]>
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 | |
7c673cae FG |
357 | if (cct->_conf->client_acl_type == "posix_acl") |
358 | acl_type = POSIX_ACL; | |
359 | ||
7c673cae FG |
360 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
361 | ||
362 | // file handles | |
363 | free_fd_set.insert(10, 1<<30); | |
364 | ||
365 | mdsmap.reset(new MDSMap); | |
366 | ||
367 | // osd interfaces | |
368 | writeback_handler.reset(new ObjecterWriteback(objecter, &objecter_finisher, | |
369 | &client_lock)); | |
370 | objectcacher.reset(new ObjectCacher(cct, "libcephfs", *writeback_handler, client_lock, | |
371 | client_flush_set_callback, // all commit callback | |
372 | (void*)this, | |
373 | cct->_conf->client_oc_size, | |
374 | cct->_conf->client_oc_max_objects, | |
375 | cct->_conf->client_oc_max_dirty, | |
376 | cct->_conf->client_oc_target_dirty, | |
377 | cct->_conf->client_oc_max_dirty_age, | |
378 | true)); | |
7c673cae FG |
379 | } |
380 | ||
381 | ||
382 | Client::~Client() | |
383 | { | |
9f95a23c | 384 | ceph_assert(ceph_mutex_is_not_locked(client_lock)); |
7c673cae | 385 | |
f67539c2 TL |
386 | // If the task is crashed or aborted and doesn't |
387 | // get any chance to run the umount and shutdow. | |
388 | { | |
389 | std::scoped_lock l{client_lock}; | |
390 | tick_thread_stopped = true; | |
391 | upkeep_cond.notify_one(); | |
392 | } | |
393 | ||
394 | if (upkeeper.joinable()) | |
395 | upkeeper.join(); | |
396 | ||
31f18b77 FG |
397 | // It is necessary to hold client_lock, because any inode destruction |
398 | // may call into ObjectCacher, which asserts that it's lock (which is | |
399 | // client_lock) is held. | |
f67539c2 | 400 | std::scoped_lock l{client_lock}; |
7c673cae FG |
401 | tear_down_cache(); |
402 | } | |
403 | ||
404 | void Client::tear_down_cache() | |
405 | { | |
406 | // fd's | |
f67539c2 TL |
407 | for (auto &[fd, fh] : fd_map) { |
408 | ldout(cct, 1) << __func__ << " forcing close of fh " << fd << " ino " << fh->inode->ino << dendl; | |
7c673cae FG |
409 | _release_fh(fh); |
410 | } | |
411 | fd_map.clear(); | |
412 | ||
413 | while (!opened_dirs.empty()) { | |
414 | dir_result_t *dirp = *opened_dirs.begin(); | |
11fdf7f2 | 415 | ldout(cct, 1) << __func__ << " forcing close of dir " << dirp << " ino " << dirp->inode->ino << dendl; |
7c673cae FG |
416 | _closedir(dirp); |
417 | } | |
418 | ||
419 | // caps! | |
420 | // *** FIXME *** | |
421 | ||
422 | // empty lru | |
7c673cae | 423 | trim_cache(); |
11fdf7f2 | 424 | ceph_assert(lru.lru_get_size() == 0); |
7c673cae FG |
425 | |
426 | // close root ino | |
11fdf7f2 | 427 | ceph_assert(inode_map.size() <= 1 + root_parents.size()); |
7c673cae | 428 | if (root && inode_map.size() == 1 + root_parents.size()) { |
b3b6e05e | 429 | root.reset(); |
7c673cae FG |
430 | } |
431 | ||
11fdf7f2 | 432 | ceph_assert(inode_map.empty()); |
7c673cae FG |
433 | } |
434 | ||
435 | inodeno_t Client::get_root_ino() | |
436 | { | |
f67539c2 | 437 | std::scoped_lock l(client_lock); |
7c673cae FG |
438 | if (use_faked_inos()) |
439 | return root->faked_ino; | |
440 | else | |
441 | return root->ino; | |
442 | } | |
443 | ||
444 | Inode *Client::get_root() | |
445 | { | |
f67539c2 | 446 | std::scoped_lock l(client_lock); |
7c673cae | 447 | root->ll_get(); |
b3b6e05e | 448 | return root.get(); |
7c673cae FG |
449 | } |
450 | ||
451 | ||
452 | // debug crapola | |
453 | ||
454 | void Client::dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected) | |
455 | { | |
456 | filepath path; | |
457 | in->make_long_path(path); | |
458 | ldout(cct, 1) << "dump_inode: " | |
459 | << (disconnected ? "DISCONNECTED ":"") | |
460 | << "inode " << in->ino | |
461 | << " " << path | |
b3b6e05e | 462 | << " ref " << in->get_nref() |
f67539c2 | 463 | << " " << *in << dendl; |
7c673cae FG |
464 | |
465 | if (f) { | |
466 | f->open_object_section("inode"); | |
467 | f->dump_stream("path") << path; | |
468 | if (disconnected) | |
469 | f->dump_int("disconnected", 1); | |
470 | in->dump(f); | |
471 | f->close_section(); | |
472 | } | |
473 | ||
474 | did.insert(in); | |
475 | if (in->dir) { | |
476 | ldout(cct, 1) << " dir " << in->dir << " size " << in->dir->dentries.size() << dendl; | |
477 | for (ceph::unordered_map<string, Dentry*>::iterator it = in->dir->dentries.begin(); | |
478 | it != in->dir->dentries.end(); | |
479 | ++it) { | |
480 | ldout(cct, 1) << " " << in->ino << " dn " << it->first << " " << it->second << " ref " << it->second->ref << dendl; | |
481 | if (f) { | |
482 | f->open_object_section("dentry"); | |
483 | it->second->dump(f); | |
484 | f->close_section(); | |
485 | } | |
486 | if (it->second->inode) | |
487 | dump_inode(f, it->second->inode.get(), did, false); | |
488 | } | |
489 | } | |
490 | } | |
491 | ||
492 | void Client::dump_cache(Formatter *f) | |
493 | { | |
494 | set<Inode*> did; | |
495 | ||
11fdf7f2 | 496 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
497 | |
498 | if (f) | |
499 | f->open_array_section("cache"); | |
500 | ||
501 | if (root) | |
b3b6e05e | 502 | dump_inode(f, root.get(), did, true); |
7c673cae FG |
503 | |
504 | // make a second pass to catch anything disconnected | |
505 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
506 | it != inode_map.end(); | |
507 | ++it) { | |
508 | if (did.count(it->second)) | |
509 | continue; | |
510 | dump_inode(f, it->second, did, true); | |
511 | } | |
512 | ||
513 | if (f) | |
514 | f->close_section(); | |
515 | } | |
516 | ||
517 | void Client::dump_status(Formatter *f) | |
518 | { | |
9f95a23c | 519 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
520 | |
521 | ldout(cct, 1) << __func__ << dendl; | |
522 | ||
523 | const epoch_t osd_epoch | |
524 | = objecter->with_osdmap(std::mem_fn(&OSDMap::get_epoch)); | |
525 | ||
526 | if (f) { | |
527 | f->open_object_section("metadata"); | |
528 | for (const auto& kv : metadata) | |
529 | f->dump_string(kv.first.c_str(), kv.second); | |
530 | f->close_section(); | |
531 | ||
532 | f->dump_int("dentry_count", lru.lru_get_size()); | |
533 | f->dump_int("dentry_pinned_count", lru.lru_get_num_pinned()); | |
534 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 535 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 | 536 | f->dump_object("inst", inst); |
11fdf7f2 TL |
537 | f->dump_object("addr", inst.addr); |
538 | f->dump_stream("inst_str") << inst.name << " " << inst.addr.get_legacy_str(); | |
539 | f->dump_string("addr_str", inst.addr.get_legacy_str()); | |
7c673cae FG |
540 | f->dump_int("inode_count", inode_map.size()); |
541 | f->dump_int("mds_epoch", mdsmap->get_epoch()); | |
542 | f->dump_int("osd_epoch", osd_epoch); | |
543 | f->dump_int("osd_epoch_barrier", cap_epoch_barrier); | |
f67539c2 | 544 | f->dump_bool("blocklisted", blocklisted); |
adb31ebb | 545 | f->dump_string("fs_name", mdsmap->get_fs_name()); |
7c673cae FG |
546 | } |
547 | } | |
548 | ||
e306af50 | 549 | void Client::_pre_init() |
7c673cae FG |
550 | { |
551 | timer.init(); | |
e306af50 TL |
552 | |
553 | objecter_finisher.start(); | |
554 | filer.reset(new Filer(objecter, &objecter_finisher)); | |
f67539c2 | 555 | objecter->enable_blocklist_events(); |
e306af50 | 556 | |
7c673cae | 557 | objectcacher->start(); |
e306af50 TL |
558 | } |
559 | ||
560 | int Client::init() | |
561 | { | |
f67539c2 TL |
562 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
563 | ceph_assert(iref_writer.is_first_writer()); | |
564 | ||
e306af50 | 565 | _pre_init(); |
9f95a23c | 566 | { |
f67539c2 | 567 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
568 | messenger->add_dispatcher_tail(this); |
569 | } | |
7c673cae | 570 | _finish_init(); |
f67539c2 | 571 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
572 | return 0; |
573 | } | |
574 | ||
575 | void Client::_finish_init() | |
576 | { | |
9f95a23c | 577 | { |
f67539c2 | 578 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
579 | // logger |
580 | PerfCountersBuilder plb(cct, "client", l_c_first, l_c_last); | |
581 | plb.add_time_avg(l_c_reply, "reply", "Latency of receiving a reply on metadata request"); | |
582 | plb.add_time_avg(l_c_lat, "lat", "Latency of processing a metadata request"); | |
583 | plb.add_time_avg(l_c_wrlat, "wrlat", "Latency of a file data write operation"); | |
584 | plb.add_time_avg(l_c_read, "rdlat", "Latency of a file data read operation"); | |
585 | plb.add_time_avg(l_c_fsync, "fsync", "Latency of a file sync operation"); | |
586 | logger.reset(plb.create_perf_counters()); | |
587 | cct->get_perfcounters_collection()->add(logger.get()); | |
588 | } | |
7c673cae | 589 | |
11fdf7f2 | 590 | cct->_conf.add_observer(this); |
7c673cae FG |
591 | |
592 | AdminSocket* admin_socket = cct->get_admin_socket(); | |
593 | int ret = admin_socket->register_command("mds_requests", | |
7c673cae FG |
594 | &m_command_hook, |
595 | "show in-progress mds requests"); | |
596 | if (ret < 0) { | |
597 | lderr(cct) << "error registering admin socket command: " | |
598 | << cpp_strerror(-ret) << dendl; | |
599 | } | |
adb31ebb TL |
600 | ret = admin_socket->register_command("mds_sessions " |
601 | "name=cap_dump,type=CephBool,req=false", | |
7c673cae FG |
602 | &m_command_hook, |
603 | "show mds session state"); | |
604 | if (ret < 0) { | |
605 | lderr(cct) << "error registering admin socket command: " | |
606 | << cpp_strerror(-ret) << dendl; | |
607 | } | |
608 | ret = admin_socket->register_command("dump_cache", | |
7c673cae FG |
609 | &m_command_hook, |
610 | "show in-memory metadata cache contents"); | |
611 | if (ret < 0) { | |
612 | lderr(cct) << "error registering admin socket command: " | |
613 | << cpp_strerror(-ret) << dendl; | |
614 | } | |
615 | ret = admin_socket->register_command("kick_stale_sessions", | |
7c673cae FG |
616 | &m_command_hook, |
617 | "kick sessions that were remote reset"); | |
618 | if (ret < 0) { | |
619 | lderr(cct) << "error registering admin socket command: " | |
620 | << cpp_strerror(-ret) << dendl; | |
621 | } | |
622 | ret = admin_socket->register_command("status", | |
7c673cae FG |
623 | &m_command_hook, |
624 | "show overall client status"); | |
625 | if (ret < 0) { | |
626 | lderr(cct) << "error registering admin socket command: " | |
627 | << cpp_strerror(-ret) << dendl; | |
628 | } | |
7c673cae FG |
629 | } |
630 | ||
631 | void Client::shutdown() | |
632 | { | |
11fdf7f2 | 633 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
634 | |
635 | // If we were not mounted, but were being used for sending | |
636 | // MDS commands, we may have sessions that need closing. | |
9f95a23c | 637 | { |
f67539c2 TL |
638 | std::scoped_lock l{client_lock}; |
639 | ||
640 | // To make sure the tick thread will be stoppped before | |
641 | // destructing the Client, just in case like the _mount() | |
642 | // failed but didn't not get a chance to stop the tick | |
643 | // thread | |
644 | tick_thread_stopped = true; | |
645 | upkeep_cond.notify_one(); | |
646 | ||
9f95a23c TL |
647 | _close_sessions(); |
648 | } | |
11fdf7f2 | 649 | cct->_conf.remove_observer(this); |
7c673cae | 650 | |
11fdf7f2 | 651 | cct->get_admin_socket()->unregister_commands(&m_command_hook); |
7c673cae FG |
652 | |
653 | if (ino_invalidate_cb) { | |
654 | ldout(cct, 10) << "shutdown stopping cache invalidator finisher" << dendl; | |
655 | async_ino_invalidator.wait_for_empty(); | |
656 | async_ino_invalidator.stop(); | |
657 | } | |
658 | ||
659 | if (dentry_invalidate_cb) { | |
660 | ldout(cct, 10) << "shutdown stopping dentry invalidator finisher" << dendl; | |
661 | async_dentry_invalidator.wait_for_empty(); | |
662 | async_dentry_invalidator.stop(); | |
663 | } | |
664 | ||
665 | if (switch_interrupt_cb) { | |
666 | ldout(cct, 10) << "shutdown stopping interrupt finisher" << dendl; | |
667 | interrupt_finisher.wait_for_empty(); | |
668 | interrupt_finisher.stop(); | |
669 | } | |
670 | ||
671 | if (remount_cb) { | |
672 | ldout(cct, 10) << "shutdown stopping remount finisher" << dendl; | |
673 | remount_finisher.wait_for_empty(); | |
674 | remount_finisher.stop(); | |
675 | } | |
676 | ||
e306af50 TL |
677 | if (ino_release_cb) { |
678 | ldout(cct, 10) << "shutdown stopping inode release finisher" << dendl; | |
679 | async_ino_releasor.wait_for_empty(); | |
680 | async_ino_releasor.stop(); | |
681 | } | |
682 | ||
7c673cae | 683 | objectcacher->stop(); // outside of client_lock! this does a join. |
f67539c2 TL |
684 | |
685 | /* | |
686 | * We are shuting down the client. | |
687 | * | |
688 | * Just declare the state to CLIENT_NEW to block and fail any | |
689 | * new comming "reader" and then try to wait all the in-flight | |
690 | * "readers" to finish. | |
691 | */ | |
692 | RWRef_t iref_writer(initialize_state, CLIENT_NEW, false); | |
693 | if (!iref_writer.is_first_writer()) | |
694 | return; | |
695 | iref_writer.wait_readers_done(); | |
696 | ||
9f95a23c | 697 | { |
f67539c2 | 698 | std::scoped_lock l(timer_lock); |
9f95a23c TL |
699 | timer.shutdown(); |
700 | } | |
f67539c2 | 701 | |
7c673cae FG |
702 | objecter_finisher.wait_for_empty(); |
703 | objecter_finisher.stop(); | |
704 | ||
705 | if (logger) { | |
706 | cct->get_perfcounters_collection()->remove(logger.get()); | |
707 | logger.reset(); | |
708 | } | |
709 | } | |
710 | ||
711 | ||
712 | // =================== | |
713 | // metadata cache stuff | |
714 | ||
715 | void Client::trim_cache(bool trim_kernel_dcache) | |
716 | { | |
181888fb FG |
717 | uint64_t max = cct->_conf->client_cache_size; |
718 | ldout(cct, 20) << "trim_cache size " << lru.lru_get_size() << " max " << max << dendl; | |
7c673cae FG |
719 | unsigned last = 0; |
720 | while (lru.lru_get_size() != last) { | |
721 | last = lru.lru_get_size(); | |
722 | ||
f67539c2 | 723 | if (!is_unmounting() && lru.lru_get_size() <= max) break; |
7c673cae FG |
724 | |
725 | // trim! | |
31f18b77 | 726 | Dentry *dn = static_cast<Dentry*>(lru.lru_get_next_expire()); |
7c673cae FG |
727 | if (!dn) |
728 | break; // done | |
f67539c2 | 729 | |
7c673cae FG |
730 | trim_dentry(dn); |
731 | } | |
732 | ||
181888fb | 733 | if (trim_kernel_dcache && lru.lru_get_size() > max) |
7c673cae FG |
734 | _invalidate_kernel_dcache(); |
735 | ||
736 | // hose root? | |
b3b6e05e | 737 | if (lru.lru_get_size() == 0 && root && root->get_nref() == 1 && inode_map.size() == 1 + root_parents.size()) { |
7c673cae | 738 | ldout(cct, 15) << "trim_cache trimmed root " << root << dendl; |
b3b6e05e | 739 | root.reset(); |
7c673cae FG |
740 | } |
741 | } | |
742 | ||
743 | void Client::trim_cache_for_reconnect(MetaSession *s) | |
744 | { | |
745 | mds_rank_t mds = s->mds_num; | |
11fdf7f2 | 746 | ldout(cct, 20) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
747 | |
748 | int trimmed = 0; | |
749 | list<Dentry*> skipped; | |
750 | while (lru.lru_get_size() > 0) { | |
751 | Dentry *dn = static_cast<Dentry*>(lru.lru_expire()); | |
752 | if (!dn) | |
753 | break; | |
754 | ||
755 | if ((dn->inode && dn->inode->caps.count(mds)) || | |
756 | dn->dir->parent_inode->caps.count(mds)) { | |
757 | trim_dentry(dn); | |
758 | trimmed++; | |
759 | } else | |
760 | skipped.push_back(dn); | |
761 | } | |
762 | ||
763 | for(list<Dentry*>::iterator p = skipped.begin(); p != skipped.end(); ++p) | |
764 | lru.lru_insert_mid(*p); | |
765 | ||
11fdf7f2 | 766 | ldout(cct, 20) << __func__ << " mds." << mds |
7c673cae FG |
767 | << " trimmed " << trimmed << " dentries" << dendl; |
768 | ||
769 | if (s->caps.size() > 0) | |
770 | _invalidate_kernel_dcache(); | |
771 | } | |
772 | ||
773 | void Client::trim_dentry(Dentry *dn) | |
774 | { | |
775 | ldout(cct, 15) << "trim_dentry unlinking dn " << dn->name | |
11fdf7f2 TL |
776 | << " in dir " |
777 | << std::hex << dn->dir->parent_inode->ino << std::dec | |
7c673cae FG |
778 | << dendl; |
779 | if (dn->inode) { | |
780 | Inode *diri = dn->dir->parent_inode; | |
7c673cae FG |
781 | clear_dir_complete_and_ordered(diri, true); |
782 | } | |
783 | unlink(dn, false, false); // drop dir, drop dentry | |
784 | } | |
785 | ||
786 | ||
1adf2230 AA |
787 | void Client::update_inode_file_size(Inode *in, int issued, uint64_t size, |
788 | uint64_t truncate_seq, uint64_t truncate_size) | |
7c673cae | 789 | { |
7c673cae FG |
790 | uint64_t prior_size = in->size; |
791 | ||
7c673cae FG |
792 | if (truncate_seq > in->truncate_seq || |
793 | (truncate_seq == in->truncate_seq && size > in->size)) { | |
794 | ldout(cct, 10) << "size " << in->size << " -> " << size << dendl; | |
795 | in->size = size; | |
796 | in->reported_size = size; | |
797 | if (truncate_seq != in->truncate_seq) { | |
798 | ldout(cct, 10) << "truncate_seq " << in->truncate_seq << " -> " | |
799 | << truncate_seq << dendl; | |
800 | in->truncate_seq = truncate_seq; | |
801 | in->oset.truncate_seq = truncate_seq; | |
802 | ||
803 | // truncate cached file data | |
804 | if (prior_size > size) { | |
805 | _invalidate_inode_cache(in, truncate_size, prior_size - truncate_size); | |
806 | } | |
807 | } | |
808 | ||
809 | // truncate inline data | |
810 | if (in->inline_version < CEPH_INLINE_NONE) { | |
811 | uint32_t len = in->inline_data.length(); | |
812 | if (size < len) | |
813 | in->inline_data.splice(size, len - size); | |
814 | } | |
815 | } | |
816 | if (truncate_seq >= in->truncate_seq && | |
817 | in->truncate_size != truncate_size) { | |
818 | if (in->is_file()) { | |
819 | ldout(cct, 10) << "truncate_size " << in->truncate_size << " -> " | |
820 | << truncate_size << dendl; | |
821 | in->truncate_size = truncate_size; | |
822 | in->oset.truncate_size = truncate_size; | |
823 | } else { | |
824 | ldout(cct, 0) << "Hmmm, truncate_seq && truncate_size changed on non-file inode!" << dendl; | |
825 | } | |
826 | } | |
1adf2230 AA |
827 | } |
828 | ||
829 | void Client::update_inode_file_time(Inode *in, int issued, uint64_t time_warp_seq, | |
830 | utime_t ctime, utime_t mtime, utime_t atime) | |
831 | { | |
832 | ldout(cct, 10) << __func__ << " " << *in << " " << ccap_string(issued) | |
833 | << " ctime " << ctime << " mtime " << mtime << dendl; | |
834 | ||
835 | if (time_warp_seq > in->time_warp_seq) | |
836 | ldout(cct, 10) << " mds time_warp_seq " << time_warp_seq | |
837 | << " is higher than local time_warp_seq " | |
838 | << in->time_warp_seq << dendl; | |
839 | ||
840 | int warn = false; | |
7c673cae FG |
841 | // be careful with size, mtime, atime |
842 | if (issued & (CEPH_CAP_FILE_EXCL| | |
843 | CEPH_CAP_FILE_WR| | |
844 | CEPH_CAP_FILE_BUFFER| | |
845 | CEPH_CAP_AUTH_EXCL| | |
846 | CEPH_CAP_XATTR_EXCL)) { | |
847 | ldout(cct, 30) << "Yay have enough caps to look at our times" << dendl; | |
848 | if (ctime > in->ctime) | |
849 | in->ctime = ctime; | |
850 | if (time_warp_seq > in->time_warp_seq) { | |
7c673cae FG |
851 | //the mds updated times, so take those! |
852 | in->mtime = mtime; | |
853 | in->atime = atime; | |
854 | in->time_warp_seq = time_warp_seq; | |
855 | } else if (time_warp_seq == in->time_warp_seq) { | |
856 | //take max times | |
857 | if (mtime > in->mtime) | |
858 | in->mtime = mtime; | |
859 | if (atime > in->atime) | |
860 | in->atime = atime; | |
861 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
862 | //ignore mds values as we have a higher seq | |
863 | } else warn = true; | |
864 | } else { | |
865 | ldout(cct, 30) << "Don't have enough caps, just taking mds' time values" << dendl; | |
866 | if (time_warp_seq >= in->time_warp_seq) { | |
867 | in->ctime = ctime; | |
868 | in->mtime = mtime; | |
869 | in->atime = atime; | |
870 | in->time_warp_seq = time_warp_seq; | |
871 | } else warn = true; | |
872 | } | |
873 | if (warn) { | |
874 | ldout(cct, 0) << "WARNING: " << *in << " mds time_warp_seq " | |
875 | << time_warp_seq << " is lower than local time_warp_seq " | |
876 | << in->time_warp_seq | |
877 | << dendl; | |
878 | } | |
879 | } | |
880 | ||
881 | void Client::_fragmap_remove_non_leaves(Inode *in) | |
882 | { | |
883 | for (map<frag_t,int>::iterator p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
884 | if (!in->dirfragtree.is_leaf(p->first)) | |
885 | in->fragmap.erase(p++); | |
886 | else | |
887 | ++p; | |
888 | } | |
889 | ||
890 | void Client::_fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds) | |
891 | { | |
892 | for (auto p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
893 | if (p->second == mds) | |
894 | in->fragmap.erase(p++); | |
895 | else | |
896 | ++p; | |
897 | } | |
898 | ||
899 | Inode * Client::add_update_inode(InodeStat *st, utime_t from, | |
900 | MetaSession *session, | |
901 | const UserPerm& request_perms) | |
902 | { | |
903 | Inode *in; | |
904 | bool was_new = false; | |
905 | if (inode_map.count(st->vino)) { | |
906 | in = inode_map[st->vino]; | |
11fdf7f2 | 907 | ldout(cct, 12) << __func__ << " had " << *in << " caps " << ccap_string(st->cap.caps) << dendl; |
7c673cae FG |
908 | } else { |
909 | in = new Inode(this, st->vino, &st->layout); | |
910 | inode_map[st->vino] = in; | |
911 | ||
912 | if (use_faked_inos()) | |
913 | _assign_faked_ino(in); | |
914 | ||
915 | if (!root) { | |
916 | root = in; | |
11fdf7f2 | 917 | if (use_faked_inos()) |
b3b6e05e | 918 | _assign_faked_root(root.get()); |
7c673cae FG |
919 | root_ancestor = in; |
920 | cwd = root; | |
f67539c2 | 921 | } else if (is_mounting()) { |
7c673cae FG |
922 | root_parents[root_ancestor] = in; |
923 | root_ancestor = in; | |
924 | } | |
925 | ||
926 | // immutable bits | |
927 | in->ino = st->vino.ino; | |
928 | in->snapid = st->vino.snapid; | |
929 | in->mode = st->mode & S_IFMT; | |
930 | was_new = true; | |
931 | } | |
932 | ||
933 | in->rdev = st->rdev; | |
934 | if (in->is_symlink()) | |
935 | in->symlink = st->symlink; | |
936 | ||
7c673cae | 937 | // only update inode if mds info is strictly newer, or it is the same and projected (odd). |
1adf2230 AA |
938 | bool new_version = false; |
939 | if (in->version == 0 || | |
940 | ((st->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
941 | (in->version & ~1) < st->version)) | |
942 | new_version = true; | |
7c673cae | 943 | |
1adf2230 AA |
944 | int issued; |
945 | in->caps_issued(&issued); | |
946 | issued |= in->caps_dirty(); | |
947 | int new_issued = ~issued & (int)st->cap.caps; | |
7c673cae | 948 | |
1adf2230 AA |
949 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
950 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
951 | in->mode = st->mode; | |
952 | in->uid = st->uid; | |
953 | in->gid = st->gid; | |
954 | in->btime = st->btime; | |
81eedcae | 955 | in->snap_btime = st->snap_btime; |
f67539c2 | 956 | in->snap_metadata = st->snap_metadata; |
1adf2230 | 957 | } |
7c673cae | 958 | |
1adf2230 AA |
959 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
960 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
961 | in->nlink = st->nlink; | |
962 | } | |
7c673cae | 963 | |
1adf2230 AA |
964 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
965 | update_inode_file_time(in, issued, st->time_warp_seq, | |
966 | st->ctime, st->mtime, st->atime); | |
967 | } | |
7c673cae | 968 | |
1adf2230 AA |
969 | if (new_version || |
970 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7c673cae | 971 | in->layout = st->layout; |
1adf2230 AA |
972 | update_inode_file_size(in, issued, st->size, st->truncate_seq, st->truncate_size); |
973 | } | |
7c673cae | 974 | |
1adf2230 AA |
975 | if (in->is_dir()) { |
976 | if (new_version || (new_issued & CEPH_CAP_FILE_SHARED)) { | |
977 | in->dirstat = st->dirstat; | |
978 | } | |
979 | // dir_layout/rstat/quota are not tracked by capability, update them only if | |
980 | // the inode stat is from auth mds | |
981 | if (new_version || (st->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
7c673cae FG |
982 | in->dir_layout = st->dir_layout; |
983 | ldout(cct, 20) << " dir hash is " << (int)in->dir_layout.dl_dir_hash << dendl; | |
1adf2230 AA |
984 | in->rstat = st->rstat; |
985 | in->quota = st->quota; | |
11fdf7f2 | 986 | in->dir_pin = st->dir_pin; |
1adf2230 AA |
987 | } |
988 | // move me if/when version reflects fragtree changes. | |
989 | if (in->dirfragtree != st->dirfragtree) { | |
990 | in->dirfragtree = st->dirfragtree; | |
991 | _fragmap_remove_non_leaves(in); | |
7c673cae | 992 | } |
7c673cae FG |
993 | } |
994 | ||
995 | if ((in->xattr_version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && | |
996 | st->xattrbl.length() && | |
997 | st->xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
998 | auto p = st->xattrbl.cbegin(); |
999 | decode(in->xattrs, p); | |
7c673cae FG |
1000 | in->xattr_version = st->xattr_version; |
1001 | } | |
1002 | ||
1adf2230 AA |
1003 | if (st->inline_version > in->inline_version) { |
1004 | in->inline_data = st->inline_data; | |
1005 | in->inline_version = st->inline_version; | |
7c673cae FG |
1006 | } |
1007 | ||
1adf2230 AA |
1008 | /* always take a newer change attr */ |
1009 | if (st->change_attr > in->change_attr) | |
1010 | in->change_attr = st->change_attr; | |
1011 | ||
1012 | if (st->version > in->version) | |
1013 | in->version = st->version; | |
1014 | ||
1015 | if (was_new) | |
1016 | ldout(cct, 12) << __func__ << " adding " << *in << " caps " << ccap_string(st->cap.caps) << dendl; | |
1017 | ||
1018 | if (!st->cap.caps) | |
1019 | return in; // as with readdir returning indoes in different snaprealms (no caps!) | |
1020 | ||
7c673cae | 1021 | if (in->snapid == CEPH_NOSNAP) { |
a8e16298 TL |
1022 | add_update_cap(in, session, st->cap.cap_id, st->cap.caps, st->cap.wanted, |
1023 | st->cap.seq, st->cap.mseq, inodeno_t(st->cap.realm), | |
1024 | st->cap.flags, request_perms); | |
28e407b8 | 1025 | if (in->auth_cap && in->auth_cap->session == session) { |
7c673cae | 1026 | in->max_size = st->max_size; |
28e407b8 AA |
1027 | in->rstat = st->rstat; |
1028 | } | |
7c673cae | 1029 | |
1adf2230 AA |
1030 | // setting I_COMPLETE needs to happen after adding the cap |
1031 | if (in->is_dir() && | |
1032 | (st->cap.caps & CEPH_CAP_FILE_SHARED) && | |
1033 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
1034 | in->dirstat.nfiles == 0 && | |
1035 | in->dirstat.nsubdirs == 0) { | |
1036 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on empty dir " << *in << dendl; | |
1037 | in->flags |= I_COMPLETE | I_DIR_ORDERED; | |
1038 | if (in->dir) { | |
1039 | ldout(cct, 10) << " dir is open on empty dir " << in->ino << " with " | |
1040 | << in->dir->dentries.size() << " entries, marking all dentries null" << dendl; | |
1041 | in->dir->readdir_cache.clear(); | |
1042 | for (const auto& p : in->dir->dentries) { | |
1043 | unlink(p.second, true, true); // keep dir, keep dentry | |
1044 | } | |
1045 | if (in->dir->dentries.empty()) | |
1046 | close_dir(in->dir); | |
7c673cae | 1047 | } |
7c673cae | 1048 | } |
1adf2230 AA |
1049 | } else { |
1050 | in->snap_caps |= st->cap.caps; | |
7c673cae FG |
1051 | } |
1052 | ||
f67539c2 | 1053 | in->fscrypt = st->fscrypt; |
7c673cae FG |
1054 | return in; |
1055 | } | |
1056 | ||
1057 | ||
1058 | /* | |
1059 | * insert_dentry_inode - insert + link a single dentry + inode into the metadata cache. | |
1060 | */ | |
1061 | Dentry *Client::insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease, | |
1062 | Inode *in, utime_t from, MetaSession *session, | |
1063 | Dentry *old_dentry) | |
1064 | { | |
1065 | Dentry *dn = NULL; | |
1066 | if (dir->dentries.count(dname)) | |
1067 | dn = dir->dentries[dname]; | |
1068 | ||
11fdf7f2 | 1069 | ldout(cct, 12) << __func__ << " '" << dname << "' vino " << in->vino() |
7c673cae FG |
1070 | << " in dir " << dir->parent_inode->vino() << " dn " << dn |
1071 | << dendl; | |
1072 | ||
1073 | if (dn && dn->inode) { | |
1074 | if (dn->inode->vino() == in->vino()) { | |
1075 | touch_dn(dn); | |
1076 | ldout(cct, 12) << " had dentry " << dname | |
1077 | << " with correct vino " << dn->inode->vino() | |
1078 | << dendl; | |
1079 | } else { | |
1080 | ldout(cct, 12) << " had dentry " << dname | |
1081 | << " with WRONG vino " << dn->inode->vino() | |
1082 | << dendl; | |
1083 | unlink(dn, true, true); // keep dir, keep dentry | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | if (!dn || !dn->inode) { | |
1088 | InodeRef tmp_ref(in); | |
1089 | if (old_dentry) { | |
1090 | if (old_dentry->dir != dir) { | |
1091 | Inode *old_diri = old_dentry->dir->parent_inode; | |
7c673cae FG |
1092 | clear_dir_complete_and_ordered(old_diri, false); |
1093 | } | |
1094 | unlink(old_dentry, dir == old_dentry->dir, false); // drop dentry, keep dir open if its the same dir | |
1095 | } | |
1096 | Inode *diri = dir->parent_inode; | |
7c673cae FG |
1097 | clear_dir_complete_and_ordered(diri, false); |
1098 | dn = link(dir, dname, in, dn); | |
1099 | } | |
1100 | ||
1101 | update_dentry_lease(dn, dlease, from, session); | |
1102 | return dn; | |
1103 | } | |
1104 | ||
1105 | void Client::update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session) | |
1106 | { | |
1107 | utime_t dttl = from; | |
1108 | dttl += (float)dlease->duration_ms / 1000.0; | |
f67539c2 TL |
1109 | |
1110 | ldout(cct, 15) << __func__ << " " << *dn << " " << *dlease << " from " << from << dendl; | |
7c673cae | 1111 | |
11fdf7f2 | 1112 | ceph_assert(dn); |
7c673cae | 1113 | |
9f95a23c | 1114 | if (dlease->mask & CEPH_LEASE_VALID) { |
7c673cae FG |
1115 | if (dttl > dn->lease_ttl) { |
1116 | ldout(cct, 10) << "got dentry lease on " << dn->name | |
1117 | << " dur " << dlease->duration_ms << "ms ttl " << dttl << dendl; | |
1118 | dn->lease_ttl = dttl; | |
1119 | dn->lease_mds = session->mds_num; | |
1120 | dn->lease_seq = dlease->seq; | |
1121 | dn->lease_gen = session->cap_gen; | |
1122 | } | |
1123 | } | |
1124 | dn->cap_shared_gen = dn->dir->parent_inode->shared_gen; | |
f91f0fd5 TL |
1125 | if (dlease->mask & CEPH_LEASE_PRIMARY_LINK) |
1126 | dn->mark_primary(); | |
f67539c2 | 1127 | dn->alternate_name = std::move(dlease->alternate_name); |
7c673cae FG |
1128 | } |
1129 | ||
1130 | ||
1131 | /* | |
1132 | * update MDS location cache for a single inode | |
1133 | */ | |
522d829b | 1134 | void Client::update_dir_dist(Inode *in, DirStat *dst, mds_rank_t from) |
7c673cae FG |
1135 | { |
1136 | // auth | |
1137 | ldout(cct, 20) << "got dirfrag map for " << in->ino << " frag " << dst->frag << " to mds " << dst->auth << dendl; | |
1138 | if (dst->auth >= 0) { | |
1139 | in->fragmap[dst->frag] = dst->auth; | |
1140 | } else { | |
1141 | in->fragmap.erase(dst->frag); | |
1142 | } | |
1143 | if (!in->dirfragtree.is_leaf(dst->frag)) { | |
1144 | in->dirfragtree.force_to_leaf(cct, dst->frag); | |
1145 | _fragmap_remove_non_leaves(in); | |
1146 | } | |
1147 | ||
522d829b TL |
1148 | // replicated, only update from auth mds reply |
1149 | if (from == dst->auth) { | |
1150 | in->dir_replicated = !dst->dist.empty(); | |
1151 | if (!dst->dist.empty()) | |
1152 | in->frag_repmap[dst->frag].assign(dst->dist.begin(), dst->dist.end()) ; | |
1153 | else | |
1154 | in->frag_repmap.erase(dst->frag); | |
1155 | } | |
7c673cae FG |
1156 | } |
1157 | ||
1158 | void Client::clear_dir_complete_and_ordered(Inode *diri, bool complete) | |
1159 | { | |
f91f0fd5 TL |
1160 | if (complete) |
1161 | diri->dir_release_count++; | |
1162 | else | |
1163 | diri->dir_ordered_count++; | |
7c673cae FG |
1164 | if (diri->flags & I_COMPLETE) { |
1165 | if (complete) { | |
1166 | ldout(cct, 10) << " clearing (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
1167 | diri->flags &= ~(I_COMPLETE | I_DIR_ORDERED); | |
1168 | } else { | |
1169 | if (diri->flags & I_DIR_ORDERED) { | |
1170 | ldout(cct, 10) << " clearing I_DIR_ORDERED on " << *diri << dendl; | |
1171 | diri->flags &= ~I_DIR_ORDERED; | |
1172 | } | |
1173 | } | |
1174 | if (diri->dir) | |
1175 | diri->dir->readdir_cache.clear(); | |
1176 | } | |
1177 | } | |
1178 | ||
1179 | /* | |
1180 | * insert results from readdir or lssnap into the metadata cache. | |
1181 | */ | |
1182 | void Client::insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri) { | |
1183 | ||
11fdf7f2 | 1184 | auto& reply = request->reply; |
7c673cae | 1185 | ConnectionRef con = request->reply->get_connection(); |
11fdf7f2 TL |
1186 | uint64_t features; |
1187 | if(session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1188 | features = (uint64_t)-1; | |
1189 | } | |
1190 | else { | |
1191 | features = con->get_features(); | |
1192 | } | |
7c673cae FG |
1193 | |
1194 | dir_result_t *dirp = request->dirp; | |
11fdf7f2 | 1195 | ceph_assert(dirp); |
7c673cae FG |
1196 | |
1197 | // the extra buffer list is only set for readdir and lssnap replies | |
11fdf7f2 | 1198 | auto p = reply->get_extra_bl().cbegin(); |
7c673cae FG |
1199 | if (!p.end()) { |
1200 | // snapdir? | |
1201 | if (request->head.op == CEPH_MDS_OP_LSSNAP) { | |
11fdf7f2 | 1202 | ceph_assert(diri); |
7c673cae FG |
1203 | diri = open_snapdir(diri); |
1204 | } | |
1205 | ||
1206 | // only open dir if we're actually adding stuff to it! | |
1207 | Dir *dir = diri->open_dir(); | |
11fdf7f2 | 1208 | ceph_assert(dir); |
7c673cae FG |
1209 | |
1210 | // dirstat | |
11fdf7f2 | 1211 | DirStat dst(p, features); |
7c673cae FG |
1212 | __u32 numdn; |
1213 | __u16 flags; | |
11fdf7f2 TL |
1214 | decode(numdn, p); |
1215 | decode(flags, p); | |
7c673cae FG |
1216 | |
1217 | bool end = ((unsigned)flags & CEPH_READDIR_FRAG_END); | |
1218 | bool hash_order = ((unsigned)flags & CEPH_READDIR_HASH_ORDER); | |
1219 | ||
1220 | frag_t fg = (unsigned)request->head.args.readdir.frag; | |
1221 | unsigned readdir_offset = dirp->next_offset; | |
1222 | string readdir_start = dirp->last_name; | |
11fdf7f2 | 1223 | ceph_assert(!readdir_start.empty() || readdir_offset == 2); |
7c673cae FG |
1224 | |
1225 | unsigned last_hash = 0; | |
1226 | if (hash_order) { | |
1227 | if (!readdir_start.empty()) { | |
1228 | last_hash = ceph_frag_value(diri->hash_dentry_name(readdir_start)); | |
1229 | } else if (flags & CEPH_READDIR_OFFSET_HASH) { | |
1230 | /* mds understands offset_hash */ | |
1231 | last_hash = (unsigned)request->head.args.readdir.offset_hash; | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | if (fg != dst.frag) { | |
1236 | ldout(cct, 10) << "insert_trace got new frag " << fg << " -> " << dst.frag << dendl; | |
1237 | fg = dst.frag; | |
1238 | if (!hash_order) { | |
1239 | readdir_offset = 2; | |
1240 | readdir_start.clear(); | |
1241 | dirp->offset = dir_result_t::make_fpos(fg, readdir_offset, false); | |
1242 | } | |
1243 | } | |
1244 | ||
1245 | ldout(cct, 10) << __func__ << " " << numdn << " readdir items, end=" << end | |
1246 | << ", hash_order=" << hash_order | |
1247 | << ", readdir_start " << readdir_start | |
1248 | << ", last_hash " << last_hash | |
1249 | << ", next_offset " << readdir_offset << dendl; | |
1250 | ||
1251 | if (diri->snapid != CEPH_SNAPDIR && | |
1252 | fg.is_leftmost() && readdir_offset == 2 && | |
1253 | !(hash_order && last_hash)) { | |
1254 | dirp->release_count = diri->dir_release_count; | |
1255 | dirp->ordered_count = diri->dir_ordered_count; | |
1256 | dirp->start_shared_gen = diri->shared_gen; | |
1257 | dirp->cache_index = 0; | |
1258 | } | |
1259 | ||
1260 | dirp->buffer_frag = fg; | |
1261 | ||
1262 | _readdir_drop_dirp_buffer(dirp); | |
1263 | dirp->buffer.reserve(numdn); | |
1264 | ||
1265 | string dname; | |
1266 | LeaseStat dlease; | |
1267 | for (unsigned i=0; i<numdn; i++) { | |
11fdf7f2 TL |
1268 | decode(dname, p); |
1269 | dlease.decode(p, features); | |
7c673cae FG |
1270 | InodeStat ist(p, features); |
1271 | ||
1272 | ldout(cct, 15) << "" << i << ": '" << dname << "'" << dendl; | |
1273 | ||
1274 | Inode *in = add_update_inode(&ist, request->sent_stamp, session, | |
1275 | request->perms); | |
1276 | Dentry *dn; | |
1277 | if (diri->dir->dentries.count(dname)) { | |
1278 | Dentry *olddn = diri->dir->dentries[dname]; | |
1279 | if (olddn->inode != in) { | |
1280 | // replace incorrect dentry | |
1281 | unlink(olddn, true, true); // keep dir, dentry | |
1282 | dn = link(dir, dname, in, olddn); | |
11fdf7f2 | 1283 | ceph_assert(dn == olddn); |
7c673cae FG |
1284 | } else { |
1285 | // keep existing dn | |
1286 | dn = olddn; | |
1287 | touch_dn(dn); | |
1288 | } | |
1289 | } else { | |
1290 | // new dn | |
1291 | dn = link(dir, dname, in, NULL); | |
1292 | } | |
f67539c2 | 1293 | dn->alternate_name = std::move(dlease.alternate_name); |
7c673cae FG |
1294 | |
1295 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1296 | if (hash_order) { | |
1297 | unsigned hash = ceph_frag_value(diri->hash_dentry_name(dname)); | |
1298 | if (hash != last_hash) | |
1299 | readdir_offset = 2; | |
1300 | last_hash = hash; | |
1301 | dn->offset = dir_result_t::make_fpos(hash, readdir_offset++, true); | |
1302 | } else { | |
1303 | dn->offset = dir_result_t::make_fpos(fg, readdir_offset++, false); | |
1304 | } | |
1305 | // add to readdir cache | |
1306 | if (dirp->release_count == diri->dir_release_count && | |
1307 | dirp->ordered_count == diri->dir_ordered_count && | |
1308 | dirp->start_shared_gen == diri->shared_gen) { | |
1309 | if (dirp->cache_index == dir->readdir_cache.size()) { | |
1310 | if (i == 0) { | |
11fdf7f2 | 1311 | ceph_assert(!dirp->inode->is_complete_and_ordered()); |
7c673cae FG |
1312 | dir->readdir_cache.reserve(dirp->cache_index + numdn); |
1313 | } | |
1314 | dir->readdir_cache.push_back(dn); | |
1315 | } else if (dirp->cache_index < dir->readdir_cache.size()) { | |
1316 | if (dirp->inode->is_complete_and_ordered()) | |
11fdf7f2 | 1317 | ceph_assert(dir->readdir_cache[dirp->cache_index] == dn); |
7c673cae FG |
1318 | else |
1319 | dir->readdir_cache[dirp->cache_index] = dn; | |
1320 | } else { | |
11fdf7f2 | 1321 | ceph_abort_msg("unexpected readdir buffer idx"); |
7c673cae FG |
1322 | } |
1323 | dirp->cache_index++; | |
1324 | } | |
1325 | // add to cached result list | |
f67539c2 | 1326 | dirp->buffer.push_back(dir_result_t::dentry(dn->offset, dname, dn->alternate_name, in)); |
7c673cae FG |
1327 | ldout(cct, 15) << __func__ << " " << hex << dn->offset << dec << ": '" << dname << "' -> " << in->ino << dendl; |
1328 | } | |
1329 | ||
1330 | if (numdn > 0) | |
1331 | dirp->last_name = dname; | |
1332 | if (end) | |
1333 | dirp->next_offset = 2; | |
1334 | else | |
1335 | dirp->next_offset = readdir_offset; | |
1336 | ||
1337 | if (dir->is_empty()) | |
1338 | close_dir(dir); | |
1339 | } | |
1340 | } | |
1341 | ||
1342 | /** insert_trace | |
1343 | * | |
1344 | * insert a trace from a MDS reply into the cache. | |
1345 | */ | |
1346 | Inode* Client::insert_trace(MetaRequest *request, MetaSession *session) | |
1347 | { | |
11fdf7f2 | 1348 | auto& reply = request->reply; |
7c673cae FG |
1349 | int op = request->get_op(); |
1350 | ||
1351 | ldout(cct, 10) << "insert_trace from " << request->sent_stamp << " mds." << session->mds_num | |
1352 | << " is_target=" << (int)reply->head.is_target | |
1353 | << " is_dentry=" << (int)reply->head.is_dentry | |
1354 | << dendl; | |
1355 | ||
11fdf7f2 | 1356 | auto p = reply->get_trace_bl().cbegin(); |
7c673cae FG |
1357 | if (request->got_unsafe) { |
1358 | ldout(cct, 10) << "insert_trace -- already got unsafe; ignoring" << dendl; | |
11fdf7f2 | 1359 | ceph_assert(p.end()); |
7c673cae FG |
1360 | return NULL; |
1361 | } | |
1362 | ||
1363 | if (p.end()) { | |
1364 | ldout(cct, 10) << "insert_trace -- no trace" << dendl; | |
1365 | ||
1366 | Dentry *d = request->dentry(); | |
1367 | if (d) { | |
1368 | Inode *diri = d->dir->parent_inode; | |
7c673cae FG |
1369 | clear_dir_complete_and_ordered(diri, true); |
1370 | } | |
1371 | ||
1372 | if (d && reply->get_result() == 0) { | |
1373 | if (op == CEPH_MDS_OP_RENAME) { | |
1374 | // rename | |
1375 | Dentry *od = request->old_dentry(); | |
1376 | ldout(cct, 10) << " unlinking rename src dn " << od << " for traceless reply" << dendl; | |
11fdf7f2 | 1377 | ceph_assert(od); |
7c673cae FG |
1378 | unlink(od, true, true); // keep dir, dentry |
1379 | } else if (op == CEPH_MDS_OP_RMDIR || | |
1380 | op == CEPH_MDS_OP_UNLINK) { | |
1381 | // unlink, rmdir | |
1382 | ldout(cct, 10) << " unlinking unlink/rmdir dn " << d << " for traceless reply" << dendl; | |
1383 | unlink(d, true, true); // keep dir, dentry | |
1384 | } | |
1385 | } | |
1386 | return NULL; | |
1387 | } | |
1388 | ||
1389 | ConnectionRef con = request->reply->get_connection(); | |
11fdf7f2 TL |
1390 | uint64_t features; |
1391 | if (session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1392 | features = (uint64_t)-1; | |
1393 | } | |
1394 | else { | |
1395 | features = con->get_features(); | |
1396 | } | |
7c673cae FG |
1397 | ldout(cct, 10) << " features 0x" << hex << features << dec << dendl; |
1398 | ||
1399 | // snap trace | |
1400 | SnapRealm *realm = NULL; | |
1401 | if (reply->snapbl.length()) | |
1402 | update_snap_trace(reply->snapbl, &realm); | |
1403 | ||
1404 | ldout(cct, 10) << " hrm " | |
1405 | << " is_target=" << (int)reply->head.is_target | |
1406 | << " is_dentry=" << (int)reply->head.is_dentry | |
1407 | << dendl; | |
1408 | ||
1409 | InodeStat dirst; | |
1410 | DirStat dst; | |
1411 | string dname; | |
1412 | LeaseStat dlease; | |
1413 | InodeStat ist; | |
1414 | ||
1415 | if (reply->head.is_dentry) { | |
1416 | dirst.decode(p, features); | |
11fdf7f2 TL |
1417 | dst.decode(p, features); |
1418 | decode(dname, p); | |
1419 | dlease.decode(p, features); | |
7c673cae FG |
1420 | } |
1421 | ||
1422 | Inode *in = 0; | |
1423 | if (reply->head.is_target) { | |
1424 | ist.decode(p, features); | |
1425 | if (cct->_conf->client_debug_getattr_caps) { | |
1426 | unsigned wanted = 0; | |
1427 | if (op == CEPH_MDS_OP_GETATTR || op == CEPH_MDS_OP_LOOKUP) | |
1428 | wanted = request->head.args.getattr.mask; | |
1429 | else if (op == CEPH_MDS_OP_OPEN || op == CEPH_MDS_OP_CREATE) | |
1430 | wanted = request->head.args.open.mask; | |
1431 | ||
1432 | if ((wanted & CEPH_CAP_XATTR_SHARED) && | |
1433 | !(ist.xattr_version > 0 && ist.xattrbl.length() > 0)) | |
11fdf7f2 | 1434 | ceph_abort_msg("MDS reply does not contain xattrs"); |
7c673cae FG |
1435 | } |
1436 | ||
1437 | in = add_update_inode(&ist, request->sent_stamp, session, | |
1438 | request->perms); | |
1439 | } | |
1440 | ||
1441 | Inode *diri = NULL; | |
1442 | if (reply->head.is_dentry) { | |
1443 | diri = add_update_inode(&dirst, request->sent_stamp, session, | |
1444 | request->perms); | |
522d829b TL |
1445 | mds_rank_t from_mds = mds_rank_t(reply->get_source().num()); |
1446 | update_dir_dist(diri, &dst, from_mds); // dir stat info is attached to .. | |
7c673cae FG |
1447 | |
1448 | if (in) { | |
1449 | Dir *dir = diri->open_dir(); | |
1450 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session, | |
1451 | (op == CEPH_MDS_OP_RENAME) ? request->old_dentry() : NULL); | |
1452 | } else { | |
1453 | Dentry *dn = NULL; | |
1454 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1455 | dn = diri->dir->dentries[dname]; | |
1456 | if (dn->inode) { | |
7c673cae FG |
1457 | clear_dir_complete_and_ordered(diri, false); |
1458 | unlink(dn, true, true); // keep dir, dentry | |
1459 | } | |
1460 | } | |
1461 | if (dlease.duration_ms > 0) { | |
1462 | if (!dn) { | |
1463 | Dir *dir = diri->open_dir(); | |
1464 | dn = link(dir, dname, NULL, NULL); | |
1465 | } | |
1466 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1467 | } | |
1468 | } | |
1469 | } else if (op == CEPH_MDS_OP_LOOKUPSNAP || | |
1470 | op == CEPH_MDS_OP_MKSNAP) { | |
1471 | ldout(cct, 10) << " faking snap lookup weirdness" << dendl; | |
1472 | // fake it for snap lookup | |
1473 | vinodeno_t vino = ist.vino; | |
1474 | vino.snapid = CEPH_SNAPDIR; | |
11fdf7f2 | 1475 | ceph_assert(inode_map.count(vino)); |
7c673cae FG |
1476 | diri = inode_map[vino]; |
1477 | ||
1478 | string dname = request->path.last_dentry(); | |
1479 | ||
1480 | LeaseStat dlease; | |
1481 | dlease.duration_ms = 0; | |
1482 | ||
1483 | if (in) { | |
1484 | Dir *dir = diri->open_dir(); | |
1485 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session); | |
1486 | } else { | |
1487 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1488 | Dentry *dn = diri->dir->dentries[dname]; | |
1489 | if (dn->inode) | |
1490 | unlink(dn, true, true); // keep dir, dentry | |
1491 | } | |
1492 | } | |
1493 | } | |
1494 | ||
1495 | if (in) { | |
1496 | if (op == CEPH_MDS_OP_READDIR || | |
1497 | op == CEPH_MDS_OP_LSSNAP) { | |
1498 | insert_readdir_results(request, session, in); | |
1499 | } else if (op == CEPH_MDS_OP_LOOKUPNAME) { | |
1500 | // hack: return parent inode instead | |
1501 | in = diri; | |
1502 | } | |
1503 | ||
1504 | if (request->dentry() == NULL && in != request->inode()) { | |
1505 | // pin the target inode if its parent dentry is not pinned | |
1506 | request->set_other_inode(in); | |
1507 | } | |
1508 | } | |
1509 | ||
1510 | if (realm) | |
1511 | put_snap_realm(realm); | |
1512 | ||
1513 | request->target = in; | |
1514 | return in; | |
1515 | } | |
1516 | ||
1517 | // ------- | |
1518 | ||
1519 | mds_rank_t Client::choose_target_mds(MetaRequest *req, Inode** phash_diri) | |
1520 | { | |
1521 | mds_rank_t mds = MDS_RANK_NONE; | |
1522 | __u32 hash = 0; | |
1523 | bool is_hash = false; | |
1524 | ||
1525 | Inode *in = NULL; | |
1526 | Dentry *de = NULL; | |
7c673cae FG |
1527 | |
1528 | if (req->resend_mds >= 0) { | |
1529 | mds = req->resend_mds; | |
1530 | req->resend_mds = -1; | |
11fdf7f2 | 1531 | ldout(cct, 10) << __func__ << " resend_mds specified as mds." << mds << dendl; |
7c673cae FG |
1532 | goto out; |
1533 | } | |
1534 | ||
1535 | if (cct->_conf->client_use_random_mds) | |
1536 | goto random_mds; | |
1537 | ||
1538 | in = req->inode(); | |
1539 | de = req->dentry(); | |
1540 | if (in) { | |
11fdf7f2 | 1541 | ldout(cct, 20) << __func__ << " starting with req->inode " << *in << dendl; |
7c673cae FG |
1542 | if (req->path.depth()) { |
1543 | hash = in->hash_dentry_name(req->path[0]); | |
11fdf7f2 | 1544 | ldout(cct, 20) << __func__ << " inode dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1545 | << " on " << req->path[0] |
1546 | << " => " << hash << dendl; | |
1547 | is_hash = true; | |
1548 | } | |
1549 | } else if (de) { | |
1550 | if (de->inode) { | |
1551 | in = de->inode.get(); | |
11fdf7f2 | 1552 | ldout(cct, 20) << __func__ << " starting with req->dentry inode " << *in << dendl; |
7c673cae FG |
1553 | } else { |
1554 | in = de->dir->parent_inode; | |
1555 | hash = in->hash_dentry_name(de->name); | |
11fdf7f2 | 1556 | ldout(cct, 20) << __func__ << " dentry dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1557 | << " on " << de->name |
1558 | << " => " << hash << dendl; | |
1559 | is_hash = true; | |
1560 | } | |
1561 | } | |
1562 | if (in) { | |
1563 | if (in->snapid != CEPH_NOSNAP) { | |
11fdf7f2 | 1564 | ldout(cct, 10) << __func__ << " " << *in << " is snapped, using nonsnap parent" << dendl; |
7c673cae FG |
1565 | while (in->snapid != CEPH_NOSNAP) { |
1566 | if (in->snapid == CEPH_SNAPDIR) | |
1567 | in = in->snapdir_parent.get(); | |
11fdf7f2 | 1568 | else if (!in->dentries.empty()) |
7c673cae FG |
1569 | /* In most cases there will only be one dentry, so getting it |
1570 | * will be the correct action. If there are multiple hard links, | |
1571 | * I think the MDS should be able to redirect as needed*/ | |
1572 | in = in->get_first_parent()->dir->parent_inode; | |
1573 | else { | |
1574 | ldout(cct, 10) << "got unlinked inode, can't look at parent" << dendl; | |
1575 | break; | |
1576 | } | |
1577 | } | |
1578 | is_hash = false; | |
1579 | } | |
1580 | ||
11fdf7f2 | 1581 | ldout(cct, 20) << __func__ << " " << *in << " is_hash=" << is_hash |
7c673cae FG |
1582 | << " hash=" << hash << dendl; |
1583 | ||
f67539c2 | 1584 | if (is_hash && S_ISDIR(in->mode) && (!in->fragmap.empty() || !in->frag_repmap.empty())) { |
7c673cae | 1585 | frag_t fg = in->dirfragtree[hash]; |
f67539c2 TL |
1586 | if (!req->auth_is_best()) { |
1587 | auto repmapit = in->frag_repmap.find(fg); | |
1588 | if (repmapit != in->frag_repmap.end()) { | |
1589 | auto& repmap = repmapit->second; | |
1590 | auto r = ceph::util::generate_random_number<uint64_t>(0, repmap.size()-1); | |
1591 | mds = repmap.at(r); | |
1592 | } | |
1593 | } else if (in->fragmap.count(fg)) { | |
7c673cae FG |
1594 | mds = in->fragmap[fg]; |
1595 | if (phash_diri) | |
1596 | *phash_diri = in; | |
91327a77 | 1597 | } else if (in->auth_cap) { |
f67539c2 | 1598 | req->send_to_auth = true; |
91327a77 AA |
1599 | mds = in->auth_cap->session->mds_num; |
1600 | } | |
1601 | if (mds >= 0) { | |
11fdf7f2 | 1602 | ldout(cct, 10) << __func__ << " from dirfragtree hash" << dendl; |
7c673cae FG |
1603 | goto out; |
1604 | } | |
1605 | } | |
1606 | ||
11fdf7f2 TL |
1607 | if (in->auth_cap && req->auth_is_best()) { |
1608 | mds = in->auth_cap->session->mds_num; | |
1609 | } else if (!in->caps.empty()) { | |
1610 | mds = in->caps.begin()->second.session->mds_num; | |
1611 | } else { | |
7c673cae | 1612 | goto random_mds; |
11fdf7f2 TL |
1613 | } |
1614 | ldout(cct, 10) << __func__ << " from caps on inode " << *in << dendl; | |
7c673cae FG |
1615 | |
1616 | goto out; | |
1617 | } | |
1618 | ||
1619 | random_mds: | |
1620 | if (mds < 0) { | |
1621 | mds = _get_random_up_mds(); | |
1622 | ldout(cct, 10) << "did not get mds through better means, so chose random mds " << mds << dendl; | |
1623 | } | |
1624 | ||
1625 | out: | |
1626 | ldout(cct, 20) << "mds is " << mds << dendl; | |
1627 | return mds; | |
1628 | } | |
1629 | ||
7c673cae FG |
1630 | void Client::connect_mds_targets(mds_rank_t mds) |
1631 | { | |
11fdf7f2 TL |
1632 | ldout(cct, 10) << __func__ << " for mds." << mds << dendl; |
1633 | ceph_assert(mds_sessions.count(mds)); | |
7c673cae | 1634 | const MDSMap::mds_info_t& info = mdsmap->get_mds_info(mds); |
f67539c2 TL |
1635 | for (const auto &rank : info.export_targets) { |
1636 | if (mds_sessions.count(rank) == 0 && | |
1637 | mdsmap->is_clientreplay_or_active_or_stopping(rank)) { | |
7c673cae | 1638 | ldout(cct, 10) << "check_mds_sessions opening mds." << mds |
f67539c2 TL |
1639 | << " export target mds." << rank << dendl; |
1640 | _open_mds_session(rank); | |
7c673cae FG |
1641 | } |
1642 | } | |
1643 | } | |
1644 | ||
adb31ebb | 1645 | void Client::dump_mds_sessions(Formatter *f, bool cap_dump) |
7c673cae FG |
1646 | { |
1647 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 1648 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 AA |
1649 | f->dump_object("inst", inst); |
1650 | f->dump_stream("inst_str") << inst; | |
1651 | f->dump_stream("addr_str") << inst.addr; | |
7c673cae | 1652 | f->open_array_section("sessions"); |
11fdf7f2 | 1653 | for (const auto &p : mds_sessions) { |
7c673cae | 1654 | f->open_object_section("session"); |
20effc67 | 1655 | p.second->dump(f, cap_dump); |
7c673cae FG |
1656 | f->close_section(); |
1657 | } | |
1658 | f->close_section(); | |
1659 | f->dump_int("mdsmap_epoch", mdsmap->get_epoch()); | |
1660 | } | |
f67539c2 | 1661 | |
7c673cae FG |
1662 | void Client::dump_mds_requests(Formatter *f) |
1663 | { | |
1664 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
1665 | p != mds_requests.end(); | |
1666 | ++p) { | |
1667 | f->open_object_section("request"); | |
1668 | p->second->dump(f); | |
1669 | f->close_section(); | |
1670 | } | |
1671 | } | |
1672 | ||
9f95a23c | 1673 | int Client::verify_reply_trace(int r, MetaSession *session, |
11fdf7f2 | 1674 | MetaRequest *request, const MConstRef<MClientReply>& reply, |
7c673cae FG |
1675 | InodeRef *ptarget, bool *pcreated, |
1676 | const UserPerm& perms) | |
1677 | { | |
1678 | // check whether this request actually did the create, and set created flag | |
1679 | bufferlist extra_bl; | |
1680 | inodeno_t created_ino; | |
1681 | bool got_created_ino = false; | |
1682 | ceph::unordered_map<vinodeno_t, Inode*>::iterator p; | |
1683 | ||
11fdf7f2 | 1684 | extra_bl = reply->get_extra_bl(); |
7c673cae | 1685 | if (extra_bl.length() >= 8) { |
9f95a23c TL |
1686 | if (session->mds_features.test(CEPHFS_FEATURE_DELEG_INO)) { |
1687 | struct openc_response_t ocres; | |
1688 | ||
1689 | decode(ocres, extra_bl); | |
1690 | created_ino = ocres.created_ino; | |
1691 | /* | |
1692 | * The userland cephfs client doesn't have a way to do an async create | |
1693 | * (yet), so just discard delegated_inos for now. Eventually we should | |
1694 | * store them and use them in create calls, even if they are synchronous, | |
1695 | * if only for testing purposes. | |
1696 | */ | |
1697 | ldout(cct, 10) << "delegated_inos: " << ocres.delegated_inos << dendl; | |
1698 | } else { | |
1699 | // u64 containing number of created ino | |
1700 | decode(created_ino, extra_bl); | |
1701 | } | |
7c673cae | 1702 | ldout(cct, 10) << "make_request created ino " << created_ino << dendl; |
9f95a23c | 1703 | got_created_ino = true; |
7c673cae FG |
1704 | } |
1705 | ||
1706 | if (pcreated) | |
1707 | *pcreated = got_created_ino; | |
1708 | ||
1709 | if (request->target) { | |
1710 | *ptarget = request->target; | |
1711 | ldout(cct, 20) << "make_request target is " << *ptarget->get() << dendl; | |
1712 | } else { | |
1713 | if (got_created_ino && (p = inode_map.find(vinodeno_t(created_ino, CEPH_NOSNAP))) != inode_map.end()) { | |
1714 | (*ptarget) = p->second; | |
1715 | ldout(cct, 20) << "make_request created, target is " << *ptarget->get() << dendl; | |
1716 | } else { | |
1717 | // we got a traceless reply, and need to look up what we just | |
1718 | // created. for now, do this by name. someday, do this by the | |
1719 | // ino... which we know! FIXME. | |
1720 | InodeRef target; | |
1721 | Dentry *d = request->dentry(); | |
1722 | if (d) { | |
1723 | if (d->dir) { | |
1724 | ldout(cct, 10) << "make_request got traceless reply, looking up #" | |
1725 | << d->dir->parent_inode->ino << "/" << d->name | |
1726 | << " got_ino " << got_created_ino | |
1727 | << " ino " << created_ino | |
1728 | << dendl; | |
1729 | r = _do_lookup(d->dir->parent_inode, d->name, request->regetattr_mask, | |
1730 | &target, perms); | |
1731 | } else { | |
1732 | // if the dentry is not linked, just do our best. see #5021. | |
11fdf7f2 | 1733 | ceph_abort_msg("how did this happen? i want logs!"); |
7c673cae FG |
1734 | } |
1735 | } else { | |
1736 | Inode *in = request->inode(); | |
1737 | ldout(cct, 10) << "make_request got traceless reply, forcing getattr on #" | |
1738 | << in->ino << dendl; | |
1739 | r = _getattr(in, request->regetattr_mask, perms, true); | |
1740 | target = in; | |
1741 | } | |
1742 | if (r >= 0) { | |
1743 | // verify ino returned in reply and trace_dist are the same | |
1744 | if (got_created_ino && | |
1745 | created_ino.val != target->ino.val) { | |
1746 | ldout(cct, 5) << "create got ino " << created_ino << " but then failed on lookup; EINTR?" << dendl; | |
f67539c2 | 1747 | r = -CEPHFS_EINTR; |
7c673cae FG |
1748 | } |
1749 | if (ptarget) | |
1750 | ptarget->swap(target); | |
1751 | } | |
1752 | } | |
1753 | } | |
1754 | ||
1755 | return r; | |
1756 | } | |
1757 | ||
1758 | ||
1759 | /** | |
1760 | * make a request | |
1761 | * | |
1762 | * Blocking helper to make an MDS request. | |
1763 | * | |
1764 | * If the ptarget flag is set, behavior changes slightly: the caller | |
1765 | * expects to get a pointer to the inode we are creating or operating | |
1766 | * on. As a result, we will follow up any traceless mutation reply | |
1767 | * with a getattr or lookup to transparently handle a traceless reply | |
1768 | * from the MDS (as when the MDS restarts and the client has to replay | |
1769 | * a request). | |
1770 | * | |
1771 | * @param request the MetaRequest to execute | |
1772 | * @param perms The user uid/gid to execute as (eventually, full group lists?) | |
1773 | * @param ptarget [optional] address to store a pointer to the target inode we want to create or operate on | |
1774 | * @param pcreated [optional; required if ptarget] where to store a bool of whether our create atomically created a file | |
1775 | * @param use_mds [optional] prefer a specific mds (-1 for default) | |
1776 | * @param pdirbl [optional; disallowed if ptarget] where to pass extra reply payload to the caller | |
1777 | */ | |
1778 | int Client::make_request(MetaRequest *request, | |
1779 | const UserPerm& perms, | |
1780 | InodeRef *ptarget, bool *pcreated, | |
1781 | mds_rank_t use_mds, | |
1782 | bufferlist *pdirbl) | |
1783 | { | |
1784 | int r = 0; | |
1785 | ||
1786 | // assign a unique tid | |
1787 | ceph_tid_t tid = ++last_tid; | |
1788 | request->set_tid(tid); | |
1789 | ||
1790 | // and timestamp | |
1791 | request->op_stamp = ceph_clock_now(); | |
1792 | ||
1793 | // make note | |
1794 | mds_requests[tid] = request->get(); | |
1795 | if (oldest_tid == 0 && request->get_op() != CEPH_MDS_OP_SETFILELOCK) | |
1796 | oldest_tid = tid; | |
1797 | ||
1798 | request->set_caller_perms(perms); | |
1799 | ||
1800 | if (cct->_conf->client_inject_fixed_oldest_tid) { | |
1801 | ldout(cct, 20) << __func__ << " injecting fixed oldest_client_tid(1)" << dendl; | |
1802 | request->set_oldest_client_tid(1); | |
1803 | } else { | |
1804 | request->set_oldest_client_tid(oldest_tid); | |
1805 | } | |
1806 | ||
1807 | // hack target mds? | |
1808 | if (use_mds >= 0) | |
1809 | request->resend_mds = use_mds; | |
1810 | ||
20effc67 | 1811 | MetaSessionRef session = NULL; |
7c673cae FG |
1812 | while (1) { |
1813 | if (request->aborted()) | |
1814 | break; | |
1815 | ||
f67539c2 TL |
1816 | if (blocklisted) { |
1817 | request->abort(-CEPHFS_EBLOCKLISTED); | |
31f18b77 FG |
1818 | break; |
1819 | } | |
1820 | ||
7c673cae | 1821 | // set up wait cond |
9f95a23c | 1822 | ceph::condition_variable caller_cond; |
7c673cae FG |
1823 | request->caller_cond = &caller_cond; |
1824 | ||
1825 | // choose mds | |
1826 | Inode *hash_diri = NULL; | |
1827 | mds_rank_t mds = choose_target_mds(request, &hash_diri); | |
1828 | int mds_state = (mds == MDS_RANK_NONE) ? MDSMap::STATE_NULL : mdsmap->get_state(mds); | |
1829 | if (mds_state != MDSMap::STATE_ACTIVE && mds_state != MDSMap::STATE_STOPPING) { | |
1830 | if (mds_state == MDSMap::STATE_NULL && mds >= mdsmap->get_max_mds()) { | |
1831 | if (hash_diri) { | |
1832 | ldout(cct, 10) << " target mds." << mds << " has stopped, remove it from fragmap" << dendl; | |
1833 | _fragmap_remove_stopped_mds(hash_diri, mds); | |
1834 | } else { | |
1835 | ldout(cct, 10) << " target mds." << mds << " has stopped, trying a random mds" << dendl; | |
1836 | request->resend_mds = _get_random_up_mds(); | |
1837 | } | |
1838 | } else { | |
1839 | ldout(cct, 10) << " target mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
1840 | wait_on_list(waiting_for_mdsmap); | |
1841 | } | |
1842 | continue; | |
1843 | } | |
1844 | ||
1845 | // open a session? | |
7c673cae FG |
1846 | if (!have_open_session(mds)) { |
1847 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 1848 | if (session->state == MetaSession::STATE_REJECTED) { |
f67539c2 | 1849 | request->abort(-CEPHFS_EPERM); |
f6b5b4d7 TL |
1850 | break; |
1851 | } | |
7c673cae FG |
1852 | // wait |
1853 | if (session->state == MetaSession::STATE_OPENING) { | |
1854 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
1855 | wait_on_context_list(session->waiting_for_open); | |
7c673cae FG |
1856 | continue; |
1857 | } | |
1858 | ||
1859 | if (!have_open_session(mds)) | |
1860 | continue; | |
1861 | } else { | |
20effc67 | 1862 | session = mds_sessions.at(mds); |
7c673cae FG |
1863 | } |
1864 | ||
1865 | // send request. | |
20effc67 | 1866 | send_request(request, session.get()); |
7c673cae FG |
1867 | |
1868 | // wait for signal | |
1869 | ldout(cct, 20) << "awaiting reply|forward|kick on " << &caller_cond << dendl; | |
1870 | request->kick = false; | |
9f95a23c TL |
1871 | std::unique_lock l{client_lock, std::adopt_lock}; |
1872 | caller_cond.wait(l, [request] { | |
1873 | return (request->reply || // reply | |
1874 | request->resend_mds >= 0 || // forward | |
1875 | request->kick); | |
1876 | }); | |
1877 | l.release(); | |
1878 | request->caller_cond = nullptr; | |
7c673cae FG |
1879 | |
1880 | // did we get a reply? | |
1881 | if (request->reply) | |
1882 | break; | |
1883 | } | |
1884 | ||
1885 | if (!request->reply) { | |
11fdf7f2 TL |
1886 | ceph_assert(request->aborted()); |
1887 | ceph_assert(!request->got_unsafe); | |
7c673cae FG |
1888 | r = request->get_abort_code(); |
1889 | request->item.remove_myself(); | |
1890 | unregister_request(request); | |
11fdf7f2 | 1891 | put_request(request); |
7c673cae FG |
1892 | return r; |
1893 | } | |
1894 | ||
1895 | // got it! | |
11fdf7f2 | 1896 | auto reply = std::move(request->reply); |
7c673cae FG |
1897 | r = reply->get_result(); |
1898 | if (r >= 0) | |
1899 | request->success = true; | |
1900 | ||
1901 | // kick dispatcher (we've got it!) | |
11fdf7f2 | 1902 | ceph_assert(request->dispatch_cond); |
9f95a23c | 1903 | request->dispatch_cond->notify_all(); |
7c673cae FG |
1904 | ldout(cct, 20) << "sendrecv kickback on tid " << tid << " " << request->dispatch_cond << dendl; |
1905 | request->dispatch_cond = 0; | |
1906 | ||
1907 | if (r >= 0 && ptarget) | |
20effc67 | 1908 | r = verify_reply_trace(r, session.get(), request, reply, ptarget, pcreated, perms); |
7c673cae FG |
1909 | |
1910 | if (pdirbl) | |
11fdf7f2 | 1911 | *pdirbl = reply->get_extra_bl(); |
7c673cae FG |
1912 | |
1913 | // -- log times -- | |
1914 | utime_t lat = ceph_clock_now(); | |
1915 | lat -= request->sent_stamp; | |
1916 | ldout(cct, 20) << "lat " << lat << dendl; | |
1917 | logger->tinc(l_c_lat, lat); | |
1918 | logger->tinc(l_c_reply, lat); | |
1919 | ||
1920 | put_request(request); | |
7c673cae FG |
1921 | return r; |
1922 | } | |
1923 | ||
1924 | void Client::unregister_request(MetaRequest *req) | |
1925 | { | |
1926 | mds_requests.erase(req->tid); | |
1927 | if (req->tid == oldest_tid) { | |
1928 | map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.upper_bound(oldest_tid); | |
1929 | while (true) { | |
1930 | if (p == mds_requests.end()) { | |
1931 | oldest_tid = 0; | |
1932 | break; | |
1933 | } | |
1934 | if (p->second->get_op() != CEPH_MDS_OP_SETFILELOCK) { | |
1935 | oldest_tid = p->first; | |
1936 | break; | |
1937 | } | |
1938 | ++p; | |
1939 | } | |
1940 | } | |
1941 | put_request(req); | |
1942 | } | |
1943 | ||
1944 | void Client::put_request(MetaRequest *request) | |
1945 | { | |
1946 | if (request->_put()) { | |
1947 | int op = -1; | |
1948 | if (request->success) | |
1949 | op = request->get_op(); | |
1950 | InodeRef other_in; | |
1951 | request->take_other_inode(&other_in); | |
1952 | delete request; | |
1953 | ||
1954 | if (other_in && | |
1955 | (op == CEPH_MDS_OP_RMDIR || | |
1956 | op == CEPH_MDS_OP_RENAME || | |
1957 | op == CEPH_MDS_OP_RMSNAP)) { | |
1958 | _try_to_trim_inode(other_in.get(), false); | |
1959 | } | |
1960 | } | |
1961 | } | |
1962 | ||
1963 | int Client::encode_inode_release(Inode *in, MetaRequest *req, | |
1964 | mds_rank_t mds, int drop, | |
1965 | int unless, int force) | |
1966 | { | |
11fdf7f2 | 1967 | ldout(cct, 20) << __func__ << " enter(in:" << *in << ", req:" << req |
f67539c2 | 1968 | << " mds:" << mds << ", drop:" << ccap_string(drop) << ", unless:" << ccap_string(unless) |
1911f103 | 1969 | << ", force:" << force << ")" << dendl; |
7c673cae | 1970 | int released = 0; |
11fdf7f2 TL |
1971 | auto it = in->caps.find(mds); |
1972 | if (it != in->caps.end()) { | |
1973 | Cap &cap = it->second; | |
7c673cae | 1974 | drop &= ~(in->dirty_caps | get_caps_used(in)); |
11fdf7f2 TL |
1975 | if ((drop & cap.issued) && |
1976 | !(unless & cap.issued)) { | |
1911f103 | 1977 | ldout(cct, 25) << "dropping caps " << ccap_string(drop) << dendl; |
11fdf7f2 TL |
1978 | cap.issued &= ~drop; |
1979 | cap.implemented &= ~drop; | |
7c673cae | 1980 | released = 1; |
7c673cae FG |
1981 | } else { |
1982 | released = force; | |
1983 | } | |
1984 | if (released) { | |
1911f103 TL |
1985 | cap.wanted = in->caps_wanted(); |
1986 | if (&cap == in->auth_cap && | |
1987 | !(cap.wanted & CEPH_CAP_ANY_FILE_WR)) { | |
1988 | in->requested_max_size = 0; | |
1989 | ldout(cct, 25) << "reset requested_max_size due to not wanting any file write cap" << dendl; | |
1990 | } | |
7c673cae FG |
1991 | ceph_mds_request_release rel; |
1992 | rel.ino = in->ino; | |
11fdf7f2 TL |
1993 | rel.cap_id = cap.cap_id; |
1994 | rel.seq = cap.seq; | |
1995 | rel.issue_seq = cap.issue_seq; | |
1996 | rel.mseq = cap.mseq; | |
1997 | rel.caps = cap.implemented; | |
1998 | rel.wanted = cap.wanted; | |
7c673cae FG |
1999 | rel.dname_len = 0; |
2000 | rel.dname_seq = 0; | |
2001 | req->cap_releases.push_back(MClientRequest::Release(rel,"")); | |
2002 | } | |
2003 | } | |
11fdf7f2 | 2004 | ldout(cct, 25) << __func__ << " exit(in:" << *in << ") released:" |
7c673cae FG |
2005 | << released << dendl; |
2006 | return released; | |
2007 | } | |
2008 | ||
2009 | void Client::encode_dentry_release(Dentry *dn, MetaRequest *req, | |
2010 | mds_rank_t mds, int drop, int unless) | |
2011 | { | |
11fdf7f2 | 2012 | ldout(cct, 20) << __func__ << " enter(dn:" |
7c673cae FG |
2013 | << dn << ")" << dendl; |
2014 | int released = 0; | |
2015 | if (dn->dir) | |
2016 | released = encode_inode_release(dn->dir->parent_inode, req, | |
2017 | mds, drop, unless, 1); | |
2018 | if (released && dn->lease_mds == mds) { | |
2019 | ldout(cct, 25) << "preemptively releasing dn to mds" << dendl; | |
11fdf7f2 | 2020 | auto& rel = req->cap_releases.back(); |
7c673cae FG |
2021 | rel.item.dname_len = dn->name.length(); |
2022 | rel.item.dname_seq = dn->lease_seq; | |
2023 | rel.dname = dn->name; | |
adb31ebb | 2024 | dn->lease_mds = -1; |
7c673cae | 2025 | } |
11fdf7f2 | 2026 | ldout(cct, 25) << __func__ << " exit(dn:" |
7c673cae FG |
2027 | << dn << ")" << dendl; |
2028 | } | |
2029 | ||
2030 | ||
2031 | /* | |
2032 | * This requires the MClientRequest *request member to be set. | |
2033 | * It will error out horribly without one. | |
2034 | * Additionally, if you set any *drop member, you'd better have | |
2035 | * set the corresponding dentry! | |
2036 | */ | |
2037 | void Client::encode_cap_releases(MetaRequest *req, mds_rank_t mds) | |
2038 | { | |
11fdf7f2 | 2039 | ldout(cct, 20) << __func__ << " enter (req: " |
7c673cae FG |
2040 | << req << ", mds: " << mds << ")" << dendl; |
2041 | if (req->inode_drop && req->inode()) | |
2042 | encode_inode_release(req->inode(), req, | |
2043 | mds, req->inode_drop, | |
2044 | req->inode_unless); | |
2045 | ||
2046 | if (req->old_inode_drop && req->old_inode()) | |
2047 | encode_inode_release(req->old_inode(), req, | |
2048 | mds, req->old_inode_drop, | |
2049 | req->old_inode_unless); | |
2050 | if (req->other_inode_drop && req->other_inode()) | |
2051 | encode_inode_release(req->other_inode(), req, | |
2052 | mds, req->other_inode_drop, | |
2053 | req->other_inode_unless); | |
2054 | ||
2055 | if (req->dentry_drop && req->dentry()) | |
2056 | encode_dentry_release(req->dentry(), req, | |
2057 | mds, req->dentry_drop, | |
2058 | req->dentry_unless); | |
2059 | ||
2060 | if (req->old_dentry_drop && req->old_dentry()) | |
2061 | encode_dentry_release(req->old_dentry(), req, | |
2062 | mds, req->old_dentry_drop, | |
2063 | req->old_dentry_unless); | |
11fdf7f2 | 2064 | ldout(cct, 25) << __func__ << " exit (req: " |
7c673cae FG |
2065 | << req << ", mds " << mds <<dendl; |
2066 | } | |
2067 | ||
2068 | bool Client::have_open_session(mds_rank_t mds) | |
2069 | { | |
11fdf7f2 TL |
2070 | const auto &it = mds_sessions.find(mds); |
2071 | return it != mds_sessions.end() && | |
20effc67 TL |
2072 | (it->second->state == MetaSession::STATE_OPEN || |
2073 | it->second->state == MetaSession::STATE_STALE); | |
7c673cae FG |
2074 | } |
2075 | ||
20effc67 | 2076 | MetaSessionRef Client::_get_mds_session(mds_rank_t mds, Connection *con) |
7c673cae | 2077 | { |
11fdf7f2 | 2078 | const auto &it = mds_sessions.find(mds); |
20effc67 | 2079 | if (it == mds_sessions.end() || it->second->con != con) { |
7c673cae | 2080 | return NULL; |
11fdf7f2 | 2081 | } else { |
20effc67 | 2082 | return it->second; |
11fdf7f2 | 2083 | } |
7c673cae FG |
2084 | } |
2085 | ||
20effc67 | 2086 | MetaSessionRef Client::_get_or_open_mds_session(mds_rank_t mds) |
7c673cae | 2087 | { |
11fdf7f2 | 2088 | auto it = mds_sessions.find(mds); |
20effc67 | 2089 | return it == mds_sessions.end() ? _open_mds_session(mds) : it->second; |
7c673cae FG |
2090 | } |
2091 | ||
2092 | /** | |
2093 | * Populate a map of strings with client-identifying metadata, | |
2094 | * such as the hostname. Call this once at initialization. | |
2095 | */ | |
2096 | void Client::populate_metadata(const std::string &mount_root) | |
2097 | { | |
2098 | // Hostname | |
f67539c2 TL |
2099 | #ifdef _WIN32 |
2100 | // TODO: move this to compat.h | |
2101 | char hostname[64]; | |
2102 | DWORD hostname_sz = 64; | |
2103 | GetComputerNameA(hostname, &hostname_sz); | |
2104 | metadata["hostname"] = hostname; | |
2105 | #else | |
7c673cae FG |
2106 | struct utsname u; |
2107 | int r = uname(&u); | |
2108 | if (r >= 0) { | |
2109 | metadata["hostname"] = u.nodename; | |
2110 | ldout(cct, 20) << __func__ << " read hostname '" << u.nodename << "'" << dendl; | |
2111 | } else { | |
2112 | ldout(cct, 1) << __func__ << " failed to read hostname (" << cpp_strerror(r) << ")" << dendl; | |
2113 | } | |
f67539c2 | 2114 | #endif |
7c673cae FG |
2115 | |
2116 | metadata["pid"] = stringify(getpid()); | |
2117 | ||
2118 | // Ceph entity id (the '0' in "client.0") | |
2119 | metadata["entity_id"] = cct->_conf->name.get_id(); | |
2120 | ||
2121 | // Our mount position | |
2122 | if (!mount_root.empty()) { | |
2123 | metadata["root"] = mount_root; | |
2124 | } | |
2125 | ||
2126 | // Ceph version | |
2127 | metadata["ceph_version"] = pretty_version_to_str(); | |
2128 | metadata["ceph_sha1"] = git_version_to_str(); | |
2129 | ||
2130 | // Apply any metadata from the user's configured overrides | |
2131 | std::vector<std::string> tokens; | |
2132 | get_str_vec(cct->_conf->client_metadata, ",", tokens); | |
2133 | for (const auto &i : tokens) { | |
2134 | auto eqpos = i.find("="); | |
2135 | // Throw out anything that isn't of the form "<str>=<str>" | |
2136 | if (eqpos == 0 || eqpos == std::string::npos || eqpos == i.size()) { | |
2137 | lderr(cct) << "Invalid metadata keyval pair: '" << i << "'" << dendl; | |
2138 | continue; | |
2139 | } | |
2140 | metadata[i.substr(0, eqpos)] = i.substr(eqpos + 1); | |
2141 | } | |
2142 | } | |
2143 | ||
2144 | /** | |
2145 | * Optionally add or override client metadata fields. | |
2146 | */ | |
2147 | void Client::update_metadata(std::string const &k, std::string const &v) | |
2148 | { | |
f67539c2 TL |
2149 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2150 | ceph_assert(iref_reader.is_state_satisfied()); | |
2151 | ||
2152 | std::scoped_lock l(client_lock); | |
7c673cae | 2153 | |
11fdf7f2 TL |
2154 | auto it = metadata.find(k); |
2155 | if (it != metadata.end()) { | |
7c673cae | 2156 | ldout(cct, 1) << __func__ << " warning, overriding metadata field '" << k |
11fdf7f2 | 2157 | << "' from '" << it->second << "' to '" << v << "'" << dendl; |
7c673cae FG |
2158 | } |
2159 | ||
2160 | metadata[k] = v; | |
2161 | } | |
2162 | ||
20effc67 | 2163 | MetaSessionRef Client::_open_mds_session(mds_rank_t mds) |
7c673cae | 2164 | { |
11fdf7f2 TL |
2165 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
2166 | auto addrs = mdsmap->get_addrs(mds); | |
2167 | auto em = mds_sessions.emplace(std::piecewise_construct, | |
2168 | std::forward_as_tuple(mds), | |
20effc67 | 2169 | std::forward_as_tuple(new MetaSession(mds, messenger->connect_to_mds(addrs), addrs))); |
11fdf7f2 | 2170 | ceph_assert(em.second); /* not already present */ |
20effc67 | 2171 | auto session = em.first->second; |
7c673cae | 2172 | |
9f95a23c | 2173 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_OPEN); |
11fdf7f2 TL |
2174 | m->metadata = metadata; |
2175 | m->supported_features = feature_bitset_t(CEPHFS_FEATURES_CLIENT_SUPPORTED); | |
f67539c2 | 2176 | m->metric_spec = feature_bitset_t(CEPHFS_METRIC_FEATURES_ALL); |
11fdf7f2 | 2177 | session->con->send_message2(std::move(m)); |
7c673cae FG |
2178 | return session; |
2179 | } | |
2180 | ||
2181 | void Client::_close_mds_session(MetaSession *s) | |
2182 | { | |
11fdf7f2 | 2183 | ldout(cct, 2) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
7c673cae | 2184 | s->state = MetaSession::STATE_CLOSING; |
9f95a23c | 2185 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
2186 | } |
2187 | ||
f6b5b4d7 | 2188 | void Client::_closed_mds_session(MetaSession *s, int err, bool rejected) |
7c673cae | 2189 | { |
11fdf7f2 | 2190 | ldout(cct, 5) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
f6b5b4d7 TL |
2191 | if (rejected && s->state != MetaSession::STATE_CLOSING) |
2192 | s->state = MetaSession::STATE_REJECTED; | |
2193 | else | |
2194 | s->state = MetaSession::STATE_CLOSED; | |
7c673cae FG |
2195 | s->con->mark_down(); |
2196 | signal_context_list(s->waiting_for_open); | |
9f95a23c | 2197 | mount_cond.notify_all(); |
f6b5b4d7 | 2198 | remove_session_caps(s, err); |
7c673cae | 2199 | kick_requests_closed(s); |
f6b5b4d7 TL |
2200 | mds_ranks_closing.erase(s->mds_num); |
2201 | if (s->state == MetaSession::STATE_CLOSED) | |
2202 | mds_sessions.erase(s->mds_num); | |
7c673cae FG |
2203 | } |
2204 | ||
11fdf7f2 | 2205 | void Client::handle_client_session(const MConstRef<MClientSession>& m) |
7c673cae FG |
2206 | { |
2207 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
11fdf7f2 | 2208 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << from << dendl; |
7c673cae | 2209 | |
f67539c2 | 2210 | std::scoped_lock cl(client_lock); |
20effc67 | 2211 | auto session = _get_mds_session(from, m->get_connection().get()); |
7c673cae FG |
2212 | if (!session) { |
2213 | ldout(cct, 10) << " discarding session message from sessionless mds " << m->get_source_inst() << dendl; | |
7c673cae FG |
2214 | return; |
2215 | } | |
2216 | ||
2217 | switch (m->get_op()) { | |
2218 | case CEPH_SESSION_OPEN: | |
11fdf7f2 TL |
2219 | { |
2220 | feature_bitset_t missing_features(CEPHFS_FEATURES_CLIENT_REQUIRED); | |
2221 | missing_features -= m->supported_features; | |
2222 | if (!missing_features.empty()) { | |
2223 | lderr(cct) << "mds." << from << " lacks required features '" | |
2224 | << missing_features << "', closing session " << dendl; | |
20effc67 TL |
2225 | _close_mds_session(session.get()); |
2226 | _closed_mds_session(session.get(), -CEPHFS_EPERM, true); | |
11fdf7f2 TL |
2227 | break; |
2228 | } | |
2229 | session->mds_features = std::move(m->supported_features); | |
2230 | ||
20effc67 | 2231 | renew_caps(session.get()); |
11fdf7f2 | 2232 | session->state = MetaSession::STATE_OPEN; |
f67539c2 | 2233 | if (is_unmounting()) |
9f95a23c | 2234 | mount_cond.notify_all(); |
11fdf7f2 TL |
2235 | else |
2236 | connect_mds_targets(from); | |
2237 | signal_context_list(session->waiting_for_open); | |
2238 | break; | |
2239 | } | |
7c673cae FG |
2240 | |
2241 | case CEPH_SESSION_CLOSE: | |
20effc67 | 2242 | _closed_mds_session(session.get()); |
7c673cae FG |
2243 | break; |
2244 | ||
2245 | case CEPH_SESSION_RENEWCAPS: | |
2246 | if (session->cap_renew_seq == m->get_seq()) { | |
a8e16298 | 2247 | bool was_stale = ceph_clock_now() >= session->cap_ttl; |
7c673cae FG |
2248 | session->cap_ttl = |
2249 | session->last_cap_renew_request + mdsmap->get_session_timeout(); | |
a8e16298 | 2250 | if (was_stale) |
20effc67 | 2251 | wake_up_session_caps(session.get(), false); |
