name, perm); |
13935 | if (r < 0) | |
13936 | return r; | |
13937 | r = may_delete(newparent, newname, perm); | |
f67539c2 | 13938 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
13939 | return r; |
13940 | } | |
13941 | ||
f67539c2 | 13942 | return _rename(parent, name, newparent, newname, perm, ""); |
7c673cae FG |
13943 | } |
13944 | ||
f67539c2 | 13945 | int Client::_link(Inode *in, Inode *dir, const char *newname, const UserPerm& perm, std::string alternate_name, InodeRef *inp) |
7c673cae | 13946 | { |
1adf2230 | 13947 | ldout(cct, 8) << "_link(" << in->ino << " to " << dir->ino << " " << newname |
7c673cae FG |
13948 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" << dendl; |
13949 | ||
13950 | if (strlen(newname) > NAME_MAX) | |
f67539c2 | 13951 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13952 | |
13953 | if (in->snapid != CEPH_NOSNAP || dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13954 | return -CEPHFS_EROFS; |
7c673cae FG |
13955 | } |
13956 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 13957 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13958 | } |
13959 | ||
b32b8144 | 13960 | in->break_all_delegs(); |
7c673cae FG |
13961 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LINK); |
13962 | ||
13963 | filepath path(newname, dir->ino); | |
13964 | req->set_filepath(path); | |
f67539c2 | 13965 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13966 | filepath existing(in->ino); |
13967 | req->set_filepath2(existing); | |
13968 | ||
13969 | req->set_inode(dir); | |
13970 | req->inode_drop = CEPH_CAP_FILE_SHARED; | |
13971 | req->inode_unless = CEPH_CAP_FILE_EXCL; | |
13972 | ||
13973 | Dentry *de; | |
13974 | int res = get_or_create(dir, newname, &de); | |
13975 | if (res < 0) | |
13976 | goto fail; | |
13977 | req->set_dentry(de); | |
13978 | ||
13979 | res = make_request(req, perm, inp); | |
13980 | ldout(cct, 10) << "link result is " << res << dendl; | |
13981 | ||
13982 | trim_cache(); | |
1adf2230 | 13983 | ldout(cct, 8) << "link(" << existing << ", " << path << ") = " << res << dendl; |
7c673cae FG |
13984 | return res; |
13985 | ||
13986 | fail: | |
13987 | put_request(req); | |
13988 | return res; | |
13989 | } | |
13990 | ||
13991 | int Client::ll_link(Inode *in, Inode *newparent, const char *newname, | |
13992 | const UserPerm& perm) | |
13993 | { | |
f67539c2 TL |
13994 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13995 | if (!mref_reader.is_state_satisfied()) | |
13996 | return -CEPHFS_ENOTCONN; | |
181888fb | 13997 | |
7c673cae FG |
13998 | vinodeno_t vino = _get_vino(in); |
13999 | vinodeno_t vnewparent = _get_vino(newparent); | |
14000 | ||
31f18b77 | 14001 | ldout(cct, 3) << "ll_link " << vino << " to " << vnewparent << " " << |
7c673cae FG |
14002 | newname << dendl; |
14003 | tout(cct) << "ll_link" << std::endl; | |
14004 | tout(cct) << vino.ino.val << std::endl; | |
14005 | tout(cct) << vnewparent << std::endl; | |
14006 | tout(cct) << newname << std::endl; | |
14007 | ||
7c673cae FG |
14008 | InodeRef target; |
14009 | ||
f67539c2 TL |
14010 | std::scoped_lock lock(client_lock); |
14011 | ||
11fdf7f2 | 14012 | if (!fuse_default_permissions) { |
7c673cae | 14013 | if (S_ISDIR(in->mode)) |
f67539c2 | 14014 | return -CEPHFS_EPERM; |
7c673cae | 14015 | |
11fdf7f2 | 14016 | int r = may_hardlink(in, perm); |
7c673cae FG |
14017 | if (r < 0) |
14018 | return r; | |
14019 | ||
14020 | r = may_create(newparent, perm); | |
14021 | if (r < 0) | |
14022 | return r; | |
14023 | } | |
14024 | ||
f67539c2 | 14025 | return _link(in, newparent, newname, perm, "", &target); |
7c673cae FG |
14026 | } |
14027 | ||
14028 | int Client::ll_num_osds(void) | |
14029 | { | |
f67539c2 | 14030 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14031 | return objecter->with_osdmap(std::mem_fn(&OSDMap::get_num_osds)); |
14032 | } | |
14033 | ||
14034 | int Client::ll_osdaddr(int osd, uint32_t *addr) | |
14035 | { | |
f67539c2 | 14036 | std::scoped_lock lock(client_lock); |
181888fb | 14037 | |
7c673cae FG |
14038 | entity_addr_t g; |
14039 | bool exists = objecter->with_osdmap([&](const OSDMap& o) { | |
14040 | if (!o.exists(osd)) | |
14041 | return false; | |
11fdf7f2 | 14042 | g = o.get_addrs(osd).front(); |
7c673cae FG |
14043 | return true; |
14044 | }); | |
14045 | if (!exists) | |
14046 | return -1; | |
14047 | uint32_t nb_addr = (g.in4_addr()).sin_addr.s_addr; | |
14048 | *addr = ntohl(nb_addr); | |
14049 | return 0; | |
14050 | } | |
181888fb | 14051 | |
7c673cae FG |
14052 | uint32_t Client::ll_stripe_unit(Inode *in) |
14053 | { | |
f67539c2 | 14054 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14055 | return in->layout.stripe_unit; |
14056 | } | |
14057 | ||
14058 | uint64_t Client::ll_snap_seq(Inode *in) | |
14059 | { | |
f67539c2 | 14060 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14061 | return in->snaprealm->seq; |
14062 | } | |
14063 | ||
14064 | int Client::ll_file_layout(Inode *in, file_layout_t *layout) | |
14065 | { | |
f67539c2 | 14066 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14067 | *layout = in->layout; |
14068 | return 0; | |
14069 | } | |
14070 | ||
14071 | int Client::ll_file_layout(Fh *fh, file_layout_t *layout) | |
14072 | { | |
14073 | return ll_file_layout(fh->inode.get(), layout); | |
14074 | } | |
14075 | ||
14076 | /* Currently we cannot take advantage of redundancy in reads, since we | |
14077 | would have to go through all possible placement groups (a | |
14078 | potentially quite large number determined by a hash), and use CRUSH | |
14079 | to calculate the appropriate set of OSDs for each placement group, | |
14080 | then index into that. An array with one entry per OSD is much more | |
14081 | tractable and works for demonstration purposes. */ | |
14082 | ||
14083 | int Client::ll_get_stripe_osd(Inode *in, uint64_t blockno, | |
14084 | file_layout_t* layout) | |
14085 | { | |
f67539c2 | 14086 | std::scoped_lock lock(client_lock); |
181888fb | 14087 | |
28e407b8 | 14088 | inodeno_t ino = in->ino; |
7c673cae FG |
14089 | uint32_t object_size = layout->object_size; |
14090 | uint32_t su = layout->stripe_unit; | |
14091 | uint32_t stripe_count = layout->stripe_count; | |
14092 | uint64_t stripes_per_object = object_size / su; | |
11fdf7f2 | 14093 | uint64_t stripeno = 0, stripepos = 0; |
7c673cae | 14094 | |
11fdf7f2 TL |
14095 | if(stripe_count) { |
14096 | stripeno = blockno / stripe_count; // which horizontal stripe (Y) | |
14097 | stripepos = blockno % stripe_count; // which object in the object set (X) | |
14098 | } | |
7c673cae FG |
14099 | uint64_t objectsetno = stripeno / stripes_per_object; // which object set |
14100 | uint64_t objectno = objectsetno * stripe_count + stripepos; // object id | |
14101 | ||
14102 | object_t oid = file_object_t(ino, objectno); | |
14103 | return objecter->with_osdmap([&](const OSDMap& o) { | |
14104 | ceph_object_layout olayout = | |
14105 | o.file_to_object_layout(oid, *layout); | |
14106 | pg_t pg = (pg_t)olayout.ol_pgid; | |
14107 | vector<int> osds; | |
14108 | int primary; | |
14109 | o.pg_to_acting_osds(pg, &osds, &primary); | |
14110 | return primary; | |
14111 | }); | |
14112 | } | |
14113 | ||
14114 | /* Return the offset of the block, internal to the object */ | |
14115 | ||
14116 | uint64_t Client::ll_get_internal_offset(Inode *in, uint64_t blockno) | |
14117 | { | |
f67539c2 | 14118 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14119 | file_layout_t *layout=&(in->layout); |
14120 | uint32_t object_size = layout->object_size; | |
14121 | uint32_t su = layout->stripe_unit; | |
14122 | uint64_t stripes_per_object = object_size / su; | |
14123 | ||
14124 | return (blockno % stripes_per_object) * su; | |
14125 | } | |
14126 | ||
14127 | int Client::ll_opendir(Inode *in, int flags, dir_result_t** dirpp, | |
14128 | const UserPerm& perms) | |
14129 | { | |
f67539c2 TL |
14130 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14131 | if (!mref_reader.is_state_satisfied()) | |
14132 | return -CEPHFS_ENOTCONN; | |
181888fb | 14133 | |
7c673cae FG |
14134 | vinodeno_t vino = _get_vino(in); |
14135 | ||
14136 | ldout(cct, 3) << "ll_opendir " << vino << dendl; | |
14137 | tout(cct) << "ll_opendir" << std::endl; | |
14138 | tout(cct) << vino.ino.val << std::endl; | |
14139 | ||
f67539c2 TL |
14140 | std::scoped_lock lock(client_lock); |
14141 | ||
11fdf7f2 | 14142 | if (!fuse_default_permissions) { |
7c673cae FG |
14143 | int r = may_open(in, flags, perms); |
14144 | if (r < 0) | |
14145 | return r; | |
14146 | } | |
14147 | ||
14148 | int r = _opendir(in, dirpp, perms); | |
f67539c2 | 14149 | tout(cct) << (uintptr_t)*dirpp << std::endl; |
7c673cae FG |
14150 | |
14151 | ldout(cct, 3) << "ll_opendir " << vino << " = " << r << " (" << *dirpp << ")" | |
14152 | << dendl; | |
14153 | return r; | |
14154 | } | |
14155 | ||
14156 | int Client::ll_releasedir(dir_result_t *dirp) | |
14157 | { | |
f67539c2 TL |
14158 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14159 | if (!mref_reader.is_state_satisfied()) | |
14160 | return -CEPHFS_ENOTCONN; | |
14161 | ||
7c673cae FG |
14162 | ldout(cct, 3) << "ll_releasedir " << dirp << dendl; |
14163 | tout(cct) << "ll_releasedir" << std::endl; | |
f67539c2 | 14164 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 14165 | |