7c673cae FG |
2252 | } |
2253 | break; | |
2254 | ||
2255 | case CEPH_SESSION_STALE: | |
28e407b8 AA |
2256 | // invalidate session caps/leases |
2257 | session->cap_gen++; | |
2258 | session->cap_ttl = ceph_clock_now(); | |
2259 | session->cap_ttl -= 1; | |
20effc67 | 2260 | renew_caps(session.get()); |
7c673cae FG |
2261 | break; |
2262 | ||
2263 | case CEPH_SESSION_RECALL_STATE: | |
f67539c2 TL |
2264 | /* |
2265 | * Call the renew caps and flush cap releases just before | |
2266 | * triming the caps in case the tick() won't get a chance | |
2267 | * to run them, which could cause the client to be blocklisted | |
2268 | * and MDS daemons trying to recall the caps again and | |
2269 | * again. | |
2270 | * | |
2271 | * In most cases it will do nothing, and the new cap releases | |
2272 | * added by trim_caps() followed will be deferred flushing | |
2273 | * by tick(). | |
2274 | */ | |
2275 | renew_and_flush_cap_releases(); | |
20effc67 | 2276 | trim_caps(session.get(), m->get_max_caps()); |
7c673cae FG |
2277 | break; |
2278 | ||
2279 | case CEPH_SESSION_FLUSHMSG: | |
a8e16298 | 2280 | /* flush cap release */ |
11fdf7f2 TL |
2281 | if (auto& m = session->release; m) { |
2282 | session->con->send_message2(std::move(m)); | |
a8e16298 | 2283 | } |
9f95a23c | 2284 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_FLUSHMSG_ACK, m->get_seq())); |
7c673cae FG |
2285 | break; |
2286 | ||
2287 | case CEPH_SESSION_FORCE_RO: | |
20effc67 | 2288 | force_session_readonly(session.get()); |
7c673cae FG |
2289 | break; |
2290 | ||
2291 | case CEPH_SESSION_REJECT: | |
11fdf7f2 TL |
2292 | { |
2293 | std::string_view error_str; | |
2294 | auto it = m->metadata.find("error_string"); | |
2295 | if (it != m->metadata.end()) | |
2296 | error_str = it->second; | |
2297 | else | |
2298 | error_str = "unknown error"; | |
2299 | lderr(cct) << "mds." << from << " rejected us (" << error_str << ")" << dendl; | |
7c673cae | 2300 | |
20effc67 | 2301 | _closed_mds_session(session.get(), -CEPHFS_EPERM, true); |
11fdf7f2 | 2302 | } |
7c673cae FG |
2303 | break; |
2304 | ||
2305 | default: | |
2306 | ceph_abort(); | |
2307 | } | |
7c673cae FG |
2308 | } |
2309 | ||
2310 | bool Client::_any_stale_sessions() const | |
2311 | { | |
9f95a23c | 2312 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae | 2313 | |
11fdf7f2 | 2314 | for (const auto &p : mds_sessions) { |
20effc67 | 2315 | if (p.second->state == MetaSession::STATE_STALE) { |
7c673cae FG |
2316 | return true; |
2317 | } | |
2318 | } | |
2319 | ||
2320 | return false; | |
2321 | } | |
2322 | ||
2323 | void Client::_kick_stale_sessions() | |
2324 | { | |
11fdf7f2 | 2325 | ldout(cct, 1) << __func__ << dendl; |
7c673cae | 2326 | |
11fdf7f2 | 2327 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
20effc67 TL |
2328 | auto s = it->second; |
2329 | if (s->state == MetaSession::STATE_REJECTED) { | |
2330 | mds_sessions.erase(it->first); | |
f6b5b4d7 TL |
2331 | continue; |
2332 | } | |
20effc67 TL |
2333 | if (s->state == MetaSession::STATE_STALE) |
2334 | _closed_mds_session(s.get()); | |
7c673cae FG |
2335 | } |
2336 | } | |
2337 | ||
2338 | void Client::send_request(MetaRequest *request, MetaSession *session, | |
2339 | bool drop_cap_releases) | |
2340 | { | |
2341 | // make the request | |
2342 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2343 | ldout(cct, 10) << __func__ << " rebuilding request " << request->get_tid() |
7c673cae | 2344 | << " for mds." << mds << dendl; |
11fdf7f2 | 2345 | auto r = build_client_request(request); |
7c673cae FG |
2346 | if (request->dentry()) { |
2347 | r->set_dentry_wanted(); | |
2348 | } | |
2349 | if (request->got_unsafe) { | |
2350 | r->set_replayed_op(); | |
2351 | if (request->target) | |
2352 | r->head.ino = request->target->ino; | |
2353 | } else { | |
2354 | encode_cap_releases(request, mds); | |
2355 | if (drop_cap_releases) // we haven't send cap reconnect yet, drop cap releases | |
2356 | request->cap_releases.clear(); | |
2357 | else | |
2358 | r->releases.swap(request->cap_releases); | |
2359 | } | |
2360 | r->set_mdsmap_epoch(mdsmap->get_epoch()); | |
2361 | if (r->head.op == CEPH_MDS_OP_SETXATTR) { | |
2362 | objecter->with_osdmap([r](const OSDMap& o) { | |
2363 | r->set_osdmap_epoch(o.get_epoch()); | |
2364 | }); | |
2365 | } | |
2366 | ||
2367 | if (request->mds == -1) { | |
2368 | request->sent_stamp = ceph_clock_now(); | |
11fdf7f2 | 2369 | ldout(cct, 20) << __func__ << " set sent_stamp to " << request->sent_stamp << dendl; |
7c673cae FG |
2370 | } |
2371 | request->mds = mds; | |
2372 | ||
2373 | Inode *in = request->inode(); | |
11fdf7f2 TL |
2374 | if (in) { |
2375 | auto it = in->caps.find(mds); | |
2376 | if (it != in->caps.end()) { | |
2377 | request->sent_on_mseq = it->second.mseq; | |
2378 | } | |
2379 | } | |
7c673cae FG |
2380 | |
2381 | session->requests.push_back(&request->item); | |
2382 | ||
11fdf7f2 TL |
2383 | ldout(cct, 10) << __func__ << " " << *r << " to mds." << mds << dendl; |
2384 | session->con->send_message2(std::move(r)); | |
7c673cae FG |
2385 | } |
2386 | ||
9f95a23c | 2387 | ref_t<MClientRequest> Client::build_client_request(MetaRequest *request) |
7c673cae | 2388 | { |
9f95a23c | 2389 | auto req = make_message<MClientRequest>(request->get_op()); |
7c673cae FG |
2390 | req->set_tid(request->tid); |
2391 | req->set_stamp(request->op_stamp); | |
2392 | memcpy(&req->head, &request->head, sizeof(ceph_mds_request_head)); | |
2393 | ||
2394 | // if the filepath's haven't been set, set them! | |
2395 | if (request->path.empty()) { | |
2396 | Inode *in = request->inode(); | |
2397 | Dentry *de = request->dentry(); | |
2398 | if (in) | |
2399 | in->make_nosnap_relative_path(request->path); | |
2400 | else if (de) { | |
2401 | if (de->inode) | |
2402 | de->inode->make_nosnap_relative_path(request->path); | |
2403 | else if (de->dir) { | |
2404 | de->dir->parent_inode->make_nosnap_relative_path(request->path); | |
2405 | request->path.push_dentry(de->name); | |
2406 | } | |
2407 | else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2408 | << " No path, inode, or appropriately-endowed dentry given!" | |
2409 | << dendl; | |
2410 | } else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2411 | << " No path, inode, or dentry given!" | |
2412 | << dendl; | |
2413 | } | |
2414 | req->set_filepath(request->get_filepath()); | |
2415 | req->set_filepath2(request->get_filepath2()); | |
f67539c2 | 2416 | req->set_alternate_name(request->alternate_name); |
7c673cae FG |
2417 | req->set_data(request->data); |
2418 | req->set_retry_attempt(request->retry_attempt++); | |
2419 | req->head.num_fwd = request->num_fwd; | |
2420 | const gid_t *_gids; | |
2421 | int gid_count = request->perms.get_gids(&_gids); | |
2422 | req->set_gid_list(gid_count, _gids); | |
2423 | return req; | |
2424 | } | |
2425 | ||
2426 | ||
2427 | ||
11fdf7f2 | 2428 | void Client::handle_client_request_forward(const MConstRef<MClientRequestForward>& fwd) |
7c673cae FG |
2429 | { |
2430 | mds_rank_t mds = mds_rank_t(fwd->get_source().num()); | |
f67539c2 TL |
2431 | |
2432 | std::scoped_lock cl(client_lock); | |
20effc67 | 2433 | auto session = _get_mds_session(mds, fwd->get_connection().get()); |
7c673cae | 2434 | if (!session) { |
7c673cae FG |
2435 | return; |
2436 | } | |
2437 | ceph_tid_t tid = fwd->get_tid(); | |
2438 | ||
2439 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2440 | ldout(cct, 10) << __func__ << " no pending request on tid " << tid << dendl; |
7c673cae FG |
2441 | return; |
2442 | } | |
2443 | ||
2444 | MetaRequest *request = mds_requests[tid]; | |
11fdf7f2 | 2445 | ceph_assert(request); |
7c673cae FG |
2446 | |
2447 | // reset retry counter | |
2448 | request->retry_attempt = 0; | |
2449 | ||
2450 | // request not forwarded, or dest mds has no session. | |
2451 | // resend. | |
11fdf7f2 | 2452 | ldout(cct, 10) << __func__ << " tid " << tid |
7c673cae FG |
2453 | << " fwd " << fwd->get_num_fwd() |
2454 | << " to mds." << fwd->get_dest_mds() | |
2455 | << ", resending to " << fwd->get_dest_mds() | |
2456 | << dendl; | |
2457 | ||
2458 | request->mds = -1; | |
2459 | request->item.remove_myself(); | |
2460 | request->num_fwd = fwd->get_num_fwd(); | |
2461 | request->resend_mds = fwd->get_dest_mds(); | |
9f95a23c | 2462 | request->caller_cond->notify_all(); |
7c673cae FG |
2463 | } |
2464 | ||
2465 | bool Client::is_dir_operation(MetaRequest *req) | |
2466 | { | |
2467 | int op = req->get_op(); | |
2468 | if (op == CEPH_MDS_OP_MKNOD || op == CEPH_MDS_OP_LINK || | |
2469 | op == CEPH_MDS_OP_UNLINK || op == CEPH_MDS_OP_RENAME || | |
2470 | op == CEPH_MDS_OP_MKDIR || op == CEPH_MDS_OP_RMDIR || | |
2471 | op == CEPH_MDS_OP_SYMLINK || op == CEPH_MDS_OP_CREATE) | |
2472 | return true; | |
2473 | return false; | |
2474 | } | |
2475 | ||
11fdf7f2 | 2476 | void Client::handle_client_reply(const MConstRef<MClientReply>& reply) |
7c673cae FG |
2477 | { |
2478 | mds_rank_t mds_num = mds_rank_t(reply->get_source().num()); | |
f67539c2 TL |
2479 | |
2480 | std::scoped_lock cl(client_lock); | |
20effc67 | 2481 | auto session = _get_mds_session(mds_num, reply->get_connection().get()); |
7c673cae | 2482 | if (!session) { |
7c673cae FG |
2483 | return; |
2484 | } | |
2485 | ||
2486 | ceph_tid_t tid = reply->get_tid(); | |
2487 | bool is_safe = reply->is_safe(); | |
2488 | ||
2489 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2490 | lderr(cct) << __func__ << " no pending request on tid " << tid |
7c673cae | 2491 | << " safe is:" << is_safe << dendl; |
7c673cae FG |
2492 | return; |
2493 | } | |
2494 | MetaRequest *request = mds_requests.at(tid); | |
2495 | ||
11fdf7f2 | 2496 | ldout(cct, 20) << __func__ << " got a reply. Safe:" << is_safe |
7c673cae FG |
2497 | << " tid " << tid << dendl; |
2498 | ||
2499 | if (request->got_unsafe && !is_safe) { | |
2500 | //duplicate response | |
2501 | ldout(cct, 0) << "got a duplicate reply on tid " << tid << " from mds " | |
2502 | << mds_num << " safe:" << is_safe << dendl; | |
7c673cae FG |
2503 | return; |
2504 | } | |
2505 | ||
f67539c2 | 2506 | if (-CEPHFS_ESTALE == reply->get_result()) { // see if we can get to proper MDS |
7c673cae FG |
2507 | ldout(cct, 20) << "got ESTALE on tid " << request->tid |
2508 | << " from mds." << request->mds << dendl; | |
2509 | request->send_to_auth = true; | |
2510 | request->resend_mds = choose_target_mds(request); | |
2511 | Inode *in = request->inode(); | |
11fdf7f2 | 2512 | std::map<mds_rank_t, Cap>::const_iterator it; |
7c673cae FG |
2513 | if (request->resend_mds >= 0 && |
2514 | request->resend_mds == request->mds && | |
2515 | (in == NULL || | |
11fdf7f2 TL |
2516 | (it = in->caps.find(request->resend_mds)) != in->caps.end() || |
2517 | request->sent_on_mseq == it->second.mseq)) { | |
2518 | ldout(cct, 20) << "have to return ESTALE" << dendl; | |
7c673cae | 2519 | } else { |
9f95a23c | 2520 | request->caller_cond->notify_all(); |
7c673cae FG |
2521 | return; |
2522 | } | |
7c673cae FG |
2523 | } |
2524 | ||
11fdf7f2 | 2525 | ceph_assert(!request->reply); |
7c673cae | 2526 | request->reply = reply; |
20effc67 | 2527 | insert_trace(request, session.get()); |
7c673cae FG |
2528 | |
2529 | // Handle unsafe reply | |
2530 | if (!is_safe) { | |
2531 | request->got_unsafe = true; | |
2532 | session->unsafe_requests.push_back(&request->unsafe_item); | |
2533 | if (is_dir_operation(request)) { | |
2534 | Inode *dir = request->inode(); | |
11fdf7f2 | 2535 | ceph_assert(dir); |
7c673cae FG |
2536 | dir->unsafe_ops.push_back(&request->unsafe_dir_item); |
2537 | } | |
2538 | if (request->target) { | |
2539 | InodeRef &in = request->target; | |
2540 | in->unsafe_ops.push_back(&request->unsafe_target_item); | |
2541 | } | |
2542 | } | |
2543 | ||
2544 | // Only signal the caller once (on the first reply): | |
2545 | // Either its an unsafe reply, or its a safe reply and no unsafe reply was sent. | |
2546 | if (!is_safe || !request->got_unsafe) { | |
9f95a23c | 2547 | ceph::condition_variable cond; |
7c673cae FG |
2548 | request->dispatch_cond = &cond; |
2549 | ||
2550 | // wake up waiter | |
11fdf7f2 | 2551 | ldout(cct, 20) << __func__ << " signalling caller " << (void*)request->caller_cond << dendl; |
9f95a23c | 2552 | request->caller_cond->notify_all(); |
7c673cae FG |
2553 | |
2554 | // wake for kick back | |
9f95a23c TL |
2555 | std::unique_lock l{client_lock, std::adopt_lock}; |
2556 | cond.wait(l, [tid, request, &cond, this] { | |
2557 | if (request->dispatch_cond) { | |
2558 | ldout(cct, 20) << "handle_client_reply awaiting kickback on tid " | |
2559 | << tid << " " << &cond << dendl; | |
2560 | } | |
2561 | return !request->dispatch_cond; | |
2562 | }); | |
2563 | l.release(); | |
7c673cae FG |
2564 | } |
2565 | ||
2566 | if (is_safe) { | |
2567 | // the filesystem change is committed to disk | |
2568 | // we're done, clean up | |
2569 | if (request->got_unsafe) { | |
2570 | request->unsafe_item.remove_myself(); | |
2571 | request->unsafe_dir_item.remove_myself(); | |
2572 | request->unsafe_target_item.remove_myself(); | |
2573 | signal_cond_list(request->waitfor_safe); | |
2574 | } | |
2575 | request->item.remove_myself(); | |
2576 | unregister_request(request); | |
2577 | } | |
f67539c2 | 2578 | if (is_unmounting()) |
9f95a23c | 2579 | mount_cond.notify_all(); |
7c673cae FG |
2580 | } |
2581 | ||
2582 | void Client::_handle_full_flag(int64_t pool) | |
2583 | { | |
2584 | ldout(cct, 1) << __func__ << ": FULL: cancelling outstanding operations " | |
2585 | << "on " << pool << dendl; | |
f67539c2 | 2586 | // Cancel all outstanding ops in this pool with -CEPHFS_ENOSPC: it is necessary |
7c673cae FG |
2587 | // to do this rather than blocking, because otherwise when we fill up we |
2588 | // potentially lock caps forever on files with dirty pages, and we need | |
2589 | // to be able to release those caps to the MDS so that it can delete files | |
2590 | // and free up space. | |
f67539c2 | 2591 | epoch_t cancelled_epoch = objecter->op_cancel_writes(-CEPHFS_ENOSPC, pool); |
7c673cae FG |
2592 | |
2593 | // For all inodes with layouts in this pool and a pending flush write op | |
2594 | // (i.e. one of the ones we will cancel), we've got to purge_set their data | |
2595 | // from ObjectCacher so that it doesn't re-issue the write in response to | |
2596 | // the ENOSPC error. | |
2597 | // Fortunately since we're cancelling everything in a given pool, we don't | |
2598 | // need to know which ops belong to which ObjectSet, we can just blow all | |
2599 | // the un-flushed cached data away and mark any dirty inodes' async_err | |
f67539c2 | 2600 | // field with -CEPHFS_ENOSPC as long as we're sure all the ops we cancelled were |
7c673cae FG |
2601 | // affecting this pool, and all the objectsets we're purging were also |
2602 | // in this pool. | |
2603 | for (unordered_map<vinodeno_t,Inode*>::iterator i = inode_map.begin(); | |
2604 | i != inode_map.end(); ++i) | |
2605 | { | |
2606 | Inode *inode = i->second; | |
2607 | if (inode->oset.dirty_or_tx | |
2608 | && (pool == -1 || inode->layout.pool_id == pool)) { | |
2609 | ldout(cct, 4) << __func__ << ": FULL: inode 0x" << std::hex << i->first << std::dec | |
2610 | << " has dirty objects, purging and setting ENOSPC" << dendl; | |
2611 | objectcacher->purge_set(&inode->oset); | |
f67539c2 | 2612 | inode->set_async_err(-CEPHFS_ENOSPC); |
7c673cae FG |
2613 | } |
2614 | } | |
2615 | ||
2616 | if (cancelled_epoch != (epoch_t)-1) { | |
2617 | set_cap_epoch_barrier(cancelled_epoch); | |
2618 | } | |
2619 | } | |
2620 | ||
11fdf7f2 | 2621 | void Client::handle_osd_map(const MConstRef<MOSDMap>& m) |
7c673cae | 2622 | { |
f67539c2 TL |
2623 | std::set<entity_addr_t> new_blocklists; |
2624 | ||
2625 | std::scoped_lock cl(client_lock); | |
2626 | objecter->consume_blocklist_events(&new_blocklists); | |
31f18b77 | 2627 | |
11fdf7f2 | 2628 | const auto myaddrs = messenger->get_myaddrs(); |
f67539c2 | 2629 | bool new_blocklist = false; |
11fdf7f2 TL |
2630 | bool prenautilus = objecter->with_osdmap( |
2631 | [&](const OSDMap& o) { | |
9f95a23c | 2632 | return o.require_osd_release < ceph_release_t::nautilus; |
11fdf7f2 | 2633 | }); |
f67539c2 | 2634 | if (!blocklisted) { |
11fdf7f2 | 2635 | for (auto a : myaddrs.v) { |
f67539c2 | 2636 | // blocklist entries are always TYPE_ANY for nautilus+ |
11fdf7f2 | 2637 | a.set_type(entity_addr_t::TYPE_ANY); |
f67539c2 TL |
2638 | if (new_blocklists.count(a)) { |
2639 | new_blocklist = true; | |
11fdf7f2 TL |
2640 | break; |
2641 | } | |
2642 | if (prenautilus) { | |
2643 | // ...except pre-nautilus, they were TYPE_LEGACY | |
2644 | a.set_type(entity_addr_t::TYPE_LEGACY); | |
f67539c2 TL |
2645 | if (new_blocklists.count(a)) { |
2646 | new_blocklist = true; | |
11fdf7f2 TL |
2647 | break; |
2648 | } | |
2649 | } | |
2650 | } | |
2651 | } | |
f67539c2 | 2652 | if (new_blocklist) { |
31f18b77 FG |
2653 | auto epoch = objecter->with_osdmap([](const OSDMap &o){ |
2654 | return o.get_epoch(); | |
2655 | }); | |
f67539c2 TL |
2656 | lderr(cct) << "I was blocklisted at osd epoch " << epoch << dendl; |
2657 | blocklisted = true; | |
31f18b77 | 2658 | |
f67539c2 | 2659 | _abort_mds_sessions(-CEPHFS_EBLOCKLISTED); |
31f18b77 FG |
2660 | |
2661 | // Since we know all our OSD ops will fail, cancel them all preemtively, | |
2662 | // so that on an unhealthy cluster we can umount promptly even if e.g. | |
2663 | // some PGs were inaccessible. | |
f67539c2 TL |
2664 | objecter->op_cancel_writes(-CEPHFS_EBLOCKLISTED); |
2665 | ||
2666 | } | |
31f18b77 | 2667 | |
f67539c2 TL |
2668 | if (blocklisted) { |
2669 | // Handle case where we were blocklisted but no longer are | |
2670 | blocklisted = objecter->with_osdmap([myaddrs](const OSDMap &o){ | |
2671 | return o.is_blocklisted(myaddrs);}); | |
31f18b77 FG |
2672 | } |
2673 | ||
f67539c2 TL |
2674 | // Always subscribe to next osdmap for blocklisted client |
2675 | // until this client is not blocklisted. | |
2676 | if (blocklisted) { | |
f64942e4 AA |
2677 | objecter->maybe_request_map(); |
2678 | } | |
2679 | ||
7c673cae FG |
2680 | if (objecter->osdmap_full_flag()) { |
2681 | _handle_full_flag(-1); | |
2682 | } else { | |
2683 | // Accumulate local list of full pools so that I can drop | |
2684 | // the objecter lock before re-entering objecter in | |
2685 | // cancel_writes | |
2686 | std::vector<int64_t> full_pools; | |
2687 | ||
2688 | objecter->with_osdmap([&full_pools](const OSDMap &o) { | |
2689 | for (const auto& kv : o.get_pools()) { | |
2690 | if (kv.second.has_flag(pg_pool_t::FLAG_FULL)) { | |
2691 | full_pools.push_back(kv.first); | |
2692 | } | |
2693 | } | |
2694 | }); | |
2695 | ||
2696 | for (auto p : full_pools) | |
2697 | _handle_full_flag(p); | |
2698 | ||
2699 | // Subscribe to subsequent maps to watch for the full flag going | |
2700 | // away. For the global full flag objecter does this for us, but | |
2701 | // it pays no attention to the per-pool full flag so in this branch | |
2702 | // we do it ourselves. | |
2703 | if (!full_pools.empty()) { | |
2704 | objecter->maybe_request_map(); | |
2705 | } | |
2706 | } | |
7c673cae FG |
2707 | } |
2708 | ||
2709 | ||
2710 | // ------------------------ | |
2711 | // incoming messages | |
2712 | ||
2713 | ||
11fdf7f2 | 2714 | bool Client::ms_dispatch2(const MessageRef &m) |
7c673cae | 2715 | { |
f67539c2 TL |
2716 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2717 | if (!iref_reader.is_state_satisfied()) { | |
7c673cae | 2718 | ldout(cct, 10) << "inactive, discarding " << *m << dendl; |
7c673cae FG |
2719 | return true; |
2720 | } | |
2721 | ||
2722 | switch (m->get_type()) { | |
2723 | // mounting and mds sessions | |
2724 | case CEPH_MSG_MDS_MAP: | |
9f95a23c | 2725 | handle_mds_map(ref_cast<MMDSMap>(m)); |
7c673cae FG |
2726 | break; |
2727 | case CEPH_MSG_FS_MAP: | |
9f95a23c | 2728 | handle_fs_map(ref_cast<MFSMap>(m)); |
7c673cae FG |
2729 | break; |
2730 | case CEPH_MSG_FS_MAP_USER: | |
9f95a23c | 2731 | handle_fs_map_user(ref_cast<MFSMapUser>(m)); |
7c673cae FG |
2732 | break; |
2733 | case CEPH_MSG_CLIENT_SESSION: | |
9f95a23c | 2734 | handle_client_session(ref_cast<MClientSession>(m)); |
7c673cae FG |
2735 | break; |
2736 | ||
2737 | case CEPH_MSG_OSD_MAP: | |
9f95a23c | 2738 | handle_osd_map(ref_cast<MOSDMap>(m)); |
7c673cae FG |
2739 | break; |
2740 | ||
2741 | // requests | |
2742 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: | |
9f95a23c | 2743 | handle_client_request_forward(ref_cast<MClientRequestForward>(m)); |
7c673cae FG |
2744 | break; |
2745 | case CEPH_MSG_CLIENT_REPLY: | |
9f95a23c | 2746 | handle_client_reply(ref_cast<MClientReply>(m)); |
11fdf7f2 TL |
2747 | break; |
2748 | ||
2749 | // reclaim reply | |
2750 | case CEPH_MSG_CLIENT_RECLAIM_REPLY: | |
9f95a23c | 2751 | handle_client_reclaim_reply(ref_cast<MClientReclaimReply>(m)); |
7c673cae FG |
2752 | break; |
2753 | ||
2754 | case CEPH_MSG_CLIENT_SNAP: | |
9f95a23c | 2755 | handle_snap(ref_cast<MClientSnap>(m)); |
7c673cae FG |
2756 | break; |
2757 | case CEPH_MSG_CLIENT_CAPS: | |
9f95a23c | 2758 | handle_caps(ref_cast<MClientCaps>(m)); |
7c673cae FG |
2759 | break; |
2760 | case CEPH_MSG_CLIENT_LEASE: | |
9f95a23c | 2761 | handle_lease(ref_cast<MClientLease>(m)); |
7c673cae FG |
2762 | break; |
2763 | case MSG_COMMAND_REPLY: | |
2764 | if (m->get_source().type() == CEPH_ENTITY_TYPE_MDS) { | |
9f95a23c | 2765 | handle_command_reply(ref_cast<MCommandReply>(m)); |
7c673cae FG |
2766 | } else { |
2767 | return false; | |
2768 | } | |
2769 | break; | |
2770 | case CEPH_MSG_CLIENT_QUOTA: | |
9f95a23c | 2771 | handle_quota(ref_cast<MClientQuota>(m)); |
7c673cae FG |
2772 | break; |
2773 | ||
2774 | default: | |
2775 | return false; | |
2776 | } | |
2777 | ||
2778 | // unmounting? | |
f67539c2 TL |
2779 | std::scoped_lock cl(client_lock); |
2780 | if (is_unmounting()) { | |
7c673cae FG |
2781 | ldout(cct, 10) << "unmounting: trim pass, size was " << lru.lru_get_size() |
2782 | << "+" << inode_map.size() << dendl; | |
f67539c2 | 2783 | uint64_t size = lru.lru_get_size() + inode_map.size(); |
7c673cae | 2784 | trim_cache(); |
f67539c2 | 2785 | if (size > lru.lru_get_size() + inode_map.size()) { |
7c673cae | 2786 | ldout(cct, 10) << "unmounting: trim pass, cache shrank, poking unmount()" << dendl; |
9f95a23c | 2787 | mount_cond.notify_all(); |
7c673cae FG |
2788 | } else { |
2789 | ldout(cct, 10) << "unmounting: trim pass, size still " << lru.lru_get_size() | |
2790 | << "+" << inode_map.size() << dendl; | |
2791 | } | |
2792 | } | |
2793 | ||
2794 | return true; | |
2795 | } | |
2796 | ||
11fdf7f2 | 2797 | void Client::handle_fs_map(const MConstRef<MFSMap>& m) |
7c673cae | 2798 | { |
f67539c2 | 2799 | std::scoped_lock cl(client_lock); |
7c673cae | 2800 | fsmap.reset(new FSMap(m->get_fsmap())); |
7c673cae FG |
2801 | |
2802 | signal_cond_list(waiting_for_fsmap); | |
2803 | ||
2804 | monclient->sub_got("fsmap", fsmap->get_epoch()); | |
2805 | } | |
2806 | ||
11fdf7f2 | 2807 | void Client::handle_fs_map_user(const MConstRef<MFSMapUser>& m) |
7c673cae | 2808 | { |
f67539c2 | 2809 | std::scoped_lock cl(client_lock); |
7c673cae FG |
2810 | fsmap_user.reset(new FSMapUser); |
2811 | *fsmap_user = m->get_fsmap(); | |
7c673cae FG |
2812 | |
2813 | monclient->sub_got("fsmap.user", fsmap_user->get_epoch()); | |
2814 | signal_cond_list(waiting_for_fsmap); | |
2815 | } | |
2816 | ||
f67539c2 TL |
2817 | // Cancel all the commands for missing or laggy GIDs |
2818 | void Client::cancel_commands(const MDSMap& newmap) | |
7c673cae | 2819 | { |
f67539c2 | 2820 | std::vector<ceph_tid_t> cancel_ops; |
7c673cae | 2821 | |
f67539c2 | 2822 | std::scoped_lock cmd_lock(command_lock); |
7c673cae | 2823 | auto &commands = command_table.get_commands(); |
f67539c2 | 2824 | for (const auto &[tid, op] : commands) { |
7c673cae | 2825 | const mds_gid_t op_mds_gid = op.mds_gid; |
f67539c2 TL |
2826 | if (newmap.is_dne_gid(op_mds_gid) || newmap.is_laggy_gid(op_mds_gid)) { |
2827 | ldout(cct, 1) << __func__ << ": cancelling command op " << tid << dendl; | |
2828 | cancel_ops.push_back(tid); | |
7c673cae FG |
2829 | if (op.outs) { |
2830 | std::ostringstream ss; | |
2831 | ss << "MDS " << op_mds_gid << " went away"; | |
2832 | *(op.outs) = ss.str(); | |
2833 | } | |
f67539c2 TL |
2834 | /* |
2835 | * No need to make the con->mark_down under | |
2836 | * client_lock here, because the con will | |
2837 | * has its own lock. | |
2838 | */ | |
7c673cae | 2839 | op.con->mark_down(); |
f67539c2 TL |
2840 | if (op.on_finish) |
2841 | op.on_finish->complete(-CEPHFS_ETIMEDOUT); | |
7c673cae FG |
2842 | } |
2843 | } | |
2844 | ||
f67539c2 TL |
2845 | for (const auto &tid : cancel_ops) |
2846 | command_table.erase(tid); | |
2847 | } | |
2848 | ||
2849 | void Client::handle_mds_map(const MConstRef<MMDSMap>& m) | |
2850 | { | |
2851 | std::unique_lock cl(client_lock); | |
2852 | if (m->get_epoch() <= mdsmap->get_epoch()) { | |
2853 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() | |
2854 | << " is identical to or older than our " | |
2855 | << mdsmap->get_epoch() << dendl; | |
2856 | return; | |
7c673cae FG |
2857 | } |
2858 | ||
f67539c2 TL |
2859 | cl.unlock(); |
2860 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() << dendl; | |
2861 | std::unique_ptr<MDSMap> _mdsmap(new MDSMap); | |
2862 | _mdsmap->decode(m->get_encoded()); | |
2863 | cancel_commands(*_mdsmap.get()); | |
2864 | cl.lock(); | |
2865 | ||
2866 | _mdsmap.swap(mdsmap); | |
2867 | ||
7c673cae | 2868 | // reset session |
11fdf7f2 | 2869 | for (auto p = mds_sessions.begin(); p != mds_sessions.end(); ) { |
7c673cae | 2870 | mds_rank_t mds = p->first; |
20effc67 | 2871 | MetaSessionRef session = p->second; |
7c673cae FG |
2872 | ++p; |
2873 | ||
f67539c2 | 2874 | int oldstate = _mdsmap->get_state(mds); |
7c673cae FG |
2875 | int newstate = mdsmap->get_state(mds); |
2876 | if (!mdsmap->is_up(mds)) { | |
2877 | session->con->mark_down(); | |
11fdf7f2 | 2878 | } else if (mdsmap->get_addrs(mds) != session->addrs) { |
f67539c2 TL |
2879 | auto old_inc = _mdsmap->get_incarnation(mds); |
2880 | auto new_inc = mdsmap->get_incarnation(mds); | |
f64942e4 AA |
2881 | if (old_inc != new_inc) { |
2882 | ldout(cct, 1) << "mds incarnation changed from " | |
2883 | << old_inc << " to " << new_inc << dendl; | |
2884 | oldstate = MDSMap::STATE_NULL; | |
2885 | } | |
7c673cae | 2886 | session->con->mark_down(); |
11fdf7f2 | 2887 | session->addrs = mdsmap->get_addrs(mds); |
7c673cae FG |
2888 | // When new MDS starts to take over, notify kernel to trim unused entries |
2889 | // in its dcache/icache. Hopefully, the kernel will release some unused | |
2890 | // inodes before the new MDS enters reconnect state. | |
20effc67 | 2891 | trim_cache_for_reconnect(session.get()); |
7c673cae FG |
2892 | } else if (oldstate == newstate) |
2893 | continue; // no change | |
f67539c2 | 2894 | |
7c673cae FG |
2895 | session->mds_state = newstate; |
2896 | if (newstate == MDSMap::STATE_RECONNECT) { | |
11fdf7f2 | 2897 | session->con = messenger->connect_to_mds(session->addrs); |
20effc67 | 2898 | send_reconnect(session.get()); |
81eedcae TL |
2899 | } else if (newstate > MDSMap::STATE_RECONNECT) { |
2900 | if (oldstate < MDSMap::STATE_RECONNECT) { | |
2901 | ldout(cct, 1) << "we may miss the MDSMap::RECONNECT, close mds session ... " << dendl; | |
20effc67 | 2902 | _closed_mds_session(session.get()); |
81eedcae TL |
2903 | continue; |
2904 | } | |
2905 | if (newstate >= MDSMap::STATE_ACTIVE) { | |
2906 | if (oldstate < MDSMap::STATE_ACTIVE) { | |
2907 | // kick new requests | |
20effc67 TL |
2908 | kick_requests(session.get()); |
2909 | kick_flushing_caps(session.get()); | |
81eedcae | 2910 | signal_context_list(session->waiting_for_open); |
20effc67 | 2911 | wake_up_session_caps(session.get(), true); |
81eedcae TL |
2912 | } |
2913 | connect_mds_targets(mds); | |
7c673cae | 2914 | } |
7c673cae FG |
2915 | } else if (newstate == MDSMap::STATE_NULL && |
2916 | mds >= mdsmap->get_max_mds()) { | |
20effc67 | 2917 | _closed_mds_session(session.get()); |
7c673cae FG |
2918 | } |
2919 | } | |
2920 | ||
2921 | // kick any waiting threads | |
2922 | signal_cond_list(waiting_for_mdsmap); | |
2923 | ||
7c673cae FG |
2924 | monclient->sub_got("mdsmap", mdsmap->get_epoch()); |
2925 | } | |
2926 | ||
2927 | void Client::send_reconnect(MetaSession *session) | |
2928 | { | |
2929 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2930 | ldout(cct, 10) << __func__ << " to mds." << mds << dendl; |
7c673cae FG |
2931 | |
2932 | // trim unused caps to reduce MDS's cache rejoin time | |
2933 | trim_cache_for_reconnect(session); | |
2934 | ||
2935 | session->readonly = false; | |
2936 | ||
11fdf7f2 | 2937 | session->release.reset(); |
7c673cae FG |
2938 | |
2939 | // reset my cap seq number | |
2940 | session->seq = 0; | |
2941 | //connect to the mds' offload targets | |
2942 | connect_mds_targets(mds); | |
2943 | //make sure unsafe requests get saved | |
2944 | resend_unsafe_requests(session); | |
2945 | ||
11fdf7f2 TL |
2946 | early_kick_flushing_caps(session); |
2947 | ||
9f95a23c | 2948 | auto m = make_message<MClientReconnect>(); |
11fdf7f2 | 2949 | bool allow_multi = session->mds_features.test(CEPHFS_FEATURE_MULTI_RECONNECT); |
7c673cae FG |
2950 | |
2951 | // i have an open session. | |
2952 | ceph::unordered_set<inodeno_t> did_snaprealm; | |
2953 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator p = inode_map.begin(); | |
2954 | p != inode_map.end(); | |
2955 | ++p) { | |
2956 | Inode *in = p->second; | |
11fdf7f2 TL |
2957 | auto it = in->caps.find(mds); |
2958 | if (it != in->caps.end()) { | |
2959 | if (allow_multi && | |
9f95a23c TL |
2960 | m->get_approx_size() >= |
2961 | static_cast<size_t>((std::numeric_limits<int>::max() >> 1))) { | |
11fdf7f2 TL |
2962 | m->mark_more(); |
2963 | session->con->send_message2(std::move(m)); | |
2964 | ||
9f95a23c | 2965 | m = make_message<MClientReconnect>(); |
11fdf7f2 TL |
2966 | } |
2967 | ||
2968 | Cap &cap = it->second; | |
7c673cae | 2969 | ldout(cct, 10) << " caps on " << p->first |
11fdf7f2 | 2970 | << " " << ccap_string(cap.issued) |
7c673cae FG |
2971 | << " wants " << ccap_string(in->caps_wanted()) |
2972 | << dendl; | |
2973 | filepath path; | |
f91f0fd5 | 2974 | in->make_short_path(path); |
7c673cae FG |
2975 | ldout(cct, 10) << " path " << path << dendl; |
2976 | ||
2977 | bufferlist flockbl; | |
2978 | _encode_filelocks(in, flockbl); | |
2979 | ||
11fdf7f2 TL |
2980 | cap.seq = 0; // reset seq. |
2981 | cap.issue_seq = 0; // reset seq. | |
2982 | cap.mseq = 0; // reset seq. | |
2983 | // cap gen should catch up with session cap_gen | |
2984 | if (cap.gen < session->cap_gen) { | |
2985 | cap.gen = session->cap_gen; | |
2986 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
2987 | } else { | |
2988 | cap.issued = cap.implemented; | |
2989 | } | |
7c673cae FG |
2990 | snapid_t snap_follows = 0; |
2991 | if (!in->cap_snaps.empty()) | |
2992 | snap_follows = in->cap_snaps.begin()->first; | |
2993 | ||
2994 | m->add_cap(p->first.ino, | |
11fdf7f2 | 2995 | cap.cap_id, |
7c673cae FG |
2996 | path.get_ino(), path.get_path(), // ino |
2997 | in->caps_wanted(), // wanted | |
11fdf7f2 | 2998 | cap.issued, // issued |
7c673cae FG |
2999 | in->snaprealm->ino, |
3000 | snap_follows, | |
3001 | flockbl); | |
3002 | ||
3003 | if (did_snaprealm.count(in->snaprealm->ino) == 0) { | |
3004 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
3005 | m->add_snaprealm(in->snaprealm->ino, in->snaprealm->seq, in->snaprealm->parent); | |
3006 | did_snaprealm.insert(in->snaprealm->ino); | |
3007 | } | |
3008 | } | |
3009 | } | |
3010 | ||
11fdf7f2 TL |
3011 | if (!allow_multi) |
3012 | m->set_encoding_version(0); // use connection features to choose encoding | |
3013 | session->con->send_message2(std::move(m)); | |
7c673cae | 3014 | |
9f95a23c | 3015 | mount_cond.notify_all(); |
11fdf7f2 TL |
3016 | |
3017 | if (session->reclaim_state == MetaSession::RECLAIMING) | |
3018 | signal_cond_list(waiting_for_reclaim); | |
7c673cae FG |
3019 | } |
3020 | ||
3021 | ||
3022 | void Client::kick_requests(MetaSession *session) | |
3023 | { | |
11fdf7f2 | 3024 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3025 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3026 | p != mds_requests.end(); | |
3027 | ++p) { | |
31f18b77 FG |
3028 | MetaRequest *req = p->second; |
3029 | if (req->got_unsafe) | |
3030 | continue; | |
3031 | if (req->aborted()) { | |
3032 | if (req->caller_cond) { | |
3033 | req->kick = true; | |
9f95a23c | 3034 | req->caller_cond->notify_all(); |
31f18b77 | 3035 | } |
7c673cae | 3036 | continue; |
31f18b77 FG |
3037 | } |
3038 | if (req->retry_attempt > 0) | |
7c673cae | 3039 | continue; // new requests only |
31f18b77 | 3040 | if (req->mds == session->mds_num) { |
7c673cae FG |
3041 | send_request(p->second, session); |
3042 | } | |
3043 | } | |
3044 | } | |
3045 | ||
3046 | void Client::resend_unsafe_requests(MetaSession *session) | |
3047 | { | |
3048 | for (xlist<MetaRequest*>::iterator iter = session->unsafe_requests.begin(); | |
3049 | !iter.end(); | |
3050 | ++iter) | |
3051 | send_request(*iter, session); | |
3052 | ||
3053 | // also re-send old requests when MDS enters reconnect stage. So that MDS can | |
3054 | // process completed requests in clientreplay stage. | |
3055 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
3056 | p != mds_requests.end(); | |
3057 | ++p) { | |
3058 | MetaRequest *req = p->second; | |
3059 | if (req->got_unsafe) | |
3060 | continue; | |
31f18b77 FG |
3061 | if (req->aborted()) |
3062 | continue; | |
7c673cae FG |
3063 | if (req->retry_attempt == 0) |
3064 | continue; // old requests only | |
3065 | if (req->mds == session->mds_num) | |
3066 | send_request(req, session, true); | |
3067 | } | |
3068 | } | |
3069 | ||
3070 | void Client::wait_unsafe_requests() | |
3071 | { | |
3072 | list<MetaRequest*> last_unsafe_reqs; | |
11fdf7f2 | 3073 | for (const auto &p : mds_sessions) { |
20effc67 TL |
3074 | const auto s = p.second; |
3075 | if (!s->unsafe_requests.empty()) { | |
3076 | MetaRequest *req = s->unsafe_requests.back(); | |
7c673cae FG |
3077 | req->get(); |
3078 | last_unsafe_reqs.push_back(req); | |
3079 | } | |
3080 | } | |
3081 | ||
3082 | for (list<MetaRequest*>::iterator p = last_unsafe_reqs.begin(); | |
3083 | p != last_unsafe_reqs.end(); | |
3084 | ++p) { | |
3085 | MetaRequest *req = *p; | |
3086 | if (req->unsafe_item.is_on_list()) | |
3087 | wait_on_list(req->waitfor_safe); | |
3088 | put_request(req); | |
3089 | } | |
3090 | } | |
3091 | ||
3092 | void Client::kick_requests_closed(MetaSession *session) | |
3093 | { | |
11fdf7f2 | 3094 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3095 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3096 | p != mds_requests.end(); ) { | |
3097 | MetaRequest *req = p->second; | |
3098 | ++p; | |
3099 | if (req->mds == session->mds_num) { | |
3100 | if (req->caller_cond) { | |
3101 | req->kick = true; | |
9f95a23c | 3102 | req->caller_cond->notify_all(); |
7c673cae FG |
3103 | } |
3104 | req->item.remove_myself(); | |
3105 | if (req->got_unsafe) { | |
11fdf7f2 | 3106 | lderr(cct) << __func__ << " removing unsafe request " << req->get_tid() << dendl; |
7c673cae | 3107 | req->unsafe_item.remove_myself(); |
eafe8130 TL |
3108 | if (is_dir_operation(req)) { |
3109 | Inode *dir = req->inode(); | |
20effc67 | 3110 | ceph_assert(dir); |
f67539c2 | 3111 | dir->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3112 | lderr(cct) << "kick_requests_closed drop req of inode(dir) : " |
3113 | << dir->ino << " " << req->get_tid() << dendl; | |
3114 | req->unsafe_dir_item.remove_myself(); | |
3115 | } | |
3116 | if (req->target) { | |
3117 | InodeRef &in = req->target; | |
f67539c2 | 3118 | in->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3119 | lderr(cct) << "kick_requests_closed drop req of inode : " |
3120 | << in->ino << " " << req->get_tid() << dendl; | |
3121 | req->unsafe_target_item.remove_myself(); | |
3122 | } | |
7c673cae FG |
3123 | signal_cond_list(req->waitfor_safe); |
3124 | unregister_request(req); | |
3125 | } | |
3126 | } | |
3127 | } | |
11fdf7f2 TL |
3128 | ceph_assert(session->requests.empty()); |
3129 | ceph_assert(session->unsafe_requests.empty()); | |
7c673cae FG |
3130 | } |
3131 | ||
3132 | ||
3133 | ||
3134 | ||
3135 | /************ | |
3136 | * leases | |
3137 | */ | |
3138 | ||
3139 | void Client::got_mds_push(MetaSession *s) | |
3140 | { | |
3141 | s->seq++; | |
3142 | ldout(cct, 10) << " mds." << s->mds_num << " seq now " << s->seq << dendl; | |
3143 | if (s->state == MetaSession::STATE_CLOSING) { | |
9f95a23c | 3144 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
3145 | } |
3146 | } | |
3147 | ||
11fdf7f2 | 3148 | void Client::handle_lease(const MConstRef<MClientLease>& m) |
7c673cae | 3149 | { |
11fdf7f2 | 3150 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 3151 | |
11fdf7f2 | 3152 | ceph_assert(m->get_action() == CEPH_MDS_LEASE_REVOKE); |
7c673cae | 3153 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
3154 | |
3155 | std::scoped_lock cl(client_lock); | |
20effc67 | 3156 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 3157 | if (!session) { |
7c673cae FG |
3158 | return; |
3159 | } | |
3160 | ||
20effc67 | 3161 | got_mds_push(session.get()); |
7c673cae FG |
3162 | |
3163 | ceph_seq_t seq = m->get_seq(); | |
3164 | ||
3165 | Inode *in; | |
3166 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); | |
3167 | if (inode_map.count(vino) == 0) { | |
3168 | ldout(cct, 10) << " don't have vino " << vino << dendl; | |
3169 | goto revoke; | |
3170 | } | |
3171 | in = inode_map[vino]; | |
3172 | ||
9f95a23c | 3173 | if (m->get_mask() & CEPH_LEASE_VALID) { |
7c673cae FG |
3174 | if (!in->dir || in->dir->dentries.count(m->dname) == 0) { |
3175 | ldout(cct, 10) << " don't have dir|dentry " << m->get_ino() << "/" << m->dname <<dendl; | |
3176 | goto revoke; | |
3177 | } | |
3178 | Dentry *dn = in->dir->dentries[m->dname]; | |
3179 | ldout(cct, 10) << " revoked DN lease on " << dn << dendl; | |
3180 | dn->lease_mds = -1; | |
3181 | } | |
3182 | ||
3183 | revoke: | |
11fdf7f2 | 3184 | { |
9f95a23c TL |
3185 | auto reply = make_message<MClientLease>(CEPH_MDS_LEASE_RELEASE, seq, |
3186 | m->get_mask(), m->get_ino(), | |
3187 | m->get_first(), m->get_last(), m->dname); | |
11fdf7f2 TL |
3188 | m->get_connection()->send_message2(std::move(reply)); |
3189 | } | |
7c673cae FG |
3190 | } |
3191 | ||
f67539c2 | 3192 | void Client::_put_inode(Inode *in, int n) |
7c673cae | 3193 | { |
f67539c2 TL |
3194 | ldout(cct, 10) << __func__ << " on " << *in << " n = " << n << dendl; |
3195 | ||
b3b6e05e TL |
3196 | int left = in->get_nref(); |
3197 | ceph_assert(left >= n + 1); | |
3198 | in->iput(n); | |
3199 | left -= n; | |
3200 | if (left == 1) { // the last one will be held by the inode_map | |
7c673cae FG |
3201 | // release any caps |
3202 | remove_all_caps(in); | |
3203 | ||
11fdf7f2 | 3204 | ldout(cct, 10) << __func__ << " deleting " << *in << dendl; |
7c673cae | 3205 | bool unclean = objectcacher->release_set(&in->oset); |
11fdf7f2 | 3206 | ceph_assert(!unclean); |
7c673cae FG |
3207 | inode_map.erase(in->vino()); |
3208 | if (use_faked_inos()) | |
3209 | _release_faked_ino(in); | |
3210 | ||
b3b6e05e | 3211 | if (root == nullptr) { |
7c673cae FG |
3212 | root_ancestor = 0; |
3213 | while (!root_parents.empty()) | |
3214 | root_parents.erase(root_parents.begin()); | |
3215 | } | |
3216 | ||
b3b6e05e | 3217 | in->iput(); |
7c673cae FG |
3218 | } |
3219 | } | |
3220 | ||
f67539c2 TL |
3221 | void Client::delay_put_inodes(bool wakeup) |
3222 | { | |
3223 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
3224 | ||
3225 | std::map<Inode*,int> release; | |
3226 | { | |
3227 | std::scoped_lock dl(delay_i_lock); | |
3228 | release.swap(delay_i_release); | |
3229 | } | |
3230 | ||
3231 | if (release.empty()) | |
3232 | return; | |
3233 | ||
3234 | for (auto &[in, cnt] : release) | |
3235 | _put_inode(in, cnt); | |
3236 | ||
3237 | if (wakeup) | |
3238 | mount_cond.notify_all(); | |
3239 | } | |
3240 | ||
3241 | void Client::put_inode(Inode *in, int n) | |
3242 | { | |
3243 | ldout(cct, 20) << __func__ << " on " << *in << " n = " << n << dendl; | |
3244 | ||
3245 | std::scoped_lock dl(delay_i_lock); | |
3246 | delay_i_release[in] += n; | |
3247 | } | |
3248 | ||
7c673cae FG |
3249 | void Client::close_dir(Dir *dir) |
3250 | { | |
3251 | Inode *in = dir->parent_inode; | |
11fdf7f2 TL |
3252 | ldout(cct, 15) << __func__ << " dir " << dir << " on " << in << dendl; |
3253 | ceph_assert(dir->is_empty()); | |
3254 | ceph_assert(in->dir == dir); | |
3255 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked | |
3256 | if (!in->dentries.empty()) | |
7c673cae FG |
3257 | in->get_first_parent()->put(); // unpin dentry |
3258 | ||
3259 | delete in->dir; | |
3260 | in->dir = 0; | |
3261 | put_inode(in); // unpin inode | |
3262 | } | |
3263 | ||
3264 | /** | |
3265 | * Don't call this with in==NULL, use get_or_create for that | |
3266 | * leave dn set to default NULL unless you're trying to add | |
3267 | * a new inode to a pre-created Dentry | |
3268 | */ | |
3269 | Dentry* Client::link(Dir *dir, const string& name, Inode *in, Dentry *dn) | |
3270 | { | |
3271 | if (!dn) { | |
3272 | // create a new Dentry | |
11fdf7f2 TL |
3273 | dn = new Dentry(dir, name); |
3274 | ||
7c673cae FG |
3275 | lru.lru_insert_mid(dn); // mid or top? |
3276 | ||
3277 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in | |
3278 | << " dn " << dn << " (new dn)" << dendl; | |
3279 | } else { | |
11fdf7f2 | 3280 | ceph_assert(!dn->inode); |
7c673cae FG |
3281 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in |
3282 | << " dn " << dn << " (old dn)" << dendl; | |
3283 | } | |
3284 | ||
3285 | if (in) { // link to inode | |
11fdf7f2 | 3286 | InodeRef tmp_ref; |
7c673cae | 3287 | // only one parent for directories! |
11fdf7f2 TL |
3288 | if (in->is_dir() && !in->dentries.empty()) { |
3289 | tmp_ref = in; // prevent unlink below from freeing the inode. | |
7c673cae | 3290 | Dentry *olddn = in->get_first_parent(); |
11fdf7f2 | 3291 | ceph_assert(olddn->dir != dir || olddn->name != name); |
7c673cae | 3292 | Inode *old_diri = olddn->dir->parent_inode; |
7c673cae FG |
3293 | clear_dir_complete_and_ordered(old_diri, true); |
3294 | unlink(olddn, true, true); // keep dir, dentry | |
3295 | } | |
3296 | ||
11fdf7f2 | 3297 | dn->link(in); |
f67539c2 | 3298 | inc_dentry_nr(); |
11fdf7f2 | 3299 | ldout(cct, 20) << "link inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3300 | } |
3301 | ||
3302 | return dn; | |
3303 | } | |
3304 | ||
3305 | void Client::unlink(Dentry *dn, bool keepdir, bool keepdentry) | |
3306 | { | |
11fdf7f2 | 3307 | InodeRef in(dn->inode); |
7c673cae FG |
3308 | ldout(cct, 15) << "unlink dir " << dn->dir->parent_inode << " '" << dn->name << "' dn " << dn |
3309 | << " inode " << dn->inode << dendl; | |
3310 | ||
3311 | // unlink from inode | |
11fdf7f2 TL |
3312 | if (dn->inode) { |
3313 | dn->unlink(); | |
f67539c2 | 3314 | dec_dentry_nr(); |
11fdf7f2 | 3315 | ldout(cct, 20) << "unlink inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3316 | } |
3317 | ||
3318 | if (keepdentry) { | |
3319 | dn->lease_mds = -1; | |
3320 | } else { | |
3321 | ldout(cct, 15) << "unlink removing '" << dn->name << "' dn " << dn << dendl; | |
3322 | ||
3323 | // unlink from dir | |
11fdf7f2 TL |
3324 | Dir *dir = dn->dir; |
3325 | dn->detach(); | |
7c673cae FG |
3326 | |
3327 | // delete den | |
3328 | lru.lru_remove(dn); | |
3329 | dn->put(); | |
11fdf7f2 TL |
3330 | |
3331 | if (dir->is_empty() && !keepdir) | |
3332 | close_dir(dir); | |
7c673cae FG |
3333 | } |
3334 | } | |
3335 | ||
3336 | /** | |
3337 | * For asynchronous flushes, check for errors from the IO and | |
3338 | * update the inode if necessary | |
3339 | */ | |
3340 | class C_Client_FlushComplete : public Context { | |
3341 | private: | |
3342 | Client *client; | |
3343 | InodeRef inode; | |
3344 | public: | |
3345 | C_Client_FlushComplete(Client *c, Inode *in) : client(c), inode(in) { } | |
3346 | void finish(int r) override { | |
9f95a23c | 3347 | ceph_assert(ceph_mutex_is_locked_by_me(client->client_lock)); |
7c673cae FG |
3348 | if (r != 0) { |
3349 | client_t const whoami = client->whoami; // For the benefit of ldout prefix | |
3350 | ldout(client->cct, 1) << "I/O error from flush on inode " << inode | |
3351 | << " 0x" << std::hex << inode->ino << std::dec | |
3352 | << ": " << r << "(" << cpp_strerror(r) << ")" << dendl; | |
3353 | inode->set_async_err(r); | |
3354 | } | |
3355 | } | |
3356 | }; | |
3357 | ||
3358 | ||
3359 | /**** | |
3360 | * caps | |
3361 | */ | |
3362 | ||
3363 | void Client::get_cap_ref(Inode *in, int cap) | |
3364 | { | |
3365 | if ((cap & CEPH_CAP_FILE_BUFFER) && | |
3366 | in->cap_refs[CEPH_CAP_FILE_BUFFER] == 0) { | |
11fdf7f2 | 3367 | ldout(cct, 5) << __func__ << " got first FILE_BUFFER ref on " << *in << dendl; |
b3b6e05e | 3368 | in->iget(); |
7c673cae FG |
3369 | } |
3370 | if ((cap & CEPH_CAP_FILE_CACHE) && | |
3371 | in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
11fdf7f2 | 3372 | ldout(cct, 5) << __func__ << " got first FILE_CACHE ref on " << *in << dendl; |
b3b6e05e | 3373 | in->iget(); |
7c673cae FG |
3374 | } |
3375 | in->get_cap_ref(cap); | |
3376 | } | |
3377 | ||
3378 | void Client::put_cap_ref(Inode *in, int cap) | |
3379 | { | |
3380 | int last = in->put_cap_ref(cap); | |
3381 | if (last) { | |
3382 | int put_nref = 0; | |
3383 | int drop = last & ~in->caps_issued(); | |
3384 | if (in->snapid == CEPH_NOSNAP) { | |
f67539c2 | 3385 | if ((last & (CEPH_CAP_FILE_WR | CEPH_CAP_FILE_BUFFER)) && |
7c673cae FG |
3386 | !in->cap_snaps.empty() && |
3387 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3388 | ldout(cct, 10) << __func__ << " finishing pending cap_snap on " << *in << dendl; |
7c673cae FG |
3389 | in->cap_snaps.rbegin()->second.writing = 0; |
3390 | finish_cap_snap(in, in->cap_snaps.rbegin()->second, get_caps_used(in)); | |
3391 | signal_cond_list(in->waitfor_caps); // wake up blocked sync writers | |
3392 | } | |
3393 | if (last & CEPH_CAP_FILE_BUFFER) { | |
3394 | for (auto &p : in->cap_snaps) | |
3395 | p.second.dirty_data = 0; | |
3396 | signal_cond_list(in->waitfor_commit); | |
11fdf7f2 | 3397 | ldout(cct, 5) << __func__ << " dropped last FILE_BUFFER ref on " << *in << dendl; |
7c673cae FG |
3398 | ++put_nref; |
3399 | } | |
3400 | } | |
3401 | if (last & CEPH_CAP_FILE_CACHE) { | |
11fdf7f2 | 3402 | ldout(cct, 5) << __func__ << " dropped last FILE_CACHE ref on " << *in << dendl; |
7c673cae FG |
3403 | ++put_nref; |
3404 | } | |
3405 | if (drop) | |
3406 | check_caps(in, 0); | |
3407 | if (put_nref) | |
3408 | put_inode(in, put_nref); | |
3409 | } | |
3410 | } | |
3411 | ||
f67539c2 TL |
3412 | // get caps for a given file handle -- the inode should have @need caps |
3413 | // issued by the mds and @want caps not revoked (or not under revocation). | |
3414 | // this routine blocks till the cap requirement is satisfied. also account | |
3415 | // (track) for capability hit when required (when cap requirement succeedes). | |
f6b5b4d7 | 3416 | int Client::get_caps(Fh *fh, int need, int want, int *phave, loff_t endoff) |
7c673cae | 3417 | { |
f6b5b4d7 TL |
3418 | Inode *in = fh->inode.get(); |
3419 | ||
7c673cae FG |
3420 | int r = check_pool_perm(in, need); |
3421 | if (r < 0) | |
3422 | return r; | |
3423 | ||
3424 | while (1) { | |
3425 | int file_wanted = in->caps_file_wanted(); | |
3426 | if ((file_wanted & need) != need) { | |
3427 | ldout(cct, 10) << "get_caps " << *in << " need " << ccap_string(need) | |
3428 | << " file_wanted " << ccap_string(file_wanted) << ", EBADF " | |
3429 | << dendl; | |
f67539c2 | 3430 | return -CEPHFS_EBADF; |
7c673cae FG |
3431 | } |
3432 | ||
f6b5b4d7 | 3433 | if ((fh->mode & CEPH_FILE_MODE_WR) && fh->gen != fd_gen) |
f67539c2 | 3434 | return -CEPHFS_EBADF; |
f6b5b4d7 TL |
3435 | |
3436 | if ((in->flags & I_ERROR_FILELOCK) && fh->has_any_filelocks()) | |
f67539c2 | 3437 | return -CEPHFS_EIO; |
f6b5b4d7 | 3438 | |
7c673cae FG |
3439 | int implemented; |
3440 | int have = in->caps_issued(&implemented); | |
3441 | ||
3442 | bool waitfor_caps = false; | |
3443 | bool waitfor_commit = false; | |
3444 | ||
3445 | if (have & need & CEPH_CAP_FILE_WR) { | |
1911f103 TL |
3446 | if (endoff > 0) { |
3447 | if ((endoff >= (loff_t)in->max_size || | |
3448 | endoff > (loff_t)(in->size << 1)) && | |
3449 | endoff > (loff_t)in->wanted_max_size) { | |
3450 | ldout(cct, 10) << "wanted_max_size " << in->wanted_max_size << " -> " << endoff << dendl; | |
3451 | in->wanted_max_size = endoff; | |
3452 | } | |
3453 | if (in->wanted_max_size > in->max_size && | |
3454 | in->wanted_max_size > in->requested_max_size) | |
3455 | check_caps(in, 0); | |
7c673cae FG |
3456 | } |
3457 | ||
3458 | if (endoff >= 0 && endoff > (loff_t)in->max_size) { | |
3459 | ldout(cct, 10) << "waiting on max_size, endoff " << endoff << " max_size " << in->max_size << " on " << *in << dendl; | |
3460 | waitfor_caps = true; | |
3461 | } | |
3462 | if (!in->cap_snaps.empty()) { | |
3463 | if (in->cap_snaps.rbegin()->second.writing) { | |
3464 | ldout(cct, 10) << "waiting on cap_snap write to complete" << dendl; | |
3465 | waitfor_caps = true; | |
3466 | } | |
3467 | for (auto &p : in->cap_snaps) { | |
3468 | if (p.second.dirty_data) { | |
3469 | waitfor_commit = true; | |
3470 | break; | |
3471 | } | |
3472 | } | |
3473 | if (waitfor_commit) { | |
3474 | _flush(in, new C_Client_FlushComplete(this, in)); | |
3475 | ldout(cct, 10) << "waiting for WRBUFFER to get dropped" << dendl; | |
3476 | } | |
3477 | } | |
3478 | } | |
3479 | ||
3480 | if (!waitfor_caps && !waitfor_commit) { | |
3481 | if ((have & need) == need) { | |
7c673cae FG |
3482 | int revoking = implemented & ~have; |
3483 | ldout(cct, 10) << "get_caps " << *in << " have " << ccap_string(have) | |
3484 | << " need " << ccap_string(need) << " want " << ccap_string(want) | |
c07f9fc5 | 3485 | << " revoking " << ccap_string(revoking) |
7c673cae | 3486 | << dendl; |
c07f9fc5 | 3487 | if ((revoking & want) == 0) { |
7c673cae FG |
3488 | *phave = need | (have & want); |
3489 | in->get_cap_ref(need); | |
f67539c2 | 3490 | cap_hit(); |
7c673cae FG |
3491 | return 0; |
3492 | } | |
3493 | } | |
3494 | ldout(cct, 10) << "waiting for caps " << *in << " need " << ccap_string(need) << " want " << ccap_string(want) << dendl; | |
3495 | waitfor_caps = true; | |
3496 | } | |
3497 | ||
3498 | if ((need & CEPH_CAP_FILE_WR) && in->auth_cap && | |
3499 | in->auth_cap->session->readonly) | |
f67539c2 | 3500 | return -CEPHFS_EROFS; |
7c673cae FG |
3501 | |
3502 | if (in->flags & I_CAP_DROPPED) { | |
3503 | int mds_wanted = in->caps_mds_wanted(); | |
3504 | if ((mds_wanted & need) != need) { | |
3505 | int ret = _renew_caps(in); | |
3506 | if (ret < 0) | |
3507 | return ret; | |
3508 | continue; | |
3509 | } | |
a8e16298 | 3510 | if (!(file_wanted & ~mds_wanted)) |
7c673cae | 3511 | in->flags &= ~I_CAP_DROPPED; |
7c673cae FG |
3512 | } |
3513 | ||
3514 | if (waitfor_caps) | |
3515 | wait_on_list(in->waitfor_caps); | |
3516 | else if (waitfor_commit) | |
3517 | wait_on_list(in->waitfor_commit); | |
3518 | } | |
3519 | } | |
3520 | ||
3521 | int Client::get_caps_used(Inode *in) | |
3522 | { | |
3523 | unsigned used = in->caps_used(); | |
3524 | if (!(used & CEPH_CAP_FILE_CACHE) && | |
3525 | !objectcacher->set_is_empty(&in->oset)) | |
3526 | used |= CEPH_CAP_FILE_CACHE; | |
3527 | return used; | |
3528 | } | |
3529 | ||
3530 | void Client::cap_delay_requeue(Inode *in) | |
3531 | { | |
11fdf7f2 | 3532 | ldout(cct, 10) << __func__ << " on " << *in << dendl; |
7c673cae FG |
3533 | in->hold_caps_until = ceph_clock_now(); |
3534 | in->hold_caps_until += cct->_conf->client_caps_release_delay; | |
28e407b8 | 3535 | delayed_list.push_back(&in->delay_cap_item); |
7c673cae FG |
3536 | } |
3537 | ||
3538 | void Client::send_cap(Inode *in, MetaSession *session, Cap *cap, | |
eafe8130 | 3539 | int flags, int used, int want, int retain, |
7c673cae FG |
3540 | int flush, ceph_tid_t flush_tid) |
3541 | { | |
3542 | int held = cap->issued | cap->implemented; | |
3543 | int revoking = cap->implemented & ~cap->issued; | |
3544 | retain &= ~revoking; | |
3545 | int dropping = cap->issued & ~retain; | |
3546 | int op = CEPH_CAP_OP_UPDATE; | |
3547 | ||
11fdf7f2 | 3548 | ldout(cct, 10) << __func__ << " " << *in |
7c673cae | 3549 | << " mds." << session->mds_num << " seq " << cap->seq |
7c673cae FG |
3550 | << " used " << ccap_string(used) |
3551 | << " want " << ccap_string(want) | |
3552 | << " flush " << ccap_string(flush) | |
3553 | << " retain " << ccap_string(retain) | |
3554 | << " held "<< ccap_string(held) | |
3555 | << " revoking " << ccap_string(revoking) | |
3556 | << " dropping " << ccap_string(dropping) | |
3557 | << dendl; | |
3558 | ||
3559 | if (cct->_conf->client_inject_release_failure && revoking) { | |
3560 | const int would_have_issued = cap->issued & retain; | |
3561 | const int would_have_implemented = cap->implemented & (cap->issued | used); | |
3562 | // Simulated bug: | |
3563 | // - tell the server we think issued is whatever they issued plus whatever we implemented | |
3564 | // - leave what we have implemented in place | |
3565 | ldout(cct, 20) << __func__ << " injecting failure to release caps" << dendl; | |
3566 | cap->issued = cap->issued | cap->implemented; | |
3567 | ||
3568 | // Make an exception for revoking xattr caps: we are injecting | |
3569 | // failure to release other caps, but allow xattr because client | |
3570 | // will block on xattr ops if it can't release these to MDS (#9800) | |
3571 | const int xattr_mask = CEPH_CAP_XATTR_SHARED | CEPH_CAP_XATTR_EXCL; | |
3572 | cap->issued ^= xattr_mask & revoking; | |
3573 | cap->implemented ^= xattr_mask & revoking; | |
3574 | ||
3575 | ldout(cct, 20) << __func__ << " issued " << ccap_string(cap->issued) << " vs " << ccap_string(would_have_issued) << dendl; | |
3576 | ldout(cct, 20) << __func__ << " implemented " << ccap_string(cap->implemented) << " vs " << ccap_string(would_have_implemented) << dendl; | |
3577 | } else { | |
3578 | // Normal behaviour | |
3579 | cap->issued &= retain; | |
3580 | cap->implemented &= cap->issued | used; | |
3581 | } | |
3582 | ||
3583 | snapid_t follows = 0; | |
3584 | ||
3585 | if (flush) | |
3586 | follows = in->snaprealm->get_snap_context().seq; | |
20effc67 | 3587 | |
9f95a23c | 3588 | auto m = make_message<MClientCaps>(op, |
7c673cae FG |
3589 | in->ino, |
3590 | 0, | |
3591 | cap->cap_id, cap->seq, | |
3592 | cap->implemented, | |
3593 | want, | |
3594 | flush, | |
3595 | cap->mseq, | |
3596 | cap_epoch_barrier); | |
3597 | m->caller_uid = in->cap_dirtier_uid; | |
3598 | m->caller_gid = in->cap_dirtier_gid; | |
3599 | ||
3600 | m->head.issue_seq = cap->issue_seq; | |
3601 | m->set_tid(flush_tid); | |
3602 | ||
3603 | m->head.uid = in->uid; | |
3604 | m->head.gid = in->gid; | |
3605 | m->head.mode = in->mode; | |
20effc67 | 3606 | |
7c673cae | 3607 | m->head.nlink = in->nlink; |
20effc67 | 3608 | |
7c673cae | 3609 | if (flush & CEPH_CAP_XATTR_EXCL) { |
11fdf7f2 | 3610 | encode(in->xattrs, m->xattrbl); |
7c673cae FG |
3611 | m->head.xattr_version = in->xattr_version; |
3612 | } | |
20effc67 | 3613 | |
7c673cae FG |
3614 | m->size = in->size; |
3615 | m->max_size = in->max_size; | |
3616 | m->truncate_seq = in->truncate_seq; | |
3617 | m->truncate_size = in->truncate_size; | |
3618 | m->mtime = in->mtime; | |
3619 | m->atime = in->atime; | |
3620 | m->ctime = in->ctime; | |
3621 | m->btime = in->btime; | |
3622 | m->time_warp_seq = in->time_warp_seq; | |
3623 | m->change_attr = in->change_attr; | |
eafe8130 TL |
3624 | |
3625 | if (!(flags & MClientCaps::FLAG_PENDING_CAPSNAP) && | |
3626 | !in->cap_snaps.empty() && | |
3627 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3628 | flags |= MClientCaps::FLAG_PENDING_CAPSNAP; | |
3629 | m->flags = flags; | |
3630 | ||
7c673cae FG |
3631 | if (flush & CEPH_CAP_FILE_WR) { |
3632 | m->inline_version = in->inline_version; | |
3633 | m->inline_data = in->inline_data; | |
3634 | } | |
3635 | ||
3636 | in->reported_size = in->size; | |
3637 | m->set_snap_follows(follows); | |
3638 | cap->wanted = want; | |
3639 | if (cap == in->auth_cap) { | |
1911f103 TL |
3640 | if (want & CEPH_CAP_ANY_FILE_WR) { |
3641 | m->set_max_size(in->wanted_max_size); | |
3642 | in->requested_max_size = in->wanted_max_size; | |
3643 | ldout(cct, 15) << "auth cap, requesting max_size " << in->requested_max_size << dendl; | |
3644 | } else { | |
3645 | in->requested_max_size = 0; | |
3646 | ldout(cct, 15) << "auth cap, reset requested_max_size due to not wanting any file write cap" << dendl; | |
3647 | } | |
7c673cae FG |
3648 | } |
3649 | ||
3650 | if (!session->flushing_caps_tids.empty()) | |
3651 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3652 | ||
11fdf7f2 | 3653 | session->con->send_message2(std::move(m)); |
7c673cae FG |
3654 | } |
3655 | ||
31f18b77 FG |
3656 | static bool is_max_size_approaching(Inode *in) |
3657 | { | |
3658 | /* mds will adjust max size according to the reported size */ | |
3659 | if (in->flushing_caps & CEPH_CAP_FILE_WR) | |
3660 | return false; | |
3661 | if (in->size >= in->max_size) | |
3662 | return true; | |
3663 | /* half of previous max_size increment has been used */ | |
3664 | if (in->max_size > in->reported_size && | |
3665 | (in->size << 1) >= in->max_size + in->reported_size) | |
3666 | return true; | |
3667 | return false; | |
3668 | } | |
7c673cae | 3669 | |
11fdf7f2 TL |
3670 | static int adjust_caps_used_for_lazyio(int used, int issued, int implemented) |
3671 | { | |
3672 | if (!(used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER))) | |
3673 | return used; | |
3674 | if (!(implemented & CEPH_CAP_FILE_LAZYIO)) | |
3675 | return used; | |
3676 | ||
3677 | if (issued & CEPH_CAP_FILE_LAZYIO) { | |
3678 | if (!(issued & CEPH_CAP_FILE_CACHE)) { | |
3679 | used &= ~CEPH_CAP_FILE_CACHE; | |
3680 | used |= CEPH_CAP_FILE_LAZYIO; | |
3681 | } | |
3682 | if (!(issued & CEPH_CAP_FILE_BUFFER)) { | |
3683 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3684 | used |= CEPH_CAP_FILE_LAZYIO; | |
3685 | } | |
3686 | } else { | |
3687 | if (!(implemented & CEPH_CAP_FILE_CACHE)) { | |
3688 | used &= ~CEPH_CAP_FILE_CACHE; | |
3689 | used |= CEPH_CAP_FILE_LAZYIO; | |
3690 | } | |
3691 | if (!(implemented & CEPH_CAP_FILE_BUFFER)) { | |
3692 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3693 | used |= CEPH_CAP_FILE_LAZYIO; | |
3694 | } | |
3695 | } | |
3696 | return used; | |
3697 | } | |
3698 | ||
7c673cae FG |
3699 | /** |
3700 | * check_caps | |
3701 | * | |
3702 | * Examine currently used and wanted versus held caps. Release, flush or ack | |
3703 | * revoked caps to the MDS as appropriate. | |
3704 | * | |
3705 | * @param in the inode to check | |
3706 | * @param flags flags to apply to cap check | |
3707 | */ | |
3708 | void Client::check_caps(Inode *in, unsigned flags) | |
3709 | { | |
3710 | unsigned wanted = in->caps_wanted(); | |
3711 | unsigned used = get_caps_used(in); | |
3712 | unsigned cap_used; | |
3713 | ||
7c673cae FG |
3714 | int implemented; |
3715 | int issued = in->caps_issued(&implemented); | |
3716 | int revoking = implemented & ~issued; | |
3717 | ||
11fdf7f2 TL |
3718 | int orig_used = used; |
3719 | used = adjust_caps_used_for_lazyio(used, issued, implemented); | |
3720 | ||
7c673cae | 3721 | int retain = wanted | used | CEPH_CAP_PIN; |
f67539c2 | 3722 | if (!is_unmounting() && in->nlink > 0) { |
a8e16298 | 3723 | if (wanted) { |
7c673cae | 3724 | retain |= CEPH_CAP_ANY; |
a8e16298 TL |
3725 | } else if (in->is_dir() && |
3726 | (issued & CEPH_CAP_FILE_SHARED) && | |
3727 | (in->flags & I_COMPLETE)) { | |
3728 | // we do this here because we don't want to drop to Fs (and then | |
3729 | // drop the Fs if we do a create!) if that alone makes us send lookups | |
3730 | // to the MDS. Doing it in in->caps_wanted() has knock-on effects elsewhere | |
3731 | wanted = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; | |
3732 | retain |= wanted; | |
3733 | } else { | |
7c673cae | 3734 | retain |= CEPH_CAP_ANY_SHARED; |
a8e16298 TL |
3735 | // keep RD only if we didn't have the file open RW, |
3736 | // because then the mds would revoke it anyway to | |
3737 | // journal max_size=0. | |
3738 | if (in->max_size == 0) | |
3739 | retain |= CEPH_CAP_ANY_RD; | |
3740 | } | |
7c673cae FG |
3741 | } |
3742 | ||
11fdf7f2 | 3743 | ldout(cct, 10) << __func__ << " on " << *in |
7c673cae FG |
3744 | << " wanted " << ccap_string(wanted) |
3745 | << " used " << ccap_string(used) | |
3746 | << " issued " << ccap_string(issued) | |
3747 | << " revoking " << ccap_string(revoking) | |
3748 | << " flags=" << flags | |
3749 | << dendl; | |
3750 | ||
3751 | if (in->snapid != CEPH_NOSNAP) | |
3752 | return; //snap caps last forever, can't write | |
3753 | ||
3754 | if (in->caps.empty()) | |
3755 | return; // guard if at end of func | |
3756 | ||
11fdf7f2 TL |
3757 | if (!(orig_used & CEPH_CAP_FILE_BUFFER) && |
3758 | (revoking & used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
94b18763 | 3759 | if (_release(in)) |
11fdf7f2 | 3760 | used &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
94b18763 | 3761 | } |
7c673cae | 3762 | |
20effc67 TL |
3763 | for (auto &[mds, cap] : in->caps) { |
3764 | auto session = mds_sessions.at(mds); | |
7c673cae FG |
3765 | |
3766 | cap_used = used; | |
11fdf7f2 | 3767 | if (in->auth_cap && &cap != in->auth_cap) |
7c673cae FG |
3768 | cap_used &= ~in->auth_cap->issued; |
3769 | ||
11fdf7f2 | 3770 | revoking = cap.implemented & ~cap.issued; |
20effc67 | 3771 | |
7c673cae | 3772 | ldout(cct, 10) << " cap mds." << mds |
11fdf7f2 TL |
3773 | << " issued " << ccap_string(cap.issued) |
3774 | << " implemented " << ccap_string(cap.implemented) | |
7c673cae FG |
3775 | << " revoking " << ccap_string(revoking) << dendl; |
3776 | ||
3777 | if (in->wanted_max_size > in->max_size && | |
3778 | in->wanted_max_size > in->requested_max_size && | |
11fdf7f2 | 3779 | &cap == in->auth_cap) |
7c673cae FG |
3780 | goto ack; |
3781 | ||
3782 | /* approaching file_max? */ | |
11fdf7f2 TL |
3783 | if ((cap.issued & CEPH_CAP_FILE_WR) && |
3784 | &cap == in->auth_cap && | |
31f18b77 | 3785 | is_max_size_approaching(in)) { |
7c673cae | 3786 | ldout(cct, 10) << "size " << in->size << " approaching max_size " << in->max_size |
31f18b77 | 3787 | << ", reported " << in->reported_size << dendl; |
7c673cae FG |
3788 | goto ack; |
3789 | } | |
3790 | ||
3791 | /* completed revocation? */ | |
3792 | if (revoking && (revoking & cap_used) == 0) { | |
11fdf7f2 | 3793 | ldout(cct, 10) << "completed revocation of " << ccap_string(cap.implemented & ~cap.issued) << dendl; |
7c673cae FG |
3794 | goto ack; |
3795 | } | |
3796 | ||
3797 | /* want more caps from mds? */ | |
11fdf7f2 | 3798 | if (wanted & ~(cap.wanted | cap.issued)) |
7c673cae FG |
3799 | goto ack; |
3800 | ||
f67539c2 | 3801 | if (!revoking && is_unmounting() && (cap_used == 0)) |
7c673cae FG |
3802 | goto ack; |
3803 | ||
11fdf7f2 | 3804 | if ((cap.issued & ~retain) == 0 && // and we don't have anything we wouldn't like |
a8e16298 | 3805 | !in->dirty_caps) // and we have no dirty caps |
7c673cae FG |
3806 | continue; |
3807 | ||
11fdf7f2 | 3808 | if (!(flags & CHECK_CAPS_NODELAY)) { |
7c673cae | 3809 | ldout(cct, 10) << "delaying cap release" << dendl; |
11fdf7f2 | 3810 | cap_delay_requeue(in); |
7c673cae FG |
3811 | continue; |
3812 | } | |
3813 | ||
3814 | ack: | |
eafe8130 TL |
3815 | if (&cap == in->auth_cap) { |
3816 | if (in->flags & I_KICK_FLUSH) { | |
3817 | ldout(cct, 20) << " reflushing caps (check_caps) on " << *in | |
3818 | << " to mds." << mds << dendl; | |
20effc67 | 3819 | kick_flushing_caps(in, session.get()); |
eafe8130 TL |
3820 | } |
3821 | if (!in->cap_snaps.empty() && | |
3822 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3823 | flush_snaps(in); | |
7c673cae FG |
3824 | } |
3825 | ||
3826 | int flushing; | |
e306af50 | 3827 | int msg_flags = 0; |
7c673cae | 3828 | ceph_tid_t flush_tid; |
11fdf7f2 | 3829 | if (in->auth_cap == &cap && in->dirty_caps) { |
7c673cae | 3830 | flushing = mark_caps_flushing(in, &flush_tid); |
e306af50 TL |
3831 | if (flags & CHECK_CAPS_SYNCHRONOUS) |
3832 | msg_flags |= MClientCaps::FLAG_SYNC; | |
7c673cae FG |
3833 | } else { |
3834 | flushing = 0; | |
3835 | flush_tid = 0; | |
3836 | } | |
3837 | ||
20effc67 TL |
3838 | in->delay_cap_item.remove_myself(); |
3839 | send_cap(in, session.get(), &cap, msg_flags, cap_used, wanted, retain, | |
eafe8130 | 3840 | flushing, flush_tid); |
7c673cae FG |
3841 | } |
3842 | } | |
3843 | ||
3844 | ||
3845 | void Client::queue_cap_snap(Inode *in, SnapContext& old_snapc) | |
3846 | { | |
3847 | int used = get_caps_used(in); | |
3848 | int dirty = in->caps_dirty(); | |
11fdf7f2 | 3849 | ldout(cct, 10) << __func__ << " " << *in << " snapc " << old_snapc << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3850 | |
3851 | if (in->cap_snaps.size() && | |
3852 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3853 | ldout(cct, 10) << __func__ << " already have pending cap_snap on " << *in << dendl; |
7c673cae FG |
3854 | return; |
3855 | } else if (in->caps_dirty() || | |
3856 | (used & CEPH_CAP_FILE_WR) || | |
3857 | (dirty & CEPH_CAP_ANY_WR)) { | |
3858 | const auto &capsnapem = in->cap_snaps.emplace(std::piecewise_construct, std::make_tuple(old_snapc.seq), std::make_tuple(in)); | |
11fdf7f2 | 3859 | ceph_assert(capsnapem.second); /* element inserted */ |
7c673cae FG |
3860 | CapSnap &capsnap = capsnapem.first->second; |
3861 | capsnap.context = old_snapc; | |
3862 | capsnap.issued = in->caps_issued(); | |
3863 | capsnap.dirty = in->caps_dirty(); | |
f67539c2 | 3864 | |
7c673cae | 3865 | capsnap.dirty_data = (used & CEPH_CAP_FILE_BUFFER); |
f67539c2 | 3866 | |
7c673cae FG |
3867 | capsnap.uid = in->uid; |
3868 | capsnap.gid = in->gid; | |
3869 | capsnap.mode = in->mode; | |
3870 | capsnap.btime = in->btime; | |
3871 | capsnap.xattrs = in->xattrs; | |
3872 | capsnap.xattr_version = in->xattr_version; | |
11fdf7f2 TL |
3873 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; |
3874 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
f67539c2 | 3875 | |
7c673cae | 3876 | if (used & CEPH_CAP_FILE_WR) { |
11fdf7f2 | 3877 | ldout(cct, 10) << __func__ << " WR used on " << *in << dendl; |
7c673cae FG |
3878 | capsnap.writing = 1; |
3879 | } else { | |
3880 | finish_cap_snap(in, capsnap, used); | |
3881 | } | |
3882 | } else { | |
11fdf7f2 | 3883 | ldout(cct, 10) << __func__ << " not dirty|writing on " << *in << dendl; |
7c673cae FG |
3884 | } |
3885 | } | |
3886 | ||
3887 | void Client::finish_cap_snap(Inode *in, CapSnap &capsnap, int used) | |
3888 | { | |
11fdf7f2 | 3889 | ldout(cct, 10) << __func__ << " " << *in << " capsnap " << (void *)&capsnap << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3890 | capsnap.size = in->size; |
3891 | capsnap.mtime = in->mtime; | |
3892 | capsnap.atime = in->atime; | |
3893 | capsnap.ctime = in->ctime; | |
3894 | capsnap.time_warp_seq = in->time_warp_seq; | |
3895 | capsnap.change_attr = in->change_attr; | |
7c673cae FG |
3896 | capsnap.dirty |= in->caps_dirty(); |
3897 | ||
11fdf7f2 TL |
3898 | /* Only reset it if it wasn't set before */ |
3899 | if (capsnap.cap_dirtier_uid == -1) { | |
3900 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; | |
3901 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
3902 | } | |
3903 | ||
7c673cae FG |
3904 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { |
3905 | capsnap.inline_data = in->inline_data; | |
3906 | capsnap.inline_version = in->inline_version; | |
3907 | } | |
3908 | ||
3909 | if (used & CEPH_CAP_FILE_BUFFER) { | |
f67539c2 | 3910 | capsnap.writing = 1; |
11fdf7f2 | 3911 | ldout(cct, 10) << __func__ << " " << *in << " cap_snap " << &capsnap << " used " << used |
7c673cae FG |
3912 | << " WRBUFFER, delaying" << dendl; |
3913 | } else { | |
3914 | capsnap.dirty_data = 0; | |
3915 | flush_snaps(in); | |
3916 | } | |
3917 | } | |
3918 | ||
eafe8130 TL |
3919 | void Client::send_flush_snap(Inode *in, MetaSession *session, |
3920 | snapid_t follows, CapSnap& capsnap) | |
3921 | { | |
9f95a23c TL |
3922 | auto m = make_message<MClientCaps>(CEPH_CAP_OP_FLUSHSNAP, |
3923 | in->ino, in->snaprealm->ino, 0, | |
3924 | in->auth_cap->mseq, cap_epoch_barrier); | |
eafe8130 TL |
3925 | m->caller_uid = capsnap.cap_dirtier_uid; |
3926 | m->caller_gid = capsnap.cap_dirtier_gid; | |
3927 | ||
3928 | m->set_client_tid(capsnap.flush_tid); | |
3929 | m->head.snap_follows = follows; | |
3930 | ||
3931 | m->head.caps = capsnap.issued; | |
3932 | m->head.dirty = capsnap.dirty; | |
3933 | ||
3934 | m->head.uid = capsnap.uid; | |
3935 | m->head.gid = capsnap.gid; | |
3936 | m->head.mode = capsnap.mode; | |
3937 | m->btime = capsnap.btime; | |
3938 | ||
3939 | m->size = capsnap.size; | |
3940 | ||
3941 | m->head.xattr_version = capsnap.xattr_version; | |
3942 | encode(capsnap.xattrs, m->xattrbl); | |
3943 | ||
3944 | m->ctime = capsnap.ctime; | |
3945 | m->btime = capsnap.btime; | |
3946 | m->mtime = capsnap.mtime; | |
3947 | m->atime = capsnap.atime; | |
3948 | m->time_warp_seq = capsnap.time_warp_seq; | |
3949 | m->change_attr = capsnap.change_attr; | |
3950 | ||
3951 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { | |
3952 | m->inline_version = in->inline_version; | |
3953 | m->inline_data = in->inline_data; | |
3954 | } | |
3955 | ||
3956 | ceph_assert(!session->flushing_caps_tids.empty()); | |
3957 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3958 | ||
3959 | session->con->send_message2(std::move(m)); | |
3960 | } | |
3961 | ||
3962 | void Client::flush_snaps(Inode *in) | |
7c673cae | 3963 | { |
eafe8130 | 3964 | ldout(cct, 10) << "flush_snaps on " << *in << dendl; |
11fdf7f2 | 3965 | ceph_assert(in->cap_snaps.size()); |
7c673cae FG |
3966 | |
3967 | // pick auth mds | |
11fdf7f2 | 3968 | ceph_assert(in->auth_cap); |
7c673cae | 3969 | MetaSession *session = in->auth_cap->session; |
7c673cae FG |
3970 | |
3971 | for (auto &p : in->cap_snaps) { | |
3972 | CapSnap &capsnap = p.second; | |
eafe8130 TL |
3973 | // only do new flush |
3974 | if (capsnap.flush_tid > 0) | |
3975 | continue; | |
7c673cae FG |
3976 | |
3977 | ldout(cct, 10) << "flush_snaps mds." << session->mds_num | |
3978 | << " follows " << p.first | |
3979 | << " size " << capsnap.size | |
3980 | << " mtime " << capsnap.mtime | |
3981 | << " dirty_data=" << capsnap.dirty_data | |
3982 | << " writing=" << capsnap.writing | |
3983 | << " on " << *in << dendl; | |
3984 | if (capsnap.dirty_data || capsnap.writing) | |
eafe8130 | 3985 | break; |
f67539c2 | 3986 | |
eafe8130 TL |
3987 | capsnap.flush_tid = ++last_flush_tid; |
3988 | session->flushing_caps_tids.insert(capsnap.flush_tid); | |
3989 | in->flushing_cap_tids[capsnap.flush_tid] = 0; | |
3990 | if (!in->flushing_cap_item.is_on_list()) | |
3991 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
7c673cae | 3992 | |
eafe8130 | 3993 | send_flush_snap(in, session, p.first, capsnap); |
7c673cae FG |
3994 | } |
3995 | } | |
3996 | ||
9f95a23c | 3997 | void Client::wait_on_list(list<ceph::condition_variable*>& ls) |
7c673cae | 3998 | { |
9f95a23c | 3999 | ceph::condition_variable cond; |
7c673cae | 4000 | ls.push_back(&cond); |
9f95a23c TL |
4001 | std::unique_lock l{client_lock, std::adopt_lock}; |
4002 | cond.wait(l); | |
4003 | l.release(); | |
7c673cae FG |
4004 | ls.remove(&cond); |
4005 | } | |
4006 | ||
9f95a23c | 4007 | void Client::signal_cond_list(list<ceph::condition_variable*>& ls) |
7c673cae | 4008 | { |
9f95a23c TL |
4009 | for (auto cond : ls) { |
4010 | cond->notify_all(); | |
4011 | } | |
7c673cae FG |
4012 | } |
4013 | ||
4014 | void Client::wait_on_context_list(list<Context*>& ls) | |
4015 | { | |
9f95a23c | 4016 | ceph::condition_variable cond; |
7c673cae FG |
4017 | bool done = false; |
4018 | int r; | |
9f95a23c TL |
4019 | ls.push_back(new C_Cond(cond, &done, &r)); |
4020 | std::unique_lock l{client_lock, std::adopt_lock}; | |
4021 | cond.wait(l, [&done] { return done;}); | |
4022 | l.release(); | |
7c673cae FG |
4023 | } |
4024 | ||
4025 | void Client::signal_context_list(list<Context*>& ls) | |
4026 | { | |
4027 | while (!ls.empty()) { | |
4028 | ls.front()->complete(0); | |
4029 | ls.pop_front(); | |
4030 | } | |
4031 | } | |
4032 | ||
a8e16298 | 4033 | void Client::wake_up_session_caps(MetaSession *s, bool reconnect) |
7c673cae | 4034 | { |
11fdf7f2 TL |
4035 | for (const auto &cap : s->caps) { |
4036 | auto &in = cap->inode; | |
a8e16298 | 4037 | if (reconnect) { |
11fdf7f2 TL |
4038 | in.requested_max_size = 0; |
4039 | in.wanted_max_size = 0; | |
a8e16298 TL |
4040 | } else { |
4041 | if (cap->gen < s->cap_gen) { | |
4042 | // mds did not re-issue stale cap. | |
4043 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
4044 | // make sure mds knows what we want. | |
11fdf7f2 TL |
4045 | if (in.caps_file_wanted() & ~cap->wanted) |
4046 | in.flags |= I_CAP_DROPPED; | |
a8e16298 TL |
4047 | } |
4048 | } | |
11fdf7f2 | 4049 | signal_cond_list(in.waitfor_caps); |
7c673cae FG |
4050 | } |
4051 | } | |
4052 | ||
4053 | ||
4054 | // flush dirty data (from objectcache) | |
4055 | ||
4056 | class C_Client_CacheInvalidate : public Context { | |
4057 | private: | |
4058 | Client *client; | |
4059 | vinodeno_t ino; | |
4060 | int64_t offset, length; | |
4061 | public: | |
4062 | C_Client_CacheInvalidate(Client *c, Inode *in, int64_t off, int64_t len) : | |
4063 | client(c), offset(off), length(len) { | |
4064 | if (client->use_faked_inos()) | |
4065 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4066 | else | |
4067 | ino = in->vino(); | |
4068 | } | |
4069 | void finish(int r) override { | |
4070 | // _async_invalidate takes the lock when it needs to, call this back from outside of lock. | |
9f95a23c | 4071 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
4072 | client->_async_invalidate(ino, offset, length); |
4073 | } | |
4074 | }; | |
4075 | ||
4076 | void Client::_async_invalidate(vinodeno_t ino, int64_t off, int64_t len) | |
4077 | { | |
f67539c2 TL |
4078 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4079 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4080 | return; |
f67539c2 | 4081 | |
11fdf7f2 | 4082 | ldout(cct, 10) << __func__ << " " << ino << " " << off << "~" << len << dendl; |
7c673cae FG |
4083 | ino_invalidate_cb(callback_handle, ino, off, len); |
4084 | } | |
4085 | ||
4086 | void Client::_schedule_invalidate_callback(Inode *in, int64_t off, int64_t len) { | |
4087 | ||
4088 | if (ino_invalidate_cb) | |
4089 | // we queue the invalidate, which calls the callback and decrements the ref | |
4090 | async_ino_invalidator.queue(new C_Client_CacheInvalidate(this, in, off, len)); | |
4091 | } | |
4092 | ||
4093 | void Client::_invalidate_inode_cache(Inode *in) | |
4094 | { | |
11fdf7f2 | 4095 | ldout(cct, 10) << __func__ << " " << *in << dendl; |
7c673cae FG |
4096 | |
4097 | // invalidate our userspace inode cache | |
94b18763 | 4098 | if (cct->_conf->client_oc) { |
7c673cae | 4099 | objectcacher->release_set(&in->oset); |
94b18763 FG |