f67539c2 | 14166 | std::scoped_lock lock(client_lock); |
181888fb | 14167 | |
7c673cae FG |
14168 | _closedir(dirp); |
14169 | return 0; | |
14170 | } | |
14171 | ||
14172 | int Client::ll_fsyncdir(dir_result_t *dirp) | |
14173 | { | |
f67539c2 TL |
14174 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14175 | if (!mref_reader.is_state_satisfied()) | |
14176 | return -CEPHFS_ENOTCONN; | |
14177 | ||
7c673cae FG |
14178 | ldout(cct, 3) << "ll_fsyncdir " << dirp << dendl; |
14179 | tout(cct) << "ll_fsyncdir" << std::endl; | |
f67539c2 | 14180 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 14181 | |
f67539c2 | 14182 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14183 | return _fsync(dirp->inode.get(), false); |
14184 | } | |
14185 | ||
14186 | int Client::ll_open(Inode *in, int flags, Fh **fhp, const UserPerm& perms) | |
14187 | { | |
11fdf7f2 | 14188 | ceph_assert(!(flags & O_CREAT)); |
7c673cae | 14189 | |
f67539c2 TL |
14190 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14191 | if (!mref_reader.is_state_satisfied()) | |
14192 | return -CEPHFS_ENOTCONN; | |
181888fb | 14193 | |
7c673cae FG |
14194 | vinodeno_t vino = _get_vino(in); |
14195 | ||
14196 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << dendl; | |
14197 | tout(cct) << "ll_open" << std::endl; | |
14198 | tout(cct) << vino.ino.val << std::endl; | |
14199 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
14200 | ||
f67539c2 TL |
14201 | std::scoped_lock lock(client_lock); |
14202 | ||
7c673cae | 14203 | int r; |
11fdf7f2 | 14204 | if (!fuse_default_permissions) { |
7c673cae FG |
14205 | r = may_open(in, flags, perms); |
14206 | if (r < 0) | |
14207 | goto out; | |
14208 | } | |
14209 | ||
14210 | r = _open(in, flags, 0, fhp /* may be NULL */, perms); | |
14211 | ||
14212 | out: | |
14213 | Fh *fhptr = fhp ? *fhp : NULL; | |
14214 | if (fhptr) { | |
14215 | ll_unclosed_fh_set.insert(fhptr); | |
14216 | } | |
f67539c2 | 14217 | tout(cct) << (uintptr_t)fhptr << std::endl; |
7c673cae FG |
14218 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << |
14219 | " = " << r << " (" << fhptr << ")" << dendl; | |
14220 | return r; | |
14221 | } | |
14222 | ||
14223 | int Client::_ll_create(Inode *parent, const char *name, mode_t mode, | |
14224 | int flags, InodeRef *in, int caps, Fh **fhp, | |
14225 | const UserPerm& perms) | |
14226 | { | |
14227 | *fhp = NULL; | |
14228 | ||
14229 | vinodeno_t vparent = _get_vino(parent); | |
14230 | ||
1adf2230 | 14231 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14232 | mode << dec << " " << ceph_flags_sys2wire(flags) << ", uid " << perms.uid() |
14233 | << ", gid " << perms.gid() << dendl; | |
14234 | tout(cct) << "ll_create" << std::endl; | |
14235 | tout(cct) << vparent.ino.val << std::endl; | |
14236 | tout(cct) << name << std::endl; | |
14237 | tout(cct) << mode << std::endl; | |
14238 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
14239 | ||
14240 | bool created = false; | |
14241 | int r = _lookup(parent, name, caps, in, perms); | |
14242 | ||
14243 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
f67539c2 | 14244 | return -CEPHFS_EEXIST; |
7c673cae | 14245 | |
f67539c2 | 14246 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
11fdf7f2 | 14247 | if (!fuse_default_permissions) { |
7c673cae FG |
14248 | r = may_create(parent, perms); |
14249 | if (r < 0) | |
14250 | goto out; | |
14251 | } | |
14252 | r = _create(parent, name, flags, mode, in, fhp, 0, 0, 0, NULL, &created, | |
f67539c2 | 14253 | perms, ""); |
7c673cae FG |
14254 | if (r < 0) |
14255 | goto out; | |
14256 | } | |
14257 | ||
14258 | if (r < 0) | |
14259 | goto out; | |
14260 | ||
11fdf7f2 | 14261 | ceph_assert(*in); |
7c673cae FG |
14262 | |
14263 | ldout(cct, 20) << "_ll_create created = " << created << dendl; | |
14264 | if (!created) { | |
11fdf7f2 | 14265 | if (!fuse_default_permissions) { |
7c673cae FG |
14266 | r = may_open(in->get(), flags, perms); |
14267 | if (r < 0) { | |
14268 | if (*fhp) { | |
14269 | int release_r = _release_fh(*fhp); | |
11fdf7f2 | 14270 | ceph_assert(release_r == 0); // during create, no async data ops should have happened |
7c673cae FG |
14271 | } |
14272 | goto out; | |
14273 | } | |
14274 | } | |
14275 | if (*fhp == NULL) { | |
14276 | r = _open(in->get(), flags, mode, fhp, perms); | |
14277 | if (r < 0) | |
14278 | goto out; | |
14279 | } | |
14280 | } | |
14281 | ||
14282 | out: | |
14283 | if (*fhp) { | |
14284 | ll_unclosed_fh_set.insert(*fhp); | |
14285 | } | |
14286 | ||
14287 | ino_t ino = 0; | |
14288 | if (r >= 0) { | |
14289 | Inode *inode = in->get(); | |
14290 | if (use_faked_inos()) | |
14291 | ino = inode->faked_ino; | |
14292 | else | |
14293 | ino = inode->ino; | |
14294 | } | |
14295 | ||
f67539c2 | 14296 | tout(cct) << (uintptr_t)*fhp << std::endl; |
7c673cae | 14297 | tout(cct) << ino << std::endl; |
1adf2230 | 14298 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14299 | mode << dec << " " << ceph_flags_sys2wire(flags) << " = " << r << " (" << |
14300 | *fhp << " " << hex << ino << dec << ")" << dendl; | |
14301 | ||
14302 | return r; | |
14303 | } | |
14304 | ||
14305 | int Client::ll_create(Inode *parent, const char *name, mode_t mode, | |
14306 | int flags, struct stat *attr, Inode **outp, Fh **fhp, | |
14307 | const UserPerm& perms) | |
14308 | { | |
f67539c2 TL |
14309 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14310 | if (!mref_reader.is_state_satisfied()) | |
14311 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14312 | |
f67539c2 TL |
14313 | std::scoped_lock lock(client_lock); |
14314 | InodeRef in; | |
181888fb | 14315 | |
7c673cae FG |
14316 | int r = _ll_create(parent, name, mode, flags, &in, CEPH_STAT_CAP_INODE_ALL, |
14317 | fhp, perms); | |
14318 | if (r >= 0) { | |
11fdf7f2 | 14319 | ceph_assert(in); |
7c673cae FG |
14320 | |
14321 | // passing an Inode in outp requires an additional ref | |
14322 | if (outp) { | |
14323 | _ll_get(in.get()); | |
14324 | *outp = in.get(); | |
14325 | } | |
14326 | fill_stat(in, attr); | |
14327 | } else { | |
14328 | attr->st_ino = 0; | |
14329 | } | |
14330 | ||
14331 | return r; | |
14332 | } | |
14333 | ||
14334 | int Client::ll_createx(Inode *parent, const char *name, mode_t mode, | |
14335 | int oflags, Inode **outp, Fh **fhp, | |
14336 | struct ceph_statx *stx, unsigned want, unsigned lflags, | |
14337 | const UserPerm& perms) | |
14338 | { | |
14339 | unsigned caps = statx_to_mask(lflags, want); | |
f67539c2 TL |
14340 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14341 | if (!mref_reader.is_state_satisfied()) | |
14342 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14343 | |
f67539c2 TL |
14344 | std::scoped_lock lock(client_lock); |
14345 | InodeRef in; | |
7c673cae FG |
14346 | |
14347 | int r = _ll_create(parent, name, mode, oflags, &in, caps, fhp, perms); | |
14348 | if (r >= 0) { | |
11fdf7f2 | 14349 | ceph_assert(in); |
7c673cae FG |
14350 | |
14351 | // passing an Inode in outp requires an additional ref | |
14352 | if (outp) { | |
14353 | _ll_get(in.get()); | |
14354 | *outp = in.get(); | |
14355 | } | |
14356 | fill_statx(in, caps, stx); | |
14357 | } else { | |
14358 | stx->stx_ino = 0; | |
14359 | stx->stx_mask = 0; | |
14360 | } | |
14361 | ||
14362 | return r; | |
14363 | } | |
14364 | ||
14365 | loff_t Client::ll_lseek(Fh *fh, loff_t offset, int whence) | |
14366 | { | |
f67539c2 TL |
14367 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14368 | if (!mref_reader.is_state_satisfied()) | |
14369 | return -CEPHFS_ENOTCONN; | |
14370 | ||
7c673cae FG |
14371 | tout(cct) << "ll_lseek" << std::endl; |
14372 | tout(cct) << offset << std::endl; | |
14373 | tout(cct) << whence << std::endl; | |
14374 | ||
f67539c2 | 14375 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14376 | return _lseek(fh, offset, whence); |
14377 | } | |
14378 | ||
14379 | int Client::ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl) | |
14380 | { | |
f67539c2 TL |
14381 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14382 | if (!mref_reader.is_state_satisfied()) | |
14383 | return -CEPHFS_ENOTCONN; | |
14384 | ||
7c673cae FG |
14385 | ldout(cct, 3) << "ll_read " << fh << " " << fh->inode->ino << " " << " " << off << "~" << len << dendl; |
14386 | tout(cct) << "ll_read" << std::endl; | |
f67539c2 | 14387 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14388 | tout(cct) << off << std::endl; |
14389 | tout(cct) << len << std::endl; | |
14390 | ||
11fdf7f2 TL |
14391 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14392 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14393 | std::scoped_lock lock(client_lock); |
14394 | ||
f6b5b4d7 TL |
14395 | int r = _read(fh, off, len, bl); |
14396 | ldout(cct, 3) << "ll_read " << fh << " " << off << "~" << len << " = " << r | |
14397 | << dendl; | |
14398 | return r; | |
7c673cae FG |
14399 | } |
14400 | ||
14401 | int Client::ll_read_block(Inode *in, uint64_t blockid, | |
14402 | char *buf, | |
14403 | uint64_t offset, | |
14404 | uint64_t length, | |
14405 | file_layout_t* layout) | |
14406 | { | |
f67539c2 TL |
14407 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14408 | if (!mref_reader.is_state_satisfied()) | |
14409 | return -CEPHFS_ENOTCONN; | |
181888fb | 14410 | |