4100 | if (!objectcacher->set_is_empty(&in->oset)) |
4101 | lderr(cct) << "failed to invalidate cache for " << *in << dendl; | |
4102 | } | |
7c673cae FG |
4103 | |
4104 | _schedule_invalidate_callback(in, 0, 0); | |
4105 | } | |
4106 | ||
4107 | void Client::_invalidate_inode_cache(Inode *in, int64_t off, int64_t len) | |
4108 | { | |
11fdf7f2 | 4109 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
4110 | |
4111 | // invalidate our userspace inode cache | |
4112 | if (cct->_conf->client_oc) { | |
4113 | vector<ObjectExtent> ls; | |
4114 | Striper::file_to_extents(cct, in->ino, &in->layout, off, len, in->truncate_size, ls); | |
28e407b8 | 4115 | objectcacher->discard_writeback(&in->oset, ls, nullptr); |
7c673cae FG |
4116 | } |
4117 | ||
4118 | _schedule_invalidate_callback(in, off, len); | |
4119 | } | |
4120 | ||
4121 | bool Client::_release(Inode *in) | |
4122 | { | |
4123 | ldout(cct, 20) << "_release " << *in << dendl; | |
4124 | if (in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
4125 | _invalidate_inode_cache(in); | |
4126 | return true; | |
4127 | } | |
4128 | return false; | |
4129 | } | |
4130 | ||
4131 | bool Client::_flush(Inode *in, Context *onfinish) | |
4132 | { | |
4133 | ldout(cct, 10) << "_flush " << *in << dendl; | |
4134 | ||
4135 | if (!in->oset.dirty_or_tx) { | |
4136 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4137 | onfinish->complete(0); | |
4138 | return true; | |
4139 | } | |
4140 | ||
4141 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
1adf2230 | 4142 | ldout(cct, 8) << __func__ << ": FULL, purging for ENOSPC" << dendl; |
7c673cae FG |
4143 | objectcacher->purge_set(&in->oset); |
4144 | if (onfinish) { | |
f67539c2 | 4145 | onfinish->complete(-CEPHFS_ENOSPC); |
7c673cae FG |
4146 | } |
4147 | return true; | |
4148 | } | |
4149 | ||
4150 | return objectcacher->flush_set(&in->oset, onfinish); | |
4151 | } | |
4152 | ||
4153 | void Client::_flush_range(Inode *in, int64_t offset, uint64_t size) | |
4154 | { | |
f67539c2 | 4155 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
4156 | if (!in->oset.dirty_or_tx) { |
4157 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4158 | return; | |
4159 | } | |
4160 | ||
11fdf7f2 | 4161 | C_SaferCond onflush("Client::_flush_range flock"); |
7c673cae | 4162 | bool ret = objectcacher->file_flush(&in->oset, &in->layout, in->snaprealm->get_snap_context(), |
11fdf7f2 | 4163 | offset, size, &onflush); |
7c673cae FG |
4164 | if (!ret) { |
4165 | // wait for flush | |
9f95a23c | 4166 | client_lock.unlock(); |
11fdf7f2 | 4167 | onflush.wait(); |
9f95a23c | 4168 | client_lock.lock(); |
7c673cae FG |
4169 | } |
4170 | } | |
4171 | ||
4172 | void Client::flush_set_callback(ObjectCacher::ObjectSet *oset) | |
4173 | { | |
f67539c2 TL |
4174 | // std::scoped_lock l(client_lock); |
4175 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); // will be called via dispatch() -> objecter -> ... | |
7c673cae | 4176 | Inode *in = static_cast<Inode *>(oset->parent); |
11fdf7f2 | 4177 | ceph_assert(in); |
7c673cae FG |
4178 | _flushed(in); |
4179 | } | |
4180 | ||
4181 | void Client::_flushed(Inode *in) | |
4182 | { | |
4183 | ldout(cct, 10) << "_flushed " << *in << dendl; | |
4184 | ||
4185 | put_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
4186 | } | |
4187 | ||
4188 | ||
4189 | ||
4190 | // checks common to add_update_cap, handle_cap_grant | |
11fdf7f2 | 4191 | void Client::check_cap_issue(Inode *in, unsigned issued) |
7c673cae FG |
4192 | { |
4193 | unsigned had = in->caps_issued(); | |
4194 | ||
4195 | if ((issued & CEPH_CAP_FILE_CACHE) && | |
4196 | !(had & CEPH_CAP_FILE_CACHE)) | |
4197 | in->cache_gen++; | |
4198 | ||
f91f0fd5 TL |
4199 | if ((issued & CEPH_CAP_FILE_SHARED) != |
4200 | (had & CEPH_CAP_FILE_SHARED)) { | |
4201 | if (issued & CEPH_CAP_FILE_SHARED) | |
4202 | in->shared_gen++; | |
7c673cae FG |
4203 | if (in->is_dir()) |
4204 | clear_dir_complete_and_ordered(in, true); | |
4205 | } | |
4206 | } | |
4207 | ||
4208 | void Client::add_update_cap(Inode *in, MetaSession *mds_session, uint64_t cap_id, | |
a8e16298 TL |
4209 | unsigned issued, unsigned wanted, unsigned seq, unsigned mseq, |
4210 | inodeno_t realm, int flags, const UserPerm& cap_perms) | |
7c673cae | 4211 | { |
11fdf7f2 TL |
4212 | if (!in->is_any_caps()) { |
4213 | ceph_assert(in->snaprealm == 0); | |
4214 | in->snaprealm = get_snap_realm(realm); | |
4215 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4216 | ldout(cct, 15) << __func__ << " first one, opened snaprealm " << in->snaprealm << dendl; | |
4217 | } else { | |
4218 | ceph_assert(in->snaprealm); | |
4219 | if ((flags & CEPH_CAP_FLAG_AUTH) && | |
4220 | realm != inodeno_t(-1) && in->snaprealm->ino != realm) { | |
4221 | in->snaprealm_item.remove_myself(); | |
4222 | auto oldrealm = in->snaprealm; | |
4223 | in->snaprealm = get_snap_realm(realm); | |
4224 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4225 | put_snap_realm(oldrealm); | |
4226 | } | |
4227 | } | |
4228 | ||
7c673cae | 4229 | mds_rank_t mds = mds_session->mds_num; |
11fdf7f2 TL |
4230 | const auto &capem = in->caps.emplace(std::piecewise_construct, std::forward_as_tuple(mds), std::forward_as_tuple(*in, mds_session)); |
4231 | Cap &cap = capem.first->second; | |
4232 | if (!capem.second) { | |
4233 | if (cap.gen < mds_session->cap_gen) | |
4234 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
7c673cae FG |
4235 | |
4236 | /* | |
4237 | * auth mds of the inode changed. we received the cap export | |
4238 | * message, but still haven't received the cap import message. | |
4239 | * handle_cap_export() updated the new auth MDS' cap. | |
4240 | * | |
4241 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
4242 | * a message that was send before the cap import message. So | |
4243 | * don't remove caps. | |
4244 | */ | |
11fdf7f2 | 4245 | if (ceph_seq_cmp(seq, cap.seq) <= 0) { |
92f5a8d4 TL |
4246 | if (&cap != in->auth_cap) |
4247 | ldout(cct, 0) << "WARNING: " << "inode " << *in << " caps on mds." << mds << " != auth_cap." << dendl; | |
4248 | ||
11fdf7f2 TL |
4249 | ceph_assert(cap.cap_id == cap_id); |
4250 | seq = cap.seq; | |
4251 | mseq = cap.mseq; | |
4252 | issued |= cap.issued; | |
7c673cae FG |
4253 | flags |= CEPH_CAP_FLAG_AUTH; |
4254 | } | |
f67539c2 TL |
4255 | } else { |
4256 | inc_pinned_icaps(); | |
7c673cae FG |
4257 | } |
4258 | ||
11fdf7f2 | 4259 | check_cap_issue(in, issued); |
7c673cae FG |
4260 | |
4261 | if (flags & CEPH_CAP_FLAG_AUTH) { | |
11fdf7f2 | 4262 | if (in->auth_cap != &cap && |
7c673cae FG |
4263 | (!in->auth_cap || ceph_seq_cmp(in->auth_cap->mseq, mseq) < 0)) { |
4264 | if (in->auth_cap && in->flushing_cap_item.is_on_list()) { | |
11fdf7f2 | 4265 | ldout(cct, 10) << __func__ << " changing auth cap: " |
7c673cae FG |
4266 | << "add myself to new auth MDS' flushing caps list" << dendl; |
4267 | adjust_session_flushing_caps(in, in->auth_cap->session, mds_session); | |
4268 | } | |
11fdf7f2 | 4269 | in->auth_cap = ∩ |
7c673cae FG |
4270 | } |
4271 | } | |
4272 | ||
11fdf7f2 TL |
4273 | unsigned old_caps = cap.issued; |
4274 | cap.cap_id = cap_id; | |
4275 | cap.issued = issued; | |
4276 | cap.implemented |= issued; | |
4277 | if (ceph_seq_cmp(mseq, cap.mseq) > 0) | |
4278 | cap.wanted = wanted; | |
a8e16298 | 4279 | else |
11fdf7f2 TL |
4280 | cap.wanted |= wanted; |
4281 | cap.seq = seq; | |
4282 | cap.issue_seq = seq; | |
4283 | cap.mseq = mseq; | |
4284 | cap.gen = mds_session->cap_gen; | |
4285 | cap.latest_perms = cap_perms; | |
4286 | ldout(cct, 10) << __func__ << " issued " << ccap_string(old_caps) << " -> " << ccap_string(cap.issued) | |
4287 | << " from mds." << mds | |
4288 | << " on " << *in | |
4289 | << dendl; | |
4290 | ||
4291 | if ((issued & ~old_caps) && in->auth_cap == &cap) { | |
7c673cae | 4292 | // non-auth MDS is revoking the newly grant caps ? |
11fdf7f2 TL |
4293 | for (auto &p : in->caps) { |
4294 | if (&p.second == &cap) | |
7c673cae | 4295 | continue; |
11fdf7f2 | 4296 | if (p.second.implemented & ~p.second.issued & issued) { |
7c673cae FG |
4297 | check_caps(in, CHECK_CAPS_NODELAY); |
4298 | break; | |
4299 | } | |
4300 | } | |
4301 | } | |
4302 | ||
4303 | if (issued & ~old_caps) | |
4304 | signal_cond_list(in->waitfor_caps); | |
4305 | } | |
4306 | ||
4307 | void Client::remove_cap(Cap *cap, bool queue_release) | |
4308 | { | |
11fdf7f2 | 4309 | auto &in = cap->inode; |
7c673cae FG |
4310 | MetaSession *session = cap->session; |
4311 | mds_rank_t mds = cap->session->mds_num; | |
4312 | ||
11fdf7f2 | 4313 | ldout(cct, 10) << __func__ << " mds." << mds << " on " << in << dendl; |
7c673cae FG |
4314 | |
4315 | if (queue_release) { | |
4316 | session->enqueue_cap_release( | |
11fdf7f2 | 4317 | in.ino, |
7c673cae FG |
4318 | cap->cap_id, |
4319 | cap->issue_seq, | |
4320 | cap->mseq, | |
4321 | cap_epoch_barrier); | |
f67539c2 TL |
4322 | } else { |
4323 | dec_pinned_icaps(); | |
7c673cae FG |
4324 | } |
4325 | ||
f67539c2 | 4326 | |
11fdf7f2 TL |
4327 | if (in.auth_cap == cap) { |
4328 | if (in.flushing_cap_item.is_on_list()) { | |
7c673cae | 4329 | ldout(cct, 10) << " removing myself from flushing_cap list" << dendl; |
11fdf7f2 | 4330 | in.flushing_cap_item.remove_myself(); |
7c673cae | 4331 | } |
11fdf7f2 | 4332 | in.auth_cap = NULL; |
7c673cae | 4333 | } |
11fdf7f2 TL |
4334 | size_t n = in.caps.erase(mds); |
4335 | ceph_assert(n == 1); | |
7c673cae FG |
4336 | cap = nullptr; |
4337 | ||
11fdf7f2 TL |
4338 | if (!in.is_any_caps()) { |
4339 | ldout(cct, 15) << __func__ << " last one, closing snaprealm " << in.snaprealm << dendl; | |
4340 | in.snaprealm_item.remove_myself(); | |
4341 | put_snap_realm(in.snaprealm); | |
4342 | in.snaprealm = 0; | |
7c673cae FG |
4343 | } |
4344 | } | |
4345 | ||
4346 | void Client::remove_all_caps(Inode *in) | |
4347 | { | |
4348 | while (!in->caps.empty()) | |
11fdf7f2 | 4349 | remove_cap(&in->caps.begin()->second, true); |
7c673cae FG |
4350 | } |
4351 | ||
f6b5b4d7 | 4352 | void Client::remove_session_caps(MetaSession *s, int err) |
7c673cae | 4353 | { |
11fdf7f2 | 4354 | ldout(cct, 10) << __func__ << " mds." << s->mds_num << dendl; |
7c673cae FG |
4355 | |
4356 | while (s->caps.size()) { | |
4357 | Cap *cap = *s->caps.begin(); | |
11fdf7f2 | 4358 | InodeRef in(&cap->inode); |
eafe8130 | 4359 | bool dirty_caps = false; |
7c673cae | 4360 | if (in->auth_cap == cap) { |
7c673cae FG |
4361 | dirty_caps = in->dirty_caps | in->flushing_caps; |
4362 | in->wanted_max_size = 0; | |
4363 | in->requested_max_size = 0; | |
f6b5b4d7 TL |
4364 | if (in->has_any_filelocks()) |
4365 | in->flags |= I_ERROR_FILELOCK; | |
7c673cae | 4366 | } |
f6b5b4d7 | 4367 | auto caps = cap->implemented; |
a8e16298 TL |
4368 | if (cap->wanted | cap->issued) |
4369 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 4370 | remove_cap(cap, false); |
eafe8130 | 4371 | in->cap_snaps.clear(); |
7c673cae | 4372 | if (dirty_caps) { |
11fdf7f2 | 4373 | lderr(cct) << __func__ << " still has dirty|flushing caps on " << *in << dendl; |
7c673cae FG |
4374 | if (in->flushing_caps) { |
4375 | num_flushing_caps--; | |
4376 | in->flushing_cap_tids.clear(); | |
4377 | } | |
4378 | in->flushing_caps = 0; | |
28e407b8 | 4379 | in->mark_caps_clean(); |
11fdf7f2 | 4380 | put_inode(in.get()); |
7c673cae | 4381 | } |
f6b5b4d7 TL |
4382 | caps &= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER; |
4383 | if (caps && !in->caps_issued_mask(caps, true)) { | |
f67539c2 | 4384 | if (err == -CEPHFS_EBLOCKLISTED) { |
f6b5b4d7 TL |
4385 | if (in->oset.dirty_or_tx) { |
4386 | lderr(cct) << __func__ << " still has dirty data on " << *in << dendl; | |
4387 | in->set_async_err(err); | |
4388 | } | |
4389 | objectcacher->purge_set(&in->oset); | |
4390 | } else { | |
4391 | objectcacher->release_set(&in->oset); | |
4392 | } | |
4393 | _schedule_invalidate_callback(in.get(), 0, 0); | |
4394 | } | |
4395 | ||
a8e16298 | 4396 | signal_cond_list(in->waitfor_caps); |
7c673cae FG |
4397 | } |
4398 | s->flushing_caps_tids.clear(); | |
9f95a23c | 4399 | sync_cond.notify_all(); |
7c673cae FG |
4400 | } |
4401 | ||
91327a77 | 4402 | int Client::_do_remount(bool retry_on_error) |
b32b8144 | 4403 | { |
adb31ebb | 4404 | uint64_t max_retries = cct->_conf.get_val<uint64_t>("mds_max_retries_on_remount_failure"); |
91327a77 | 4405 | |
b32b8144 FG |
4406 | errno = 0; |
4407 | int r = remount_cb(callback_handle); | |
91327a77 AA |
4408 | if (r == 0) { |
4409 | retries_on_invalidate = 0; | |
4410 | } else { | |
b32b8144 FG |
4411 | int e = errno; |
4412 | client_t whoami = get_nodeid(); | |
4413 | if (r == -1) { | |
4414 | lderr(cct) << | |
4415 | "failed to remount (to trim kernel dentries): " | |
4416 | "errno = " << e << " (" << strerror(e) << ")" << dendl; | |
4417 | } else { | |
4418 | lderr(cct) << | |
4419 | "failed to remount (to trim kernel dentries): " | |
4420 | "return code = " << r << dendl; | |
4421 | } | |
91327a77 | 4422 | bool should_abort = |
11fdf7f2 TL |
4423 | (cct->_conf.get_val<bool>("client_die_on_failed_remount") || |
4424 | cct->_conf.get_val<bool>("client_die_on_failed_dentry_invalidate")) && | |
91327a77 | 4425 | !(retry_on_error && (++retries_on_invalidate < max_retries)); |
f67539c2 | 4426 | if (should_abort && !is_unmounting()) { |
b32b8144 FG |
4427 | lderr(cct) << "failed to remount for kernel dentry trimming; quitting!" << dendl; |
4428 | ceph_abort(); | |
4429 | } | |
4430 | } | |
4431 | return r; | |
4432 | } | |
4433 | ||
7c673cae FG |
4434 | class C_Client_Remount : public Context { |
4435 | private: | |
4436 | Client *client; | |
4437 | public: | |
4438 | explicit C_Client_Remount(Client *c) : client(c) {} | |
4439 | void finish(int r) override { | |
11fdf7f2 | 4440 | ceph_assert(r == 0); |
91327a77 | 4441 | client->_do_remount(true); |
7c673cae FG |
4442 | } |
4443 | }; | |
4444 | ||
4445 | void Client::_invalidate_kernel_dcache() | |
4446 | { | |
f67539c2 TL |
4447 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4448 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4449 | return; |
f67539c2 | 4450 | |
94b18763 FG |
4451 | if (can_invalidate_dentries) { |
4452 | if (dentry_invalidate_cb && root->dir) { | |
4453 | for (ceph::unordered_map<string, Dentry*>::iterator p = root->dir->dentries.begin(); | |
4454 | p != root->dir->dentries.end(); | |
4455 | ++p) { | |
4456 | if (p->second->inode) | |
4457 | _schedule_invalidate_dentry_callback(p->second, false); | |
4458 | } | |
7c673cae FG |
4459 | } |
4460 | } else if (remount_cb) { | |
4461 | // Hacky: | |
4462 | // when remounting a file system, linux kernel trims all unused dentries in the fs | |
4463 | remount_finisher.queue(new C_Client_Remount(this)); | |
4464 | } | |
4465 | } | |
4466 | ||
91327a77 AA |
4467 | void Client::_trim_negative_child_dentries(InodeRef& in) |
4468 | { | |
4469 | if (!in->is_dir()) | |
4470 | return; | |
4471 | ||
4472 | Dir* dir = in->dir; | |
4473 | if (dir && dir->dentries.size() == dir->num_null_dentries) { | |
4474 | for (auto p = dir->dentries.begin(); p != dir->dentries.end(); ) { | |
4475 | Dentry *dn = p->second; | |
4476 | ++p; | |
11fdf7f2 | 4477 | ceph_assert(!dn->inode); |
91327a77 AA |
4478 | if (dn->lru_is_expireable()) |
4479 | unlink(dn, true, false); // keep dir, drop dentry | |
4480 | } | |
4481 | if (dir->dentries.empty()) { | |
4482 | close_dir(dir); | |
4483 | } | |
4484 | } | |
4485 | ||
4486 | if (in->flags & I_SNAPDIR_OPEN) { | |
4487 | InodeRef snapdir = open_snapdir(in.get()); | |
4488 | _trim_negative_child_dentries(snapdir); | |
4489 | } | |
4490 | } | |
4491 | ||
e306af50 TL |
4492 | class C_Client_CacheRelease : public Context { |
4493 | private: | |
4494 | Client *client; | |
4495 | vinodeno_t ino; | |
4496 | public: | |
4497 | C_Client_CacheRelease(Client *c, Inode *in) : | |
4498 | client(c) { | |
4499 | if (client->use_faked_inos()) | |
4500 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4501 | else | |
4502 | ino = in->vino(); | |
4503 | } | |
4504 | void finish(int r) override { | |
4505 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); | |
4506 | client->_async_inode_release(ino); | |
4507 | } | |
4508 | }; | |
4509 | ||
4510 | void Client::_async_inode_release(vinodeno_t ino) | |
4511 | { | |
f67539c2 TL |
4512 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4513 | if (!mref_reader.is_state_satisfied()) | |
e306af50 | 4514 | return; |
f67539c2 | 4515 | |
e306af50 TL |
4516 | ldout(cct, 10) << __func__ << " " << ino << dendl; |
4517 | ino_release_cb(callback_handle, ino); | |
4518 | } | |
4519 | ||
4520 | void Client::_schedule_ino_release_callback(Inode *in) { | |
4521 | ||
4522 | if (ino_release_cb) | |
4523 | // we queue the invalidate, which calls the callback and decrements the ref | |
4524 | async_ino_releasor.queue(new C_Client_CacheRelease(this, in)); | |
4525 | } | |
4526 | ||
28e407b8 | 4527 | void Client::trim_caps(MetaSession *s, uint64_t max) |
7c673cae FG |
4528 | { |
4529 | mds_rank_t mds = s->mds_num; | |
28e407b8 | 4530 | size_t caps_size = s->caps.size(); |
11fdf7f2 | 4531 | ldout(cct, 10) << __func__ << " mds." << mds << " max " << max |
7c673cae FG |
4532 | << " caps " << caps_size << dendl; |
4533 | ||
28e407b8 AA |
4534 | uint64_t trimmed = 0; |
4535 | auto p = s->caps.begin(); | |
4536 | std::set<Dentry *> to_trim; /* this avoids caps other than the one we're | |
4537 | * looking at from getting deleted during traversal. */ | |
7c673cae FG |
4538 | while ((caps_size - trimmed) > max && !p.end()) { |
4539 | Cap *cap = *p; | |
11fdf7f2 | 4540 | InodeRef in(&cap->inode); |
7c673cae FG |
4541 | |
4542 | // Increment p early because it will be invalidated if cap | |
4543 | // is deleted inside remove_cap | |
4544 | ++p; | |
4545 | ||
4546 | if (in->caps.size() > 1 && cap != in->auth_cap) { | |
4547 | int mine = cap->issued | cap->implemented; | |
4548 | int oissued = in->auth_cap ? in->auth_cap->issued : 0; | |
4549 | // disposable non-auth cap | |
b32b8144 | 4550 | if (!(get_caps_used(in.get()) & ~oissued & mine)) { |
7c673cae | 4551 | ldout(cct, 20) << " removing unused, unneeded non-auth cap on " << *in << dendl; |
28e407b8 | 4552 | cap = (remove_cap(cap, true), nullptr); |
7c673cae FG |
4553 | trimmed++; |
4554 | } | |
4555 | } else { | |
4556 | ldout(cct, 20) << " trying to trim dentries for " << *in << dendl; | |
91327a77 | 4557 | _trim_negative_child_dentries(in); |
7c673cae | 4558 | bool all = true; |
11fdf7f2 TL |
4559 | auto q = in->dentries.begin(); |
4560 | while (q != in->dentries.end()) { | |
4561 | Dentry *dn = *q; | |
4562 | ++q; | |
7c673cae FG |
4563 | if (dn->lru_is_expireable()) { |
4564 | if (can_invalidate_dentries && | |
b3b6e05e | 4565 | dn->dir->parent_inode->ino == CEPH_INO_ROOT) { |
7c673cae FG |
4566 | // Only issue one of these per DN for inodes in root: handle |
4567 | // others more efficiently by calling for root-child DNs at | |
4568 | // the end of this function. | |
4569 | _schedule_invalidate_dentry_callback(dn, true); | |
4570 | } | |
28e407b8 AA |
4571 | ldout(cct, 20) << " queueing dentry for trimming: " << dn->name << dendl; |
4572 | to_trim.insert(dn); | |
7c673cae FG |
4573 | } else { |
4574 | ldout(cct, 20) << " not expirable: " << dn->name << dendl; | |
4575 | all = false; | |
4576 | } | |
4577 | } | |
b3b6e05e | 4578 | if (in->ll_ref == 1 && in->ino != CEPH_INO_ROOT) { |
f91f0fd5 TL |
4579 | _schedule_ino_release_callback(in.get()); |
4580 | } | |
b3b6e05e | 4581 | if (all && in->ino != CEPH_INO_ROOT) { |
7c673cae FG |
4582 | ldout(cct, 20) << __func__ << " counting as trimmed: " << *in << dendl; |
4583 | trimmed++; | |
4584 | } | |
4585 | } | |
4586 | } | |
28e407b8 AA |
4587 | ldout(cct, 20) << " trimming queued dentries: " << dendl; |
4588 | for (const auto &dn : to_trim) { | |
4589 | trim_dentry(dn); | |
4590 | } | |
4591 | to_trim.clear(); | |
7c673cae | 4592 | |
b32b8144 | 4593 | caps_size = s->caps.size(); |
11fdf7f2 | 4594 | if (caps_size > (size_t)max) |
7c673cae FG |
4595 | _invalidate_kernel_dcache(); |
4596 | } | |
4597 | ||
4598 | void Client::force_session_readonly(MetaSession *s) | |
4599 | { | |
4600 | s->readonly = true; | |
4601 | for (xlist<Cap*>::iterator p = s->caps.begin(); !p.end(); ++p) { | |
11fdf7f2 TL |
4602 | auto &in = (*p)->inode; |
4603 | if (in.caps_wanted() & CEPH_CAP_FILE_WR) | |
4604 | signal_cond_list(in.waitfor_caps); | |
7c673cae FG |
4605 | } |
4606 | } | |
4607 | ||
7c673cae FG |
4608 | int Client::mark_caps_flushing(Inode *in, ceph_tid_t* ptid) |
4609 | { | |
4610 | MetaSession *session = in->auth_cap->session; | |
4611 | ||
4612 | int flushing = in->dirty_caps; | |
11fdf7f2 | 4613 | ceph_assert(flushing); |
7c673cae FG |
4614 | |
4615 | ceph_tid_t flush_tid = ++last_flush_tid; | |
4616 | in->flushing_cap_tids[flush_tid] = flushing; | |
4617 | ||
4618 | if (!in->flushing_caps) { | |
11fdf7f2 | 4619 | ldout(cct, 10) << __func__ << " " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4620 | num_flushing_caps++; |
4621 | } else { | |
11fdf7f2 | 4622 | ldout(cct, 10) << __func__ << " (more) " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4623 | } |
4624 | ||
4625 | in->flushing_caps |= flushing; | |
28e407b8 | 4626 | in->mark_caps_clean(); |
7c673cae FG |
4627 | |
4628 | if (!in->flushing_cap_item.is_on_list()) | |
4629 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
4630 | session->flushing_caps_tids.insert(flush_tid); | |
4631 | ||
4632 | *ptid = flush_tid; | |
4633 | return flushing; | |
4634 | } | |
4635 | ||
4636 | void Client::adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s) | |
4637 | { | |
4638 | for (auto &p : in->cap_snaps) { | |
4639 | CapSnap &capsnap = p.second; | |
4640 | if (capsnap.flush_tid > 0) { | |
4641 | old_s->flushing_caps_tids.erase(capsnap.flush_tid); | |
4642 | new_s->flushing_caps_tids.insert(capsnap.flush_tid); | |
4643 | } | |
4644 | } | |
4645 | for (map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4646 | it != in->flushing_cap_tids.end(); | |
4647 | ++it) { | |
4648 | old_s->flushing_caps_tids.erase(it->first); | |
4649 | new_s->flushing_caps_tids.insert(it->first); | |
4650 | } | |
4651 | new_s->flushing_caps.push_back(&in->flushing_cap_item); | |
4652 | } | |
4653 | ||
4654 | /* | |
20effc67 TL |
4655 | * Flush all the dirty caps back to the MDS. Because the callers |
4656 | * generally wait on the result of this function (syncfs and umount | |
4657 | * cases), we set CHECK_CAPS_SYNCHRONOUS on the last check_caps call. | |
7c673cae FG |
4658 | */ |
4659 | void Client::flush_caps_sync() | |
4660 | { | |
4661 | ldout(cct, 10) << __func__ << dendl; | |
20effc67 TL |
4662 | for (auto &q : mds_sessions) { |
4663 | auto s = q.second; | |
4664 | xlist<Inode*>::iterator p = s->dirty_list.begin(); | |
4665 | while (!p.end()) { | |
4666 | unsigned flags = CHECK_CAPS_NODELAY; | |
4667 | Inode *in = *p; | |
7c673cae | 4668 | |
20effc67 TL |
4669 | ++p; |
4670 | if (p.end()) | |
4671 | flags |= CHECK_CAPS_SYNCHRONOUS; | |
4672 | check_caps(in, flags); | |
4673 | } | |
7c673cae FG |
4674 | } |
4675 | } | |
4676 | ||
7c673cae FG |
4677 | void Client::wait_sync_caps(Inode *in, ceph_tid_t want) |
4678 | { | |
4679 | while (in->flushing_caps) { | |
4680 | map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
11fdf7f2 | 4681 | ceph_assert(it != in->flushing_cap_tids.end()); |
7c673cae FG |
4682 | if (it->first > want) |
4683 | break; | |
11fdf7f2 | 4684 | ldout(cct, 10) << __func__ << " on " << *in << " flushing " |
7c673cae FG |
4685 | << ccap_string(it->second) << " want " << want |
4686 | << " last " << it->first << dendl; | |
4687 | wait_on_list(in->waitfor_caps); | |
4688 | } | |
4689 | } | |
4690 | ||
4691 | void Client::wait_sync_caps(ceph_tid_t want) | |
4692 | { | |
4693 | retry: | |
11fdf7f2 | 4694 | ldout(cct, 10) << __func__ << " want " << want << " (last is " << last_flush_tid << ", " |
7c673cae | 4695 | << num_flushing_caps << " total flushing)" << dendl; |
11fdf7f2 | 4696 | for (auto &p : mds_sessions) { |
20effc67 | 4697 | auto s = p.second; |
7c673cae FG |
4698 | if (s->flushing_caps_tids.empty()) |
4699 | continue; | |
4700 | ceph_tid_t oldest_tid = *s->flushing_caps_tids.begin(); | |
4701 | if (oldest_tid <= want) { | |
11fdf7f2 | 4702 | ldout(cct, 10) << " waiting on mds." << p.first << " tid " << oldest_tid |
7c673cae | 4703 | << " (want " << want << ")" << dendl; |
9f95a23c TL |
4704 | std::unique_lock l{client_lock, std::adopt_lock}; |
4705 | sync_cond.wait(l); | |
4706 | l.release(); | |
7c673cae FG |
4707 | goto retry; |
4708 | } | |
4709 | } | |
4710 | } | |
4711 | ||
eafe8130 TL |
4712 | void Client::kick_flushing_caps(Inode *in, MetaSession *session) |
4713 | { | |
4714 | in->flags &= ~I_KICK_FLUSH; | |
4715 | ||
4716 | Cap *cap = in->auth_cap; | |
4717 | ceph_assert(cap->session == session); | |
4718 | ||
4719 | ceph_tid_t last_snap_flush = 0; | |
4720 | for (auto p = in->flushing_cap_tids.rbegin(); | |
4721 | p != in->flushing_cap_tids.rend(); | |
4722 | ++p) { | |
4723 | if (!p->second) { | |
4724 | last_snap_flush = p->first; | |
4725 | break; | |
4726 | } | |
4727 | } | |
4728 | ||
4729 | int wanted = in->caps_wanted(); | |
4730 | int used = get_caps_used(in) | in->caps_dirty(); | |
4731 | auto it = in->cap_snaps.begin(); | |
4732 | for (auto& p : in->flushing_cap_tids) { | |
4733 | if (p.second) { | |
4734 | int msg_flags = p.first < last_snap_flush ? MClientCaps::FLAG_PENDING_CAPSNAP : 0; | |
4735 | send_cap(in, session, cap, msg_flags, used, wanted, (cap->issued | cap->implemented), | |
4736 | p.second, p.first); | |
4737 | } else { | |
4738 | ceph_assert(it != in->cap_snaps.end()); | |
4739 | ceph_assert(it->second.flush_tid == p.first); | |
4740 | send_flush_snap(in, session, it->first, it->second); | |
4741 | ++it; | |
4742 | } | |
4743 | } | |
4744 | } | |
4745 | ||
7c673cae FG |
4746 | void Client::kick_flushing_caps(MetaSession *session) |
4747 | { | |
4748 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 4749 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
4750 | |
4751 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { | |
4752 | Inode *in = *p; | |
eafe8130 TL |
4753 | if (in->flags & I_KICK_FLUSH) { |
4754 | ldout(cct, 20) << " reflushing caps on " << *in << " to mds." << mds << dendl; | |
4755 | kick_flushing_caps(in, session); | |
4756 | } | |
7c673cae | 4757 | } |
7c673cae FG |
4758 | } |
4759 | ||
4760 | void Client::early_kick_flushing_caps(MetaSession *session) | |
4761 | { | |
7c673cae FG |
4762 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { |
4763 | Inode *in = *p; | |
11fdf7f2 TL |
4764 | Cap *cap = in->auth_cap; |
4765 | ceph_assert(cap); | |
7c673cae FG |
4766 | |
4767 | // if flushing caps were revoked, we re-send the cap flush in client reconnect | |
4768 | // stage. This guarantees that MDS processes the cap flush message before issuing | |
4769 | // the flushing caps to other client. | |
eafe8130 TL |
4770 | if ((in->flushing_caps & in->auth_cap->issued) == in->flushing_caps) { |
4771 | in->flags |= I_KICK_FLUSH; | |
7c673cae | 4772 | continue; |
eafe8130 | 4773 | } |
7c673cae FG |
4774 | |
4775 | ldout(cct, 20) << " reflushing caps (early_kick) on " << *in | |
4776 | << " to mds." << session->mds_num << dendl; | |
11fdf7f2 TL |
4777 | // send_reconnect() also will reset these sequence numbers. make sure |
4778 | // sequence numbers in cap flush message match later reconnect message. | |
4779 | cap->seq = 0; | |
4780 | cap->issue_seq = 0; | |
4781 | cap->mseq = 0; | |
4782 | cap->issued = cap->implemented; | |
4783 | ||
eafe8130 | 4784 | kick_flushing_caps(in, session); |
7c673cae FG |
4785 | } |
4786 | } | |
4787 | ||
7c673cae FG |
4788 | void Client::invalidate_snaprealm_and_children(SnapRealm *realm) |
4789 | { | |
4790 | list<SnapRealm*> q; | |
4791 | q.push_back(realm); | |
4792 | ||
4793 | while (!q.empty()) { | |
4794 | realm = q.front(); | |
4795 | q.pop_front(); | |
4796 | ||
11fdf7f2 | 4797 | ldout(cct, 10) << __func__ << " " << *realm << dendl; |
7c673cae FG |
4798 | realm->invalidate_cache(); |
4799 | ||
4800 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4801 | p != realm->pchildren.end(); | |
4802 | ++p) | |
4803 | q.push_back(*p); | |
4804 | } | |
4805 | } | |
4806 | ||
4807 | SnapRealm *Client::get_snap_realm(inodeno_t r) | |
4808 | { | |
4809 | SnapRealm *realm = snap_realms[r]; | |
4810 | if (!realm) | |
4811 | snap_realms[r] = realm = new SnapRealm(r); | |
11fdf7f2 | 4812 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4813 | realm->nref++; |
4814 | return realm; | |
4815 | } | |
4816 | ||
4817 | SnapRealm *Client::get_snap_realm_maybe(inodeno_t r) | |
4818 | { | |
4819 | if (snap_realms.count(r) == 0) { | |
11fdf7f2 | 4820 | ldout(cct, 20) << __func__ << " " << r << " fail" << dendl; |
7c673cae FG |
4821 | return NULL; |
4822 | } | |
4823 | SnapRealm *realm = snap_realms[r]; | |
11fdf7f2 | 4824 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4825 | realm->nref++; |
4826 | return realm; | |
4827 | } | |
4828 | ||
4829 | void Client::put_snap_realm(SnapRealm *realm) | |
4830 | { | |
11fdf7f2 | 4831 | ldout(cct, 20) << __func__ << " " << realm->ino << " " << realm |
7c673cae FG |
4832 | << " " << realm->nref << " -> " << (realm->nref - 1) << dendl; |
4833 | if (--realm->nref == 0) { | |
4834 | snap_realms.erase(realm->ino); | |
4835 | if (realm->pparent) { | |
4836 | realm->pparent->pchildren.erase(realm); | |
4837 | put_snap_realm(realm->pparent); | |
4838 | } | |
4839 | delete realm; | |
4840 | } | |
4841 | } | |
4842 | ||
4843 | bool Client::adjust_realm_parent(SnapRealm *realm, inodeno_t parent) | |
4844 | { | |
4845 | if (realm->parent != parent) { | |
11fdf7f2 | 4846 | ldout(cct, 10) << __func__ << " " << *realm |
7c673cae FG |
4847 | << " " << realm->parent << " -> " << parent << dendl; |
4848 | realm->parent = parent; | |
4849 | if (realm->pparent) { | |
4850 | realm->pparent->pchildren.erase(realm); | |
4851 | put_snap_realm(realm->pparent); | |
4852 | } | |
4853 | realm->pparent = get_snap_realm(parent); | |
4854 | realm->pparent->pchildren.insert(realm); | |
4855 | return true; | |
4856 | } | |
4857 | return false; | |
4858 | } | |
4859 | ||
4860 | static bool has_new_snaps(const SnapContext& old_snapc, | |
4861 | const SnapContext& new_snapc) | |
4862 | { | |
4863 | return !new_snapc.snaps.empty() && new_snapc.snaps[0] > old_snapc.seq; | |
4864 | } | |
4865 | ||
4866 | ||
11fdf7f2 | 4867 | void Client::update_snap_trace(const bufferlist& bl, SnapRealm **realm_ret, bool flush) |
7c673cae FG |
4868 | { |
4869 | SnapRealm *first_realm = NULL; | |
11fdf7f2 | 4870 | ldout(cct, 10) << __func__ << " len " << bl.length() << dendl; |
7c673cae FG |
4871 | |
4872 | map<SnapRealm*, SnapContext> dirty_realms; | |
4873 | ||
11fdf7f2 | 4874 | auto p = bl.cbegin(); |
7c673cae FG |
4875 | while (!p.end()) { |
4876 | SnapRealmInfo info; | |
11fdf7f2 | 4877 | decode(info, p); |
7c673cae FG |
4878 | SnapRealm *realm = get_snap_realm(info.ino()); |
4879 | ||
4880 | bool invalidate = false; | |
4881 | ||
4882 | if (info.seq() > realm->seq) { | |
11fdf7f2 | 4883 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() << " > " << realm->seq |
7c673cae FG |
4884 | << dendl; |
4885 | ||
4886 | if (flush) { | |
4887 | // writeback any dirty caps _before_ updating snap list (i.e. with old snap info) | |
4888 | // flush me + children | |
4889 | list<SnapRealm*> q; | |
4890 | q.push_back(realm); | |
4891 | while (!q.empty()) { | |
4892 | SnapRealm *realm = q.front(); | |
4893 | q.pop_front(); | |
4894 | ||
4895 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4896 | p != realm->pchildren.end(); | |
4897 | ++p) | |
4898 | q.push_back(*p); | |
4899 | ||
4900 | if (dirty_realms.count(realm) == 0) { | |
4901 | realm->nref++; | |
4902 | dirty_realms[realm] = realm->get_snap_context(); | |
4903 | } | |
4904 | } | |
4905 | } | |
4906 | ||
4907 | // update | |
4908 | realm->seq = info.seq(); | |
4909 | realm->created = info.created(); | |
4910 | realm->parent_since = info.parent_since(); | |
4911 | realm->prior_parent_snaps = info.prior_parent_snaps; | |
4912 | realm->my_snaps = info.my_snaps; | |
4913 | invalidate = true; | |
4914 | } | |
4915 | ||
4916 | // _always_ verify parent | |
4917 | if (adjust_realm_parent(realm, info.parent())) | |
4918 | invalidate = true; | |
4919 | ||
4920 | if (invalidate) { | |
4921 | invalidate_snaprealm_and_children(realm); | |
11fdf7f2 | 4922 | ldout(cct, 15) << __func__ << " " << *realm << " self|parent updated" << dendl; |
7c673cae FG |
4923 | ldout(cct, 15) << " snapc " << realm->get_snap_context() << dendl; |
4924 | } else { | |
11fdf7f2 | 4925 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() |
7c673cae FG |
4926 | << " <= " << realm->seq << " and same parent, SKIPPING" << dendl; |
4927 | } | |
f67539c2 | 4928 | |
7c673cae FG |
4929 | if (!first_realm) |
4930 | first_realm = realm; | |
4931 | else | |
4932 | put_snap_realm(realm); | |
4933 | } | |
4934 | ||
f67539c2 | 4935 | for (auto &[realm, snapc] : dirty_realms) { |
7c673cae | 4936 | // if there are new snaps ? |
f67539c2 | 4937 | if (has_new_snaps(snapc, realm->get_snap_context())) { |
7c673cae | 4938 | ldout(cct, 10) << " flushing caps on " << *realm << dendl; |
f67539c2 TL |
4939 | for (auto&& in : realm->inodes_with_caps) { |
4940 | queue_cap_snap(in, snapc); | |
7c673cae FG |
4941 | } |
4942 | } else { | |
4943 | ldout(cct, 10) << " no new snap on " << *realm << dendl; | |
4944 | } | |
4945 | put_snap_realm(realm); | |
4946 | } | |
4947 | ||
4948 | if (realm_ret) | |
4949 | *realm_ret = first_realm; | |
4950 | else | |
4951 | put_snap_realm(first_realm); | |
4952 | } | |
4953 | ||
11fdf7f2 | 4954 | void Client::handle_snap(const MConstRef<MClientSnap>& m) |
7c673cae | 4955 | { |
11fdf7f2 | 4956 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 4957 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
4958 | |
4959 | std::scoped_lock cl(client_lock); | |
20effc67 | 4960 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 4961 | if (!session) { |
7c673cae FG |
4962 | return; |
4963 | } | |
4964 | ||
20effc67 | 4965 | got_mds_push(session.get()); |
7c673cae FG |
4966 | |
4967 | map<Inode*, SnapContext> to_move; | |
4968 | SnapRealm *realm = 0; | |
4969 | ||
4970 | if (m->head.op == CEPH_SNAP_OP_SPLIT) { | |
11fdf7f2 | 4971 | ceph_assert(m->head.split); |
7c673cae | 4972 | SnapRealmInfo info; |
11fdf7f2 TL |
4973 | auto p = m->bl.cbegin(); |
4974 | decode(info, p); | |
4975 | ceph_assert(info.ino() == m->head.split); | |
7c673cae FG |
4976 | |
4977 | // flush, then move, ino's. | |
4978 | realm = get_snap_realm(info.ino()); | |
4979 | ldout(cct, 10) << " splitting off " << *realm << dendl; | |
11fdf7f2 TL |
4980 | for (auto& ino : m->split_inos) { |
4981 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
7c673cae FG |
4982 | if (inode_map.count(vino)) { |
4983 | Inode *in = inode_map[vino]; | |
4984 | if (!in->snaprealm || in->snaprealm == realm) | |
4985 | continue; | |
4986 | if (in->snaprealm->created > info.created()) { | |
4987 | ldout(cct, 10) << " NOT moving " << *in << " from _newer_ realm " | |
4988 | << *in->snaprealm << dendl; | |
4989 | continue; | |
4990 | } | |
4991 | ldout(cct, 10) << " moving " << *in << " from " << *in->snaprealm << dendl; | |
4992 | ||
4993 | ||
4994 | in->snaprealm_item.remove_myself(); | |
4995 | to_move[in] = in->snaprealm->get_snap_context(); | |
4996 | put_snap_realm(in->snaprealm); | |
4997 | } | |
4998 | } | |
4999 | ||
5000 | // move child snaprealms, too | |
11fdf7f2 TL |
5001 | for (auto& child_realm : m->split_realms) { |
5002 | ldout(cct, 10) << "adjusting snaprealm " << child_realm << " parent" << dendl; | |
5003 | SnapRealm *child = get_snap_realm_maybe(child_realm); | |
7c673cae FG |
5004 | if (!child) |
5005 | continue; | |
5006 | adjust_realm_parent(child, realm->ino); | |
5007 | put_snap_realm(child); | |
5008 | } | |
5009 | } | |
5010 | ||
5011 | update_snap_trace(m->bl, NULL, m->head.op != CEPH_SNAP_OP_DESTROY); | |
5012 | ||
5013 | if (realm) { | |
5014 | for (auto p = to_move.begin(); p != to_move.end(); ++p) { | |
5015 | Inode *in = p->first; | |
5016 | in->snaprealm = realm; | |
5017 | realm->inodes_with_caps.push_back(&in->snaprealm_item); | |
5018 | realm->nref++; | |
5019 | // queue for snap writeback | |
5020 | if (has_new_snaps(p->second, realm->get_snap_context())) | |
5021 | queue_cap_snap(in, p->second); | |
5022 | } | |
5023 | put_snap_realm(realm); | |
5024 | } | |
7c673cae FG |
5025 | } |
5026 | ||
11fdf7f2 | 5027 | void Client::handle_quota(const MConstRef<MClientQuota>& m) |
7c673cae FG |
5028 | { |
5029 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5030 | |
5031 | std::scoped_lock cl(client_lock); | |
20effc67 | 5032 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 5033 | if (!session) { |
7c673cae FG |
5034 | return; |
5035 | } | |
5036 | ||
20effc67 | 5037 | got_mds_push(session.get()); |
7c673cae | 5038 | |
11fdf7f2 | 5039 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << mds << dendl; |
7c673cae FG |
5040 | |
5041 | vinodeno_t vino(m->ino, CEPH_NOSNAP); | |
5042 | if (inode_map.count(vino)) { | |
5043 | Inode *in = NULL; | |
5044 | in = inode_map[vino]; | |
5045 | ||
5046 | if (in) { | |
5047 | in->quota = m->quota; | |
5048 | in->rstat = m->rstat; | |
5049 | } | |
5050 | } | |
7c673cae FG |
5051 | } |
5052 | ||
11fdf7f2 | 5053 | void Client::handle_caps(const MConstRef<MClientCaps>& m) |
7c673cae FG |
5054 | { |
5055 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5056 | |
5057 | std::scoped_lock cl(client_lock); | |
20effc67 | 5058 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 5059 | if (!session) { |
7c673cae FG |
5060 | return; |
5061 | } | |
5062 | ||
5063 | if (m->osd_epoch_barrier && !objecter->have_map(m->osd_epoch_barrier)) { | |
5064 | // Pause RADOS operations until we see the required epoch | |
5065 | objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
5066 | } | |
5067 | ||
5068 | if (m->osd_epoch_barrier > cap_epoch_barrier) { | |
5069 | // Record the barrier so that we will transmit it to MDS when releasing | |
5070 | set_cap_epoch_barrier(m->osd_epoch_barrier); | |
5071 | } | |
5072 | ||
20effc67 | 5073 | got_mds_push(session.get()); |
7c673cae | 5074 | |
11fdf7f2 | 5075 | Inode *in; |
7c673cae | 5076 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); |
11fdf7f2 TL |
5077 | if (auto it = inode_map.find(vino); it != inode_map.end()) { |
5078 | in = it->second; | |
5079 | } else { | |
7c673cae | 5080 | if (m->get_op() == CEPH_CAP_OP_IMPORT) { |
11fdf7f2 | 5081 | ldout(cct, 5) << __func__ << " don't have vino " << vino << " on IMPORT, immediately releasing" << dendl; |
7c673cae FG |
5082 | session->enqueue_cap_release( |
5083 | m->get_ino(), | |
5084 | m->get_cap_id(), | |
5085 | m->get_seq(), | |
5086 | m->get_mseq(), | |
5087 | cap_epoch_barrier); | |
5088 | } else { | |
11fdf7f2 | 5089 | ldout(cct, 5) << __func__ << " don't have vino " << vino << ", dropping" << dendl; |
7c673cae | 5090 | } |
7c673cae FG |
5091 | |
5092 | // in case the mds is waiting on e.g. a revocation | |
5093 | flush_cap_releases(); | |
5094 | return; | |
5095 | } | |
5096 | ||
5097 | switch (m->get_op()) { | |
20effc67 TL |
5098 | case CEPH_CAP_OP_EXPORT: return handle_cap_export(session.get(), in, m); |
5099 | case CEPH_CAP_OP_FLUSHSNAP_ACK: return handle_cap_flushsnap_ack(session.get(), in, m); | |
5100 | case CEPH_CAP_OP_IMPORT: /* no return */ handle_cap_import(session.get(), in, m); | |
7c673cae FG |
5101 | } |
5102 | ||
11fdf7f2 TL |
5103 | if (auto it = in->caps.find(mds); it != in->caps.end()) { |
5104 | Cap &cap = in->caps.at(mds); | |
7c673cae | 5105 | |
11fdf7f2 | 5106 | switch (m->get_op()) { |
20effc67 | 5107 | case CEPH_CAP_OP_TRUNC: return handle_cap_trunc(session.get(), in, m); |
11fdf7f2 TL |
5108 | case CEPH_CAP_OP_IMPORT: |
5109 | case CEPH_CAP_OP_REVOKE: | |
20effc67 TL |
5110 | case CEPH_CAP_OP_GRANT: return handle_cap_grant(session.get(), in, &cap, m); |
5111 | case CEPH_CAP_OP_FLUSH_ACK: return handle_cap_flush_ack(session.get(), in, &cap, m); | |
11fdf7f2 TL |
5112 | } |
5113 | } else { | |
5114 | ldout(cct, 5) << __func__ << " don't have " << *in << " cap on mds." << mds << dendl; | |
5115 | return; | |
7c673cae FG |
5116 | } |
5117 | } | |
5118 | ||
11fdf7f2 | 5119 | void Client::handle_cap_import(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5120 | { |
5121 | mds_rank_t mds = session->mds_num; | |
5122 | ||
11fdf7f2 | 5123 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5124 | << " IMPORT from mds." << mds << dendl; |
5125 | ||
5126 | const mds_rank_t peer_mds = mds_rank_t(m->peer.mds); | |
5127 | Cap *cap = NULL; | |
5128 | UserPerm cap_perms; | |
11fdf7f2 TL |
5129 | if (auto it = in->caps.find(peer_mds); m->peer.cap_id && it != in->caps.end()) { |
5130 | cap = &it->second; | |
5131 | cap_perms = cap->latest_perms; | |
7c673cae FG |
5132 | } |
5133 | ||
5134 | // add/update it | |
5135 | SnapRealm *realm = NULL; | |
5136 | update_snap_trace(m->snapbl, &realm); | |
5137 | ||
1911f103 TL |
5138 | int issued = m->get_caps(); |
5139 | int wanted = m->get_wanted(); | |
7c673cae | 5140 | add_update_cap(in, session, m->get_cap_id(), |
1911f103 | 5141 | issued, wanted, m->get_seq(), m->get_mseq(), |
a8e16298 | 5142 | m->get_realm(), CEPH_CAP_FLAG_AUTH, cap_perms); |
7c673cae FG |
5143 | |
5144 | if (cap && cap->cap_id == m->peer.cap_id) { | |
5145 | remove_cap(cap, (m->peer.flags & CEPH_CAP_FLAG_RELEASE)); | |
5146 | } | |
5147 | ||
5148 | if (realm) | |
5149 | put_snap_realm(realm); | |
5150 | ||
eafe8130 | 5151 | if (in->auth_cap && in->auth_cap->session == session) { |
1911f103 TL |
5152 | if (!(wanted & CEPH_CAP_ANY_FILE_WR) || |
5153 | in->requested_max_size > m->get_max_size()) { | |
5154 | in->requested_max_size = 0; | |
5155 | ldout(cct, 15) << "reset requested_max_size after cap import" << dendl; | |
5156 | } | |
7c673cae | 5157 | // reflush any/all caps (if we are now the auth_cap) |
eafe8130 | 5158 | kick_flushing_caps(in, session); |
7c673cae FG |
5159 | } |
5160 | } | |
5161 | ||
11fdf7f2 | 5162 | void Client::handle_cap_export(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5163 | { |
5164 | mds_rank_t mds = session->mds_num; | |
5165 | ||
11fdf7f2 | 5166 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5167 | << " EXPORT from mds." << mds << dendl; |
5168 | ||
11fdf7f2 TL |
5169 | auto it = in->caps.find(mds); |
5170 | if (it != in->caps.end()) { | |
5171 | Cap &cap = it->second; | |
5172 | if (cap.cap_id == m->get_cap_id()) { | |
5173 | if (m->peer.cap_id) { | |
5174 | const auto peer_mds = mds_rank_t(m->peer.mds); | |
20effc67 | 5175 | auto tsession = _get_or_open_mds_session(peer_mds); |
11fdf7f2 TL |
5176 | auto it = in->caps.find(peer_mds); |
5177 | if (it != in->caps.end()) { | |
5178 | Cap &tcap = it->second; | |
5179 | if (tcap.cap_id == m->peer.cap_id && | |
5180 | ceph_seq_cmp(tcap.seq, m->peer.seq) < 0) { | |
5181 | tcap.cap_id = m->peer.cap_id; | |
5182 | tcap.seq = m->peer.seq - 1; | |
5183 | tcap.issue_seq = tcap.seq; | |
5184 | tcap.issued |= cap.issued; | |
5185 | tcap.implemented |= cap.issued; | |
5186 | if (&cap == in->auth_cap) | |
5187 | in->auth_cap = &tcap; | |
5188 | if (in->auth_cap == &tcap && in->flushing_cap_item.is_on_list()) | |
20effc67 | 5189 | adjust_session_flushing_caps(in, session, tsession.get()); |
11fdf7f2 TL |
5190 | } |
5191 | } else { | |
20effc67 | 5192 | add_update_cap(in, tsession.get(), m->peer.cap_id, cap.issued, 0, |
11fdf7f2 TL |
5193 | m->peer.seq - 1, m->peer.mseq, (uint64_t)-1, |
5194 | &cap == in->auth_cap ? CEPH_CAP_FLAG_AUTH : 0, | |
5195 | cap.latest_perms); | |
5196 | } | |
7c673cae | 5197 | } else { |
11fdf7f2 TL |
5198 | if (cap.wanted | cap.issued) |
5199 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 5200 | } |
7c673cae | 5201 | |
11fdf7f2 TL |
5202 | remove_cap(&cap, false); |
5203 | } | |
7c673cae | 5204 | } |
7c673cae FG |
5205 | } |
5206 | ||
11fdf7f2 | 5207 | void Client::handle_cap_trunc(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5208 | { |
5209 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 5210 | ceph_assert(in->caps.count(mds)); |
7c673cae | 5211 | |
11fdf7f2 | 5212 | ldout(cct, 10) << __func__ << " on ino " << *in |
7c673cae FG |
5213 | << " size " << in->size << " -> " << m->get_size() |
5214 | << dendl; | |
5215 | ||
1adf2230 AA |
5216 | int issued; |
5217 | in->caps_issued(&issued); | |
5218 | issued |= in->caps_dirty(); | |
5219 | update_inode_file_size(in, issued, m->get_size(), | |
5220 | m->get_truncate_seq(), m->get_truncate_size()); | |
7c673cae FG |
5221 | } |
5222 | ||
11fdf7f2 | 5223 | void Client::handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5224 | { |
5225 | ceph_tid_t flush_ack_tid = m->get_client_tid(); | |
5226 | int dirty = m->get_dirty(); | |
5227 | int cleaned = 0; | |
5228 | int flushed = 0; | |
5229 | ||
11fdf7f2 TL |
5230 | auto it = in->flushing_cap_tids.begin(); |
5231 | if (it->first < flush_ack_tid) { | |
5232 | ldout(cct, 0) << __func__ << " mds." << session->mds_num | |
5233 | << " got unexpected flush ack tid " << flush_ack_tid | |
5234 | << " expected is " << it->first << dendl; | |
5235 | } | |
5236 | for (; it != in->flushing_cap_tids.end(); ) { | |
eafe8130 TL |
5237 | if (!it->second) { |
5238 | // cap snap | |
5239 | ++it; | |
5240 | continue; | |
5241 | } | |
7c673cae FG |
5242 | if (it->first == flush_ack_tid) |
5243 | cleaned = it->second; | |
5244 | if (it->first <= flush_ack_tid) { | |
5245 | session->flushing_caps_tids.erase(it->first); | |
5246 | in->flushing_cap_tids.erase(it++); | |
5247 | ++flushed; | |
5248 | continue; | |
5249 | } | |
5250 | cleaned &= ~it->second; | |
5251 | if (!cleaned) | |
5252 | break; | |
5253 | ++it; | |
5254 | } | |
5255 | ||
11fdf7f2 | 5256 | ldout(cct, 5) << __func__ << " mds." << session->mds_num |
7c673cae FG |
5257 | << " cleaned " << ccap_string(cleaned) << " on " << *in |
5258 | << " with " << ccap_string(dirty) << dendl; | |
5259 | ||
5260 | if (flushed) { | |
5261 | signal_cond_list(in->waitfor_caps); | |
5262 | if (session->flushing_caps_tids.empty() || | |
5263 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5264 | sync_cond.notify_all(); |
7c673cae FG |
5265 | } |
5266 | ||
5267 | if (!dirty) { | |
5268 | in->cap_dirtier_uid = -1; | |
5269 | in->cap_dirtier_gid = -1; | |
5270 | } | |
5271 | ||
5272 | if (!cleaned) { | |
5273 | ldout(cct, 10) << " tid " << m->get_client_tid() << " != any cap bit tids" << dendl; | |
5274 | } else { | |
5275 | if (in->flushing_caps) { | |
5276 | ldout(cct, 5) << " flushing_caps " << ccap_string(in->flushing_caps) | |
5277 | << " -> " << ccap_string(in->flushing_caps & ~cleaned) << dendl; | |
5278 | in->flushing_caps &= ~cleaned; | |
5279 | if (in->flushing_caps == 0) { | |
5280 | ldout(cct, 10) << " " << *in << " !flushing" << dendl; | |
5281 | num_flushing_caps--; | |
eafe8130 | 5282 | if (in->flushing_cap_tids.empty()) |
7c673cae FG |
5283 | in->flushing_cap_item.remove_myself(); |
5284 | } | |
5285 | if (!in->caps_dirty()) | |
5286 | put_inode(in); | |
5287 | } | |
5288 | } | |
7c673cae FG |
5289 | } |
5290 | ||
5291 | ||
11fdf7f2 | 5292 | void Client::handle_cap_flushsnap_ack(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae | 5293 | { |
eafe8130 | 5294 | ceph_tid_t flush_ack_tid = m->get_client_tid(); |
7c673cae | 5295 | mds_rank_t mds = session->mds_num; |
11fdf7f2 | 5296 | ceph_assert(in->caps.count(mds)); |
7c673cae FG |
5297 | snapid_t follows = m->get_snap_follows(); |
5298 | ||
11fdf7f2 TL |
5299 | if (auto it = in->cap_snaps.find(follows); it != in->cap_snaps.end()) { |
5300 | auto& capsnap = it->second; | |
eafe8130 TL |
5301 | if (flush_ack_tid != capsnap.flush_tid) { |
5302 | ldout(cct, 10) << " tid " << flush_ack_tid << " != " << capsnap.flush_tid << dendl; | |
7c673cae | 5303 | } else { |
eafe8130 | 5304 | InodeRef tmp_ref(in); |
11fdf7f2 | 5305 | ldout(cct, 5) << __func__ << " mds." << mds << " flushed snap follows " << follows |
7c673cae | 5306 | << " on " << *in << dendl; |
7c673cae | 5307 | session->flushing_caps_tids.erase(capsnap.flush_tid); |
eafe8130 TL |
5308 | in->flushing_cap_tids.erase(capsnap.flush_tid); |
5309 | if (in->flushing_caps == 0 && in->flushing_cap_tids.empty()) | |
5310 | in->flushing_cap_item.remove_myself(); | |
11fdf7f2 | 5311 | in->cap_snaps.erase(it); |
eafe8130 TL |
5312 | |
5313 | signal_cond_list(in->waitfor_caps); | |
5314 | if (session->flushing_caps_tids.empty() || | |
5315 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5316 | sync_cond.notify_all(); |
7c673cae FG |
5317 | } |
5318 | } else { | |
11fdf7f2 | 5319 | ldout(cct, 5) << __func__ << " DUP(?) mds." << mds << " flushed snap follows " << follows |
7c673cae FG |
5320 | << " on " << *in << dendl; |
5321 | // we may not have it if we send multiple FLUSHSNAP requests and (got multiple FLUSHEDSNAPs back) | |
5322 | } | |
7c673cae FG |
5323 | } |
5324 | ||
5325 | class C_Client_DentryInvalidate : public Context { | |
5326 | private: | |
5327 | Client *client; | |
5328 | vinodeno_t dirino; | |
5329 | vinodeno_t ino; | |
5330 | string name; | |
5331 | public: | |
5332 | C_Client_DentryInvalidate(Client *c, Dentry *dn, bool del) : | |
5333 | client(c), name(dn->name) { | |
5334 | if (client->use_faked_inos()) { | |
5335 | dirino.ino = dn->dir->parent_inode->faked_ino; | |
5336 | if (del) | |
5337 | ino.ino = dn->inode->faked_ino; | |
5338 | } else { | |
5339 | dirino = dn->dir->parent_inode->vino(); | |
5340 | if (del) | |
5341 | ino = dn->inode->vino(); | |
5342 | } | |
5343 | if (!del) | |
5344 | ino.ino = inodeno_t(); | |
5345 | } | |
5346 | void finish(int r) override { | |
5347 | // _async_dentry_invalidate is responsible for its own locking | |
9f95a23c | 5348 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
5349 | client->_async_dentry_invalidate(dirino, ino, name); |
5350 | } | |
5351 | }; | |
5352 | ||
5353 | void Client::_async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name) | |
5354 | { | |
f67539c2 TL |
5355 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
5356 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 5357 | return; |
f67539c2 | 5358 | |
11fdf7f2 | 5359 | ldout(cct, 10) << __func__ << " '" << name << "' ino " << ino |
7c673cae | 5360 | << " in dir " << dirino << dendl; |
e306af50 | 5361 | dentry_invalidate_cb(callback_handle, dirino, ino, name.c_str(), name.length()); |
7c673cae FG |
5362 | } |
5363 | ||
5364 | void Client::_schedule_invalidate_dentry_callback(Dentry *dn, bool del) | |
5365 | { | |
5366 | if (dentry_invalidate_cb && dn->inode->ll_ref > 0) | |
5367 | async_dentry_invalidator.queue(new C_Client_DentryInvalidate(this, dn, del)); | |
5368 | } | |
5369 | ||
5370 | void Client::_try_to_trim_inode(Inode *in, bool sched_inval) | |
5371 | { | |
b3b6e05e | 5372 | int ref = in->get_nref(); |
494da23a | 5373 | ldout(cct, 5) << __func__ << " in " << *in <<dendl; |
7c673cae FG |
5374 | |
5375 | if (in->dir && !in->dir->dentries.empty()) { | |
5376 | for (auto p = in->dir->dentries.begin(); | |
5377 | p != in->dir->dentries.end(); ) { | |
5378 | Dentry *dn = p->second; | |
5379 | ++p; | |
5380 | /* rmsnap removes whole subtree, need trim inodes recursively. | |
5381 | * we don't need to invalidate dentries recursively. because | |
5382 | * invalidating a directory dentry effectively invalidate | |
5383 | * whole subtree */ | |
5384 | if (in->snapid != CEPH_NOSNAP && dn->inode && dn->inode->is_dir()) | |
5385 | _try_to_trim_inode(dn->inode.get(), false); | |
5386 | ||
5387 | if (dn->lru_is_expireable()) | |
5388 | unlink(dn, true, false); // keep dir, drop dentry | |
5389 | } | |
5390 | if (in->dir->dentries.empty()) { | |
5391 | close_dir(in->dir); | |
5392 | --ref; | |
5393 | } | |
5394 | } | |
5395 | ||
b3b6e05e | 5396 | if (ref > 1 && (in->flags & I_SNAPDIR_OPEN)) { |
7c673cae FG |
5397 | InodeRef snapdir = open_snapdir(in); |
5398 | _try_to_trim_inode(snapdir.get(), false); | |
5399 | --ref; | |
5400 | } | |
5401 | ||
b3b6e05e | 5402 | if (ref > 1) { |
11fdf7f2 TL |
5403 | auto q = in->dentries.begin(); |
5404 | while (q != in->dentries.end()) { | |
5405 | Dentry *dn = *q; | |
5406 | ++q; | |
494da23a TL |
5407 | if( in->ll_ref > 0 && sched_inval) { |
5408 | // FIXME: we play lots of unlink/link tricks when handling MDS replies, | |
5409 | // so in->dentries doesn't always reflect the state of kernel's dcache. | |
5410 | _schedule_invalidate_dentry_callback(dn, true); | |
5411 | } | |
7c673cae FG |
5412 | unlink(dn, true, true); |
5413 | } | |
5414 | } | |
5415 | } | |
5416 | ||
11fdf7f2 | 5417 | void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5418 | { |
5419 | mds_rank_t mds = session->mds_num; | |
5420 | int used = get_caps_used(in); | |
5421 | int wanted = in->caps_wanted(); | |
a4b75251 | 5422 | int flags = 0; |
7c673cae | 5423 | |
a8e16298 TL |
5424 | const unsigned new_caps = m->get_caps(); |
5425 | const bool was_stale = session->cap_gen > cap->gen; | |
11fdf7f2 | 5426 | ldout(cct, 5) << __func__ << " on in " << m->get_ino() |
7c673cae FG |
5427 | << " mds." << mds << " seq " << m->get_seq() |
5428 | << " caps now " << ccap_string(new_caps) | |
a8e16298 | 5429 | << " was " << ccap_string(cap->issued) |
92f5a8d4 | 5430 | << (was_stale ? " (stale)" : "") << dendl; |
a8e16298 TL |
5431 | |
5432 | if (was_stale) | |
5433 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
7c673cae | 5434 | cap->seq = m->get_seq(); |
28e407b8 | 5435 | cap->gen = session->cap_gen; |
7c673cae | 5436 | |
11fdf7f2 | 5437 | check_cap_issue(in, new_caps); |
a8e16298 | 5438 | |
7c673cae | 5439 | // update inode |
1adf2230 AA |
5440 | int issued; |
5441 | in->caps_issued(&issued); | |
5442 | issued |= in->caps_dirty(); | |
7c673cae | 5443 | |
1adf2230 AA |
5444 | if ((new_caps & CEPH_CAP_AUTH_SHARED) && |
5445 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
5446 | in->mode = m->head.mode; |
5447 | in->uid = m->head.uid; | |
5448 | in->gid = m->head.gid; | |
5449 | in->btime = m->btime; | |
5450 | } | |
5451 | bool deleted_inode = false; | |
1adf2230 AA |
5452 | if ((new_caps & CEPH_CAP_LINK_SHARED) && |
5453 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
7c673cae | 5454 | in->nlink = m->head.nlink; |
20effc67 | 5455 | if (in->nlink == 0) |
7c673cae FG |
5456 | deleted_inode = true; |
5457 | } | |
1adf2230 | 5458 | if (!(issued & CEPH_CAP_XATTR_EXCL) && |
7c673cae FG |
5459 | m->xattrbl.length() && |
5460 | m->head.xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
5461 | auto p = m->xattrbl.cbegin(); |
5462 | decode(in->xattrs, p); | |
7c673cae FG |
5463 | in->xattr_version = m->head.xattr_version; |
5464 | } | |
28e407b8 AA |
5465 | |
5466 | if ((new_caps & CEPH_CAP_FILE_SHARED) && m->dirstat_is_valid()) { | |
5467 | in->dirstat.nfiles = m->get_nfiles(); | |
5468 | in->dirstat.nsubdirs = m->get_nsubdirs(); | |
5469 | } | |
5470 | ||
1adf2230 AA |
5471 | if (new_caps & CEPH_CAP_ANY_RD) { |
5472 | update_inode_file_time(in, issued, m->get_time_warp_seq(), | |
5473 | m->get_ctime(), m->get_mtime(), m->get_atime()); | |
5474 | } | |
5475 | ||
5476 | if (new_caps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { | |
5477 | in->layout = m->get_layout(); | |
5478 | update_inode_file_size(in, issued, m->get_size(), | |
5479 | m->get_truncate_seq(), m->get_truncate_size()); | |
5480 | } | |
5481 | ||
5482 | if (m->inline_version > in->inline_version) { | |
5483 | in->inline_data = m->inline_data; | |
5484 | in->inline_version = m->inline_version; | |
5485 | } | |
5486 | ||
5487 | /* always take a newer change attr */ | |
5488 | if (m->get_change_attr() > in->change_attr) | |
5489 | in->change_attr = m->get_change_attr(); | |
7c673cae FG |
5490 | |
5491 | // max_size | |
5492 | if (cap == in->auth_cap && | |
1adf2230 AA |
5493 | (new_caps & CEPH_CAP_ANY_FILE_WR) && |
5494 | (m->get_max_size() != in->max_size)) { | |
7c673cae FG |
5495 | ldout(cct, 10) << "max_size " << in->max_size << " -> " << m->get_max_size() << dendl; |
5496 | in->max_size = m->get_max_size(); | |
5497 | if (in->max_size > in->wanted_max_size) { | |
5498 | in->wanted_max_size = 0; | |
5499 | in->requested_max_size = 0; | |
5500 | } | |
5501 | } | |
5502 | ||
5503 | bool check = false; | |
a8e16298 TL |
5504 | if ((was_stale || m->get_op() == CEPH_CAP_OP_IMPORT) && |
5505 | (wanted & ~(cap->wanted | new_caps))) { | |
5506 | // If mds is importing cap, prior cap messages that update 'wanted' | |
5507 | // may get dropped by mds (migrate seq mismatch). | |
5508 | // | |
5509 | // We don't send cap message to update 'wanted' if what we want are | |
5510 | // already issued. If mds revokes caps, cap message that releases caps | |
5511 | // also tells mds what we want. But if caps got revoked by mds forcedly | |
5512 | // (session stale). We may haven't told mds what we want. | |
7c673cae | 5513 | check = true; |
a8e16298 | 5514 | } |
7c673cae | 5515 | |
7c673cae FG |
5516 | |
5517 | // update caps | |
a8e16298 | 5518 | auto revoked = cap->issued & ~new_caps; |
b32b8144 FG |
5519 | if (revoked) { |
5520 | ldout(cct, 10) << " revocation of " << ccap_string(revoked) << dendl; | |
7c673cae FG |
5521 | cap->issued = new_caps; |
5522 | cap->implemented |= new_caps; | |
5523 | ||
b32b8144 FG |
5524 | // recall delegations if we're losing caps necessary for them |
5525 | if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_RD)) | |
5526 | in->recall_deleg(false); | |
5527 | else if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_WR)) | |
5528 | in->recall_deleg(true); | |
5529 | ||
11fdf7f2 TL |
5530 | used = adjust_caps_used_for_lazyio(used, cap->issued, cap->implemented); |
5531 | if ((used & revoked & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO)) && | |
28e407b8 | 5532 | !_flush(in, new C_Client_FlushComplete(this, in))) { |
7c673cae | 5533 | // waitin' for flush |
11fdf7f2 | 5534 | } else if (used & revoked & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) { |
a4b75251 TL |
5535 | if (_release(in)) { |
5536 | check = true; | |
5537 | flags = CHECK_CAPS_NODELAY; | |
5538 | } | |
7c673cae FG |
5539 | } else { |
5540 | cap->wanted = 0; // don't let check_caps skip sending a response to MDS | |
5541 | check = true; | |
a4b75251 | 5542 | flags = CHECK_CAPS_NODELAY; |
7c673cae | 5543 | } |
a8e16298 TL |
5544 | } else if (cap->issued == new_caps) { |
5545 | ldout(cct, 10) << " caps unchanged at " << ccap_string(cap->issued) << dendl; | |
7c673cae | 5546 | } else { |
a8e16298 | 5547 | ldout(cct, 10) << " grant, new caps are " << ccap_string(new_caps & ~cap->issued) << dendl; |
7c673cae FG |
5548 | cap->issued = new_caps; |
5549 | cap->implemented |= new_caps; | |
5550 | ||
5551 | if (cap == in->auth_cap) { | |
5552 | // non-auth MDS is revoking the newly grant caps ? | |
11fdf7f2 TL |
5553 | for (const auto &p : in->caps) { |
5554 | if (&p.second == cap) | |
7c673cae | 5555 | continue; |
11fdf7f2 | 5556 | if (p.second.implemented & ~p.second.issued & new_caps) { |
7c673cae FG |
5557 | check = true; |
5558 | break; | |
5559 | } | |
5560 | } | |
5561 | } | |
5562 | } | |
5563 | ||
5564 | if (check) | |
a4b75251 | 5565 | check_caps(in, flags); |
7c673cae FG |
5566 | |
5567 | // wake up waiters | |
5568 | if (new_caps) | |
5569 | signal_cond_list(in->waitfor_caps); | |
5570 | ||
5571 | // may drop inode's last ref | |
5572 | if (deleted_inode) | |
5573 | _try_to_trim_inode(in, true); | |
7c673cae FG |
5574 | } |
5575 | ||
7c673cae FG |
5576 | int Client::inode_permission(Inode *in, const UserPerm& perms, unsigned want) |
5577 | { | |
b3b6e05e TL |
5578 | if (perms.uid() == 0) { |
5579 | // Executable are overridable when there is at least one exec bit set | |
5580 | if((want & MAY_EXEC) && !(in->mode & S_IXUGO)) | |
5581 | return -CEPHFS_EACCES; | |
7c673cae | 5582 | return 0; |
b3b6e05e | 5583 | } |
7c673cae FG |
5584 | |
5585 | if (perms.uid() != in->uid && (in->mode & S_IRWXG)) { | |
5586 | int ret = _posix_acl_permission(in, perms, want); | |
f67539c2 | 5587 | if (ret != -CEPHFS_EAGAIN) |
7c673cae FG |
5588 | return ret; |
5589 | } | |
5590 | ||
5591 | // check permissions before doing anything else | |
5592 | if (!in->check_mode(perms, want)) | |
f67539c2 | 5593 | return -CEPHFS_EACCES; |
7c673cae FG |
5594 | return 0; |
5595 | } | |
5596 | ||
5597 | int Client::xattr_permission(Inode *in, const char *name, unsigned want, | |
5598 | const UserPerm& perms) | |
5599 | { | |
5600 | int r = _getattr_for_perm(in, perms); | |
5601 | if (r < 0) | |
5602 | goto out; | |
5603 | ||
5604 | r = 0; | |
5605 | if (strncmp(name, "system.", 7) == 0) { | |
5606 | if ((want & MAY_WRITE) && (perms.uid() != 0 && perms.uid() != in->uid)) | |
f67539c2 | 5607 | r = -CEPHFS_EPERM; |
7c673cae FG |
5608 | } else { |
5609 | r = inode_permission(in, perms, want); | |
5610 | } | |
5611 | out: | |
1adf2230 | 5612 | ldout(cct, 5) << __func__ << " " << in << " = " << r << dendl; |
7c673cae FG |
5613 | return r; |
5614 | } | |
5615 | ||
20effc67 | 5616 | std::ostream& operator<<(std::ostream &out, const UserPerm& perm) { |
7c673cae FG |
5617 | out << "UserPerm(uid: " << perm.uid() << ", gid: " << perm.gid() << ")"; |
5618 | return out; | |
5619 | } | |
5620 | ||
5621 | int Client::may_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
5622 | const UserPerm& perms) | |
5623 | { | |
181888fb | 5624 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5625 | int r = _getattr_for_perm(in, perms); |
5626 | if (r < 0) | |
5627 | goto out; | |
5628 | ||
5629 | if (mask & CEPH_SETATTR_SIZE) { | |
5630 | r = inode_permission(in, perms, MAY_WRITE); | |
5631 | if (r < 0) | |
5632 | goto out; | |
5633 | } | |
5634 | ||
f67539c2 | 5635 | r = -CEPHFS_EPERM; |
7c673cae FG |
5636 | if (mask & CEPH_SETATTR_UID) { |
5637 | if (perms.uid() != 0 && (perms.uid() != in->uid || stx->stx_uid != in->uid)) | |
5638 | goto out; | |
5639 | } | |
5640 | if (mask & CEPH_SETATTR_GID) { | |
5641 | if (perms.uid() != 0 && (perms.uid() != in->uid || | |
5642 | (!perms.gid_in_groups(stx->stx_gid) && stx->stx_gid != in->gid))) | |
5643 | goto out; | |
5644 | } | |
5645 | ||
5646 | if (mask & CEPH_SETATTR_MODE) { | |
5647 | if (perms.uid() != 0 && perms.uid() != in->uid) | |
5648 | goto out; | |
5649 | ||
5650 | gid_t i_gid = (mask & CEPH_SETATTR_GID) ? stx->stx_gid : in->gid; | |
5651 | if (perms.uid() != 0 && !perms.gid_in_groups(i_gid)) | |
5652 | stx->stx_mode &= ~S_ISGID; | |
5653 | } | |
5654 | ||
5655 | if (mask & (CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME | | |
5656 | CEPH_SETATTR_MTIME | CEPH_SETATTR_ATIME)) { | |
5657 | if (perms.uid() != 0 && perms.uid() != in->uid) { | |
5658 | int check_mask = CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME; | |
5659 | if (!(mask & CEPH_SETATTR_MTIME_NOW)) | |
5660 | check_mask |= CEPH_SETATTR_MTIME; | |
5661 | if (!(mask & CEPH_SETATTR_ATIME_NOW)) | |
5662 | check_mask |= CEPH_SETATTR_ATIME; | |
5663 | if (check_mask & mask) { | |
5664 | goto out; | |
5665 | } else { | |
5666 | r = inode_permission(in, perms, MAY_WRITE); | |
5667 | if (r < 0) | |
5668 | goto out; | |
5669 | } | |
5670 | } | |
5671 | } | |
5672 | r = 0; | |
5673 | out: | |
5674 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5675 | return r; | |
5676 | } | |
5677 | ||
5678 | int Client::may_open(Inode *in, int flags, const UserPerm& perms) | |
5679 | { | |
181888fb | 5680 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5681 | unsigned want = 0; |
5682 | ||
5683 | if ((flags & O_ACCMODE) == O_WRONLY) | |
5684 | want = MAY_WRITE; | |
5685 | else if ((flags & O_ACCMODE) == O_RDWR) | |
5686 | want = MAY_READ | MAY_WRITE; | |
5687 | else if ((flags & O_ACCMODE) == O_RDONLY) | |
5688 | want = MAY_READ; | |
5689 | if (flags & O_TRUNC) | |
5690 | want |= MAY_WRITE; | |
5691 | ||
5692 | int r = 0; | |
5693 | switch (in->mode & S_IFMT) { | |
5694 | case S_IFLNK: | |
f67539c2 | 5695 | r = -CEPHFS_ELOOP; |
7c673cae FG |
5696 | goto out; |
5697 | case S_IFDIR: | |
5698 | if (want & MAY_WRITE) { | |
f67539c2 | 5699 | r = -CEPHFS_EISDIR; |
7c673cae FG |
5700 | goto out; |
5701 | } | |
5702 | break; | |
5703 | } | |
5704 | ||
5705 | r = _getattr_for_perm(in, perms); | |
5706 | if (r < 0) | |
5707 | goto out; | |
5708 | ||
5709 | r = inode_permission(in, perms, want); | |
5710 | out: | |
5711 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5712 | return r; | |
5713 | } | |
5714 | ||
5715 | int Client::may_lookup(Inode *dir, const UserPerm& perms) | |
5716 | { | |
181888fb | 5717 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5718 | int r = _getattr_for_perm(dir, perms); |
5719 | if (r < 0) | |
5720 | goto out; | |
5721 | ||
5722 | r = inode_permission(dir, perms, MAY_EXEC); | |
5723 | out: | |
5724 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5725 | return r; | |
5726 | } | |
5727 | ||
5728 | int Client::may_create(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 | MAY_WRITE); | |
5736 | out: | |
5737 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5738 | return r; | |
5739 | } | |
5740 | ||
5741 | int Client::may_delete(Inode *dir, const char *name, const UserPerm& perms) | |
5742 | { | |
181888fb | 5743 | ldout(cct, 20) << __func__ << " " << *dir << "; " << "; name " << name << "; " << 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 | if (r < 0) | |
5750 | goto out; | |
5751 | ||
f67539c2 | 5752 | /* 'name == NULL' means rmsnap w/o permission checks */ |
7c673cae FG |
5753 | if (perms.uid() != 0 && name && (dir->mode & S_ISVTX)) { |
5754 | InodeRef otherin; | |
5755 | r = _lookup(dir, name, CEPH_CAP_AUTH_SHARED, &otherin, perms); | |
5756 | if (r < 0) | |
5757 | goto out; | |
5758 | if (dir->uid != perms.uid() && otherin->uid != perms.uid()) | |
f67539c2 | 5759 | r = -CEPHFS_EPERM; |
7c673cae FG |
5760 | } |
5761 | out: | |
5762 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5763 | return r; | |
5764 | } | |
5765 | ||
f67539c2 TL |
5766 | int Client::may_delete(const char *relpath, const UserPerm& perms) { |
5767 | ldout(cct, 20) << __func__ << " " << relpath << "; " << perms << dendl; | |
5768 | ||
5769 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
5770 | if (!mref_reader.is_state_satisfied()) | |
5771 | return -ENOTCONN; | |
5772 | ||
5773 | filepath path(relpath); | |
5774 | string name = path.last_dentry(); | |
5775 | path.pop_dentry(); | |
5776 | InodeRef dir; | |
5777 | ||
5778 | std::scoped_lock lock(client_lock); | |
5779 | int r = path_walk(path, &dir, perms); | |
5780 | if (r < 0) | |
5781 | return r; | |
5782 | if (cct->_conf->client_permissions) { | |
5783 | int r = may_delete(dir.get(), name.c_str(), perms); | |
5784 | if (r < 0) | |
5785 | return r; | |
5786 | } | |
5787 | ||
5788 | return 0; | |
5789 | } | |
5790 | ||
7c673cae FG |
5791 | int Client::may_hardlink(Inode *in, const UserPerm& perms) |
5792 | { | |
181888fb | 5793 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5794 | int r = _getattr_for_perm(in, perms); |
5795 | if (r < 0) | |
5796 | goto out; | |
5797 | ||
5798 | if (perms.uid() == 0 || perms.uid() == in->uid) { | |
5799 | r = 0; | |
5800 | goto out; | |
5801 | } | |
5802 | ||
f67539c2 | 5803 | r = -CEPHFS_EPERM; |
7c673cae FG |
5804 | if (!S_ISREG(in->mode)) |
5805 | goto out; | |
5806 | ||
5807 | if (in->mode & S_ISUID) | |
5808 | goto out; | |
5809 | ||
5810 | if ((in->mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
5811 | goto out; | |
5812 | ||
5813 | r = inode_permission(in, perms, MAY_READ | MAY_WRITE); | |
5814 | out: | |
5815 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5816 | return r; | |
5817 | } | |
5818 | ||
5819 | int Client::_getattr_for_perm(Inode *in, const UserPerm& perms) | |
5820 | { | |
5821 | int mask = CEPH_STAT_CAP_MODE; | |
5822 | bool force = false; | |
5823 | if (acl_type != NO_ACL) { | |
5824 | mask |= CEPH_STAT_CAP_XATTR; | |
5825 | force = in->xattr_version == 0; | |
5826 | } | |
5827 | return _getattr(in, mask, perms, force); | |
5828 | } | |
5829 | ||
5830 | vinodeno_t Client::_get_vino(Inode *in) | |
5831 | { | |
5832 | /* The caller must hold the client lock */ | |
5833 | return vinodeno_t(in->ino, in->snapid); | |
5834 | } | |
5835 | ||
7c673cae FG |
5836 | /** |
5837 | * Resolve an MDS spec to a list of MDS daemon GIDs. | |
5838 | * | |
5839 | * The spec is a string representing a GID, rank, filesystem:rank, or name/id. | |
5840 | * It may be '*' in which case it matches all GIDs. | |
5841 | * | |
5842 | * If no error is returned, the `targets` vector will be populated with at least | |
5843 | * one MDS. | |
5844 | */ | |
5845 | int Client::resolve_mds( | |
5846 | const std::string &mds_spec, | |
5847 | std::vector<mds_gid_t> *targets) | |
5848 | { | |
11fdf7f2 TL |
5849 | ceph_assert(fsmap); |
5850 | ceph_assert(targets != nullptr); | |
7c673cae FG |
5851 | |
5852 | mds_role_t role; | |
f67539c2 TL |
5853 | CachedStackStringStream css; |
5854 | int role_r = fsmap->parse_role(mds_spec, &role, *css); | |
7c673cae FG |
5855 | if (role_r == 0) { |
5856 | // We got a role, resolve it to a GID | |
f67539c2 TL |
5857 | auto& info = fsmap->get_filesystem(role.fscid)->mds_map.get_info(role.rank); |
5858 | ldout(cct, 10) << __func__ << ": resolved " << mds_spec << " to role '" | |
5859 | << role << "' aka " << info.human_name() << dendl; | |
5860 | targets->push_back(info.global_id); | |
7c673cae FG |
5861 | return 0; |
5862 | } | |
5863 | ||
5864 | std::string strtol_err; | |
5865 | long long rank_or_gid = strict_strtoll(mds_spec.c_str(), 10, &strtol_err); | |
5866 | if (strtol_err.empty()) { | |
5867 | // It is a possible GID | |
5868 | const mds_gid_t mds_gid = mds_gid_t(rank_or_gid); | |
5869 | if (fsmap->gid_exists(mds_gid)) { | |
f67539c2 TL |
5870 | auto& info = fsmap->get_info_gid(mds_gid); |
5871 | ldout(cct, 10) << __func__ << ": validated gid " << mds_gid << " aka " | |
5872 | << info.human_name() << dendl; | |
7c673cae | 5873 | targets->push_back(mds_gid); |
f67539c2 | 5874 | return 0; |
7c673cae | 5875 | } else { |
f67539c2 | 5876 | lderr(cct) << __func__ << ": gid " << mds_gid << " not in MDS map" |
7c673cae | 5877 | << dendl; |
f67539c2 TL |
5878 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
5879 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5880 | } |
5881 | } else if (mds_spec == "*") { | |
5882 | // It is a wildcard: use all MDSs | |
f67539c2 | 5883 | const auto& mds_info = fsmap->get_mds_info(); |
7c673cae | 5884 | |
f67539c2 | 5885 | ldout(cct, 10) << __func__ << ": resolving `*' to all MDS daemons" << dendl; |
7c673cae | 5886 | if (mds_info.empty()) { |
f67539c2 TL |
5887 | lderr(cct) << __func__ << ": no MDS daemons found" << dendl; |
5888 | lderr(cct) << "FSMap: " << *fsmap << dendl; | |
5889 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5890 | } |
5891 | ||
f67539c2 TL |
5892 | for (const auto& [gid, info] : mds_info) { |
5893 | ldout(cct, 10) << __func__ << ": appending " << info.human_name() << " to targets" << dendl; | |
5894 | targets->push_back(gid); | |
7c673cae | 5895 | } |
f67539c2 | 5896 | return 0; |
7c673cae FG |
5897 | } else { |
5898 | // It did not parse as an integer, it is not a wildcard, it must be a name | |
5899 | const mds_gid_t mds_gid = fsmap->find_mds_gid_by_name(mds_spec); | |
5900 | if (mds_gid == 0) { | |
f67539c2 | 5901 | lderr(cct) << __func__ << ": no MDS daemons found by name `" << mds_spec << "'" << dendl; |
7c673cae | 5902 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
f67539c2 | 5903 | return -CEPHFS_ENOENT; |
7c673cae | 5904 | } else { |
f67539c2 TL |
5905 | auto& info = fsmap->get_info_gid(mds_gid); |
5906 | ldout(cct, 10) << __func__ << ": resolved name '" << mds_spec | |
5907 | << "' to " << info.human_name() << dendl; | |
7c673cae FG |
5908 | targets->push_back(mds_gid); |
5909 | } | |
f67539c2 | 5910 | return 0; |
7c673cae | 5911 | } |
7c673cae FG |
5912 | } |
5913 | ||
5914 | ||
5915 | /** | |
5916 | * Authenticate with mon and establish global ID | |
5917 | */ | |
5918 | int Client::authenticate() | |
5919 | { | |
9f95a23c | 5920 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
5921 | |
5922 | if (monclient->is_authenticated()) { | |
5923 | return 0; | |
5924 | } | |
5925 | ||
9f95a23c | 5926 | client_lock.unlock(); |
7c673cae | 5927 | int r = monclient->authenticate(cct->_conf->client_mount_timeout); |
9f95a23c | 5928 | client_lock.lock(); |
7c673cae FG |
5929 | if (r < 0) { |
5930 | return r; | |
5931 | } | |
5932 | ||
5933 | whoami = monclient->get_global_id(); | |
5934 | messenger->set_myname(entity_name_t::CLIENT(whoami.v)); | |
5935 | ||
5936 | return 0; | |
5937 | } | |
5938 | ||
5939 | int Client::fetch_fsmap(bool user) | |
5940 | { | |
f67539c2 TL |
5941 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
5942 | ||
7c673cae FG |
5943 | // Retrieve FSMap to enable looking up daemon addresses. We need FSMap |
5944 | // rather than MDSMap because no one MDSMap contains all the daemons, and | |
5945 | // a `tell` can address any daemon. | |
5946 | version_t fsmap_latest; | |
f67539c2 | 5947 | bs::error_code ec; |
7c673cae | 5948 | do { |
9f95a23c | 5949 | client_lock.unlock(); |
f67539c2 TL |
5950 | std::tie(fsmap_latest, std::ignore) = |
5951 | monclient->get_version("fsmap", ca::use_blocked[ec]); | |
9f95a23c | 5952 | client_lock.lock(); |
f67539c2 | 5953 | } while (ec == bs::errc::resource_unavailable_try_again); |
7c673cae | 5954 | |
f67539c2 TL |
5955 | if (ec) { |
5956 | lderr(cct) << "Failed to learn FSMap version: " << ec << dendl; | |
5957 | return ceph::from_error_code(ec); | |
7c673cae FG |
5958 | } |
5959 | ||
5960 | ldout(cct, 10) << __func__ << " learned FSMap version " << fsmap_latest << dendl; | |
5961 | ||
5962 | if (user) { | |
5963 | if (!fsmap_user || fsmap_user->get_epoch() < fsmap_latest) { | |
5964 | monclient->sub_want("fsmap.user", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5965 | monclient->renew_subs(); | |
5966 | wait_on_list(waiting_for_fsmap); | |
5967 | } | |
11fdf7f2 TL |
5968 | ceph_assert(fsmap_user); |
5969 | ceph_assert(fsmap_user->get_epoch() >= fsmap_latest); | |
7c673cae FG |
5970 | } else { |
5971 | if (!fsmap || fsmap->get_epoch() < fsmap_latest) { | |
5972 | monclient->sub_want("fsmap", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5973 | monclient->renew_subs(); | |
5974 | wait_on_list(waiting_for_fsmap); | |
5975 | } | |
11fdf7f2 TL |
5976 | ceph_assert(fsmap); |
5977 | ceph_assert(fsmap->get_epoch() >= fsmap_latest); | |
7c673cae FG |
5978 | } |
5979 | ldout(cct, 10) << __func__ << " finished waiting for FSMap version " | |
5980 | << fsmap_latest << dendl; | |
5981 | return 0; | |
5982 | } | |
5983 | ||
5984 | /** | |
5985 | * | |
5986 | * @mds_spec one of ID, rank, GID, "*" | |
5987 | * | |
5988 | */ | |
5989 | int Client::mds_command( | |
5990 | const std::string &mds_spec, | |
5991 | const vector<string>& cmd, | |
5992 | const bufferlist& inbl, | |
5993 | bufferlist *outbl, | |
5994 | string *outs, | |
5995 | Context *onfinish) | |
5996 | { | |
f67539c2 TL |
5997 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
5998 | if (!iref_reader.is_state_satisfied()) | |
5999 | return -CEPHFS_ENOTCONN; | |
7c673cae | 6000 | |
f67539c2 | 6001 | std::unique_lock cl(client_lock); |
7c673cae FG |
6002 | |
6003 | int r; | |
6004 | r = authenticate(); | |
6005 | if (r < 0) { | |
6006 | return r; | |
6007 | } | |
6008 | ||
6009 | r = fetch_fsmap(false); | |
6010 | if (r < 0) { | |
6011 | return r; | |
6012 | } | |
6013 | ||
6014 | // Look up MDS target(s) of the command | |
6015 | std::vector<mds_gid_t> targets; | |
6016 | r = resolve_mds(mds_spec, &targets); | |
6017 | if (r < 0) { | |
6018 | return r; | |
6019 | } | |
6020 | ||
6021 | // If daemons are laggy, we won't send them commands. If all | |
6022 | // are laggy then we fail. | |
6023 | std::vector<mds_gid_t> non_laggy; | |
f67539c2 | 6024 | for (const auto& gid : targets) { |
7c673cae FG |
6025 | const auto info = fsmap->get_info_gid(gid); |
6026 | if (!info.laggy()) { | |
6027 | non_laggy.push_back(gid); | |
6028 | } | |
6029 | } | |
6030 | if (non_laggy.size() == 0) { | |
6031 | *outs = "All targeted MDS daemons are laggy"; | |
f67539c2 | 6032 | return -CEPHFS_ENOENT; |
7c673cae FG |
6033 | } |
6034 | ||
6035 | if (metadata.empty()) { | |
6036 | // We are called on an unmounted client, so metadata | |
6037 | // won't be initialized yet. | |
6038 | populate_metadata(""); | |
6039 | } | |
6040 | ||
6041 | // Send commands to targets | |
6042 | C_GatherBuilder gather(cct, onfinish); | |
f67539c2 | 6043 | for (const auto& target_gid : non_laggy) { |
7c673cae FG |
6044 | const auto info = fsmap->get_info_gid(target_gid); |
6045 | ||
6046 | // Open a connection to the target MDS | |
11fdf7f2 | 6047 | ConnectionRef conn = messenger->connect_to_mds(info.get_addrs()); |
7c673cae | 6048 | |
f67539c2 TL |
6049 | cl.unlock(); |
6050 | { | |
6051 | std::scoped_lock cmd_lock(command_lock); | |
6052 | // Generate MDSCommandOp state | |
6053 | auto &op = command_table.start_command(); | |
7c673cae | 6054 | |
f67539c2 TL |
6055 | op.on_finish = gather.new_sub(); |
6056 | op.cmd = cmd; | |
6057 | op.outbl = outbl; | |
6058 | op.outs = outs; | |
6059 | op.inbl = inbl; | |
6060 | op.mds_gid = target_gid; | |
6061 | op.con = conn; | |
7c673cae | 6062 | |
f67539c2 TL |
6063 | ldout(cct, 4) << __func__ << ": new command op to " << target_gid |
6064 | << " tid=" << op.tid << cmd << dendl; | |
7c673cae | 6065 | |
f67539c2 TL |
6066 | // Construct and send MCommand |
6067 | MessageRef m = op.get_message(monclient->get_fsid()); | |
6068 | conn->send_message2(std::move(m)); | |
6069 | } | |
6070 | cl.lock(); | |
7c673cae FG |
6071 | } |
6072 | gather.activate(); | |
6073 | ||
6074 | return 0; | |
6075 | } | |
6076 | ||
11fdf7f2 | 6077 | void Client::handle_command_reply(const MConstRef<MCommandReply>& m) |
7c673cae FG |
6078 | { |
6079 | ceph_tid_t const tid = m->get_tid(); | |
6080 | ||
6081 | ldout(cct, 10) << __func__ << ": tid=" << m->get_tid() << dendl; | |
6082 | ||
f67539c2 | 6083 | std::scoped_lock cmd_lock(command_lock); |
7c673cae FG |
6084 | if (!command_table.exists(tid)) { |
6085 | ldout(cct, 1) << __func__ << ": unknown tid " << tid << ", dropping" << dendl; | |
7c673cae FG |
6086 | return; |
6087 | } | |
6088 | ||
6089 | auto &op = command_table.get_command(tid); | |
6090 | if (op.outbl) { | |
11fdf7f2 | 6091 | *op.outbl = m->get_data(); |
7c673cae FG |
6092 | } |
6093 | if (op.outs) { | |
6094 | *op.outs = m->rs; | |
6095 | } | |
6096 | ||
6097 | if (op.on_finish) { | |
6098 | op.on_finish->complete(m->r); | |
6099 | } | |
6100 | ||
6101 | command_table.erase(tid); | |
7c673cae FG |
6102 | } |
6103 | ||
6104 | // ------------------- | |
6105 | // MOUNT | |
6106 | ||
11fdf7f2 | 6107 | int Client::subscribe_mdsmap(const std::string &fs_name) |
7c673cae | 6108 | { |
7c673cae FG |
6109 | int r = authenticate(); |
6110 | if (r < 0) { | |
6111 | lderr(cct) << "authentication failed: " << cpp_strerror(r) << dendl; | |
6112 | return r; | |
6113 | } | |
6114 | ||
11fdf7f2 TL |
6115 | std::string resolved_fs_name; |
6116 | if (fs_name.empty()) { | |
9f95a23c TL |
6117 | resolved_fs_name = cct->_conf.get_val<std::string>("client_fs"); |
6118 | if (resolved_fs_name.empty()) | |
6119 | // Try the backwards compatibility fs name option | |
6120 | resolved_fs_name = cct->_conf.get_val<std::string>("client_mds_namespace"); | |
11fdf7f2 TL |
6121 | } else { |
6122 | resolved_fs_name = fs_name; | |
6123 | } | |
6124 | ||
7c673cae | 6125 | std::string want = "mdsmap"; |
11fdf7f2 | 6126 | if (!resolved_fs_name.empty()) { |
7c673cae FG |
6127 | r = fetch_fsmap(true); |
6128 | if (r < 0) | |
6129 | return r; | |
11fdf7f2 TL |
6130 | fscid = fsmap_user->get_fs_cid(resolved_fs_name); |
6131 | if (fscid == FS_CLUSTER_ID_NONE) { | |
f67539c2 | 6132 | return -CEPHFS_ENOENT; |
11fdf7f2 | 6133 | } |
7c673cae FG |
6134 | |
6135 | std::ostringstream oss; | |
11fdf7f2 | 6136 | oss << want << "." << fscid; |
7c673cae FG |
6137 | want = oss.str(); |
6138 | } | |
6139 | ldout(cct, 10) << "Subscribing to map '" << want << "'" << dendl; | |
6140 | ||
6141 | monclient->sub_want(want, 0, 0); | |
6142 | monclient->renew_subs(); | |
6143 | ||
11fdf7f2 TL |
6144 | return 0; |
6145 | } | |
6146 | ||
6147 | int Client::mount(const std::string &mount_root, const UserPerm& perms, | |
6148 | bool require_mds, const std::string &fs_name) | |
6149 | { | |
f67539c2 | 6150 | ceph_assert(is_initialized()); |
11fdf7f2 | 6151 | |
f67539c2 TL |
6152 | /* |
6153 | * To make sure that the _unmount() must wait until the mount() | |
6154 | * is done. | |
6155 | */ | |
6156 | RWRef_t mref_writer(mount_state, CLIENT_MOUNTING, false); | |
6157 | if (!mref_writer.is_first_writer()) // already mounting or mounted | |
11fdf7f2 | 6158 | return 0; |
11fdf7f2 | 6159 | |
f67539c2 | 6160 | std::unique_lock cl(client_lock); |
11fdf7f2 TL |
6161 | |
6162 | int r = subscribe_mdsmap(fs_name); | |
6163 | if (r < 0) { | |
6164 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
6165 | return r; | |
6166 | } | |
6167 | ||
f67539c2 TL |
6168 | start_tick_thread(); // start tick thread |
6169 | ||
7c673cae FG |
6170 | if (require_mds) { |
6171 | while (1) { | |
6172 | auto availability = mdsmap->is_cluster_available(); | |
6173 | if (availability == MDSMap::STUCK_UNAVAILABLE) { | |
6174 | // Error out | |
6175 | ldout(cct, 10) << "mds cluster unavailable: epoch=" << mdsmap->get_epoch() << dendl; | |
6176 | return CEPH_FUSE_NO_MDS_UP; | |
6177 | } else if (availability == MDSMap::AVAILABLE) { | |
6178 | // Continue to mount | |
6179 | break; | |
6180 | } else if (availability == MDSMap::TRANSIENT_UNAVAILABLE) { | |
6181 | // Else, wait. MDSMonitor will update the map to bring | |
6182 | // us to a conclusion eventually. | |
6183 | wait_on_list(waiting_for_mdsmap); | |
6184 | } else { | |
6185 | // Unexpected value! | |
6186 | ceph_abort(); | |
6187 | } | |
6188 | } | |
6189 | } | |
6190 | ||
6191 | populate_metadata(mount_root.empty() ? "/" : mount_root); | |
6192 | ||
6193 | filepath fp(CEPH_INO_ROOT); | |
6194 | if (!mount_root.empty()) { | |
6195 | fp = filepath(mount_root.c_str()); | |
6196 | } | |
6197 | while (true) { | |
6198 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
6199 | req->set_filepath(fp); | |
6200 | req->head.args.getattr.mask = CEPH_STAT_CAP_INODE_ALL; | |
6201 | int res = make_request(req, perms); | |
6202 | if (res < 0) { | |
f67539c2 | 6203 | if (res == -CEPHFS_EACCES && root) { |
7c673cae FG |
6204 | ldout(cct, 1) << __func__ << " EACCES on parent of mount point; quotas may not work" << dendl; |
6205 | break; | |
6206 | } | |
6207 | return res; | |
6208 | } | |
6209 | ||
6210 | if (fp.depth()) | |
6211 | fp.pop_dentry(); | |
6212 | else | |
6213 | break; | |
6214 | } | |
6215 | ||
11fdf7f2 | 6216 | ceph_assert(root); |
b3b6e05e | 6217 | _ll_get(root.get()); |
7c673cae | 6218 | |
7c673cae FG |
6219 | // trace? |
6220 | if (!cct->_conf->client_trace.empty()) { | |
6221 | traceout.open(cct->_conf->client_trace.c_str()); | |
6222 | if (traceout.is_open()) { | |
6223 | ldout(cct, 1) << "opened trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6224 | } else { | |
6225 | ldout(cct, 1) << "FAILED to open trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6226 | } | |
6227 | } | |
6228 | ||
6229 | /* | |
6230 | ldout(cct, 3) << "op: // client trace data structs" << dendl; | |
6231 | ldout(cct, 3) << "op: struct stat st;" << dendl; | |
6232 | ldout(cct, 3) << "op: struct utimbuf utim;" << dendl; | |
6233 | ldout(cct, 3) << "op: int readlinkbuf_len = 1000;" << dendl; | |
6234 | ldout(cct, 3) << "op: char readlinkbuf[readlinkbuf_len];" << dendl; | |
6235 | ldout(cct, 3) << "op: map<string, inode_t*> dir_contents;" << dendl; | |
6236 | ldout(cct, 3) << "op: map<int, int> open_files;" << dendl; | |
6237 | ldout(cct, 3) << "op: int fd;" << dendl; | |
6238 | */ | |
f67539c2 TL |
6239 | |
6240 | mref_writer.update_state(CLIENT_MOUNTED); | |
7c673cae FG |
6241 | return 0; |
6242 | } | |
6243 | ||
6244 | // UNMOUNT | |
6245 | ||
6246 | void Client::_close_sessions() | |
6247 | { | |
f6b5b4d7 | 6248 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
20effc67 | 6249 | if (it->second->state == MetaSession::STATE_REJECTED) |
f6b5b4d7 TL |
6250 | mds_sessions.erase(it++); |
6251 | else | |
6252 | ++it; | |
6253 | } | |
6254 | ||
7c673cae FG |
6255 | while (!mds_sessions.empty()) { |
6256 | // send session closes! | |
11fdf7f2 | 6257 | for (auto &p : mds_sessions) { |
20effc67 TL |
6258 | if (p.second->state != MetaSession::STATE_CLOSING) { |
6259 | _close_mds_session(p.second.get()); | |
f6b5b4d7 | 6260 | mds_ranks_closing.insert(p.first); |
7c673cae FG |
6261 | } |
6262 | } | |
6263 | ||
6264 | // wait for sessions to close | |
f6b5b4d7 TL |
6265 | double timo = cct->_conf.get_val<std::chrono::seconds>("client_shutdown_timeout").count(); |
6266 | ldout(cct, 2) << "waiting for " << mds_ranks_closing.size() << " mds session(s) to close (timeout: " | |
6267 | << timo << "s)" << dendl; | |
9f95a23c | 6268 | std::unique_lock l{client_lock, std::adopt_lock}; |
f6b5b4d7 TL |
6269 | if (!timo) { |
6270 | mount_cond.wait(l); | |
6271 | } else if (!mount_cond.wait_for(l, ceph::make_timespan(timo), [this] { return mds_ranks_closing.empty(); })) { | |
6272 | ldout(cct, 1) << mds_ranks_closing.size() << " mds(s) did not respond to session close -- timing out." << dendl; | |
6273 | while (!mds_ranks_closing.empty()) { | |
6274 | auto session = mds_sessions.at(*mds_ranks_closing.begin()); | |
6275 | // this prunes entry from mds_sessions and mds_ranks_closing | |
20effc67 | 6276 | _closed_mds_session(session.get(), -CEPHFS_ETIMEDOUT); |
f6b5b4d7 TL |
6277 | } |
6278 | } | |
6279 | ||
6280 | mds_ranks_closing.clear(); | |
9f95a23c | 6281 | l.release(); |
7c673cae FG |
6282 | } |
6283 | } | |
6284 | ||
522d829b TL |
6285 | void Client::flush_mdlog_sync(Inode *in) |
6286 | { | |
6287 | if (in->unsafe_ops.empty()) { | |
6288 | return; | |
6289 | } | |
6290 | ||
6291 | std::set<mds_rank_t> anchor; | |
6292 | for (auto &&p : in->unsafe_ops) { | |
6293 | anchor.emplace(p->mds); | |
6294 | } | |
6295 | if (in->auth_cap) { | |
6296 | anchor.emplace(in->auth_cap->session->mds_num); | |
6297 | } | |
6298 | ||
6299 | for (auto &rank : anchor) { | |
6300 | auto session = &mds_sessions.at(rank); | |
20effc67 | 6301 | flush_mdlog(session->get()); |
522d829b TL |
6302 | } |
6303 | } | |
6304 | ||
31f18b77 FG |
6305 | void Client::flush_mdlog_sync() |
6306 | { | |
522d829b | 6307 | if (mds_requests.empty()) |
31f18b77 | 6308 | return; |
11fdf7f2 | 6309 | for (auto &p : mds_sessions) { |
20effc67 | 6310 | flush_mdlog(p.second.get()); |
31f18b77 FG |
6311 | } |
6312 | } | |
6313 | ||
6314 | void Client::flush_mdlog(MetaSession *session) | |
6315 | { | |
6316 | // Only send this to Luminous or newer MDS daemons, older daemons | |
6317 | // will crash if they see an unknown CEPH_SESSION_* value in this msg. | |
6318 | const uint64_t features = session->con->get_features(); | |
6319 | if (HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
9f95a23c | 6320 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_FLUSH_MDLOG); |
11fdf7f2 | 6321 | session->con->send_message2(std::move(m)); |
31f18b77 FG |
6322 | } |
6323 | } | |
6324 | ||
6325 | ||
11fdf7f2 TL |
6326 | void Client::_abort_mds_sessions(int err) |
6327 | { | |
6328 | for (auto p = mds_requests.begin(); p != mds_requests.end(); ) { | |
6329 | auto req = p->second; | |
6330 | ++p; | |
6331 | // unsafe requests will be removed during close session below. | |
6332 | if (req->got_unsafe) | |
6333 | continue; | |
6334 | ||
6335 | req->abort(err); | |
6336 | if (req->caller_cond) { | |
6337 | req->kick = true; | |
9f95a23c | 6338 | req->caller_cond->notify_all(); |
11fdf7f2 TL |
6339 | } |
6340 | } | |
6341 | ||
6342 | // Process aborts on any requests that were on this waitlist. | |
6343 | // Any requests that were on a waiting_for_open session waitlist | |
6344 | // will get kicked during close session below. | |
6345 | signal_cond_list(waiting_for_mdsmap); | |
6346 | ||
6347 | // Force-close all sessions | |
6348 | while(!mds_sessions.empty()) { | |
20effc67 TL |
6349 | auto session = mds_sessions.begin()->second; |
6350 | _closed_mds_session(session.get(), err); | |
11fdf7f2 TL |
6351 | } |
6352 | } | |
6353 | ||
6354 | void Client::_unmount(bool abort) | |
7c673cae | 6355 | { |
f67539c2 TL |
6356 | /* |
6357 | * We are unmounting the client. | |
6358 | * | |
6359 | * Just declare the state to STATE_UNMOUNTING to block and fail | |
6360 | * any new comming "reader" and then try to wait all the in-flight | |
6361 | * "readers" to finish. | |
6362 | */ | |
6363 | RWRef_t mref_writer(mount_state, CLIENT_UNMOUNTING, false); | |
6364 | if (!mref_writer.is_first_writer()) | |
181888fb | 6365 | return; |
f67539c2 | 6366 | mref_writer.wait_readers_done(); |
7c673cae | 6367 | |
f67539c2 TL |
6368 | std::unique_lock lock{client_lock}; |
6369 | ||
6370 | if (abort || blocklisted) { | |
6371 | ldout(cct, 2) << "unmounting (" << (abort ? "abort)" : "blocklisted)") << dendl; | |
11fdf7f2 TL |
6372 | } else { |
6373 | ldout(cct, 2) << "unmounting" << dendl; | |
6374 | } | |
7c673cae | 6375 | |
b32b8144 FG |
6376 | deleg_timeout = 0; |
6377 | ||
11fdf7f2 | 6378 | if (abort) { |
f67539c2 | 6379 | mount_aborted = true; |
11fdf7f2 | 6380 | // Abort all mds sessions |
f67539c2 | 6381 | _abort_mds_sessions(-CEPHFS_ENOTCONN); |
11fdf7f2 | 6382 | |
f67539c2 | 6383 | objecter->op_cancel_writes(-CEPHFS_ENOTCONN); |
11fdf7f2 TL |
6384 | } else { |
6385 | // flush the mdlog for pending requests, if any | |
6386 | flush_mdlog_sync(); | |
6387 | } | |
6388 | ||
9f95a23c TL |
6389 | mount_cond.wait(lock, [this] { |
6390 | if (!mds_requests.empty()) { | |
6391 | ldout(cct, 10) << "waiting on " << mds_requests.size() << " requests" | |
6392 | << dendl; | |
6393 | } | |
6394 | return mds_requests.empty(); | |
6395 | }); | |
7c673cae FG |
6396 | |
6397 | cwd.reset(); | |
b3b6e05e | 6398 | root.reset(); |
7c673cae FG |
6399 | |
6400 | // clean up any unclosed files | |
6401 | while (!fd_map.empty()) { | |
6402 | Fh *fh = fd_map.begin()->second; | |
6403 | fd_map.erase(fd_map.begin()); | |
6404 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *fh->inode << dendl; | |
6405 | _release_fh(fh); | |
6406 | } | |
6407 | ||
6408 | while (!ll_unclosed_fh_set.empty()) { | |
6409 | set<Fh*>::iterator it = ll_unclosed_fh_set.begin(); | |
6410 | Fh *fh = *it; | |
6411 | ll_unclosed_fh_set.erase(fh); | |
6412 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *(fh->inode) << dendl; | |
6413 | _release_fh(fh); | |
6414 | } | |
6415 | ||
6416 | while (!opened_dirs.empty()) { | |
6417 | dir_result_t *dirp = *opened_dirs.begin(); | |
6418 | ldout(cct, 0) << " destroyed lost open dir " << dirp << " on " << *dirp->inode << dendl; | |
6419 | _closedir(dirp); | |
6420 | } | |
6421 | ||
6422 | _ll_drop_pins(); | |
6423 | ||
7c673cae FG |
6424 | if (cct->_conf->client_oc) { |
6425 | // flush/release all buffered data | |
11fdf7f2 TL |
6426 | std::list<InodeRef> anchor; |
6427 | for (auto& p : inode_map) { | |
6428 | Inode *in = p.second; | |
7c673cae | 6429 | if (!in) { |
11fdf7f2 TL |
6430 | ldout(cct, 0) << "null inode_map entry ino " << p.first << dendl; |
6431 | ceph_assert(in); | |
7c673cae | 6432 | } |
11fdf7f2 TL |
6433 | |
6434 | // prevent inode from getting freed | |
6435 | anchor.emplace_back(in); | |
6436 | ||
f67539c2 | 6437 | if (abort || blocklisted) { |
11fdf7f2 TL |
6438 | objectcacher->purge_set(&in->oset); |
6439 | } else if (!in->caps.empty()) { | |
7c673cae FG |
6440 | _release(in); |
6441 | _flush(in, new C_Client_FlushComplete(this, in)); | |
6442 | } | |
6443 | } | |
6444 | } | |
6445 | ||
f67539c2 | 6446 | if (abort || blocklisted) { |
20effc67 TL |
6447 | for (auto &q : mds_sessions) { |
6448 | auto s = q.second; | |
6449 | for (auto p = s->dirty_list.begin(); !p.end(); ) { | |
6450 | Inode *in = *p; | |
6451 | ++p; | |
6452 | if (in->dirty_caps) { | |
6453 | ldout(cct, 0) << " drop dirty caps on " << *in << dendl; | |
6454 | in->mark_caps_clean(); | |
6455 | put_inode(in); | |
6456 | } | |
11fdf7f2 TL |
6457 | } |
6458 | } | |
6459 | } else { | |
6460 | flush_caps_sync(); | |
6461 | wait_sync_caps(last_flush_tid); | |
6462 | } | |
7c673cae FG |
6463 | |
6464 | // empty lru cache | |
7c673cae FG |
6465 | trim_cache(); |
6466 | ||
f67539c2 TL |
6467 | delay_put_inodes(); |
6468 | ||
7c673cae FG |
6469 | while (lru.lru_get_size() > 0 || |
6470 | !inode_map.empty()) { | |
6471 | ldout(cct, 2) << "cache still has " << lru.lru_get_size() | |
6472 | << "+" << inode_map.size() << " items" | |
6473 | << ", waiting (for caps to release?)" | |
6474 | << dendl; | |
f67539c2 | 6475 | |
9f95a23c TL |
6476 | if (auto r = mount_cond.wait_for(lock, ceph::make_timespan(5)); |
6477 | r == std::cv_status::timeout) { | |
7c673cae FG |
6478 | dump_cache(NULL); |
6479 | } | |
6480 | } | |
11fdf7f2 TL |
6481 | ceph_assert(lru.lru_get_size() == 0); |
6482 | ceph_assert(inode_map.empty()); | |
7c673cae FG |
6483 | |
6484 | // stop tracing | |
6485 | if (!cct->_conf->client_trace.empty()) { | |
6486 | ldout(cct, 1) << "closing trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6487 | traceout.close(); | |
6488 | } | |
6489 | ||
f67539c2 TL |
6490 | // stop the tick thread |
6491 | tick_thread_stopped = true; | |
6492 | upkeep_cond.notify_one(); | |
6493 | ||
7c673cae FG |
6494 | _close_sessions(); |
6495 | ||
f67539c2 | 6496 | mref_writer.update_state(CLIENT_UNMOUNTED); |
7c673cae FG |
6497 | |
6498 | ldout(cct, 2) << "unmounted." << dendl; | |
6499 | } | |
6500 | ||
b32b8144 FG |
6501 | void Client::unmount() |
6502 | { | |
11fdf7f2 TL |
6503 | _unmount(false); |
6504 | } | |
6505 | ||
6506 | void Client::abort_conn() | |
6507 | { | |
11fdf7f2 | 6508 | _unmount(true); |
b32b8144 FG |
6509 | } |
6510 | ||
7c673cae FG |
6511 | void Client::flush_cap_releases() |
6512 | { | |
f67539c2 TL |
6513 | uint64_t nr_caps = 0; |
6514 | ||
7c673cae | 6515 | // send any cap releases |
11fdf7f2 | 6516 | for (auto &p : mds_sessions) { |
20effc67 TL |
6517 | auto session = p.second; |
6518 | if (session->release && mdsmap->is_clientreplay_or_active_or_stopping( | |
11fdf7f2 | 6519 | p.first)) { |
20effc67 | 6520 | nr_caps += session->release->caps.size(); |
7c673cae FG |
6521 | if (cct->_conf->client_inject_release_failure) { |
6522 | ldout(cct, 20) << __func__ << " injecting failure to send cap release message" << dendl; | |
7c673cae | 6523 | } else { |
20effc67 | 6524 | session->con->send_message2(std::move(session->release)); |
7c673cae | 6525 | } |
20effc67 | 6526 | session->release.reset(); |
7c673cae FG |
6527 | } |
6528 | } | |
f67539c2 TL |
6529 | |
6530 | if (nr_caps > 0) { | |
6531 | dec_pinned_icaps(nr_caps); | |
6532 | } | |
7c673cae FG |
6533 | } |
6534 | ||
f67539c2 | 6535 | void Client::renew_and_flush_cap_releases() |
7c673cae | 6536 | { |
f67539c2 TL |
6537 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
6538 | ||
6539 | if (!mount_aborted && mdsmap->get_epoch()) { | |
6540 | // renew caps? | |
6541 | utime_t el = ceph_clock_now() - last_cap_renew; | |
6542 | if (unlikely(el > mdsmap->get_session_timeout() / 3.0)) | |
6543 | renew_caps(); | |
6544 | ||
6545 | flush_cap_releases(); | |
7c673cae | 6546 | } |
f67539c2 TL |
6547 | } |
6548 | ||
6549 | void Client::tick() | |
6550 | { | |
6551 | ldout(cct, 20) << "tick" << dendl; | |
7c673cae | 6552 | |
7c673cae FG |
6553 | utime_t now = ceph_clock_now(); |
6554 | ||
f67539c2 TL |
6555 | /* |
6556 | * If the mount() is not finished | |
6557 | */ | |
6558 | if (is_mounting() && !mds_requests.empty()) { | |
7c673cae | 6559 | MetaRequest *req = mds_requests.begin()->second; |
f67539c2 | 6560 | |
7c673cae | 6561 | if (req->op_stamp + cct->_conf->client_mount_timeout < now) { |
f67539c2 | 6562 | req->abort(-CEPHFS_ETIMEDOUT); |
7c673cae | 6563 | if (req->caller_cond) { |
f67539c2 TL |
6564 | req->kick = true; |
6565 | req->caller_cond->notify_all(); | |
7c673cae FG |
6566 | } |
6567 | signal_cond_list(waiting_for_mdsmap); | |
11fdf7f2 | 6568 | for (auto &p : mds_sessions) { |
20effc67 | 6569 | signal_context_list(p.second->waiting_for_open); |
11fdf7f2 | 6570 | } |
7c673cae FG |
6571 | } |
6572 | } | |
6573 | ||
f67539c2 | 6574 | renew_and_flush_cap_releases(); |
7c673cae FG |
6575 | |
6576 | // delayed caps | |
28e407b8 | 6577 | xlist<Inode*>::iterator p = delayed_list.begin(); |
7c673cae FG |
6578 | while (!p.end()) { |
6579 | Inode *in = *p; | |
6580 | ++p; | |
f67539c2 | 6581 | if (!mount_aborted && in->hold_caps_until > now) |
7c673cae | 6582 | break; |
28e407b8 | 6583 | delayed_list.pop_front(); |
f67539c2 TL |
6584 | if (!mount_aborted) |
6585 | check_caps(in, CHECK_CAPS_NODELAY); | |
7c673cae FG |
6586 | } |
6587 | ||
f67539c2 TL |
6588 | if (!mount_aborted) |
6589 | collect_and_send_metrics(); | |
6590 | ||
6591 | delay_put_inodes(is_unmounting()); | |
7c673cae | 6592 | trim_cache(true); |
f6b5b4d7 | 6593 | |
f67539c2 | 6594 | if (blocklisted && (is_mounted() || is_unmounting()) && |
f6b5b4d7 TL |
6595 | last_auto_reconnect + 30 * 60 < now && |
6596 | cct->_conf.get_val<bool>("client_reconnect_stale")) { | |
6597 | messenger->client_reset(); | |
6598 | fd_gen++; // invalidate open files | |
f67539c2 | 6599 | blocklisted = false; |
f6b5b4d7 TL |
6600 | _kick_stale_sessions(); |
6601 | last_auto_reconnect = now; | |
6602 | } | |
7c673cae FG |
6603 | } |
6604 | ||
f67539c2 TL |
6605 | void Client::start_tick_thread() |
6606 | { | |
6607 | upkeeper = std::thread([this]() { | |
6608 | using time = ceph::coarse_mono_time; | |
6609 | using sec = std::chrono::seconds; | |
6610 | ||
6611 | auto last_tick = time::min(); | |
6612 | ||
6613 | std::unique_lock cl(client_lock); | |
6614 | while (!tick_thread_stopped) { | |
6615 | auto now = clock::now(); | |
6616 | auto since = now - last_tick; | |
6617 | ||
6618 | auto t_interval = clock::duration(cct->_conf.get_val<sec>("client_tick_interval")); | |
6619 | auto d_interval = clock::duration(cct->_conf.get_val<sec>("client_debug_inject_tick_delay")); | |
6620 | ||
6621 | auto interval = std::max(t_interval, d_interval); | |
6622 | if (likely(since >= interval*.90)) { | |
6623 | tick(); | |
6624 | last_tick = clock::now(); | |
6625 | } else { | |
6626 | interval -= since; | |
6627 | } | |
6628 | ||
6629 | ldout(cct, 20) << "upkeep thread waiting interval " << interval << dendl; | |
6630 | if (!tick_thread_stopped) | |
6631 | upkeep_cond.wait_for(cl, interval); | |
6632 | } | |
6633 | }); | |
6634 | } | |
6635 | ||
6636 | void Client::collect_and_send_metrics() { | |
6637 | ldout(cct, 20) << __func__ << dendl; | |
6638 | ||
6639 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6640 | ||
6641 | // right now, we only track and send global metrics. its sufficient | |
6642 | // to send these metrics to MDS rank0. | |
6643 | collect_and_send_global_metrics(); | |
6644 | } | |
6645 | ||
6646 | void Client::collect_and_send_global_metrics() { | |
6647 | ldout(cct, 20) << __func__ << dendl; | |
6648 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6649 | ||
6650 | if (!have_open_session((mds_rank_t)0)) { | |
6651 | ldout(cct, 5) << __func__ << ": no session with rank=0 -- not sending metric" | |
6652 | << dendl; | |
6653 | return; | |
6654 | } | |
6655 | auto session = _get_or_open_mds_session((mds_rank_t)0); | |
6656 | if (!session->mds_features.test(CEPHFS_FEATURE_METRIC_COLLECT)) { | |
6657 | ldout(cct, 5) << __func__ << ": rank=0 does not support metrics" << dendl; | |
6658 | return; | |
6659 | } | |
6660 | ||
6661 | ClientMetricMessage metric; | |
6662 | std::vector<ClientMetricMessage> message; | |
6663 | ||
6664 | // read latency | |
6665 | metric = ClientMetricMessage(ReadLatencyPayload(logger->tget(l_c_read))); | |
6666 | message.push_back(metric); | |
6667 | ||
6668 | // write latency | |
6669 | metric = ClientMetricMessage(WriteLatencyPayload(logger->tget(l_c_wrlat))); | |
6670 | message.push_back(metric); | |
6671 | ||
6672 | // metadata latency | |
6673 | metric = ClientMetricMessage(MetadataLatencyPayload(logger->tget(l_c_lat))); | |
6674 | message.push_back(metric); | |
6675 | ||
6676 | // cap hit ratio -- nr_caps is unused right now | |
6677 | auto [cap_hits, cap_misses] = get_cap_hit_rates(); | |
6678 | metric = ClientMetricMessage(CapInfoPayload(cap_hits, cap_misses, 0)); | |
6679 | message.push_back(metric); | |
6680 | ||
6681 | // dentry lease hit ratio | |
6682 | auto [dlease_hits, dlease_misses, nr] = get_dlease_hit_rates(); | |
6683 | metric = ClientMetricMessage(DentryLeasePayload(dlease_hits, dlease_misses, nr)); | |
6684 | message.push_back(metric); | |
6685 | ||
6686 | // opened files | |
6687 | { | |
6688 | auto [opened_files, total_inodes] = get_opened_files_rates(); | |
6689 | metric = ClientMetricMessage(OpenedFilesPayload(opened_files, total_inodes)); | |
6690 | } | |
6691 | message.push_back(metric); | |
6692 | ||
6693 | // pinned i_caps | |
6694 | { | |
6695 | auto [pinned_icaps, total_inodes] = get_pinned_icaps_rates(); | |
6696 | metric = ClientMetricMessage(PinnedIcapsPayload(pinned_icaps, total_inodes)); | |
6697 | } | |
6698 | message.push_back(metric); | |
6699 | ||
6700 | // opened inodes | |
6701 | { | |
6702 | auto [opened_inodes, total_inodes] = get_opened_inodes_rates(); | |
6703 | metric = ClientMetricMessage(OpenedInodesPayload(opened_inodes, total_inodes)); | |
6704 | } | |
6705 | message.push_back(metric); | |
6706 | ||
a4b75251 TL |
6707 | // read io sizes |
6708 | metric = ClientMetricMessage(ReadIoSizesPayload(total_read_ops, | |
6709 | total_read_size)); | |
6710 | message.push_back(metric); | |
6711 | ||
6712 | // write io sizes | |
6713 | metric = ClientMetricMessage(WriteIoSizesPayload(total_write_ops, | |
6714 | total_write_size)); | |
6715 | message.push_back(metric); | |
6716 | ||
f67539c2 TL |
6717 | session->con->send_message2(make_message<MClientMetrics>(std::move(message))); |
6718 | } | |
6719 | ||
7c673cae FG |
6720 | void Client::renew_caps() |
6721 | { | |
6722 | ldout(cct, 10) << "renew_caps()" << dendl; | |
6723 | last_cap_renew = ceph_clock_now(); | |
6724 | ||
11fdf7f2 TL |
6725 | for (auto &p : mds_sessions) { |
6726 | ldout(cct, 15) << "renew_caps requesting from mds." << p.first << dendl; | |
6727 | if (mdsmap->get_state(p.first) >= MDSMap::STATE_REJOIN) | |
20effc67 | 6728 | renew_caps(p.second.get()); |
7c673cae FG |
6729 | } |
6730 | } | |
6731 | ||
6732 | void Client::renew_caps(MetaSession *session) | |
6733 | { | |
6734 | ldout(cct, 10) << "renew_caps mds." << session->mds_num << dendl; | |
6735 | session->last_cap_renew_request = ceph_clock_now(); | |
6736 | uint64_t seq = ++session->cap_renew_seq; | |
9f95a23c | 6737 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_RENEWCAPS, seq)); |
7c673cae FG |
6738 | } |
6739 | ||
6740 | ||
6741 | // =============================================================== | |
6742 | // high level (POSIXy) interface | |
6743 | ||
6744 | int Client::_do_lookup(Inode *dir, const string& name, int mask, | |
6745 | InodeRef *target, const UserPerm& perms) | |
6746 | { | |
6747 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
6748 | MetaRequest *req = new MetaRequest(op); | |
6749 | filepath path; | |
6750 | dir->make_nosnap_relative_path(path); | |
6751 | path.push_dentry(name); | |
6752 | req->set_filepath(path); | |
6753 | req->set_inode(dir); | |
6754 | if (cct->_conf->client_debug_getattr_caps && op == CEPH_MDS_OP_LOOKUP) | |
6755 | mask |= DEBUG_GETATTR_CAPS; | |
6756 | req->head.args.getattr.mask = mask; | |
6757 | ||
11fdf7f2 | 6758 | ldout(cct, 10) << __func__ << " on " << path << dendl; |
7c673cae FG |
6759 | |
6760 | int r = make_request(req, perms, target); | |
11fdf7f2 | 6761 | ldout(cct, 10) << __func__ << " res is " << r << dendl; |
7c673cae FG |
6762 | return r; |
6763 | } | |
6764 | ||
f67539c2 TL |
6765 | bool Client::_dentry_valid(const Dentry *dn) |
6766 | { | |
6767 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6768 | ||
6769 | // is dn lease valid? | |
6770 | utime_t now = ceph_clock_now(); | |
6771 | if (dn->lease_mds >= 0 && dn->lease_ttl > now && | |
6772 | mds_sessions.count(dn->lease_mds)) { | |
20effc67 TL |
6773 | auto s = mds_sessions.at(dn->lease_mds); |
6774 | if (s->cap_ttl > now && s->cap_gen == dn->lease_gen) { | |
f67539c2 TL |
6775 | dlease_hit(); |
6776 | return true; | |
6777 | } | |
6778 | ||
20effc67 | 6779 | ldout(cct, 20) << " bad lease, cap_ttl " << s->cap_ttl << ", cap_gen " << s->cap_gen |
f67539c2 TL |
6780 | << " vs lease_gen " << dn->lease_gen << dendl; |
6781 | } | |
6782 | ||
6783 | dlease_miss(); | |
6784 | return false; | |
6785 | } | |
6786 | ||
7c673cae | 6787 | int Client::_lookup(Inode *dir, const string& dname, int mask, InodeRef *target, |
f67539c2 | 6788 | const UserPerm& perms, std::string* alternate_name) |
7c673cae FG |
6789 | { |
6790 | int r = 0; | |
6791 | Dentry *dn = NULL; | |
f67539c2 | 6792 | bool did_lookup_request = false; |
f91f0fd5 TL |
6793 | // can only request shared caps |
6794 | mask &= CEPH_CAP_ANY_SHARED | CEPH_STAT_RSTAT; | |
7c673cae | 6795 | |
7c673cae | 6796 | if (dname == "..") { |
11fdf7f2 TL |
6797 | if (dir->dentries.empty()) { |
6798 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); | |
6799 | filepath path(dir->ino); | |
6800 | req->set_filepath(path); | |
6801 | ||
6802 | InodeRef tmptarget; | |
6803 | int r = make_request(req, perms, &tmptarget, NULL, rand() % mdsmap->get_num_in_mds()); | |
6804 | ||
6805 | if (r == 0) { | |
f91f0fd5 | 6806 | *target = std::move(tmptarget); |
11fdf7f2 TL |
6807 | ldout(cct, 8) << __func__ << " found target " << (*target)->ino << dendl; |
6808 | } else { | |
6809 | *target = dir; | |
6810 | } | |
6811 | } | |
7c673cae FG |
6812 | else |
6813 | *target = dir->get_first_parent()->dir->parent_inode; //dirs can't be hard-linked | |
6814 | goto done; | |
6815 | } | |
6816 | ||
6817 | if (dname == ".") { | |
6818 | *target = dir; | |
6819 | goto done; | |
6820 | } | |
6821 | ||
11fdf7f2 | 6822 | if (!dir->is_dir()) { |
f67539c2 | 6823 | r = -CEPHFS_ENOTDIR; |
11fdf7f2 TL |
6824 | goto done; |
6825 | } | |
6826 | ||
7c673cae | 6827 | if (dname.length() > NAME_MAX) { |
f67539c2 | 6828 | r = -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
6829 | goto done; |
6830 | } | |
6831 | ||
6832 | if (dname == cct->_conf->client_snapdir && | |
6833 | dir->snapid == CEPH_NOSNAP) { | |
6834 | *target = open_snapdir(dir); | |
6835 | goto done; | |
6836 | } | |
6837 | ||
f67539c2 | 6838 | relookup: |
7c673cae FG |
6839 | if (dir->dir && |
6840 | dir->dir->dentries.count(dname)) { | |
6841 | dn = dir->dir->dentries[dname]; | |
6842 | ||
f67539c2 TL |
6843 | ldout(cct, 20) << __func__ << " have " << *dn << " from mds." << dn->lease_mds |
6844 | << " ttl " << dn->lease_ttl << " seq " << dn->lease_seq << dendl; | |
7c673cae | 6845 | |
94b18763 | 6846 | if (!dn->inode || dn->inode->caps_issued_mask(mask, true)) { |
f67539c2 TL |
6847 | if (_dentry_valid(dn)) { |
6848 | // touch this mds's dir cap too, even though we don't _explicitly_ use it here, to | |
6849 | // make trim_caps() behave. | |
6850 | dir->try_touch_cap(dn->lease_mds); | |
6851 | goto hit_dn; | |
7c673cae | 6852 | } |
92f5a8d4 | 6853 | // dir shared caps? |
94b18763 | 6854 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 6855 | if (dn->cap_shared_gen == dir->shared_gen && |
94b18763 | 6856 | (!dn->inode || dn->inode->caps_issued_mask(mask, true))) |
7c673cae FG |
6857 | goto hit_dn; |
6858 | if (!dn->inode && (dir->flags & I_COMPLETE)) { | |
11fdf7f2 | 6859 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " |
7c673cae | 6860 | << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 6861 | return -CEPHFS_ENOENT; |
7c673cae FG |
6862 | } |
6863 | } | |
6864 | } else { | |
6865 | ldout(cct, 20) << " no cap on " << dn->inode->vino() << dendl; | |
6866 | } | |
6867 | } else { | |
6868 | // can we conclude ENOENT locally? | |
94b18763 | 6869 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true) && |
7c673cae | 6870 | (dir->flags & I_COMPLETE)) { |
11fdf7f2 | 6871 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 6872 | return -CEPHFS_ENOENT; |
7c673cae FG |
6873 | } |
6874 | } | |
6875 | ||
f67539c2 TL |
6876 | if (did_lookup_request) { |
6877 | r = 0; | |
6878 | goto done; | |
6879 | } | |
7c673cae | 6880 | r = _do_lookup(dir, dname, mask, target, perms); |
f67539c2 TL |
6881 | did_lookup_request = true; |
6882 | if (r == 0) { | |
6883 | /* complete lookup to get dentry for alternate_name */ | |
6884 | goto relookup; | |
6885 | } else { | |
6886 | goto done; | |
6887 | } | |
6888 | ||
6889 | hit_dn: | |
6890 | if (dn->inode) { | |
7c673cae | 6891 | *target = dn->inode; |
f67539c2 TL |
6892 | if (alternate_name) |
6893 | *alternate_name = dn->alternate_name; | |
7c673cae | 6894 | } else { |
f67539c2 | 6895 | r = -CEPHFS_ENOENT; |
7c673cae FG |
6896 | } |
6897 | touch_dn(dn); | |
f67539c2 | 6898 | goto done; |
7c673cae FG |
6899 | |
6900 | done: | |
6901 | if (r < 0) | |
11fdf7f2 | 6902 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << r << dendl; |
7c673cae | 6903 | else |
11fdf7f2 | 6904 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << **target << dendl; |
7c673cae FG |
6905 | return r; |
6906 | } | |
6907 | ||
6908 | int Client::get_or_create(Inode *dir, const char* name, | |
6909 | Dentry **pdn, bool expect_null) | |
6910 | { | |
6911 | // lookup | |
11fdf7f2 | 6912 | ldout(cct, 20) << __func__ << " " << *dir << " name " << name << dendl; |
7c673cae FG |
6913 | dir->open_dir(); |
6914 | if (dir->dir->dentries.count(name)) { | |
6915 | Dentry *dn = dir->dir->dentries[name]; | |
f67539c2 TL |
6916 | if (_dentry_valid(dn)) { |
6917 | if (expect_null) | |
6918 | return -CEPHFS_EEXIST; | |
7c673cae FG |
6919 | } |
6920 | *pdn = dn; | |
6921 | } else { | |
6922 | // otherwise link up a new one | |
6923 | *pdn = link(dir->dir, name, NULL, NULL); | |
6924 | } | |
6925 | ||
6926 | // success | |
6927 | return 0; | |
6928 | } | |
6929 | ||
f67539c2 TL |
6930 | int Client::walk(std::string_view path, walk_dentry_result* wdr, const UserPerm& perms, bool followsym) |
6931 | { | |
6932 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
6933 | if (!mref_reader.is_state_satisfied()) | |
6934 | return -CEPHFS_ENOTCONN; | |
6935 | ||
6936 | ldout(cct, 10) << __func__ << ": " << path << dendl; | |
6937 | ||
6938 | std::scoped_lock lock(client_lock); | |
6939 | ||
6940 | return path_walk(path, wdr, perms, followsym); | |
6941 | } | |
6942 | ||
7c673cae | 6943 | int Client::path_walk(const filepath& origpath, InodeRef *end, |
b3b6e05e | 6944 | const UserPerm& perms, bool followsym, int mask, InodeRef dirinode) |
f67539c2 TL |
6945 | { |
6946 | walk_dentry_result wdr; | |
b3b6e05e | 6947 | int rc = path_walk(origpath, &wdr, perms, followsym, mask, dirinode); |
f67539c2 TL |
6948 | *end = std::move(wdr.in); |
6949 | return rc; | |
6950 | } | |
6951 | ||
b3b6e05e TL |
6952 | int Client::path_walk(const filepath& origpath, walk_dentry_result* result, const UserPerm& perms, |
6953 | bool followsym, int mask, InodeRef dirinode) | |
7c673cae FG |
6954 | { |
6955 | filepath path = origpath; | |
6956 | InodeRef cur; | |
f67539c2 | 6957 | std::string alternate_name; |
7c673cae FG |
6958 | if (origpath.absolute()) |
6959 | cur = root; | |
b3b6e05e | 6960 | else if (!dirinode) |
7c673cae | 6961 | cur = cwd; |
b3b6e05e TL |
6962 | else { |
6963 | cur = dirinode; | |
6964 | } | |
11fdf7f2 | 6965 | ceph_assert(cur); |
7c673cae | 6966 | |
b3b6e05e | 6967 | ldout(cct, 20) << __func__ << " cur=" << *cur << dendl; |
11fdf7f2 | 6968 | ldout(cct, 10) << __func__ << " " << path << dendl; |
7c673cae FG |
6969 | |
6970 | int symlinks = 0; | |
6971 | ||
6972 | unsigned i=0; | |
6973 | while (i < path.depth() && cur) { | |
6974 | int caps = 0; | |
6975 | const string &dname = path[i]; | |
6976 | ldout(cct, 10) << " " << i << " " << *cur << " " << dname << dendl; | |
6977 | ldout(cct, 20) << " (path is " << path << ")" << dendl; | |
6978 | InodeRef next; | |
6979 | if (cct->_conf->client_permissions) { | |
6980 | int r = may_lookup(cur.get(), perms); | |
6981 | if (r < 0) | |
6982 | return r; | |
6983 | caps = CEPH_CAP_AUTH_SHARED; | |
6984 | } | |
6985 | ||
6986 | /* Get extra requested caps on the last component */ | |
6987 | if (i == (path.depth() - 1)) | |
6988 | caps |= mask; | |
f67539c2 | 6989 | int r = _lookup(cur.get(), dname, caps, &next, perms, &alternate_name); |
7c673cae FG |
6990 | if (r < 0) |
6991 | return r; | |
6992 | // only follow trailing symlink if followsym. always follow | |
6993 | // 'directory' symlinks. | |
6994 | if (next && next->is_symlink()) { | |
6995 | symlinks++; | |
6996 | ldout(cct, 20) << " symlink count " << symlinks << ", value is '" << next->symlink << "'" << dendl; | |
6997 | if (symlinks > MAXSYMLINKS) { | |
f67539c2 | 6998 | return -CEPHFS_ELOOP; |
7c673cae FG |
6999 | } |
7000 | ||
7001 | if (i < path.depth() - 1) { | |
7002 | // dir symlink | |
7003 | // replace consumed components of path with symlink dir target | |
7004 | filepath resolved(next->symlink.c_str()); | |
7005 | resolved.append(path.postfixpath(i + 1)); | |
7006 | path = resolved; | |
7007 | i = 0; | |
7008 | if (next->symlink[0] == '/') { | |
7009 | cur = root; | |
7010 | } | |
7011 | continue; | |
7012 | } else if (followsym) { | |
7013 | if (next->symlink[0] == '/') { | |
7014 | path = next->symlink.c_str(); | |
7015 | i = 0; | |
7016 | // reset position | |
7017 | cur = root; | |
7018 | } else { | |
7019 | filepath more(next->symlink.c_str()); | |
7020 | // we need to remove the symlink component from off of the path | |
7021 | // before adding the target that the symlink points to. remain | |
7022 | // at the same position in the path. | |
7023 | path.pop_dentry(); | |
7024 | path.append(more); | |
7025 | } | |
7026 | continue; | |
7027 | } | |
7028 | } | |
7029 | cur.swap(next); | |
7030 | i++; | |
7031 | } | |
7032 | if (!cur) | |
f67539c2 TL |
7033 | return -CEPHFS_ENOENT; |
7034 | if (result) { | |
7035 | result->in = std::move(cur); | |
7036 | result->alternate_name = std::move(alternate_name); | |
7037 | } | |
7c673cae FG |
7038 | return 0; |
7039 | } | |
7040 | ||
7041 | ||
7042 | // namespace ops | |
7043 | ||
f67539c2 | 7044 | int Client::link(const char *relexisting, const char *relpath, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7045 | { |
f67539c2 TL |
7046 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7047 | if (!mref_reader.is_state_satisfied()) | |
7048 | return -CEPHFS_ENOTCONN; | |
7049 | ||
7c673cae FG |
7050 | tout(cct) << "link" << std::endl; |
7051 | tout(cct) << relexisting << std::endl; | |
7052 | tout(cct) << relpath << std::endl; | |
7053 | ||
7054 | filepath existing(relexisting); | |
7055 | ||
7056 | InodeRef in, dir; | |
f67539c2 TL |
7057 | |
7058 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7059 | int r = path_walk(existing, &in, perm, true); |
7060 | if (r < 0) | |
7061 | return r; | |
7062 | if (std::string(relpath) == "/") { | |
f67539c2 | 7063 | r = -CEPHFS_EEXIST; |
7c673cae FG |
7064 | return r; |
7065 | } | |
7066 | filepath path(relpath); | |
7067 | string name = path.last_dentry(); | |
7068 | path.pop_dentry(); | |
7069 | ||
7070 | r = path_walk(path, &dir, perm, true); | |
7071 | if (r < 0) | |
7072 | return r; | |
7073 | if (cct->_conf->client_permissions) { | |
7074 | if (S_ISDIR(in->mode)) { | |
f67539c2 | 7075 | r = -CEPHFS_EPERM; |
7c673cae FG |
7076 | return r; |
7077 | } | |
7078 | r = may_hardlink(in.get(), perm); | |
7079 | if (r < 0) | |
7080 | return r; | |
7081 | r = may_create(dir.get(), perm); | |
7082 | if (r < 0) | |
7083 | return r; | |
7084 | } | |
f67539c2 | 7085 | r = _link(in.get(), dir.get(), name.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7086 | return r; |
7087 | } | |
7088 | ||
7089 | int Client::unlink(const char *relpath, const UserPerm& perm) | |
b3b6e05e TL |
7090 | { |
7091 | return unlinkat(CEPHFS_AT_FDCWD, relpath, 0, perm); | |
7092 | } | |
7093 | ||
7094 | int Client::unlinkat(int dirfd, const char *relpath, int flags, const UserPerm& perm) | |
7c673cae | 7095 | { |
f67539c2 | 7096 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7097 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7098 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7099 | } |
f67539c2 | 7100 | |
11fdf7f2 | 7101 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7102 | tout(cct) << dirfd << std::endl; |
7c673cae | 7103 | tout(cct) << relpath << std::endl; |
b3b6e05e | 7104 | tout(cct) << flags << std::endl; |
7c673cae | 7105 | |
b3b6e05e TL |
7106 | if (std::string(relpath) == "/") { |
7107 | return flags & AT_REMOVEDIR ? -CEPHFS_EBUSY : -CEPHFS_EISDIR; | |
7108 | } | |
7c673cae FG |
7109 | |
7110 | filepath path(relpath); | |
7111 | string name = path.last_dentry(); | |
7112 | path.pop_dentry(); | |
7113 | InodeRef dir; | |
f67539c2 TL |
7114 | |
7115 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7116 | |
7117 | InodeRef dirinode; | |
7118 | int r = get_fd_inode(dirfd, &dirinode); | |
7119 | if (r < 0) { | |
7120 | return r; | |
7121 | } | |
7122 | ||
7123 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7124 | if (r < 0) { | |
7c673cae | 7125 | return r; |
b3b6e05e | 7126 | } |
7c673cae FG |
7127 | if (cct->_conf->client_permissions) { |
7128 | r = may_delete(dir.get(), name.c_str(), perm); | |
b3b6e05e | 7129 | if (r < 0) { |
7c673cae | 7130 | return r; |
b3b6e05e | 7131 | } |
7c673cae | 7132 | } |
b3b6e05e TL |
7133 | if (flags & AT_REMOVEDIR) { |
7134 | r = _rmdir(dir.get(), name.c_str(), perm); | |
7135 | } else { | |
7136 | r = _unlink(dir.get(), name.c_str(), perm); | |
7137 | } | |
7138 | return r; | |
7c673cae FG |
7139 | } |
7140 | ||
f67539c2 | 7141 | int Client::rename(const char *relfrom, const char *relto, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7142 | { |
f67539c2 TL |
7143 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7144 | if (!mref_reader.is_state_satisfied()) | |
7145 | return -CEPHFS_ENOTCONN; | |
7146 | ||
11fdf7f2 | 7147 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7148 | tout(cct) << relfrom << std::endl; |
7149 | tout(cct) << relto << std::endl; | |
7150 | ||
7151 | if (std::string(relfrom) == "/" || std::string(relto) == "/") | |
f67539c2 | 7152 | return -CEPHFS_EBUSY; |
7c673cae FG |
7153 | |
7154 | filepath from(relfrom); | |
7155 | filepath to(relto); | |
7156 | string fromname = from.last_dentry(); | |
7157 | from.pop_dentry(); | |
7158 | string toname = to.last_dentry(); | |
7159 | to.pop_dentry(); | |
7160 | ||
7161 | InodeRef fromdir, todir; | |
f67539c2 TL |
7162 | |
7163 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7164 | int r = path_walk(from, &fromdir, perm); |
7165 | if (r < 0) | |
7166 | goto out; | |
7167 | r = path_walk(to, &todir, perm); | |
7168 | if (r < 0) | |
7169 | goto out; | |
7170 | ||
7171 | if (cct->_conf->client_permissions) { | |
7172 | int r = may_delete(fromdir.get(), fromname.c_str(), perm); | |
7173 | if (r < 0) | |
7174 | return r; | |
7175 | r = may_delete(todir.get(), toname.c_str(), perm); | |
f67539c2 | 7176 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
7177 | return r; |
7178 | } | |
f67539c2 | 7179 | r = _rename(fromdir.get(), fromname.c_str(), todir.get(), toname.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7180 | out: |
7181 | return r; | |
7182 | } | |
7183 | ||
7184 | // dirs | |
7185 | ||
f67539c2 | 7186 | int Client::mkdir(const char *relpath, mode_t mode, const UserPerm& perm, std::string alternate_name) |
b3b6e05e TL |
7187 | { |
7188 | return mkdirat(CEPHFS_AT_FDCWD, relpath, mode, perm, alternate_name); | |
7189 | } | |
7190 | ||
7191 | int Client::mkdirat(int dirfd, const char *relpath, mode_t mode, const UserPerm& perm, | |
7192 | std::string alternate_name) | |
7c673cae | 7193 | { |
f67539c2 TL |
7194 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7195 | if (!mref_reader.is_state_satisfied()) | |
7196 | return -CEPHFS_ENOTCONN; | |
7197 | ||
11fdf7f2 | 7198 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7199 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7200 | tout(cct) << relpath << std::endl; |
7201 | tout(cct) << mode << std::endl; | |
11fdf7f2 | 7202 | ldout(cct, 10) << __func__ << ": " << relpath << dendl; |
7c673cae | 7203 | |
b3b6e05e | 7204 | if (std::string(relpath) == "/") { |
f67539c2 | 7205 | return -CEPHFS_EEXIST; |
b3b6e05e | 7206 | } |
7c673cae FG |
7207 | |
7208 | filepath path(relpath); | |
7209 | string name = path.last_dentry(); | |
7210 | path.pop_dentry(); | |
7211 | InodeRef dir; | |
f67539c2 TL |
7212 | |
7213 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7214 | |
7215 | InodeRef dirinode; | |
7216 | int r = get_fd_inode(dirfd, &dirinode); | |
7217 | if (r < 0) { | |
7c673cae | 7218 | return r; |
b3b6e05e TL |
7219 | } |
7220 | ||
7221 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7222 | if (r < 0) { | |
7223 | return r; | |
7224 | } | |
7c673cae FG |
7225 | if (cct->_conf->client_permissions) { |
7226 | r = may_create(dir.get(), perm); | |
b3b6e05e | 7227 | if (r < 0) { |
7c673cae | 7228 | return r; |
b3b6e05e | 7229 | } |
7c673cae | 7230 | } |
f67539c2 | 7231 | return _mkdir(dir.get(), name.c_str(), mode, perm, 0, {}, std::move(alternate_name)); |
7c673cae FG |
7232 | } |
7233 | ||
7234 | int Client::mkdirs(const char *relpath, mode_t mode, const UserPerm& perms) | |
7235 | { | |
f67539c2 TL |
7236 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7237 | if (!mref_reader.is_state_satisfied()) | |
7238 | return -CEPHFS_ENOTCONN; | |
7239 | ||
7c673cae | 7240 | ldout(cct, 10) << "Client::mkdirs " << relpath << dendl; |
11fdf7f2 | 7241 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7242 | tout(cct) << relpath << std::endl; |
7243 | tout(cct) << mode << std::endl; | |
7244 | ||
7245 | //get through existing parts of path | |
7246 | filepath path(relpath); | |
7247 | unsigned int i; | |
7248 | int r = 0, caps = 0; | |
7249 | InodeRef cur, next; | |
f67539c2 TL |
7250 | |
7251 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7252 | cur = cwd; |
7253 | for (i=0; i<path.depth(); ++i) { | |
7254 | if (cct->_conf->client_permissions) { | |
7255 | r = may_lookup(cur.get(), perms); | |
7256 | if (r < 0) | |
7257 | break; | |
7258 | caps = CEPH_CAP_AUTH_SHARED; | |
7259 | } | |
7260 | r = _lookup(cur.get(), path[i].c_str(), caps, &next, perms); | |
7261 | if (r < 0) | |
7262 | break; | |
7263 | cur.swap(next); | |
7264 | } | |
f67539c2 | 7265 | if (r!=-CEPHFS_ENOENT) return r; |
11fdf7f2 | 7266 | ldout(cct, 20) << __func__ << " got through " << i << " directories on path " << relpath << dendl; |
7c673cae FG |
7267 | //make new directory at each level |
7268 | for (; i<path.depth(); ++i) { | |
7269 | if (cct->_conf->client_permissions) { | |
7270 | r = may_create(cur.get(), perms); | |
7271 | if (r < 0) | |
7272 | return r; | |
7273 | } | |
7274 | //make new dir | |
7275 | r = _mkdir(cur.get(), path[i].c_str(), mode, perms, &next); | |
c07f9fc5 | 7276 | |
7c673cae | 7277 | //check proper creation/existence |
f67539c2 | 7278 | if(-CEPHFS_EEXIST == r && i < path.depth() - 1) { |
c07f9fc5 FG |
7279 | r = _lookup(cur.get(), path[i].c_str(), CEPH_CAP_AUTH_SHARED, &next, perms); |
7280 | } | |
7281 | if (r < 0) | |
7282 | return r; | |
7c673cae FG |
7283 | //move to new dir and continue |
7284 | cur.swap(next); | |
11fdf7f2 | 7285 | ldout(cct, 20) << __func__ << ": successfully created directory " |
7c673cae FG |
7286 | << filepath(cur->ino).get_path() << dendl; |
7287 | } | |
7288 | return 0; | |
7289 | } | |
7290 | ||
7291 | int Client::rmdir(const char *relpath, const UserPerm& perms) | |
7292 | { | |
b3b6e05e | 7293 | return unlinkat(CEPHFS_AT_FDCWD, relpath, AT_REMOVEDIR, perms); |
7c673cae FG |
7294 | } |
7295 | ||
7296 | int Client::mknod(const char *relpath, mode_t mode, const UserPerm& perms, dev_t rdev) | |
f67539c2 TL |
7297 | { |
7298 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
7299 | if (!mref_reader.is_state_satisfied()) | |
7300 | return -CEPHFS_ENOTCONN; | |
7301 | ||
11fdf7f2 | 7302 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7303 | tout(cct) << relpath << std::endl; |
7304 | tout(cct) << mode << std::endl; | |
7305 | tout(cct) << rdev << std::endl; | |
7306 | ||
7307 | if (std::string(relpath) == "/") | |
f67539c2 | 7308 | return -CEPHFS_EEXIST; |
7c673cae FG |
7309 | |
7310 | filepath path(relpath); | |
7311 | string name = path.last_dentry(); | |
7312 | path.pop_dentry(); | |
7313 | InodeRef dir; | |
f67539c2 TL |
7314 | |
7315 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7316 | int r = path_walk(path, &dir, perms); |
7317 | if (r < 0) | |
7318 | return r; | |
7319 | if (cct->_conf->client_permissions) { | |
7320 | int r = may_create(dir.get(), perms); | |
7321 | if (r < 0) | |
7322 | return r; | |
7323 | } | |
7324 | return _mknod(dir.get(), name.c_str(), mode, rdev, perms); | |
7325 | } | |
7326 | ||
7327 | // symlinks | |
7328 | ||
f67539c2 | 7329 | int Client::symlink(const char *target, const char *relpath, const UserPerm& perms, std::string alternate_name) |
b3b6e05e TL |
7330 | { |
7331 | return symlinkat(target, CEPHFS_AT_FDCWD, relpath, perms, alternate_name); | |
7332 | } | |
7333 | ||
7334 | int Client::symlinkat(const char *target, int dirfd, const char *relpath, const UserPerm& perms, | |
7335 | std::string alternate_name) | |
7c673cae | 7336 | { |
f67539c2 | 7337 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7338 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7339 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7340 | } |
f67539c2 | 7341 | |
11fdf7f2 | 7342 | tout(cct) << __func__ << std::endl; |
7c673cae | 7343 | tout(cct) << target << std::endl; |
b3b6e05e | 7344 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7345 | tout(cct) << relpath << std::endl; |
7346 | ||
b3b6e05e | 7347 | if (std::string(relpath) == "/") { |
f67539c2 | 7348 | return -CEPHFS_EEXIST; |
b3b6e05e | 7349 | } |
7c673cae FG |
7350 | |
7351 | filepath path(relpath); | |
7352 | string name = path.last_dentry(); | |
7353 | path.pop_dentry(); | |
7354 | InodeRef dir; | |
f67539c2 TL |
7355 | |
7356 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7357 | |
7358 | InodeRef dirinode; | |
7359 | int r = get_fd_inode(dirfd, &dirinode); | |
7360 | if (r < 0) { | |
7c673cae | 7361 | return r; |
b3b6e05e TL |
7362 | } |
7363 | r = path_walk(path, &dir, perms, true, 0, dirinode); | |
7364 | if (r < 0) { | |
7365 | return r; | |
7366 | } | |
7c673cae FG |
7367 | if (cct->_conf->client_permissions) { |
7368 | int r = may_create(dir.get(), perms); | |
b3b6e05e | 7369 | if (r < 0) { |
7c673cae | 7370 | return r; |
b3b6e05e | 7371 | } |
7c673cae | 7372 | } |
f67539c2 | 7373 | return _symlink(dir.get(), name.c_str(), target, perms, std::move(alternate_name)); |
7c673cae FG |
7374 | } |
7375 | ||
7376 | int Client::readlink(const char *relpath, char *buf, loff_t size, const UserPerm& perms) | |
7377 | { | |
b3b6e05e TL |
7378 | return readlinkat(CEPHFS_AT_FDCWD, relpath, buf, size, perms); |
7379 | } | |
7380 | ||
7381 | int Client::readlinkat(int dirfd, const char *relpath, char *buf, loff_t size, const UserPerm& perms) { | |
f67539c2 | 7382 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7383 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7384 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7385 | } |
f67539c2 | 7386 | |
11fdf7f2 | 7387 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7388 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7389 | tout(cct) << relpath << std::endl; |
7390 | ||
b3b6e05e | 7391 | InodeRef dirinode; |
f67539c2 | 7392 | std::scoped_lock lock(client_lock); |
b3b6e05e TL |
7393 | int r = get_fd_inode(dirfd, &dirinode); |
7394 | if (r < 0) { | |
7c673cae | 7395 | return r; |
b3b6e05e TL |
7396 | } |
7397 | ||
7398 | InodeRef in; | |
7399 | filepath path(relpath); | |
7400 | r = path_walk(path, &in, perms, false, 0, dirinode); | |
7401 | if (r < 0) { | |
7402 | return r; | |
7403 | } | |
7c673cae FG |
7404 | |
7405 | return _readlink(in.get(), buf, size); | |
7406 | } | |
7407 | ||
7408 | int Client::_readlink(Inode *in, char *buf, size_t size) | |
7409 | { | |
7410 | if (!in->is_symlink()) | |
f67539c2 | 7411 | return -CEPHFS_EINVAL; |
7c673cae FG |
7412 | |
7413 | // copy into buf (at most size bytes) | |
7414 | int r = in->symlink.length(); | |
7415 | if (r > (int)size) | |
7416 | r = size; | |
7417 | memcpy(buf, in->symlink.c_str(), r); | |
7418 | return r; | |
7419 | } | |
7420 | ||
7421 | ||
7422 | // inode stuff | |
7423 | ||
7424 | int Client::_getattr(Inode *in, int mask, const UserPerm& perms, bool force) | |
7425 | { | |
94b18763 | 7426 | bool yes = in->caps_issued_mask(mask, true); |
7c673cae | 7427 | |
11fdf7f2 | 7428 | ldout(cct, 10) << __func__ << " mask " << ccap_string(mask) << " issued=" << yes << dendl; |
7c673cae FG |
7429 | if (yes && !force) |
7430 | return 0; | |
7431 | ||
7432 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
7433 | filepath path; | |
7434 | in->make_nosnap_relative_path(path); | |
7435 | req->set_filepath(path); | |
7436 | req->set_inode(in); | |
7437 | req->head.args.getattr.mask = mask; | |
7438 | ||
7439 | int res = make_request(req, perms); | |
11fdf7f2 | 7440 | ldout(cct, 10) << __func__ << " result=" << res << dendl; |
7c673cae FG |
7441 | return res; |
7442 | } | |
7443 | ||
7444 | int Client::_do_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
7445 | const UserPerm& perms, InodeRef *inp) | |
7446 | { | |
7447 | int issued = in->caps_issued(); | |
20effc67 TL |
7448 | union ceph_mds_request_args args; |
7449 | bool kill_sguid = false; | |
7450 | int inode_drop = 0; | |
7c673cae | 7451 | |
11fdf7f2 | 7452 | ldout(cct, 10) << __func__ << " mask " << mask << " issued " << |
7c673cae FG |
7453 | ccap_string(issued) << dendl; |
7454 | ||
7455 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 7456 | return -CEPHFS_EROFS; |
7c673cae FG |
7457 | } |
7458 | if ((mask & CEPH_SETATTR_SIZE) && | |
f67539c2 TL |
7459 | (uint64_t)stx->stx_size > in->size && |
7460 | is_quota_bytes_exceeded(in, (uint64_t)stx->stx_size - in->size, | |
7c673cae | 7461 | perms)) { |
f67539c2 | 7462 | return -CEPHFS_EDQUOT; |
7c673cae FG |
7463 | } |
7464 | ||
20effc67 TL |
7465 | memset(&args, 0, sizeof(args)); |
7466 | ||
7c673cae FG |
7467 | // make the change locally? |
7468 | if ((in->cap_dirtier_uid >= 0 && perms.uid() != in->cap_dirtier_uid) || | |
7469 | (in->cap_dirtier_gid >= 0 && perms.gid() != in->cap_dirtier_gid)) { | |
7470 | ldout(cct, 10) << __func__ << " caller " << perms.uid() << ":" << perms.gid() | |
7471 | << " != cap dirtier " << in->cap_dirtier_uid << ":" | |
7472 | << in->cap_dirtier_gid << ", forcing sync setattr" | |
7473 | << dendl; | |
7474 | /* | |
7475 | * This works because we implicitly flush the caps as part of the | |
7476 | * request, so the cap update check will happen with the writeback | |
7477 | * cap context, and then the setattr check will happen with the | |
7478 | * caller's context. | |
7479 | * | |
7480 | * In reality this pattern is likely pretty rare (different users | |
7481 | * setattr'ing the same file). If that turns out not to be the | |
7482 | * case later, we can build a more complex pipelined cap writeback | |
7483 | * infrastructure... | |
7484 | */ | |
20effc67 | 7485 | mask |= CEPH_SETATTR_CTIME; |
7c673cae FG |
7486 | } |
7487 | ||
7488 | if (!mask) { | |
7489 | // caller just needs us to bump the ctime | |
7490 | in->ctime = ceph_clock_now(); | |
7491 | in->cap_dirtier_uid = perms.uid(); | |
7492 | in->cap_dirtier_gid = perms.gid(); | |
7493 | if (issued & CEPH_CAP_AUTH_EXCL) | |
28e407b8 | 7494 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7495 | else if (issued & CEPH_CAP_FILE_EXCL) |
28e407b8 | 7496 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
7c673cae | 7497 | else if (issued & CEPH_CAP_XATTR_EXCL) |
28e407b8 | 7498 | in->mark_caps_dirty(CEPH_CAP_XATTR_EXCL); |
7c673cae FG |
7499 | else |
7500 | mask |= CEPH_SETATTR_CTIME; | |
7501 | } | |
7502 | ||
7503 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
20effc67 | 7504 | kill_sguid = mask & (CEPH_SETATTR_SIZE|CEPH_SETATTR_KILL_SGUID); |
7c673cae FG |
7505 | |
7506 | mask &= ~CEPH_SETATTR_KILL_SGUID; | |
20effc67 TL |
7507 | } else if (mask & CEPH_SETATTR_SIZE) { |
7508 | /* If we don't have Ax, then we must ask the server to clear them on truncate */ | |
7509 | mask |= CEPH_SETATTR_KILL_SGUID; | |
7510 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7511 | } | |
7512 | ||
7513 | if (mask & CEPH_SETATTR_UID) { | |
7514 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
7c673cae | 7515 | |
20effc67 | 7516 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { |
7c673cae FG |
7517 | in->ctime = ceph_clock_now(); |
7518 | in->cap_dirtier_uid = perms.uid(); | |
7519 | in->cap_dirtier_gid = perms.gid(); | |
7520 | in->uid = stx->stx_uid; | |
28e407b8 | 7521 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7522 | mask &= ~CEPH_SETATTR_UID; |
7523 | kill_sguid = true; | |
20effc67 TL |
7524 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7525 | in->uid != stx->stx_uid) { | |
7526 | args.setattr.uid = stx->stx_uid; | |
7527 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7528 | } else { | |
7529 | mask &= ~CEPH_SETATTR_UID; | |
7c673cae | 7530 | } |
20effc67 TL |
7531 | } |
7532 | ||
7533 | if (mask & CEPH_SETATTR_GID) { | |
7534 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
7535 | ||
7536 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
7537 | in->ctime = ceph_clock_now(); |
7538 | in->cap_dirtier_uid = perms.uid(); | |
7539 | in->cap_dirtier_gid = perms.gid(); | |
7540 | in->gid = stx->stx_gid; | |
28e407b8 | 7541 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7542 | mask &= ~CEPH_SETATTR_GID; |
7543 | kill_sguid = true; | |
20effc67 TL |
7544 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7545 | in->gid != stx->stx_gid) { | |
7546 | args.setattr.gid = stx->stx_gid; | |
7547 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7548 | } else { | |
7549 | mask &= ~CEPH_SETATTR_GID; | |
7c673cae | 7550 | } |
20effc67 | 7551 | } |
7c673cae | 7552 | |
20effc67 TL |
7553 | if (mask & CEPH_SETATTR_MODE) { |
7554 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
7555 | ||
7556 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
7557 | in->ctime = ceph_clock_now(); |
7558 | in->cap_dirtier_uid = perms.uid(); | |
7559 | in->cap_dirtier_gid = perms.gid(); | |
7560 | in->mode = (in->mode & ~07777) | (stx->stx_mode & 07777); | |
28e407b8 | 7561 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7562 | mask &= ~CEPH_SETATTR_MODE; |
20effc67 TL |
7563 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7564 | in->mode != stx->stx_mode) { | |
7565 | args.setattr.mode = stx->stx_mode; | |
7566 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7567 | } else { | |
7568 | mask &= ~CEPH_SETATTR_MODE; | |
7c673cae | 7569 | } |
20effc67 TL |
7570 | } else if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL) && |
7571 | kill_sguid && S_ISREG(in->mode) && | |
7572 | (in->mode & (S_IXUSR|S_IXGRP|S_IXOTH))) { | |
7573 | /* Must squash the any setuid/setgid bits with an ownership change */ | |
7574 | in->mode &= ~(S_ISUID|S_ISGID); | |
7575 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); | |
7576 | } | |
7577 | ||
7578 | if (mask & CEPH_SETATTR_BTIME) { | |
7579 | ldout(cct,10) << "changing btime to " << in->btime << dendl; | |
7c673cae | 7580 | |
20effc67 | 7581 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { |
7c673cae FG |
7582 | in->ctime = ceph_clock_now(); |
7583 | in->cap_dirtier_uid = perms.uid(); | |
7584 | in->cap_dirtier_gid = perms.gid(); | |
7585 | in->btime = utime_t(stx->stx_btime); | |
28e407b8 | 7586 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7587 | mask &= ~CEPH_SETATTR_BTIME; |
20effc67 TL |
7588 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7589 | in->btime != utime_t(stx->stx_btime)) { | |
7590 | args.setattr.btime = utime_t(stx->stx_btime); | |
7591 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7592 | } else { | |
7593 | mask &= ~CEPH_SETATTR_BTIME; | |
7594 | } | |
7595 | } | |
7596 | ||
7597 | if (mask & CEPH_SETATTR_SIZE) { | |
7598 | if ((uint64_t)stx->stx_size >= mdsmap->get_max_filesize()) { | |
7599 | //too big! | |
7600 | ldout(cct,10) << "unable to set size to " << stx->stx_size << ". Too large!" << dendl; | |
7601 | return -CEPHFS_EFBIG; | |
7602 | } | |
7603 | ||
7604 | ldout(cct,10) << "changing size to " << stx->stx_size << dendl; | |
7605 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL) && | |
7606 | !(mask & CEPH_SETATTR_KILL_SGUID) && | |
7607 | stx->stx_size >= in->size) { | |
7608 | if (stx->stx_size > in->size) { | |
7609 | in->size = in->reported_size = stx->stx_size; | |
7610 | in->cap_dirtier_uid = perms.uid(); | |
7611 | in->cap_dirtier_gid = perms.gid(); | |
7612 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); | |
7613 | mask &= ~(CEPH_SETATTR_SIZE); | |
7614 | mask |= CEPH_SETATTR_MTIME; | |
7615 | } else { | |
7616 | // ignore it when size doesn't change | |
7617 | mask &= ~(CEPH_SETATTR_SIZE); | |
7618 | } | |
7619 | } else { | |
7620 | args.setattr.size = stx->stx_size; | |
7621 | inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
7622 | CEPH_CAP_FILE_WR; | |
7623 | } | |
7624 | } | |
7625 | ||
7626 | if (mask & CEPH_SETATTR_MTIME) { | |
7627 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
7628 | in->mtime = utime_t(stx->stx_mtime); | |
7629 | in->ctime = ceph_clock_now(); | |
7630 | in->cap_dirtier_uid = perms.uid(); | |
7631 | in->cap_dirtier_gid = perms.gid(); | |
7632 | in->time_warp_seq++; | |
7633 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); | |
7634 | mask &= ~CEPH_SETATTR_MTIME; | |
7635 | } else if (in->caps_issued_mask(CEPH_CAP_FILE_WR) && | |
7636 | utime_t(stx->stx_mtime) > in->mtime) { | |
7637 | in->mtime = utime_t(stx->stx_mtime); | |
7638 | in->ctime = ceph_clock_now(); | |
7639 | in->cap_dirtier_uid = perms.uid(); | |
7640 | in->cap_dirtier_gid = perms.gid(); | |
7641 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); | |
7642 | mask &= ~CEPH_SETATTR_MTIME; | |
7643 | } else if (!in->caps_issued_mask(CEPH_CAP_FILE_SHARED) || | |
7644 | in->mtime != utime_t(stx->stx_mtime)) { | |
7645 | args.setattr.mtime = utime_t(stx->stx_mtime); | |
7646 | inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
7647 | CEPH_CAP_FILE_WR; | |
7648 | } else { | |
7649 | mask &= ~CEPH_SETATTR_MTIME; | |
7c673cae | 7650 | } |
7c673cae FG |
7651 | } |
7652 | ||
20effc67 TL |
7653 | if (mask & CEPH_SETATTR_ATIME) { |
7654 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
7655 | in->atime = utime_t(stx->stx_atime); | |
7c673cae FG |
7656 | in->ctime = ceph_clock_now(); |
7657 | in->cap_dirtier_uid = perms.uid(); | |
7658 | in->cap_dirtier_gid = perms.gid(); | |
7659 | in->time_warp_seq++; | |
28e407b8 | 7660 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
20effc67 TL |
7661 | mask &= ~CEPH_SETATTR_ATIME; |
7662 | } else if (in->caps_issued_mask(CEPH_CAP_FILE_WR) && | |
7663 | utime_t(stx->stx_atime) > in->atime) { | |
7664 | in->atime = utime_t(stx->stx_atime); | |
7665 | in->ctime = ceph_clock_now(); | |
7666 | in->cap_dirtier_uid = perms.uid(); | |
7667 | in->cap_dirtier_gid = perms.gid(); | |
7668 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); | |
7669 | mask &= ~CEPH_SETATTR_ATIME; | |
7670 | } else if (!in->caps_issued_mask(CEPH_CAP_FILE_SHARED) || | |
7671 | in->atime != utime_t(stx->stx_atime)) { | |
7672 | args.setattr.atime = utime_t(stx->stx_atime); | |
7673 | inode_drop |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
7674 | CEPH_CAP_FILE_WR; | |
7675 | } else { | |
7676 | mask &= ~CEPH_SETATTR_ATIME; | |
7c673cae FG |
7677 | } |
7678 | } | |
20effc67 | 7679 | |
7c673cae FG |
7680 | if (!mask) { |
7681 | in->change_attr++; | |
7682 | return 0; | |
7683 | } | |
7684 | ||
7c673cae FG |
7685 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETATTR); |
7686 | ||
7687 | filepath path; | |
7688 | ||
7689 | in->make_nosnap_relative_path(path); | |
7690 | req->set_filepath(path); | |
7691 | req->set_inode(in); | |
7692 | ||
20effc67 TL |
7693 | req->head.args = args; |
7694 | req->inode_drop = inode_drop; | |
7c673cae | 7695 | req->head.args.setattr.mask = mask; |
7c673cae FG |
7696 | req->regetattr_mask = mask; |
7697 | ||
7698 | int res = make_request(req, perms, inp); | |
7699 | ldout(cct, 10) << "_setattr result=" << res << dendl; | |
7700 | return res; | |
7701 | } | |
7702 | ||
7703 | /* Note that we only care about attrs that setattr cares about */ | |
7704 | void Client::stat_to_statx(struct stat *st, struct ceph_statx *stx) | |
7705 | { | |
7706 | stx->stx_size = st->st_size; | |
7707 | stx->stx_mode = st->st_mode; | |
7708 | stx->stx_uid = st->st_uid; | |
7709 | stx->stx_gid = st->st_gid; | |
11fdf7f2 TL |
7710 | #ifdef __APPLE__ |
7711 | stx->stx_mtime = st->st_mtimespec; | |
7712 | stx->stx_atime = st->st_atimespec; | |
f67539c2 TL |
7713 | #elif __WIN32 |
7714 | stx->stx_mtime.tv_sec = st->st_mtime; | |
7715 | stx->stx_atime.tv_sec = st->st_atime; | |
11fdf7f2 | 7716 | #else |
7c673cae FG |
7717 | stx->stx_mtime = st->st_mtim; |
7718 | stx->stx_atime = st->st_atim; | |
11fdf7f2 | 7719 | #endif |
7c673cae FG |
7720 | } |
7721 | ||
7722 | int Client::__setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
7723 | const UserPerm& perms, InodeRef *inp) | |
7724 | { | |
7725 | int ret = _do_setattr(in, stx, mask, perms, inp); | |
7726 | if (ret < 0) | |
7727 | return ret; | |
7728 | if (mask & CEPH_SETATTR_MODE) | |
7729 | ret = _posix_acl_chmod(in, stx->stx_mode, perms); | |
7730 | return ret; | |
7731 | } | |
7732 | ||
7733 | int Client::_setattrx(InodeRef &in, struct ceph_statx *stx, int mask, | |
7734 | const UserPerm& perms) | |
7735 | { | |
7736 | mask &= (CEPH_SETATTR_MODE | CEPH_SETATTR_UID | | |
7737 | CEPH_SETATTR_GID | CEPH_SETATTR_MTIME | | |
7738 | CEPH_SETATTR_ATIME | CEPH_SETATTR_SIZE | | |
7739 | CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME); | |
7740 | if (cct->_conf->client_permissions) { | |
7741 | int r = may_setattr(in.get(), stx, mask, perms); | |
7742 | if (r < 0) | |
7743 | return r; | |
7744 | } | |
7745 | return __setattrx(in.get(), stx, mask, perms); | |
7746 | } | |
7747 | ||
7748 | int Client::_setattr(InodeRef &in, struct stat *attr, int mask, | |
7749 | const UserPerm& perms) | |
7750 | { | |
7751 | struct ceph_statx stx; | |
7752 | ||
7753 | stat_to_statx(attr, &stx); | |
7754 | mask &= ~CEPH_SETATTR_BTIME; | |
181888fb FG |
7755 | |
7756 | if ((mask & CEPH_SETATTR_UID) && attr->st_uid == static_cast<uid_t>(-1)) { | |
7757 | mask &= ~CEPH_SETATTR_UID; | |
7758 | } | |
7759 | if ((mask & CEPH_SETATTR_GID) && attr->st_gid == static_cast<uid_t>(-1)) { | |
7760 | mask &= ~CEPH_SETATTR_GID; | |
7761 | } | |
7762 | ||
7c673cae FG |
7763 | return _setattrx(in, &stx, mask, perms); |
7764 | } | |
7765 | ||
7766 | int Client::setattr(const char *relpath, struct stat *attr, int mask, | |
7767 | const UserPerm& perms) | |
7768 | { | |
f67539c2 TL |
7769 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7770 | if (!mref_reader.is_state_satisfied()) | |
7771 | return -CEPHFS_ENOTCONN; | |
7772 | ||
11fdf7f2 | 7773 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7774 | tout(cct) << relpath << std::endl; |
7775 | tout(cct) << mask << std::endl; | |
7776 | ||
7777 | filepath path(relpath); | |
7778 | InodeRef in; | |
f67539c2 TL |
7779 | |
7780 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7781 | int r = path_walk(path, &in, perms); |
7782 | if (r < 0) | |
7783 | return r; | |
7784 | return _setattr(in, attr, mask, perms); | |
7785 | } | |
7786 | ||
7787 | int Client::setattrx(const char *relpath, struct ceph_statx *stx, int mask, | |
7788 | const UserPerm& perms, int flags) | |
7789 | { | |
f67539c2 TL |
7790 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7791 | if (!mref_reader.is_state_satisfied()) | |
7792 | return -CEPHFS_ENOTCONN; | |
7793 | ||
11fdf7f2 | 7794 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7795 | tout(cct) << relpath << std::endl; |
7796 | tout(cct) << mask << std::endl; | |
7797 | ||
7798 | filepath path(relpath); | |
7799 | InodeRef in; | |
f67539c2 TL |
7800 | |
7801 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7802 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW)); |
7803 | if (r < 0) | |
7804 | return r; | |
7805 | return _setattrx(in, stx, mask, perms); | |
7806 | } | |
7807 | ||
7808 | int Client::fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms) | |
7809 | { | |
f67539c2 TL |
7810 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7811 | if (!mref_reader.is_state_satisfied()) | |
7812 | return -CEPHFS_ENOTCONN; | |
7813 | ||
11fdf7f2 | 7814 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7815 | tout(cct) << fd << std::endl; |
7816 | tout(cct) << mask << std::endl; | |
7817 | ||
f67539c2 | 7818 | std::scoped_lock lock(client_lock); |
7c673cae FG |
7819 | Fh *f = get_filehandle(fd); |
7820 | if (!f) | |
f67539c2 | 7821 | return -CEPHFS_EBADF; |
7c673cae FG |
7822 | #if defined(__linux__) && defined(O_PATH) |
7823 | if (f->flags & O_PATH) | |
f67539c2 | 7824 | return -CEPHFS_EBADF; |
7c673cae FG |
7825 | #endif |
7826 | return _setattr(f->inode, attr, mask, perms); | |
7827 | } | |
7828 | ||
7829 | int Client::fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms) | |
7830 | { | |
f67539c2 TL |
7831 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7832 | if (!mref_reader.is_state_satisfied()) | |
7833 | return -CEPHFS_ENOTCONN; | |
7834 | ||
11fdf7f2 | 7835 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7836 | tout(cct) << fd << std::endl; |
7837 | tout(cct) << mask << std::endl; | |
7838 | ||
f67539c2 | 7839 | std::scoped_lock lock(client_lock); |
7c673cae FG |
7840 | Fh *f = get_filehandle(fd); |
7841 | if (!f) | |
f67539c2 | 7842 | return -CEPHFS_EBADF; |
7c673cae FG |
7843 | #if defined(__linux__) && defined(O_PATH) |
7844 | if (f->flags & O_PATH) | |
f67539c2 | 7845 | return -CEPHFS_EBADF; |
7c673cae FG |
7846 | #endif |
7847 | return _setattrx(f->inode, stx, mask, perms); | |
7848 | } | |
7849 | ||
7850 | int Client::stat(const char *relpath, struct stat *stbuf, const UserPerm& perms, | |
7851 | frag_info_t *dirstat, int mask) | |
7852 | { | |
f67539c2 TL |
7853 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7854 | if (!mref_reader.is_state_satisfied()) | |
7855 | return -CEPHFS_ENOTCONN; | |
7856 | ||
11fdf7f2 | 7857 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7858 | tout(cct) << "stat" << std::endl; |
7859 | tout(cct) << relpath << std::endl; | |
181888fb | 7860 | |
7c673cae FG |
7861 | filepath path(relpath); |
7862 | InodeRef in; | |
f67539c2 TL |
7863 | |
7864 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7865 | int r = path_walk(path, &in, perms, true, mask); |
7866 | if (r < 0) | |
7867 | return r; | |
7868 | r = _getattr(in, mask, perms); | |
7869 | if (r < 0) { | |
11fdf7f2 | 7870 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7871 | return r; |
7872 | } | |
7873 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 7874 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7875 | return r; |
7876 | } | |
7877 | ||
7878 | unsigned Client::statx_to_mask(unsigned int flags, unsigned int want) | |
7879 | { | |
7880 | unsigned mask = 0; | |
7881 | ||
7882 | /* if NO_ATTR_SYNC is set, then we don't need any -- just use what's in cache */ | |
7883 | if (flags & AT_NO_ATTR_SYNC) | |
7884 | goto out; | |
7885 | ||
7886 | /* Always set PIN to distinguish from AT_NO_ATTR_SYNC case */ | |
7887 | mask |= CEPH_CAP_PIN; | |
7888 | if (want & (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7889 | mask |= CEPH_CAP_AUTH_SHARED; | |
7890 | if (want & (CEPH_STATX_NLINK|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7891 | mask |= CEPH_CAP_LINK_SHARED; | |
adb31ebb | 7892 | 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 |
7893 | mask |= CEPH_CAP_FILE_SHARED; |
7894 | if (want & (CEPH_STATX_VERSION|CEPH_STATX_CTIME)) | |
7895 | mask |= CEPH_CAP_XATTR_SHARED; | |
7896 | out: | |
7897 | return mask; | |
7898 | } | |
7899 | ||
7900 | int Client::statx(const char *relpath, struct ceph_statx *stx, | |
7901 | const UserPerm& perms, | |
7902 | unsigned int want, unsigned int flags) | |
7903 | { | |
b3b6e05e | 7904 | return statxat(CEPHFS_AT_FDCWD, relpath, stx, perms, want, flags); |
7c673cae FG |
7905 | } |
7906 | ||
7907 | int Client::lstat(const char *relpath, struct stat *stbuf, | |
7908 | const UserPerm& perms, frag_info_t *dirstat, int mask) | |
7909 | { | |
f67539c2 TL |
7910 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7911 | if (!mref_reader.is_state_satisfied()) | |
7912 | return -CEPHFS_ENOTCONN; | |
7913 | ||
11fdf7f2 | 7914 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
11fdf7f2 | 7915 | tout(cct) << __func__ << std::endl; |
7c673cae | 7916 | tout(cct) << relpath << std::endl; |
181888fb | 7917 | |
7c673cae FG |
7918 | filepath path(relpath); |
7919 | InodeRef in; | |
f67539c2 TL |
7920 | |
7921 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7922 | // don't follow symlinks |
7923 | int r = path_walk(path, &in, perms, false, mask); | |
7924 | if (r < 0) | |
7925 | return r; | |
7926 | r = _getattr(in, mask, perms); | |
7927 | if (r < 0) { | |
11fdf7f2 | 7928 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7929 | return r; |
7930 | } | |
7931 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 7932 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7933 | return r; |
7934 | } | |
7935 | ||
7936 | int Client::fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat, nest_info_t *rstat) | |
7937 | { | |
11fdf7f2 | 7938 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
7939 | << " mode 0" << oct << in->mode << dec |
7940 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
7941 | memset(st, 0, sizeof(struct stat)); | |
7942 | if (use_faked_inos()) | |
7943 | st->st_ino = in->faked_ino; | |
7944 | else | |
7945 | st->st_ino = in->ino; | |
7946 | st->st_dev = in->snapid; | |
7947 | st->st_mode = in->mode; | |
7948 | st->st_rdev = in->rdev; | |
28e407b8 AA |
7949 | if (in->is_dir()) { |
7950 | switch (in->nlink) { | |
7951 | case 0: | |
7952 | st->st_nlink = 0; /* dir is unlinked */ | |
7953 | break; | |
7954 | case 1: | |
7955 | st->st_nlink = 1 /* parent dentry */ | |