b32b8144 | 14411 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14412 | object_t oid = file_object_t(vino.ino, blockid); |
14413 | C_SaferCond onfinish; | |
14414 | bufferlist bl; | |
14415 | ||
14416 | objecter->read(oid, | |
14417 | object_locator_t(layout->pool_id), | |
14418 | offset, | |
14419 | length, | |
14420 | vino.snapid, | |
14421 | &bl, | |
14422 | CEPH_OSD_FLAG_READ, | |
14423 | &onfinish); | |
14424 | ||
7c673cae | 14425 | int r = onfinish.wait(); |
7c673cae | 14426 | if (r >= 0) { |
9f95a23c | 14427 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
14428 | r = bl.length(); |
14429 | } | |
14430 | ||
14431 | return r; | |
14432 | } | |
14433 | ||
14434 | /* It appears that the OSD doesn't return success unless the entire | |
14435 | buffer was written, return the write length on success. */ | |
14436 | ||
14437 | int Client::ll_write_block(Inode *in, uint64_t blockid, | |
14438 | char* buf, uint64_t offset, | |
14439 | uint64_t length, file_layout_t* layout, | |
14440 | uint64_t snapseq, uint32_t sync) | |
14441 | { | |
7c673cae | 14442 | vinodeno_t vino = ll_get_vino(in); |
7c673cae | 14443 | int r = 0; |
11fdf7f2 | 14444 | std::unique_ptr<C_SaferCond> onsafe = nullptr; |
f67539c2 TL |
14445 | |
14446 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
14447 | if (!mref_reader.is_state_satisfied()) | |
14448 | return -CEPHFS_ENOTCONN; | |
14449 | ||
7c673cae | 14450 | if (length == 0) { |
f67539c2 | 14451 | return -CEPHFS_EINVAL; |
7c673cae FG |
14452 | } |
14453 | if (true || sync) { | |
14454 | /* if write is stable, the epilogue is waiting on | |
14455 | * flock */ | |
11fdf7f2 | 14456 | onsafe.reset(new C_SaferCond("Client::ll_write_block flock")); |
7c673cae FG |
14457 | } |
14458 | object_t oid = file_object_t(vino.ino, blockid); | |
14459 | SnapContext fakesnap; | |
11fdf7f2 TL |
14460 | ceph::bufferlist bl; |
14461 | if (length > 0) { | |
14462 | bl.push_back(buffer::copy(buf, length)); | |
14463 | } | |
7c673cae FG |
14464 | |
14465 | ldout(cct, 1) << "ll_block_write for " << vino.ino << "." << blockid | |
14466 | << dendl; | |
14467 | ||
14468 | fakesnap.seq = snapseq; | |
14469 | ||
14470 | /* lock just in time */ | |
7c673cae FG |
14471 | objecter->write(oid, |
14472 | object_locator_t(layout->pool_id), | |
14473 | offset, | |
14474 | length, | |
14475 | fakesnap, | |
14476 | bl, | |
14477 | ceph::real_clock::now(), | |
14478 | 0, | |
11fdf7f2 | 14479 | onsafe.get()); |
7c673cae | 14480 | |
11fdf7f2 TL |
14481 | if (nullptr != onsafe) { |
14482 | r = onsafe->wait(); | |
7c673cae FG |
14483 | } |
14484 | ||
14485 | if (r < 0) { | |
14486 | return r; | |
14487 | } else { | |
14488 | return length; | |
14489 | } | |
14490 | } | |
14491 | ||
14492 | int Client::ll_commit_blocks(Inode *in, | |
14493 | uint64_t offset, | |
14494 | uint64_t length) | |
14495 | { | |
7c673cae FG |
14496 | /* |
14497 | BarrierContext *bctx; | |
b32b8144 | 14498 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14499 | uint64_t ino = vino.ino; |
14500 | ||
14501 | ldout(cct, 1) << "ll_commit_blocks for " << vino.ino << " from " | |
14502 | << offset << " to " << length << dendl; | |
14503 | ||
14504 | if (length == 0) { | |
f67539c2 | 14505 | return -CEPHFS_EINVAL; |
7c673cae FG |
14506 | } |
14507 | ||
f67539c2 | 14508 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14509 | map<uint64_t, BarrierContext*>::iterator p = barriers.find(ino); |
14510 | if (p != barriers.end()) { | |
14511 | barrier_interval civ(offset, offset + length); | |
14512 | p->second->commit_barrier(civ); | |
14513 | } | |
14514 | */ | |
14515 | return 0; | |
14516 | } | |
14517 | ||
14518 | int Client::ll_write(Fh *fh, loff_t off, loff_t len, const char *data) | |
14519 | { | |
7c673cae FG |
14520 | ldout(cct, 3) << "ll_write " << fh << " " << fh->inode->ino << " " << off << |
14521 | "~" << len << dendl; | |
14522 | tout(cct) << "ll_write" << std::endl; | |
f67539c2 | 14523 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14524 | tout(cct) << off << std::endl; |
14525 | tout(cct) << len << std::endl; | |
14526 | ||
f67539c2 TL |
14527 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14528 | if (!mref_reader.is_state_satisfied()) | |
14529 | return -CEPHFS_ENOTCONN; | |
181888fb | 14530 | |
11fdf7f2 TL |
14531 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14532 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14533 | std::scoped_lock lock(client_lock); |
14534 | ||
7c673cae FG |
14535 | int r = _write(fh, off, len, data, NULL, 0); |
14536 | ldout(cct, 3) << "ll_write " << fh << " " << off << "~" << len << " = " << r | |
14537 | << dendl; | |
14538 | return r; | |
14539 | } | |
14540 | ||
11fdf7f2 TL |
14541 | int64_t Client::ll_writev(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) |
14542 | { | |
f67539c2 TL |
14543 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14544 | if (!mref_reader.is_state_satisfied()) | |
14545 | return -CEPHFS_ENOTCONN; | |
14546 | ||
20effc67 TL |
14547 | std::scoped_lock cl(client_lock); |
14548 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, true, false); | |
11fdf7f2 TL |
14549 | } |
14550 | ||
14551 | int64_t Client::ll_readv(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) | |
14552 | { | |
f67539c2 TL |
14553 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14554 | if (!mref_reader.is_state_satisfied()) | |
14555 | return -CEPHFS_ENOTCONN; | |
14556 | ||
20effc67 TL |
14557 | std::scoped_lock cl(client_lock); |
14558 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, false, false); | |
11fdf7f2 TL |
14559 | } |
14560 | ||
7c673cae FG |
14561 | int Client::ll_flush(Fh *fh) |
14562 | { | |
f67539c2 TL |
14563 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14564 | if (!mref_reader.is_state_satisfied()) | |
14565 | return -CEPHFS_ENOTCONN; | |
14566 | ||
7c673cae FG |
14567 | ldout(cct, 3) << "ll_flush " << fh << " " << fh->inode->ino << " " << dendl; |
14568 | tout(cct) << "ll_flush" << std::endl; | |
f67539c2 | 14569 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14570 | |
f67539c2 | 14571 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14572 | return _flush(fh); |
14573 | } | |
14574 | ||
14575 | int Client::ll_fsync(Fh *fh, bool syncdataonly) | |
14576 | { | |
f67539c2 TL |
14577 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14578 | if (!mref_reader.is_state_satisfied()) | |
14579 | return -CEPHFS_ENOTCONN; | |
14580 | ||
7c673cae FG |
14581 | ldout(cct, 3) << "ll_fsync " << fh << " " << fh->inode->ino << " " << dendl; |
14582 | tout(cct) << "ll_fsync" << std::endl; | |
f67539c2 | 14583 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14584 | |
f67539c2 | 14585 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14586 | int r = _fsync(fh, syncdataonly); |
14587 | if (r) { | |
14588 | // If we're returning an error, clear it from the FH | |
14589 | fh->take_async_err(); | |
14590 | } | |
14591 | return r; | |
14592 | } | |
14593 | ||
28e407b8 AA |
14594 | int Client::ll_sync_inode(Inode *in, bool syncdataonly) |
14595 | { | |
f67539c2 TL |
14596 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14597 | if (!mref_reader.is_state_satisfied()) | |
14598 | return -CEPHFS_ENOTCONN; | |
14599 | ||
28e407b8 AA |
14600 | ldout(cct, 3) << "ll_sync_inode " << *in << " " << dendl; |
14601 | tout(cct) << "ll_sync_inode" << std::endl; | |
f67539c2 | 14602 | tout(cct) << (uintptr_t)in << std::endl; |
28e407b8 | 14603 | |
f67539c2 | 14604 | std::scoped_lock lock(client_lock); |
28e407b8 AA |
14605 | return _fsync(in, syncdataonly); |
14606 | } | |
14607 | ||
7c673cae FG |
14608 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
14609 | { | |
f67539c2 TL |
14610 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
14611 | ||
7c673cae | 14612 | if (offset < 0 || length <= 0) |
f67539c2 | 14613 | return -CEPHFS_EINVAL; |
7c673cae FG |
14614 | |
14615 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
f67539c2 | 14616 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14617 | |
14618 | if ((mode & FALLOC_FL_PUNCH_HOLE) && !(mode & FALLOC_FL_KEEP_SIZE)) | |
f67539c2 | 14619 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14620 | |
14621 | Inode *in = fh->inode.get(); | |
14622 | ||
14623 | if (objecter->osdmap_pool_full(in->layout.pool_id) && | |
14624 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
f67539c2 | 14625 | return -CEPHFS_ENOSPC; |
7c673cae FG |
14626 | } |
14627 | ||
14628 | if (in->snapid != CEPH_NOSNAP) | |
f67539c2 | 14629 | return -CEPHFS_EROFS; |
7c673cae FG |
14630 | |
14631 | if ((fh->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 14632 | return -CEPHFS_EBADF; |
7c673cae FG |
14633 | |
14634 | uint64_t size = offset + length; | |
14635 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && | |
14636 | size > in->size && | |
11fdf7f2 | 14637 | is_quota_bytes_exceeded(in, size - in->size, fh->actor_perms)) { |
f67539c2 | 14638 | return -CEPHFS_EDQUOT; |
7c673cae FG |
14639 | } |
14640 | ||
14641 | int have; | |
f6b5b4d7 | 14642 | int r = get_caps(fh, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER, &have, -1); |
7c673cae FG |
14643 | if (r < 0) |
14644 | return r; | |
14645 | ||
11fdf7f2 | 14646 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
7c673cae FG |
14647 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
14648 | if (in->inline_version < CEPH_INLINE_NONE && | |