7956 | + 1 /* <dir>/. */ | |
7957 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
7958 | break; | |
7959 | default: | |
7960 | ceph_abort(); | |
7961 | } | |
7962 | } else { | |
7963 | st->st_nlink = in->nlink; | |
7964 | } | |
7c673cae FG |
7965 | st->st_uid = in->uid; |
7966 | st->st_gid = in->gid; | |
7967 | if (in->ctime > in->mtime) { | |
7968 | stat_set_ctime_sec(st, in->ctime.sec()); | |
7969 | stat_set_ctime_nsec(st, in->ctime.nsec()); | |
7970 | } else { | |
7971 | stat_set_ctime_sec(st, in->mtime.sec()); | |
7972 | stat_set_ctime_nsec(st, in->mtime.nsec()); | |
7973 | } | |
7974 | stat_set_atime_sec(st, in->atime.sec()); | |
7975 | stat_set_atime_nsec(st, in->atime.nsec()); | |
7976 | stat_set_mtime_sec(st, in->mtime.sec()); | |
7977 | stat_set_mtime_nsec(st, in->mtime.nsec()); | |
7978 | if (in->is_dir()) { | |
7979 | if (cct->_conf->client_dirsize_rbytes) | |
7980 | st->st_size = in->rstat.rbytes; | |
7981 | else | |
7982 | st->st_size = in->dirstat.size(); | |
f67539c2 TL |
7983 | // The Windows "stat" structure provides just a subset of the fields that are |
7984 | // available on Linux. | |
7985 | #ifndef _WIN32 | |
7c673cae | 7986 | st->st_blocks = 1; |
f67539c2 | 7987 | #endif |
7c673cae FG |
7988 | } else { |
7989 | st->st_size = in->size; | |
f67539c2 | 7990 | #ifndef _WIN32 |
7c673cae | 7991 | st->st_blocks = (in->size + 511) >> 9; |
f67539c2 | 7992 | #endif |
7c673cae | 7993 | } |
f67539c2 | 7994 | #ifndef _WIN32 |
11fdf7f2 | 7995 | st->st_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
f67539c2 | 7996 | #endif |
7c673cae FG |
7997 | |
7998 | if (dirstat) | |
7999 | *dirstat = in->dirstat; | |
8000 | if (rstat) | |
8001 | *rstat = in->rstat; | |
8002 | ||
8003 | return in->caps_issued(); | |
8004 | } | |
8005 | ||
8006 | void Client::fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx) | |
8007 | { | |
11fdf7f2 | 8008 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
8009 | << " mode 0" << oct << in->mode << dec |
8010 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
8011 | memset(stx, 0, sizeof(struct ceph_statx)); | |
8012 | ||
8013 | /* | |
8014 | * If mask is 0, then the caller set AT_NO_ATTR_SYNC. Reset the mask | |
8015 | * so that all bits are set. | |
8016 | */ | |
8017 | if (!mask) | |
8018 | mask = ~0; | |
8019 | ||
8020 | /* These are always considered to be available */ | |
8021 | stx->stx_dev = in->snapid; | |
11fdf7f2 | 8022 | stx->stx_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
7c673cae FG |
8023 | |
8024 | /* Type bits are always set, even when CEPH_STATX_MODE is not */ | |
8025 | stx->stx_mode = S_IFMT & in->mode; | |
8026 | stx->stx_ino = use_faked_inos() ? in->faked_ino : (ino_t)in->ino; | |
8027 | stx->stx_rdev = in->rdev; | |
8028 | stx->stx_mask |= (CEPH_STATX_INO|CEPH_STATX_RDEV); | |
8029 | ||
8030 | if (mask & CEPH_CAP_AUTH_SHARED) { | |
8031 | stx->stx_uid = in->uid; | |
8032 | stx->stx_gid = in->gid; | |
8033 | stx->stx_mode = in->mode; | |
8034 | in->btime.to_timespec(&stx->stx_btime); | |
8035 | stx->stx_mask |= (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME); | |
8036 | } | |
8037 | ||
8038 | if (mask & CEPH_CAP_LINK_SHARED) { | |
28e407b8 AA |
8039 | if (in->is_dir()) { |
8040 | switch (in->nlink) { | |
8041 | case 0: | |
8042 | stx->stx_nlink = 0; /* dir is unlinked */ | |
8043 | break; | |
8044 | case 1: | |
8045 | stx->stx_nlink = 1 /* parent dentry */ | |
8046 | + 1 /* <dir>/. */ | |
8047 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
8048 | break; | |
8049 | default: | |
8050 | ceph_abort(); | |
8051 | } | |
8052 | } else { | |
8053 | stx->stx_nlink = in->nlink; | |
8054 | } | |
7c673cae FG |
8055 | stx->stx_mask |= CEPH_STATX_NLINK; |
8056 | } | |
8057 | ||
8058 | if (mask & CEPH_CAP_FILE_SHARED) { | |
8059 | ||
8060 | in->atime.to_timespec(&stx->stx_atime); | |
8061 | in->mtime.to_timespec(&stx->stx_mtime); | |
8062 | ||
8063 | if (in->is_dir()) { | |
8064 | if (cct->_conf->client_dirsize_rbytes) | |
8065 | stx->stx_size = in->rstat.rbytes; | |
8066 | else | |
8067 | stx->stx_size = in->dirstat.size(); | |
8068 | stx->stx_blocks = 1; | |
8069 | } else { | |
8070 | stx->stx_size = in->size; | |
8071 | stx->stx_blocks = (in->size + 511) >> 9; | |
8072 | } | |
8073 | stx->stx_mask |= (CEPH_STATX_ATIME|CEPH_STATX_MTIME| | |
8074 | CEPH_STATX_SIZE|CEPH_STATX_BLOCKS); | |
8075 | } | |
8076 | ||
8077 | /* Change time and change_attr both require all shared caps to view */ | |
8078 | if ((mask & CEPH_STAT_CAP_INODE_ALL) == CEPH_STAT_CAP_INODE_ALL) { | |
8079 | stx->stx_version = in->change_attr; | |
8080 | if (in->ctime > in->mtime) | |
8081 | in->ctime.to_timespec(&stx->stx_ctime); | |
8082 | else | |
8083 | in->mtime.to_timespec(&stx->stx_ctime); | |
8084 | stx->stx_mask |= (CEPH_STATX_CTIME|CEPH_STATX_VERSION); | |
8085 | } | |
8086 | ||
8087 | } | |
8088 | ||
8089 | void Client::touch_dn(Dentry *dn) | |
8090 | { | |
8091 | lru.lru_touch(dn); | |
8092 | } | |
8093 | ||
8094 | int Client::chmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
8095 | { | |
b3b6e05e | 8096 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, 0, perms); |
7c673cae FG |
8097 | } |
8098 | ||
8099 | int Client::fchmod(int fd, mode_t mode, const UserPerm& perms) | |
8100 | { | |
f67539c2 TL |
8101 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8102 | if (!mref_reader.is_state_satisfied()) | |
8103 | return -CEPHFS_ENOTCONN; | |
8104 | ||
11fdf7f2 | 8105 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8106 | tout(cct) << fd << std::endl; |
8107 | tout(cct) << mode << std::endl; | |
181888fb | 8108 | |
f67539c2 | 8109 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8110 | Fh *f = get_filehandle(fd); |
8111 | if (!f) | |
f67539c2 | 8112 | return -CEPHFS_EBADF; |
7c673cae FG |
8113 | #if defined(__linux__) && defined(O_PATH) |
8114 | if (f->flags & O_PATH) | |
f67539c2 | 8115 | return -CEPHFS_EBADF; |
7c673cae FG |
8116 | #endif |
8117 | struct stat attr; | |
8118 | attr.st_mode = mode; | |
8119 | return _setattr(f->inode, &attr, CEPH_SETATTR_MODE, perms); | |
8120 | } | |
8121 | ||
b3b6e05e TL |
8122 | int Client::chmodat(int dirfd, const char *relpath, mode_t mode, int flags, |
8123 | const UserPerm& perms) { | |
f67539c2 | 8124 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8125 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8126 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8127 | } |
f67539c2 | 8128 | |
11fdf7f2 | 8129 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8130 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8131 | tout(cct) << relpath << std::endl; |
8132 | tout(cct) << mode << std::endl; | |
b3b6e05e | 8133 | tout(cct) << flags << std::endl; |
181888fb | 8134 | |
7c673cae FG |
8135 | filepath path(relpath); |
8136 | InodeRef in; | |
b3b6e05e | 8137 | InodeRef dirinode; |
f67539c2 TL |
8138 | |
8139 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8140 | int r = get_fd_inode(dirfd, &dirinode); |
8141 | if (r < 0) { | |
8142 | return r; | |
8143 | } | |
8144 | ||
8145 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8146 | if (r < 0) { | |
7c673cae | 8147 | return r; |
b3b6e05e | 8148 | } |
7c673cae FG |
8149 | struct stat attr; |
8150 | attr.st_mode = mode; | |
8151 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
8152 | } | |
8153 | ||
b3b6e05e TL |
8154 | int Client::lchmod(const char *relpath, mode_t mode, const UserPerm& perms) |
8155 | { | |
8156 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, AT_SYMLINK_NOFOLLOW, perms); | |
8157 | } | |
8158 | ||
7c673cae FG |
8159 | int Client::chown(const char *relpath, uid_t new_uid, gid_t new_gid, |
8160 | const UserPerm& perms) | |
8161 | { | |
b3b6e05e | 8162 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, 0, perms); |
7c673cae FG |
8163 | } |
8164 | ||
8165 | int Client::fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms) | |
8166 | { | |
f67539c2 TL |
8167 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8168 | if (!mref_reader.is_state_satisfied()) | |
8169 | return -CEPHFS_ENOTCONN; | |
8170 | ||
11fdf7f2 | 8171 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8172 | tout(cct) << fd << std::endl; |
8173 | tout(cct) << new_uid << std::endl; | |
8174 | tout(cct) << new_gid << std::endl; | |
181888fb | 8175 | |
f67539c2 | 8176 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8177 | Fh *f = get_filehandle(fd); |
8178 | if (!f) | |
f67539c2 | 8179 | return -CEPHFS_EBADF; |
7c673cae FG |
8180 | #if defined(__linux__) && defined(O_PATH) |
8181 | if (f->flags & O_PATH) | |
f67539c2 | 8182 | return -CEPHFS_EBADF; |
7c673cae FG |
8183 | #endif |
8184 | struct stat attr; | |
8185 | attr.st_uid = new_uid; | |
8186 | attr.st_gid = new_gid; | |
8187 | int mask = 0; | |
8188 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
8189 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
8190 | return _setattr(f->inode, &attr, mask, perms); | |
8191 | } | |
8192 | ||
8193 | int Client::lchown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
8194 | const UserPerm& perms) | |
8195 | { | |
b3b6e05e TL |
8196 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, AT_SYMLINK_NOFOLLOW, perms); |
8197 | } | |
8198 | ||
8199 | int Client::chownat(int dirfd, const char *relpath, uid_t new_uid, gid_t new_gid, | |
8200 | int flags, const UserPerm& perms) { | |
f67539c2 | 8201 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8202 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8203 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8204 | } |
f67539c2 | 8205 | |
11fdf7f2 | 8206 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8207 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8208 | tout(cct) << relpath << std::endl; |
8209 | tout(cct) << new_uid << std::endl; | |
8210 | tout(cct) << new_gid << std::endl; | |
b3b6e05e | 8211 | tout(cct) << flags << std::endl; |
181888fb | 8212 | |
7c673cae FG |
8213 | filepath path(relpath); |
8214 | InodeRef in; | |
b3b6e05e | 8215 | InodeRef dirinode; |
f67539c2 TL |
8216 | |
8217 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8218 | int r = get_fd_inode(dirfd, &dirinode); |
8219 | if (r < 0) { | |
7c673cae | 8220 | return r; |
b3b6e05e TL |
8221 | } |
8222 | ||
8223 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8224 | if (r < 0) { | |
8225 | return r; | |
8226 | } | |
7c673cae FG |
8227 | struct stat attr; |
8228 | attr.st_uid = new_uid; | |
8229 | attr.st_gid = new_gid; | |
b3b6e05e | 8230 | return _setattr(in, &attr, CEPH_SETATTR_UID|CEPH_SETATTR_GID, perms); |
7c673cae FG |
8231 | } |
8232 | ||
11fdf7f2 TL |
8233 | static void attr_set_atime_and_mtime(struct stat *attr, |
8234 | const utime_t &atime, | |
8235 | const utime_t &mtime) | |
8236 | { | |
8237 | stat_set_atime_sec(attr, atime.tv.tv_sec); | |
8238 | stat_set_atime_nsec(attr, atime.tv.tv_nsec); | |
8239 | stat_set_mtime_sec(attr, mtime.tv.tv_sec); | |
8240 | stat_set_mtime_nsec(attr, mtime.tv.tv_nsec); | |
8241 | } | |
8242 | ||
8243 | // for [l]utime() invoke the timeval variant as the timespec | |
8244 | // variant are not yet implemented. for futime[s](), invoke | |
8245 | // the timespec variant. | |
7c673cae FG |
8246 | int Client::utime(const char *relpath, struct utimbuf *buf, |
8247 | const UserPerm& perms) | |
8248 | { | |
11fdf7f2 TL |
8249 | struct timeval tv[2]; |
8250 | tv[0].tv_sec = buf->actime; | |
8251 | tv[0].tv_usec = 0; | |
8252 | tv[1].tv_sec = buf->modtime; | |
8253 | tv[1].tv_usec = 0; | |
8254 | ||
8255 | return utimes(relpath, tv, perms); | |
8256 | } | |
8257 | ||
8258 | int Client::lutime(const char *relpath, struct utimbuf *buf, | |
8259 | const UserPerm& perms) | |
8260 | { | |
8261 | struct timeval tv[2]; | |
8262 | tv[0].tv_sec = buf->actime; | |
8263 | tv[0].tv_usec = 0; | |
8264 | tv[1].tv_sec = buf->modtime; | |
8265 | tv[1].tv_usec = 0; | |
8266 | ||
8267 | return lutimes(relpath, tv, perms); | |
8268 | } | |
8269 | ||
8270 | int Client::futime(int fd, struct utimbuf *buf, const UserPerm& perms) | |
8271 | { | |
8272 | struct timespec ts[2]; | |
8273 | ts[0].tv_sec = buf->actime; | |
8274 | ts[0].tv_nsec = 0; | |
8275 | ts[1].tv_sec = buf->modtime; | |
8276 | ts[1].tv_nsec = 0; | |
8277 | ||
8278 | return futimens(fd, ts, perms); | |
8279 | } | |
8280 | ||
8281 | int Client::utimes(const char *relpath, struct timeval times[2], | |
8282 | const UserPerm& perms) | |
8283 | { | |
f67539c2 TL |
8284 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8285 | if (!mref_reader.is_state_satisfied()) | |
8286 | return -CEPHFS_ENOTCONN; | |
8287 | ||
11fdf7f2 | 8288 | tout(cct) << __func__ << std::endl; |
7c673cae | 8289 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8290 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8291 | << std::endl; | |
8292 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8293 | << std::endl; | |
181888fb | 8294 | |
7c673cae FG |
8295 | filepath path(relpath); |
8296 | InodeRef in; | |
f67539c2 TL |
8297 | |
8298 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8299 | int r = path_walk(path, &in, perms); |
8300 | if (r < 0) | |
8301 | return r; | |
8302 | struct stat attr; | |
11fdf7f2 TL |
8303 | utime_t atime(times[0]); |
8304 | utime_t mtime(times[1]); | |
8305 | ||
8306 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8307 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8308 | } | |
8309 | ||
11fdf7f2 TL |
8310 | int Client::lutimes(const char *relpath, struct timeval times[2], |
8311 | const UserPerm& perms) | |
7c673cae | 8312 | { |
f67539c2 TL |
8313 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8314 | if (!mref_reader.is_state_satisfied()) | |
8315 | return -CEPHFS_ENOTCONN; | |
8316 | ||
11fdf7f2 | 8317 | tout(cct) << __func__ << std::endl; |
7c673cae | 8318 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8319 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8320 | << std::endl; | |
8321 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8322 | << std::endl; | |
181888fb | 8323 | |
7c673cae FG |
8324 | filepath path(relpath); |
8325 | InodeRef in; | |
f67539c2 TL |
8326 | |
8327 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8328 | int r = path_walk(path, &in, perms, false); |
8329 | if (r < 0) | |
8330 | return r; | |
8331 | struct stat attr; | |
11fdf7f2 TL |
8332 | utime_t atime(times[0]); |
8333 | utime_t mtime(times[1]); | |
8334 | ||
8335 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8336 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8337 | } | |
8338 | ||
11fdf7f2 TL |
8339 | int Client::futimes(int fd, struct timeval times[2], const UserPerm& perms) |
8340 | { | |
8341 | struct timespec ts[2]; | |
8342 | ts[0].tv_sec = times[0].tv_sec; | |
8343 | ts[0].tv_nsec = times[0].tv_usec * 1000; | |
8344 | ts[1].tv_sec = times[1].tv_sec; | |
8345 | ts[1].tv_nsec = times[1].tv_usec * 1000; | |
8346 | ||
8347 | return futimens(fd, ts, perms); | |
8348 | } | |
8349 | ||
8350 | int Client::futimens(int fd, struct timespec times[2], const UserPerm& perms) | |
8351 | { | |
f67539c2 TL |
8352 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8353 | if (!mref_reader.is_state_satisfied()) | |
8354 | return -CEPHFS_ENOTCONN; | |
8355 | ||
11fdf7f2 TL |
8356 | tout(cct) << __func__ << std::endl; |
8357 | tout(cct) << fd << std::endl; | |
8358 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8359 | << std::endl; | |
8360 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8361 | << std::endl; | |
8362 | ||
f67539c2 | 8363 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
8364 | Fh *f = get_filehandle(fd); |
8365 | if (!f) | |
f67539c2 | 8366 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8367 | #if defined(__linux__) && defined(O_PATH) |
8368 | if (f->flags & O_PATH) | |
f67539c2 | 8369 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8370 | #endif |
8371 | struct stat attr; | |
8372 | utime_t atime(times[0]); | |
8373 | utime_t mtime(times[1]); | |
8374 | ||
8375 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8376 | return _setattr(f->inode, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8377 | } | |
8378 | ||
b3b6e05e TL |
8379 | int Client::utimensat(int dirfd, const char *relpath, struct timespec times[2], int flags, |
8380 | const UserPerm& perms) { | |
8381 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8382 | if (!mref_reader.is_state_satisfied()) { | |
8383 | return -CEPHFS_ENOTCONN; | |
8384 | } | |
8385 | ||
8386 | tout(cct) << __func__ << std::endl; | |
8387 | tout(cct) << dirfd << std::endl; | |
8388 | tout(cct) << relpath << std::endl; | |
8389 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8390 | << std::endl; | |
8391 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8392 | << std::endl; | |
8393 | tout(cct) << flags << std::endl; | |
8394 | ||
8395 | filepath path(relpath); | |
8396 | InodeRef in; | |
8397 | InodeRef dirinode; | |
8398 | ||
8399 | std::scoped_lock lock(client_lock); | |
8400 | int r = get_fd_inode(dirfd, &dirinode); | |
8401 | if (r < 0) { | |
8402 | return r; | |
8403 | } | |
8404 | ||
8405 | #if defined(__linux__) && defined(O_PATH) | |
8406 | if (flags & O_PATH) { | |
8407 | return -CEPHFS_EBADF; | |
8408 | } | |
8409 | #endif | |
8410 | ||
8411 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8412 | if (r < 0) { | |
8413 | return r; | |
8414 | } | |
8415 | struct stat attr; | |
8416 | utime_t atime(times[0]); | |
8417 | utime_t mtime(times[1]); | |
8418 | ||
8419 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8420 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8421 | } | |
8422 | ||
7c673cae FG |
8423 | int Client::flock(int fd, int operation, uint64_t owner) |
8424 | { | |
f67539c2 TL |
8425 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8426 | if (!mref_reader.is_state_satisfied()) | |
8427 | return -CEPHFS_ENOTCONN; | |
8428 | ||
11fdf7f2 | 8429 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8430 | tout(cct) << fd << std::endl; |
8431 | tout(cct) << operation << std::endl; | |
8432 | tout(cct) << owner << std::endl; | |
181888fb | 8433 | |
f67539c2 | 8434 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8435 | Fh *f = get_filehandle(fd); |
8436 | if (!f) | |
f67539c2 | 8437 | return -CEPHFS_EBADF; |
7c673cae FG |
8438 | |
8439 | return _flock(f, operation, owner); | |
8440 | } | |
8441 | ||
8442 | int Client::opendir(const char *relpath, dir_result_t **dirpp, const UserPerm& perms) | |
8443 | { | |
f67539c2 TL |
8444 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8445 | if (!mref_reader.is_state_satisfied()) | |
8446 | return -CEPHFS_ENOTCONN; | |
8447 | ||
11fdf7f2 | 8448 | tout(cct) << __func__ << std::endl; |
7c673cae | 8449 | tout(cct) << relpath << std::endl; |
181888fb | 8450 | |
7c673cae FG |
8451 | filepath path(relpath); |
8452 | InodeRef in; | |
f67539c2 TL |
8453 | |
8454 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8455 | int r = path_walk(path, &in, perms, true); |
8456 | if (r < 0) | |
8457 | return r; | |
8458 | if (cct->_conf->client_permissions) { | |
8459 | int r = may_open(in.get(), O_RDONLY, perms); | |
8460 | if (r < 0) | |
8461 | return r; | |
8462 | } | |
8463 | r = _opendir(in.get(), dirpp, perms); | |
8464 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
f67539c2 TL |
8465 | if (r != -CEPHFS_ENOTDIR) |
8466 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
7c673cae FG |
8467 | return r; |
8468 | } | |
8469 | ||
b3b6e05e TL |
8470 | int Client::fdopendir(int dirfd, dir_result_t **dirpp, const UserPerm &perms) { |
8471 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8472 | if (!mref_reader.is_state_satisfied()) { | |
8473 | return -CEPHFS_ENOTCONN; | |
8474 | } | |
8475 | ||
8476 | tout(cct) << __func__ << std::endl; | |
8477 | tout(cct) << dirfd << std::endl; | |
8478 | ||
8479 | InodeRef dirinode; | |
8480 | std::scoped_lock locker(client_lock); | |
8481 | int r = get_fd_inode(dirfd, &dirinode); | |
8482 | if (r < 0) { | |
8483 | return r; | |
8484 | } | |
8485 | ||
8486 | if (cct->_conf->client_permissions) { | |
8487 | r = may_open(dirinode.get(), O_RDONLY, perms); | |
8488 | if (r < 0) { | |
8489 | return r; | |
8490 | } | |
8491 | } | |
8492 | r = _opendir(dirinode.get(), dirpp, perms); | |
8493 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
8494 | if (r != -CEPHFS_ENOTDIR) { | |
8495 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
8496 | } | |
8497 | return r; | |
8498 | } | |
8499 | ||
7c673cae FG |
8500 | int Client::_opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms) |
8501 | { | |
8502 | if (!in->is_dir()) | |
f67539c2 | 8503 | return -CEPHFS_ENOTDIR; |
7c673cae FG |
8504 | *dirpp = new dir_result_t(in, perms); |
8505 | opened_dirs.insert(*dirpp); | |
11fdf7f2 | 8506 | ldout(cct, 8) << __func__ << "(" << in->ino << ") = " << 0 << " (" << *dirpp << ")" << dendl; |
7c673cae FG |
8507 | return 0; |
8508 | } | |
8509 | ||
8510 | ||
8511 | int Client::closedir(dir_result_t *dir) | |
8512 | { | |
11fdf7f2 | 8513 | tout(cct) << __func__ << std::endl; |
f67539c2 | 8514 | tout(cct) << (uintptr_t)dir << std::endl; |
7c673cae | 8515 | |
11fdf7f2 | 8516 | ldout(cct, 3) << __func__ << "(" << dir << ") = 0" << dendl; |
f67539c2 | 8517 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8518 | _closedir(dir); |
8519 | return 0; | |
8520 | } | |
8521 | ||
8522 | void Client::_closedir(dir_result_t *dirp) | |
8523 | { | |
11fdf7f2 | 8524 | ldout(cct, 10) << __func__ << "(" << dirp << ")" << dendl; |
f67539c2 | 8525 | |
7c673cae | 8526 | if (dirp->inode) { |
11fdf7f2 | 8527 | ldout(cct, 10) << __func__ << " detaching inode " << dirp->inode << dendl; |
7c673cae FG |
8528 | dirp->inode.reset(); |
8529 | } | |
8530 | _readdir_drop_dirp_buffer(dirp); | |
8531 | opened_dirs.erase(dirp); | |
8532 | delete dirp; | |
8533 | } | |
8534 | ||
8535 | void Client::rewinddir(dir_result_t *dirp) | |
8536 | { | |
11fdf7f2 | 8537 | ldout(cct, 3) << __func__ << "(" << dirp << ")" << dendl; |
181888fb | 8538 | |
f67539c2 TL |
8539 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8540 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8541 | return; |
8542 | ||
f67539c2 | 8543 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8544 | dir_result_t *d = static_cast<dir_result_t*>(dirp); |
8545 | _readdir_drop_dirp_buffer(d); | |
8546 | d->reset(); | |
8547 | } | |
8548 | ||
8549 | loff_t Client::telldir(dir_result_t *dirp) | |
8550 | { | |
8551 | dir_result_t *d = static_cast<dir_result_t*>(dirp); | |
11fdf7f2 | 8552 | ldout(cct, 3) << __func__ << "(" << dirp << ") = " << d->offset << dendl; |
7c673cae FG |
8553 | return d->offset; |
8554 | } | |
8555 | ||
8556 | void Client::seekdir(dir_result_t *dirp, loff_t offset) | |
8557 | { | |
11fdf7f2 | 8558 | ldout(cct, 3) << __func__ << "(" << dirp << ", " << offset << ")" << dendl; |
7c673cae | 8559 | |
f67539c2 TL |
8560 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8561 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8562 | return; |
8563 | ||
f67539c2 TL |
8564 | std::scoped_lock lock(client_lock); |
8565 | ||
7c673cae FG |
8566 | if (offset == dirp->offset) |
8567 | return; | |
8568 | ||
8569 | if (offset > dirp->offset) | |
8570 | dirp->release_count = 0; // bump if we do a forward seek | |
8571 | else | |
8572 | dirp->ordered_count = 0; // disable filling readdir cache | |
8573 | ||
8574 | if (dirp->hash_order()) { | |
8575 | if (dirp->offset > offset) { | |
8576 | _readdir_drop_dirp_buffer(dirp); | |
8577 | dirp->reset(); | |
8578 | } | |
8579 | } else { | |
8580 | if (offset == 0 || | |
8581 | dirp->buffer_frag != frag_t(dir_result_t::fpos_high(offset)) || | |
8582 | dirp->offset_low() > dir_result_t::fpos_low(offset)) { | |
8583 | _readdir_drop_dirp_buffer(dirp); | |
8584 | dirp->reset(); | |
8585 | } | |
8586 | } | |
8587 | ||
8588 | dirp->offset = offset; | |
8589 | } | |
8590 | ||
8591 | ||
8592 | //struct dirent { | |
8593 | // ino_t d_ino; /* inode number */ | |
8594 | // off_t d_off; /* offset to the next dirent */ | |
8595 | // unsigned short d_reclen; /* length of this record */ | |
8596 | // unsigned char d_type; /* type of file */ | |
8597 | // char d_name[256]; /* filename */ | |
8598 | //}; | |
8599 | void Client::fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off) | |
8600 | { | |
8601 | strncpy(de->d_name, name, 255); | |
8602 | de->d_name[255] = '\0'; | |
f67539c2 | 8603 | #if !defined(__CYGWIN__) && !(defined(_WIN32)) |
7c673cae | 8604 | de->d_ino = ino; |
11fdf7f2 | 8605 | #if !defined(__APPLE__) && !defined(__FreeBSD__) |
7c673cae FG |
8606 | de->d_off = next_off; |
8607 | #endif | |
8608 | de->d_reclen = 1; | |
8609 | de->d_type = IFTODT(type); | |
11fdf7f2 | 8610 | ldout(cct, 10) << __func__ << " '" << de->d_name << "' -> " << inodeno_t(de->d_ino) |
7c673cae FG |
8611 | << " type " << (int)de->d_type << " w/ next_off " << hex << next_off << dec << dendl; |
8612 | #endif | |
8613 | } | |
8614 | ||
8615 | void Client::_readdir_next_frag(dir_result_t *dirp) | |
8616 | { | |
8617 | frag_t fg = dirp->buffer_frag; | |
8618 | ||
8619 | if (fg.is_rightmost()) { | |
11fdf7f2 | 8620 | ldout(cct, 10) << __func__ << " advance from " << fg << " to END" << dendl; |
7c673cae FG |
8621 | dirp->set_end(); |
8622 | return; | |
8623 | } | |
8624 | ||
8625 | // advance | |
8626 | fg = fg.next(); | |
11fdf7f2 | 8627 | ldout(cct, 10) << __func__ << " advance from " << dirp->buffer_frag << " to " << fg << dendl; |
7c673cae FG |
8628 | |
8629 | if (dirp->hash_order()) { | |
8630 | // keep last_name | |
8631 | int64_t new_offset = dir_result_t::make_fpos(fg.value(), 2, true); | |
8632 | if (dirp->offset < new_offset) // don't decrease offset | |
8633 | dirp->offset = new_offset; | |
8634 | } else { | |
8635 | dirp->last_name.clear(); | |
8636 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); | |
8637 | _readdir_rechoose_frag(dirp); | |
8638 | } | |
8639 | } | |
8640 | ||
8641 | void Client::_readdir_rechoose_frag(dir_result_t *dirp) | |
8642 | { | |
11fdf7f2 | 8643 | ceph_assert(dirp->inode); |
7c673cae FG |
8644 | |
8645 | if (dirp->hash_order()) | |
8646 | return; | |
8647 | ||
8648 | frag_t cur = frag_t(dirp->offset_high()); | |
8649 | frag_t fg = dirp->inode->dirfragtree[cur.value()]; | |
8650 | if (fg != cur) { | |
11fdf7f2 | 8651 | ldout(cct, 10) << __func__ << " frag " << cur << " maps to " << fg << dendl; |
7c673cae FG |
8652 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); |
8653 | dirp->last_name.clear(); | |
8654 | dirp->next_offset = 2; | |
8655 | } | |
8656 | } | |
8657 | ||
8658 | void Client::_readdir_drop_dirp_buffer(dir_result_t *dirp) | |
8659 | { | |
11fdf7f2 | 8660 | ldout(cct, 10) << __func__ << " " << dirp << dendl; |
7c673cae FG |
8661 | dirp->buffer.clear(); |
8662 | } | |
8663 | ||
8664 | int Client::_readdir_get_frag(dir_result_t *dirp) | |
8665 | { | |
11fdf7f2 TL |
8666 | ceph_assert(dirp); |
8667 | ceph_assert(dirp->inode); | |
7c673cae FG |
8668 | |
8669 | // get the current frag. | |
8670 | frag_t fg; | |
8671 | if (dirp->hash_order()) | |
8672 | fg = dirp->inode->dirfragtree[dirp->offset_high()]; | |
8673 | else | |
8674 | fg = frag_t(dirp->offset_high()); | |
8675 | ||
11fdf7f2 | 8676 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " fg " << fg |
7c673cae FG |
8677 | << " offset " << hex << dirp->offset << dec << dendl; |
8678 | ||
8679 | int op = CEPH_MDS_OP_READDIR; | |
8680 | if (dirp->inode && dirp->inode->snapid == CEPH_SNAPDIR) | |
8681 | op = CEPH_MDS_OP_LSSNAP; | |
8682 | ||
8683 | InodeRef& diri = dirp->inode; | |
8684 | ||
8685 | MetaRequest *req = new MetaRequest(op); | |
8686 | filepath path; | |
8687 | diri->make_nosnap_relative_path(path); | |
8688 | req->set_filepath(path); | |
8689 | req->set_inode(diri.get()); | |
8690 | req->head.args.readdir.frag = fg; | |
8691 | req->head.args.readdir.flags = CEPH_READDIR_REPLY_BITFLAGS; | |
8692 | if (dirp->last_name.length()) { | |
94b18763 | 8693 | req->path2.set_path(dirp->last_name); |
7c673cae FG |
8694 | } else if (dirp->hash_order()) { |
8695 | req->head.args.readdir.offset_hash = dirp->offset_high(); | |
8696 | } | |
8697 | req->dirp = dirp; | |
8698 | ||
8699 | bufferlist dirbl; | |
8700 | int res = make_request(req, dirp->perms, NULL, NULL, -1, &dirbl); | |
8701 | ||
f67539c2 | 8702 | if (res == -CEPHFS_EAGAIN) { |
11fdf7f2 | 8703 | ldout(cct, 10) << __func__ << " got EAGAIN, retrying" << dendl; |
7c673cae FG |
8704 | _readdir_rechoose_frag(dirp); |
8705 | return _readdir_get_frag(dirp); | |
8706 | } | |
8707 | ||
8708 | if (res == 0) { | |
11fdf7f2 | 8709 | ldout(cct, 10) << __func__ << " " << dirp << " got frag " << dirp->buffer_frag |
7c673cae FG |
8710 | << " size " << dirp->buffer.size() << dendl; |
8711 | } else { | |
11fdf7f2 | 8712 | ldout(cct, 10) << __func__ << " got error " << res << ", setting end flag" << dendl; |
7c673cae FG |
8713 | dirp->set_end(); |
8714 | } | |
8715 | ||
8716 | return res; | |
8717 | } | |
8718 | ||
8719 | struct dentry_off_lt { | |
8720 | bool operator()(const Dentry* dn, int64_t off) const { | |
8721 | return dir_result_t::fpos_cmp(dn->offset, off) < 0; | |
8722 | } | |
8723 | }; | |
8724 | ||
8725 | int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, | |
8726 | int caps, bool getref) | |
8727 | { | |
f67539c2 | 8728 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11fdf7f2 | 8729 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino |
7c673cae FG |
8730 | << " last_name " << dirp->last_name << " offset " << hex << dirp->offset << dec |
8731 | << dendl; | |
8732 | Dir *dir = dirp->inode->dir; | |
8733 | ||
8734 | if (!dir) { | |
8735 | ldout(cct, 10) << " dir is empty" << dendl; | |
8736 | dirp->set_end(); | |
8737 | return 0; | |
8738 | } | |
8739 | ||
8740 | vector<Dentry*>::iterator pd = std::lower_bound(dir->readdir_cache.begin(), | |
8741 | dir->readdir_cache.end(), | |
8742 | dirp->offset, dentry_off_lt()); | |
8743 | ||
8744 | string dn_name; | |
8745 | while (true) { | |
adb31ebb | 8746 | int mask = caps; |
7c673cae | 8747 | if (!dirp->inode->is_complete_and_ordered()) |
f67539c2 | 8748 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8749 | if (pd == dir->readdir_cache.end()) |
8750 | break; | |
8751 | Dentry *dn = *pd; | |
8752 | if (dn->inode == NULL) { | |
8753 | ldout(cct, 15) << " skipping null '" << dn->name << "'" << dendl; | |
8754 | ++pd; | |
8755 | continue; | |
8756 | } | |
8757 | if (dn->cap_shared_gen != dir->parent_inode->shared_gen) { | |
8758 | ldout(cct, 15) << " skipping mismatch shared gen '" << dn->name << "'" << dendl; | |
8759 | ++pd; | |
8760 | continue; | |
8761 | } | |
8762 | ||
92f5a8d4 | 8763 | int idx = pd - dir->readdir_cache.begin(); |
adb31ebb TL |
8764 | if (dn->inode->is_dir()) { |
8765 | mask |= CEPH_STAT_RSTAT; | |
8766 | } | |
8767 | int r = _getattr(dn->inode, mask, dirp->perms); | |
7c673cae FG |
8768 | if (r < 0) |
8769 | return r; | |
92f5a8d4 TL |
8770 | |
8771 | // the content of readdir_cache may change after _getattr(), so pd may be invalid iterator | |
8772 | pd = dir->readdir_cache.begin() + idx; | |
8773 | if (pd >= dir->readdir_cache.end() || *pd != dn) | |
f67539c2 | 8774 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8775 | |
8776 | struct ceph_statx stx; | |
8777 | struct dirent de; | |
8778 | fill_statx(dn->inode, caps, &stx); | |
8779 | ||
8780 | uint64_t next_off = dn->offset + 1; | |
eafe8130 | 8781 | fill_dirent(&de, dn->name.c_str(), stx.stx_mode, stx.stx_ino, next_off); |
7c673cae FG |
8782 | ++pd; |
8783 | if (pd == dir->readdir_cache.end()) | |
8784 | next_off = dir_result_t::END; | |
8785 | ||
8786 | Inode *in = NULL; | |
7c673cae FG |
8787 | if (getref) { |
8788 | in = dn->inode.get(); | |
8789 | _ll_get(in); | |
8790 | } | |
8791 | ||
8792 | dn_name = dn->name; // fill in name while we have lock | |
8793 | ||
9f95a23c | 8794 | client_lock.unlock(); |
7c673cae | 8795 | r = cb(p, &de, &stx, next_off, in); // _next_ offset |
9f95a23c | 8796 | client_lock.lock(); |
7c673cae FG |
8797 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << dn->offset << dec |
8798 | << " = " << r << dendl; | |
8799 | if (r < 0) { | |
8800 | return r; | |
8801 | } | |
8802 | ||
8803 | dirp->offset = next_off; | |
8804 | if (dirp->at_end()) | |
8805 | dirp->next_offset = 2; | |
8806 | else | |
8807 | dirp->next_offset = dirp->offset_low(); | |
8808 | dirp->last_name = dn_name; // we successfully returned this one; update! | |
28e407b8 | 8809 | dirp->release_count = 0; // last_name no longer match cache index |
7c673cae FG |
8810 | if (r > 0) |
8811 | return r; | |
8812 | } | |
8813 | ||
11fdf7f2 | 8814 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " at end" << dendl; |
7c673cae FG |
8815 | dirp->set_end(); |
8816 | return 0; | |
8817 | } | |
8818 | ||
8819 | int Client::readdir_r_cb(dir_result_t *d, add_dirent_cb_t cb, void *p, | |
8820 | unsigned want, unsigned flags, bool getref) | |
8821 | { | |
8822 | int caps = statx_to_mask(flags, want); | |
8823 | ||
f67539c2 TL |
8824 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8825 | if (!mref_reader.is_state_satisfied()) | |
8826 | return -CEPHFS_ENOTCONN; | |
7c673cae | 8827 | |
f67539c2 | 8828 | std::unique_lock cl(client_lock); |
181888fb | 8829 | |
7c673cae FG |
8830 | dir_result_t *dirp = static_cast<dir_result_t*>(d); |
8831 | ||
11fdf7f2 | 8832 | ldout(cct, 10) << __func__ << " " << *dirp->inode << " offset " << hex << dirp->offset |
7c673cae FG |
8833 | << dec << " at_end=" << dirp->at_end() |
8834 | << " hash_order=" << dirp->hash_order() << dendl; | |
8835 | ||
8836 | struct dirent de; | |
8837 | struct ceph_statx stx; | |
8838 | memset(&de, 0, sizeof(de)); | |
8839 | memset(&stx, 0, sizeof(stx)); | |
8840 | ||
8841 | InodeRef& diri = dirp->inode; | |
8842 | ||
8843 | if (dirp->at_end()) | |
8844 | return 0; | |
8845 | ||
8846 | if (dirp->offset == 0) { | |
8847 | ldout(cct, 15) << " including ." << dendl; | |
11fdf7f2 | 8848 | ceph_assert(diri->dentries.size() < 2); // can't have multiple hard-links to a dir |
7c673cae FG |
8849 | uint64_t next_off = 1; |
8850 | ||
8851 | int r; | |
adb31ebb | 8852 | r = _getattr(diri, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
8853 | if (r < 0) |
8854 | return r; | |
8855 | ||
8856 | fill_statx(diri, caps, &stx); | |
8857 | fill_dirent(&de, ".", S_IFDIR, stx.stx_ino, next_off); | |
8858 | ||
8859 | Inode *inode = NULL; | |
8860 | if (getref) { | |
8861 | inode = diri.get(); | |
8862 | _ll_get(inode); | |
8863 | } | |
8864 | ||
f67539c2 | 8865 | cl.unlock(); |
7c673cae | 8866 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 8867 | cl.lock(); |
7c673cae FG |
8868 | if (r < 0) |
8869 | return r; | |
8870 | ||
8871 | dirp->offset = next_off; | |
8872 | if (r > 0) | |
8873 | return r; | |
8874 | } | |
8875 | if (dirp->offset == 1) { | |
8876 | ldout(cct, 15) << " including .." << dendl; | |
8877 | uint64_t next_off = 2; | |
8878 | InodeRef in; | |
11fdf7f2 | 8879 | if (diri->dentries.empty()) |
7c673cae FG |
8880 | in = diri; |
8881 | else | |
94b18763 | 8882 | in = diri->get_first_parent()->dir->parent_inode; |
7c673cae FG |
8883 | |
8884 | int r; | |
adb31ebb | 8885 | r = _getattr(in, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
8886 | if (r < 0) |
8887 | return r; | |
8888 | ||
8889 | fill_statx(in, caps, &stx); | |
8890 | fill_dirent(&de, "..", S_IFDIR, stx.stx_ino, next_off); | |
8891 | ||
8892 | Inode *inode = NULL; | |
8893 | if (getref) { | |
8894 | inode = in.get(); | |
8895 | _ll_get(inode); | |
8896 | } | |
8897 | ||
f67539c2 | 8898 | cl.unlock(); |
7c673cae | 8899 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 8900 | cl.lock(); |
7c673cae FG |
8901 | if (r < 0) |
8902 | return r; | |
8903 | ||
8904 | dirp->offset = next_off; | |
8905 | if (r > 0) | |
8906 | return r; | |
8907 | } | |
8908 | ||
8909 | // can we read from our cache? | |
8910 | ldout(cct, 10) << "offset " << hex << dirp->offset << dec | |
8911 | << " snapid " << dirp->inode->snapid << " (complete && ordered) " | |
8912 | << dirp->inode->is_complete_and_ordered() | |
8913 | << " issued " << ccap_string(dirp->inode->caps_issued()) | |
8914 | << dendl; | |
8915 | if (dirp->inode->snapid != CEPH_SNAPDIR && | |
8916 | dirp->inode->is_complete_and_ordered() && | |
94b18763 | 8917 | dirp->inode->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 8918 | int err = _readdir_cache_cb(dirp, cb, p, caps, getref); |
f67539c2 | 8919 | if (err != -CEPHFS_EAGAIN) |
7c673cae FG |
8920 | return err; |
8921 | } | |
8922 | ||
8923 | while (1) { | |
8924 | if (dirp->at_end()) | |
8925 | return 0; | |
8926 | ||
8927 | bool check_caps = true; | |
8928 | if (!dirp->is_cached()) { | |
8929 | int r = _readdir_get_frag(dirp); | |
8930 | if (r) | |
8931 | return r; | |
8932 | // _readdir_get_frag () may updates dirp->offset if the replied dirfrag is | |
8933 | // different than the requested one. (our dirfragtree was outdated) | |
8934 | check_caps = false; | |
8935 | } | |
8936 | frag_t fg = dirp->buffer_frag; | |
8937 | ||
8938 | ldout(cct, 10) << "frag " << fg << " buffer size " << dirp->buffer.size() | |
8939 | << " offset " << hex << dirp->offset << dendl; | |
8940 | ||
8941 | for (auto it = std::lower_bound(dirp->buffer.begin(), dirp->buffer.end(), | |
8942 | dirp->offset, dir_result_t::dentry_off_lt()); | |
8943 | it != dirp->buffer.end(); | |
8944 | ++it) { | |
8945 | dir_result_t::dentry &entry = *it; | |
8946 | ||
8947 | uint64_t next_off = entry.offset + 1; | |
8948 | ||
8949 | int r; | |
8950 | if (check_caps) { | |
adb31ebb TL |
8951 | int mask = caps; |
8952 | if(entry.inode->is_dir()){ | |
8953 | mask |= CEPH_STAT_RSTAT; | |
8954 | } | |
8955 | r = _getattr(entry.inode, mask, dirp->perms); | |
7c673cae FG |
8956 | if (r < 0) |
8957 | return r; | |
8958 | } | |
8959 | ||
8960 | fill_statx(entry.inode, caps, &stx); | |
8961 | fill_dirent(&de, entry.name.c_str(), stx.stx_mode, stx.stx_ino, next_off); | |
8962 | ||
8963 | Inode *inode = NULL; | |
8964 | if (getref) { | |
8965 | inode = entry.inode.get(); | |
8966 | _ll_get(inode); | |
8967 | } | |
8968 | ||
f67539c2 | 8969 | cl.unlock(); |
7c673cae | 8970 | r = cb(p, &de, &stx, next_off, inode); // _next_ offset |
f67539c2 | 8971 | cl.lock(); |
7c673cae FG |
8972 | |
8973 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << next_off - 1 << dec | |
8974 | << " = " << r << dendl; | |
8975 | if (r < 0) | |
8976 | return r; | |
8977 | ||
8978 | dirp->offset = next_off; | |
8979 | if (r > 0) | |
8980 | return r; | |
8981 | } | |
8982 | ||
8983 | if (dirp->next_offset > 2) { | |
8984 | ldout(cct, 10) << " fetching next chunk of this frag" << dendl; | |
8985 | _readdir_drop_dirp_buffer(dirp); | |
8986 | continue; // more! | |
8987 | } | |
8988 | ||
8989 | if (!fg.is_rightmost()) { | |
8990 | // next frag! | |
8991 | _readdir_next_frag(dirp); | |
8992 | continue; | |
8993 | } | |
8994 | ||
8995 | if (diri->shared_gen == dirp->start_shared_gen && | |
8996 | diri->dir_release_count == dirp->release_count) { | |
8997 | if (diri->dir_ordered_count == dirp->ordered_count) { | |
8998 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
8999 | if (diri->dir) { | |
11fdf7f2 | 9000 | ceph_assert(diri->dir->readdir_cache.size() >= dirp->cache_index); |
7c673cae FG |
9001 | diri->dir->readdir_cache.resize(dirp->cache_index); |
9002 | } | |
9003 | diri->flags |= I_COMPLETE | I_DIR_ORDERED; | |
9004 | } else { | |
9005 | ldout(cct, 10) << " marking I_COMPLETE on " << *diri << dendl; | |
9006 | diri->flags |= I_COMPLETE; | |
9007 | } | |
9008 | } | |
9009 | ||
9010 | dirp->set_end(); | |
9011 | return 0; | |
9012 | } | |
9013 | ceph_abort(); | |
9014 | return 0; | |
9015 | } | |
9016 | ||
9017 | ||
9018 | int Client::readdir_r(dir_result_t *d, struct dirent *de) | |
9019 | { | |
9020 | return readdirplus_r(d, de, 0, 0, 0, NULL); | |
9021 | } | |
9022 | ||
9023 | /* | |
9024 | * readdirplus_r | |
9025 | * | |
9026 | * returns | |
9027 | * 1 if we got a dirent | |
9028 | * 0 for end of directory | |
9029 | * <0 on error | |
9030 | */ | |
9031 | ||
9032 | struct single_readdir { | |
9033 | struct dirent *de; | |
9034 | struct ceph_statx *stx; | |
9035 | Inode *inode; | |
9036 | bool full; | |
9037 | }; | |
9038 | ||
9039 | static int _readdir_single_dirent_cb(void *p, struct dirent *de, | |
9040 | struct ceph_statx *stx, off_t off, | |
9041 | Inode *in) | |
9042 | { | |
9043 | single_readdir *c = static_cast<single_readdir *>(p); | |
9044 | ||
9045 | if (c->full) | |
9046 | return -1; // already filled this dirent | |
9047 | ||
9048 | *c->de = *de; | |
9049 | if (c->stx) | |
9050 | *c->stx = *stx; | |
9051 | c->inode = in; | |
9052 | c->full = true; | |
9053 | return 1; | |
9054 | } | |
9055 | ||
9056 | struct dirent *Client::readdir(dir_result_t *d) | |
9057 | { | |
9058 | int ret; | |
f91f0fd5 | 9059 | auto& de = d->de; |
7c673cae FG |
9060 | single_readdir sr; |
9061 | sr.de = &de; | |
9062 | sr.stx = NULL; | |
9063 | sr.inode = NULL; | |
9064 | sr.full = false; | |
9065 | ||
9066 | // our callback fills the dirent and sets sr.full=true on first | |
9067 | // call, and returns -1 the second time around. | |
9068 | ret = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr); | |
9069 | if (ret < -1) { | |
9070 | errno = -ret; // this sucks. | |
9071 | return (dirent *) NULL; | |
9072 | } | |
9073 | if (sr.full) { | |
9074 | return &de; | |
9075 | } | |
9076 | return (dirent *) NULL; | |
9077 | } | |
9078 | ||
9079 | int Client::readdirplus_r(dir_result_t *d, struct dirent *de, | |
9080 | struct ceph_statx *stx, unsigned want, | |
9081 | unsigned flags, Inode **out) | |
9082 | { | |
9083 | single_readdir sr; | |
9084 | sr.de = de; | |
9085 | sr.stx = stx; | |
9086 | sr.inode = NULL; | |
9087 | sr.full = false; | |
9088 | ||
9089 | // our callback fills the dirent and sets sr.full=true on first | |
9090 | // call, and returns -1 the second time around. | |
9091 | int r = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr, want, flags, out); | |
9092 | if (r < -1) | |
9093 | return r; | |
9094 | if (out) | |
9095 | *out = sr.inode; | |
9096 | if (sr.full) | |
9097 | return 1; | |
9098 | return 0; | |
9099 | } | |
9100 | ||
9101 | ||
9102 | /* getdents */ | |
9103 | struct getdents_result { | |
9104 | char *buf; | |
9105 | int buflen; | |
9106 | int pos; | |
9107 | bool fullent; | |
9108 | }; | |
9109 | ||
9110 | static int _readdir_getdent_cb(void *p, struct dirent *de, | |
9111 | struct ceph_statx *stx, off_t off, Inode *in) | |
9112 | { | |
9113 | struct getdents_result *c = static_cast<getdents_result *>(p); | |
9114 | ||
9115 | int dlen; | |
9116 | if (c->fullent) | |
9117 | dlen = sizeof(*de); | |
9118 | else | |
9119 | dlen = strlen(de->d_name) + 1; | |
9120 | ||
9121 | if (c->pos + dlen > c->buflen) | |
9122 | return -1; // doesn't fit | |
9123 | ||
9124 | if (c->fullent) { | |
9125 | memcpy(c->buf + c->pos, de, sizeof(*de)); | |
9126 | } else { | |
9127 | memcpy(c->buf + c->pos, de->d_name, dlen); | |
9128 | } | |
9129 | c->pos += dlen; | |
9130 | return 0; | |
9131 | } | |
9132 | ||
9133 | int Client::_getdents(dir_result_t *dir, char *buf, int buflen, bool fullent) | |
9134 | { | |
9135 | getdents_result gr; | |
9136 | gr.buf = buf; | |
9137 | gr.buflen = buflen; | |
9138 | gr.fullent = fullent; | |
9139 | gr.pos = 0; | |
9140 | ||
9141 | int r = readdir_r_cb(dir, _readdir_getdent_cb, (void *)&gr); | |
9142 | ||
9143 | if (r < 0) { // some error | |
9144 | if (r == -1) { // buffer ran out of space | |
9145 | if (gr.pos) { // but we got some entries already! | |
9146 | return gr.pos; | |
9147 | } // or we need a larger buffer | |
f67539c2 | 9148 | return -CEPHFS_ERANGE; |
7c673cae FG |
9149 | } else { // actual error, return it |
9150 | return r; | |
9151 | } | |
9152 | } | |
9153 | return gr.pos; | |
9154 | } | |
9155 | ||
9156 | ||
9157 | /* getdir */ | |
9158 | struct getdir_result { | |
9159 | list<string> *contents; | |
9160 | int num; | |
9161 | }; | |
9162 | ||
9163 | static int _getdir_cb(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in) | |
9164 | { | |
9165 | getdir_result *r = static_cast<getdir_result *>(p); | |
9166 | ||
9167 | r->contents->push_back(de->d_name); | |
9168 | r->num++; | |
9169 | return 0; | |
9170 | } | |
9171 | ||
9172 | int Client::getdir(const char *relpath, list<string>& contents, | |
9173 | const UserPerm& perms) | |
9174 | { | |
9175 | ldout(cct, 3) << "getdir(" << relpath << ")" << dendl; | |
f67539c2 TL |
9176 | tout(cct) << "getdir" << std::endl; |
9177 | tout(cct) << relpath << std::endl; | |
7c673cae FG |
9178 | |
9179 | dir_result_t *d; | |
9180 | int r = opendir(relpath, &d, perms); | |
9181 | if (r < 0) | |
9182 | return r; | |
9183 | ||
9184 | getdir_result gr; | |
9185 | gr.contents = &contents; | |
9186 | gr.num = 0; | |
9187 | r = readdir_r_cb(d, _getdir_cb, (void *)&gr); | |
9188 | ||
9189 | closedir(d); | |
9190 | ||
9191 | if (r < 0) | |
9192 | return r; | |
9193 | return gr.num; | |
9194 | } | |
9195 | ||
9196 | ||
9197 | /****** file i/o **********/ | |
f67539c2 | 9198 | |
b3b6e05e | 9199 | // common parts for open and openat. call with client_lock locked. |
20effc67 | 9200 | int Client::create_and_open(int dirfd, const char *relpath, int flags, |
b3b6e05e TL |
9201 | const UserPerm& perms, mode_t mode, int stripe_unit, |
9202 | int stripe_count, int object_size, const char *data_pool, | |
9203 | std::string alternate_name) { | |
9204 | ceph_assert(ceph_mutex_is_locked(client_lock)); | |
f91f0fd5 | 9205 | int cflags = ceph_flags_sys2wire(flags); |
f91f0fd5 | 9206 | tout(cct) << cflags << std::endl; |
7c673cae FG |
9207 | |
9208 | Fh *fh = NULL; | |
9209 | ||
9210 | #if defined(__linux__) && defined(O_PATH) | |
9211 | /* When the O_PATH is being specified, others flags than O_DIRECTORY | |
9212 | * and O_NOFOLLOW are ignored. Please refer do_entry_open() function | |
9213 | * in kernel (fs/open.c). */ | |
9214 | if (flags & O_PATH) | |
9215 | flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; | |
9216 | #endif | |
9217 | ||
9218 | filepath path(relpath); | |
9219 | InodeRef in; | |
9220 | bool created = false; | |
9221 | /* O_CREATE with O_EXCL enforces O_NOFOLLOW. */ | |
9222 | bool followsym = !((flags & O_NOFOLLOW) || ((flags & O_CREAT) && (flags & O_EXCL))); | |
f91f0fd5 TL |
9223 | int mask = ceph_caps_for_mode(ceph_flags_to_mode(cflags)); |
9224 | ||
b3b6e05e | 9225 | InodeRef dirinode = nullptr; |
20effc67 TL |
9226 | int r = get_fd_inode(dirfd, &dirinode); |
9227 | if (r < 0) { | |
9228 | return r; | |
b3b6e05e | 9229 | } |
7c673cae | 9230 | |
20effc67 | 9231 | r = path_walk(path, &in, perms, followsym, mask, dirinode); |
7c673cae | 9232 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) |
f67539c2 | 9233 | return -CEPHFS_EEXIST; |
7c673cae FG |
9234 | |
9235 | #if defined(__linux__) && defined(O_PATH) | |
9236 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW) && !(flags & O_PATH)) | |
9237 | #else | |
b3b6e05e | 9238 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW)) |
7c673cae | 9239 | #endif |
f67539c2 | 9240 | return -CEPHFS_ELOOP; |
7c673cae | 9241 | |
f67539c2 | 9242 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
7c673cae FG |
9243 | filepath dirpath = path; |
9244 | string dname = dirpath.last_dentry(); | |
9245 | dirpath.pop_dentry(); | |
9246 | InodeRef dir; | |
9247 | r = path_walk(dirpath, &dir, perms, true, | |
b3b6e05e TL |
9248 | cct->_conf->client_permissions ? CEPH_CAP_AUTH_SHARED : 0, dirinode); |
9249 | if (r < 0) { | |
7c673cae | 9250 | goto out; |
b3b6e05e | 9251 | } |
7c673cae FG |
9252 | if (cct->_conf->client_permissions) { |
9253 | r = may_create(dir.get(), perms); | |
9254 | if (r < 0) | |
b3b6e05e | 9255 | goto out; |
7c673cae FG |
9256 | } |
9257 | r = _create(dir.get(), dname.c_str(), flags, mode, &in, &fh, stripe_unit, | |
f67539c2 TL |
9258 | stripe_count, object_size, data_pool, &created, perms, |
9259 | std::move(alternate_name)); | |
7c673cae FG |
9260 | } |
9261 | if (r < 0) | |
9262 | goto out; | |
9263 | ||
9264 | if (!created) { | |
9265 | // posix says we can only check permissions of existing files | |
9266 | if (cct->_conf->client_permissions) { | |
9267 | r = may_open(in.get(), flags, perms); | |
9268 | if (r < 0) | |
b3b6e05e | 9269 | goto out; |
7c673cae FG |
9270 | } |
9271 | } | |
9272 | ||
9273 | if (!fh) | |
9274 | r = _open(in.get(), flags, mode, &fh, perms); | |
9275 | if (r >= 0) { | |
9276 | // allocate a integer file descriptor | |
11fdf7f2 | 9277 | ceph_assert(fh); |
7c673cae | 9278 | r = get_fd(); |
11fdf7f2 | 9279 | ceph_assert(fd_map.count(r) == 0); |
7c673cae FG |
9280 | fd_map[r] = fh; |
9281 | } | |
9282 | ||
9283 | out: | |
b3b6e05e TL |
9284 | return r; |
9285 | } | |
9286 | ||
9287 | int Client::open(const char *relpath, int flags, const UserPerm& perms, | |
9288 | mode_t mode, int stripe_unit, int stripe_count, | |
9289 | int object_size, const char *data_pool, std::string alternate_name) | |
9290 | { | |
9291 | return openat(CEPHFS_AT_FDCWD, relpath, flags, perms, mode, stripe_unit, | |
9292 | stripe_count, object_size, data_pool, alternate_name); | |
9293 | } | |
9294 | ||
b3b6e05e TL |
9295 | int Client::openat(int dirfd, const char *relpath, int flags, const UserPerm& perms, |
9296 | mode_t mode, int stripe_unit, int stripe_count, int object_size, | |
9297 | const char *data_pool, std::string alternate_name) { | |
9298 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9299 | if (!mref_reader.is_state_satisfied()) { | |
9300 | return -CEPHFS_ENOTCONN; | |
9301 | } | |
9302 | ||
9303 | ldout(cct, 3) << "openat enter(" << relpath << ")" << dendl; | |
9304 | tout(cct) << dirfd << std::endl; | |
9305 | tout(cct) << relpath << std::endl; | |
9306 | tout(cct) << flags << std::endl; | |
9307 | tout(cct) << mode << std::endl; | |
9308 | ||
9309 | std::scoped_lock locker(client_lock); | |
9310 | int r = create_and_open(dirfd, relpath, flags, perms, mode, stripe_unit, stripe_count, | |
9311 | object_size, data_pool, alternate_name); | |
9312 | ||
7c673cae | 9313 | tout(cct) << r << std::endl; |
b3b6e05e | 9314 | ldout(cct, 3) << "openat exit(" << relpath << ")" << dendl; |
7c673cae FG |
9315 | return r; |
9316 | } | |
9317 | ||
7c673cae FG |
9318 | int Client::lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name, |
9319 | const UserPerm& perms) | |
9320 | { | |
11fdf7f2 | 9321 | ldout(cct, 3) << __func__ << " enter(" << ino << ", #" << dirino << "/" << name << ")" << dendl; |
7c673cae | 9322 | |
f67539c2 TL |
9323 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9324 | if (!mref_reader.is_state_satisfied()) | |
9325 | return -CEPHFS_ENOTCONN; | |
181888fb | 9326 | |
f67539c2 | 9327 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9328 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPHASH); |
9329 | filepath path(ino); | |
9330 | req->set_filepath(path); | |
9331 | ||
9332 | uint32_t h = ceph_str_hash(CEPH_STR_HASH_RJENKINS, name, strlen(name)); | |
9333 | char f[30]; | |
9334 | sprintf(f, "%u", h); | |
9335 | filepath path2(dirino); | |
9336 | path2.push_dentry(string(f)); | |
9337 | req->set_filepath2(path2); | |
9338 | ||
9339 | int r = make_request(req, perms, NULL, NULL, | |
9340 | rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9341 | ldout(cct, 3) << __func__ << " exit(" << ino << ", #" << dirino << "/" << name << ") = " << r << dendl; |
7c673cae FG |
9342 | return r; |
9343 | } | |
9344 | ||
9345 | ||
9346 | /** | |
9347 | * Load inode into local cache. | |
9348 | * | |
9349 | * If inode pointer is non-NULL, and take a reference on | |
9350 | * the resulting Inode object in one operation, so that caller | |
9351 | * can safely assume inode will still be there after return. | |
9352 | */ | |
f67539c2 | 9353 | int Client::_lookup_vino(vinodeno_t vino, const UserPerm& perms, Inode **inode) |
7c673cae | 9354 | { |
f67539c2 | 9355 | ldout(cct, 8) << __func__ << " enter(" << vino << ")" << dendl; |
7c673cae | 9356 | |
f67539c2 TL |
9357 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9358 | if (!mref_reader.is_state_satisfied()) | |
9359 | return -CEPHFS_ENOTCONN; | |
181888fb | 9360 | |
b3b6e05e TL |
9361 | if (is_reserved_vino(vino)) |
9362 | return -CEPHFS_ESTALE; | |
9363 | ||
7c673cae | 9364 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPINO); |
f67539c2 | 9365 | filepath path(vino.ino); |
7c673cae FG |
9366 | req->set_filepath(path); |
9367 | ||
f67539c2 TL |
9368 | /* |
9369 | * The MDS expects either a "real" snapid here or 0. The special value | |
9370 | * carveouts for the snapid are all at the end of the range so we can | |
9371 | * just look for any snapid below this value. | |
9372 | */ | |
9373 | if (vino.snapid < CEPH_NOSNAP) | |
9374 | req->head.args.lookupino.snapid = vino.snapid; | |
9375 | ||
7c673cae FG |
9376 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); |
9377 | if (r == 0 && inode != NULL) { | |
7c673cae | 9378 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
11fdf7f2 | 9379 | ceph_assert(p != inode_map.end()); |
7c673cae FG |
9380 | *inode = p->second; |
9381 | _ll_get(*inode); | |
9382 | } | |
f67539c2 | 9383 | ldout(cct, 8) << __func__ << " exit(" << vino << ") = " << r << dendl; |
7c673cae FG |
9384 | return r; |
9385 | } | |
9386 | ||
1adf2230 AA |
9387 | int Client::lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode) |
9388 | { | |
f67539c2 TL |
9389 | vinodeno_t vino(ino, CEPH_NOSNAP); |
9390 | std::scoped_lock lock(client_lock); | |
9391 | return _lookup_vino(vino, perms, inode); | |
1adf2230 | 9392 | } |
7c673cae FG |
9393 | |
9394 | /** | |
9395 | * Find the parent inode of `ino` and insert it into | |
9396 | * our cache. Conditionally also set `parent` to a referenced | |
9397 | * Inode* if caller provides non-NULL value. | |
9398 | */ | |
1adf2230 | 9399 | int Client::_lookup_parent(Inode *ino, const UserPerm& perms, Inode **parent) |
7c673cae | 9400 | { |
11fdf7f2 | 9401 | ldout(cct, 8) << __func__ << " enter(" << ino->ino << ")" << dendl; |
7c673cae | 9402 | |
7c673cae FG |
9403 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); |
9404 | filepath path(ino->ino); | |
9405 | req->set_filepath(path); | |
9406 | ||
9407 | InodeRef target; | |
9408 | int r = make_request(req, perms, &target, NULL, rand() % mdsmap->get_num_in_mds()); | |
9409 | // Give caller a reference to the parent ino if they provided a pointer. | |
9410 | if (parent != NULL) { | |
9411 | if (r == 0) { | |
9412 | *parent = target.get(); | |
9413 | _ll_get(*parent); | |
11fdf7f2 | 9414 | ldout(cct, 8) << __func__ << " found parent " << (*parent)->ino << dendl; |
7c673cae FG |
9415 | } else { |
9416 | *parent = NULL; | |
9417 | } | |
9418 | } | |
11fdf7f2 | 9419 | ldout(cct, 8) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9420 | return r; |
9421 | } | |
9422 | ||
7c673cae FG |
9423 | /** |
9424 | * Populate the parent dentry for `ino`, provided it is | |
9425 | * a child of `parent`. | |
9426 | */ | |
1adf2230 | 9427 | int Client::_lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
7c673cae | 9428 | { |
11fdf7f2 TL |
9429 | ceph_assert(parent->is_dir()); |
9430 | ldout(cct, 3) << __func__ << " enter(" << ino->ino << ")" << dendl; | |
7c673cae | 9431 | |
f67539c2 TL |
9432 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9433 | if (!mref_reader.is_state_satisfied()) | |
9434 | return -CEPHFS_ENOTCONN; | |
181888fb | 9435 | |
7c673cae FG |
9436 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
9437 | req->set_filepath2(filepath(parent->ino)); | |
9438 | req->set_filepath(filepath(ino->ino)); | |
9439 | req->set_inode(ino); | |
9440 | ||
9441 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9442 | ldout(cct, 3) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9443 | return r; |
9444 | } | |
9445 | ||
1adf2230 AA |
9446 | int Client::lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
9447 | { | |
f67539c2 | 9448 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
9449 | return _lookup_name(ino, parent, perms); |
9450 | } | |
7c673cae | 9451 | |
11fdf7f2 | 9452 | Fh *Client::_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms) |
7c673cae | 9453 | { |
11fdf7f2 | 9454 | ceph_assert(in); |
f6b5b4d7 | 9455 | Fh *f = new Fh(in, flags, cmode, fd_gen, perms); |
7c673cae | 9456 | |
11fdf7f2 | 9457 | ldout(cct, 10) << __func__ << " " << in->ino << " mode " << cmode << dendl; |
7c673cae FG |
9458 | |
9459 | if (in->snapid != CEPH_NOSNAP) { | |
9460 | in->snap_cap_refs++; | |
9461 | ldout(cct, 5) << "open success, fh is " << f << " combined IMMUTABLE SNAP caps " | |
9462 | << ccap_string(in->caps_issued()) << dendl; | |
9463 | } | |
9464 | ||
11fdf7f2 | 9465 | const auto& conf = cct->_conf; |
7c673cae FG |
9466 | f->readahead.set_trigger_requests(1); |
9467 | f->readahead.set_min_readahead_size(conf->client_readahead_min); | |
9468 | uint64_t max_readahead = Readahead::NO_LIMIT; | |
9469 | if (conf->client_readahead_max_bytes) { | |
11fdf7f2 | 9470 | max_readahead = std::min(max_readahead, (uint64_t)conf->client_readahead_max_bytes); |
7c673cae FG |
9471 | } |
9472 | if (conf->client_readahead_max_periods) { | |
11fdf7f2 | 9473 | max_readahead = std::min(max_readahead, in->layout.get_period()*(uint64_t)conf->client_readahead_max_periods); |
7c673cae FG |
9474 | } |
9475 | f->readahead.set_max_readahead_size(max_readahead); | |
9476 | vector<uint64_t> alignments; | |
9477 | alignments.push_back(in->layout.get_period()); | |
9478 | alignments.push_back(in->layout.stripe_unit); | |
9479 | f->readahead.set_alignments(alignments); | |
9480 | ||
9481 | return f; | |
9482 | } | |
9483 | ||
9484 | int Client::_release_fh(Fh *f) | |
9485 | { | |
9486 | //ldout(cct, 3) << "op: client->close(open_files[ " << fh << " ]);" << dendl; | |
9487 | //ldout(cct, 3) << "op: open_files.erase( " << fh << " );" << dendl; | |
9488 | Inode *in = f->inode.get(); | |
11fdf7f2 | 9489 | ldout(cct, 8) << __func__ << " " << f << " mode " << f->mode << " on " << *in << dendl; |
7c673cae | 9490 | |
b32b8144 FG |
9491 | in->unset_deleg(f); |
9492 | ||
7c673cae FG |
9493 | if (in->snapid == CEPH_NOSNAP) { |
9494 | if (in->put_open_ref(f->mode)) { | |
9495 | _flush(in, new C_Client_FlushComplete(this, in)); | |
9496 | check_caps(in, 0); | |
9497 | } | |
9498 | } else { | |
11fdf7f2 | 9499 | ceph_assert(in->snap_cap_refs > 0); |
7c673cae FG |
9500 | in->snap_cap_refs--; |
9501 | } | |
9502 | ||
9503 | _release_filelocks(f); | |
9504 | ||
9505 | // Finally, read any async err (i.e. from flushes) | |
9506 | int err = f->take_async_err(); | |
9507 | if (err != 0) { | |
11fdf7f2 | 9508 | ldout(cct, 1) << __func__ << " " << f << " on inode " << *in << " caught async_err = " |
7c673cae FG |
9509 | << cpp_strerror(err) << dendl; |
9510 | } else { | |
11fdf7f2 | 9511 | ldout(cct, 10) << __func__ << " " << f << " on inode " << *in << " no async_err state" << dendl; |
7c673cae FG |
9512 | } |
9513 | ||
9514 | _put_fh(f); | |
9515 | ||
9516 | return err; | |
9517 | } | |
9518 | ||
9519 | void Client::_put_fh(Fh *f) | |
9520 | { | |
9521 | int left = f->put(); | |
9522 | if (!left) { | |
9523 | delete f; | |
9524 | } | |
9525 | } | |
9526 | ||
9527 | int Client::_open(Inode *in, int flags, mode_t mode, Fh **fhp, | |
9528 | const UserPerm& perms) | |
9529 | { | |
9530 | if (in->snapid != CEPH_NOSNAP && | |
9531 | (flags & (O_WRONLY | O_RDWR | O_CREAT | O_TRUNC | O_APPEND))) { | |
f67539c2 | 9532 | return -CEPHFS_EROFS; |
7c673cae FG |
9533 | } |
9534 | ||
9535 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
9536 | int cflags = ceph_flags_sys2wire(flags); |
9537 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
9538 | cflags |= CEPH_O_LAZY; | |
9539 | ||
9540 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
9541 | int want = ceph_caps_for_mode(cmode); |
9542 | int result = 0; | |
9543 | ||
9544 | in->get_open_ref(cmode); // make note of pending open, since it effects _wanted_ caps. | |
9545 | ||
b32b8144 | 9546 | if ((flags & O_TRUNC) == 0 && in->caps_issued_mask(want)) { |
7c673cae FG |
9547 | // update wanted? |
9548 | check_caps(in, CHECK_CAPS_NODELAY); | |
9549 | } else { | |
b32b8144 | 9550 | |
7c673cae FG |
9551 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); |
9552 | filepath path; | |
9553 | in->make_nosnap_relative_path(path); | |
9554 | req->set_filepath(path); | |
11fdf7f2 | 9555 | req->head.args.open.flags = cflags & ~CEPH_O_CREAT; |
7c673cae FG |
9556 | req->head.args.open.mode = mode; |
9557 | req->head.args.open.pool = -1; | |
9558 | if (cct->_conf->client_debug_getattr_caps) | |
9559 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9560 | else | |
9561 | req->head.args.open.mask = 0; | |
9562 | req->head.args.open.old_size = in->size; // for O_TRUNC | |
9563 | req->set_inode(in); | |
9564 | result = make_request(req, perms); | |
b32b8144 FG |
9565 | |
9566 | /* | |
9567 | * NFS expects that delegations will be broken on a conflicting open, | |
9568 | * not just when there is actual conflicting access to the file. SMB leases | |
9569 | * and oplocks also have similar semantics. | |
9570 | * | |
9571 | * Ensure that clients that have delegations enabled will wait on minimal | |
9572 | * caps during open, just to ensure that other clients holding delegations | |
9573 | * return theirs first. | |
9574 | */ | |
9575 | if (deleg_timeout && result == 0) { | |
9576 | int need = 0, have; | |
9577 | ||
9578 | if (cmode & CEPH_FILE_MODE_WR) | |
9579 | need |= CEPH_CAP_FILE_WR; | |
9580 | if (cmode & CEPH_FILE_MODE_RD) | |
9581 | need |= CEPH_CAP_FILE_RD; | |
9582 | ||
f6b5b4d7 TL |
9583 | Fh fh(in, flags, cmode, fd_gen, perms); |
9584 | result = get_caps(&fh, need, want, &have, -1); | |
b32b8144 | 9585 | if (result < 0) { |
1adf2230 | 9586 | ldout(cct, 8) << "Unable to get caps after open of inode " << *in << |
b32b8144 FG |
9587 | " . Denying open: " << |
9588 | cpp_strerror(result) << dendl; | |
b32b8144 FG |
9589 | } else { |
9590 | put_cap_ref(in, need); | |
9591 | } | |
9592 | } | |
7c673cae FG |
9593 | } |
9594 | ||
9595 | // success? | |
9596 | if (result >= 0) { | |
9597 | if (fhp) | |
9598 | *fhp = _create_fh(in, flags, cmode, perms); | |
9599 | } else { | |
9600 | in->put_open_ref(cmode); | |
9601 | } | |
9602 | ||
9603 | trim_cache(); | |
9604 | ||
9605 | return result; | |
9606 | } | |
9607 | ||
9608 | int Client::_renew_caps(Inode *in) | |
9609 | { | |
9610 | int wanted = in->caps_file_wanted(); | |
9611 | if (in->is_any_caps() && | |
9612 | ((wanted & CEPH_CAP_ANY_WR) == 0 || in->auth_cap)) { | |
9613 | check_caps(in, CHECK_CAPS_NODELAY); | |
9614 | return 0; | |
9615 | } | |
9616 | ||
9617 | int flags = 0; | |
9618 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
9619 | flags = O_RDWR; | |
9620 | else if (wanted & CEPH_CAP_FILE_RD) | |
9621 | flags = O_RDONLY; | |
9622 | else if (wanted & CEPH_CAP_FILE_WR) | |
9623 | flags = O_WRONLY; | |
9624 | ||
9625 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); | |
9626 | filepath path; | |
9627 | in->make_nosnap_relative_path(path); | |
9628 | req->set_filepath(path); | |
9629 | req->head.args.open.flags = flags; | |
9630 | req->head.args.open.pool = -1; | |
9631 | if (cct->_conf->client_debug_getattr_caps) | |
9632 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9633 | else | |
9634 | req->head.args.open.mask = 0; | |
9635 | req->set_inode(in); | |
9636 | ||
9637 | // duplicate in case Cap goes away; not sure if that race is a concern? | |
9638 | const UserPerm *pperm = in->get_best_perms(); | |
9639 | UserPerm perms; | |
9640 | if (pperm != NULL) | |
9641 | perms = *pperm; | |
9642 | int ret = make_request(req, perms); | |
9643 | return ret; | |
9644 | } | |
9645 | ||
b3b6e05e | 9646 | int Client::_close(int fd) |
7c673cae FG |
9647 | { |
9648 | ldout(cct, 3) << "close enter(" << fd << ")" << dendl; | |
7c673cae FG |
9649 | tout(cct) << "close" << std::endl; |
9650 | tout(cct) << fd << std::endl; | |
9651 | ||
9652 | Fh *fh = get_filehandle(fd); | |
9653 | if (!fh) | |
f67539c2 | 9654 | return -CEPHFS_EBADF; |
7c673cae FG |
9655 | int err = _release_fh(fh); |
9656 | fd_map.erase(fd); | |
9657 | put_fd(fd); | |
9658 | ldout(cct, 3) << "close exit(" << fd << ")" << dendl; | |
9659 | return err; | |
9660 | } | |
9661 | ||
b3b6e05e TL |
9662 | int Client::close(int fd) { |
9663 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9664 | if (!mref_reader.is_state_satisfied()) | |
9665 | return -CEPHFS_ENOTCONN; | |
9666 | ||
9667 | std::scoped_lock lock(client_lock); | |
9668 | return _close(fd); | |
9669 | } | |
7c673cae FG |
9670 | |
9671 | // ------------ | |
9672 | // read, write | |
9673 | ||
9674 | loff_t Client::lseek(int fd, loff_t offset, int whence) | |
9675 | { | |
f67539c2 TL |
9676 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9677 | if (!mref_reader.is_state_satisfied()) | |
9678 | return -CEPHFS_ENOTCONN; | |
9679 | ||
7c673cae FG |
9680 | tout(cct) << "lseek" << std::endl; |
9681 | tout(cct) << fd << std::endl; | |
9682 | tout(cct) << offset << std::endl; | |
9683 | tout(cct) << whence << std::endl; | |
9684 | ||
f67539c2 | 9685 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9686 | Fh *f = get_filehandle(fd); |
9687 | if (!f) | |
f67539c2 | 9688 | return -CEPHFS_EBADF; |
7c673cae FG |
9689 | #if defined(__linux__) && defined(O_PATH) |
9690 | if (f->flags & O_PATH) | |
f67539c2 | 9691 | return -CEPHFS_EBADF; |
7c673cae FG |
9692 | #endif |
9693 | return _lseek(f, offset, whence); | |
9694 | } | |
9695 | ||
9696 | loff_t Client::_lseek(Fh *f, loff_t offset, int whence) | |
9697 | { | |
9698 | Inode *in = f->inode.get(); | |
9f95a23c | 9699 | bool whence_check = false; |
11fdf7f2 | 9700 | loff_t pos = -1; |
7c673cae | 9701 | |
9f95a23c TL |
9702 | switch (whence) { |
9703 | case SEEK_END: | |
9704 | whence_check = true; | |
9705 | break; | |
9706 | ||
9707 | #ifdef SEEK_DATA | |
9708 | case SEEK_DATA: | |
9709 | whence_check = true; | |
9710 | break; | |
9711 | #endif | |
9712 | ||
9713 | #ifdef SEEK_HOLE | |
9714 | case SEEK_HOLE: | |
9715 | whence_check = true; | |
9716 | break; | |
9717 | #endif | |
9718 | } | |
9719 | ||
9720 | if (whence_check) { | |
9721 | int r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
9722 | if (r < 0) | |
92f5a8d4 | 9723 | return r; |
92f5a8d4 TL |
9724 | } |
9725 | ||
7c673cae FG |
9726 | switch (whence) { |
9727 | case SEEK_SET: | |
11fdf7f2 | 9728 | pos = offset; |
7c673cae FG |
9729 | break; |
9730 | ||
9731 | case SEEK_CUR: | |
92f5a8d4 | 9732 | pos = f->pos + offset; |
7c673cae FG |
9733 | break; |
9734 | ||
9735 | case SEEK_END: | |
11fdf7f2 | 9736 | pos = in->size + offset; |
7c673cae FG |
9737 | break; |
9738 | ||
9f95a23c | 9739 | #ifdef SEEK_DATA |
92f5a8d4 | 9740 | case SEEK_DATA: |
9f95a23c | 9741 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9742 | return -CEPHFS_ENXIO; |
92f5a8d4 TL |
9743 | pos = offset; |
9744 | break; | |
9f95a23c | 9745 | #endif |
92f5a8d4 | 9746 | |
9f95a23c | 9747 | #ifdef SEEK_HOLE |
92f5a8d4 | 9748 | case SEEK_HOLE: |
9f95a23c | 9749 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9750 | return -CEPHFS_ENXIO; |
9f95a23c | 9751 | pos = in->size; |
92f5a8d4 | 9752 | break; |
9f95a23c | 9753 | #endif |
92f5a8d4 | 9754 | |
7c673cae | 9755 | default: |
92f5a8d4 | 9756 | ldout(cct, 1) << __func__ << ": invalid whence value " << whence << dendl; |
f67539c2 | 9757 | return -CEPHFS_EINVAL; |
7c673cae FG |
9758 | } |
9759 | ||
11fdf7f2 | 9760 | if (pos < 0) { |
f67539c2 | 9761 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
9762 | } else { |
9763 | f->pos = pos; | |
9764 | } | |
9765 | ||
1adf2230 | 9766 | ldout(cct, 8) << "_lseek(" << f << ", " << offset << ", " << whence << ") = " << f->pos << dendl; |
7c673cae FG |
9767 | return f->pos; |
9768 | } | |
9769 | ||
9770 | ||
9771 | void Client::lock_fh_pos(Fh *f) | |
9772 | { | |
11fdf7f2 | 9773 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae FG |
9774 | |
9775 | if (f->pos_locked || !f->pos_waiters.empty()) { | |
9f95a23c | 9776 | ceph::condition_variable cond; |
7c673cae | 9777 | f->pos_waiters.push_back(&cond); |
11fdf7f2 | 9778 | ldout(cct, 10) << __func__ << " BLOCKING on " << f << dendl; |
9f95a23c TL |
9779 | std::unique_lock l{client_lock, std::adopt_lock}; |
9780 | cond.wait(l, [f, me=&cond] { | |
9781 | return !f->pos_locked && f->pos_waiters.front() == me; | |
9782 | }); | |
9783 | l.release(); | |
11fdf7f2 TL |
9784 | ldout(cct, 10) << __func__ << " UNBLOCKING on " << f << dendl; |
9785 | ceph_assert(f->pos_waiters.front() == &cond); | |
7c673cae FG |
9786 | f->pos_waiters.pop_front(); |
9787 | } | |
9788 | ||
9789 | f->pos_locked = true; | |
9790 | } | |
9791 | ||
9792 | void Client::unlock_fh_pos(Fh *f) | |
9793 | { | |
f67539c2 TL |
9794 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9795 | ||
11fdf7f2 | 9796 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae | 9797 | f->pos_locked = false; |
f67539c2 TL |
9798 | if (!f->pos_waiters.empty()) { |
9799 | // only wake up the oldest waiter | |
9800 | auto cond = f->pos_waiters.front(); | |
9801 | cond->notify_one(); | |
9802 | } | |
7c673cae FG |
9803 | } |
9804 | ||
9805 | int Client::uninline_data(Inode *in, Context *onfinish) | |
9806 | { | |
9807 | if (!in->inline_data.length()) { | |
9808 | onfinish->complete(0); | |
9809 | return 0; | |
9810 | } | |
9811 | ||
9812 | char oid_buf[32]; | |
9813 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (long long unsigned)in->ino); | |
9814 | object_t oid = oid_buf; | |
9815 | ||
9816 | ObjectOperation create_ops; | |
9817 | create_ops.create(false); | |
9818 | ||
9819 | objecter->mutate(oid, | |
9820 | OSDMap::file_to_object_locator(in->layout), | |
9821 | create_ops, | |
9822 | in->snaprealm->get_snap_context(), | |
9823 | ceph::real_clock::now(), | |
9824 | 0, | |
9825 | NULL); | |
9826 | ||
9827 | bufferlist inline_version_bl; | |
11fdf7f2 | 9828 | encode(in->inline_version, inline_version_bl); |
7c673cae FG |
9829 | |
9830 | ObjectOperation uninline_ops; | |
9831 | uninline_ops.cmpxattr("inline_version", | |
9832 | CEPH_OSD_CMPXATTR_OP_GT, | |
9833 | CEPH_OSD_CMPXATTR_MODE_U64, | |
9834 | inline_version_bl); | |
9835 | bufferlist inline_data = in->inline_data; | |
9836 | uninline_ops.write(0, inline_data, in->truncate_size, in->truncate_seq); | |
9837 | uninline_ops.setxattr("inline_version", stringify(in->inline_version)); | |
9838 | ||
9839 | objecter->mutate(oid, | |
9840 | OSDMap::file_to_object_locator(in->layout), | |
9841 | uninline_ops, | |
9842 | in->snaprealm->get_snap_context(), | |
9843 | ceph::real_clock::now(), | |
9844 | 0, | |
9845 | onfinish); | |
9846 | ||
9847 | return 0; | |
9848 | } | |
9849 | ||
9850 | // | |
9851 | ||
9852 | // blocking osd interface | |
9853 | ||
9854 | int Client::read(int fd, char *buf, loff_t size, loff_t offset) | |
9855 | { | |
f67539c2 TL |
9856 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9857 | if (!mref_reader.is_state_satisfied()) | |
9858 | return -CEPHFS_ENOTCONN; | |
9859 | ||
7c673cae FG |
9860 | tout(cct) << "read" << std::endl; |
9861 | tout(cct) << fd << std::endl; | |
9862 | tout(cct) << size << std::endl; | |
9863 | tout(cct) << offset << std::endl; | |
9864 | ||
f67539c2 | 9865 | std::unique_lock lock(client_lock); |
7c673cae FG |
9866 | Fh *f = get_filehandle(fd); |
9867 | if (!f) | |
f67539c2 | 9868 | return -CEPHFS_EBADF; |
7c673cae FG |
9869 | #if defined(__linux__) && defined(O_PATH) |
9870 | if (f->flags & O_PATH) | |
f67539c2 | 9871 | return -CEPHFS_EBADF; |
7c673cae FG |
9872 | #endif |
9873 | bufferlist bl; | |
11fdf7f2 TL |
9874 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
9875 | size = std::min(size, (loff_t)INT_MAX); | |
7c673cae FG |
9876 | int r = _read(f, offset, size, &bl); |
9877 | ldout(cct, 3) << "read(" << fd << ", " << (void*)buf << ", " << size << ", " << offset << ") = " << r << dendl; | |
9878 | if (r >= 0) { | |
f6b5b4d7 | 9879 | lock.unlock(); |
9f95a23c | 9880 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
9881 | r = bl.length(); |
9882 | } | |
9883 | return r; | |
9884 | } | |
9885 | ||
9886 | int Client::preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset) | |
9887 | { | |
9888 | if (iovcnt < 0) | |
f67539c2 | 9889 | return -CEPHFS_EINVAL; |
7c673cae FG |
9890 | return _preadv_pwritev(fd, iov, iovcnt, offset, false); |
9891 | } | |
9892 | ||
11fdf7f2 | 9893 | int64_t Client::_read(Fh *f, int64_t offset, uint64_t size, bufferlist *bl) |
7c673cae | 9894 | { |
f67539c2 TL |
9895 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9896 | ||
11fdf7f2 TL |
9897 | int want, have = 0; |
9898 | bool movepos = false; | |
9899 | std::unique_ptr<C_SaferCond> onuninline; | |
adb31ebb | 9900 | int64_t rc = 0; |
11fdf7f2 | 9901 | const auto& conf = cct->_conf; |
7c673cae | 9902 | Inode *in = f->inode.get(); |
11fdf7f2 TL |
9903 | utime_t lat; |
9904 | utime_t start = ceph_clock_now(); | |
7c673cae FG |
9905 | |
9906 | if ((f->mode & CEPH_FILE_MODE_RD) == 0) | |
f67539c2 | 9907 | return -CEPHFS_EBADF; |
7c673cae FG |
9908 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
9909 | ||
7c673cae FG |
9910 | if (offset < 0) { |
9911 | lock_fh_pos(f); | |
9912 | offset = f->pos; | |
9913 | movepos = true; | |
9914 | } | |
9915 | loff_t start_pos = offset; | |
9916 | ||
9917 | if (in->inline_version == 0) { | |
adb31ebb | 9918 | auto r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); |
c07f9fc5 | 9919 | if (r < 0) { |
adb31ebb | 9920 | rc = r; |
11fdf7f2 | 9921 | goto done; |
c07f9fc5 | 9922 | } |
11fdf7f2 | 9923 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
9924 | } |
9925 | ||
9926 | retry: | |
11fdf7f2 TL |
9927 | if (f->mode & CEPH_FILE_MODE_LAZY) |
9928 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
9929 | else | |
9930 | want = CEPH_CAP_FILE_CACHE; | |
adb31ebb TL |
9931 | { |
9932 | auto r = get_caps(f, CEPH_CAP_FILE_RD, want, &have, -1); | |
9933 | if (r < 0) { | |
9934 | rc = r; | |
9935 | goto done; | |
9936 | } | |
c07f9fc5 | 9937 | } |
7c673cae | 9938 | if (f->flags & O_DIRECT) |
11fdf7f2 | 9939 | have &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
9940 | |
9941 | if (in->inline_version < CEPH_INLINE_NONE) { | |
9942 | if (!(have & CEPH_CAP_FILE_CACHE)) { | |
11fdf7f2 TL |
9943 | onuninline.reset(new C_SaferCond("Client::_read_uninline_data flock")); |
9944 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
9945 | } else { |
9946 | uint32_t len = in->inline_data.length(); | |
7c673cae FG |
9947 | uint64_t endoff = offset + size; |
9948 | if (endoff > in->size) | |
9949 | endoff = in->size; | |
9950 | ||
9951 | if (offset < len) { | |
9952 | if (endoff <= len) { | |
9953 | bl->substr_of(in->inline_data, offset, endoff - offset); | |
9954 | } else { | |
9955 | bl->substr_of(in->inline_data, offset, len - offset); | |
9956 | bl->append_zero(endoff - len); | |
9957 | } | |
adb31ebb | 9958 | rc = endoff - offset; |
7c673cae FG |
9959 | } else if ((uint64_t)offset < endoff) { |
9960 | bl->append_zero(endoff - offset); | |
adb31ebb | 9961 | rc = endoff - offset; |
11fdf7f2 | 9962 | } else { |
adb31ebb | 9963 | rc = 0; |
7c673cae | 9964 | } |
7c673cae FG |
9965 | goto success; |
9966 | } | |
9967 | } | |
9968 | ||
9969 | if (!conf->client_debug_force_sync_read && | |
11fdf7f2 TL |
9970 | conf->client_oc && |
9971 | (have & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
9972 | |
9973 | if (f->flags & O_RSYNC) { | |
9974 | _flush_range(in, offset, size); | |
9975 | } | |
adb31ebb TL |
9976 | rc = _read_async(f, offset, size, bl); |
9977 | if (rc < 0) | |
7c673cae FG |
9978 | goto done; |
9979 | } else { | |
9980 | if (f->flags & O_DIRECT) | |
9981 | _flush_range(in, offset, size); | |
9982 | ||
9983 | bool checkeof = false; | |
adb31ebb TL |
9984 | rc = _read_sync(f, offset, size, bl, &checkeof); |
9985 | if (rc < 0) | |
7c673cae FG |
9986 | goto done; |
9987 | if (checkeof) { | |
adb31ebb TL |
9988 | offset += rc; |
9989 | size -= rc; | |
7c673cae FG |
9990 | |
9991 | put_cap_ref(in, CEPH_CAP_FILE_RD); | |
9992 | have = 0; | |
9993 | // reverify size | |
adb31ebb TL |
9994 | { |
9995 | auto r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
9996 | if (r < 0) { | |
9997 | rc = r; | |
9998 | goto done; | |
9999 | } | |
10000 | } | |
7c673cae FG |
10001 | |
10002 | // eof? short read. | |
10003 | if ((uint64_t)offset < in->size) | |
10004 | goto retry; | |
10005 | } | |
10006 | } | |
10007 | ||
10008 | success: | |
adb31ebb | 10009 | ceph_assert(rc >= 0); |
a4b75251 | 10010 | update_read_io_size(bl->length()); |
7c673cae FG |
10011 | if (movepos) { |
10012 | // adjust fd pos | |
adb31ebb | 10013 | f->pos = start_pos + rc; |
7c673cae | 10014 | } |
11fdf7f2 TL |
10015 | |
10016 | lat = ceph_clock_now(); | |
10017 | lat -= start; | |
10018 | logger->tinc(l_c_read, lat); | |
7c673cae FG |
10019 | |
10020 | done: | |
10021 | // done! | |
11fdf7f2 | 10022 | |
7c673cae | 10023 | if (onuninline) { |
9f95a23c | 10024 | client_lock.unlock(); |
11fdf7f2 | 10025 | int ret = onuninline->wait(); |
9f95a23c | 10026 | client_lock.lock(); |
f67539c2 | 10027 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
10028 | in->inline_data.clear(); |
10029 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 10030 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10031 | check_caps(in, 0); |
10032 | } else | |
adb31ebb | 10033 | rc = ret; |
7c673cae | 10034 | } |
11fdf7f2 | 10035 | if (have) { |
7c673cae | 10036 | put_cap_ref(in, CEPH_CAP_FILE_RD); |
11fdf7f2 TL |
10037 | } |
10038 | if (movepos) { | |
10039 | unlock_fh_pos(f); | |
10040 | } | |
adb31ebb | 10041 | return rc; |
7c673cae FG |
10042 | } |
10043 | ||
10044 | Client::C_Readahead::C_Readahead(Client *c, Fh *f) : | |
10045 | client(c), f(f) { | |
10046 | f->get(); | |
10047 | f->readahead.inc_pending(); | |
10048 | } | |
10049 | ||
10050 | Client::C_Readahead::~C_Readahead() { | |
10051 | f->readahead.dec_pending(); | |
10052 | client->_put_fh(f); | |
10053 | } | |
10054 | ||
10055 | void Client::C_Readahead::finish(int r) { | |
10056 | lgeneric_subdout(client->cct, client, 20) << "client." << client->get_nodeid() << " " << "C_Readahead on " << f->inode << dendl; | |
10057 | client->put_cap_ref(f->inode.get(), CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
a4b75251 TL |
10058 | if (r > 0) { |
10059 | client->update_read_io_size(r); | |
10060 | } | |
7c673cae FG |
10061 | } |
10062 | ||
10063 | int Client::_read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl) | |
10064 | { | |
f67539c2 TL |
10065 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10066 | ||
11fdf7f2 | 10067 | const auto& conf = cct->_conf; |
7c673cae FG |
10068 | Inode *in = f->inode.get(); |
10069 | ||
11fdf7f2 | 10070 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
10071 | |
10072 | // trim read based on file size? | |
10073 | if (off >= in->size) | |
10074 | return 0; | |
10075 | if (len == 0) | |
10076 | return 0; | |
10077 | if (off + len > in->size) { | |
10078 | len = in->size - off; | |
10079 | } | |
10080 | ||
10081 | ldout(cct, 10) << " min_bytes=" << f->readahead.get_min_readahead_size() | |
10082 | << " max_bytes=" << f->readahead.get_max_readahead_size() | |
10083 | << " max_periods=" << conf->client_readahead_max_periods << dendl; | |
10084 | ||
10085 | // read (and possibly block) | |
11fdf7f2 TL |
10086 | int r = 0; |
10087 | C_SaferCond onfinish("Client::_read_async flock"); | |
7c673cae | 10088 | r = objectcacher->file_read(&in->oset, &in->layout, in->snapid, |
11fdf7f2 | 10089 | off, len, bl, 0, &onfinish); |
7c673cae FG |
10090 | if (r == 0) { |
10091 | get_cap_ref(in, CEPH_CAP_FILE_CACHE); | |
9f95a23c | 10092 | client_lock.unlock(); |
11fdf7f2 | 10093 | r = onfinish.wait(); |
9f95a23c | 10094 | client_lock.lock(); |
7c673cae | 10095 | put_cap_ref(in, CEPH_CAP_FILE_CACHE); |
a4b75251 | 10096 | update_read_io_size(bl->length()); |
7c673cae FG |
10097 | } |
10098 | ||
10099 | if(f->readahead.get_min_readahead_size() > 0) { | |
10100 | pair<uint64_t, uint64_t> readahead_extent = f->readahead.update(off, len, in->size); | |
10101 | if (readahead_extent.second > 0) { | |
10102 | ldout(cct, 20) << "readahead " << readahead_extent.first << "~" << readahead_extent.second | |
10103 | << " (caller wants " << off << "~" << len << ")" << dendl; | |
10104 | Context *onfinish2 = new C_Readahead(this, f); | |
10105 | int r2 = objectcacher->file_read(&in->oset, &in->layout, in->snapid, | |
10106 | readahead_extent.first, readahead_extent.second, | |
10107 | NULL, 0, onfinish2); | |
10108 | if (r2 == 0) { | |
10109 | ldout(cct, 20) << "readahead initiated, c " << onfinish2 << dendl; | |
10110 | get_cap_ref(in, CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
10111 | } else { | |
10112 | ldout(cct, 20) << "readahead was no-op, already cached" << dendl; | |
10113 | delete onfinish2; | |
10114 | } | |
10115 | } | |
10116 | } | |
10117 | ||
10118 | return r; | |
10119 | } | |
10120 | ||
10121 | int Client::_read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, | |
10122 | bool *checkeof) | |
10123 | { | |
f67539c2 TL |
10124 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10125 | ||
7c673cae FG |
10126 | Inode *in = f->inode.get(); |
10127 | uint64_t pos = off; | |
10128 | int left = len; | |
10129 | int read = 0; | |
10130 | ||
11fdf7f2 | 10131 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae | 10132 | |
f67539c2 TL |
10133 | // 0 success, 1 continue and < 0 error happen. |
10134 | auto wait_and_copy = [&](C_SaferCond &onfinish, bufferlist &tbl, int wanted) { | |
11fdf7f2 | 10135 | int r = onfinish.wait(); |
7c673cae FG |
10136 | |
10137 | // if we get ENOENT from OSD, assume 0 bytes returned | |
f67539c2 | 10138 | if (r == -CEPHFS_ENOENT) |
7c673cae FG |
10139 | r = 0; |
10140 | if (r < 0) | |
10141 | return r; | |
f67539c2 | 10142 | |
7c673cae FG |
10143 | if (tbl.length()) { |
10144 | r = tbl.length(); | |
10145 | ||
10146 | read += r; | |
10147 | pos += r; | |
10148 | left -= r; | |
10149 | bl->claim_append(tbl); | |
10150 | } | |
10151 | // short read? | |
10152 | if (r >= 0 && r < wanted) { | |
10153 | if (pos < in->size) { | |
10154 | // zero up to known EOF | |
10155 | int64_t some = in->size - pos; | |
10156 | if (some > left) | |
10157 | some = left; | |
11fdf7f2 TL |
10158 | auto z = buffer::ptr_node::create(some); |
10159 | z->zero(); | |
10160 | bl->push_back(std::move(z)); | |
7c673cae FG |
10161 | read += some; |
10162 | pos += some; | |
10163 | left -= some; | |
10164 | if (left == 0) | |
f67539c2 | 10165 | return 0; |
7c673cae FG |
10166 | } |
10167 | ||
10168 | *checkeof = true; | |
f67539c2 | 10169 | return 0; |
7c673cae | 10170 | } |
f67539c2 TL |
10171 | return 1; |
10172 | }; | |
7c673cae | 10173 | |
f67539c2 TL |
10174 | while (left > 0) { |
10175 | C_SaferCond onfinish("Client::_read_sync flock"); | |
10176 | bufferlist tbl; | |
7c673cae | 10177 | |
f67539c2 TL |
10178 | int wanted = left; |
10179 | filer->read_trunc(in->ino, &in->layout, in->snapid, | |
10180 | pos, left, &tbl, 0, | |
10181 | in->truncate_size, in->truncate_seq, | |
10182 | &onfinish); | |
10183 | client_lock.unlock(); | |
10184 | int r = wait_and_copy(onfinish, tbl, wanted); | |
10185 | client_lock.lock(); | |
10186 | if (!r) | |
10187 | return read; | |
10188 | if (r < 0) | |
10189 | return r; | |
7c673cae | 10190 | } |
f67539c2 | 10191 | return read; |
7c673cae FG |
10192 | } |
10193 | ||
10194 | int Client::write(int fd, const char *buf, loff_t size, loff_t offset) | |
10195 | { | |
f67539c2 TL |
10196 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10197 | if (!mref_reader.is_state_satisfied()) | |
10198 | return -CEPHFS_ENOTCONN; | |
10199 | ||
7c673cae FG |
10200 | tout(cct) << "write" << std::endl; |
10201 | tout(cct) << fd << std::endl; | |
10202 | tout(cct) << size << std::endl; | |
10203 | tout(cct) << offset << std::endl; | |
10204 | ||
f67539c2 | 10205 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10206 | Fh *fh = get_filehandle(fd); |
10207 | if (!fh) | |
f67539c2 | 10208 | return -CEPHFS_EBADF; |
7c673cae FG |
10209 | #if defined(__linux__) && defined(O_PATH) |
10210 | if (fh->flags & O_PATH) | |
f67539c2 | 10211 | return -CEPHFS_EBADF; |
7c673cae | 10212 | #endif |
11fdf7f2 TL |
10213 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
10214 | size = std::min(size, (loff_t)INT_MAX); | |
10215 | int r = _write(fh, offset, size, buf, NULL, false); | |
7c673cae FG |
10216 | ldout(cct, 3) << "write(" << fd << ", \"...\", " << size << ", " << offset << ") = " << r << dendl; |
10217 | return r; | |
10218 | } | |
10219 | ||
10220 | int Client::pwritev(int fd, const struct iovec *iov, int iovcnt, int64_t offset) | |
10221 | { | |
10222 | if (iovcnt < 0) | |
f67539c2 | 10223 | return -CEPHFS_EINVAL; |
7c673cae FG |
10224 | return _preadv_pwritev(fd, iov, iovcnt, offset, true); |
10225 | } | |
10226 | ||
11fdf7f2 | 10227 | int64_t Client::_preadv_pwritev_locked(Fh *fh, const struct iovec *iov, |
20effc67 TL |
10228 | unsigned iovcnt, int64_t offset, |
10229 | bool write, bool clamp_to_int) | |
7c673cae | 10230 | { |
20effc67 TL |
10231 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10232 | ||
7c673cae FG |
10233 | #if defined(__linux__) && defined(O_PATH) |
10234 | if (fh->flags & O_PATH) | |
f67539c2 | 10235 | return -CEPHFS_EBADF; |
7c673cae FG |
10236 | #endif |
10237 | loff_t totallen = 0; | |
10238 | for (unsigned i = 0; i < iovcnt; i++) { | |
10239 | totallen += iov[i].iov_len; | |
10240 | } | |
11fdf7f2 TL |
10241 | |
10242 | /* | |
10243 | * Some of the API functions take 64-bit size values, but only return | |
10244 | * 32-bit signed integers. Clamp the I/O sizes in those functions so that | |
10245 | * we don't do I/Os larger than the values we can return. | |
10246 | */ | |
10247 | if (clamp_to_int) { | |
10248 | totallen = std::min(totallen, (loff_t)INT_MAX); | |
10249 | } | |
7c673cae | 10250 | if (write) { |
11fdf7f2 TL |
10251 | int64_t w = _write(fh, offset, totallen, NULL, iov, iovcnt); |
10252 | ldout(cct, 3) << "pwritev(" << fh << ", \"...\", " << totallen << ", " << offset << ") = " << w << dendl; | |
7c673cae FG |
10253 | return w; |
10254 | } else { | |
10255 | bufferlist bl; | |
11fdf7f2 TL |
10256 | int64_t r = _read(fh, offset, totallen, &bl); |
10257 | ldout(cct, 3) << "preadv(" << fh << ", " << offset << ") = " << r << dendl; | |
7c673cae FG |
10258 | if (r <= 0) |
10259 | return r; | |
10260 | ||
20effc67 | 10261 | client_lock.unlock(); |
9f95a23c | 10262 | auto iter = bl.cbegin(); |
7c673cae FG |
10263 | for (unsigned j = 0, resid = r; j < iovcnt && resid > 0; j++) { |
10264 | /* | |
f67539c2 TL |
10265 | * This piece of code aims to handle the case that bufferlist |
10266 | * does not have enough data to fill in the iov | |
7c673cae | 10267 | */ |
9f95a23c TL |
10268 | const auto round_size = std::min<unsigned>(resid, iov[j].iov_len); |
10269 | iter.copy(round_size, reinterpret_cast<char*>(iov[j].iov_base)); | |
10270 | resid -= round_size; | |
10271 | /* iter is self-updating */ | |
7c673cae | 10272 | } |
20effc67 | 10273 | client_lock.lock(); |
f67539c2 | 10274 | return r; |
7c673cae FG |
10275 | } |
10276 | } | |
10277 | ||
11fdf7f2 TL |
10278 | int Client::_preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write) |
10279 | { | |
f67539c2 TL |
10280 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10281 | if (!mref_reader.is_state_satisfied()) | |
10282 | return -CEPHFS_ENOTCONN; | |
10283 | ||
11fdf7f2 TL |
10284 | tout(cct) << fd << std::endl; |
10285 | tout(cct) << offset << std::endl; | |
10286 | ||
20effc67 | 10287 | std::scoped_lock cl(client_lock); |
11fdf7f2 TL |
10288 | Fh *fh = get_filehandle(fd); |
10289 | if (!fh) | |
f67539c2 | 10290 | return -CEPHFS_EBADF; |
20effc67 | 10291 | return _preadv_pwritev_locked(fh, iov, iovcnt, offset, write, true); |
11fdf7f2 TL |
10292 | } |
10293 | ||
10294 | int64_t Client::_write(Fh *f, int64_t offset, uint64_t size, const char *buf, | |
10295 | const struct iovec *iov, int iovcnt) | |
7c673cae | 10296 | { |
f67539c2 TL |
10297 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10298 | ||
f64942e4 AA |
10299 | uint64_t fpos = 0; |
10300 | ||
7c673cae | 10301 | if ((uint64_t)(offset+size) > mdsmap->get_max_filesize()) //too large! |
f67539c2 | 10302 | return -CEPHFS_EFBIG; |
7c673cae FG |
10303 | |
10304 | //ldout(cct, 7) << "write fh " << fh << " size " << size << " offset " << offset << dendl; | |
10305 | Inode *in = f->inode.get(); | |
10306 | ||
10307 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
f67539c2 | 10308 | return -CEPHFS_ENOSPC; |
7c673cae FG |
10309 | } |
10310 | ||
11fdf7f2 | 10311 | ceph_assert(in->snapid == CEPH_NOSNAP); |
7c673cae FG |
10312 | |
10313 | // was Fh opened as writeable? | |
10314 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 10315 | return -CEPHFS_EBADF; |
7c673cae | 10316 | |
7c673cae FG |
10317 | // use/adjust fd pos? |
10318 | if (offset < 0) { | |
10319 | lock_fh_pos(f); | |
10320 | /* | |
10321 | * FIXME: this is racy in that we may block _after_ this point waiting for caps, and size may | |
10322 | * change out from under us. | |
10323 | */ | |
10324 | if (f->flags & O_APPEND) { | |
9f95a23c | 10325 | auto r = _lseek(f, 0, SEEK_END); |
7c673cae FG |
10326 | if (r < 0) { |
10327 | unlock_fh_pos(f); | |
10328 | return r; | |
10329 | } | |
10330 | } | |
10331 | offset = f->pos; | |
f64942e4 | 10332 | fpos = offset+size; |
7c673cae FG |
10333 | unlock_fh_pos(f); |
10334 | } | |
10335 | ||
11fdf7f2 TL |
10336 | // check quota |
10337 | uint64_t endoff = offset + size; | |
10338 | if (endoff > in->size && is_quota_bytes_exceeded(in, endoff - in->size, | |
10339 | f->actor_perms)) { | |
f67539c2 | 10340 | return -CEPHFS_EDQUOT; |
11fdf7f2 TL |
10341 | } |
10342 | ||
7c673cae FG |
10343 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
10344 | ||
10345 | ldout(cct, 10) << "cur file size is " << in->size << dendl; | |
10346 | ||
10347 | // time it. | |
10348 | utime_t start = ceph_clock_now(); | |
10349 | ||
10350 | if (in->inline_version == 0) { | |
10351 | int r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); | |
10352 | if (r < 0) | |
10353 | return r; | |
11fdf7f2 | 10354 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
10355 | } |
10356 | ||
10357 | // copy into fresh buffer (since our write may be resub, async) | |
10358 | bufferlist bl; | |
10359 | if (buf) { | |
10360 | if (size > 0) | |
10361 | bl.append(buf, size); | |
10362 | } else if (iov){ | |
10363 | for (int i = 0; i < iovcnt; i++) { | |
10364 | if (iov[i].iov_len > 0) { | |
10365 | bl.append((const char *)iov[i].iov_base, iov[i].iov_len); | |
10366 | } | |
10367 | } | |
10368 | } | |
10369 | ||
10370 | utime_t lat; | |
10371 | uint64_t totalwritten; | |
11fdf7f2 TL |
10372 | int want, have; |
10373 | if (f->mode & CEPH_FILE_MODE_LAZY) | |
10374 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
10375 | else | |
10376 | want = CEPH_CAP_FILE_BUFFER; | |
f6b5b4d7 | 10377 | int r = get_caps(f, CEPH_CAP_FILE_WR|CEPH_CAP_AUTH_SHARED, want, &have, endoff); |
7c673cae FG |
10378 | if (r < 0) |
10379 | return r; | |
10380 | ||
10381 | /* clear the setuid/setgid bits, if any */ | |
181888fb | 10382 | if (unlikely(in->mode & (S_ISUID|S_ISGID)) && size > 0) { |
7c673cae FG |
10383 | struct ceph_statx stx = { 0 }; |
10384 | ||
10385 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10386 | r = __setattrx(in, &stx, CEPH_SETATTR_KILL_SGUID, f->actor_perms); | |
10387 | if (r < 0) | |
10388 | return r; | |
10389 | } else { | |
10390 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10391 | } | |
10392 | ||
10393 | if (f->flags & O_DIRECT) | |
11fdf7f2 | 10394 | have &= ~(CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
10395 | |
10396 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
10397 | ||
11fdf7f2 TL |
10398 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
10399 | ||
7c673cae FG |
10400 | if (in->inline_version < CEPH_INLINE_NONE) { |
10401 | if (endoff > cct->_conf->client_max_inline_size || | |
10402 | endoff > CEPH_INLINE_MAX_SIZE || | |
10403 | !(have & CEPH_CAP_FILE_BUFFER)) { | |
11fdf7f2 TL |
10404 | onuninline.reset(new C_SaferCond("Client::_write_uninline_data flock")); |
10405 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
10406 | } else { |
10407 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10408 | ||
10409 | uint32_t len = in->inline_data.length(); | |
10410 | ||
10411 | if (endoff < len) | |
9f95a23c | 10412 | in->inline_data.begin(endoff).copy(len - endoff, bl); // XXX |
7c673cae FG |
10413 | |
10414 | if (offset < len) | |
10415 | in->inline_data.splice(offset, len - offset); | |
10416 | else if (offset > len) | |
10417 | in->inline_data.append_zero(offset - len); | |
10418 | ||
10419 | in->inline_data.append(bl); | |
10420 | in->inline_version++; | |
10421 | ||
10422 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10423 | ||
10424 | goto success; | |
10425 | } | |
10426 | } | |
10427 | ||
11fdf7f2 TL |
10428 | if (cct->_conf->client_oc && |
10429 | (have & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
10430 | // do buffered write |
10431 | if (!in->oset.dirty_or_tx) | |
10432 | get_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
10433 | ||
10434 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10435 | ||
10436 | // async, caching, non-blocking. | |
10437 | r = objectcacher->file_write(&in->oset, &in->layout, | |
10438 | in->snaprealm->get_snap_context(), | |
10439 | offset, size, bl, ceph::real_clock::now(), | |
10440 | 0); | |
10441 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10442 | ||
10443 | if (r < 0) | |
10444 | goto done; | |
10445 | ||
10446 | // flush cached write if O_SYNC is set on file fh | |
10447 | // O_DSYNC == O_SYNC on linux < 2.6.33 | |
10448 | // O_SYNC = __O_SYNC | O_DSYNC on linux >= 2.6.33 | |
10449 | if ((f->flags & O_SYNC) || (f->flags & O_DSYNC)) { | |
10450 | _flush_range(in, offset, size); | |
10451 | } | |
10452 | } else { | |
10453 | if (f->flags & O_DIRECT) | |
10454 | _flush_range(in, offset, size); | |
10455 | ||
10456 | // simple, non-atomic sync write | |
11fdf7f2 | 10457 | C_SaferCond onfinish("Client::_write flock"); |
f67539c2 | 10458 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
10459 | |
10460 | filer->write_trunc(in->ino, &in->layout, in->snaprealm->get_snap_context(), | |
10461 | offset, size, bl, ceph::real_clock::now(), 0, | |
10462 | in->truncate_size, in->truncate_seq, | |
11fdf7f2 | 10463 | &onfinish); |
9f95a23c | 10464 | client_lock.unlock(); |
f6b5b4d7 | 10465 | r = onfinish.wait(); |
9f95a23c | 10466 | client_lock.lock(); |
f67539c2 | 10467 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
f6b5b4d7 TL |
10468 | if (r < 0) |
10469 | goto done; | |
7c673cae FG |
10470 | } |
10471 | ||
10472 | // if we get here, write was successful, update client metadata | |
10473 | success: | |
a4b75251 | 10474 | update_write_io_size(size); |
7c673cae FG |
10475 | // time |
10476 | lat = ceph_clock_now(); | |
10477 | lat -= start; | |
10478 | logger->tinc(l_c_wrlat, lat); | |
10479 | ||
f64942e4 AA |
10480 | if (fpos) { |
10481 | lock_fh_pos(f); | |
10482 | f->pos = fpos; | |
10483 | unlock_fh_pos(f); | |
10484 | } | |
7c673cae | 10485 | totalwritten = size; |
11fdf7f2 | 10486 | r = (int64_t)totalwritten; |
7c673cae FG |
10487 | |
10488 | // extend file? | |
10489 | if (totalwritten + offset > in->size) { | |
10490 | in->size = totalwritten + offset; | |
28e407b8 | 10491 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 10492 | |
11fdf7f2 | 10493 | if (is_quota_bytes_approaching(in, f->actor_perms)) { |
7c673cae | 10494 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
10495 | } else if (is_max_size_approaching(in)) { |
10496 | check_caps(in, 0); | |
7c673cae FG |
10497 | } |
10498 | ||
10499 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", extending file size" << dendl; | |
10500 | } else { | |
10501 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", leaving file size at " << in->size << dendl; | |
10502 | } | |
10503 | ||
10504 | // mtime | |
91327a77 | 10505 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 10506 | in->change_attr++; |
28e407b8 | 10507 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10508 | |
10509 | done: | |
10510 | ||
11fdf7f2 | 10511 | if (nullptr != onuninline) { |
9f95a23c | 10512 | client_lock.unlock(); |
11fdf7f2 | 10513 | int uninline_ret = onuninline->wait(); |
9f95a23c | 10514 | client_lock.lock(); |
7c673cae | 10515 | |
f67539c2 | 10516 | if (uninline_ret >= 0 || uninline_ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
10517 | in->inline_data.clear(); |
10518 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 10519 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10520 | check_caps(in, 0); |
10521 | } else | |
10522 | r = uninline_ret; | |
10523 | } | |
10524 | ||
10525 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
10526 | return r; | |
10527 | } | |
10528 | ||
10529 | int Client::_flush(Fh *f) | |
10530 | { | |
10531 | Inode *in = f->inode.get(); | |
10532 | int err = f->take_async_err(); | |
10533 | if (err != 0) { | |
10534 | ldout(cct, 1) << __func__ << ": " << f << " on inode " << *in << " caught async_err = " | |
10535 | << cpp_strerror(err) << dendl; | |
10536 | } else { | |
10537 | ldout(cct, 10) << __func__ << ": " << f << " on inode " << *in << " no async_err state" << dendl; | |
10538 | } | |
10539 | ||
10540 | return err; | |
10541 | } | |
10542 | ||
10543 | int Client::truncate(const char *relpath, loff_t length, const UserPerm& perms) | |
10544 | { | |
10545 | struct ceph_statx stx; | |
10546 | stx.stx_size = length; | |
10547 | return setattrx(relpath, &stx, CEPH_SETATTR_SIZE, perms); | |
10548 | } | |
10549 | ||
10550 | int Client::ftruncate(int fd, loff_t length, const UserPerm& perms) | |
10551 | { | |
f67539c2 TL |
10552 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10553 | if (!mref_reader.is_state_satisfied()) | |
10554 | return -CEPHFS_ENOTCONN; | |
10555 | ||
11fdf7f2 | 10556 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
10557 | tout(cct) << fd << std::endl; |
10558 | tout(cct) << length << std::endl; | |
10559 | ||
f67539c2 | 10560 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10561 | Fh *f = get_filehandle(fd); |
10562 | if (!f) | |
f67539c2 | 10563 | return -CEPHFS_EBADF; |
7c673cae FG |
10564 | #if defined(__linux__) && defined(O_PATH) |
10565 | if (f->flags & O_PATH) | |
f67539c2 | 10566 | return -CEPHFS_EBADF; |
7c673cae | 10567 | #endif |
adb31ebb | 10568 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) |
f67539c2 | 10569 | return -CEPHFS_EBADF; |
7c673cae FG |
10570 | struct stat attr; |
10571 | attr.st_size = length; | |
10572 | return _setattr(f->inode, &attr, CEPH_SETATTR_SIZE, perms); | |
10573 | } | |
10574 | ||
10575 | int Client::fsync(int fd, bool syncdataonly) | |
10576 | { | |
f67539c2 TL |
10577 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10578 | if (!mref_reader.is_state_satisfied()) | |
10579 | return -CEPHFS_ENOTCONN; | |
10580 | ||
7c673cae FG |
10581 | tout(cct) << "fsync" << std::endl; |
10582 | tout(cct) << fd << std::endl; | |
10583 | tout(cct) << syncdataonly << std::endl; | |
10584 | ||
f67539c2 | 10585 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10586 | Fh *f = get_filehandle(fd); |
10587 | if (!f) | |
f67539c2 | 10588 | return -CEPHFS_EBADF; |
7c673cae FG |
10589 | #if defined(__linux__) && defined(O_PATH) |
10590 | if (f->flags & O_PATH) | |
f67539c2 | 10591 | return -CEPHFS_EBADF; |
7c673cae FG |
10592 | #endif |
10593 | int r = _fsync(f, syncdataonly); | |
10594 | if (r == 0) { | |
10595 | // The IOs in this fsync were okay, but maybe something happened | |
10596 | // in the background that we shoudl be reporting? | |
10597 | r = f->take_async_err(); | |
1adf2230 | 10598 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly |
7c673cae FG |
10599 | << ") = 0, async_err = " << r << dendl; |
10600 | } else { | |
10601 | // Assume that an error we encountered during fsync, even reported | |
10602 | // synchronously, would also have applied the error to the Fh, and we | |
10603 | // should clear it here to avoid returning the same error again on next | |
10604 | // call. | |
1adf2230 | 10605 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly << ") = " |
7c673cae FG |
10606 | << r << dendl; |
10607 | f->take_async_err(); | |
10608 | } | |
10609 | return r; | |
10610 | } | |
10611 | ||
10612 | int Client::_fsync(Inode *in, bool syncdataonly) | |
10613 | { | |
f67539c2 TL |
10614 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10615 | ||
7c673cae | 10616 | int r = 0; |
11fdf7f2 | 10617 | std::unique_ptr<C_SaferCond> object_cacher_completion = nullptr; |
7c673cae FG |
10618 | ceph_tid_t flush_tid = 0; |
10619 | InodeRef tmp_ref; | |
11fdf7f2 TL |
10620 | utime_t lat; |
10621 | utime_t start = ceph_clock_now(); | |
7c673cae | 10622 | |
1adf2230 | 10623 | ldout(cct, 8) << "_fsync on " << *in << " " << (syncdataonly ? "(dataonly)":"(data+metadata)") << dendl; |
7c673cae FG |
10624 | |
10625 | if (cct->_conf->client_oc) { | |
11fdf7f2 TL |
10626 | object_cacher_completion.reset(new C_SaferCond("Client::_fsync::lock")); |
10627 | tmp_ref = in; // take a reference; C_SaferCond doesn't and _flush won't either | |
10628 | _flush(in, object_cacher_completion.get()); | |
7c673cae FG |
10629 | ldout(cct, 15) << "using return-valued form of _fsync" << dendl; |
10630 | } | |
10631 | ||
10632 | if (!syncdataonly && in->dirty_caps) { | |
10633 | check_caps(in, CHECK_CAPS_NODELAY|CHECK_CAPS_SYNCHRONOUS); | |
10634 | if (in->flushing_caps) | |
10635 | flush_tid = last_flush_tid; | |
10636 | } else ldout(cct, 10) << "no metadata needs to commit" << dendl; | |
10637 | ||
10638 | if (!syncdataonly && !in->unsafe_ops.empty()) { | |
522d829b | 10639 | flush_mdlog_sync(in); |
28e407b8 | 10640 | |
7c673cae FG |
10641 | MetaRequest *req = in->unsafe_ops.back(); |
10642 | ldout(cct, 15) << "waiting on unsafe requests, last tid " << req->get_tid() << dendl; | |
10643 | ||
10644 | req->get(); | |
10645 | wait_on_list(req->waitfor_safe); | |
10646 | put_request(req); | |
10647 | } | |
10648 | ||
11fdf7f2 | 10649 | if (nullptr != object_cacher_completion) { // wait on a real reply instead of guessing |
9f95a23c | 10650 | client_lock.unlock(); |
7c673cae | 10651 | ldout(cct, 15) << "waiting on data to flush" << dendl; |
11fdf7f2 | 10652 | r = object_cacher_completion->wait(); |
9f95a23c | 10653 | client_lock.lock(); |
7c673cae FG |
10654 | ldout(cct, 15) << "got " << r << " from flush writeback" << dendl; |
10655 | } else { | |
10656 | // FIXME: this can starve | |
10657 | while (in->cap_refs[CEPH_CAP_FILE_BUFFER] > 0) { | |
10658 | ldout(cct, 10) << "ino " << in->ino << " has " << in->cap_refs[CEPH_CAP_FILE_BUFFER] | |
10659 | << " uncommitted, waiting" << dendl; | |
10660 | wait_on_list(in->waitfor_commit); | |
10661 | } | |
10662 | } | |
10663 | ||
10664 | if (!r) { | |
10665 | if (flush_tid > 0) | |
10666 | wait_sync_caps(in, flush_tid); | |
10667 | ||
10668 | ldout(cct, 10) << "ino " << in->ino << " has no uncommitted writes" << dendl; | |
10669 | } else { | |
1adf2230 | 10670 | ldout(cct, 8) << "ino " << in->ino << " failed to commit to disk! " |
7c673cae FG |
10671 | << cpp_strerror(-r) << dendl; |
10672 | } | |
11fdf7f2 TL |
10673 | |
10674 | lat = ceph_clock_now(); | |
10675 | lat -= start; | |
10676 | logger->tinc(l_c_fsync, lat); | |
7c673cae FG |
10677 | |
10678 | return r; | |
10679 | } | |
10680 | ||
10681 | int Client::_fsync(Fh *f, bool syncdataonly) | |
10682 | { | |
1adf2230 | 10683 | ldout(cct, 8) << "_fsync(" << f << ", " << (syncdataonly ? "dataonly)":"data+metadata)") << dendl; |
7c673cae FG |
10684 | return _fsync(f->inode.get(), syncdataonly); |
10685 | } | |
10686 | ||
10687 | int Client::fstat(int fd, struct stat *stbuf, const UserPerm& perms, int mask) | |
10688 | { | |
f67539c2 TL |
10689 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10690 | if (!mref_reader.is_state_satisfied()) | |
10691 | return -CEPHFS_ENOTCONN; | |
10692 | ||
7c673cae FG |
10693 | tout(cct) << "fstat mask " << hex << mask << dec << std::endl; |
10694 | tout(cct) << fd << std::endl; | |
10695 | ||
f67539c2 | 10696 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10697 | Fh *f = get_filehandle(fd); |
10698 | if (!f) | |
f67539c2 | 10699 | return -CEPHFS_EBADF; |
7c673cae FG |
10700 | int r = _getattr(f->inode, mask, perms); |
10701 | if (r < 0) | |
10702 | return r; | |
10703 | fill_stat(f->inode, stbuf, NULL); | |
1adf2230 | 10704 | ldout(cct, 5) << "fstat(" << fd << ", " << stbuf << ") = " << r << dendl; |
7c673cae FG |
10705 | return r; |
10706 | } | |
10707 | ||
10708 | int Client::fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms, | |
10709 | unsigned int want, unsigned int flags) | |
10710 | { | |
f67539c2 TL |
10711 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10712 | if (!mref_reader.is_state_satisfied()) | |
10713 | return -CEPHFS_ENOTCONN; | |
10714 | ||
7c673cae FG |
10715 | tout(cct) << "fstatx flags " << hex << flags << " want " << want << dec << std::endl; |
10716 | tout(cct) << fd << std::endl; | |
10717 | ||
f67539c2 | 10718 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10719 | Fh *f = get_filehandle(fd); |
10720 | if (!f) | |
f67539c2 | 10721 | return -CEPHFS_EBADF; |
7c673cae FG |
10722 | |
10723 | unsigned mask = statx_to_mask(flags, want); | |
10724 | ||
10725 | int r = 0; | |
b3b6e05e | 10726 | if (mask) { |
7c673cae FG |
10727 | r = _getattr(f->inode, mask, perms); |
10728 | if (r < 0) { | |
10729 | ldout(cct, 3) << "fstatx exit on error!" << dendl; | |
10730 | return r; | |
10731 | } | |
10732 | } | |
10733 | ||
10734 | fill_statx(f->inode, mask, stx); | |
10735 | ldout(cct, 3) << "fstatx(" << fd << ", " << stx << ") = " << r << dendl; | |
10736 | return r; | |
10737 | } | |
10738 | ||
b3b6e05e TL |
10739 | int Client::statxat(int dirfd, const char *relpath, |
10740 | struct ceph_statx *stx, const UserPerm& perms, | |
10741 | unsigned int want, unsigned int flags) { | |
10742 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
10743 | if (!mref_reader.is_state_satisfied()) { | |
10744 | return -CEPHFS_ENOTCONN; | |
10745 | } | |
10746 | ||
10747 | tout(cct) << __func__ << " flags " << hex << flags << " want " << want << dec << std::endl; | |
10748 | tout(cct) << dirfd << std::endl; | |
10749 | tout(cct) << relpath << std::endl; | |
10750 | ||
10751 | unsigned mask = statx_to_mask(flags, want); | |
10752 | ||
10753 | InodeRef dirinode; | |
10754 | std::scoped_lock lock(client_lock); | |
10755 | int r = get_fd_inode(dirfd, &dirinode); | |
10756 | if (r < 0) { | |
10757 | return r; | |
10758 | } | |
10759 | ||
10760 | InodeRef in; | |
10761 | filepath path(relpath); | |
10762 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask, dirinode); | |
10763 | if (r < 0) { | |
10764 | return r; | |
10765 | } | |
10766 | r = _getattr(in, mask, perms); | |
10767 | if (r < 0) { | |
10768 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; | |
10769 | return r; | |
10770 | } | |
10771 | ||
10772 | fill_statx(in, mask, stx); | |
10773 | ldout(cct, 3) << __func__ << " dirfd" << dirfd << ", r= " << r << dendl; | |
10774 | return r; | |
10775 | } | |
10776 | ||
7c673cae FG |
10777 | // not written yet, but i want to link! |
10778 | ||
10779 | int Client::chdir(const char *relpath, std::string &new_cwd, | |
10780 | const UserPerm& perms) | |
10781 | { | |
f67539c2 TL |
10782 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10783 | if (!mref_reader.is_state_satisfied()) | |
10784 | return -CEPHFS_ENOTCONN; | |
10785 | ||
7c673cae FG |
10786 | tout(cct) << "chdir" << std::endl; |
10787 | tout(cct) << relpath << std::endl; | |
181888fb | 10788 | |
7c673cae FG |
10789 | filepath path(relpath); |
10790 | InodeRef in; | |
f67539c2 TL |
10791 | |
10792 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
10793 | int r = path_walk(path, &in, perms); |
10794 | if (r < 0) | |
10795 | return r; | |
92f5a8d4 TL |
10796 | |
10797 | if (!(in.get()->is_dir())) | |
f67539c2 | 10798 | return -CEPHFS_ENOTDIR; |
92f5a8d4 | 10799 | |
7c673cae FG |
10800 | if (cwd != in) |
10801 | cwd.swap(in); | |
10802 | ldout(cct, 3) << "chdir(" << relpath << ") cwd now " << cwd->ino << dendl; | |
10803 | ||
b5b8bbf5 | 10804 | _getcwd(new_cwd, perms); |
7c673cae FG |
10805 | return 0; |
10806 | } | |
10807 | ||
b5b8bbf5 | 10808 | void Client::_getcwd(string& dir, const UserPerm& perms) |
7c673cae FG |
10809 | { |
10810 | filepath path; | |
11fdf7f2 | 10811 | ldout(cct, 10) << __func__ << " " << *cwd << dendl; |
7c673cae FG |
10812 | |
10813 | Inode *in = cwd.get(); | |
b3b6e05e | 10814 | while (in != root.get()) { |
11fdf7f2 | 10815 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked |
7c673cae FG |
10816 | |
10817 | // A cwd or ancester is unlinked | |
11fdf7f2 | 10818 | if (in->dentries.empty()) { |
7c673cae FG |
10819 | return; |
10820 | } | |
10821 | ||
10822 | Dentry *dn = in->get_first_parent(); | |
10823 | ||
10824 | ||
10825 | if (!dn) { | |
10826 | // look it up | |
11fdf7f2 | 10827 | ldout(cct, 10) << __func__ << " looking up parent for " << *in << dendl; |
7c673cae FG |
10828 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
10829 | filepath path(in->ino); | |
10830 | req->set_filepath(path); | |
10831 | req->set_inode(in); | |
10832 | int res = make_request(req, perms); | |
10833 | if (res < 0) | |
10834 | break; | |
10835 | ||
10836 | // start over | |
10837 | path = filepath(); | |
10838 | in = cwd.get(); | |
10839 | continue; | |
10840 | } | |
10841 | path.push_front_dentry(dn->name); | |
10842 | in = dn->dir->parent_inode; | |
10843 | } | |
10844 | dir = "/"; | |
10845 | dir += path.get_path(); | |
10846 | } | |
10847 | ||
b5b8bbf5 FG |
10848 | void Client::getcwd(string& dir, const UserPerm& perms) |
10849 | { | |
f67539c2 TL |
10850 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10851 | if (!mref_reader.is_state_satisfied()) | |
10852 | return; | |
10853 | ||
10854 | std::scoped_lock l(client_lock); | |
10855 | ||
10856 | _getcwd(dir, perms); | |
b5b8bbf5 FG |
10857 | } |
10858 | ||
7c673cae FG |
10859 | int Client::statfs(const char *path, struct statvfs *stbuf, |
10860 | const UserPerm& perms) | |
10861 | { | |
f67539c2 TL |
10862 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10863 | if (!mref_reader.is_state_satisfied()) | |
10864 | return -CEPHFS_ENOTCONN; | |
10865 | ||
11fdf7f2 | 10866 | tout(cct) << __func__ << std::endl; |
91327a77 | 10867 | unsigned long int total_files_on_fs; |
7c673cae FG |
10868 | |
10869 | ceph_statfs stats; | |
10870 | C_SaferCond cond; | |
d2e6a577 | 10871 | |
f67539c2 | 10872 | std::unique_lock lock(client_lock); |
d2e6a577 FG |
10873 | const vector<int64_t> &data_pools = mdsmap->get_data_pools(); |
10874 | if (data_pools.size() == 1) { | |
10875 | objecter->get_fs_stats(stats, data_pools[0], &cond); | |
10876 | } else { | |
20effc67 | 10877 | objecter->get_fs_stats(stats, std::optional<int64_t>(), &cond); |
d2e6a577 | 10878 | } |
7c673cae | 10879 | |
f67539c2 | 10880 | lock.unlock(); |
7c673cae | 10881 | int rval = cond.wait(); |
f67539c2 TL |
10882 | lock.lock(); |
10883 | ||
20effc67 | 10884 | ceph_assert(root); |
91327a77 | 10885 | total_files_on_fs = root->rstat.rfiles + root->rstat.rsubdirs; |
7c673cae FG |
10886 | |
10887 | if (rval < 0) { | |
10888 | ldout(cct, 1) << "underlying call to statfs returned error: " | |
10889 | << cpp_strerror(rval) | |
10890 | << dendl; | |
10891 | return rval; | |
10892 | } | |
10893 | ||
10894 | memset(stbuf, 0, sizeof(*stbuf)); | |
10895 | ||
10896 | /* | |
10897 | * we're going to set a block size of 4MB so we can represent larger | |
10898 | * FSes without overflowing. Additionally convert the space | |
10899 | * measurements from KB to bytes while making them in terms of | |
10900 | * blocks. We use 4MB only because it is big enough, and because it | |
10901 | * actually *is* the (ceph) default block size. | |
10902 | */ | |
10903 | const int CEPH_BLOCK_SHIFT = 22; | |
10904 | stbuf->f_frsize = 1 << CEPH_BLOCK_SHIFT; | |
10905 | stbuf->f_bsize = 1 << CEPH_BLOCK_SHIFT; | |
91327a77 | 10906 | stbuf->f_files = total_files_on_fs; |
f67539c2 | 10907 | stbuf->f_ffree = -1; |
7c673cae FG |
10908 | stbuf->f_favail = -1; |
10909 | stbuf->f_fsid = -1; // ?? | |
10910 | stbuf->f_flag = 0; // ?? | |
10911 | stbuf->f_namemax = NAME_MAX; | |
10912 | ||
10913 | // Usually quota_root will == root_ancestor, but if the mount root has no | |
10914 | // quota but we can see a parent of it that does have a quota, we'll | |
10915 | // respect that one instead. | |
11fdf7f2 | 10916 | ceph_assert(root != nullptr); |
b3b6e05e | 10917 | InodeRef quota_root = root->quota.is_enable() ? root : get_quota_root(root.get(), perms); |
7c673cae FG |
10918 | |
10919 | // get_quota_root should always give us something | |
10920 | // because client quotas are always enabled | |
11fdf7f2 | 10921 | ceph_assert(quota_root != nullptr); |
7c673cae FG |
10922 | |
10923 | if (quota_root && cct->_conf->client_quota_df && quota_root->quota.max_bytes) { | |
10924 | ||
10925 | // Skip the getattr if any sessions are stale, as we don't want to | |
10926 | // block `df` if this client has e.g. been evicted, or if the MDS cluster | |
10927 | // is unhealthy. | |
10928 | if (!_any_stale_sessions()) { | |
10929 | int r = _getattr(quota_root, 0, perms, true); | |
10930 | if (r != 0) { | |
10931 | // Ignore return value: error getting latest inode metadata is not a good | |
10932 | // reason to break "df". | |
10933 | lderr(cct) << "Error in getattr on quota root 0x" | |
10934 | << std::hex << quota_root->ino << std::dec | |
10935 | << " statfs result may be outdated" << dendl; | |
10936 | } | |
10937 | } | |
10938 | ||
10939 | // Special case: if there is a size quota set on the Inode acting | |
10940 | // as the root for this client mount, then report the quota status | |
10941 | // as the filesystem statistics. | |
10942 | const fsblkcnt_t total = quota_root->quota.max_bytes >> CEPH_BLOCK_SHIFT; | |
10943 | const fsblkcnt_t used = quota_root->rstat.rbytes >> CEPH_BLOCK_SHIFT; | |
31f18b77 FG |
10944 | // It is possible for a quota to be exceeded: arithmetic here must |
10945 | // handle case where used > total. | |
10946 | const fsblkcnt_t free = total > used ? total - used : 0; | |
7c673cae FG |
10947 | |
10948 | stbuf->f_blocks = total; | |
10949 | stbuf->f_bfree = free; | |
10950 | stbuf->f_bavail = free; | |
10951 | } else { | |
d2e6a577 | 10952 | // General case: report the cluster statistics returned from RADOS. Because |
7c673cae FG |
10953 | // multiple pools may be used without one filesystem namespace via |
10954 | // layouts, this is the most correct thing we can do. | |
10955 | stbuf->f_blocks = stats.kb >> (CEPH_BLOCK_SHIFT - 10); | |
10956 | stbuf->f_bfree = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
10957 | stbuf->f_bavail = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
10958 | } | |
10959 | ||
10960 | return rval; | |
10961 | } | |
10962 | ||
10963 | int Client::_do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep, | |
10964 | struct flock *fl, uint64_t owner, bool removing) | |
10965 | { | |
11fdf7f2 | 10966 | ldout(cct, 10) << __func__ << " ino " << in->ino |
7c673cae FG |
10967 | << (lock_type == CEPH_LOCK_FCNTL ? " fcntl" : " flock") |
10968 | << " type " << fl->l_type << " owner " << owner | |
10969 | << " " << fl->l_start << "~" << fl->l_len << dendl; | |
10970 | ||
f6b5b4d7 | 10971 | if (in->flags & I_ERROR_FILELOCK) |
f67539c2 | 10972 | return -CEPHFS_EIO; |
f6b5b4d7 | 10973 | |
7c673cae FG |
10974 | int lock_cmd; |
10975 | if (F_RDLCK == fl->l_type) | |
10976 | lock_cmd = CEPH_LOCK_SHARED; | |
10977 | else if (F_WRLCK == fl->l_type) | |
10978 | lock_cmd = CEPH_LOCK_EXCL; | |
10979 | else if (F_UNLCK == fl->l_type) | |
10980 | lock_cmd = CEPH_LOCK_UNLOCK; | |
10981 | else | |
f67539c2 | 10982 | return -CEPHFS_EIO; |
7c673cae FG |
10983 | |
10984 | if (op != CEPH_MDS_OP_SETFILELOCK || lock_cmd == CEPH_LOCK_UNLOCK) | |
10985 | sleep = 0; | |
10986 | ||
10987 | /* | |
10988 | * Set the most significant bit, so that MDS knows the 'owner' | |
10989 | * is sufficient to identify the owner of lock. (old code uses | |
10990 | * both 'owner' and 'pid') | |
10991 | */ | |
10992 | owner |= (1ULL << 63); | |
10993 | ||
10994 | MetaRequest *req = new MetaRequest(op); | |
10995 | filepath path; | |
10996 | in->make_nosnap_relative_path(path); | |
10997 | req->set_filepath(path); | |
10998 | req->set_inode(in); | |
10999 | ||
11000 | req->head.args.filelock_change.rule = lock_type; | |
11001 | req->head.args.filelock_change.type = lock_cmd; | |
11002 | req->head.args.filelock_change.owner = owner; | |
11003 | req->head.args.filelock_change.pid = fl->l_pid; | |
11004 | req->head.args.filelock_change.start = fl->l_start; | |
11005 | req->head.args.filelock_change.length = fl->l_len; | |
11006 | req->head.args.filelock_change.wait = sleep; | |
11007 | ||
11008 | int ret; | |
11009 | bufferlist bl; | |
11010 | ||
11011 | if (sleep && switch_interrupt_cb) { | |
11012 | // enable interrupt | |
11013 | switch_interrupt_cb(callback_handle, req->get()); | |
11014 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
7c673cae FG |
11015 | // disable interrupt |
11016 | switch_interrupt_cb(callback_handle, NULL); | |
31f18b77 FG |
11017 | if (ret == 0 && req->aborted()) { |
11018 | // effect of this lock request has been revoked by the 'lock intr' request | |
11019 | ret = req->get_abort_code(); | |
11020 | } | |
7c673cae FG |
11021 | put_request(req); |
11022 | } else { | |
11023 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
11024 | } | |
11025 | ||
11026 | if (ret == 0) { | |
11027 | if (op == CEPH_MDS_OP_GETFILELOCK) { | |
11028 | ceph_filelock filelock; | |
11fdf7f2 TL |
11029 | auto p = bl.cbegin(); |
11030 | decode(filelock, p); | |
7c673cae FG |
11031 | |
11032 | if (CEPH_LOCK_SHARED == filelock.type) | |
11033 | fl->l_type = F_RDLCK; | |
11034 | else if (CEPH_LOCK_EXCL == filelock.type) | |
11035 | fl->l_type = F_WRLCK; | |
11036 | else | |
11037 | fl->l_type = F_UNLCK; | |
11038 | ||
11039 | fl->l_whence = SEEK_SET; | |
11040 | fl->l_start = filelock.start; | |
11041 | fl->l_len = filelock.length; | |
11042 | fl->l_pid = filelock.pid; | |
11043 | } else if (op == CEPH_MDS_OP_SETFILELOCK) { | |
11044 | ceph_lock_state_t *lock_state; | |
11045 | if (lock_type == CEPH_LOCK_FCNTL) { | |
11046 | if (!in->fcntl_locks) | |
11fdf7f2 TL |
11047 | in->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
11048 | lock_state = in->fcntl_locks.get(); | |
7c673cae FG |
11049 | } else if (lock_type == CEPH_LOCK_FLOCK) { |
11050 | if (!in->flock_locks) | |
11fdf7f2 TL |
11051 | in->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
11052 | lock_state = in->flock_locks.get(); | |
7c673cae FG |
11053 | } else { |
11054 | ceph_abort(); | |
f67539c2 | 11055 | return -CEPHFS_EINVAL; |
7c673cae FG |
11056 | } |
11057 | _update_lock_state(fl, owner, lock_state); | |
11058 | ||
11059 | if (!removing) { | |
11060 | if (lock_type == CEPH_LOCK_FCNTL) { | |
11061 | if (!fh->fcntl_locks) | |
11fdf7f2 TL |
11062 | fh->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
11063 | lock_state = fh->fcntl_locks.get(); | |
7c673cae FG |
11064 | } else { |
11065 | if (!fh->flock_locks) | |
11fdf7f2 TL |
11066 | fh->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
11067 | lock_state = fh->flock_locks.get(); | |
7c673cae FG |
11068 | } |
11069 | _update_lock_state(fl, owner, lock_state); | |
11070 | } | |
11071 | } else | |
11072 | ceph_abort(); | |
11073 | } | |
11074 | return ret; | |
11075 | } | |
11076 | ||
11077 | int Client::_interrupt_filelock(MetaRequest *req) | |
11078 | { | |
31f18b77 FG |
11079 | // Set abort code, but do not kick. The abort code prevents the request |
11080 | // from being re-sent. | |
f67539c2 | 11081 | req->abort(-CEPHFS_EINTR); |
31f18b77 FG |
11082 | if (req->mds < 0) |
11083 | return 0; // haven't sent the request | |
11084 | ||
7c673cae FG |
11085 | Inode *in = req->inode(); |
11086 | ||
11087 | int lock_type; | |
11088 | if (req->head.args.filelock_change.rule == CEPH_LOCK_FLOCK) | |
11089 | lock_type = CEPH_LOCK_FLOCK_INTR; | |
11090 | else if (req->head.args.filelock_change.rule == CEPH_LOCK_FCNTL) | |
11091 | lock_type = CEPH_LOCK_FCNTL_INTR; | |
11092 | else { | |
11093 | ceph_abort(); | |
f67539c2 | 11094 | return -CEPHFS_EINVAL; |
7c673cae FG |
11095 | } |
11096 | ||
11097 | MetaRequest *intr_req = new MetaRequest(CEPH_MDS_OP_SETFILELOCK); | |
11098 | filepath path; | |
11099 | in->make_nosnap_relative_path(path); | |
11100 | intr_req->set_filepath(path); | |
11101 | intr_req->set_inode(in); | |
11102 | intr_req->head.args.filelock_change = req->head.args.filelock_change; | |
11103 | intr_req->head.args.filelock_change.rule = lock_type; | |
11104 | intr_req->head.args.filelock_change.type = CEPH_LOCK_UNLOCK; | |
11105 | ||
11106 | UserPerm perms(req->get_uid(), req->get_gid()); | |
11107 | return make_request(intr_req, perms, NULL, NULL, -1); | |
11108 | } | |
11109 | ||
11110 | void Client::_encode_filelocks(Inode *in, bufferlist& bl) | |
11111 | { | |
11112 | if (!in->fcntl_locks && !in->flock_locks) | |
11113 | return; | |
11114 | ||
11115 | unsigned nr_fcntl_locks = in->fcntl_locks ? in->fcntl_locks->held_locks.size() : 0; | |
11fdf7f2 | 11116 | encode(nr_fcntl_locks, bl); |
7c673cae | 11117 | if (nr_fcntl_locks) { |
11fdf7f2 | 11118 | auto &lock_state = in->fcntl_locks; |
20effc67 | 11119 | for(auto p = lock_state->held_locks.begin(); |
7c673cae FG |
11120 | p != lock_state->held_locks.end(); |
11121 | ++p) | |
11fdf7f2 | 11122 | encode(p->second, bl); |
7c673cae FG |
11123 | } |
11124 | ||
11125 | unsigned nr_flock_locks = in->flock_locks ? in->flock_locks->held_locks.size() : 0; | |
11fdf7f2 | 11126 | encode(nr_flock_locks, bl); |
7c673cae | 11127 | if (nr_flock_locks) { |
11fdf7f2 | 11128 | auto &lock_state = in->flock_locks; |
20effc67 | 11129 | for(auto p = lock_state->held_locks.begin(); |
7c673cae FG |
11130 | p != lock_state->held_locks.end(); |
11131 | ++p) | |
11fdf7f2 | 11132 | encode(p->second, bl); |
7c673cae FG |
11133 | } |
11134 | ||
11fdf7f2 | 11135 | ldout(cct, 10) << __func__ << " ino " << in->ino << ", " << nr_fcntl_locks |
7c673cae FG |
11136 | << " fcntl locks, " << nr_flock_locks << " flock locks" << dendl; |
11137 | } | |
11138 | ||
11139 | void Client::_release_filelocks(Fh *fh) | |
11140 | { | |
11141 | if (!fh->fcntl_locks && !fh->flock_locks) | |
11142 | return; | |
11143 | ||
11144 | Inode *in = fh->inode.get(); | |
11fdf7f2 | 11145 | ldout(cct, 10) << __func__ << " " << fh << " ino " << in->ino << dendl; |
7c673cae | 11146 | |
f6b5b4d7 TL |
11147 | list<ceph_filelock> activated_locks; |
11148 | ||
7c673cae FG |
11149 | list<pair<int, ceph_filelock> > to_release; |
11150 | ||
11151 | if (fh->fcntl_locks) { | |
11fdf7f2 | 11152 | auto &lock_state = fh->fcntl_locks; |
f6b5b4d7 TL |
11153 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11154 | auto q = p++; | |
11155 | if (in->flags & I_ERROR_FILELOCK) { | |
11156 | lock_state->remove_lock(q->second, activated_locks); | |
11157 | } else { | |
11158 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FCNTL, q->second)); | |
11159 | } | |
11160 | } | |
11fdf7f2 | 11161 | lock_state.reset(); |
7c673cae FG |
11162 | } |
11163 | if (fh->flock_locks) { | |
11fdf7f2 | 11164 | auto &lock_state = fh->flock_locks; |
f6b5b4d7 TL |
11165 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11166 | auto q = p++; | |
11167 | if (in->flags & I_ERROR_FILELOCK) { | |
11168 | lock_state->remove_lock(q->second, activated_locks); | |
11169 | } else { | |
11170 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FLOCK, q->second)); | |
11171 | } | |
11172 | } | |
11fdf7f2 | 11173 | lock_state.reset(); |
7c673cae FG |
11174 | } |
11175 | ||
f6b5b4d7 TL |
11176 | if ((in->flags & I_ERROR_FILELOCK) && !in->has_any_filelocks()) |
11177 | in->flags &= ~I_ERROR_FILELOCK; | |
7c673cae | 11178 | |
f6b5b4d7 | 11179 | if (to_release.empty()) |
11fdf7f2 TL |
11180 | return; |
11181 | ||
7c673cae FG |
11182 | struct flock fl; |
11183 | memset(&fl, 0, sizeof(fl)); | |
11184 | fl.l_whence = SEEK_SET; | |
11185 | fl.l_type = F_UNLCK; | |
11186 | ||
11187 | for (list<pair<int, ceph_filelock> >::iterator p = to_release.begin(); | |
11188 | p != to_release.end(); | |
11189 | ++p) { | |
11190 | fl.l_start = p->second.start; | |
11191 | fl.l_len = p->second.length; | |
11192 | fl.l_pid = p->second.pid; | |
11193 | _do_filelock(in, fh, p->first, CEPH_MDS_OP_SETFILELOCK, 0, &fl, | |
11194 | p->second.owner, true); | |
11195 | } | |
11196 | } | |
11197 | ||
11198 | void Client::_update_lock_state(struct flock *fl, uint64_t owner, | |
11199 | ceph_lock_state_t *lock_state) | |
11200 | { | |
11201 | int lock_cmd; | |
11202 | if (F_RDLCK == fl->l_type) | |
11203 | lock_cmd = CEPH_LOCK_SHARED; | |
11204 | else if (F_WRLCK == fl->l_type) | |
11205 | lock_cmd = CEPH_LOCK_EXCL; | |
11206 | else | |
11207 | lock_cmd = CEPH_LOCK_UNLOCK;; | |
11208 | ||
11209 | ceph_filelock filelock; | |
11210 | filelock.start = fl->l_start; | |
11211 | filelock.length = fl->l_len; | |
11212 | filelock.client = 0; | |
11213 | // see comment in _do_filelock() | |
11214 | filelock.owner = owner | (1ULL << 63); | |
11215 | filelock.pid = fl->l_pid; | |
11216 | filelock.type = lock_cmd; | |
11217 | ||
11218 | if (filelock.type == CEPH_LOCK_UNLOCK) { | |
11219 | list<ceph_filelock> activated_locks; | |
11220 | lock_state->remove_lock(filelock, activated_locks); | |
11221 | } else { | |
11222 | bool r = lock_state->add_lock(filelock, false, false, NULL); | |
11fdf7f2 | 11223 | ceph_assert(r); |
7c673cae FG |
11224 | } |
11225 | } | |
11226 | ||
11227 | int Client::_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
11228 | { | |
11229 | Inode *in = fh->inode.get(); | |
11230 | ldout(cct, 10) << "_getlk " << fh << " ino " << in->ino << dendl; | |
11231 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_GETFILELOCK, 0, fl, owner); | |
11232 | return ret; | |
11233 | } | |
11234 | ||
11235 | int Client::_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
11236 | { | |
11237 | Inode *in = fh->inode.get(); | |
11238 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << dendl; | |
11239 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_SETFILELOCK, sleep, fl, owner); | |
11240 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11241 | return ret; | |
11242 | } | |
11243 | ||
11244 | int Client::_flock(Fh *fh, int cmd, uint64_t owner) | |
11245 | { | |
11246 | Inode *in = fh->inode.get(); | |
11247 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << dendl; | |
11248 | ||
11249 | int sleep = !(cmd & LOCK_NB); | |
11250 | cmd &= ~LOCK_NB; | |
11251 | ||
11252 | int type; | |
11253 | switch (cmd) { | |
11254 | case LOCK_SH: | |
11255 | type = F_RDLCK; | |
11256 | break; | |
11257 | case LOCK_EX: | |
11258 | type = F_WRLCK; | |
11259 | break; | |
11260 | case LOCK_UN: | |
11261 | type = F_UNLCK; | |
11262 | break; | |
11263 | default: | |
f67539c2 | 11264 | return -CEPHFS_EINVAL; |
7c673cae FG |
11265 | } |
11266 | ||
11267 | struct flock fl; | |
11268 | memset(&fl, 0, sizeof(fl)); | |
11269 | fl.l_type = type; | |
11270 | fl.l_whence = SEEK_SET; | |
11271 | ||
11272 | int ret = _do_filelock(in, fh, CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK, sleep, &fl, owner); | |
11273 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11274 | return ret; | |
11275 | } | |
11276 | ||
f67539c2 TL |
11277 | int Client::get_snap_info(const char *path, const UserPerm &perms, SnapInfo *snap_info) { |
11278 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11279 | if (!mref_reader.is_state_satisfied()) { | |
11280 | return -CEPHFS_ENOTCONN; | |
11281 | } | |
11282 | ||
20effc67 | 11283 | std::scoped_lock lock(client_lock); |
f67539c2 TL |
11284 | InodeRef in; |
11285 | int r = Client::path_walk(path, &in, perms, true); | |
11286 | if (r < 0) { | |
11287 | return r; | |
11288 | } | |
11289 | ||
11290 | if (in->snapid == CEPH_NOSNAP) { | |
11291 | return -CEPHFS_EINVAL; | |
11292 | } | |
11293 | ||
11294 | snap_info->id = in->snapid; | |
11295 | snap_info->metadata = in->snap_metadata; | |
11296 | return 0; | |
11297 | } | |
11298 | ||
11299 | int Client::ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms) | |
11300 | { | |
11301 | /* Since the only thing this does is wrap a call to statfs, and | |
11302 | statfs takes a lock, it doesn't seem we have a need to split it | |
11303 | out. */ | |
7c673cae FG |
11304 | return statfs(0, stbuf, perms); |
11305 | } | |
11306 | ||
20effc67 | 11307 | void Client::_ll_register_callbacks(struct ceph_client_callback_args *args) |
7c673cae FG |
11308 | { |
11309 | if (!args) | |
11310 | return; | |
20effc67 | 11311 | |
11fdf7f2 | 11312 | ldout(cct, 10) << __func__ << " cb " << args->handle |
7c673cae FG |
11313 | << " invalidate_ino_cb " << args->ino_cb |
11314 | << " invalidate_dentry_cb " << args->dentry_cb | |
7c673cae FG |
11315 | << " switch_interrupt_cb " << args->switch_intr_cb |
11316 | << " remount_cb " << args->remount_cb | |
11317 | << dendl; | |
11318 | callback_handle = args->handle; | |
11319 | if (args->ino_cb) { | |
11320 | ino_invalidate_cb = args->ino_cb; | |
11321 | async_ino_invalidator.start(); | |
11322 | } | |
11323 | if (args->dentry_cb) { | |
11324 | dentry_invalidate_cb = args->dentry_cb; | |
11325 | async_dentry_invalidator.start(); | |
11326 | } | |
11327 | if (args->switch_intr_cb) { | |
11328 | switch_interrupt_cb = args->switch_intr_cb; | |
11329 | interrupt_finisher.start(); | |
11330 | } | |
11331 | if (args->remount_cb) { | |
11332 | remount_cb = args->remount_cb; | |
11333 | remount_finisher.start(); | |
11334 | } | |
e306af50 TL |
11335 | if (args->ino_release_cb) { |
11336 | ino_release_cb = args->ino_release_cb; | |
11337 | async_ino_releasor.start(); | |
11338 | } | |
11339 | if (args->umask_cb) | |
11340 | umask_cb = args->umask_cb; | |
7c673cae FG |
11341 | } |
11342 | ||
20effc67 TL |
11343 | // This is deprecated, use ll_register_callbacks2() instead. |
11344 | void Client::ll_register_callbacks(struct ceph_client_callback_args *args) | |
11345 | { | |
11346 | ceph_assert(!is_mounting() && !is_mounted() && !is_unmounting()); | |
11347 | ||
11348 | _ll_register_callbacks(args); | |
11349 | } | |
11350 | ||
11351 | int Client::ll_register_callbacks2(struct ceph_client_callback_args *args) | |
11352 | { | |
11353 | if (is_mounting() || is_mounted() || is_unmounting()) | |
11354 | return -CEPHFS_EBUSY; | |
11355 | ||
11356 | _ll_register_callbacks(args); | |
11357 | return 0; | |
11358 | } | |
11359 | ||
7c673cae FG |
11360 | int Client::test_dentry_handling(bool can_invalidate) |
11361 | { | |
11362 | int r = 0; | |
11363 | ||
f67539c2 TL |
11364 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
11365 | if (!iref_reader.is_state_satisfied()) | |
11366 | return -CEPHFS_ENOTCONN; | |
11367 | ||
7c673cae FG |
11368 | can_invalidate_dentries = can_invalidate; |
11369 | ||
11370 | if (can_invalidate_dentries) { | |
11fdf7f2 | 11371 | ceph_assert(dentry_invalidate_cb); |
7c673cae | 11372 | ldout(cct, 1) << "using dentry_invalidate_cb" << dendl; |
b32b8144 | 11373 | r = 0; |
11fdf7f2 TL |
11374 | } else { |
11375 | ceph_assert(remount_cb); | |
7c673cae | 11376 | ldout(cct, 1) << "using remount_cb" << dendl; |
91327a77 | 11377 | r = _do_remount(false); |
b32b8144 | 11378 | } |
11fdf7f2 | 11379 | |
7c673cae FG |
11380 | return r; |
11381 | } | |
11382 | ||
11383 | int Client::_sync_fs() | |
11384 | { | |
f67539c2 TL |
11385 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11386 | ||
11fdf7f2 | 11387 | ldout(cct, 10) << __func__ << dendl; |
7c673cae FG |
11388 | |
11389 | // flush file data | |
11fdf7f2 TL |
11390 | std::unique_ptr<C_SaferCond> cond = nullptr; |
11391 | if (cct->_conf->client_oc) { | |
11392 | cond.reset(new C_SaferCond("Client::_sync_fs:lock")); | |
11393 | objectcacher->flush_all(cond.get()); | |
11394 | } | |
7c673cae FG |
11395 | |
11396 | // flush caps | |
11397 | flush_caps_sync(); | |
11398 | ceph_tid_t flush_tid = last_flush_tid; | |
11399 | ||
11400 | // wait for unsafe mds requests | |
11401 | wait_unsafe_requests(); | |
11402 | ||
11403 | wait_sync_caps(flush_tid); | |
11404 | ||
11fdf7f2 | 11405 | if (nullptr != cond) { |
9f95a23c | 11406 | client_lock.unlock(); |
11fdf7f2 TL |
11407 | ldout(cct, 15) << __func__ << " waiting on data to flush" << dendl; |
11408 | cond->wait(); | |
11409 | ldout(cct, 15) << __func__ << " flush finished" << dendl; | |
9f95a23c | 11410 | client_lock.lock(); |
7c673cae FG |
11411 | } |
11412 | ||
11413 | return 0; | |
11414 | } | |
11415 | ||
11416 | int Client::sync_fs() | |
11417 | { | |
f67539c2 TL |
11418 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11419 | if (!mref_reader.is_state_satisfied()) | |
11420 | return -CEPHFS_ENOTCONN; | |
181888fb | 11421 | |
f67539c2 | 11422 | std::scoped_lock l(client_lock); |
181888fb | 11423 | |
7c673cae FG |
11424 | return _sync_fs(); |
11425 | } | |
11426 | ||
11427 | int64_t Client::drop_caches() | |
11428 | { | |
f67539c2 | 11429 | std::scoped_lock l(client_lock); |
7c673cae FG |
11430 | return objectcacher->release_all(); |
11431 | } | |
11432 | ||
11fdf7f2 TL |
11433 | int Client::_lazyio(Fh *fh, int enable) |
11434 | { | |
11435 | Inode *in = fh->inode.get(); | |
11436 | ldout(cct, 20) << __func__ << " " << *in << " " << !!enable << dendl; | |
11437 | ||
11438 | if (!!(fh->mode & CEPH_FILE_MODE_LAZY) == !!enable) | |
11439 | return 0; | |
11440 | ||
11441 | int orig_mode = fh->mode; | |
11442 | if (enable) { | |
11443 | fh->mode |= CEPH_FILE_MODE_LAZY; | |
11444 | in->get_open_ref(fh->mode); | |
11445 | in->put_open_ref(orig_mode); | |
11446 | check_caps(in, CHECK_CAPS_NODELAY); | |
11447 | } else { | |
11448 | fh->mode &= ~CEPH_FILE_MODE_LAZY; | |
11449 | in->get_open_ref(fh->mode); | |
11450 | in->put_open_ref(orig_mode); | |
11451 | check_caps(in, 0); | |
11452 | } | |
11453 | ||
11454 | return 0; | |
11455 | } | |
11456 | ||
11457 | int Client::lazyio(int fd, int enable) | |
11458 | { | |
f67539c2 | 11459 | std::scoped_lock l(client_lock); |
11fdf7f2 TL |
11460 | Fh *f = get_filehandle(fd); |
11461 | if (!f) | |
f67539c2 | 11462 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
11463 | |
11464 | return _lazyio(f, enable); | |
11465 | } | |
11466 | ||
11467 | int Client::ll_lazyio(Fh *fh, int enable) | |
11468 | { | |
11fdf7f2 TL |
11469 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << !!enable << dendl; |
11470 | tout(cct) << __func__ << std::endl; | |
11471 | ||
f67539c2 | 11472 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
11473 | return _lazyio(fh, enable); |
11474 | } | |
7c673cae | 11475 | |
92f5a8d4 | 11476 | int Client::lazyio_propagate(int fd, loff_t offset, size_t count) |
7c673cae | 11477 | { |
f67539c2 | 11478 | std::scoped_lock l(client_lock); |
92f5a8d4 | 11479 | ldout(cct, 3) << "op: client->lazyio_propagate(" << fd |
7c673cae FG |
11480 | << ", " << offset << ", " << count << ")" << dendl; |
11481 | ||
11482 | Fh *f = get_filehandle(fd); | |
11483 | if (!f) | |
f67539c2 | 11484 | return -CEPHFS_EBADF; |
7c673cae FG |
11485 | |
11486 | // for now | |
11487 | _fsync(f, true); | |
11488 | ||
11489 | return 0; | |
11490 | } | |
11491 | ||
11492 | int Client::lazyio_synchronize(int fd, loff_t offset, size_t count) | |
11493 | { | |
f67539c2 | 11494 | std::scoped_lock l(client_lock); |
7c673cae FG |
11495 | ldout(cct, 3) << "op: client->lazyio_synchronize(" << fd |
11496 | << ", " << offset << ", " << count << ")" << dendl; | |
11497 | ||
11498 | Fh *f = get_filehandle(fd); | |
11499 | if (!f) | |
f67539c2 | 11500 | return -CEPHFS_EBADF; |
7c673cae FG |
11501 | Inode *in = f->inode.get(); |
11502 | ||
11503 | _fsync(f, true); | |
92f5a8d4 TL |
11504 | if (_release(in)) { |
11505 | int r =_getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
11506 | if (r < 0) | |
11507 | return r; | |
11508 | } | |
7c673cae FG |
11509 | return 0; |
11510 | } | |
11511 | ||
11512 | ||
11513 | // ============================= | |
11514 | // snaps | |
11515 | ||
f67539c2 TL |
11516 | int Client::mksnap(const char *relpath, const char *name, const UserPerm& perm, |
11517 | mode_t mode, const std::map<std::string, std::string> &metadata) | |
7c673cae | 11518 | { |
f67539c2 TL |
11519 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11520 | if (!mref_reader.is_state_satisfied()) | |
11521 | return -CEPHFS_ENOTCONN; | |
181888fb | 11522 | |
f67539c2 | 11523 | std::scoped_lock l(client_lock); |
181888fb | 11524 | |
7c673cae FG |
11525 | filepath path(relpath); |
11526 | InodeRef in; | |
11527 | int r = path_walk(path, &in, perm); | |
11528 | if (r < 0) | |
11529 | return r; | |
11530 | if (cct->_conf->client_permissions) { | |
11531 | r = may_create(in.get(), perm); | |
11532 | if (r < 0) | |
11533 | return r; | |
11534 | } | |
11535 | Inode *snapdir = open_snapdir(in.get()); | |
f67539c2 | 11536 | return _mkdir(snapdir, name, mode, perm, nullptr, metadata); |
7c673cae | 11537 | } |
181888fb | 11538 | |
f67539c2 | 11539 | int Client::rmsnap(const char *relpath, const char *name, const UserPerm& perms, bool check_perms) |
7c673cae | 11540 | { |
f67539c2 TL |
11541 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11542 | if (!mref_reader.is_state_satisfied()) | |
11543 | return -CEPHFS_ENOTCONN; | |
181888fb | 11544 | |
f67539c2 | 11545 | std::scoped_lock l(client_lock); |
181888fb | 11546 | |
7c673cae FG |
11547 | filepath path(relpath); |
11548 | InodeRef in; | |
11549 | int r = path_walk(path, &in, perms); | |
11550 | if (r < 0) | |
11551 | return r; | |
f67539c2 | 11552 | Inode *snapdir = open_snapdir(in.get()); |
7c673cae | 11553 | if (cct->_conf->client_permissions) { |
f67539c2 | 11554 | r = may_delete(snapdir, check_perms ? name : NULL, perms); |
7c673cae FG |
11555 | if (r < 0) |
11556 | return r; | |
11557 | } | |
7c673cae FG |
11558 | return _rmdir(snapdir, name, perms); |
11559 | } | |
11560 | ||
11561 | // ============================= | |
11562 | // expose caps | |
11563 | ||
f67539c2 TL |
11564 | int Client::get_caps_issued(int fd) |
11565 | { | |
11566 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11567 | if (!mref_reader.is_state_satisfied()) | |
11568 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11569 | |
f67539c2 | 11570 | std::scoped_lock lock(client_lock); |
181888fb | 11571 | |
7c673cae FG |
11572 | Fh *f = get_filehandle(fd); |
11573 | if (!f) | |
f67539c2 | 11574 | return -CEPHFS_EBADF; |
7c673cae FG |
11575 | |
11576 | return f->inode->caps_issued(); | |
11577 | } | |
11578 | ||
11579 | int Client::get_caps_issued(const char *path, const UserPerm& perms) | |
11580 | { | |
f67539c2 TL |
11581 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11582 | if (!mref_reader.is_state_satisfied()) | |
11583 | return -CEPHFS_ENOTCONN; | |
181888fb | 11584 | |
f67539c2 | 11585 | std::scoped_lock lock(client_lock); |
181888fb | 11586 | |
7c673cae FG |
11587 | filepath p(path); |
11588 | InodeRef in; | |
11589 | int r = path_walk(p, &in, perms, true); | |
11590 | if (r < 0) | |
11591 | return r; | |
11592 | return in->caps_issued(); | |
11593 | } | |
11594 | ||
11595 | // ========================================= | |
11596 | // low level | |
11597 | ||
11598 | Inode *Client::open_snapdir(Inode *diri) | |
11599 | { | |
11600 | Inode *in; | |
11601 | vinodeno_t vino(diri->ino, CEPH_SNAPDIR); | |
11602 | if (!inode_map.count(vino)) { | |
11603 | in = new Inode(this, vino, &diri->layout); | |
11604 | ||
11605 | in->ino = diri->ino; | |
11606 | in->snapid = CEPH_SNAPDIR; | |
11607 | in->mode = diri->mode; | |
11608 | in->uid = diri->uid; | |
11609 | in->gid = diri->gid; | |
494da23a | 11610 | in->nlink = 1; |
7c673cae FG |
11611 | in->mtime = diri->mtime; |
11612 | in->ctime = diri->ctime; | |
11613 | in->btime = diri->btime; | |
f6b5b4d7 | 11614 | in->atime = diri->atime; |
7c673cae FG |
11615 | in->size = diri->size; |
11616 | in->change_attr = diri->change_attr; | |
11617 | ||
11618 | in->dirfragtree.clear(); | |
11619 | in->snapdir_parent = diri; | |
11620 | diri->flags |= I_SNAPDIR_OPEN; | |
11621 | inode_map[vino] = in; | |
11622 | if (use_faked_inos()) | |
11623 | _assign_faked_ino(in); | |
11624 | ldout(cct, 10) << "open_snapdir created snapshot inode " << *in << dendl; | |
11625 | } else { | |
11626 | in = inode_map[vino]; | |
11627 | ldout(cct, 10) << "open_snapdir had snapshot inode " << *in << dendl; | |
11628 | } | |
11629 | return in; | |
11630 | } | |
11631 | ||
11632 | int Client::ll_lookup(Inode *parent, const char *name, struct stat *attr, | |
11633 | Inode **out, const UserPerm& perms) | |
11634 | { | |
f67539c2 TL |
11635 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11636 | if (!mref_reader.is_state_satisfied()) | |
11637 | return -CEPHFS_ENOTCONN; | |
11638 | ||
31f18b77 | 11639 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 TL |
11640 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
11641 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11642 | tout(cct) << name << std::endl; |
11643 | ||
f67539c2 | 11644 | std::scoped_lock lock(client_lock); |
181888fb | 11645 | |
7c673cae | 11646 | int r = 0; |
11fdf7f2 TL |
11647 | if (!fuse_default_permissions) { |
11648 | if (strcmp(name, ".") && strcmp(name, "..")) { | |
11649 | r = may_lookup(parent, perms); | |
11650 | if (r < 0) | |
11651 | return r; | |
11652 | } | |
7c673cae FG |
11653 | } |
11654 | ||
11655 | string dname(name); | |
11656 | InodeRef in; | |
11657 | ||
11658 | r = _lookup(parent, dname, CEPH_STAT_CAP_INODE_ALL, &in, perms); | |
11659 | if (r < 0) { | |
11660 | attr->st_ino = 0; | |
11661 | goto out; | |
11662 | } | |
11663 | ||
11fdf7f2 | 11664 | ceph_assert(in); |
7c673cae FG |
11665 | fill_stat(in, attr); |
11666 | _ll_get(in.get()); | |
11667 | ||
11668 | out: | |
11fdf7f2 | 11669 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11670 | << " -> " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; |
11671 | tout(cct) << attr->st_ino << std::endl; | |
11672 | *out = in.get(); | |
11673 | return r; | |
11674 | } | |
11675 | ||
f67539c2 TL |
11676 | int Client::ll_lookup_vino( |
11677 | vinodeno_t vino, | |
1adf2230 AA |
11678 | const UserPerm& perms, |
11679 | Inode **inode) | |
11680 | { | |
81eedcae | 11681 | ceph_assert(inode != NULL); |
f67539c2 TL |
11682 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11683 | if (!mref_reader.is_state_satisfied()) | |
11684 | return -CEPHFS_ENOTCONN; | |
81eedcae | 11685 | |
b3b6e05e TL |
11686 | if (is_reserved_vino(vino)) |
11687 | return -CEPHFS_ESTALE; | |
11688 | ||
f67539c2 TL |
11689 | std::scoped_lock lock(client_lock); |
11690 | ldout(cct, 3) << __func__ << " " << vino << dendl; | |
1adf2230 | 11691 | |
f67539c2 TL |
11692 | // Check the cache first |
11693 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
11694 | if (p != inode_map.end()) { | |
11695 | *inode = p->second; | |
11696 | _ll_get(*inode); | |
81eedcae TL |
11697 | return 0; |
11698 | } | |
11699 | ||
f67539c2 | 11700 | uint64_t snapid = vino.snapid; |
81eedcae | 11701 | |
f67539c2 TL |
11702 | // for snapdir, find the non-snapped dir inode |
11703 | if (snapid == CEPH_SNAPDIR) | |
11704 | vino.snapid = CEPH_NOSNAP; | |
11705 | ||
11706 | int r = _lookup_vino(vino, perms, inode); | |
11707 | if (r) | |
1adf2230 | 11708 | return r; |
f67539c2 | 11709 | ceph_assert(*inode != NULL); |
81eedcae | 11710 | |
f67539c2 TL |
11711 | if (snapid == CEPH_SNAPDIR) { |
11712 | Inode *tmp = *inode; | |
1adf2230 | 11713 | |
f67539c2 TL |
11714 | // open the snapdir and put the inode ref |
11715 | *inode = open_snapdir(tmp); | |
11716 | _ll_forget(tmp, 1); | |
11717 | _ll_get(*inode); | |
1adf2230 | 11718 | } |
1adf2230 AA |
11719 | return 0; |
11720 | } | |
11721 | ||
f67539c2 TL |
11722 | int Client::ll_lookup_inode( |
11723 | struct inodeno_t ino, | |
11724 | const UserPerm& perms, | |
11725 | Inode **inode) | |
11726 | { | |
11727 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
11728 | return ll_lookup_vino(vino, perms, inode); | |
11729 | } | |
11730 | ||
7c673cae FG |
11731 | int Client::ll_lookupx(Inode *parent, const char *name, Inode **out, |
11732 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
11733 | const UserPerm& perms) | |
11734 | { | |
f67539c2 TL |
11735 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11736 | if (!mref_reader.is_state_satisfied()) | |
11737 | return -CEPHFS_ENOTCONN; | |
11738 | ||
31f18b77 | 11739 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 | 11740 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
7c673cae FG |
11741 | tout(cct) << "ll_lookupx" << std::endl; |
11742 | tout(cct) << name << std::endl; | |
11743 | ||
f67539c2 | 11744 | std::scoped_lock lock(client_lock); |
181888fb | 11745 | |
7c673cae | 11746 | int r = 0; |
11fdf7f2 | 11747 | if (!fuse_default_permissions) { |
7c673cae FG |
11748 | r = may_lookup(parent, perms); |
11749 | if (r < 0) | |
11750 | return r; | |
11751 | } | |
11752 | ||
11753 | string dname(name); | |
11754 | InodeRef in; | |
11755 | ||
11756 | unsigned mask = statx_to_mask(flags, want); | |
11757 | r = _lookup(parent, dname, mask, &in, perms); | |
11758 | if (r < 0) { | |
11759 | stx->stx_ino = 0; | |
11760 | stx->stx_mask = 0; | |
11761 | } else { | |
11fdf7f2 | 11762 | ceph_assert(in); |
7c673cae FG |
11763 | fill_statx(in, mask, stx); |
11764 | _ll_get(in.get()); | |
11765 | } | |
11766 | ||
11fdf7f2 | 11767 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11768 | << " -> " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; |
11769 | tout(cct) << stx->stx_ino << std::endl; | |
11770 | *out = in.get(); | |
11771 | return r; | |
11772 | } | |
11773 | ||
11774 | int Client::ll_walk(const char* name, Inode **out, struct ceph_statx *stx, | |
11775 | unsigned int want, unsigned int flags, const UserPerm& perms) | |
11776 | { | |
f67539c2 TL |
11777 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11778 | if (!mref_reader.is_state_satisfied()) | |
11779 | return -CEPHFS_ENOTCONN; | |
181888fb | 11780 | |
7c673cae FG |
11781 | filepath fp(name, 0); |
11782 | InodeRef in; | |
11783 | int rc; | |
11784 | unsigned mask = statx_to_mask(flags, want); | |
11785 | ||
11fdf7f2 TL |
11786 | ldout(cct, 3) << __func__ << " " << name << dendl; |
11787 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11788 | tout(cct) << name << std::endl; |
11789 | ||
f67539c2 | 11790 | std::scoped_lock lock(client_lock); |
7c673cae FG |
11791 | rc = path_walk(fp, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); |
11792 | if (rc < 0) { | |
11793 | /* zero out mask, just in case... */ | |
11794 | stx->stx_mask = 0; | |
11795 | stx->stx_ino = 0; | |
11796 | *out = NULL; | |
11797 | return rc; | |
11798 | } else { | |
11fdf7f2 | 11799 | ceph_assert(in); |
7c673cae FG |
11800 | fill_statx(in, mask, stx); |
11801 | _ll_get(in.get()); | |
11802 | *out = in.get(); | |
11803 | return 0; | |
11804 | } | |
11805 | } | |
11806 | ||
11807 | void Client::_ll_get(Inode *in) | |
11808 | { | |
11809 | if (in->ll_ref == 0) { | |
b3b6e05e | 11810 | in->iget(); |
11fdf7f2 TL |
11811 | if (in->is_dir() && !in->dentries.empty()) { |
11812 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
11813 | in->get_first_parent()->get(); // pin dentry |
11814 | } | |
11fdf7f2 TL |
11815 | if (in->snapid != CEPH_NOSNAP) |
11816 | ll_snap_ref[in->snapid]++; | |
7c673cae FG |
11817 | } |
11818 | in->ll_get(); | |
11fdf7f2 | 11819 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " -> " << in->ll_ref << dendl; |
7c673cae FG |
11820 | } |
11821 | ||
494da23a | 11822 | int Client::_ll_put(Inode *in, uint64_t num) |
7c673cae FG |
11823 | { |
11824 | in->ll_put(num); | |
11fdf7f2 | 11825 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " " << num << " -> " << in->ll_ref << dendl; |
7c673cae | 11826 | if (in->ll_ref == 0) { |
11fdf7f2 TL |
11827 | if (in->is_dir() && !in->dentries.empty()) { |
11828 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
11829 | in->get_first_parent()->put(); // unpin dentry |
11830 | } | |
11fdf7f2 TL |
11831 | if (in->snapid != CEPH_NOSNAP) { |
11832 | auto p = ll_snap_ref.find(in->snapid); | |
11833 | ceph_assert(p != ll_snap_ref.end()); | |
11834 | ceph_assert(p->second > 0); | |
11835 | if (--p->second == 0) | |
11836 | ll_snap_ref.erase(p); | |
11837 | } | |
7c673cae FG |
11838 | put_inode(in); |
11839 | return 0; | |
11840 | } else { | |
11841 | return in->ll_ref; | |
11842 | } | |
11843 | } | |
11844 | ||
11845 | void Client::_ll_drop_pins() | |
11846 | { | |
11fdf7f2 | 11847 | ldout(cct, 10) << __func__ << dendl; |
1adf2230 | 11848 | std::set<InodeRef> to_be_put; //this set will be deconstructed item by item when exit |
7c673cae FG |
11849 | ceph::unordered_map<vinodeno_t, Inode*>::iterator next; |
11850 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
11851 | it != inode_map.end(); | |
11852 | it = next) { | |
11853 | Inode *in = it->second; | |
11854 | next = it; | |
11855 | ++next; | |
1adf2230 AA |
11856 | if (in->ll_ref){ |
11857 | to_be_put.insert(in); | |
7c673cae | 11858 | _ll_put(in, in->ll_ref); |
1adf2230 | 11859 | } |
7c673cae FG |
11860 | } |
11861 | } | |
11862 | ||
494da23a | 11863 | bool Client::_ll_forget(Inode *in, uint64_t count) |
7c673cae | 11864 | { |
11fdf7f2 | 11865 | inodeno_t ino = in->ino; |
7c673cae | 11866 | |
11fdf7f2 TL |
11867 | ldout(cct, 8) << __func__ << " " << ino << " " << count << dendl; |
11868 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11869 | tout(cct) << ino.val << std::endl; |
11870 | tout(cct) << count << std::endl; | |
11871 | ||
181888fb | 11872 | // Ignore forget if we're no longer mounted |
f67539c2 TL |
11873 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11874 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
11875 | return true; |
11876 | ||
7c673cae FG |
11877 | if (ino == 1) return true; // ignore forget on root. |
11878 | ||
11879 | bool last = false; | |
11880 | if (in->ll_ref < count) { | |
11881 | ldout(cct, 1) << "WARNING: ll_forget on " << ino << " " << count | |
11882 | << ", which only has ll_ref=" << in->ll_ref << dendl; | |
11883 | _ll_put(in, in->ll_ref); | |
11884 | last = true; | |
11885 | } else { | |
11886 | if (_ll_put(in, count) == 0) | |
11887 | last = true; | |
11888 | } | |
11889 | ||
11890 | return last; | |
11891 | } | |
11892 | ||