14649 | (have & CEPH_CAP_FILE_BUFFER)) { | |
14650 | bufferlist bl; | |
9f95a23c | 14651 | auto inline_iter = in->inline_data.cbegin(); |
7c673cae FG |
14652 | int len = in->inline_data.length(); |
14653 | if (offset < len) { | |
14654 | if (offset > 0) | |
9f95a23c | 14655 | inline_iter.copy(offset, bl); |
7c673cae FG |
14656 | int size = length; |
14657 | if (offset + size > len) | |
14658 | size = len - offset; | |
14659 | if (size > 0) | |
14660 | bl.append_zero(size); | |
9f95a23c TL |
14661 | if (offset + size < len) { |
14662 | inline_iter += size; | |
14663 | inline_iter.copy(len - offset - size, bl); | |
14664 | } | |
7c673cae FG |
14665 | in->inline_data = bl; |
14666 | in->inline_version++; | |
14667 | } | |
91327a77 | 14668 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14669 | in->change_attr++; |
28e407b8 | 14670 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14671 | } else { |
14672 | if (in->inline_version < CEPH_INLINE_NONE) { | |
11fdf7f2 TL |
14673 | onuninline.reset(new C_SaferCond("Client::_fallocate_uninline_data flock")); |
14674 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
14675 | } |
14676 | ||
11fdf7f2 | 14677 | C_SaferCond onfinish("Client::_punch_hole flock"); |
7c673cae | 14678 | |
7c673cae FG |
14679 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
14680 | ||
14681 | _invalidate_inode_cache(in, offset, length); | |
14682 | filer->zero(in->ino, &in->layout, | |
14683 | in->snaprealm->get_snap_context(), | |
14684 | offset, length, | |
14685 | ceph::real_clock::now(), | |
11fdf7f2 | 14686 | 0, true, &onfinish); |
91327a77 | 14687 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14688 | in->change_attr++; |
28e407b8 | 14689 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14690 | |
9f95a23c | 14691 | client_lock.unlock(); |
11fdf7f2 | 14692 | onfinish.wait(); |
9f95a23c | 14693 | client_lock.lock(); |
f67539c2 | 14694 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
14695 | } |
14696 | } else if (!(mode & FALLOC_FL_KEEP_SIZE)) { | |
14697 | uint64_t size = offset + length; | |
14698 | if (size > in->size) { | |
14699 | in->size = size; | |
91327a77 | 14700 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14701 | in->change_attr++; |
28e407b8 | 14702 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14703 | |
11fdf7f2 | 14704 | if (is_quota_bytes_approaching(in, fh->actor_perms)) { |
7c673cae | 14705 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
14706 | } else if (is_max_size_approaching(in)) { |
14707 | check_caps(in, 0); | |
7c673cae FG |
14708 | } |
14709 | } | |
14710 | } | |
14711 | ||
11fdf7f2 | 14712 | if (nullptr != onuninline) { |
9f95a23c | 14713 | client_lock.unlock(); |
11fdf7f2 | 14714 | int ret = onuninline->wait(); |
9f95a23c | 14715 | client_lock.lock(); |
7c673cae | 14716 | |
f67539c2 | 14717 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
14718 | in->inline_data.clear(); |
14719 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 14720 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14721 | check_caps(in, 0); |
14722 | } else | |
11fdf7f2 | 14723 | r = ret; |
7c673cae FG |
14724 | } |
14725 | ||
14726 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
14727 | return r; | |
14728 | } | |
7c673cae | 14729 | |
11fdf7f2 | 14730 | int Client::ll_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
7c673cae | 14731 | { |
f67539c2 TL |
14732 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14733 | if (!mref_reader.is_state_satisfied()) | |
14734 | return -CEPHFS_ENOTCONN; | |
14735 | ||
11fdf7f2 TL |
14736 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << dendl; |
14737 | tout(cct) << __func__ << " " << mode << " " << offset << " " << length << std::endl; | |
f67539c2 | 14738 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14739 | |
f67539c2 | 14740 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14741 | return _fallocate(fh, mode, offset, length); |
14742 | } | |
14743 | ||
14744 | int Client::fallocate(int fd, int mode, loff_t offset, loff_t length) | |
14745 | { | |
f67539c2 TL |
14746 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14747 | if (!mref_reader.is_state_satisfied()) | |
14748 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14749 | |
f67539c2 | 14750 | tout(cct) << __func__ << " " << " " << fd << mode << " " << offset << " " << length << std::endl; |
181888fb | 14751 | |
f67539c2 | 14752 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14753 | Fh *fh = get_filehandle(fd); |
14754 | if (!fh) | |
f67539c2 | 14755 | return -CEPHFS_EBADF; |
7c673cae FG |
14756 | #if defined(__linux__) && defined(O_PATH) |
14757 | if (fh->flags & O_PATH) | |
f67539c2 | 14758 | return -CEPHFS_EBADF; |
7c673cae FG |
14759 | #endif |
14760 | return _fallocate(fh, mode, offset, length); | |
14761 | } | |
14762 | ||
14763 | int Client::ll_release(Fh *fh) | |
14764 | { | |
f67539c2 TL |
14765 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14766 | if (!mref_reader.is_state_satisfied()) | |
14767 | return -CEPHFS_ENOTCONN; | |
91327a77 | 14768 | |
11fdf7f2 | 14769 | ldout(cct, 3) << __func__ << " (fh)" << fh << " " << fh->inode->ino << " " << |
7c673cae | 14770 | dendl; |
11fdf7f2 | 14771 | tout(cct) << __func__ << " (fh)" << std::endl; |
f67539c2 TL |
14772 | tout(cct) << (uintptr_t)fh << std::endl; |
14773 | ||
14774 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
14775 | |
14776 | if (ll_unclosed_fh_set.count(fh)) | |
14777 | ll_unclosed_fh_set.erase(fh); | |
14778 | return _release_fh(fh); | |
14779 | } | |
14780 | ||
14781 | int Client::ll_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
14782 | { | |
f67539c2 TL |
14783 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14784 | if (!mref_reader.is_state_satisfied()) | |
14785 | return -CEPHFS_ENOTCONN; | |
7c673cae FG |
14786 | |
14787 | ldout(cct, 3) << "ll_getlk (fh)" << fh << " " << fh->inode->ino << dendl; | |
f67539c2 | 14788 | tout(cct) << "ll_getk (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14789 | |
f67539c2 | 14790 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14791 | return _getlk(fh, fl, owner); |
14792 | } | |
14793 | ||
14794 | int Client::ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
14795 | { | |
f67539c2 TL |
14796 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14797 | if (!mref_reader.is_state_satisfied()) | |
14798 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14799 | |
11fdf7f2 | 14800 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14801 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14802 | |
f67539c2 | 14803 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14804 | return _setlk(fh, fl, owner, sleep); |
14805 | } | |
14806 | ||
14807 | int Client::ll_flock(Fh *fh, int cmd, uint64_t owner) | |
14808 | { | |
f67539c2 TL |
14809 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14810 | if (!mref_reader.is_state_satisfied()) | |
14811 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14812 | |
11fdf7f2 | 14813 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14814 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14815 | |
f67539c2 | 14816 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14817 | return _flock(fh, cmd, owner); |
14818 | } | |
14819 | ||
b32b8144 FG |
14820 | int Client::set_deleg_timeout(uint32_t timeout) |
14821 | { | |
f67539c2 | 14822 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14823 | |
14824 | /* | |
f67539c2 | 14825 | * The whole point is to prevent blocklisting so we must time out the |
b32b8144 FG |
14826 | * delegation before the session autoclose timeout kicks in. |
14827 | */ | |
14828 | if (timeout >= mdsmap->get_session_autoclose()) | |
f67539c2 | 14829 | return -CEPHFS_EINVAL; |
b32b8144 FG |
14830 | |
14831 | deleg_timeout = timeout; | |
14832 | return 0; | |
14833 | } | |
14834 | ||
14835 | int Client::ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv) | |
14836 | { | |
f67539c2 | 14837 | int ret = -CEPHFS_EINVAL; |
b32b8144 | 14838 | |
f67539c2 TL |
14839 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14840 | if (!mref_reader.is_state_satisfied()) | |
14841 | return -CEPHFS_ENOTCONN; | |
b32b8144 | 14842 | |
f67539c2 | 14843 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14844 | |
14845 | Inode *inode = fh->inode.get(); | |
14846 | ||
14847 | switch(cmd) { | |
14848 | case CEPH_DELEGATION_NONE: | |
14849 | inode->unset_deleg(fh); | |
14850 | ret = 0; | |
14851 | break; | |
14852 | default: | |
14853 | try { | |
14854 | ret = inode->set_deleg(fh, cmd, cb, priv); | |
11fdf7f2 | 14855 | } catch (std::bad_alloc&) { |
f67539c2 | 14856 | ret = -CEPHFS_ENOMEM; |
b32b8144 FG |
14857 | } |
14858 | break; | |
14859 | } | |
14860 | return ret; | |
14861 | } | |
14862 | ||
7c673cae FG |
14863 | class C_Client_RequestInterrupt : public Context { |
14864 | private: | |
14865 | Client *client; | |
14866 | MetaRequest *req; | |
14867 | public: | |
14868 | C_Client_RequestInterrupt(Client *c, MetaRequest *r) : client(c), req(r) { | |
14869 | req->get(); | |
14870 | } | |
14871 | void finish(int r) override { | |
f67539c2 | 14872 | std::scoped_lock l(client->client_lock); |
11fdf7f2 | 14873 | ceph_assert(req->head.op == CEPH_MDS_OP_SETFILELOCK); |
7c673cae FG |
14874 | client->_interrupt_filelock(req); |
14875 | client->put_request(req); | |
14876 | } | |
14877 | }; | |
14878 | ||
14879 | void Client::ll_interrupt(void *d) | |
14880 | { | |
14881 | MetaRequest *req = static_cast<MetaRequest*>(d); | |
11fdf7f2 TL |
14882 | ldout(cct, 3) << __func__ << " tid " << req->get_tid() << dendl; |
14883 | tout(cct) << __func__ << " tid " << req->get_tid() << std::endl; | |
7c673cae FG |
14884 | interrupt_finisher.queue(new C_Client_RequestInterrupt(this, req)); |
14885 | } | |
14886 | ||
14887 | // ========================================= | |
14888 | // layout | |
14889 | ||
14890 | // expose file layouts | |
14891 | ||
14892 | int Client::describe_layout(const char *relpath, file_layout_t *lp, | |
14893 | const UserPerm& perms) | |
14894 | { | |
f67539c2 TL |
14895 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14896 | if (!mref_reader.is_state_satisfied()) | |
14897 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14898 | |
f67539c2 | 14899 | std::scoped_lock lock(client_lock); |
181888fb | 14900 | |
7c673cae FG |
14901 | filepath path(relpath); |
14902 | InodeRef in; | |
14903 | int r = path_walk(path, &in, perms); | |
14904 | if (r < 0) | |
14905 | return r; | |
14906 | ||
14907 | *lp = in->layout; | |
14908 | ||
11fdf7f2 | 14909 | ldout(cct, 3) << __func__ << "(" << relpath << ") = 0" << dendl; |
7c673cae FG |
14910 | return 0; |
14911 | } | |
14912 | ||
14913 | int Client::fdescribe_layout(int fd, file_layout_t *lp) | |
14914 | { | |
f67539c2 TL |
14915 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14916 | if (!mref_reader.is_state_satisfied()) | |
14917 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14918 | |
f67539c2 | 14919 | std::scoped_lock lock(client_lock); |
181888fb | 14920 | |
7c673cae FG |
14921 | Fh *f = get_filehandle(fd); |
14922 | if (!f) | |
f67539c2 | 14923 | return -CEPHFS_EBADF; |
7c673cae FG |
14924 | Inode *in = f->inode.get(); |
14925 | ||
14926 | *lp = in->layout; | |
14927 | ||
11fdf7f2 | 14928 | ldout(cct, 3) << __func__ << "(" << fd << ") = 0" << dendl; |
7c673cae FG |
14929 | return 0; |
14930 | } | |
14931 | ||
d2e6a577 FG |
14932 | int64_t Client::get_default_pool_id() |
14933 | { | |
f67539c2 TL |
14934 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14935 | if (!mref_reader.is_state_satisfied()) | |
14936 | return -CEPHFS_ENOTCONN; | |
181888fb | 14937 | |
f67539c2 | 14938 | std::scoped_lock lock(client_lock); |
181888fb | 14939 | |
d2e6a577 FG |
14940 | /* first data pool is the default */ |
14941 | return mdsmap->get_first_data_pool(); | |
14942 | } | |
7c673cae FG |
14943 | |
14944 | // expose osdmap | |
14945 | ||
14946 | int64_t Client::get_pool_id(const char *pool_name) | |
14947 | { | |
f67539c2 TL |
14948 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14949 | if (!mref_reader.is_state_satisfied()) | |
14950 | return -CEPHFS_ENOTCONN; | |
181888fb | 14951 | |
f67539c2 | 14952 | std::scoped_lock lock(client_lock); |
181888fb | 14953 | |
7c673cae FG |
14954 | return objecter->with_osdmap(std::mem_fn(&OSDMap::lookup_pg_pool_name), |
14955 | pool_name); | |
14956 | } | |
14957 | ||
14958 | string Client::get_pool_name(int64_t pool) | |
14959 | { | |
f67539c2 TL |
14960 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14961 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
14962 | return string(); |
14963 | ||
f67539c2 TL |
14964 | std::scoped_lock lock(client_lock); |
14965 | ||
7c673cae FG |
14966 | return objecter->with_osdmap([pool](const OSDMap& o) { |
14967 | return o.have_pg_pool(pool) ? o.get_pool_name(pool) : string(); | |
14968 | }); | |
14969 | } | |
14970 | ||
14971 | int Client::get_pool_replication(int64_t pool) | |
14972 | { | |
f67539c2 TL |
14973 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14974 | if (!mref_reader.is_state_satisfied()) | |
14975 | return -CEPHFS_ENOTCONN; | |
181888fb | 14976 | |
f67539c2 | 14977 | std::scoped_lock lock(client_lock); |
181888fb | 14978 | |
7c673cae | 14979 | return objecter->with_osdmap([pool](const OSDMap& o) { |
f67539c2 | 14980 | return o.have_pg_pool(pool) ? o.get_pg_pool(pool)->get_size() : -CEPHFS_ENOENT; |
7c673cae FG |
14981 | }); |
14982 | } | |
14983 | ||
14984 | int Client::get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds) | |
14985 | { | |
f67539c2 TL |
14986 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14987 | if (!mref_reader.is_state_satisfied()) | |
14988 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14989 | |
f67539c2 | 14990 | std::scoped_lock lock(client_lock); |
181888fb | 14991 | |
7c673cae FG |
14992 | Fh *f = get_filehandle(fd); |
14993 | if (!f) | |
f67539c2 | 14994 | return -CEPHFS_EBADF; |
7c673cae FG |
14995 | Inode *in = f->inode.get(); |
14996 | ||
14997 | vector<ObjectExtent> extents; | |
14998 | Striper::file_to_extents(cct, in->ino, &in->layout, off, 1, in->truncate_size, extents); | |
11fdf7f2 | 14999 | ceph_assert(extents.size() == 1); |
7c673cae FG |
15000 | |
15001 | objecter->with_osdmap([&](const OSDMap& o) { | |
15002 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
15003 | o.pg_to_acting_osds(pg, osds); | |
15004 | }); | |
15005 | ||
15006 | if (osds.empty()) | |
f67539c2 | 15007 | return -CEPHFS_EINVAL; |
7c673cae FG |
15008 | |
15009 | /* | |
15010 | * Return the remainder of the extent (stripe unit) | |
15011 | * | |
15012 | * If length = 1 is passed to Striper::file_to_extents we get a single | |
15013 | * extent back, but its length is one so we still need to compute the length | |
15014 | * to the end of the stripe unit. | |
15015 | * | |
15016 | * If length = su then we may get 1 or 2 objects back in the extents vector | |
15017 | * which would have to be examined. Even then, the offsets are local to the | |
15018 | * object, so matching up to the file offset is extra work. | |
15019 | * | |
15020 | * It seems simpler to stick with length = 1 and manually compute the | |
15021 | * remainder. | |
15022 | */ | |
15023 | if (len) { | |
15024 | uint64_t su = in->layout.stripe_unit; | |
15025 | *len = su - (off % su); | |
15026 | } | |
15027 | ||
15028 | return 0; | |
15029 | } | |
15030 | ||
15031 | int Client::get_osd_crush_location(int id, vector<pair<string, string> >& path) | |
15032 | { | |
f67539c2 TL |
15033 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15034 | if (!mref_reader.is_state_satisfied()) | |
15035 | return -CEPHFS_ENOTCONN; | |
181888fb | 15036 | |
f67539c2 | 15037 | std::scoped_lock lock(client_lock); |
181888fb | 15038 | |
7c673cae | 15039 | if (id < 0) |
f67539c2 | 15040 | return -CEPHFS_EINVAL; |
7c673cae FG |
15041 | return objecter->with_osdmap([&](const OSDMap& o) { |
15042 | return o.crush->get_full_location_ordered(id, path); | |
15043 | }); | |
15044 | } | |
15045 | ||
15046 | int Client::get_file_stripe_address(int fd, loff_t offset, | |
15047 | vector<entity_addr_t>& address) | |
15048 | { | |
f67539c2 TL |
15049 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15050 | if (!mref_reader.is_state_satisfied()) | |
15051 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15052 | |
f67539c2 | 15053 | std::scoped_lock lock(client_lock); |
181888fb | 15054 | |
7c673cae FG |
15055 | Fh *f = get_filehandle(fd); |
15056 | if (!f) | |
f67539c2 | 15057 | return -CEPHFS_EBADF; |
7c673cae FG |
15058 | Inode *in = f->inode.get(); |
15059 | ||
15060 | // which object? | |
15061 | vector<ObjectExtent> extents; | |
15062 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, 1, | |
15063 | in->truncate_size, extents); | |
11fdf7f2 | 15064 | ceph_assert(extents.size() == 1); |
7c673cae FG |
15065 | |
15066 | // now we have the object and its 'layout' | |
15067 | return objecter->with_osdmap([&](const OSDMap& o) { | |
15068 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
15069 | vector<int> osds; | |
15070 | o.pg_to_acting_osds(pg, osds); | |
15071 | if (osds.empty()) | |
f67539c2 | 15072 | return -CEPHFS_EINVAL; |
7c673cae | 15073 | for (unsigned i = 0; i < osds.size(); i++) { |
11fdf7f2 | 15074 | entity_addr_t addr = o.get_addrs(osds[i]).front(); |
7c673cae FG |
15075 | address.push_back(addr); |
15076 | } | |
15077 | return 0; | |
15078 | }); | |
15079 | } | |
15080 | ||
15081 | int Client::get_osd_addr(int osd, entity_addr_t& addr) | |
15082 | { | |
f67539c2 TL |
15083 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15084 | if (!mref_reader.is_state_satisfied()) | |
15085 | return -CEPHFS_ENOTCONN; | |
181888fb | 15086 | |
f67539c2 | 15087 | std::scoped_lock lock(client_lock); |
181888fb | 15088 | |
7c673cae FG |
15089 | return objecter->with_osdmap([&](const OSDMap& o) { |
15090 | if (!o.exists(osd)) | |
f67539c2 | 15091 | return -CEPHFS_ENOENT; |
7c673cae | 15092 | |
11fdf7f2 | 15093 | addr = o.get_addrs(osd).front(); |
7c673cae FG |
15094 | return 0; |
15095 | }); | |
15096 | } | |
15097 | ||
15098 | int Client::enumerate_layout(int fd, vector<ObjectExtent>& result, | |
15099 | loff_t length, loff_t offset) | |
15100 | { | |
f67539c2 TL |
15101 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15102 | if (!mref_reader.is_state_satisfied()) | |
15103 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15104 | |
f67539c2 | 15105 | std::scoped_lock lock(client_lock); |
181888fb | 15106 | |
7c673cae FG |
15107 | Fh *f = get_filehandle(fd); |
15108 | if (!f) | |