494da23a | 11893 | bool Client::ll_forget(Inode *in, uint64_t count) |
1adf2230 | 11894 | { |
f67539c2 | 11895 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
11896 | return _ll_forget(in, count); |
11897 | } | |
11898 | ||
7c673cae FG |
11899 | bool Client::ll_put(Inode *in) |
11900 | { | |
11901 | /* ll_forget already takes the lock */ | |
11902 | return ll_forget(in, 1); | |
11903 | } | |
11904 | ||
11fdf7f2 TL |
11905 | int Client::ll_get_snap_ref(snapid_t snap) |
11906 | { | |
f67539c2 | 11907 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
11908 | auto p = ll_snap_ref.find(snap); |
11909 | if (p != ll_snap_ref.end()) | |
11910 | return p->second; | |
11911 | return 0; | |
11912 | } | |
11913 | ||
7c673cae FG |
11914 | snapid_t Client::ll_get_snapid(Inode *in) |
11915 | { | |
f67539c2 | 11916 | std::scoped_lock lock(client_lock); |
7c673cae FG |
11917 | return in->snapid; |
11918 | } | |
11919 | ||
11920 | Inode *Client::ll_get_inode(ino_t ino) | |
11921 | { | |
f67539c2 TL |
11922 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11923 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
11924 | return NULL; |
11925 | ||
f67539c2 TL |
11926 | std::scoped_lock lock(client_lock); |
11927 | ||
7c673cae FG |
11928 | vinodeno_t vino = _map_faked_ino(ino); |
11929 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
11930 | if (p == inode_map.end()) | |
11931 | return NULL; | |
11932 | Inode *in = p->second; | |
11933 | _ll_get(in); | |
11934 | return in; | |
11935 | } | |
11936 | ||
11937 | Inode *Client::ll_get_inode(vinodeno_t vino) | |
11938 | { | |
f67539c2 TL |
11939 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11940 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
11941 | return NULL; |
11942 | ||
b3b6e05e TL |
11943 | if (is_reserved_vino(vino)) |
11944 | return NULL; | |
11945 | ||
f67539c2 TL |
11946 | std::scoped_lock lock(client_lock); |
11947 | ||
7c673cae FG |
11948 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
11949 | if (p == inode_map.end()) | |
11950 | return NULL; | |
11951 | Inode *in = p->second; | |
11952 | _ll_get(in); | |
11953 | return in; | |
11954 | } | |
11955 | ||
11956 | int Client::_ll_getattr(Inode *in, int caps, const UserPerm& perms) | |
11957 | { | |
11958 | vinodeno_t vino = _get_vino(in); | |
11959 | ||
11fdf7f2 TL |
11960 | ldout(cct, 8) << __func__ << " " << vino << dendl; |
11961 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11962 | tout(cct) << vino.ino.val << std::endl; |
11963 | ||
11964 | if (vino.snapid < CEPH_NOSNAP) | |
11965 | return 0; | |
11966 | else | |
11967 | return _getattr(in, caps, perms); | |
11968 | } | |
11969 | ||
11970 | int Client::ll_getattr(Inode *in, struct stat *attr, const UserPerm& perms) | |
11971 | { | |
f67539c2 TL |
11972 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11973 | if (!mref_reader.is_state_satisfied()) | |
11974 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11975 | |
f67539c2 | 11976 | std::scoped_lock lock(client_lock); |
181888fb | 11977 | |
7c673cae FG |
11978 | int res = _ll_getattr(in, CEPH_STAT_CAP_INODE_ALL, perms); |
11979 | ||
11980 | if (res == 0) | |
11981 | fill_stat(in, attr); | |
11fdf7f2 | 11982 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11983 | return res; |
11984 | } | |
11985 | ||
11986 | int Client::ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want, | |
11987 | unsigned int flags, const UserPerm& perms) | |
11988 | { | |
f67539c2 TL |
11989 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11990 | if (!mref_reader.is_state_satisfied()) | |
11991 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11992 | |
f67539c2 | 11993 | std::scoped_lock lock(client_lock); |
181888fb | 11994 | |
7c673cae FG |
11995 | int res = 0; |
11996 | unsigned mask = statx_to_mask(flags, want); | |
11997 | ||
94b18763 | 11998 | if (mask && !in->caps_issued_mask(mask, true)) |
7c673cae FG |
11999 | res = _ll_getattr(in, mask, perms); |
12000 | ||
12001 | if (res == 0) | |
12002 | fill_statx(in, mask, stx); | |
11fdf7f2 | 12003 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12004 | return res; |
12005 | } | |
12006 | ||
12007 | int Client::_ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
12008 | const UserPerm& perms, InodeRef *inp) | |
12009 | { | |
12010 | vinodeno_t vino = _get_vino(in); | |
12011 | ||
11fdf7f2 | 12012 | ldout(cct, 8) << __func__ << " " << vino << " mask " << hex << mask << dec |
7c673cae | 12013 | << dendl; |
11fdf7f2 | 12014 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
12015 | tout(cct) << vino.ino.val << std::endl; |
12016 | tout(cct) << stx->stx_mode << std::endl; | |
12017 | tout(cct) << stx->stx_uid << std::endl; | |
12018 | tout(cct) << stx->stx_gid << std::endl; | |
12019 | tout(cct) << stx->stx_size << std::endl; | |
12020 | tout(cct) << stx->stx_mtime << std::endl; | |
12021 | tout(cct) << stx->stx_atime << std::endl; | |
12022 | tout(cct) << stx->stx_btime << std::endl; | |
12023 | tout(cct) << mask << std::endl; | |
12024 | ||
11fdf7f2 | 12025 | if (!fuse_default_permissions) { |
7c673cae FG |
12026 | int res = may_setattr(in, stx, mask, perms); |
12027 | if (res < 0) | |
12028 | return res; | |
12029 | } | |
12030 | ||
12031 | mask &= ~(CEPH_SETATTR_MTIME_NOW | CEPH_SETATTR_ATIME_NOW); | |
12032 | ||
12033 | return __setattrx(in, stx, mask, perms, inp); | |
12034 | } | |
12035 | ||
12036 | int Client::ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
12037 | const UserPerm& perms) | |
12038 | { | |
f67539c2 TL |
12039 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12040 | if (!mref_reader.is_state_satisfied()) | |
12041 | return -CEPHFS_ENOTCONN; | |
181888fb | 12042 | |
f67539c2 | 12043 | std::scoped_lock lock(client_lock); |
181888fb | 12044 | |
7c673cae FG |
12045 | InodeRef target(in); |
12046 | int res = _ll_setattrx(in, stx, mask, perms, &target); | |
12047 | if (res == 0) { | |
11fdf7f2 | 12048 | ceph_assert(in == target.get()); |
7c673cae FG |
12049 | fill_statx(in, in->caps_issued(), stx); |
12050 | } | |
12051 | ||
11fdf7f2 | 12052 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12053 | return res; |
12054 | } | |
12055 | ||
12056 | int Client::ll_setattr(Inode *in, struct stat *attr, int mask, | |
12057 | const UserPerm& perms) | |
12058 | { | |
12059 | struct ceph_statx stx; | |
12060 | stat_to_statx(attr, &stx); | |
12061 | ||
f67539c2 TL |
12062 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12063 | if (!mref_reader.is_state_satisfied()) | |
12064 | return -CEPHFS_ENOTCONN; | |
181888fb | 12065 | |
f67539c2 | 12066 | std::scoped_lock lock(client_lock); |
181888fb | 12067 | |
7c673cae FG |
12068 | InodeRef target(in); |
12069 | int res = _ll_setattrx(in, &stx, mask, perms, &target); | |
12070 | if (res == 0) { | |
11fdf7f2 | 12071 | ceph_assert(in == target.get()); |
7c673cae FG |
12072 | fill_stat(in, attr); |
12073 | } | |
12074 | ||
11fdf7f2 | 12075 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12076 | return res; |
12077 | } | |
12078 | ||
12079 | ||
12080 | // ---------- | |
12081 | // xattrs | |
12082 | ||
12083 | int Client::getxattr(const char *path, const char *name, void *value, size_t size, | |
12084 | const UserPerm& perms) | |
12085 | { | |
f67539c2 TL |
12086 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12087 | if (!mref_reader.is_state_satisfied()) | |
12088 | return -CEPHFS_ENOTCONN; | |
181888fb | 12089 | |
f67539c2 | 12090 | std::scoped_lock lock(client_lock); |
181888fb | 12091 | |
7c673cae FG |
12092 | InodeRef in; |
12093 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12094 | if (r < 0) | |
12095 | return r; | |
12096 | return _getxattr(in, name, value, size, perms); | |
12097 | } | |
12098 | ||
12099 | int Client::lgetxattr(const char *path, const char *name, void *value, size_t size, | |
12100 | const UserPerm& perms) | |
12101 | { | |
f67539c2 TL |
12102 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12103 | if (!mref_reader.is_state_satisfied()) | |
12104 | return -CEPHFS_ENOTCONN; | |
181888fb | 12105 | |
f67539c2 | 12106 | std::scoped_lock lock(client_lock); |
181888fb | 12107 | |
7c673cae FG |
12108 | InodeRef in; |
12109 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12110 | if (r < 0) | |
12111 | return r; | |
12112 | return _getxattr(in, name, value, size, perms); | |
12113 | } | |
12114 | ||
12115 | int Client::fgetxattr(int fd, const char *name, void *value, size_t size, | |
12116 | const UserPerm& perms) | |
12117 | { | |
f67539c2 TL |
12118 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12119 | if (!mref_reader.is_state_satisfied()) | |
12120 | return -CEPHFS_ENOTCONN; | |
181888fb | 12121 | |
f67539c2 | 12122 | std::scoped_lock lock(client_lock); |
181888fb | 12123 | |
7c673cae FG |
12124 | Fh *f = get_filehandle(fd); |
12125 | if (!f) | |
f67539c2 | 12126 | return -CEPHFS_EBADF; |
7c673cae FG |
12127 | return _getxattr(f->inode, name, value, size, perms); |
12128 | } | |
12129 | ||
12130 | int Client::listxattr(const char *path, char *list, size_t size, | |
12131 | const UserPerm& perms) | |
12132 | { | |
f67539c2 TL |
12133 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12134 | if (!mref_reader.is_state_satisfied()) | |
12135 | return -CEPHFS_ENOTCONN; | |
181888fb | 12136 | |
f67539c2 | 12137 | std::scoped_lock lock(client_lock); |
181888fb | 12138 | |
7c673cae FG |
12139 | InodeRef in; |
12140 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12141 | if (r < 0) | |
12142 | return r; | |
12143 | return Client::_listxattr(in.get(), list, size, perms); | |
12144 | } | |
12145 | ||
12146 | int Client::llistxattr(const char *path, char *list, size_t size, | |
12147 | const UserPerm& perms) | |
12148 | { | |
f67539c2 TL |
12149 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12150 | if (!mref_reader.is_state_satisfied()) | |
12151 | return -CEPHFS_ENOTCONN; | |
181888fb | 12152 | |
f67539c2 | 12153 | std::scoped_lock lock(client_lock); |
181888fb | 12154 | |
7c673cae FG |
12155 | InodeRef in; |
12156 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12157 | if (r < 0) | |
12158 | return r; | |
12159 | return Client::_listxattr(in.get(), list, size, perms); | |
12160 | } | |
12161 | ||
12162 | int Client::flistxattr(int fd, char *list, size_t size, const UserPerm& perms) | |
12163 | { | |
f67539c2 TL |
12164 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12165 | if (!mref_reader.is_state_satisfied()) | |
12166 | return -CEPHFS_ENOTCONN; | |
181888fb | 12167 | |
f67539c2 | 12168 | std::scoped_lock lock(client_lock); |
181888fb | 12169 | |
7c673cae FG |
12170 | Fh *f = get_filehandle(fd); |
12171 | if (!f) | |
f67539c2 | 12172 | return -CEPHFS_EBADF; |
7c673cae FG |
12173 | return Client::_listxattr(f->inode.get(), list, size, perms); |
12174 | } | |
12175 | ||
12176 | int Client::removexattr(const char *path, const char *name, | |
12177 | const UserPerm& perms) | |
12178 | { | |
f67539c2 TL |
12179 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12180 | if (!mref_reader.is_state_satisfied()) | |
12181 | return -CEPHFS_ENOTCONN; | |
181888fb | 12182 | |
f67539c2 | 12183 | std::scoped_lock lock(client_lock); |
181888fb | 12184 | |
7c673cae FG |
12185 | InodeRef in; |
12186 | int r = Client::path_walk(path, &in, perms, true); | |
12187 | if (r < 0) | |
12188 | return r; | |
12189 | return _removexattr(in, name, perms); | |
12190 | } | |
12191 | ||
12192 | int Client::lremovexattr(const char *path, const char *name, | |
12193 | const UserPerm& perms) | |
12194 | { | |
f67539c2 TL |
12195 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12196 | if (!mref_reader.is_state_satisfied()) | |
12197 | return -CEPHFS_ENOTCONN; | |
181888fb | 12198 | |
f67539c2 | 12199 | std::scoped_lock lock(client_lock); |
181888fb | 12200 | |
7c673cae FG |
12201 | InodeRef in; |
12202 | int r = Client::path_walk(path, &in, perms, false); | |
12203 | if (r < 0) | |
12204 | return r; | |
12205 | return _removexattr(in, name, perms); | |
12206 | } | |
12207 | ||
12208 | int Client::fremovexattr(int fd, const char *name, const UserPerm& perms) | |
12209 | { | |
f67539c2 TL |
12210 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12211 | if (!mref_reader.is_state_satisfied()) | |
12212 | return -CEPHFS_ENOTCONN; | |
181888fb | 12213 | |
f67539c2 | 12214 | std::scoped_lock lock(client_lock); |
181888fb | 12215 | |
7c673cae FG |
12216 | Fh *f = get_filehandle(fd); |
12217 | if (!f) | |
f67539c2 | 12218 | return -CEPHFS_EBADF; |
7c673cae FG |
12219 | return _removexattr(f->inode, name, perms); |
12220 | } | |
12221 | ||
12222 | int Client::setxattr(const char *path, const char *name, const void *value, | |
12223 | size_t size, int flags, const UserPerm& perms) | |
12224 | { | |
f67539c2 TL |
12225 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12226 | if (!mref_reader.is_state_satisfied()) | |
12227 | return -CEPHFS_ENOTCONN; | |
12228 | ||
7c673cae FG |
12229 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
12230 | ||
f67539c2 | 12231 | std::scoped_lock lock(client_lock); |
181888fb | 12232 | |
7c673cae FG |
12233 | InodeRef in; |
12234 | int r = Client::path_walk(path, &in, perms, true); | |
12235 | if (r < 0) | |
12236 | return r; | |
12237 | return _setxattr(in, name, value, size, flags, perms); | |
12238 | } | |
12239 | ||
12240 | int Client::lsetxattr(const char *path, const char *name, const void *value, | |
12241 | size_t size, int flags, const UserPerm& perms) | |
12242 | { | |
f67539c2 TL |
12243 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12244 | if (!mref_reader.is_state_satisfied()) | |
12245 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12246 | |
f67539c2 | 12247 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12248 | |
f67539c2 | 12249 | std::scoped_lock lock(client_lock); |
181888fb | 12250 | |
7c673cae FG |
12251 | InodeRef in; |
12252 | int r = Client::path_walk(path, &in, perms, false); | |
12253 | if (r < 0) | |
12254 | return r; | |
12255 | return _setxattr(in, name, value, size, flags, perms); | |
12256 | } | |
12257 | ||
12258 | int Client::fsetxattr(int fd, const char *name, const void *value, size_t size, | |
12259 | int flags, const UserPerm& perms) | |
12260 | { | |
f67539c2 TL |
12261 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12262 | if (!mref_reader.is_state_satisfied()) | |
12263 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12264 | |
f67539c2 | 12265 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12266 | |
f67539c2 | 12267 | std::scoped_lock lock(client_lock); |
181888fb | 12268 | |
7c673cae FG |
12269 | Fh *f = get_filehandle(fd); |
12270 | if (!f) | |
f67539c2 | 12271 | return -CEPHFS_EBADF; |
7c673cae FG |
12272 | return _setxattr(f->inode, name, value, size, flags, perms); |
12273 | } | |
12274 | ||
12275 | int Client::_getxattr(Inode *in, const char *name, void *value, size_t size, | |
12276 | const UserPerm& perms) | |
12277 | { | |
12278 | int r; | |
12279 | ||
12280 | const VXattr *vxattr = _match_vxattr(in, name); | |
12281 | if (vxattr) { | |
f67539c2 | 12282 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12283 | |
12284 | // Do a force getattr to get the latest quota before returning | |
12285 | // a value to userspace. | |
28e407b8 AA |
12286 | int flags = 0; |
12287 | if (vxattr->flags & VXATTR_RSTAT) { | |
12288 | flags |= CEPH_STAT_RSTAT; | |
12289 | } | |
adb31ebb TL |
12290 | if (vxattr->flags & VXATTR_DIRSTAT) { |
12291 | flags |= CEPH_CAP_FILE_SHARED; | |
12292 | } | |
f67539c2 | 12293 | r = _getattr(in, flags | CEPH_STAT_CAP_XATTR, perms, true); |
7c673cae FG |
12294 | if (r != 0) { |
12295 | // Error from getattr! | |
12296 | return r; | |
12297 | } | |
12298 | ||
12299 | // call pointer-to-member function | |
12300 | char buf[256]; | |
12301 | if (!(vxattr->exists_cb && !(this->*(vxattr->exists_cb))(in))) { | |
12302 | r = (this->*(vxattr->getxattr_cb))(in, buf, sizeof(buf)); | |
12303 | } else { | |
f67539c2 | 12304 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12305 | } |
12306 | ||
12307 | if (size != 0) { | |
12308 | if (r > (int)size) { | |
f67539c2 | 12309 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12310 | } else if (r > 0) { |
12311 | memcpy(value, buf, r); | |
12312 | } | |
12313 | } | |
12314 | goto out; | |
12315 | } | |
12316 | ||
12317 | if (acl_type == NO_ACL && !strncmp(name, "system.", 7)) { | |
f67539c2 | 12318 | r = -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12319 | goto out; |
12320 | } | |
12321 | ||
12322 | r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
12323 | if (r == 0) { | |
12324 | string n(name); | |
f67539c2 | 12325 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12326 | if (in->xattrs.count(n)) { |
12327 | r = in->xattrs[n].length(); | |
12328 | if (r > 0 && size != 0) { | |
12329 | if (size >= (unsigned)r) | |
12330 | memcpy(value, in->xattrs[n].c_str(), r); | |
12331 | else | |
f67539c2 | 12332 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12333 | } |
12334 | } | |
12335 | } | |
12336 | out: | |
1adf2230 | 12337 | ldout(cct, 8) << "_getxattr(" << in->ino << ", \"" << name << "\", " << size << ") = " << r << dendl; |
7c673cae FG |
12338 | return r; |
12339 | } | |
12340 | ||
12341 | int Client::_getxattr(InodeRef &in, const char *name, void *value, size_t size, | |
12342 | const UserPerm& perms) | |
12343 | { | |
12344 | if (cct->_conf->client_permissions) { | |
12345 | int r = xattr_permission(in.get(), name, MAY_READ, perms); | |
12346 | if (r < 0) | |
12347 | return r; | |
12348 | } | |
12349 | return _getxattr(in.get(), name, value, size, perms); | |
12350 | } | |
12351 | ||
12352 | int Client::ll_getxattr(Inode *in, const char *name, void *value, | |
12353 | size_t size, const UserPerm& perms) | |
12354 | { | |
f67539c2 TL |
12355 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12356 | if (!mref_reader.is_state_satisfied()) | |
12357 | return -CEPHFS_ENOTCONN; | |
181888fb | 12358 | |
7c673cae FG |
12359 | vinodeno_t vino = _get_vino(in); |
12360 | ||
11fdf7f2 TL |
12361 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12362 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12363 | tout(cct) << vino.ino.val << std::endl; |
12364 | tout(cct) << name << std::endl; | |
12365 | ||
f67539c2 | 12366 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12367 | if (!fuse_default_permissions) { |
7c673cae FG |
12368 | int r = xattr_permission(in, name, MAY_READ, perms); |
12369 | if (r < 0) | |
12370 | return r; | |
12371 | } | |
12372 | ||
12373 | return _getxattr(in, name, value, size, perms); | |
12374 | } | |
12375 | ||
12376 | int Client::_listxattr(Inode *in, char *name, size_t size, | |
12377 | const UserPerm& perms) | |
12378 | { | |
81eedcae | 12379 | bool len_only = (size == 0); |
7c673cae | 12380 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); |
81eedcae TL |
12381 | if (r != 0) { |
12382 | goto out; | |
12383 | } | |
7c673cae | 12384 | |
81eedcae | 12385 | r = 0; |
f67539c2 TL |
12386 | for ([[maybe_unused]] const auto &[xattr_name, xattr_value_bl] : in->xattrs) { |
12387 | if (xattr_name.rfind("ceph.", 0) == 0) { | |
12388 | continue; | |
12389 | } | |
12390 | ||
12391 | size_t this_len = xattr_name.length() + 1; | |
81eedcae TL |
12392 | r += this_len; |
12393 | if (len_only) | |
12394 | continue; | |
7c673cae | 12395 | |
81eedcae | 12396 | if (this_len > size) { |
f67539c2 | 12397 | r = -CEPHFS_ERANGE; |
81eedcae TL |
12398 | goto out; |
12399 | } | |
12400 | ||
f67539c2 | 12401 | memcpy(name, xattr_name.c_str(), this_len); |
81eedcae TL |
12402 | name += this_len; |
12403 | size -= this_len; | |
12404 | } | |
81eedcae | 12405 | out: |
11fdf7f2 | 12406 | ldout(cct, 8) << __func__ << "(" << in->ino << ", " << size << ") = " << r << dendl; |
7c673cae FG |
12407 | return r; |
12408 | } | |
12409 | ||
12410 | int Client::ll_listxattr(Inode *in, char *names, size_t size, | |
12411 | const UserPerm& perms) | |
12412 | { | |
f67539c2 TL |
12413 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12414 | if (!mref_reader.is_state_satisfied()) | |
12415 | return -CEPHFS_ENOTCONN; | |
181888fb | 12416 | |
7c673cae FG |
12417 | vinodeno_t vino = _get_vino(in); |
12418 | ||
11fdf7f2 TL |
12419 | ldout(cct, 3) << __func__ << " " << vino << " size " << size << dendl; |
12420 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12421 | tout(cct) << vino.ino.val << std::endl; |
12422 | tout(cct) << size << std::endl; | |
12423 | ||
f67539c2 | 12424 | std::scoped_lock lock(client_lock); |
7c673cae FG |
12425 | return _listxattr(in, names, size, perms); |
12426 | } | |
12427 | ||
12428 | int Client::_do_setxattr(Inode *in, const char *name, const void *value, | |
12429 | size_t size, int flags, const UserPerm& perms) | |
12430 | { | |
12431 | ||
12432 | int xattr_flags = 0; | |
12433 | if (!value) | |
12434 | xattr_flags |= CEPH_XATTR_REMOVE; | |
12435 | if (flags & XATTR_CREATE) | |
12436 | xattr_flags |= CEPH_XATTR_CREATE; | |
12437 | if (flags & XATTR_REPLACE) | |
12438 | xattr_flags |= CEPH_XATTR_REPLACE; | |
12439 | ||
12440 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETXATTR); | |
12441 | filepath path; | |
12442 | in->make_nosnap_relative_path(path); | |
12443 | req->set_filepath(path); | |
12444 | req->set_string2(name); | |
12445 | req->set_inode(in); | |
12446 | req->head.args.setxattr.flags = xattr_flags; | |
12447 | ||
12448 | bufferlist bl; | |
20effc67 | 12449 | ceph_assert(value || size == 0); |
7c673cae FG |
12450 | bl.append((const char*)value, size); |
12451 | req->set_data(bl); | |
12452 | ||
12453 | int res = make_request(req, perms); | |
12454 | ||
12455 | trim_cache(); | |
11fdf7f2 | 12456 | ldout(cct, 3) << __func__ << "(" << in->ino << ", \"" << name << "\") = " << |
7c673cae FG |
12457 | res << dendl; |
12458 | return res; | |
12459 | } | |
12460 | ||
12461 | int Client::_setxattr(Inode *in, const char *name, const void *value, | |
12462 | size_t size, int flags, const UserPerm& perms) | |
12463 | { | |
12464 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12465 | return -CEPHFS_EROFS; |
7c673cae FG |
12466 | } |
12467 | ||
f6b5b4d7 TL |
12468 | if (size == 0) { |
12469 | value = ""; | |
12470 | } else if (value == NULL) { | |
f67539c2 | 12471 | return -CEPHFS_EINVAL; |
f6b5b4d7 TL |
12472 | } |
12473 | ||
7c673cae FG |
12474 | bool posix_acl_xattr = false; |
12475 | if (acl_type == POSIX_ACL) | |
12476 | posix_acl_xattr = !strncmp(name, "system.", 7); | |
12477 | ||
12478 | if (strncmp(name, "user.", 5) && | |
12479 | strncmp(name, "security.", 9) && | |
12480 | strncmp(name, "trusted.", 8) && | |
12481 | strncmp(name, "ceph.", 5) && | |
12482 | !posix_acl_xattr) | |
f67539c2 | 12483 | return -CEPHFS_EOPNOTSUPP; |
7c673cae | 12484 | |
11fdf7f2 TL |
12485 | bool check_realm = false; |
12486 | ||
7c673cae FG |
12487 | if (posix_acl_xattr) { |
12488 | if (!strcmp(name, ACL_EA_ACCESS)) { | |
12489 | mode_t new_mode = in->mode; | |
12490 | if (value) { | |
12491 | int ret = posix_acl_equiv_mode(value, size, &new_mode); | |
12492 | if (ret < 0) | |
12493 | return ret; | |
12494 | if (ret == 0) { | |
12495 | value = NULL; | |
12496 | size = 0; | |
12497 | } | |
12498 | if (new_mode != in->mode) { | |
12499 | struct ceph_statx stx; | |
12500 | stx.stx_mode = new_mode; | |
12501 | ret = _do_setattr(in, &stx, CEPH_SETATTR_MODE, perms, NULL); | |
12502 | if (ret < 0) | |
12503 | return ret; | |
12504 | } | |
12505 | } | |
12506 | } else if (!strcmp(name, ACL_EA_DEFAULT)) { | |
12507 | if (value) { | |
12508 | if (!S_ISDIR(in->mode)) | |
f67539c2 | 12509 | return -CEPHFS_EACCES; |
7c673cae FG |
12510 | int ret = posix_acl_check(value, size); |
12511 | if (ret < 0) | |
f67539c2 | 12512 | return -CEPHFS_EINVAL; |
7c673cae FG |
12513 | if (ret == 0) { |
12514 | value = NULL; | |
12515 | size = 0; | |
12516 | } | |
12517 | } | |
12518 | } else { | |
f67539c2 | 12519 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12520 | } |
12521 | } else { | |
12522 | const VXattr *vxattr = _match_vxattr(in, name); | |
11fdf7f2 TL |
12523 | if (vxattr) { |
12524 | if (vxattr->readonly) | |
f67539c2 | 12525 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12526 | if (vxattr->name.compare(0, 10, "ceph.quota") == 0 && value) |
12527 | check_realm = true; | |
12528 | } | |
7c673cae FG |
12529 | } |
12530 | ||
11fdf7f2 TL |
12531 | int ret = _do_setxattr(in, name, value, size, flags, perms); |
12532 | if (ret >= 0 && check_realm) { | |
12533 | // check if snaprealm was created for quota inode | |
12534 | if (in->quota.is_enable() && | |
12535 | !(in->snaprealm && in->snaprealm->ino == in->ino)) | |
f67539c2 | 12536 | ret = -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12537 | } |
12538 | ||
12539 | return ret; | |
7c673cae FG |
12540 | } |
12541 | ||
12542 | int Client::_setxattr(InodeRef &in, const char *name, const void *value, | |
12543 | size_t size, int flags, const UserPerm& perms) | |
12544 | { | |
12545 | if (cct->_conf->client_permissions) { | |
12546 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12547 | if (r < 0) | |
12548 | return r; | |
12549 | } | |
12550 | return _setxattr(in.get(), name, value, size, flags, perms); | |
12551 | } | |
12552 | ||
12553 | int Client::_setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap) | |
12554 | { | |
12555 | string tmp; | |
12556 | if (name == "layout") { | |
12557 | string::iterator begin = value.begin(); | |
12558 | string::iterator end = value.end(); | |
12559 | keys_and_values<string::iterator> p; // create instance of parser | |
12560 | std::map<string, string> m; // map to receive results | |
12561 | if (!qi::parse(begin, end, p, m)) { // returns true if successful | |
f67539c2 | 12562 | return -CEPHFS_EINVAL; |
7c673cae FG |
12563 | } |
12564 | if (begin != end) | |
f67539c2 | 12565 | return -CEPHFS_EINVAL; |
7c673cae FG |
12566 | for (map<string,string>::iterator q = m.begin(); q != m.end(); ++q) { |
12567 | if (q->first == "pool") { | |
12568 | tmp = q->second; | |
12569 | break; | |
12570 | } | |
12571 | } | |
12572 | } else if (name == "layout.pool") { | |
12573 | tmp = value; | |
12574 | } | |
12575 | ||
12576 | if (tmp.length()) { | |
12577 | int64_t pool; | |
12578 | try { | |
12579 | pool = boost::lexical_cast<unsigned>(tmp); | |
12580 | if (!osdmap->have_pg_pool(pool)) | |
f67539c2 | 12581 | return -CEPHFS_ENOENT; |
7c673cae FG |
12582 | } catch (boost::bad_lexical_cast const&) { |
12583 | pool = osdmap->lookup_pg_pool_name(tmp); | |
12584 | if (pool < 0) { | |
f67539c2 | 12585 | return -CEPHFS_ENOENT; |
7c673cae FG |
12586 | } |
12587 | } | |
12588 | } | |
12589 | ||
12590 | return 0; | |
12591 | } | |
12592 | ||
12593 | void Client::_setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t size) | |
12594 | { | |
12595 | // For setting pool of layout, MetaRequest need osdmap epoch. | |
12596 | // There is a race which create a new data pool but client and mds both don't have. | |
12597 | // Make client got the latest osdmap which make mds quickly judge whether get newer osdmap. | |
f67539c2 | 12598 | ldout(cct, 15) << __func__ << ": name = " << name << dendl; |
7c673cae FG |
12599 | if (strcmp(name, "ceph.file.layout.pool") == 0 || strcmp(name, "ceph.dir.layout.pool") == 0 || |
12600 | strcmp(name, "ceph.file.layout") == 0 || strcmp(name, "ceph.dir.layout") == 0) { | |
12601 | string rest(strstr(name, "layout")); | |
12602 | string v((const char*)value, size); | |
12603 | int r = objecter->with_osdmap([&](const OSDMap& o) { | |
12604 | return _setxattr_check_data_pool(rest, v, &o); | |
12605 | }); | |
12606 | ||
f67539c2 TL |
12607 | if (r == -CEPHFS_ENOENT) { |
12608 | bs::error_code ec; | |
12609 | ldout(cct, 20) << __func__ << ": waiting for latest osdmap" << dendl; | |
12610 | objecter->wait_for_latest_osdmap(ca::use_blocked[ec]); | |
12611 | ldout(cct, 20) << __func__ << ": got latest osdmap: " << ec << dendl; | |
7c673cae FG |
12612 | } |
12613 | } | |
12614 | } | |
12615 | ||
12616 | int Client::ll_setxattr(Inode *in, const char *name, const void *value, | |
12617 | size_t size, int flags, const UserPerm& perms) | |
12618 | { | |
f67539c2 TL |
12619 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12620 | if (!mref_reader.is_state_satisfied()) | |
12621 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12622 | |
f67539c2 | 12623 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12624 | |
7c673cae FG |
12625 | vinodeno_t vino = _get_vino(in); |
12626 | ||
11fdf7f2 TL |
12627 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12628 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12629 | tout(cct) << vino.ino.val << std::endl; |
12630 | tout(cct) << name << std::endl; | |
12631 | ||
f67539c2 | 12632 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12633 | if (!fuse_default_permissions) { |
7c673cae FG |
12634 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12635 | if (r < 0) | |
12636 | return r; | |
12637 | } | |
12638 | return _setxattr(in, name, value, size, flags, perms); | |
12639 | } | |
12640 | ||
12641 | int Client::_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12642 | { | |
12643 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12644 | return -CEPHFS_EROFS; |
7c673cae FG |
12645 | } |
12646 | ||
12647 | // same xattrs supported by kernel client | |
12648 | if (strncmp(name, "user.", 5) && | |
12649 | strncmp(name, "system.", 7) && | |
12650 | strncmp(name, "security.", 9) && | |
12651 | strncmp(name, "trusted.", 8) && | |
12652 | strncmp(name, "ceph.", 5)) | |
f67539c2 | 12653 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12654 | |
12655 | const VXattr *vxattr = _match_vxattr(in, name); | |
12656 | if (vxattr && vxattr->readonly) | |
f67539c2 | 12657 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12658 | |
12659 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_RMXATTR); | |
12660 | filepath path; | |
12661 | in->make_nosnap_relative_path(path); | |
12662 | req->set_filepath(path); | |
12663 | req->set_filepath2(name); | |
12664 | req->set_inode(in); | |
12665 | ||
12666 | int res = make_request(req, perms); | |
12667 | ||
12668 | trim_cache(); | |
1adf2230 | 12669 | ldout(cct, 8) << "_removexattr(" << in->ino << ", \"" << name << "\") = " << res << dendl; |
7c673cae FG |
12670 | return res; |
12671 | } | |
12672 | ||
12673 | int Client::_removexattr(InodeRef &in, const char *name, const UserPerm& perms) | |
12674 | { | |
12675 | if (cct->_conf->client_permissions) { | |
12676 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12677 | if (r < 0) | |
12678 | return r; | |
12679 | } | |
12680 | return _removexattr(in.get(), name, perms); | |
12681 | } | |
12682 | ||
12683 | int Client::ll_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12684 | { | |
f67539c2 TL |
12685 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12686 | if (!mref_reader.is_state_satisfied()) | |
12687 | return -CEPHFS_ENOTCONN; | |
181888fb | 12688 | |
7c673cae FG |
12689 | vinodeno_t vino = _get_vino(in); |
12690 | ||
12691 | ldout(cct, 3) << "ll_removexattr " << vino << " " << name << dendl; | |
12692 | tout(cct) << "ll_removexattr" << std::endl; | |
12693 | tout(cct) << vino.ino.val << std::endl; | |
12694 | tout(cct) << name << std::endl; | |
12695 | ||
f67539c2 | 12696 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12697 | if (!fuse_default_permissions) { |
7c673cae FG |
12698 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12699 | if (r < 0) | |
12700 | return r; | |
12701 | } | |
12702 | ||
12703 | return _removexattr(in, name, perms); | |
12704 | } | |
12705 | ||
12706 | bool Client::_vxattrcb_quota_exists(Inode *in) | |
12707 | { | |
11fdf7f2 | 12708 | return in->quota.is_enable() && |
f6b5b4d7 TL |
12709 | (in->snapid != CEPH_NOSNAP || |
12710 | (in->snaprealm && in->snaprealm->ino == in->ino)); | |
7c673cae FG |
12711 | } |
12712 | size_t Client::_vxattrcb_quota(Inode *in, char *val, size_t size) | |
12713 | { | |
12714 | return snprintf(val, size, | |
12715 | "max_bytes=%lld max_files=%lld", | |
12716 | (long long int)in->quota.max_bytes, | |
12717 | (long long int)in->quota.max_files); | |
12718 | } | |
12719 | size_t Client::_vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size) | |
12720 | { | |
12721 | return snprintf(val, size, "%lld", (long long int)in->quota.max_bytes); | |
12722 | } | |
12723 | size_t Client::_vxattrcb_quota_max_files(Inode *in, char *val, size_t size) | |
12724 | { | |
12725 | return snprintf(val, size, "%lld", (long long int)in->quota.max_files); | |
12726 | } | |
12727 | ||
12728 | bool Client::_vxattrcb_layout_exists(Inode *in) | |
12729 | { | |
12730 | return in->layout != file_layout_t(); | |
12731 | } | |
12732 | size_t Client::_vxattrcb_layout(Inode *in, char *val, size_t size) | |
12733 | { | |
12734 | int r = snprintf(val, size, | |
11fdf7f2 | 12735 | "stripe_unit=%llu stripe_count=%llu object_size=%llu pool=", |
7c673cae FG |
12736 | (unsigned long long)in->layout.stripe_unit, |
12737 | (unsigned long long)in->layout.stripe_count, | |
12738 | (unsigned long long)in->layout.object_size); | |
12739 | objecter->with_osdmap([&](const OSDMap& o) { | |
12740 | if (o.have_pg_pool(in->layout.pool_id)) | |
12741 | r += snprintf(val + r, size - r, "%s", | |
12742 | o.get_pool_name(in->layout.pool_id).c_str()); | |
12743 | else | |
12744 | r += snprintf(val + r, size - r, "%" PRIu64, | |
12745 | (uint64_t)in->layout.pool_id); | |
12746 | }); | |
12747 | if (in->layout.pool_ns.length()) | |
12748 | r += snprintf(val + r, size - r, " pool_namespace=%s", | |
12749 | in->layout.pool_ns.c_str()); | |
12750 | return r; | |
12751 | } | |
12752 | size_t Client::_vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size) | |
12753 | { | |
11fdf7f2 | 12754 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_unit); |
7c673cae FG |
12755 | } |
12756 | size_t Client::_vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size) | |
12757 | { | |
11fdf7f2 | 12758 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_count); |
7c673cae FG |
12759 | } |
12760 | size_t Client::_vxattrcb_layout_object_size(Inode *in, char *val, size_t size) | |
12761 | { | |
11fdf7f2 | 12762 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.object_size); |
7c673cae FG |
12763 | } |
12764 | size_t Client::_vxattrcb_layout_pool(Inode *in, char *val, size_t size) | |
12765 | { | |
12766 | size_t r; | |
12767 | objecter->with_osdmap([&](const OSDMap& o) { | |
12768 | if (o.have_pg_pool(in->layout.pool_id)) | |
12769 | r = snprintf(val, size, "%s", o.get_pool_name( | |
12770 | in->layout.pool_id).c_str()); | |
12771 | else | |
12772 | r = snprintf(val, size, "%" PRIu64, (uint64_t)in->layout.pool_id); | |
12773 | }); | |
12774 | return r; | |
12775 | } | |
12776 | size_t Client::_vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size) | |
12777 | { | |
12778 | return snprintf(val, size, "%s", in->layout.pool_ns.c_str()); | |
12779 | } | |
12780 | size_t Client::_vxattrcb_dir_entries(Inode *in, char *val, size_t size) | |
12781 | { | |
11fdf7f2 | 12782 | return snprintf(val, size, "%llu", (unsigned long long)(in->dirstat.nfiles + in->dirstat.nsubdirs)); |
7c673cae FG |
12783 | } |
12784 | size_t Client::_vxattrcb_dir_files(Inode *in, char *val, size_t size) | |
12785 | { | |
11fdf7f2 | 12786 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nfiles); |
7c673cae FG |
12787 | } |
12788 | size_t Client::_vxattrcb_dir_subdirs(Inode *in, char *val, size_t size) | |
12789 | { | |
11fdf7f2 | 12790 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nsubdirs); |
7c673cae FG |
12791 | } |
12792 | size_t Client::_vxattrcb_dir_rentries(Inode *in, char *val, size_t size) | |
12793 | { | |
11fdf7f2 | 12794 | return snprintf(val, size, "%llu", (unsigned long long)(in->rstat.rfiles + in->rstat.rsubdirs)); |
7c673cae FG |
12795 | } |
12796 | size_t Client::_vxattrcb_dir_rfiles(Inode *in, char *val, size_t size) | |
12797 | { | |
11fdf7f2 | 12798 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rfiles); |
7c673cae FG |
12799 | } |
12800 | size_t Client::_vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size) | |
12801 | { | |
11fdf7f2 | 12802 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsubdirs); |
7c673cae | 12803 | } |
f67539c2 TL |
12804 | size_t Client::_vxattrcb_dir_rsnaps(Inode *in, char *val, size_t size) |
12805 | { | |
12806 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsnaps); | |
12807 | } | |
7c673cae FG |
12808 | size_t Client::_vxattrcb_dir_rbytes(Inode *in, char *val, size_t size) |
12809 | { | |
11fdf7f2 | 12810 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rbytes); |
7c673cae FG |
12811 | } |
12812 | size_t Client::_vxattrcb_dir_rctime(Inode *in, char *val, size_t size) | |
12813 | { | |
81eedcae | 12814 | return snprintf(val, size, "%ld.%09ld", (long)in->rstat.rctime.sec(), |
7c673cae FG |
12815 | (long)in->rstat.rctime.nsec()); |
12816 | } | |
11fdf7f2 TL |
12817 | bool Client::_vxattrcb_dir_pin_exists(Inode *in) |
12818 | { | |
f67539c2 | 12819 | return in->dir_pin != -CEPHFS_ENODATA; |
11fdf7f2 TL |
12820 | } |
12821 | size_t Client::_vxattrcb_dir_pin(Inode *in, char *val, size_t size) | |
12822 | { | |
12823 | return snprintf(val, size, "%ld", (long)in->dir_pin); | |
12824 | } | |
7c673cae | 12825 | |
81eedcae TL |
12826 | bool Client::_vxattrcb_snap_btime_exists(Inode *in) |
12827 | { | |
12828 | return !in->snap_btime.is_zero(); | |
12829 | } | |
12830 | ||
12831 | size_t Client::_vxattrcb_snap_btime(Inode *in, char *val, size_t size) | |
12832 | { | |
12833 | return snprintf(val, size, "%llu.%09lu", | |
12834 | (long long unsigned)in->snap_btime.sec(), | |
12835 | (long unsigned)in->snap_btime.nsec()); | |
12836 | } | |
12837 | ||
20effc67 TL |
12838 | size_t Client::_vxattrcb_caps(Inode *in, char *val, size_t size) |
12839 | { | |
12840 | int issued; | |
12841 | ||
12842 | in->caps_issued(&issued); | |
12843 | return snprintf(val, size, "%s/0x%x", ccap_string(issued).c_str(), issued); | |
12844 | } | |
12845 | ||
f67539c2 TL |
12846 | bool Client::_vxattrcb_mirror_info_exists(Inode *in) |
12847 | { | |
12848 | // checking one of the xattrs would suffice | |
12849 | return in->xattrs.count("ceph.mirror.info.cluster_id") != 0; | |
12850 | } | |
12851 | ||
12852 | size_t Client::_vxattrcb_mirror_info(Inode *in, char *val, size_t size) | |
12853 | { | |
12854 | return snprintf(val, size, "cluster_id=%.*s fs_id=%.*s", | |
12855 | in->xattrs["ceph.mirror.info.cluster_id"].length(), | |
12856 | in->xattrs["ceph.mirror.info.cluster_id"].c_str(), | |
12857 | in->xattrs["ceph.mirror.info.fs_id"].length(), | |
12858 | in->xattrs["ceph.mirror.info.fs_id"].c_str()); | |
12859 | } | |
12860 | ||
adb31ebb TL |
12861 | size_t Client::_vxattrcb_cluster_fsid(Inode *in, char *val, size_t size) |
12862 | { | |
12863 | return snprintf(val, size, "%s", monclient->get_fsid().to_string().c_str()); | |
12864 | } | |
12865 | ||
12866 | size_t Client::_vxattrcb_client_id(Inode *in, char *val, size_t size) | |
12867 | { | |
12868 | auto name = messenger->get_myname(); | |
20effc67 | 12869 | return snprintf(val, size, "%s%" PRId64, name.type_str(), name.num()); |
adb31ebb TL |
12870 | } |
12871 | ||
7c673cae FG |
12872 | #define CEPH_XATTR_NAME(_type, _name) "ceph." #_type "." #_name |
12873 | #define CEPH_XATTR_NAME2(_type, _name, _name2) "ceph." #_type "." #_name "." #_name2 | |
12874 | ||
adb31ebb | 12875 | #define XATTR_NAME_CEPH(_type, _name, _flags) \ |
28e407b8 AA |
12876 | { \ |
12877 | name: CEPH_XATTR_NAME(_type, _name), \ | |
12878 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
12879 | readonly: true, \ | |
28e407b8 AA |
12880 | exists_cb: NULL, \ |
12881 | flags: _flags, \ | |
7c673cae FG |
12882 | } |
12883 | #define XATTR_LAYOUT_FIELD(_type, _name, _field) \ | |
12884 | { \ | |
12885 | name: CEPH_XATTR_NAME2(_type, _name, _field), \ | |
12886 | getxattr_cb: &Client::_vxattrcb_ ## _name ## _ ## _field, \ | |
12887 | readonly: false, \ | |
7c673cae | 12888 | exists_cb: &Client::_vxattrcb_layout_exists, \ |
28e407b8 | 12889 | flags: 0, \ |
7c673cae FG |
12890 | } |
12891 | #define XATTR_QUOTA_FIELD(_type, _name) \ | |
12892 | { \ | |
12893 | name: CEPH_XATTR_NAME(_type, _name), \ | |
12894 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
12895 | readonly: false, \ | |
7c673cae | 12896 | exists_cb: &Client::_vxattrcb_quota_exists, \ |
28e407b8 | 12897 | flags: 0, \ |
7c673cae FG |
12898 | } |
12899 | ||
12900 | const Client::VXattr Client::_dir_vxattrs[] = { | |
12901 | { | |
12902 | name: "ceph.dir.layout", | |
12903 | getxattr_cb: &Client::_vxattrcb_layout, | |
12904 | readonly: false, | |
7c673cae | 12905 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 12906 | flags: 0, |
7c673cae FG |
12907 | }, |
12908 | XATTR_LAYOUT_FIELD(dir, layout, stripe_unit), | |
12909 | XATTR_LAYOUT_FIELD(dir, layout, stripe_count), | |
12910 | XATTR_LAYOUT_FIELD(dir, layout, object_size), | |
12911 | XATTR_LAYOUT_FIELD(dir, layout, pool), | |
12912 | XATTR_LAYOUT_FIELD(dir, layout, pool_namespace), | |
adb31ebb TL |
12913 | XATTR_NAME_CEPH(dir, entries, VXATTR_DIRSTAT), |
12914 | XATTR_NAME_CEPH(dir, files, VXATTR_DIRSTAT), | |
12915 | XATTR_NAME_CEPH(dir, subdirs, VXATTR_DIRSTAT), | |
12916 | XATTR_NAME_CEPH(dir, rentries, VXATTR_RSTAT), | |
12917 | XATTR_NAME_CEPH(dir, rfiles, VXATTR_RSTAT), | |
12918 | XATTR_NAME_CEPH(dir, rsubdirs, VXATTR_RSTAT), | |
f67539c2 | 12919 | XATTR_NAME_CEPH(dir, rsnaps, VXATTR_RSTAT), |
adb31ebb TL |
12920 | XATTR_NAME_CEPH(dir, rbytes, VXATTR_RSTAT), |
12921 | XATTR_NAME_CEPH(dir, rctime, VXATTR_RSTAT), | |
7c673cae FG |
12922 | { |
12923 | name: "ceph.quota", | |
12924 | getxattr_cb: &Client::_vxattrcb_quota, | |
12925 | readonly: false, | |
7c673cae | 12926 | exists_cb: &Client::_vxattrcb_quota_exists, |
28e407b8 | 12927 | flags: 0, |
7c673cae FG |
12928 | }, |
12929 | XATTR_QUOTA_FIELD(quota, max_bytes), | |
12930 | XATTR_QUOTA_FIELD(quota, max_files), | |
11fdf7f2 TL |
12931 | { |
12932 | name: "ceph.dir.pin", | |
12933 | getxattr_cb: &Client::_vxattrcb_dir_pin, | |
12934 | readonly: false, | |
11fdf7f2 TL |
12935 | exists_cb: &Client::_vxattrcb_dir_pin_exists, |
12936 | flags: 0, | |
12937 | }, | |
81eedcae TL |
12938 | { |
12939 | name: "ceph.snap.btime", | |
12940 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
12941 | readonly: true, | |
81eedcae TL |
12942 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
12943 | flags: 0, | |
12944 | }, | |
f67539c2 TL |
12945 | { |
12946 | name: "ceph.mirror.info", | |
12947 | getxattr_cb: &Client::_vxattrcb_mirror_info, | |
12948 | readonly: false, | |
12949 | exists_cb: &Client::_vxattrcb_mirror_info_exists, | |
12950 | flags: 0, | |
12951 | }, | |
20effc67 TL |
12952 | { |
12953 | name: "ceph.caps", | |
12954 | getxattr_cb: &Client::_vxattrcb_caps, | |
12955 | readonly: true, | |
12956 | exists_cb: NULL, | |
12957 | flags: 0, | |
12958 | }, | |
7c673cae FG |
12959 | { name: "" } /* Required table terminator */ |
12960 | }; | |
12961 | ||
12962 | const Client::VXattr Client::_file_vxattrs[] = { | |
12963 | { | |
12964 | name: "ceph.file.layout", | |
12965 | getxattr_cb: &Client::_vxattrcb_layout, | |
12966 | readonly: false, | |
7c673cae | 12967 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 12968 | flags: 0, |
7c673cae FG |
12969 | }, |
12970 | XATTR_LAYOUT_FIELD(file, layout, stripe_unit), | |
12971 | XATTR_LAYOUT_FIELD(file, layout, stripe_count), | |
12972 | XATTR_LAYOUT_FIELD(file, layout, object_size), | |
12973 | XATTR_LAYOUT_FIELD(file, layout, pool), | |
12974 | XATTR_LAYOUT_FIELD(file, layout, pool_namespace), | |
81eedcae TL |
12975 | { |
12976 | name: "ceph.snap.btime", | |
12977 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
12978 | readonly: true, | |
81eedcae TL |
12979 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
12980 | flags: 0, | |
12981 | }, | |
20effc67 TL |
12982 | { |
12983 | name: "ceph.caps", | |
12984 | getxattr_cb: &Client::_vxattrcb_caps, | |
12985 | readonly: true, | |
12986 | exists_cb: NULL, | |
12987 | flags: 0, | |
12988 | }, | |
7c673cae FG |
12989 | { name: "" } /* Required table terminator */ |
12990 | }; | |
12991 | ||
adb31ebb TL |
12992 | const Client::VXattr Client::_common_vxattrs[] = { |
12993 | { | |
12994 | name: "ceph.cluster_fsid", | |
12995 | getxattr_cb: &Client::_vxattrcb_cluster_fsid, | |
12996 | readonly: true, | |
12997 | exists_cb: nullptr, | |
12998 | flags: 0, | |
12999 | }, | |
13000 | { | |
13001 | name: "ceph.client_id", | |
13002 | getxattr_cb: &Client::_vxattrcb_client_id, | |
13003 | readonly: true, | |
13004 | exists_cb: nullptr, | |
13005 | flags: 0, | |
13006 | }, | |
13007 | { name: "" } /* Required table terminator */ | |
13008 | }; | |
13009 | ||
7c673cae FG |
13010 | const Client::VXattr *Client::_get_vxattrs(Inode *in) |
13011 | { | |
13012 | if (in->is_dir()) | |
13013 | return _dir_vxattrs; | |
13014 | else if (in->is_file()) | |
13015 | return _file_vxattrs; | |
13016 | return NULL; | |
13017 | } | |
13018 | ||
13019 | const Client::VXattr *Client::_match_vxattr(Inode *in, const char *name) | |
13020 | { | |
13021 | if (strncmp(name, "ceph.", 5) == 0) { | |
13022 | const VXattr *vxattr = _get_vxattrs(in); | |
13023 | if (vxattr) { | |
13024 | while (!vxattr->name.empty()) { | |
13025 | if (vxattr->name == name) | |
13026 | return vxattr; | |
13027 | vxattr++; | |
13028 | } | |
13029 | } | |
adb31ebb TL |
13030 | |
13031 | // for common vxattrs | |
13032 | vxattr = _common_vxattrs; | |
13033 | while (!vxattr->name.empty()) { | |
13034 | if (vxattr->name == name) | |
13035 | return vxattr; | |
13036 | vxattr++; | |
13037 | } | |
7c673cae | 13038 | } |
adb31ebb | 13039 | |
7c673cae FG |
13040 | return NULL; |
13041 | } | |
13042 | ||
7c673cae FG |
13043 | int Client::ll_readlink(Inode *in, char *buf, size_t buflen, const UserPerm& perms) |
13044 | { | |
f67539c2 TL |
13045 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13046 | if (!mref_reader.is_state_satisfied()) | |
13047 | return -CEPHFS_ENOTCONN; | |
181888fb | 13048 | |
7c673cae FG |
13049 | vinodeno_t vino = _get_vino(in); |
13050 | ||
13051 | ldout(cct, 3) << "ll_readlink " << vino << dendl; | |
13052 | tout(cct) << "ll_readlink" << std::endl; | |
13053 | tout(cct) << vino.ino.val << std::endl; | |
13054 | ||
f67539c2 | 13055 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
13056 | for (auto dn : in->dentries) { |
13057 | touch_dn(dn); | |
7c673cae FG |
13058 | } |
13059 | ||
13060 | int r = _readlink(in, buf, buflen); // FIXME: no permission checking! | |
13061 | ldout(cct, 3) << "ll_readlink " << vino << " = " << r << dendl; | |
13062 | return r; | |
13063 | } | |
13064 | ||
13065 | int Client::_mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev, | |
13066 | const UserPerm& perms, InodeRef *inp) | |
13067 | { | |
1adf2230 | 13068 | ldout(cct, 8) << "_mknod(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
13069 | << mode << dec << ", " << rdev << ", uid " << perms.uid() |
13070 | << ", gid " << perms.gid() << ")" << dendl; | |
13071 | ||
13072 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13073 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13074 | |
13075 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13076 | return -CEPHFS_EROFS; |
7c673cae FG |
13077 | } |
13078 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13079 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13080 | } |
13081 | ||
13082 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_MKNOD); | |
13083 | ||
13084 | filepath path; | |
13085 | dir->make_nosnap_relative_path(path); | |
13086 | path.push_dentry(name); | |
13087 | req->set_filepath(path); | |
13088 | req->set_inode(dir); | |
13089 | req->head.args.mknod.rdev = rdev; | |
13090 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13091 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13092 | ||
13093 | bufferlist xattrs_bl; | |
13094 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
13095 | if (res < 0) | |
13096 | goto fail; | |
13097 | req->head.args.mknod.mode = mode; | |
13098 | if (xattrs_bl.length() > 0) | |
13099 | req->set_data(xattrs_bl); | |
13100 | ||
13101 | Dentry *de; | |
13102 | res = get_or_create(dir, name, &de); | |
13103 | if (res < 0) | |
13104 | goto fail; | |
13105 | req->set_dentry(de); | |
13106 | ||
13107 | res = make_request(req, perms, inp); | |
13108 | ||
13109 | trim_cache(); | |
13110 | ||
1adf2230 | 13111 | ldout(cct, 8) << "mknod(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13112 | return res; |
13113 | ||
13114 | fail: | |
13115 | put_request(req); | |
13116 | return res; | |
13117 | } | |
13118 | ||
13119 | int Client::ll_mknod(Inode *parent, const char *name, mode_t mode, | |
13120 | dev_t rdev, struct stat *attr, Inode **out, | |
13121 | const UserPerm& perms) | |
13122 | { | |
f67539c2 TL |
13123 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13124 | if (!mref_reader.is_state_satisfied()) | |
13125 | return -CEPHFS_ENOTCONN; | |
181888fb | 13126 | |
7c673cae FG |
13127 | vinodeno_t vparent = _get_vino(parent); |
13128 | ||
13129 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name << dendl; | |
13130 | tout(cct) << "ll_mknod" << std::endl; | |
13131 | tout(cct) << vparent.ino.val << std::endl; | |
13132 | tout(cct) << name << std::endl; | |
13133 | tout(cct) << mode << std::endl; | |
13134 | tout(cct) << rdev << std::endl; | |
13135 | ||
f67539c2 | 13136 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 13137 | if (!fuse_default_permissions) { |
7c673cae FG |
13138 | int r = may_create(parent, perms); |
13139 | if (r < 0) | |
13140 | return r; | |
13141 | } | |
13142 | ||
13143 | InodeRef in; | |
13144 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13145 | if (r == 0) { | |
13146 | fill_stat(in, attr); | |
13147 | _ll_get(in.get()); | |
13148 | } | |
13149 | tout(cct) << attr->st_ino << std::endl; | |
13150 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name | |
13151 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13152 | *out = in.get(); | |
13153 | return r; | |
13154 | } | |
13155 | ||
13156 | int Client::ll_mknodx(Inode *parent, const char *name, mode_t mode, | |
13157 | dev_t rdev, Inode **out, | |
13158 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13159 | const UserPerm& perms) | |
13160 | { | |
f67539c2 TL |
13161 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13162 | if (!mref_reader.is_state_satisfied()) | |
13163 | return -CEPHFS_ENOTCONN; | |
7c673cae | 13164 | |
f67539c2 | 13165 | unsigned caps = statx_to_mask(flags, want); |
181888fb | 13166 | |
7c673cae FG |
13167 | vinodeno_t vparent = _get_vino(parent); |
13168 | ||
13169 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name << dendl; | |
13170 | tout(cct) << "ll_mknodx" << std::endl; | |
13171 | tout(cct) << vparent.ino.val << std::endl; | |
13172 | tout(cct) << name << std::endl; | |
13173 | tout(cct) << mode << std::endl; | |
13174 | tout(cct) << rdev << std::endl; | |
13175 | ||
f67539c2 TL |
13176 | std::scoped_lock lock(client_lock); |
13177 | ||
11fdf7f2 | 13178 | if (!fuse_default_permissions) { |
7c673cae FG |
13179 | int r = may_create(parent, perms); |
13180 | if (r < 0) | |
13181 | return r; | |
13182 | } | |
13183 | ||
13184 | InodeRef in; | |
13185 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13186 | if (r == 0) { | |
13187 | fill_statx(in, caps, stx); | |
13188 | _ll_get(in.get()); | |
13189 | } | |
13190 | tout(cct) << stx->stx_ino << std::endl; | |
13191 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name | |
13192 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13193 | *out = in.get(); | |
13194 | return r; | |
13195 | } | |
13196 | ||
13197 | int Client::_create(Inode *dir, const char *name, int flags, mode_t mode, | |
13198 | InodeRef *inp, Fh **fhp, int stripe_unit, int stripe_count, | |
13199 | int object_size, const char *data_pool, bool *created, | |
f67539c2 | 13200 | const UserPerm& perms, std::string alternate_name) |
7c673cae | 13201 | { |
1adf2230 | 13202 | ldout(cct, 8) << "_create(" << dir->ino << " " << name << ", 0" << oct << |
7c673cae FG |
13203 | mode << dec << ")" << dendl; |
13204 | ||
13205 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13206 | return -CEPHFS_ENAMETOOLONG; |
7c673cae | 13207 | if (dir->snapid != CEPH_NOSNAP) { |
f67539c2 | 13208 | return -CEPHFS_EROFS; |
7c673cae FG |
13209 | } |
13210 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13211 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13212 | } |
13213 | ||
13214 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
13215 | int cflags = ceph_flags_sys2wire(flags); |
13216 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
13217 | cflags |= CEPH_O_LAZY; | |
13218 | ||
13219 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
13220 | |
13221 | int64_t pool_id = -1; | |
13222 | if (data_pool && *data_pool) { | |
13223 | pool_id = objecter->with_osdmap( | |
13224 | std::mem_fn(&OSDMap::lookup_pg_pool_name), data_pool); | |
13225 | if (pool_id < 0) | |
f67539c2 | 13226 | return -CEPHFS_EINVAL; |
7c673cae | 13227 | if (pool_id > 0xffffffffll) |
f67539c2 | 13228 | return -CEPHFS_ERANGE; // bummer! |
7c673cae FG |
13229 | } |
13230 | ||
13231 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_CREATE); | |
13232 | ||
13233 | filepath path; | |
13234 | dir->make_nosnap_relative_path(path); | |
13235 | path.push_dentry(name); | |
13236 | req->set_filepath(path); | |
f67539c2 | 13237 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae | 13238 | req->set_inode(dir); |
11fdf7f2 | 13239 | req->head.args.open.flags = cflags | CEPH_O_CREAT; |
7c673cae FG |
13240 | |
13241 | req->head.args.open.stripe_unit = stripe_unit; | |
13242 | req->head.args.open.stripe_count = stripe_count; | |
13243 | req->head.args.open.object_size = object_size; | |
13244 | if (cct->_conf->client_debug_getattr_caps) | |
13245 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
13246 | else | |
13247 | req->head.args.open.mask = 0; | |
13248 | req->head.args.open.pool = pool_id; | |
13249 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13250 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13251 | ||
13252 | mode |= S_IFREG; | |
13253 | bufferlist xattrs_bl; | |
13254 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
13255 | if (res < 0) | |
13256 | goto fail; | |
13257 | req->head.args.open.mode = mode; | |
13258 | if (xattrs_bl.length() > 0) | |
13259 | req->set_data(xattrs_bl); | |
13260 | ||
13261 | Dentry *de; | |
13262 | res = get_or_create(dir, name, &de); | |
13263 | if (res < 0) | |
13264 | goto fail; | |
13265 | req->set_dentry(de); | |
13266 | ||
13267 | res = make_request(req, perms, inp, created); | |
13268 | if (res < 0) { | |
13269 | goto reply_error; | |
13270 | } | |
13271 | ||
13272 | /* If the caller passed a value in fhp, do the open */ | |
13273 | if(fhp) { | |
13274 | (*inp)->get_open_ref(cmode); | |
13275 | *fhp = _create_fh(inp->get(), flags, cmode, perms); | |
13276 | } | |
13277 | ||
13278 | reply_error: | |
13279 | trim_cache(); | |
13280 | ||
1adf2230 | 13281 | ldout(cct, 8) << "create(" << path << ", 0" << oct << mode << dec |
7c673cae FG |
13282 | << " layout " << stripe_unit |
13283 | << ' ' << stripe_count | |
13284 | << ' ' << object_size | |
13285 | <<") = " << res << dendl; | |
13286 | return res; | |
13287 | ||
13288 | fail: | |
13289 | put_request(req); | |
13290 | return res; | |
13291 | } | |
13292 | ||
7c673cae | 13293 | int Client::_mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm, |
f67539c2 TL |
13294 | InodeRef *inp, const std::map<std::string, std::string> &metadata, |
13295 | std::string alternate_name) | |
7c673cae | 13296 | { |
1adf2230 | 13297 | ldout(cct, 8) << "_mkdir(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
13298 | << mode << dec << ", uid " << perm.uid() |
13299 | << ", gid " << perm.gid() << ")" << dendl; | |
13300 | ||
13301 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13302 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13303 | |
13304 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13305 | return -CEPHFS_EROFS; |
7c673cae FG |
13306 | } |
13307 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 13308 | return -CEPHFS_EDQUOT; |
7c673cae | 13309 | } |
f67539c2 TL |
13310 | |
13311 | bool is_snap_op = dir->snapid == CEPH_SNAPDIR; | |
13312 | MetaRequest *req = new MetaRequest(is_snap_op ? | |
7c673cae FG |
13313 | CEPH_MDS_OP_MKSNAP : CEPH_MDS_OP_MKDIR); |
13314 | ||
13315 | filepath path; | |
13316 | dir->make_nosnap_relative_path(path); | |
13317 | path.push_dentry(name); | |
13318 | req->set_filepath(path); | |
13319 | req->set_inode(dir); | |
13320 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13321 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
f67539c2 | 13322 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13323 | |
13324 | mode |= S_IFDIR; | |
f67539c2 TL |
13325 | bufferlist bl; |
13326 | int res = _posix_acl_create(dir, &mode, bl, perm); | |
7c673cae FG |
13327 | if (res < 0) |
13328 | goto fail; | |
13329 | req->head.args.mkdir.mode = mode; | |
f67539c2 TL |
13330 | if (is_snap_op) { |
13331 | SnapPayload payload; | |
13332 | // clear the bufferlist that may have been populated by the call | |
13333 | // to _posix_acl_create(). MDS mksnap does not make use of it. | |
13334 | // So, reuse it to pass metadata payload. | |
13335 | bl.clear(); | |
13336 | payload.metadata = metadata; | |
13337 | encode(payload, bl); | |
13338 | } | |
13339 | if (bl.length() > 0) { | |
13340 | req->set_data(bl); | |
13341 | } | |
7c673cae FG |
13342 | |
13343 | Dentry *de; | |
13344 | res = get_or_create(dir, name, &de); | |
13345 | if (res < 0) | |
13346 | goto fail; | |
13347 | req->set_dentry(de); | |
13348 | ||
13349 | ldout(cct, 10) << "_mkdir: making request" << dendl; | |
13350 | res = make_request(req, perm, inp); | |
13351 | ldout(cct, 10) << "_mkdir result is " << res << dendl; | |
13352 | ||
13353 | trim_cache(); | |
13354 | ||
1adf2230 | 13355 | ldout(cct, 8) << "_mkdir(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13356 | return res; |
13357 | ||
13358 | fail: | |
13359 | put_request(req); | |
13360 | return res; | |
13361 | } | |
13362 | ||
13363 | int Client::ll_mkdir(Inode *parent, const char *name, mode_t mode, | |
13364 | struct stat *attr, Inode **out, const UserPerm& perm) | |
13365 | { | |
f67539c2 TL |
13366 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13367 | if (!mref_reader.is_state_satisfied()) | |
13368 | return -CEPHFS_ENOTCONN; | |
181888fb | 13369 | |
7c673cae FG |
13370 | vinodeno_t vparent = _get_vino(parent); |
13371 | ||
13372 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name << dendl; | |
13373 | tout(cct) << "ll_mkdir" << std::endl; | |
13374 | tout(cct) << vparent.ino.val << std::endl; | |
13375 | tout(cct) << name << std::endl; | |
13376 | tout(cct) << mode << std::endl; | |
13377 | ||
f67539c2 TL |
13378 | std::scoped_lock lock(client_lock); |
13379 | ||
11fdf7f2 | 13380 | if (!fuse_default_permissions) { |
7c673cae FG |
13381 | int r = may_create(parent, perm); |
13382 | if (r < 0) | |
13383 | return r; | |
13384 | } | |
13385 | ||
13386 | InodeRef in; | |
13387 | int r = _mkdir(parent, name, mode, perm, &in); | |
13388 | if (r == 0) { | |
13389 | fill_stat(in, attr); | |
13390 | _ll_get(in.get()); | |
13391 | } | |
13392 | tout(cct) << attr->st_ino << std::endl; | |
13393 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name | |
13394 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13395 | *out = in.get(); | |
13396 | return r; | |
13397 | } | |
13398 | ||
13399 | int Client::ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out, | |
13400 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13401 | const UserPerm& perms) | |
13402 | { | |
f67539c2 TL |
13403 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13404 | if (!mref_reader.is_state_satisfied()) | |
13405 | return -CEPHFS_ENOTCONN; | |
181888fb | 13406 | |
7c673cae FG |
13407 | vinodeno_t vparent = _get_vino(parent); |
13408 | ||
13409 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name << dendl; | |
13410 | tout(cct) << "ll_mkdirx" << std::endl; | |
13411 | tout(cct) << vparent.ino.val << std::endl; | |
13412 | tout(cct) << name << std::endl; | |
13413 | tout(cct) << mode << std::endl; | |
13414 | ||
f67539c2 TL |
13415 | std::scoped_lock lock(client_lock); |
13416 | ||
11fdf7f2 | 13417 | if (!fuse_default_permissions) { |
7c673cae FG |
13418 | int r = may_create(parent, perms); |
13419 | if (r < 0) | |
13420 | return r; | |
13421 | } | |
13422 | ||
13423 | InodeRef in; | |
13424 | int r = _mkdir(parent, name, mode, perms, &in); | |
13425 | if (r == 0) { | |
13426 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13427 | _ll_get(in.get()); | |
13428 | } else { | |
13429 | stx->stx_ino = 0; | |
13430 | stx->stx_mask = 0; | |
13431 | } | |
13432 | tout(cct) << stx->stx_ino << std::endl; | |
13433 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name | |
13434 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13435 | *out = in.get(); | |
13436 | return r; | |
13437 | } | |
13438 | ||
13439 | int Client::_symlink(Inode *dir, const char *name, const char *target, | |
f67539c2 | 13440 | const UserPerm& perms, std::string alternate_name, InodeRef *inp) |
7c673cae | 13441 | { |
1adf2230 | 13442 | ldout(cct, 8) << "_symlink(" << dir->ino << " " << name << ", " << target |
7c673cae FG |
13443 | << ", uid " << perms.uid() << ", gid " << perms.gid() << ")" |
13444 | << dendl; | |
13445 | ||
13446 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13447 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13448 | |
13449 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13450 | return -CEPHFS_EROFS; |
7c673cae FG |
13451 | } |
13452 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13453 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13454 | } |
13455 | ||
13456 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SYMLINK); | |
13457 | ||
13458 | filepath path; | |
13459 | dir->make_nosnap_relative_path(path); | |
13460 | path.push_dentry(name); | |
13461 | req->set_filepath(path); | |
f67539c2 | 13462 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13463 | req->set_inode(dir); |
13464 | req->set_string2(target); | |
13465 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13466 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13467 | ||
13468 | Dentry *de; | |
13469 | int res = get_or_create(dir, name, &de); | |
13470 | if (res < 0) | |
13471 | goto fail; | |
13472 | req->set_dentry(de); | |
13473 | ||
13474 | res = make_request(req, perms, inp); | |
13475 | ||
13476 | trim_cache(); | |
1adf2230 | 13477 | ldout(cct, 8) << "_symlink(\"" << path << "\", \"" << target << "\") = " << |
7c673cae FG |
13478 | res << dendl; |
13479 | return res; | |
13480 | ||
13481 | fail: | |
13482 | put_request(req); | |
13483 | return res; | |
13484 | } | |
13485 | ||
13486 | int Client::ll_symlink(Inode *parent, const char *name, const char *value, | |
13487 | struct stat *attr, Inode **out, const UserPerm& perms) | |
13488 | { | |
f67539c2 TL |
13489 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13490 | if (!mref_reader.is_state_satisfied()) | |
13491 | return -CEPHFS_ENOTCONN; | |
181888fb | 13492 | |
7c673cae FG |
13493 | vinodeno_t vparent = _get_vino(parent); |
13494 | ||
13495 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name << " -> " << value | |
13496 | << dendl; | |
13497 | tout(cct) << "ll_symlink" << std::endl; | |
13498 | tout(cct) << vparent.ino.val << std::endl; | |
13499 | tout(cct) << name << std::endl; | |
13500 | tout(cct) << value << std::endl; | |
13501 | ||
f67539c2 TL |
13502 | std::scoped_lock lock(client_lock); |
13503 | ||
11fdf7f2 | 13504 | if (!fuse_default_permissions) { |
7c673cae FG |
13505 | int r = may_create(parent, perms); |
13506 | if (r < 0) | |
13507 | return r; | |
13508 | } | |
13509 | ||
13510 | InodeRef in; | |
f67539c2 | 13511 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13512 | if (r == 0) { |
13513 | fill_stat(in, attr); | |
13514 | _ll_get(in.get()); | |
13515 | } | |
13516 | tout(cct) << attr->st_ino << std::endl; | |
13517 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name | |
13518 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13519 | *out = in.get(); | |
13520 | return r; | |
13521 | } | |
13522 | ||
13523 | int Client::ll_symlinkx(Inode *parent, const char *name, const char *value, | |
13524 | Inode **out, struct ceph_statx *stx, unsigned want, | |
13525 | unsigned flags, const UserPerm& perms) | |
13526 | { | |
f67539c2 TL |
13527 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13528 | if (!mref_reader.is_state_satisfied()) | |
13529 | return -CEPHFS_ENOTCONN; | |
181888fb | 13530 | |
7c673cae FG |
13531 | vinodeno_t vparent = _get_vino(parent); |
13532 | ||
13533 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name << " -> " << value | |
13534 | << dendl; | |
13535 | tout(cct) << "ll_symlinkx" << std::endl; | |
13536 | tout(cct) << vparent.ino.val << std::endl; | |
13537 | tout(cct) << name << std::endl; | |
13538 | tout(cct) << value << std::endl; | |
13539 | ||
f67539c2 TL |
13540 | std::scoped_lock lock(client_lock); |
13541 | ||
11fdf7f2 | 13542 | if (!fuse_default_permissions) { |
7c673cae FG |
13543 | int r = may_create(parent, perms); |
13544 | if (r < 0) | |
13545 | return r; | |
13546 | } | |
13547 | ||
13548 | InodeRef in; | |
f67539c2 | 13549 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13550 | if (r == 0) { |
13551 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13552 | _ll_get(in.get()); | |
13553 | } | |
13554 | tout(cct) << stx->stx_ino << std::endl; | |
13555 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name | |
13556 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13557 | *out = in.get(); | |
13558 | return r; | |
13559 | } | |
13560 | ||
13561 | int Client::_unlink(Inode *dir, const char *name, const UserPerm& perm) | |
13562 | { | |
1adf2230 | 13563 | ldout(cct, 8) << "_unlink(" << dir->ino << " " << name |
7c673cae FG |
13564 | << " uid " << perm.uid() << " gid " << perm.gid() |
13565 | << ")" << dendl; | |
13566 | ||
13567 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13568 | return -CEPHFS_EROFS; |
7c673cae FG |
13569 | } |
13570 | ||
13571 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_UNLINK); | |
13572 | ||
13573 | filepath path; | |
13574 | dir->make_nosnap_relative_path(path); | |
13575 | path.push_dentry(name); | |
13576 | req->set_filepath(path); | |
13577 | ||
13578 | InodeRef otherin; | |
b32b8144 | 13579 | Inode *in; |
7c673cae | 13580 | Dentry *de; |
b32b8144 | 13581 | |
7c673cae FG |
13582 | int res = get_or_create(dir, name, &de); |
13583 | if (res < 0) | |
13584 | goto fail; | |
13585 | req->set_dentry(de); | |
13586 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13587 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13588 | ||
13589 | res = _lookup(dir, name, 0, &otherin, perm); | |
13590 | if (res < 0) | |
13591 | goto fail; | |
b32b8144 FG |
13592 | |
13593 | in = otherin.get(); | |
13594 | req->set_other_inode(in); | |
13595 | in->break_all_delegs(); | |
7c673cae FG |
13596 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
13597 | ||
13598 | req->set_inode(dir); | |
13599 | ||
13600 | res = make_request(req, perm); | |
13601 | ||
13602 | trim_cache(); | |
1adf2230 | 13603 | ldout(cct, 8) << "unlink(" << path << ") = " << res << dendl; |
7c673cae FG |
13604 | return res; |
13605 | ||
13606 | fail: | |
13607 | put_request(req); | |
13608 | return res; | |
13609 | } | |
13610 | ||
13611 | int Client::ll_unlink(Inode *in, const char *name, const UserPerm& perm) | |
13612 | { | |
f67539c2 TL |
13613 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13614 | if (!mref_reader.is_state_satisfied()) | |
13615 | return -CEPHFS_ENOTCONN; | |
181888fb | 13616 | |
7c673cae FG |
13617 | vinodeno_t vino = _get_vino(in); |
13618 | ||
13619 | ldout(cct, 3) << "ll_unlink " << vino << " " << name << dendl; | |
13620 | tout(cct) << "ll_unlink" << std::endl; | |
13621 | tout(cct) << vino.ino.val << std::endl; | |
13622 | tout(cct) << name << std::endl; | |
13623 | ||
f67539c2 TL |
13624 | std::scoped_lock lock(client_lock); |
13625 | ||
11fdf7f2 | 13626 | if (!fuse_default_permissions) { |
7c673cae FG |
13627 | int r = may_delete(in, name, perm); |
13628 | if (r < 0) | |
13629 | return r; | |
13630 | } | |
13631 | return _unlink(in, name, perm); | |
13632 | } | |
13633 | ||
13634 | int Client::_rmdir(Inode *dir, const char *name, const UserPerm& perms) | |
13635 | { | |
1adf2230 | 13636 | ldout(cct, 8) << "_rmdir(" << dir->ino << " " << name << " uid " |
7c673cae FG |
13637 | << perms.uid() << " gid " << perms.gid() << ")" << dendl; |
13638 | ||
13639 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13640 | return -CEPHFS_EROFS; |
7c673cae | 13641 | } |
b32b8144 FG |
13642 | |
13643 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_RMSNAP : CEPH_MDS_OP_RMDIR; | |
13644 | MetaRequest *req = new MetaRequest(op); | |
7c673cae FG |
13645 | filepath path; |
13646 | dir->make_nosnap_relative_path(path); | |
13647 | path.push_dentry(name); | |
13648 | req->set_filepath(path); | |
11fdf7f2 | 13649 | req->set_inode(dir); |
7c673cae FG |
13650 | |
13651 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13652 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13653 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
13654 | ||
13655 | InodeRef in; | |
13656 | ||
13657 | Dentry *de; | |
13658 | int res = get_or_create(dir, name, &de); | |
13659 | if (res < 0) | |
13660 | goto fail; | |
b32b8144 FG |
13661 | if (op == CEPH_MDS_OP_RMDIR) |
13662 | req->set_dentry(de); | |
13663 | else | |
13664 | de->get(); | |
13665 | ||
7c673cae FG |
13666 | res = _lookup(dir, name, 0, &in, perms); |
13667 | if (res < 0) | |
13668 | goto fail; | |
11fdf7f2 TL |
13669 | |
13670 | if (op == CEPH_MDS_OP_RMSNAP) { | |
7c673cae | 13671 | unlink(de, true, true); |
b32b8144 | 13672 | de->put(); |
7c673cae | 13673 | } |
11fdf7f2 | 13674 | req->set_other_inode(in.get()); |
7c673cae FG |
13675 | |
13676 | res = make_request(req, perms); | |
13677 | ||
13678 | trim_cache(); | |
1adf2230 | 13679 | ldout(cct, 8) << "rmdir(" << path << ") = " << res << dendl; |
7c673cae FG |
13680 | return res; |
13681 | ||
13682 | fail: | |
13683 | put_request(req); | |
13684 | return res; | |
13685 | } | |
13686 | ||
13687 | int Client::ll_rmdir(Inode *in, const char *name, const UserPerm& perms) | |
13688 | { | |
f67539c2 TL |
13689 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13690 | if (!mref_reader.is_state_satisfied()) | |
13691 | return -CEPHFS_ENOTCONN; | |
181888fb | 13692 | |
7c673cae FG |
13693 | vinodeno_t vino = _get_vino(in); |
13694 | ||
13695 | ldout(cct, 3) << "ll_rmdir " << vino << " " << name << dendl; | |
13696 | tout(cct) << "ll_rmdir" << std::endl; | |
13697 | tout(cct) << vino.ino.val << std::endl; | |
13698 | tout(cct) << name << std::endl; | |
13699 | ||
f67539c2 TL |
13700 | std::scoped_lock lock(client_lock); |
13701 | ||
11fdf7f2 | 13702 | if (!fuse_default_permissions) { |
7c673cae FG |
13703 | int r = may_delete(in, name, perms); |
13704 | if (r < 0) | |
13705 | return r; | |
13706 | } | |
13707 | ||
13708 | return _rmdir(in, name, perms); | |
13709 | } | |
13710 | ||
f67539c2 | 13711 | int Client::_rename(Inode *fromdir, const char *fromname, Inode *todir, const char *toname, const UserPerm& perm, std::string alternate_name) |
7c673cae | 13712 | { |
1adf2230 | 13713 | ldout(cct, 8) << "_rename(" << fromdir->ino << " " << fromname << " to " |
7c673cae FG |
13714 | << todir->ino << " " << toname |
13715 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" | |
13716 | << dendl; | |
13717 | ||
13718 | if (fromdir->snapid != todir->snapid) | |
f67539c2 | 13719 | return -CEPHFS_EXDEV; |
7c673cae FG |
13720 | |
13721 | int op = CEPH_MDS_OP_RENAME; | |
13722 | if (fromdir->snapid != CEPH_NOSNAP) { | |
13723 | if (fromdir == todir && fromdir->snapid == CEPH_SNAPDIR) | |
13724 | op = CEPH_MDS_OP_RENAMESNAP; | |
13725 | else | |
f67539c2 TL |
13726 | return -CEPHFS_EROFS; |
13727 | } | |
13728 | if (fromdir != todir) { | |
13729 | Inode *fromdir_root = | |
13730 | fromdir->quota.is_enable() ? fromdir : get_quota_root(fromdir, perm); | |
13731 | Inode *todir_root = | |
13732 | todir->quota.is_enable() ? todir : get_quota_root(todir, perm); | |
13733 | if (fromdir_root != todir_root) { | |
13734 | return -CEPHFS_EXDEV; | |
13735 | } | |
7c673cae | 13736 | } |
7c673cae FG |
13737 | |
13738 | InodeRef target; | |
13739 | MetaRequest *req = new MetaRequest(op); | |
13740 | ||
13741 | filepath from; | |
13742 | fromdir->make_nosnap_relative_path(from); | |
13743 | from.push_dentry(fromname); | |
13744 | filepath to; | |
13745 | todir->make_nosnap_relative_path(to); | |
13746 | to.push_dentry(toname); | |
13747 | req->set_filepath(to); | |
13748 | req->set_filepath2(from); | |
f67539c2 | 13749 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13750 | |
13751 | Dentry *oldde; | |
13752 | int res = get_or_create(fromdir, fromname, &oldde); | |
13753 | if (res < 0) | |
13754 | goto fail; | |
13755 | Dentry *de; | |
13756 | res = get_or_create(todir, toname, &de); | |
13757 | if (res < 0) | |
13758 | goto fail; | |
13759 | ||
13760 | if (op == CEPH_MDS_OP_RENAME) { | |
13761 | req->set_old_dentry(oldde); | |
13762 | req->old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
13763 | req->old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
13764 | ||
13765 | req->set_dentry(de); | |
13766 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13767 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13768 | ||
13769 | InodeRef oldin, otherin; | |
f67539c2 | 13770 | res = _lookup(fromdir, fromname, 0, &oldin, perm); |
7c673cae FG |
13771 | if (res < 0) |
13772 | goto fail; | |
b32b8144 FG |
13773 | |
13774 | Inode *oldinode = oldin.get(); | |
13775 | oldinode->break_all_delegs(); | |
13776 | req->set_old_inode(oldinode); | |
7c673cae FG |
13777 | req->old_inode_drop = CEPH_CAP_LINK_SHARED; |
13778 | ||
13779 | res = _lookup(todir, toname, 0, &otherin, perm); | |
b32b8144 FG |
13780 | switch (res) { |
13781 | case 0: | |
13782 | { | |
13783 | Inode *in = otherin.get(); | |
13784 | req->set_other_inode(in); | |
13785 | in->break_all_delegs(); | |
13786 | } | |
7c673cae | 13787 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
b32b8144 | 13788 | break; |
f67539c2 | 13789 | case -CEPHFS_ENOENT: |
b32b8144 FG |
13790 | break; |
13791 | default: | |
13792 | goto fail; | |
7c673cae FG |
13793 | } |
13794 | ||
13795 | req->set_inode(todir); | |
13796 | } else { | |
13797 | // renamesnap reply contains no tracedn, so we need to invalidate | |
13798 | // dentry manually | |
13799 | unlink(oldde, true, true); | |
13800 | unlink(de, true, true); | |
11fdf7f2 TL |
13801 | |
13802 | req->set_inode(todir); | |
7c673cae FG |
13803 | } |
13804 | ||
13805 | res = make_request(req, perm, &target); | |
13806 | ldout(cct, 10) << "rename result is " << res << dendl; | |
13807 | ||
13808 | // renamed item from our cache | |
13809 | ||