f67539c2 | 15109 | return -CEPHFS_EBADF; |
7c673cae FG |
15110 | Inode *in = f->inode.get(); |
15111 | ||
15112 | // map to a list of extents | |
15113 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, length, in->truncate_size, result); | |
15114 | ||
11fdf7f2 | 15115 | ldout(cct, 3) << __func__ << "(" << fd << ", " << length << ", " << offset << ") = 0" << dendl; |
7c673cae FG |
15116 | return 0; |
15117 | } | |
15118 | ||
15119 | ||
f67539c2 | 15120 | /* find an osd with the same ip. -CEPHFS_ENXIO if none. */ |
7c673cae FG |
15121 | int Client::get_local_osd() |
15122 | { | |
f67539c2 TL |
15123 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15124 | if (!mref_reader.is_state_satisfied()) | |
15125 | return -CEPHFS_ENOTCONN; | |
181888fb | 15126 | |
f67539c2 | 15127 | std::scoped_lock lock(client_lock); |
181888fb | 15128 | |
7c673cae FG |
15129 | objecter->with_osdmap([this](const OSDMap& o) { |
15130 | if (o.get_epoch() != local_osd_epoch) { | |
11fdf7f2 | 15131 | local_osd = o.find_osd_on_ip(messenger->get_myaddrs().front()); |
7c673cae FG |
15132 | local_osd_epoch = o.get_epoch(); |
15133 | } | |
15134 | }); | |
15135 | return local_osd; | |
15136 | } | |
15137 | ||
15138 | ||
15139 | ||
15140 | ||
15141 | ||
15142 | ||
15143 | // =============================== | |
15144 | ||
15145 | void Client::ms_handle_connect(Connection *con) | |
15146 | { | |
11fdf7f2 | 15147 | ldout(cct, 10) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15148 | } |
15149 | ||
15150 | bool Client::ms_handle_reset(Connection *con) | |
15151 | { | |
11fdf7f2 | 15152 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15153 | return false; |
15154 | } | |
15155 | ||
15156 | void Client::ms_handle_remote_reset(Connection *con) | |
15157 | { | |
f67539c2 | 15158 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 15159 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15160 | switch (con->get_peer_type()) { |
15161 | case CEPH_ENTITY_TYPE_MDS: | |
15162 | { | |
15163 | // kludge to figure out which mds this is; fixme with a Connection* state | |
15164 | mds_rank_t mds = MDS_RANK_NONE; | |
20effc67 | 15165 | MetaSessionRef s = NULL; |
11fdf7f2 | 15166 | for (auto &p : mds_sessions) { |
b3b6e05e | 15167 | if (mdsmap->have_inst(p.first) && mdsmap->get_addrs(p.first) == con->get_peer_addrs()) { |
11fdf7f2 | 15168 | mds = p.first; |
20effc67 | 15169 | s = p.second; |
7c673cae FG |
15170 | } |
15171 | } | |
15172 | if (mds >= 0) { | |
20effc67 | 15173 | ceph_assert(s != NULL); |
7c673cae FG |
15174 | switch (s->state) { |
15175 | case MetaSession::STATE_CLOSING: | |
15176 | ldout(cct, 1) << "reset from mds we were closing; we'll call that closed" << dendl; | |
20effc67 | 15177 | _closed_mds_session(s.get()); |
7c673cae FG |
15178 | break; |
15179 | ||
15180 | case MetaSession::STATE_OPENING: | |
15181 | { | |
15182 | ldout(cct, 1) << "reset from mds we were opening; retrying" << dendl; | |
15183 | list<Context*> waiters; | |
15184 | waiters.swap(s->waiting_for_open); | |
20effc67 TL |
15185 | _closed_mds_session(s.get()); |
15186 | auto news = _get_or_open_mds_session(mds); | |
7c673cae FG |
15187 | news->waiting_for_open.swap(waiters); |
15188 | } | |
15189 | break; | |
15190 | ||
15191 | case MetaSession::STATE_OPEN: | |
15192 | { | |
f67539c2 | 15193 | objecter->maybe_request_map(); /* to check if we are blocklisted */ |
f6b5b4d7 | 15194 | if (cct->_conf.get_val<bool>("client_reconnect_stale")) { |
7c673cae | 15195 | ldout(cct, 1) << "reset from mds we were open; close mds session for reconnect" << dendl; |
20effc67 | 15196 | _closed_mds_session(s.get()); |
7c673cae FG |
15197 | } else { |
15198 | ldout(cct, 1) << "reset from mds we were open; mark session as stale" << dendl; | |
15199 | s->state = MetaSession::STATE_STALE; | |
15200 | } | |
15201 | } | |
15202 | break; | |
15203 | ||
15204 | case MetaSession::STATE_NEW: | |
15205 | case MetaSession::STATE_CLOSED: | |
15206 | default: | |
15207 | break; | |
15208 | } | |
15209 | } | |
15210 | } | |
15211 | break; | |
15212 | } | |
15213 | } | |
15214 | ||
15215 | bool Client::ms_handle_refused(Connection *con) | |
15216 | { | |
11fdf7f2 | 15217 | ldout(cct, 1) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15218 | return false; |
15219 | } | |
15220 | ||
7c673cae FG |
15221 | Inode *Client::get_quota_root(Inode *in, const UserPerm& perms) |
15222 | { | |
11fdf7f2 TL |
15223 | Inode *quota_in = root_ancestor; |
15224 | SnapRealm *realm = in->snaprealm; | |
15225 | while (realm) { | |
15226 | ldout(cct, 10) << __func__ << " realm " << realm->ino << dendl; | |
15227 | if (realm->ino != in->ino) { | |
15228 | auto p = inode_map.find(vinodeno_t(realm->ino, CEPH_NOSNAP)); | |
15229 | if (p == inode_map.end()) | |
15230 | break; | |
7c673cae | 15231 | |
11fdf7f2 TL |
15232 | if (p->second->quota.is_enable()) { |
15233 | quota_in = p->second; | |
15234 | break; | |
7c673cae | 15235 | } |
7c673cae | 15236 | } |
11fdf7f2 | 15237 | realm = realm->pparent; |
7c673cae | 15238 | } |
11fdf7f2 TL |
15239 | ldout(cct, 10) << __func__ << " " << in->vino() << " -> " << quota_in->vino() << dendl; |
15240 | return quota_in; | |
7c673cae FG |
15241 | } |
15242 | ||
15243 | /** | |
15244 | * Traverse quota ancestors of the Inode, return true | |
15245 | * if any of them passes the passed function | |
15246 | */ | |
15247 | bool Client::check_quota_condition(Inode *in, const UserPerm& perms, | |
15248 | std::function<bool (const Inode &in)> test) | |
15249 | { | |
15250 | while (true) { | |
11fdf7f2 | 15251 | ceph_assert(in != NULL); |
7c673cae FG |
15252 | if (test(*in)) { |
15253 | return true; | |
15254 | } | |
15255 | ||
15256 | if (in == root_ancestor) { | |
15257 | // We're done traversing, drop out | |
15258 | return false; | |
15259 | } else { | |
15260 | // Continue up the tree | |
15261 | in = get_quota_root(in, perms); | |
15262 | } | |
15263 | } | |
15264 | ||
15265 | return false; | |
15266 | } | |
15267 | ||
15268 | bool Client::is_quota_files_exceeded(Inode *in, const UserPerm& perms) | |
15269 | { | |
15270 | return check_quota_condition(in, perms, | |
15271 | [](const Inode &in) { | |
15272 | return in.quota.max_files && in.rstat.rsize() >= in.quota.max_files; | |
15273 | }); | |
15274 | } | |
15275 | ||
15276 | bool Client::is_quota_bytes_exceeded(Inode *in, int64_t new_bytes, | |
11fdf7f2 | 15277 | const UserPerm& perms) |
7c673cae FG |
15278 | { |
15279 | return check_quota_condition(in, perms, | |
11fdf7f2 | 15280 | [&new_bytes](const Inode &in) { |
7c673cae FG |
15281 | return in.quota.max_bytes && (in.rstat.rbytes + new_bytes) |
15282 | > in.quota.max_bytes; | |
15283 | }); | |
15284 | } | |
15285 | ||
11fdf7f2 | 15286 | bool Client::is_quota_bytes_approaching(Inode *in, const UserPerm& perms) |
7c673cae | 15287 | { |
9f95a23c TL |
15288 | ceph_assert(in->size >= in->reported_size); |
15289 | const uint64_t size = in->size - in->reported_size; | |
11fdf7f2 | 15290 | return check_quota_condition(in, perms, |
9f95a23c | 15291 | [&size](const Inode &in) { |
11fdf7f2 TL |
15292 | if (in.quota.max_bytes) { |
15293 | if (in.rstat.rbytes >= in.quota.max_bytes) { | |
15294 | return true; | |
15295 | } | |
15296 | ||
11fdf7f2 | 15297 | const uint64_t space = in.quota.max_bytes - in.rstat.rbytes; |
11fdf7f2 TL |
15298 | return (space >> 4) < size; |
15299 | } else { | |
15300 | return false; | |
15301 | } | |
15302 | }); | |
7c673cae FG |
15303 | } |
15304 | ||
15305 | enum { | |
15306 | POOL_CHECKED = 1, | |
15307 | POOL_CHECKING = 2, | |
15308 | POOL_READ = 4, | |
15309 | POOL_WRITE = 8, | |
15310 | }; | |
15311 | ||
15312 | int Client::check_pool_perm(Inode *in, int need) | |
15313 | { | |
f67539c2 TL |
15314 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
15315 | ||
7c673cae FG |
15316 | if (!cct->_conf->client_check_pool_perm) |
15317 | return 0; | |
15318 | ||
f67539c2 TL |
15319 | /* Only need to do this for regular files */ |
15320 | if (!in->is_file()) | |
15321 | return 0; | |
15322 | ||
7c673cae FG |
15323 | int64_t pool_id = in->layout.pool_id; |
15324 | std::string pool_ns = in->layout.pool_ns; | |
15325 | std::pair<int64_t, std::string> perm_key(pool_id, pool_ns); | |
15326 | int have = 0; | |
15327 | while (true) { | |
15328 | auto it = pool_perms.find(perm_key); | |
15329 | if (it == pool_perms.end()) | |
15330 | break; | |
15331 | if (it->second == POOL_CHECKING) { | |
15332 | // avoid concurrent checkings | |
15333 | wait_on_list(waiting_for_pool_perm); | |
15334 | } else { | |
15335 | have = it->second; | |
11fdf7f2 | 15336 | ceph_assert(have & POOL_CHECKED); |
7c673cae FG |
15337 | break; |
15338 | } | |
15339 | } | |
15340 | ||
15341 | if (!have) { | |
15342 | if (in->snapid != CEPH_NOSNAP) { | |
15343 | // pool permission check needs to write to the first object. But for snapshot, | |
20effc67 | 15344 | // head of the first object may have already been deleted. To avoid creating |
7c673cae FG |
15345 | // orphan object, skip the check for now. |
15346 | return 0; | |
15347 | } | |
15348 | ||
15349 | pool_perms[perm_key] = POOL_CHECKING; | |
15350 | ||
15351 | char oid_buf[32]; | |
15352 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (unsigned long long)in->ino); | |
15353 | object_t oid = oid_buf; | |
15354 | ||
15355 | SnapContext nullsnapc; | |
15356 | ||
15357 | C_SaferCond rd_cond; | |
15358 | ObjectOperation rd_op; | |
f67539c2 | 15359 | rd_op.stat(nullptr, nullptr, nullptr); |
7c673cae FG |
15360 | |
15361 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), rd_op, | |
15362 | nullsnapc, ceph::real_clock::now(), 0, &rd_cond); | |
15363 | ||
15364 | C_SaferCond wr_cond; | |
15365 | ObjectOperation wr_op; | |
15366 | wr_op.create(true); | |
15367 | ||
15368 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), wr_op, | |
15369 | nullsnapc, ceph::real_clock::now(), 0, &wr_cond); | |
15370 | ||
9f95a23c | 15371 | client_lock.unlock(); |
7c673cae FG |
15372 | int rd_ret = rd_cond.wait(); |
15373 | int wr_ret = wr_cond.wait(); | |
9f95a23c | 15374 | client_lock.lock(); |
7c673cae FG |
15375 | |
15376 | bool errored = false; | |
15377 | ||
f67539c2 | 15378 | if (rd_ret == 0 || rd_ret == -CEPHFS_ENOENT) |
7c673cae | 15379 | have |= POOL_READ; |
f67539c2 | 15380 | else if (rd_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15381 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15382 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15383 | errored = true; | |
15384 | } | |
15385 | ||
f67539c2 | 15386 | if (wr_ret == 0 || wr_ret == -CEPHFS_EEXIST) |
7c673cae | 15387 | have |= POOL_WRITE; |
f67539c2 | 15388 | else if (wr_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15389 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15390 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15391 | errored = true; | |
15392 | } | |
15393 | ||
15394 | if (errored) { | |
15395 | // Indeterminate: erase CHECKING state so that subsequent calls re-check. | |
15396 | // Raise EIO because actual error code might be misleading for | |
15397 | // userspace filesystem user. | |
15398 | pool_perms.erase(perm_key); | |
15399 | signal_cond_list(waiting_for_pool_perm); | |
f67539c2 | 15400 | return -CEPHFS_EIO; |
7c673cae FG |
15401 | } |
15402 | ||
15403 | pool_perms[perm_key] = have | POOL_CHECKED; | |
15404 | signal_cond_list(waiting_for_pool_perm); | |
15405 | } | |
15406 | ||
15407 | if ((need & CEPH_CAP_FILE_RD) && !(have & POOL_READ)) { | |
11fdf7f2 | 15408 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15409 | << " need " << ccap_string(need) << ", but no read perm" << dendl; |
f67539c2 | 15410 | return -CEPHFS_EPERM; |
7c673cae FG |
15411 | } |
15412 | if ((need & CEPH_CAP_FILE_WR) && !(have & POOL_WRITE)) { | |
11fdf7f2 | 15413 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15414 | << " need " << ccap_string(need) << ", but no write perm" << dendl; |
f67539c2 | 15415 | return -CEPHFS_EPERM; |
7c673cae FG |
15416 | } |
15417 | ||
15418 | return 0; | |
15419 | } | |
15420 | ||
15421 | int Client::_posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want) | |
15422 | { | |
15423 | if (acl_type == POSIX_ACL) { | |
15424 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15425 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15426 | ||
15427 | return posix_acl_permits(access_acl, in->uid, in->gid, perms, want); | |
15428 | } | |
15429 | } | |
f67539c2 | 15430 | return -CEPHFS_EAGAIN; |
7c673cae FG |
15431 | } |
15432 | ||
15433 | int Client::_posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms) | |
15434 | { | |
15435 | if (acl_type == NO_ACL) | |
15436 | return 0; | |
15437 | ||
15438 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
15439 | if (r < 0) | |
15440 | goto out; | |
15441 | ||
15442 | if (acl_type == POSIX_ACL) { | |
15443 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15444 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15445 | bufferptr acl(access_acl.c_str(), access_acl.length()); | |
15446 | r = posix_acl_access_chmod(acl, mode); | |
15447 | if (r < 0) | |
15448 | goto out; | |
15449 | r = _do_setxattr(in, ACL_EA_ACCESS, acl.c_str(), acl.length(), 0, perms); | |
15450 | } else { | |
15451 | r = 0; | |
15452 | } | |
15453 | } | |
15454 | out: | |
15455 | ldout(cct, 10) << __func__ << " ino " << in->ino << " result=" << r << dendl; | |
15456 | return r; | |
15457 | } | |
15458 | ||
15459 | int Client::_posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl, | |
15460 | const UserPerm& perms) | |
15461 | { | |
15462 | if (acl_type == NO_ACL) | |
15463 | return 0; | |
15464 | ||
15465 | if (S_ISLNK(*mode)) | |
15466 | return 0; | |
15467 | ||
15468 | int r = _getattr(dir, CEPH_STAT_CAP_XATTR, perms, dir->xattr_version == 0); | |
15469 | if (r < 0) | |
15470 | goto out; | |
15471 | ||
15472 | if (acl_type == POSIX_ACL) { | |
15473 | if (dir->xattrs.count(ACL_EA_DEFAULT)) { | |
15474 | map<string, bufferptr> xattrs; | |
15475 | ||
15476 | const bufferptr& default_acl = dir->xattrs[ACL_EA_DEFAULT]; | |
15477 | bufferptr acl(default_acl.c_str(), default_acl.length()); | |
15478 | r = posix_acl_inherit_mode(acl, mode); | |
15479 | if (r < 0) | |
15480 | goto out; | |
15481 | ||
15482 | if (r > 0) { | |
15483 | r = posix_acl_equiv_mode(acl.c_str(), acl.length(), mode); | |
15484 | if (r < 0) | |
15485 | goto out; | |
15486 | if (r > 0) | |
15487 | xattrs[ACL_EA_ACCESS] = acl; | |
15488 | } | |
15489 | ||
15490 | if (S_ISDIR(*mode)) | |
15491 | xattrs[ACL_EA_DEFAULT] = dir->xattrs[ACL_EA_DEFAULT]; | |
15492 | ||
15493 | r = xattrs.size(); | |
15494 | if (r > 0) | |
11fdf7f2 | 15495 | encode(xattrs, xattrs_bl); |
7c673cae FG |
15496 | } else { |
15497 | if (umask_cb) | |
15498 | *mode &= ~umask_cb(callback_handle); | |
15499 | r = 0; | |
15500 | } | |
15501 | } | |
15502 | out: | |
15503 | ldout(cct, 10) << __func__ << " dir ino " << dir->ino << " result=" << r << dendl; | |
15504 | return r; | |
15505 | } | |
15506 | ||
15507 | void Client::set_filer_flags(int flags) | |
15508 | { | |
f67539c2 | 15509 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15510 | ceph_assert(flags == 0 || |
7c673cae FG |
15511 | flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
15512 | objecter->add_global_op_flags(flags); | |
15513 | } | |
15514 | ||
15515 | void Client::clear_filer_flags(int flags) | |
15516 | { | |
f67539c2 | 15517 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15518 | ceph_assert(flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
7c673cae FG |
15519 | objecter->clear_global_op_flag(flags); |
15520 | } | |
15521 | ||
11fdf7f2 TL |
15522 | // called before mount |
15523 | void Client::set_uuid(const std::string& uuid) | |
15524 | { | |
f67539c2 TL |
15525 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15526 | ceph_assert(iref_reader.is_state_satisfied()); | |
15527 | ||
15528 | std::scoped_lock l(client_lock); | |
20effc67 | 15529 | ceph_assert(!uuid.empty()); |
11fdf7f2 TL |
15530 | |
15531 | metadata["uuid"] = uuid; | |
15532 | _close_sessions(); | |
15533 | } | |
15534 | ||
15535 | // called before mount. 0 means infinite | |
15536 | void Client::set_session_timeout(unsigned timeout) | |
15537 | { | |
f67539c2 TL |
15538 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15539 | ceph_assert(iref_reader.is_state_satisfied()); | |
15540 | ||
15541 | std::scoped_lock l(client_lock); | |
11fdf7f2 TL |
15542 | |
15543 | metadata["timeout"] = stringify(timeout); | |
15544 | } | |
15545 | ||
15546 | // called before mount | |
15547 | int Client::start_reclaim(const std::string& uuid, unsigned flags, | |
15548 | const std::string& fs_name) | |
15549 | { | |
f67539c2 TL |
15550 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15551 | if (!iref_reader.is_state_satisfied()) | |
15552 | return -CEPHFS_ENOTCONN; | |
11fdf7f2 TL |
15553 | |
15554 | if (uuid.empty()) | |
f67539c2 | 15555 | return -CEPHFS_EINVAL; |
11fdf7f2 | 15556 | |
f67539c2 | 15557 | std::unique_lock l(client_lock); |
11fdf7f2 TL |
15558 | { |
15559 | auto it = metadata.find("uuid"); | |
15560 | if (it != metadata.end() && it->second == uuid) | |
f67539c2 | 15561 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15562 | } |
15563 | ||
15564 | int r = subscribe_mdsmap(fs_name); | |
15565 | if (r < 0) { | |
15566 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
15567 | return r; | |
15568 | } | |
15569 | ||
15570 | if (metadata.empty()) | |
15571 | populate_metadata(""); | |
15572 | ||
15573 | while (mdsmap->get_epoch() == 0) | |
15574 | wait_on_list(waiting_for_mdsmap); | |
15575 | ||
15576 | reclaim_errno = 0; | |
15577 | for (unsigned mds = 0; mds < mdsmap->get_num_in_mds(); ) { | |
15578 | if (!mdsmap->is_up(mds)) { | |
15579 | ldout(cct, 10) << "mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
15580 | wait_on_list(waiting_for_mdsmap); | |
15581 | continue; | |
15582 | } | |
15583 | ||
20effc67 | 15584 | MetaSessionRef session; |
11fdf7f2 TL |
15585 | if (!have_open_session(mds)) { |
15586 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 15587 | if (session->state == MetaSession::STATE_REJECTED) |
f67539c2 | 15588 | return -CEPHFS_EPERM; |
11fdf7f2 TL |
15589 | if (session->state != MetaSession::STATE_OPENING) { |
15590 | // umounting? | |
f67539c2 | 15591 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15592 | } |
15593 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
15594 | wait_on_context_list(session->waiting_for_open); | |
11fdf7f2 TL |
15595 | continue; |
15596 | } | |
15597 | ||
20effc67 | 15598 | session = mds_sessions.at(mds); |
11fdf7f2 | 15599 | if (!session->mds_features.test(CEPHFS_FEATURE_RECLAIM_CLIENT)) |
f67539c2 | 15600 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
15601 | |
15602 | if (session->reclaim_state == MetaSession::RECLAIM_NULL || | |
15603 | session->reclaim_state == MetaSession::RECLAIMING) { | |
15604 | session->reclaim_state = MetaSession::RECLAIMING; | |
9f95a23c | 15605 | auto m = make_message<MClientReclaim>(uuid, flags); |
11fdf7f2 TL |
15606 | session->con->send_message2(std::move(m)); |
15607 | wait_on_list(waiting_for_reclaim); | |
15608 | } else if (session->reclaim_state == MetaSession::RECLAIM_FAIL) { | |
f67539c2 | 15609 | return reclaim_errno ? : -CEPHFS_ENOTRECOVERABLE; |
11fdf7f2 TL |
15610 | } else { |
15611 | mds++; | |
15612 | } | |
15613 | } | |
15614 | ||
15615 | // didn't find target session in any mds | |
15616 | if (reclaim_target_addrs.empty()) { | |
15617 | if (flags & CEPH_RECLAIM_RESET) | |
f67539c2 TL |
15618 | return -CEPHFS_ENOENT; |
15619 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15620 | } |
15621 | ||
15622 | if (flags & CEPH_RECLAIM_RESET) | |
15623 | return 0; | |
15624 | ||
f67539c2 TL |
15625 | // use blocklist to check if target session was killed |
15626 | // (config option mds_session_blocklist_on_evict needs to be true) | |
15627 | ldout(cct, 10) << __func__ << ": waiting for OSD epoch " << reclaim_osd_epoch << dendl; | |
15628 | bs::error_code ec; | |
15629 | l.unlock(); | |
15630 | objecter->wait_for_map(reclaim_osd_epoch, ca::use_blocked[ec]); | |
15631 | l.lock(); | |
11fdf7f2 | 15632 | |
f67539c2 TL |
15633 | if (ec) |
15634 | return ceph::from_error_code(ec); | |
15635 | ||
15636 | bool blocklisted = objecter->with_osdmap( | |
11fdf7f2 | 15637 | [this](const OSDMap &osd_map) -> bool { |
f67539c2 | 15638 | return osd_map.is_blocklisted(reclaim_target_addrs); |
11fdf7f2 | 15639 | }); |
f67539c2 TL |
15640 | if (blocklisted) |
15641 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15642 | |
15643 | metadata["reclaiming_uuid"] = uuid; | |
15644 | return 0; | |
15645 | } | |
15646 | ||
15647 | void Client::finish_reclaim() | |
15648 | { | |
15649 | auto it = metadata.find("reclaiming_uuid"); | |
15650 | if (it == metadata.end()) { | |
15651 | for (auto &p : mds_sessions) | |
20effc67 | 15652 | p.second->reclaim_state = MetaSession::RECLAIM_NULL; |
11fdf7f2 TL |
15653 | return; |
15654 | } | |
15655 | ||
15656 | for (auto &p : mds_sessions) { | |
20effc67 | 15657 | p.second->reclaim_state = MetaSession::RECLAIM_NULL; |
9f95a23c | 15658 | auto m = make_message<MClientReclaim>("", MClientReclaim::FLAG_FINISH); |
20effc67 | 15659 | p.second->con->send_message2(std::move(m)); |
11fdf7f2 TL |
15660 | } |
15661 | ||
15662 | metadata["uuid"] = it->second; | |
15663 | metadata.erase(it); | |
15664 | } | |
15665 | ||
15666 | void Client::handle_client_reclaim_reply(const MConstRef<MClientReclaimReply>& reply) | |
15667 | { | |
15668 | mds_rank_t from = mds_rank_t(reply->get_source().num()); | |
15669 | ldout(cct, 10) << __func__ << " " << *reply << " from mds." << from << dendl; | |
15670 | ||
f67539c2 | 15671 | std::scoped_lock cl(client_lock); |
20effc67 | 15672 | auto session = _get_mds_session(from, reply->get_connection().get()); |
11fdf7f2 TL |
15673 | if (!session) { |
15674 | ldout(cct, 10) << " discarding reclaim reply from sessionless mds." << from << dendl; | |
15675 | return; | |
15676 | } | |
15677 | ||
15678 | if (reply->get_result() >= 0) { | |
15679 | session->reclaim_state = MetaSession::RECLAIM_OK; | |
15680 | if (reply->get_epoch() > reclaim_osd_epoch) | |
15681 | reclaim_osd_epoch = reply->get_epoch(); | |
15682 | if (!reply->get_addrs().empty()) | |
15683 | reclaim_target_addrs = reply->get_addrs(); | |
15684 | } else { | |
15685 | session->reclaim_state = MetaSession::RECLAIM_FAIL; | |
15686 | reclaim_errno = reply->get_result(); | |
15687 | } | |
15688 | ||
15689 | signal_cond_list(waiting_for_reclaim); | |
15690 | } | |
15691 | ||
7c673cae FG |
15692 | /** |
15693 | * This is included in cap release messages, to cause | |
15694 | * the MDS to wait until this OSD map epoch. It is necessary | |
15695 | * in corner cases where we cancel RADOS ops, so that | |
15696 | * nobody else tries to do IO to the same objects in | |
15697 | * the same epoch as the cancelled ops. | |
15698 | */ | |
15699 | void Client::set_cap_epoch_barrier(epoch_t e) | |
15700 | { | |
15701 | ldout(cct, 5) << __func__ << " epoch = " << e << dendl; | |
15702 | cap_epoch_barrier = e; | |
15703 | } | |
15704 | ||
15705 | const char** Client::get_tracked_conf_keys() const | |
15706 | { | |
15707 | static const char* keys[] = { | |
15708 | "client_cache_size", | |
15709 | "client_cache_mid", | |
15710 | "client_acl_type", | |
b32b8144 FG |
15711 | "client_deleg_timeout", |
15712 | "client_deleg_break_on_open", | |
f67539c2 TL |
15713 | "client_oc_size", |
15714 | "client_oc_max_objects", | |
15715 | "client_oc_max_dirty", | |
15716 | "client_oc_target_dirty", | |
15717 | "client_oc_max_dirty_age", | |
7c673cae FG |
15718 | NULL |
15719 | }; | |
15720 | return keys; | |
15721 | } | |
15722 | ||
11fdf7f2 | 15723 | void Client::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
15724 | const std::set <std::string> &changed) |
15725 | { | |
f67539c2 | 15726 | std::scoped_lock lock(client_lock); |
7c673cae | 15727 | |
181888fb | 15728 | if (changed.count("client_cache_mid")) { |
7c673cae FG |
15729 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
15730 | } | |
15731 | if (changed.count("client_acl_type")) { | |
15732 | acl_type = NO_ACL; | |
15733 | if (cct->_conf->client_acl_type == "posix_acl") | |
15734 | acl_type = POSIX_ACL; | |
15735 | } | |
f67539c2 TL |
15736 | if (changed.count("client_oc_size")) { |
15737 | objectcacher->set_max_size(cct->_conf->client_oc_size); | |
15738 | } | |
15739 | if (changed.count("client_oc_max_objects")) { | |
15740 | objectcacher->set_max_objects(cct->_conf->client_oc_max_objects); | |
15741 | } | |
15742 | if (changed.count("client_oc_max_dirty")) { | |
15743 | objectcacher->set_max_dirty(cct->_conf->client_oc_max_dirty); | |
15744 | } | |
15745 | if (changed.count("client_oc_target_dirty")) { | |
15746 | objectcacher->set_target_dirty(cct->_conf->client_oc_target_dirty); | |
15747 | } | |
15748 | if (changed.count("client_oc_max_dirty_age")) { | |
15749 | objectcacher->set_max_dirty_age(cct->_conf->client_oc_max_dirty_age); | |
15750 | } | |
33c7a0ef TL |
15751 | if (changed.count("client_collect_and_send_global_metrics")) { |
15752 | _collect_and_send_global_metrics = cct->_conf.get_val<bool>( | |
15753 | "client_collect_and_send_global_metrics"); | |
15754 | } | |
7c673cae FG |
15755 | } |
15756 | ||
7c673cae FG |
15757 | void intrusive_ptr_add_ref(Inode *in) |
15758 | { | |
b3b6e05e | 15759 | in->iget(); |
7c673cae | 15760 | } |
f67539c2 | 15761 | |
7c673cae FG |
15762 | void intrusive_ptr_release(Inode *in) |
15763 | { | |
15764 | in->client->put_inode(in); | |
15765 | } | |
15766 | ||
15767 | mds_rank_t Client::_get_random_up_mds() const | |
15768 | { | |
9f95a23c | 15769 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
15770 | |
15771 | std::set<mds_rank_t> up; | |
15772 | mdsmap->get_up_mds_set(up); | |
15773 | ||
15774 | if (up.empty()) | |
15775 | return MDS_RANK_NONE; | |
15776 | std::set<mds_rank_t>::const_iterator p = up.begin(); | |
15777 | for (int n = rand() % up.size(); n; n--) | |
15778 | ++p; | |
15779 | return *p; | |
15780 | } | |
15781 | ||
15782 | ||
f67539c2 TL |
15783 | StandaloneClient::StandaloneClient(Messenger *m, MonClient *mc, |
15784 | boost::asio::io_context& ictx) | |
15785 | : Client(m, mc, new Objecter(m->cct, m, mc, ictx)) | |
7c673cae FG |
15786 | { |
15787 | monclient->set_messenger(m); | |
15788 | objecter->set_client_incarnation(0); | |
15789 | } | |
15790 | ||
15791 | StandaloneClient::~StandaloneClient() | |
15792 | { | |
15793 | delete objecter; | |
15794 | objecter = nullptr; | |
15795 | } | |
15796 | ||
15797 | int StandaloneClient::init() | |
15798 | { | |
f67539c2 TL |
15799 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
15800 | ceph_assert(iref_writer.is_first_writer()); | |
15801 | ||
e306af50 | 15802 | _pre_init(); |
7c673cae FG |
15803 | objecter->init(); |
15804 | ||
9f95a23c | 15805 | client_lock.lock(); |
7c673cae FG |
15806 | |
15807 | messenger->add_dispatcher_tail(objecter); | |
15808 | messenger->add_dispatcher_tail(this); | |
15809 | ||
15810 | monclient->set_want_keys(CEPH_ENTITY_TYPE_MDS | CEPH_ENTITY_TYPE_OSD); | |
15811 | int r = monclient->init(); | |
15812 | if (r < 0) { | |
15813 | // need to do cleanup because we're in an intermediate init state | |
f67539c2 TL |
15814 | { |
15815 | std::scoped_lock l(timer_lock); | |
15816 | timer.shutdown(); | |
15817 | } | |
15818 | ||
9f95a23c | 15819 | client_lock.unlock(); |
7c673cae FG |
15820 | objecter->shutdown(); |
15821 | objectcacher->stop(); | |
15822 | monclient->shutdown(); | |
15823 | return r; | |
15824 | } | |
15825 | objecter->start(); | |
15826 | ||
9f95a23c | 15827 | client_lock.unlock(); |
7c673cae | 15828 | _finish_init(); |
f67539c2 | 15829 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
15830 | |
15831 | return 0; | |
15832 | } | |
15833 | ||
15834 | void StandaloneClient::shutdown() | |
15835 | { | |
15836 | Client::shutdown(); | |
15837 | objecter->shutdown(); | |
15838 | monclient->shutdown(); | |
15839 | } |