13810 | trim_cache(); | |
1adf2230 | 13811 | ldout(cct, 8) << "_rename(" << from << ", " << to << ") = " << res << dendl; |
7c673cae FG |
13812 | return res; |
13813 | ||
13814 | fail: | |
13815 | put_request(req); | |
13816 | return res; | |
13817 | } | |
13818 | ||
13819 | int Client::ll_rename(Inode *parent, const char *name, Inode *newparent, | |
13820 | const char *newname, const UserPerm& perm) | |
13821 | { | |
f67539c2 TL |
13822 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13823 | if (!mref_reader.is_state_satisfied()) | |
13824 | return -CEPHFS_ENOTCONN; | |
181888fb | 13825 | |
7c673cae FG |
13826 | vinodeno_t vparent = _get_vino(parent); |
13827 | vinodeno_t vnewparent = _get_vino(newparent); | |
13828 | ||
13829 | ldout(cct, 3) << "ll_rename " << vparent << " " << name << " to " | |
13830 | << vnewparent << " " << newname << dendl; | |
13831 | tout(cct) << "ll_rename" << std::endl; | |
13832 | tout(cct) << vparent.ino.val << std::endl; | |
13833 | tout(cct) << name << std::endl; | |
13834 | tout(cct) << vnewparent.ino.val << std::endl; | |
13835 | tout(cct) << newname << std::endl; | |
13836 | ||
f67539c2 TL |
13837 | std::scoped_lock lock(client_lock); |
13838 | ||
11fdf7f2 | 13839 | if (!fuse_default_permissions) { |
7c673cae FG |
13840 | int r = may_delete(parent, name, perm); |
13841 | if (r < 0) | |
13842 | return r; | |
13843 | r = may_delete(newparent, newname, perm); | |
f67539c2 | 13844 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
13845 | return r; |
13846 | } | |
13847 | ||
f67539c2 | 13848 | return _rename(parent, name, newparent, newname, perm, ""); |
7c673cae FG |
13849 | } |
13850 | ||
f67539c2 | 13851 | int Client::_link(Inode *in, Inode *dir, const char *newname, const UserPerm& perm, std::string alternate_name, InodeRef *inp) |
7c673cae | 13852 | { |
1adf2230 | 13853 | ldout(cct, 8) << "_link(" << in->ino << " to " << dir->ino << " " << newname |
7c673cae FG |
13854 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" << dendl; |
13855 | ||
13856 | if (strlen(newname) > NAME_MAX) | |
f67539c2 | 13857 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13858 | |
13859 | if (in->snapid != CEPH_NOSNAP || dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13860 | return -CEPHFS_EROFS; |
7c673cae FG |
13861 | } |
13862 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 13863 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13864 | } |
13865 | ||
b32b8144 | 13866 | in->break_all_delegs(); |
7c673cae FG |
13867 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LINK); |
13868 | ||
13869 | filepath path(newname, dir->ino); | |
13870 | req->set_filepath(path); | |
f67539c2 | 13871 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13872 | filepath existing(in->ino); |
13873 | req->set_filepath2(existing); | |
13874 | ||
13875 | req->set_inode(dir); | |
13876 | req->inode_drop = CEPH_CAP_FILE_SHARED; | |
13877 | req->inode_unless = CEPH_CAP_FILE_EXCL; | |
13878 | ||
13879 | Dentry *de; | |
13880 | int res = get_or_create(dir, newname, &de); | |
13881 | if (res < 0) | |
13882 | goto fail; | |
13883 | req->set_dentry(de); | |
13884 | ||
13885 | res = make_request(req, perm, inp); | |
13886 | ldout(cct, 10) << "link result is " << res << dendl; | |
13887 | ||
13888 | trim_cache(); | |
1adf2230 | 13889 | ldout(cct, 8) << "link(" << existing << ", " << path << ") = " << res << dendl; |
7c673cae FG |
13890 | return res; |
13891 | ||
13892 | fail: | |
13893 | put_request(req); | |
13894 | return res; | |
13895 | } | |
13896 | ||
13897 | int Client::ll_link(Inode *in, Inode *newparent, const char *newname, | |
13898 | const UserPerm& perm) | |
13899 | { | |
f67539c2 TL |
13900 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13901 | if (!mref_reader.is_state_satisfied()) | |
13902 | return -CEPHFS_ENOTCONN; | |
181888fb | 13903 | |
7c673cae FG |
13904 | vinodeno_t vino = _get_vino(in); |
13905 | vinodeno_t vnewparent = _get_vino(newparent); | |
13906 | ||
31f18b77 | 13907 | ldout(cct, 3) << "ll_link " << vino << " to " << vnewparent << " " << |
7c673cae FG |
13908 | newname << dendl; |
13909 | tout(cct) << "ll_link" << std::endl; | |
13910 | tout(cct) << vino.ino.val << std::endl; | |
13911 | tout(cct) << vnewparent << std::endl; | |
13912 | tout(cct) << newname << std::endl; | |
13913 | ||
7c673cae FG |
13914 | InodeRef target; |
13915 | ||
f67539c2 TL |
13916 | std::scoped_lock lock(client_lock); |
13917 | ||
11fdf7f2 | 13918 | if (!fuse_default_permissions) { |
7c673cae | 13919 | if (S_ISDIR(in->mode)) |
f67539c2 | 13920 | return -CEPHFS_EPERM; |
7c673cae | 13921 | |
11fdf7f2 | 13922 | int r = may_hardlink(in, perm); |
7c673cae FG |
13923 | if (r < 0) |
13924 | return r; | |
13925 | ||
13926 | r = may_create(newparent, perm); | |
13927 | if (r < 0) | |
13928 | return r; | |
13929 | } | |
13930 | ||
f67539c2 | 13931 | return _link(in, newparent, newname, perm, "", &target); |
7c673cae FG |
13932 | } |
13933 | ||
13934 | int Client::ll_num_osds(void) | |
13935 | { | |
f67539c2 | 13936 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13937 | return objecter->with_osdmap(std::mem_fn(&OSDMap::get_num_osds)); |
13938 | } | |
13939 | ||
13940 | int Client::ll_osdaddr(int osd, uint32_t *addr) | |
13941 | { | |
f67539c2 | 13942 | std::scoped_lock lock(client_lock); |
181888fb | 13943 | |
7c673cae FG |
13944 | entity_addr_t g; |
13945 | bool exists = objecter->with_osdmap([&](const OSDMap& o) { | |
13946 | if (!o.exists(osd)) | |
13947 | return false; | |
11fdf7f2 | 13948 | g = o.get_addrs(osd).front(); |
7c673cae FG |
13949 | return true; |
13950 | }); | |
13951 | if (!exists) | |
13952 | return -1; | |
13953 | uint32_t nb_addr = (g.in4_addr()).sin_addr.s_addr; | |
13954 | *addr = ntohl(nb_addr); | |
13955 | return 0; | |
13956 | } | |
181888fb | 13957 | |
7c673cae FG |
13958 | uint32_t Client::ll_stripe_unit(Inode *in) |
13959 | { | |
f67539c2 | 13960 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13961 | return in->layout.stripe_unit; |
13962 | } | |
13963 | ||
13964 | uint64_t Client::ll_snap_seq(Inode *in) | |
13965 | { | |
f67539c2 | 13966 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13967 | return in->snaprealm->seq; |
13968 | } | |
13969 | ||
13970 | int Client::ll_file_layout(Inode *in, file_layout_t *layout) | |
13971 | { | |
f67539c2 | 13972 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13973 | *layout = in->layout; |
13974 | return 0; | |
13975 | } | |
13976 | ||
13977 | int Client::ll_file_layout(Fh *fh, file_layout_t *layout) | |
13978 | { | |
13979 | return ll_file_layout(fh->inode.get(), layout); | |
13980 | } | |
13981 | ||
13982 | /* Currently we cannot take advantage of redundancy in reads, since we | |
13983 | would have to go through all possible placement groups (a | |
13984 | potentially quite large number determined by a hash), and use CRUSH | |
13985 | to calculate the appropriate set of OSDs for each placement group, | |
13986 | then index into that. An array with one entry per OSD is much more | |
13987 | tractable and works for demonstration purposes. */ | |
13988 | ||
13989 | int Client::ll_get_stripe_osd(Inode *in, uint64_t blockno, | |
13990 | file_layout_t* layout) | |
13991 | { | |
f67539c2 | 13992 | std::scoped_lock lock(client_lock); |
181888fb | 13993 | |
28e407b8 | 13994 | inodeno_t ino = in->ino; |
7c673cae FG |
13995 | uint32_t object_size = layout->object_size; |
13996 | uint32_t su = layout->stripe_unit; | |
13997 | uint32_t stripe_count = layout->stripe_count; | |
13998 | uint64_t stripes_per_object = object_size / su; | |
11fdf7f2 | 13999 | uint64_t stripeno = 0, stripepos = 0; |
7c673cae | 14000 | |
11fdf7f2 TL |
14001 | if(stripe_count) { |
14002 | stripeno = blockno / stripe_count; // which horizontal stripe (Y) | |
14003 | stripepos = blockno % stripe_count; // which object in the object set (X) | |
14004 | } | |
7c673cae FG |
14005 | uint64_t objectsetno = stripeno / stripes_per_object; // which object set |
14006 | uint64_t objectno = objectsetno * stripe_count + stripepos; // object id | |
14007 | ||
14008 | object_t oid = file_object_t(ino, objectno); | |
14009 | return objecter->with_osdmap([&](const OSDMap& o) { | |
14010 | ceph_object_layout olayout = | |
14011 | o.file_to_object_layout(oid, *layout); | |
14012 | pg_t pg = (pg_t)olayout.ol_pgid; | |
14013 | vector<int> osds; | |
14014 | int primary; | |
14015 | o.pg_to_acting_osds(pg, &osds, &primary); | |
14016 | return primary; | |
14017 | }); | |
14018 | } | |
14019 | ||
14020 | /* Return the offset of the block, internal to the object */ | |
14021 | ||
14022 | uint64_t Client::ll_get_internal_offset(Inode *in, uint64_t blockno) | |
14023 | { | |
f67539c2 | 14024 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14025 | file_layout_t *layout=&(in->layout); |
14026 | uint32_t object_size = layout->object_size; | |
14027 | uint32_t su = layout->stripe_unit; | |
14028 | uint64_t stripes_per_object = object_size / su; | |
14029 | ||
14030 | return (blockno % stripes_per_object) * su; | |
14031 | } | |
14032 | ||
14033 | int Client::ll_opendir(Inode *in, int flags, dir_result_t** dirpp, | |
14034 | const UserPerm& perms) | |
14035 | { | |
f67539c2 TL |
14036 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14037 | if (!mref_reader.is_state_satisfied()) | |
14038 | return -CEPHFS_ENOTCONN; | |
181888fb | 14039 | |
7c673cae FG |
14040 | vinodeno_t vino = _get_vino(in); |
14041 | ||
14042 | ldout(cct, 3) << "ll_opendir " << vino << dendl; | |
14043 | tout(cct) << "ll_opendir" << std::endl; | |
14044 | tout(cct) << vino.ino.val << std::endl; | |
14045 | ||
f67539c2 TL |
14046 | std::scoped_lock lock(client_lock); |
14047 | ||
11fdf7f2 | 14048 | if (!fuse_default_permissions) { |
7c673cae FG |
14049 | int r = may_open(in, flags, perms); |
14050 | if (r < 0) | |
14051 | return r; | |
14052 | } | |
14053 | ||
14054 | int r = _opendir(in, dirpp, perms); | |
f67539c2 | 14055 | tout(cct) << (uintptr_t)*dirpp << std::endl; |
7c673cae FG |
14056 | |
14057 | ldout(cct, 3) << "ll_opendir " << vino << " = " << r << " (" << *dirpp << ")" | |
14058 | << dendl; | |
14059 | return r; | |
14060 | } | |
14061 | ||
14062 | int Client::ll_releasedir(dir_result_t *dirp) | |
14063 | { | |
f67539c2 TL |
14064 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14065 | if (!mref_reader.is_state_satisfied()) | |
14066 | return -CEPHFS_ENOTCONN; | |
14067 | ||
7c673cae FG |
14068 | ldout(cct, 3) << "ll_releasedir " << dirp << dendl; |
14069 | tout(cct) << "ll_releasedir" << std::endl; | |
f67539c2 | 14070 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 14071 | |
f67539c2 | 14072 | std::scoped_lock lock(client_lock); |
181888fb | 14073 | |
7c673cae FG |
14074 | _closedir(dirp); |
14075 | return 0; | |
14076 | } | |
14077 | ||
14078 | int Client::ll_fsyncdir(dir_result_t *dirp) | |
14079 | { | |
f67539c2 TL |
14080 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14081 | if (!mref_reader.is_state_satisfied()) | |
14082 | return -CEPHFS_ENOTCONN; | |
14083 | ||
7c673cae FG |
14084 | ldout(cct, 3) << "ll_fsyncdir " << dirp << dendl; |
14085 | tout(cct) << "ll_fsyncdir" << std::endl; | |
f67539c2 | 14086 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 14087 | |
f67539c2 | 14088 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14089 | return _fsync(dirp->inode.get(), false); |
14090 | } | |
14091 | ||
14092 | int Client::ll_open(Inode *in, int flags, Fh **fhp, const UserPerm& perms) | |
14093 | { | |
11fdf7f2 | 14094 | ceph_assert(!(flags & O_CREAT)); |
7c673cae | 14095 | |
f67539c2 TL |
14096 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14097 | if (!mref_reader.is_state_satisfied()) | |
14098 | return -CEPHFS_ENOTCONN; | |
181888fb | 14099 | |
7c673cae FG |
14100 | vinodeno_t vino = _get_vino(in); |
14101 | ||
14102 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << dendl; | |
14103 | tout(cct) << "ll_open" << std::endl; | |
14104 | tout(cct) << vino.ino.val << std::endl; | |
14105 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
14106 | ||
f67539c2 TL |
14107 | std::scoped_lock lock(client_lock); |
14108 | ||
7c673cae | 14109 | int r; |
11fdf7f2 | 14110 | if (!fuse_default_permissions) { |
7c673cae FG |
14111 | r = may_open(in, flags, perms); |
14112 | if (r < 0) | |
14113 | goto out; | |
14114 | } | |
14115 | ||
14116 | r = _open(in, flags, 0, fhp /* may be NULL */, perms); | |
14117 | ||
14118 | out: | |
14119 | Fh *fhptr = fhp ? *fhp : NULL; | |
14120 | if (fhptr) { | |
14121 | ll_unclosed_fh_set.insert(fhptr); | |
14122 | } | |
f67539c2 | 14123 | tout(cct) << (uintptr_t)fhptr << std::endl; |
7c673cae FG |
14124 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << |
14125 | " = " << r << " (" << fhptr << ")" << dendl; | |
14126 | return r; | |
14127 | } | |
14128 | ||
14129 | int Client::_ll_create(Inode *parent, const char *name, mode_t mode, | |
14130 | int flags, InodeRef *in, int caps, Fh **fhp, | |
14131 | const UserPerm& perms) | |
14132 | { | |
14133 | *fhp = NULL; | |
14134 | ||
14135 | vinodeno_t vparent = _get_vino(parent); | |
14136 | ||
1adf2230 | 14137 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14138 | mode << dec << " " << ceph_flags_sys2wire(flags) << ", uid " << perms.uid() |
14139 | << ", gid " << perms.gid() << dendl; | |
14140 | tout(cct) << "ll_create" << std::endl; | |
14141 | tout(cct) << vparent.ino.val << std::endl; | |
14142 | tout(cct) << name << std::endl; | |
14143 | tout(cct) << mode << std::endl; | |
14144 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
14145 | ||
14146 | bool created = false; | |
14147 | int r = _lookup(parent, name, caps, in, perms); | |
14148 | ||
14149 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
f67539c2 | 14150 | return -CEPHFS_EEXIST; |
7c673cae | 14151 | |
f67539c2 | 14152 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
11fdf7f2 | 14153 | if (!fuse_default_permissions) { |
7c673cae FG |
14154 | r = may_create(parent, perms); |
14155 | if (r < 0) | |
14156 | goto out; | |
14157 | } | |
14158 | r = _create(parent, name, flags, mode, in, fhp, 0, 0, 0, NULL, &created, | |
f67539c2 | 14159 | perms, ""); |
7c673cae FG |
14160 | if (r < 0) |
14161 | goto out; | |
14162 | } | |
14163 | ||
14164 | if (r < 0) | |
14165 | goto out; | |
14166 | ||
11fdf7f2 | 14167 | ceph_assert(*in); |
7c673cae FG |
14168 | |
14169 | ldout(cct, 20) << "_ll_create created = " << created << dendl; | |
14170 | if (!created) { | |
11fdf7f2 | 14171 | if (!fuse_default_permissions) { |
7c673cae FG |
14172 | r = may_open(in->get(), flags, perms); |
14173 | if (r < 0) { | |
14174 | if (*fhp) { | |
14175 | int release_r = _release_fh(*fhp); | |
11fdf7f2 | 14176 | ceph_assert(release_r == 0); // during create, no async data ops should have happened |
7c673cae FG |
14177 | } |
14178 | goto out; | |
14179 | } | |
14180 | } | |
14181 | if (*fhp == NULL) { | |
14182 | r = _open(in->get(), flags, mode, fhp, perms); | |
14183 | if (r < 0) | |
14184 | goto out; | |
14185 | } | |
14186 | } | |
14187 | ||
14188 | out: | |
14189 | if (*fhp) { | |
14190 | ll_unclosed_fh_set.insert(*fhp); | |
14191 | } | |
14192 | ||
14193 | ino_t ino = 0; | |
14194 | if (r >= 0) { | |
14195 | Inode *inode = in->get(); | |
14196 | if (use_faked_inos()) | |
14197 | ino = inode->faked_ino; | |
14198 | else | |
14199 | ino = inode->ino; | |
14200 | } | |
14201 | ||
f67539c2 | 14202 | tout(cct) << (uintptr_t)*fhp << std::endl; |
7c673cae | 14203 | tout(cct) << ino << std::endl; |
1adf2230 | 14204 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14205 | mode << dec << " " << ceph_flags_sys2wire(flags) << " = " << r << " (" << |
14206 | *fhp << " " << hex << ino << dec << ")" << dendl; | |
14207 | ||
14208 | return r; | |
14209 | } | |
14210 | ||
14211 | int Client::ll_create(Inode *parent, const char *name, mode_t mode, | |
14212 | int flags, struct stat *attr, Inode **outp, Fh **fhp, | |
14213 | const UserPerm& perms) | |
14214 | { | |
f67539c2 TL |
14215 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14216 | if (!mref_reader.is_state_satisfied()) | |
14217 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14218 | |
f67539c2 TL |
14219 | std::scoped_lock lock(client_lock); |
14220 | InodeRef in; | |
181888fb | 14221 | |
7c673cae FG |
14222 | int r = _ll_create(parent, name, mode, flags, &in, CEPH_STAT_CAP_INODE_ALL, |
14223 | fhp, perms); | |
14224 | if (r >= 0) { | |
11fdf7f2 | 14225 | ceph_assert(in); |
7c673cae FG |
14226 | |
14227 | // passing an Inode in outp requires an additional ref | |
14228 | if (outp) { | |
14229 | _ll_get(in.get()); | |
14230 | *outp = in.get(); | |
14231 | } | |
14232 | fill_stat(in, attr); | |
14233 | } else { | |
14234 | attr->st_ino = 0; | |
14235 | } | |
14236 | ||
14237 | return r; | |
14238 | } | |
14239 | ||
14240 | int Client::ll_createx(Inode *parent, const char *name, mode_t mode, | |
14241 | int oflags, Inode **outp, Fh **fhp, | |
14242 | struct ceph_statx *stx, unsigned want, unsigned lflags, | |
14243 | const UserPerm& perms) | |
14244 | { | |
14245 | unsigned caps = statx_to_mask(lflags, want); | |
f67539c2 TL |
14246 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14247 | if (!mref_reader.is_state_satisfied()) | |
14248 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14249 | |
f67539c2 TL |
14250 | std::scoped_lock lock(client_lock); |
14251 | InodeRef in; | |
7c673cae FG |
14252 | |
14253 | int r = _ll_create(parent, name, mode, oflags, &in, caps, fhp, perms); | |
14254 | if (r >= 0) { | |
11fdf7f2 | 14255 | ceph_assert(in); |
7c673cae FG |
14256 | |
14257 | // passing an Inode in outp requires an additional ref | |
14258 | if (outp) { | |
14259 | _ll_get(in.get()); | |
14260 | *outp = in.get(); | |
14261 | } | |
14262 | fill_statx(in, caps, stx); | |
14263 | } else { | |
14264 | stx->stx_ino = 0; | |
14265 | stx->stx_mask = 0; | |
14266 | } | |
14267 | ||
14268 | return r; | |
14269 | } | |
14270 | ||
14271 | loff_t Client::ll_lseek(Fh *fh, loff_t offset, int whence) | |
14272 | { | |
f67539c2 TL |
14273 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14274 | if (!mref_reader.is_state_satisfied()) | |
14275 | return -CEPHFS_ENOTCONN; | |
14276 | ||
7c673cae FG |
14277 | tout(cct) << "ll_lseek" << std::endl; |
14278 | tout(cct) << offset << std::endl; | |
14279 | tout(cct) << whence << std::endl; | |
14280 | ||
f67539c2 | 14281 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14282 | return _lseek(fh, offset, whence); |
14283 | } | |
14284 | ||
14285 | int Client::ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl) | |
14286 | { | |
f67539c2 TL |
14287 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14288 | if (!mref_reader.is_state_satisfied()) | |
14289 | return -CEPHFS_ENOTCONN; | |
14290 | ||
7c673cae FG |
14291 | ldout(cct, 3) << "ll_read " << fh << " " << fh->inode->ino << " " << " " << off << "~" << len << dendl; |
14292 | tout(cct) << "ll_read" << std::endl; | |
f67539c2 | 14293 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14294 | tout(cct) << off << std::endl; |
14295 | tout(cct) << len << std::endl; | |
14296 | ||
11fdf7f2 TL |
14297 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14298 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14299 | std::scoped_lock lock(client_lock); |
14300 | ||
f6b5b4d7 TL |
14301 | int r = _read(fh, off, len, bl); |
14302 | ldout(cct, 3) << "ll_read " << fh << " " << off << "~" << len << " = " << r | |
14303 | << dendl; | |
14304 | return r; | |
7c673cae FG |
14305 | } |
14306 | ||
14307 | int Client::ll_read_block(Inode *in, uint64_t blockid, | |
14308 | char *buf, | |
14309 | uint64_t offset, | |
14310 | uint64_t length, | |
14311 | file_layout_t* layout) | |
14312 | { | |
f67539c2 TL |
14313 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14314 | if (!mref_reader.is_state_satisfied()) | |
14315 | return -CEPHFS_ENOTCONN; | |
181888fb | 14316 | |
b32b8144 | 14317 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14318 | object_t oid = file_object_t(vino.ino, blockid); |
14319 | C_SaferCond onfinish; | |
14320 | bufferlist bl; | |
14321 | ||
14322 | objecter->read(oid, | |
14323 | object_locator_t(layout->pool_id), | |
14324 | offset, | |
14325 | length, | |
14326 | vino.snapid, | |
14327 | &bl, | |
14328 | CEPH_OSD_FLAG_READ, | |
14329 | &onfinish); | |
14330 | ||
7c673cae | 14331 | int r = onfinish.wait(); |
7c673cae | 14332 | if (r >= 0) { |
9f95a23c | 14333 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
14334 | r = bl.length(); |
14335 | } | |
14336 | ||
14337 | return r; | |
14338 | } | |
14339 | ||
14340 | /* It appears that the OSD doesn't return success unless the entire | |
14341 | buffer was written, return the write length on success. */ | |
14342 | ||
14343 | int Client::ll_write_block(Inode *in, uint64_t blockid, | |
14344 | char* buf, uint64_t offset, | |
14345 | uint64_t length, file_layout_t* layout, | |
14346 | uint64_t snapseq, uint32_t sync) | |
14347 | { | |
7c673cae | 14348 | vinodeno_t vino = ll_get_vino(in); |
7c673cae | 14349 | int r = 0; |
11fdf7f2 | 14350 | std::unique_ptr<C_SaferCond> onsafe = nullptr; |
f67539c2 TL |
14351 | |
14352 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
14353 | if (!mref_reader.is_state_satisfied()) | |
14354 | return -CEPHFS_ENOTCONN; | |
14355 | ||
7c673cae | 14356 | if (length == 0) { |
f67539c2 | 14357 | return -CEPHFS_EINVAL; |
7c673cae FG |
14358 | } |
14359 | if (true || sync) { | |
14360 | /* if write is stable, the epilogue is waiting on | |
14361 | * flock */ | |
11fdf7f2 | 14362 | onsafe.reset(new C_SaferCond("Client::ll_write_block flock")); |
7c673cae FG |
14363 | } |
14364 | object_t oid = file_object_t(vino.ino, blockid); | |
14365 | SnapContext fakesnap; | |
11fdf7f2 TL |
14366 | ceph::bufferlist bl; |
14367 | if (length > 0) { | |
14368 | bl.push_back(buffer::copy(buf, length)); | |
14369 | } | |
7c673cae FG |
14370 | |
14371 | ldout(cct, 1) << "ll_block_write for " << vino.ino << "." << blockid | |
14372 | << dendl; | |
14373 | ||
14374 | fakesnap.seq = snapseq; | |
14375 | ||
14376 | /* lock just in time */ | |
7c673cae FG |
14377 | objecter->write(oid, |
14378 | object_locator_t(layout->pool_id), | |
14379 | offset, | |
14380 | length, | |
14381 | fakesnap, | |
14382 | bl, | |
14383 | ceph::real_clock::now(), | |
14384 | 0, | |
11fdf7f2 | 14385 | onsafe.get()); |
7c673cae | 14386 | |
11fdf7f2 TL |
14387 | if (nullptr != onsafe) { |
14388 | r = onsafe->wait(); | |
7c673cae FG |
14389 | } |
14390 | ||
14391 | if (r < 0) { | |
14392 | return r; | |
14393 | } else { | |
14394 | return length; | |
14395 | } | |
14396 | } | |
14397 | ||
14398 | int Client::ll_commit_blocks(Inode *in, | |
14399 | uint64_t offset, | |
14400 | uint64_t length) | |
14401 | { | |
7c673cae FG |
14402 | /* |
14403 | BarrierContext *bctx; | |
b32b8144 | 14404 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14405 | uint64_t ino = vino.ino; |
14406 | ||
14407 | ldout(cct, 1) << "ll_commit_blocks for " << vino.ino << " from " | |
14408 | << offset << " to " << length << dendl; | |
14409 | ||
14410 | if (length == 0) { | |
f67539c2 | 14411 | return -CEPHFS_EINVAL; |
7c673cae FG |
14412 | } |
14413 | ||
f67539c2 | 14414 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14415 | map<uint64_t, BarrierContext*>::iterator p = barriers.find(ino); |
14416 | if (p != barriers.end()) { | |
14417 | barrier_interval civ(offset, offset + length); | |
14418 | p->second->commit_barrier(civ); | |
14419 | } | |
14420 | */ | |
14421 | return 0; | |
14422 | } | |
14423 | ||
14424 | int Client::ll_write(Fh *fh, loff_t off, loff_t len, const char *data) | |
14425 | { | |
7c673cae FG |
14426 | ldout(cct, 3) << "ll_write " << fh << " " << fh->inode->ino << " " << off << |
14427 | "~" << len << dendl; | |
14428 | tout(cct) << "ll_write" << std::endl; | |
f67539c2 | 14429 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14430 | tout(cct) << off << std::endl; |
14431 | tout(cct) << len << std::endl; | |
14432 | ||
f67539c2 TL |
14433 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14434 | if (!mref_reader.is_state_satisfied()) | |
14435 | return -CEPHFS_ENOTCONN; | |
181888fb | 14436 | |
11fdf7f2 TL |
14437 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14438 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14439 | std::scoped_lock lock(client_lock); |
14440 | ||
7c673cae FG |
14441 | int r = _write(fh, off, len, data, NULL, 0); |
14442 | ldout(cct, 3) << "ll_write " << fh << " " << off << "~" << len << " = " << r | |
14443 | << dendl; | |
14444 | return r; | |
14445 | } | |
14446 | ||
11fdf7f2 TL |
14447 | int64_t Client::ll_writev(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) |
14448 | { | |
f67539c2 TL |
14449 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14450 | if (!mref_reader.is_state_satisfied()) | |
14451 | return -CEPHFS_ENOTCONN; | |
14452 | ||
20effc67 TL |
14453 | std::scoped_lock cl(client_lock); |
14454 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, true, false); | |
11fdf7f2 TL |
14455 | } |
14456 | ||
14457 | int64_t Client::ll_readv(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) | |
14458 | { | |
f67539c2 TL |
14459 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14460 | if (!mref_reader.is_state_satisfied()) | |
14461 | return -CEPHFS_ENOTCONN; | |
14462 | ||
20effc67 TL |
14463 | std::scoped_lock cl(client_lock); |
14464 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, false, false); | |
11fdf7f2 TL |
14465 | } |
14466 | ||
7c673cae FG |
14467 | int Client::ll_flush(Fh *fh) |
14468 | { | |
f67539c2 TL |
14469 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14470 | if (!mref_reader.is_state_satisfied()) | |
14471 | return -CEPHFS_ENOTCONN; | |
14472 | ||
7c673cae FG |
14473 | ldout(cct, 3) << "ll_flush " << fh << " " << fh->inode->ino << " " << dendl; |
14474 | tout(cct) << "ll_flush" << std::endl; | |
f67539c2 | 14475 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14476 | |
f67539c2 | 14477 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14478 | return _flush(fh); |
14479 | } | |
14480 | ||
14481 | int Client::ll_fsync(Fh *fh, bool syncdataonly) | |
14482 | { | |
f67539c2 TL |
14483 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14484 | if (!mref_reader.is_state_satisfied()) | |
14485 | return -CEPHFS_ENOTCONN; | |
14486 | ||
7c673cae FG |
14487 | ldout(cct, 3) << "ll_fsync " << fh << " " << fh->inode->ino << " " << dendl; |
14488 | tout(cct) << "ll_fsync" << std::endl; | |
f67539c2 | 14489 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14490 | |
f67539c2 | 14491 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14492 | int r = _fsync(fh, syncdataonly); |
14493 | if (r) { | |
14494 | // If we're returning an error, clear it from the FH | |
14495 | fh->take_async_err(); | |
14496 | } | |
14497 | return r; | |
14498 | } | |
14499 | ||
28e407b8 AA |
14500 | int Client::ll_sync_inode(Inode *in, bool syncdataonly) |
14501 | { | |
f67539c2 TL |
14502 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14503 | if (!mref_reader.is_state_satisfied()) | |
14504 | return -CEPHFS_ENOTCONN; | |
14505 | ||
28e407b8 AA |
14506 | ldout(cct, 3) << "ll_sync_inode " << *in << " " << dendl; |
14507 | tout(cct) << "ll_sync_inode" << std::endl; | |
f67539c2 | 14508 | tout(cct) << (uintptr_t)in << std::endl; |
28e407b8 | 14509 | |
f67539c2 | 14510 | std::scoped_lock lock(client_lock); |
28e407b8 AA |
14511 | return _fsync(in, syncdataonly); |
14512 | } | |
14513 | ||
7c673cae FG |
14514 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
14515 | { | |
f67539c2 TL |
14516 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
14517 | ||
7c673cae | 14518 | if (offset < 0 || length <= 0) |
f67539c2 | 14519 | return -CEPHFS_EINVAL; |
7c673cae FG |
14520 | |
14521 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
f67539c2 | 14522 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14523 | |
14524 | if ((mode & FALLOC_FL_PUNCH_HOLE) && !(mode & FALLOC_FL_KEEP_SIZE)) | |
f67539c2 | 14525 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14526 | |
14527 | Inode *in = fh->inode.get(); | |
14528 | ||
14529 | if (objecter->osdmap_pool_full(in->layout.pool_id) && | |
14530 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
f67539c2 | 14531 | return -CEPHFS_ENOSPC; |
7c673cae FG |
14532 | } |
14533 | ||
14534 | if (in->snapid != CEPH_NOSNAP) | |
f67539c2 | 14535 | return -CEPHFS_EROFS; |
7c673cae FG |
14536 | |
14537 | if ((fh->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 14538 | return -CEPHFS_EBADF; |
7c673cae FG |
14539 | |
14540 | uint64_t size = offset + length; | |
14541 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && | |
14542 | size > in->size && | |
11fdf7f2 | 14543 | is_quota_bytes_exceeded(in, size - in->size, fh->actor_perms)) { |
f67539c2 | 14544 | return -CEPHFS_EDQUOT; |
7c673cae FG |
14545 | } |
14546 | ||
14547 | int have; | |
f6b5b4d7 | 14548 | int r = get_caps(fh, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER, &have, -1); |
7c673cae FG |
14549 | if (r < 0) |
14550 | return r; | |
14551 | ||
11fdf7f2 | 14552 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
7c673cae FG |
14553 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
14554 | if (in->inline_version < CEPH_INLINE_NONE && | |
14555 | (have & CEPH_CAP_FILE_BUFFER)) { | |
14556 | bufferlist bl; | |
9f95a23c | 14557 | auto inline_iter = in->inline_data.cbegin(); |
7c673cae FG |
14558 | int len = in->inline_data.length(); |
14559 | if (offset < len) { | |
14560 | if (offset > 0) | |
9f95a23c | 14561 | inline_iter.copy(offset, bl); |
7c673cae FG |
14562 | int size = length; |
14563 | if (offset + size > len) | |
14564 | size = len - offset; | |
14565 | if (size > 0) | |
14566 | bl.append_zero(size); | |
9f95a23c TL |
14567 | if (offset + size < len) { |
14568 | inline_iter += size; | |
14569 | inline_iter.copy(len - offset - size, bl); | |
14570 | } | |
7c673cae FG |
14571 | in->inline_data = bl; |
14572 | in->inline_version++; | |
14573 | } | |
91327a77 | 14574 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14575 | in->change_attr++; |
28e407b8 | 14576 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14577 | } else { |
14578 | if (in->inline_version < CEPH_INLINE_NONE) { | |
11fdf7f2 TL |
14579 | onuninline.reset(new C_SaferCond("Client::_fallocate_uninline_data flock")); |
14580 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
14581 | } |
14582 | ||
11fdf7f2 | 14583 | C_SaferCond onfinish("Client::_punch_hole flock"); |
7c673cae | 14584 | |
7c673cae FG |
14585 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
14586 | ||
14587 | _invalidate_inode_cache(in, offset, length); | |
14588 | filer->zero(in->ino, &in->layout, | |
14589 | in->snaprealm->get_snap_context(), | |
14590 | offset, length, | |
14591 | ceph::real_clock::now(), | |
11fdf7f2 | 14592 | 0, true, &onfinish); |
91327a77 | 14593 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14594 | in->change_attr++; |
28e407b8 | 14595 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14596 | |
9f95a23c | 14597 | client_lock.unlock(); |
11fdf7f2 | 14598 | onfinish.wait(); |
9f95a23c | 14599 | client_lock.lock(); |
f67539c2 | 14600 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
14601 | } |
14602 | } else if (!(mode & FALLOC_FL_KEEP_SIZE)) { | |
14603 | uint64_t size = offset + length; | |
14604 | if (size > in->size) { | |
14605 | in->size = size; | |
91327a77 | 14606 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14607 | in->change_attr++; |
28e407b8 | 14608 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14609 | |
11fdf7f2 | 14610 | if (is_quota_bytes_approaching(in, fh->actor_perms)) { |
7c673cae | 14611 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
14612 | } else if (is_max_size_approaching(in)) { |
14613 | check_caps(in, 0); | |
7c673cae FG |
14614 | } |
14615 | } | |
14616 | } | |
14617 | ||
11fdf7f2 | 14618 | if (nullptr != onuninline) { |
9f95a23c | 14619 | client_lock.unlock(); |
11fdf7f2 | 14620 | int ret = onuninline->wait(); |
9f95a23c | 14621 | client_lock.lock(); |
7c673cae | 14622 | |
f67539c2 | 14623 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
14624 | in->inline_data.clear(); |
14625 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 14626 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14627 | check_caps(in, 0); |
14628 | } else | |
11fdf7f2 | 14629 | r = ret; |
7c673cae FG |
14630 | } |
14631 | ||
14632 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
14633 | return r; | |
14634 | } | |
7c673cae | 14635 | |
11fdf7f2 | 14636 | int Client::ll_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
7c673cae | 14637 | { |
f67539c2 TL |
14638 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14639 | if (!mref_reader.is_state_satisfied()) | |
14640 | return -CEPHFS_ENOTCONN; | |
14641 | ||
11fdf7f2 TL |
14642 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << dendl; |
14643 | tout(cct) << __func__ << " " << mode << " " << offset << " " << length << std::endl; | |
f67539c2 | 14644 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14645 | |
f67539c2 | 14646 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14647 | return _fallocate(fh, mode, offset, length); |
14648 | } | |
14649 | ||
14650 | int Client::fallocate(int fd, int mode, loff_t offset, loff_t length) | |
14651 | { | |
f67539c2 TL |
14652 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14653 | if (!mref_reader.is_state_satisfied()) | |
14654 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14655 | |
f67539c2 | 14656 | tout(cct) << __func__ << " " << " " << fd << mode << " " << offset << " " << length << std::endl; |
181888fb | 14657 | |
f67539c2 | 14658 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14659 | Fh *fh = get_filehandle(fd); |
14660 | if (!fh) | |
f67539c2 | 14661 | return -CEPHFS_EBADF; |
7c673cae FG |
14662 | #if defined(__linux__) && defined(O_PATH) |
14663 | if (fh->flags & O_PATH) | |
f67539c2 | 14664 | return -CEPHFS_EBADF; |
7c673cae FG |
14665 | #endif |
14666 | return _fallocate(fh, mode, offset, length); | |
14667 | } | |
14668 | ||
14669 | int Client::ll_release(Fh *fh) | |
14670 | { | |
f67539c2 TL |
14671 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14672 | if (!mref_reader.is_state_satisfied()) | |
14673 | return -CEPHFS_ENOTCONN; | |
91327a77 | 14674 | |
11fdf7f2 | 14675 | ldout(cct, 3) << __func__ << " (fh)" << fh << " " << fh->inode->ino << " " << |
7c673cae | 14676 | dendl; |
11fdf7f2 | 14677 | tout(cct) << __func__ << " (fh)" << std::endl; |
f67539c2 TL |
14678 | tout(cct) << (uintptr_t)fh << std::endl; |
14679 | ||
14680 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
14681 | |
14682 | if (ll_unclosed_fh_set.count(fh)) | |
14683 | ll_unclosed_fh_set.erase(fh); | |
14684 | return _release_fh(fh); | |
14685 | } | |
14686 | ||
14687 | int Client::ll_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
14688 | { | |
f67539c2 TL |
14689 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14690 | if (!mref_reader.is_state_satisfied()) | |
14691 | return -CEPHFS_ENOTCONN; | |
7c673cae FG |
14692 | |
14693 | ldout(cct, 3) << "ll_getlk (fh)" << fh << " " << fh->inode->ino << dendl; | |
f67539c2 | 14694 | tout(cct) << "ll_getk (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14695 | |
f67539c2 | 14696 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14697 | return _getlk(fh, fl, owner); |
14698 | } | |
14699 | ||
14700 | int Client::ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
14701 | { | |
f67539c2 TL |
14702 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14703 | if (!mref_reader.is_state_satisfied()) | |
14704 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14705 | |
11fdf7f2 | 14706 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14707 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14708 | |
f67539c2 | 14709 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14710 | return _setlk(fh, fl, owner, sleep); |
14711 | } | |
14712 | ||
14713 | int Client::ll_flock(Fh *fh, int cmd, uint64_t owner) | |
14714 | { | |
f67539c2 TL |
14715 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14716 | if (!mref_reader.is_state_satisfied()) | |
14717 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14718 | |
11fdf7f2 | 14719 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14720 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14721 | |
f67539c2 | 14722 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14723 | return _flock(fh, cmd, owner); |
14724 | } | |
14725 | ||
b32b8144 FG |
14726 | int Client::set_deleg_timeout(uint32_t timeout) |
14727 | { | |
f67539c2 | 14728 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14729 | |
14730 | /* | |
f67539c2 | 14731 | * The whole point is to prevent blocklisting so we must time out the |
b32b8144 FG |
14732 | * delegation before the session autoclose timeout kicks in. |
14733 | */ | |
14734 | if (timeout >= mdsmap->get_session_autoclose()) | |
f67539c2 | 14735 | return -CEPHFS_EINVAL; |
b32b8144 FG |
14736 | |
14737 | deleg_timeout = timeout; | |
14738 | return 0; | |
14739 | } | |
14740 | ||
14741 | int Client::ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv) | |
14742 | { | |
f67539c2 | 14743 | int ret = -CEPHFS_EINVAL; |
b32b8144 | 14744 | |
f67539c2 TL |
14745 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14746 | if (!mref_reader.is_state_satisfied()) | |
14747 | return -CEPHFS_ENOTCONN; | |
b32b8144 | 14748 | |
f67539c2 | 14749 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14750 | |
14751 | Inode *inode = fh->inode.get(); | |
14752 | ||
14753 | switch(cmd) { | |
14754 | case CEPH_DELEGATION_NONE: | |
14755 | inode->unset_deleg(fh); | |
14756 | ret = 0; | |
14757 | break; | |
14758 | default: | |
14759 | try { | |
14760 | ret = inode->set_deleg(fh, cmd, cb, priv); | |
11fdf7f2 | 14761 | } catch (std::bad_alloc&) { |
f67539c2 | 14762 | ret = -CEPHFS_ENOMEM; |
b32b8144 FG |
14763 | } |
14764 | break; | |
14765 | } | |
14766 | return ret; | |
14767 | } | |
14768 | ||
7c673cae FG |
14769 | class C_Client_RequestInterrupt : public Context { |
14770 | private: | |
14771 | Client *client; | |
14772 | MetaRequest *req; | |
14773 | public: | |
14774 | C_Client_RequestInterrupt(Client *c, MetaRequest *r) : client(c), req(r) { | |
14775 | req->get(); | |
14776 | } | |
14777 | void finish(int r) override { | |
f67539c2 | 14778 | std::scoped_lock l(client->client_lock); |
11fdf7f2 | 14779 | ceph_assert(req->head.op == CEPH_MDS_OP_SETFILELOCK); |
7c673cae FG |
14780 | client->_interrupt_filelock(req); |
14781 | client->put_request(req); | |
14782 | } | |
14783 | }; | |
14784 | ||
14785 | void Client::ll_interrupt(void *d) | |
14786 | { | |
14787 | MetaRequest *req = static_cast<MetaRequest*>(d); | |
11fdf7f2 TL |
14788 | ldout(cct, 3) << __func__ << " tid " << req->get_tid() << dendl; |
14789 | tout(cct) << __func__ << " tid " << req->get_tid() << std::endl; | |
7c673cae FG |
14790 | interrupt_finisher.queue(new C_Client_RequestInterrupt(this, req)); |
14791 | } | |
14792 | ||
14793 | // ========================================= | |
14794 | // layout | |
14795 | ||
14796 | // expose file layouts | |
14797 | ||
14798 | int Client::describe_layout(const char *relpath, file_layout_t *lp, | |
14799 | const UserPerm& perms) | |
14800 | { | |
f67539c2 TL |
14801 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14802 | if (!mref_reader.is_state_satisfied()) | |
14803 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14804 | |
f67539c2 | 14805 | std::scoped_lock lock(client_lock); |
181888fb | 14806 | |
7c673cae FG |
14807 | filepath path(relpath); |
14808 | InodeRef in; | |
14809 | int r = path_walk(path, &in, perms); | |
14810 | if (r < 0) | |
14811 | return r; | |
14812 | ||
14813 | *lp = in->layout; | |
14814 | ||
11fdf7f2 | 14815 | ldout(cct, 3) << __func__ << "(" << relpath << ") = 0" << dendl; |
7c673cae FG |
14816 | return 0; |
14817 | } | |
14818 | ||
14819 | int Client::fdescribe_layout(int fd, file_layout_t *lp) | |
14820 | { | |
f67539c2 TL |
14821 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14822 | if (!mref_reader.is_state_satisfied()) | |
14823 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14824 | |
f67539c2 | 14825 | std::scoped_lock lock(client_lock); |
181888fb | 14826 | |
7c673cae FG |
14827 | Fh *f = get_filehandle(fd); |
14828 | if (!f) | |
f67539c2 | 14829 | return -CEPHFS_EBADF; |
7c673cae FG |
14830 | Inode *in = f->inode.get(); |
14831 | ||
14832 | *lp = in->layout; | |
14833 | ||
11fdf7f2 | 14834 | ldout(cct, 3) << __func__ << "(" << fd << ") = 0" << dendl; |
7c673cae FG |
14835 | return 0; |
14836 | } | |
14837 | ||
d2e6a577 FG |
14838 | int64_t Client::get_default_pool_id() |
14839 | { | |
f67539c2 TL |
14840 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14841 | if (!mref_reader.is_state_satisfied()) | |
14842 | return -CEPHFS_ENOTCONN; | |
181888fb | 14843 | |
f67539c2 | 14844 | std::scoped_lock lock(client_lock); |
181888fb | 14845 | |
d2e6a577 FG |
14846 | /* first data pool is the default */ |
14847 | return mdsmap->get_first_data_pool(); | |
14848 | } | |
7c673cae FG |
14849 | |
14850 | // expose osdmap | |
14851 | ||
14852 | int64_t Client::get_pool_id(const char *pool_name) | |
14853 | { | |
f67539c2 TL |
14854 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14855 | if (!mref_reader.is_state_satisfied()) | |
14856 | return -CEPHFS_ENOTCONN; | |
181888fb | 14857 | |
f67539c2 | 14858 | std::scoped_lock lock(client_lock); |
181888fb | 14859 | |
7c673cae FG |
14860 | return objecter->with_osdmap(std::mem_fn(&OSDMap::lookup_pg_pool_name), |
14861 | pool_name); | |
14862 | } | |
14863 | ||
14864 | string Client::get_pool_name(int64_t pool) | |
14865 | { | |
f67539c2 TL |
14866 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14867 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
14868 | return string(); |
14869 | ||
f67539c2 TL |
14870 | std::scoped_lock lock(client_lock); |
14871 | ||
7c673cae FG |
14872 | return objecter->with_osdmap([pool](const OSDMap& o) { |
14873 | return o.have_pg_pool(pool) ? o.get_pool_name(pool) : string(); | |
14874 | }); | |
14875 | } | |
14876 | ||
14877 | int Client::get_pool_replication(int64_t pool) | |
14878 | { | |
f67539c2 TL |
14879 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14880 | if (!mref_reader.is_state_satisfied()) | |
14881 | return -CEPHFS_ENOTCONN; | |
181888fb | 14882 | |
f67539c2 | 14883 | std::scoped_lock lock(client_lock); |
181888fb | 14884 | |
7c673cae | 14885 | return objecter->with_osdmap([pool](const OSDMap& o) { |
f67539c2 | 14886 | return o.have_pg_pool(pool) ? o.get_pg_pool(pool)->get_size() : -CEPHFS_ENOENT; |
7c673cae FG |
14887 | }); |
14888 | } | |
14889 | ||
14890 | int Client::get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds) | |
14891 | { | |
f67539c2 TL |
14892 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14893 | if (!mref_reader.is_state_satisfied()) | |
14894 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14895 | |
f67539c2 | 14896 | std::scoped_lock lock(client_lock); |
181888fb | 14897 | |
7c673cae FG |
14898 | Fh *f = get_filehandle(fd); |
14899 | if (!f) | |
f67539c2 | 14900 | return -CEPHFS_EBADF; |
7c673cae FG |
14901 | Inode *in = f->inode.get(); |
14902 | ||
14903 | vector<ObjectExtent> extents; | |
14904 | Striper::file_to_extents(cct, in->ino, &in->layout, off, 1, in->truncate_size, extents); | |
11fdf7f2 | 14905 | ceph_assert(extents.size() == 1); |
7c673cae FG |
14906 | |
14907 | objecter->with_osdmap([&](const OSDMap& o) { | |
14908 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
14909 | o.pg_to_acting_osds(pg, osds); | |
14910 | }); | |
14911 | ||
14912 | if (osds.empty()) | |
f67539c2 | 14913 | return -CEPHFS_EINVAL; |
7c673cae FG |
14914 | |
14915 | /* | |
14916 | * Return the remainder of the extent (stripe unit) | |
14917 | * | |
14918 | * If length = 1 is passed to Striper::file_to_extents we get a single | |
14919 | * extent back, but its length is one so we still need to compute the length | |
14920 | * to the end of the stripe unit. | |
14921 | * | |
14922 | * If length = su then we may get 1 or 2 objects back in the extents vector | |
14923 | * which would have to be examined. Even then, the offsets are local to the | |
14924 | * object, so matching up to the file offset is extra work. | |
14925 | * | |
14926 | * It seems simpler to stick with length = 1 and manually compute the | |
14927 | * remainder. | |
14928 | */ | |
14929 | if (len) { | |
14930 | uint64_t su = in->layout.stripe_unit; | |
14931 | *len = su - (off % su); | |
14932 | } | |
14933 | ||
14934 | return 0; | |
14935 | } | |
14936 | ||
14937 | int Client::get_osd_crush_location(int id, vector<pair<string, string> >& path) | |
14938 | { | |
f67539c2 TL |
14939 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14940 | if (!mref_reader.is_state_satisfied()) | |
14941 | return -CEPHFS_ENOTCONN; | |
181888fb | 14942 | |
f67539c2 | 14943 | std::scoped_lock lock(client_lock); |
181888fb | 14944 | |
7c673cae | 14945 | if (id < 0) |
f67539c2 | 14946 | return -CEPHFS_EINVAL; |
7c673cae FG |
14947 | return objecter->with_osdmap([&](const OSDMap& o) { |
14948 | return o.crush->get_full_location_ordered(id, path); | |
14949 | }); | |
14950 | } | |
14951 | ||
14952 | int Client::get_file_stripe_address(int fd, loff_t offset, | |
14953 | vector<entity_addr_t>& address) | |
14954 | { | |
f67539c2 TL |
14955 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14956 | if (!mref_reader.is_state_satisfied()) | |
14957 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14958 | |
f67539c2 | 14959 | std::scoped_lock lock(client_lock); |
181888fb | 14960 | |
7c673cae FG |
14961 | Fh *f = get_filehandle(fd); |
14962 | if (!f) | |
f67539c2 | 14963 | return -CEPHFS_EBADF; |
7c673cae FG |
14964 | Inode *in = f->inode.get(); |
14965 | ||
14966 | // which object? | |
14967 | vector<ObjectExtent> extents; | |
14968 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, 1, | |
14969 | in->truncate_size, extents); | |
11fdf7f2 | 14970 | ceph_assert(extents.size() == 1); |
7c673cae FG |
14971 | |
14972 | // now we have the object and its 'layout' | |
14973 | return objecter->with_osdmap([&](const OSDMap& o) { | |
14974 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
14975 | vector<int> osds; | |
14976 | o.pg_to_acting_osds(pg, osds); | |
14977 | if (osds.empty()) | |
f67539c2 | 14978 | return -CEPHFS_EINVAL; |
7c673cae | 14979 | for (unsigned i = 0; i < osds.size(); i++) { |
11fdf7f2 | 14980 | entity_addr_t addr = o.get_addrs(osds[i]).front(); |
7c673cae FG |
14981 | address.push_back(addr); |
14982 | } | |
14983 | return 0; | |
14984 | }); | |
14985 | } | |
14986 | ||
14987 | int Client::get_osd_addr(int osd, entity_addr_t& addr) | |
14988 | { | |
f67539c2 TL |
14989 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14990 | if (!mref_reader.is_state_satisfied()) | |
14991 | return -CEPHFS_ENOTCONN; | |
181888fb | 14992 | |
f67539c2 | 14993 | std::scoped_lock lock(client_lock); |
181888fb | 14994 | |
7c673cae FG |
14995 | return objecter->with_osdmap([&](const OSDMap& o) { |
14996 | if (!o.exists(osd)) | |
f67539c2 | 14997 | return -CEPHFS_ENOENT; |
7c673cae | 14998 | |
11fdf7f2 | 14999 | addr = o.get_addrs(osd).front(); |
7c673cae FG |
15000 | return 0; |
15001 | }); | |
15002 | } | |
15003 | ||
15004 | int Client::enumerate_layout(int fd, vector<ObjectExtent>& result, | |
15005 | loff_t length, loff_t offset) | |
15006 | { | |
f67539c2 TL |
15007 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15008 | if (!mref_reader.is_state_satisfied()) | |
15009 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15010 | |
f67539c2 | 15011 | std::scoped_lock lock(client_lock); |
181888fb | 15012 | |
7c673cae FG |
15013 | Fh *f = get_filehandle(fd); |
15014 | if (!f) | |
f67539c2 | 15015 | return -CEPHFS_EBADF; |
7c673cae FG |
15016 | Inode *in = f->inode.get(); |
15017 | ||
15018 | // map to a list of extents | |
15019 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, length, in->truncate_size, result); | |
15020 | ||
11fdf7f2 | 15021 | ldout(cct, 3) << __func__ << "(" << fd << ", " << length << ", " << offset << ") = 0" << dendl; |
7c673cae FG |
15022 | return 0; |
15023 | } | |
15024 | ||
15025 | ||
f67539c2 | 15026 | /* find an osd with the same ip. -CEPHFS_ENXIO if none. */ |
7c673cae FG |
15027 | int Client::get_local_osd() |
15028 | { | |
f67539c2 TL |
15029 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15030 | if (!mref_reader.is_state_satisfied()) | |
15031 | return -CEPHFS_ENOTCONN; | |
181888fb | 15032 | |
f67539c2 | 15033 | std::scoped_lock lock(client_lock); |
181888fb | 15034 | |
7c673cae FG |
15035 | objecter->with_osdmap([this](const OSDMap& o) { |
15036 | if (o.get_epoch() != local_osd_epoch) { | |
11fdf7f2 | 15037 | local_osd = o.find_osd_on_ip(messenger->get_myaddrs().front()); |
7c673cae FG |
15038 | local_osd_epoch = o.get_epoch(); |
15039 | } | |
15040 | }); | |
15041 | return local_osd; | |
15042 | } | |
15043 | ||
15044 | ||
15045 | ||
15046 | ||
15047 | ||
15048 | ||
15049 | // =============================== | |
15050 | ||
15051 | void Client::ms_handle_connect(Connection *con) | |
15052 | { | |
11fdf7f2 | 15053 | ldout(cct, 10) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15054 | } |
15055 | ||
15056 | bool Client::ms_handle_reset(Connection *con) | |
15057 | { | |
11fdf7f2 | 15058 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15059 | return false; |
15060 | } | |
15061 | ||
15062 | void Client::ms_handle_remote_reset(Connection *con) | |
15063 | { | |
f67539c2 | 15064 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 15065 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15066 | switch (con->get_peer_type()) { |
15067 | case CEPH_ENTITY_TYPE_MDS: | |
15068 | { | |
15069 | // kludge to figure out which mds this is; fixme with a Connection* state | |
15070 | mds_rank_t mds = MDS_RANK_NONE; | |
20effc67 | 15071 | MetaSessionRef s = NULL; |
11fdf7f2 | 15072 | for (auto &p : mds_sessions) { |
b3b6e05e | 15073 | if (mdsmap->have_inst(p.first) && mdsmap->get_addrs(p.first) == con->get_peer_addrs()) { |
11fdf7f2 | 15074 | mds = p.first; |
20effc67 | 15075 | s = p.second; |
7c673cae FG |
15076 | } |
15077 | } | |
15078 | if (mds >= 0) { | |
20effc67 | 15079 | ceph_assert(s != NULL); |
7c673cae FG |
15080 | switch (s->state) { |
15081 | case MetaSession::STATE_CLOSING: | |
15082 | ldout(cct, 1) << "reset from mds we were closing; we'll call that closed" << dendl; | |
20effc67 | 15083 | _closed_mds_session(s.get()); |
7c673cae FG |
15084 | break; |
15085 | ||
15086 | case MetaSession::STATE_OPENING: | |
15087 | { | |
15088 | ldout(cct, 1) << "reset from mds we were opening; retrying" << dendl; | |
15089 | list<Context*> waiters; | |
15090 | waiters.swap(s->waiting_for_open); | |
20effc67 TL |
15091 | _closed_mds_session(s.get()); |
15092 | auto news = _get_or_open_mds_session(mds); | |
7c673cae FG |
15093 | news->waiting_for_open.swap(waiters); |
15094 | } | |
15095 | break; | |
15096 | ||
15097 | case MetaSession::STATE_OPEN: | |
15098 | { | |
f67539c2 | 15099 | objecter->maybe_request_map(); /* to check if we are blocklisted */ |
f6b5b4d7 | 15100 | if (cct->_conf.get_val<bool>("client_reconnect_stale")) { |
7c673cae | 15101 | ldout(cct, 1) << "reset from mds we were open; close mds session for reconnect" << dendl; |
20effc67 | 15102 | _closed_mds_session(s.get()); |
7c673cae FG |
15103 | } else { |
15104 | ldout(cct, 1) << "reset from mds we were open; mark session as stale" << dendl; | |
15105 | s->state = MetaSession::STATE_STALE; | |
15106 | } | |
15107 | } | |
15108 | break; | |
15109 | ||
15110 | case MetaSession::STATE_NEW: | |
15111 | case MetaSession::STATE_CLOSED: | |
15112 | default: | |
15113 | break; | |
15114 | } | |
15115 | } | |
15116 | } | |
15117 | break; | |
15118 | } | |
15119 | } | |
15120 | ||
15121 | bool Client::ms_handle_refused(Connection *con) | |
15122 | { | |
11fdf7f2 | 15123 | ldout(cct, 1) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15124 | return false; |
15125 | } | |
15126 | ||
7c673cae FG |
15127 | Inode *Client::get_quota_root(Inode *in, const UserPerm& perms) |
15128 | { | |
11fdf7f2 TL |
15129 | Inode *quota_in = root_ancestor; |
15130 | SnapRealm *realm = in->snaprealm; | |
15131 | while (realm) { | |
15132 | ldout(cct, 10) << __func__ << " realm " << realm->ino << dendl; | |
15133 | if (realm->ino != in->ino) { | |
15134 | auto p = inode_map.find(vinodeno_t(realm->ino, CEPH_NOSNAP)); | |
15135 | if (p == inode_map.end()) | |
15136 | break; | |
7c673cae | 15137 | |
11fdf7f2 TL |
15138 | if (p->second->quota.is_enable()) { |
15139 | quota_in = p->second; | |
15140 | break; | |
7c673cae | 15141 | } |
7c673cae | 15142 | } |
11fdf7f2 | 15143 | realm = realm->pparent; |
7c673cae | 15144 | } |
11fdf7f2 TL |
15145 | ldout(cct, 10) << __func__ << " " << in->vino() << " -> " << quota_in->vino() << dendl; |
15146 | return quota_in; | |
7c673cae FG |
15147 | } |
15148 | ||
15149 | /** | |
15150 | * Traverse quota ancestors of the Inode, return true | |
15151 | * if any of them passes the passed function | |
15152 | */ | |
15153 | bool Client::check_quota_condition(Inode *in, const UserPerm& perms, | |
15154 | std::function<bool (const Inode &in)> test) | |
15155 | { | |
15156 | while (true) { | |
11fdf7f2 | 15157 | ceph_assert(in != NULL); |
7c673cae FG |
15158 | if (test(*in)) { |
15159 | return true; | |
15160 | } | |
15161 | ||
15162 | if (in == root_ancestor) { | |
15163 | // We're done traversing, drop out | |
15164 | return false; | |
15165 | } else { | |
15166 | // Continue up the tree | |
15167 | in = get_quota_root(in, perms); | |
15168 | } | |
15169 | } | |
15170 | ||
15171 | return false; | |
15172 | } | |
15173 | ||
15174 | bool Client::is_quota_files_exceeded(Inode *in, const UserPerm& perms) | |
15175 | { | |
15176 | return check_quota_condition(in, perms, | |
15177 | [](const Inode &in) { | |
15178 | return in.quota.max_files && in.rstat.rsize() >= in.quota.max_files; | |
15179 | }); | |
15180 | } | |
15181 | ||
15182 | bool Client::is_quota_bytes_exceeded(Inode *in, int64_t new_bytes, | |
11fdf7f2 | 15183 | const UserPerm& perms) |
7c673cae FG |
15184 | { |
15185 | return check_quota_condition(in, perms, | |
11fdf7f2 | 15186 | [&new_bytes](const Inode &in) { |
7c673cae FG |
15187 | return in.quota.max_bytes && (in.rstat.rbytes + new_bytes) |
15188 | > in.quota.max_bytes; | |
15189 | }); | |
15190 | } | |
15191 | ||
11fdf7f2 | 15192 | bool Client::is_quota_bytes_approaching(Inode *in, const UserPerm& perms) |
7c673cae | 15193 | { |
9f95a23c TL |
15194 | ceph_assert(in->size >= in->reported_size); |
15195 | const uint64_t size = in->size - in->reported_size; | |
11fdf7f2 | 15196 | return check_quota_condition(in, perms, |
9f95a23c | 15197 | [&size](const Inode &in) { |
11fdf7f2 TL |
15198 | if (in.quota.max_bytes) { |
15199 | if (in.rstat.rbytes >= in.quota.max_bytes) { | |
15200 | return true; | |
15201 | } | |
15202 | ||
11fdf7f2 | 15203 | const uint64_t space = in.quota.max_bytes - in.rstat.rbytes; |
11fdf7f2 TL |
15204 | return (space >> 4) < size; |
15205 | } else { | |
15206 | return false; | |
15207 | } | |
15208 | }); | |
7c673cae FG |
15209 | } |
15210 | ||
15211 | enum { | |
15212 | POOL_CHECKED = 1, | |
15213 | POOL_CHECKING = 2, | |
15214 | POOL_READ = 4, | |
15215 | POOL_WRITE = 8, | |
15216 | }; | |
15217 | ||
15218 | int Client::check_pool_perm(Inode *in, int need) | |
15219 | { | |
f67539c2 TL |
15220 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
15221 | ||
7c673cae FG |
15222 | if (!cct->_conf->client_check_pool_perm) |
15223 | return 0; | |
15224 | ||
f67539c2 TL |
15225 | /* Only need to do this for regular files */ |
15226 | if (!in->is_file()) | |
15227 | return 0; | |
15228 | ||
7c673cae FG |
15229 | int64_t pool_id = in->layout.pool_id; |
15230 | std::string pool_ns = in->layout.pool_ns; | |
15231 | std::pair<int64_t, std::string> perm_key(pool_id, pool_ns); | |
15232 | int have = 0; | |
15233 | while (true) { | |
15234 | auto it = pool_perms.find(perm_key); | |
15235 | if (it == pool_perms.end()) | |
15236 | break; | |
15237 | if (it->second == POOL_CHECKING) { | |
15238 | // avoid concurrent checkings | |
15239 | wait_on_list(waiting_for_pool_perm); | |
15240 | } else { | |
15241 | have = it->second; | |
11fdf7f2 | 15242 | ceph_assert(have & POOL_CHECKED); |
7c673cae FG |
15243 | break; |
15244 | } | |
15245 | } | |
15246 | ||
15247 | if (!have) { | |
15248 | if (in->snapid != CEPH_NOSNAP) { | |
15249 | // pool permission check needs to write to the first object. But for snapshot, | |
20effc67 | 15250 | // head of the first object may have already been deleted. To avoid creating |
7c673cae FG |
15251 | // orphan object, skip the check for now. |
15252 | return 0; | |
15253 | } | |
15254 | ||
15255 | pool_perms[perm_key] = POOL_CHECKING; | |
15256 | ||
15257 | char oid_buf[32]; | |
15258 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (unsigned long long)in->ino); | |
15259 | object_t oid = oid_buf; | |
15260 | ||
15261 | SnapContext nullsnapc; | |
15262 | ||
15263 | C_SaferCond rd_cond; | |
15264 | ObjectOperation rd_op; | |
f67539c2 | 15265 | rd_op.stat(nullptr, nullptr, nullptr); |
7c673cae FG |
15266 | |
15267 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), rd_op, | |
15268 | nullsnapc, ceph::real_clock::now(), 0, &rd_cond); | |
15269 | ||
15270 | C_SaferCond wr_cond; | |
15271 | ObjectOperation wr_op; | |
15272 | wr_op.create(true); | |
15273 | ||
15274 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), wr_op, | |
15275 | nullsnapc, ceph::real_clock::now(), 0, &wr_cond); | |
15276 | ||
9f95a23c | 15277 | client_lock.unlock(); |
7c673cae FG |
15278 | int rd_ret = rd_cond.wait(); |
15279 | int wr_ret = wr_cond.wait(); | |
9f95a23c | 15280 | client_lock.lock(); |
7c673cae FG |
15281 | |
15282 | bool errored = false; | |
15283 | ||
f67539c2 | 15284 | if (rd_ret == 0 || rd_ret == -CEPHFS_ENOENT) |
7c673cae | 15285 | have |= POOL_READ; |
f67539c2 | 15286 | else if (rd_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15287 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15288 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15289 | errored = true; | |
15290 | } | |
15291 | ||
f67539c2 | 15292 | if (wr_ret == 0 || wr_ret == -CEPHFS_EEXIST) |
7c673cae | 15293 | have |= POOL_WRITE; |
f67539c2 | 15294 | else if (wr_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15295 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15296 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15297 | errored = true; | |
15298 | } | |
15299 | ||
15300 | if (errored) { | |
15301 | // Indeterminate: erase CHECKING state so that subsequent calls re-check. | |
15302 | // Raise EIO because actual error code might be misleading for | |
15303 | // userspace filesystem user. | |
15304 | pool_perms.erase(perm_key); | |
15305 | signal_cond_list(waiting_for_pool_perm); | |
f67539c2 | 15306 | return -CEPHFS_EIO; |
7c673cae FG |
15307 | } |
15308 | ||
15309 | pool_perms[perm_key] = have | POOL_CHECKED; | |
15310 | signal_cond_list(waiting_for_pool_perm); | |
15311 | } | |
15312 | ||
15313 | if ((need & CEPH_CAP_FILE_RD) && !(have & POOL_READ)) { | |
11fdf7f2 | 15314 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15315 | << " need " << ccap_string(need) << ", but no read perm" << dendl; |
f67539c2 | 15316 | return -CEPHFS_EPERM; |
7c673cae FG |
15317 | } |
15318 | if ((need & CEPH_CAP_FILE_WR) && !(have & POOL_WRITE)) { | |
11fdf7f2 | 15319 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15320 | << " need " << ccap_string(need) << ", but no write perm" << dendl; |
f67539c2 | 15321 | return -CEPHFS_EPERM; |
7c673cae FG |
15322 | } |
15323 | ||
15324 | return 0; | |
15325 | } | |
15326 | ||
15327 | int Client::_posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want) | |
15328 | { | |
15329 | if (acl_type == POSIX_ACL) { | |
15330 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15331 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15332 | ||
15333 | return posix_acl_permits(access_acl, in->uid, in->gid, perms, want); | |
15334 | } | |
15335 | } | |
f67539c2 | 15336 | return -CEPHFS_EAGAIN; |
7c673cae FG |
15337 | } |
15338 | ||
15339 | int Client::_posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms) | |
15340 | { | |
15341 | if (acl_type == NO_ACL) | |
15342 | return 0; | |
15343 | ||
15344 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
15345 | if (r < 0) | |
15346 | goto out; | |
15347 | ||
15348 | if (acl_type == POSIX_ACL) { | |
15349 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15350 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15351 | bufferptr acl(access_acl.c_str(), access_acl.length()); | |
15352 | r = posix_acl_access_chmod(acl, mode); | |
15353 | if (r < 0) | |
15354 | goto out; | |
15355 | r = _do_setxattr(in, ACL_EA_ACCESS, acl.c_str(), acl.length(), 0, perms); | |
15356 | } else { | |
15357 | r = 0; | |
15358 | } | |
15359 | } | |
15360 | out: | |
15361 | ldout(cct, 10) << __func__ << " ino " << in->ino << " result=" << r << dendl; | |
15362 | return r; | |
15363 | } | |
15364 | ||
15365 | int Client::_posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl, | |
15366 | const UserPerm& perms) | |
15367 | { | |
15368 | if (acl_type == NO_ACL) | |
15369 | return 0; | |
15370 | ||
15371 | if (S_ISLNK(*mode)) | |
15372 | return 0; | |
15373 | ||
15374 | int r = _getattr(dir, CEPH_STAT_CAP_XATTR, perms, dir->xattr_version == 0); | |
15375 | if (r < 0) | |
15376 | goto out; | |
15377 | ||
15378 | if (acl_type == POSIX_ACL) { | |
15379 | if (dir->xattrs.count(ACL_EA_DEFAULT)) { | |
15380 | map<string, bufferptr> xattrs; | |
15381 | ||
15382 | const bufferptr& default_acl = dir->xattrs[ACL_EA_DEFAULT]; | |
15383 | bufferptr acl(default_acl.c_str(), default_acl.length()); | |
15384 | r = posix_acl_inherit_mode(acl, mode); | |
15385 | if (r < 0) | |
15386 | goto out; | |
15387 | ||
15388 | if (r > 0) { | |
15389 | r = posix_acl_equiv_mode(acl.c_str(), acl.length(), mode); | |
15390 | if (r < 0) | |
15391 | goto out; | |
15392 | if (r > 0) | |
15393 | xattrs[ACL_EA_ACCESS] = acl; | |
15394 | } | |
15395 | ||
15396 | if (S_ISDIR(*mode)) | |
15397 | xattrs[ACL_EA_DEFAULT] = dir->xattrs[ACL_EA_DEFAULT]; | |
15398 | ||
15399 | r = xattrs.size(); | |
15400 | if (r > 0) | |
11fdf7f2 | 15401 | encode(xattrs, xattrs_bl); |
7c673cae FG |
15402 | } else { |
15403 | if (umask_cb) | |
15404 | *mode &= ~umask_cb(callback_handle); | |
15405 | r = 0; | |
15406 | } | |
15407 | } | |
15408 | out: | |
15409 | ldout(cct, 10) << __func__ << " dir ino " << dir->ino << " result=" << r << dendl; | |
15410 | return r; | |
15411 | } | |
15412 | ||
15413 | void Client::set_filer_flags(int flags) | |
15414 | { | |
f67539c2 | 15415 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15416 | ceph_assert(flags == 0 || |
7c673cae FG |
15417 | flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
15418 | objecter->add_global_op_flags(flags); | |
15419 | } | |
15420 | ||
15421 | void Client::clear_filer_flags(int flags) | |
15422 | { | |
f67539c2 | 15423 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15424 | ceph_assert(flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
7c673cae FG |
15425 | objecter->clear_global_op_flag(flags); |
15426 | } | |
15427 | ||
11fdf7f2 TL |
15428 | // called before mount |
15429 | void Client::set_uuid(const std::string& uuid) | |
15430 | { | |
f67539c2 TL |
15431 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15432 | ceph_assert(iref_reader.is_state_satisfied()); | |
15433 | ||
15434 | std::scoped_lock l(client_lock); | |
20effc67 | 15435 | ceph_assert(!uuid.empty()); |
11fdf7f2 TL |
15436 | |
15437 | metadata["uuid"] = uuid; | |
15438 | _close_sessions(); | |
15439 | } | |
15440 | ||
15441 | // called before mount. 0 means infinite | |
15442 | void Client::set_session_timeout(unsigned timeout) | |
15443 | { | |
f67539c2 TL |
15444 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15445 | ceph_assert(iref_reader.is_state_satisfied()); | |
15446 | ||
15447 | std::scoped_lock l(client_lock); | |
11fdf7f2 TL |
15448 | |
15449 | metadata["timeout"] = stringify(timeout); | |
15450 | } | |
15451 | ||
15452 | // called before mount | |
15453 | int Client::start_reclaim(const std::string& uuid, unsigned flags, | |
15454 | const std::string& fs_name) | |
15455 | { | |
f67539c2 TL |
15456 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15457 | if (!iref_reader.is_state_satisfied()) | |
15458 | return -CEPHFS_ENOTCONN; | |
11fdf7f2 TL |
15459 | |
15460 | if (uuid.empty()) | |
f67539c2 | 15461 | return -CEPHFS_EINVAL; |
11fdf7f2 | 15462 | |
f67539c2 | 15463 | std::unique_lock l(client_lock); |
11fdf7f2 TL |
15464 | { |
15465 | auto it = metadata.find("uuid"); | |
15466 | if (it != metadata.end() && it->second == uuid) | |
f67539c2 | 15467 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15468 | } |
15469 | ||
15470 | int r = subscribe_mdsmap(fs_name); | |
15471 | if (r < 0) { | |
15472 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
15473 | return r; | |
15474 | } | |
15475 | ||
15476 | if (metadata.empty()) | |
15477 | populate_metadata(""); | |
15478 | ||
15479 | while (mdsmap->get_epoch() == 0) | |
15480 | wait_on_list(waiting_for_mdsmap); | |
15481 | ||
15482 | reclaim_errno = 0; | |
15483 | for (unsigned mds = 0; mds < mdsmap->get_num_in_mds(); ) { | |
15484 | if (!mdsmap->is_up(mds)) { | |
15485 | ldout(cct, 10) << "mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
15486 | wait_on_list(waiting_for_mdsmap); | |
15487 | continue; | |
15488 | } | |
15489 | ||
20effc67 | 15490 | MetaSessionRef session; |
11fdf7f2 TL |
15491 | if (!have_open_session(mds)) { |
15492 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 15493 | if (session->state == MetaSession::STATE_REJECTED) |
f67539c2 | 15494 | return -CEPHFS_EPERM; |
11fdf7f2 TL |
15495 | if (session->state != MetaSession::STATE_OPENING) { |
15496 | // umounting? | |
f67539c2 | 15497 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15498 | } |
15499 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
15500 | wait_on_context_list(session->waiting_for_open); | |
11fdf7f2 TL |
15501 | continue; |
15502 | } | |
15503 | ||
20effc67 | 15504 | session = mds_sessions.at(mds); |
11fdf7f2 | 15505 | if (!session->mds_features.test(CEPHFS_FEATURE_RECLAIM_CLIENT)) |
f67539c2 | 15506 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
15507 | |
15508 | if (session->reclaim_state == MetaSession::RECLAIM_NULL || | |
15509 | session->reclaim_state == MetaSession::RECLAIMING) { | |
15510 | session->reclaim_state = MetaSession::RECLAIMING; | |
9f95a23c | 15511 | auto m = make_message<MClientReclaim>(uuid, flags); |
11fdf7f2 TL |
15512 | session->con->send_message2(std::move(m)); |
15513 | wait_on_list(waiting_for_reclaim); | |
15514 | } else if (session->reclaim_state == MetaSession::RECLAIM_FAIL) { | |
f67539c2 | 15515 | return reclaim_errno ? : -CEPHFS_ENOTRECOVERABLE; |
11fdf7f2 TL |
15516 | } else { |
15517 | mds++; | |
15518 | } | |
15519 | } | |
15520 | ||
15521 | // didn't find target session in any mds | |
15522 | if (reclaim_target_addrs.empty()) { | |
15523 | if (flags & CEPH_RECLAIM_RESET) | |
f67539c2 TL |
15524 | return -CEPHFS_ENOENT; |
15525 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15526 | } |
15527 | ||
15528 | if (flags & CEPH_RECLAIM_RESET) | |
15529 | return 0; | |
15530 | ||
f67539c2 TL |
15531 | // use blocklist to check if target session was killed |
15532 | // (config option mds_session_blocklist_on_evict needs to be true) | |
15533 | ldout(cct, 10) << __func__ << ": waiting for OSD epoch " << reclaim_osd_epoch << dendl; | |
15534 | bs::error_code ec; | |
15535 | l.unlock(); | |
15536 | objecter->wait_for_map(reclaim_osd_epoch, ca::use_blocked[ec]); | |
15537 | l.lock(); | |
11fdf7f2 | 15538 | |
f67539c2 TL |
15539 | if (ec) |
15540 | return ceph::from_error_code(ec); | |
15541 | ||
15542 | bool blocklisted = objecter->with_osdmap( | |
11fdf7f2 | 15543 | [this](const OSDMap &osd_map) -> bool { |
f67539c2 | 15544 | return osd_map.is_blocklisted(reclaim_target_addrs); |
11fdf7f2 | 15545 | }); |
f67539c2 TL |
15546 | if (blocklisted) |
15547 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15548 | |
15549 | metadata["reclaiming_uuid"] = uuid; | |
15550 | return 0; | |
15551 | } | |
15552 | ||
15553 | void Client::finish_reclaim() | |
15554 | { | |
15555 | auto it = metadata.find("reclaiming_uuid"); | |
15556 | if (it == metadata.end()) { | |
15557 | for (auto &p : mds_sessions) | |
20effc67 | 15558 | p.second->reclaim_state = MetaSession::RECLAIM_NULL; |
11fdf7f2 TL |
15559 | return; |
15560 | } | |
15561 | ||
15562 | for (auto &p : mds_sessions) { | |
20effc67 | 15563 | p.second->reclaim_state = MetaSession::RECLAIM_NULL; |
9f95a23c | 15564 | auto m = make_message<MClientReclaim>("", MClientReclaim::FLAG_FINISH); |
20effc67 | 15565 | p.second->con->send_message2(std::move(m)); |
11fdf7f2 TL |
15566 | } |
15567 | ||
15568 | metadata["uuid"] = it->second; | |
15569 | metadata.erase(it); | |
15570 | } | |
15571 | ||
15572 | void Client::handle_client_reclaim_reply(const MConstRef<MClientReclaimReply>& reply) | |
15573 | { | |
15574 | mds_rank_t from = mds_rank_t(reply->get_source().num()); | |
15575 | ldout(cct, 10) << __func__ << " " << *reply << " from mds." << from << dendl; | |
15576 | ||
f67539c2 | 15577 | std::scoped_lock cl(client_lock); |
20effc67 | 15578 | auto session = _get_mds_session(from, reply->get_connection().get()); |
11fdf7f2 TL |
15579 | if (!session) { |
15580 | ldout(cct, 10) << " discarding reclaim reply from sessionless mds." << from << dendl; | |
15581 | return; | |
15582 | } | |
15583 | ||
15584 | if (reply->get_result() >= 0) { | |
15585 | session->reclaim_state = MetaSession::RECLAIM_OK; | |
15586 | if (reply->get_epoch() > reclaim_osd_epoch) | |
15587 | reclaim_osd_epoch = reply->get_epoch(); | |
15588 | if (!reply->get_addrs().empty()) | |
15589 | reclaim_target_addrs = reply->get_addrs(); | |
15590 | } else { | |
15591 | session->reclaim_state = MetaSession::RECLAIM_FAIL; | |
15592 | reclaim_errno = reply->get_result(); | |
15593 | } | |
15594 | ||
15595 | signal_cond_list(waiting_for_reclaim); | |
15596 | } | |
15597 | ||
7c673cae FG |
15598 | /** |
15599 | * This is included in cap release messages, to cause | |
15600 | * the MDS to wait until this OSD map epoch. It is necessary | |
15601 | * in corner cases where we cancel RADOS ops, so that | |
15602 | * nobody else tries to do IO to the same objects in | |
15603 | * the same epoch as the cancelled ops. | |
15604 | */ | |
15605 | void Client::set_cap_epoch_barrier(epoch_t e) | |
15606 | { | |
15607 | ldout(cct, 5) << __func__ << " epoch = " << e << dendl; | |
15608 | cap_epoch_barrier = e; | |
15609 | } | |
15610 | ||
15611 | const char** Client::get_tracked_conf_keys() const | |
15612 | { | |
15613 | static const char* keys[] = { | |
15614 | "client_cache_size", | |
15615 | "client_cache_mid", | |
15616 | "client_acl_type", | |
b32b8144 FG |
15617 | "client_deleg_timeout", |
15618 | "client_deleg_break_on_open", | |
f67539c2 TL |
15619 | "client_oc_size", |
15620 | "client_oc_max_objects", | |
15621 | "client_oc_max_dirty", | |
15622 | "client_oc_target_dirty", | |
15623 | "client_oc_max_dirty_age", | |
7c673cae FG |
15624 | NULL |
15625 | }; | |
15626 | return keys; | |
15627 | } | |
15628 | ||
11fdf7f2 | 15629 | void Client::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
15630 | const std::set <std::string> &changed) |
15631 | { | |
f67539c2 | 15632 | std::scoped_lock lock(client_lock); |
7c673cae | 15633 | |
181888fb | 15634 | if (changed.count("client_cache_mid")) { |
7c673cae FG |
15635 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
15636 | } | |
15637 | if (changed.count("client_acl_type")) { | |
15638 | acl_type = NO_ACL; | |
15639 | if (cct->_conf->client_acl_type == "posix_acl") | |
15640 | acl_type = POSIX_ACL; | |
15641 | } | |
f67539c2 TL |
15642 | if (changed.count("client_oc_size")) { |
15643 | objectcacher->set_max_size(cct->_conf->client_oc_size); | |
15644 | } | |
15645 | if (changed.count("client_oc_max_objects")) { | |
15646 | objectcacher->set_max_objects(cct->_conf->client_oc_max_objects); | |
15647 | } | |
15648 | if (changed.count("client_oc_max_dirty")) { | |
15649 | objectcacher->set_max_dirty(cct->_conf->client_oc_max_dirty); | |
15650 | } | |
15651 | if (changed.count("client_oc_target_dirty")) { | |
15652 | objectcacher->set_target_dirty(cct->_conf->client_oc_target_dirty); | |
15653 | } | |
15654 | if (changed.count("client_oc_max_dirty_age")) { | |
15655 | objectcacher->set_max_dirty_age(cct->_conf->client_oc_max_dirty_age); | |
15656 | } | |
7c673cae FG |
15657 | } |
15658 | ||
7c673cae FG |
15659 | void intrusive_ptr_add_ref(Inode *in) |
15660 | { | |
b3b6e05e | 15661 | in->iget(); |
7c673cae | 15662 | } |
f67539c2 | 15663 | |
7c673cae FG |
15664 | void intrusive_ptr_release(Inode *in) |
15665 | { | |
15666 | in->client->put_inode(in); | |
15667 | } | |
15668 | ||
15669 | mds_rank_t Client::_get_random_up_mds() const | |
15670 | { | |
9f95a23c | 15671 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
15672 | |
15673 | std::set<mds_rank_t> up; | |
15674 | mdsmap->get_up_mds_set(up); | |
15675 | ||
15676 | if (up.empty()) | |
15677 | return MDS_RANK_NONE; | |
15678 | std::set<mds_rank_t>::const_iterator p = up.begin(); | |
15679 | for (int n = rand() % up.size(); n; n--) | |
15680 | ++p; | |
15681 | return *p; | |
15682 | } | |
15683 | ||
15684 | ||
f67539c2 TL |
15685 | StandaloneClient::StandaloneClient(Messenger *m, MonClient *mc, |
15686 | boost::asio::io_context& ictx) | |
15687 | : Client(m, mc, new Objecter(m->cct, m, mc, ictx)) | |
7c673cae FG |
15688 | { |
15689 | monclient->set_messenger(m); | |
15690 | objecter->set_client_incarnation(0); | |
15691 | } | |
15692 | ||
15693 | StandaloneClient::~StandaloneClient() | |
15694 | { | |
15695 | delete objecter; | |
15696 | objecter = nullptr; | |
15697 | } | |
15698 | ||
15699 | int StandaloneClient::init() | |
15700 | { | |
f67539c2 TL |
15701 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
15702 | ceph_assert(iref_writer.is_first_writer()); | |
15703 | ||
e306af50 | 15704 | _pre_init(); |
7c673cae FG |
15705 | objecter->init(); |
15706 | ||
9f95a23c | 15707 | client_lock.lock(); |
7c673cae FG |
15708 | |
15709 | messenger->add_dispatcher_tail(objecter); | |
15710 | messenger->add_dispatcher_tail(this); | |
15711 | ||
15712 | monclient->set_want_keys(CEPH_ENTITY_TYPE_MDS | CEPH_ENTITY_TYPE_OSD); | |
15713 | int r = monclient->init(); | |
15714 | if (r < 0) { | |
15715 | // need to do cleanup because we're in an intermediate init state | |
f67539c2 TL |
15716 | { |
15717 | std::scoped_lock l(timer_lock); | |
15718 | timer.shutdown(); | |
15719 | } | |
15720 | ||
9f95a23c | 15721 | client_lock.unlock(); |
7c673cae FG |
15722 | objecter->shutdown(); |
15723 | objectcacher->stop(); | |
15724 | monclient->shutdown(); | |
15725 | return r; | |
15726 | } | |
15727 | objecter->start(); | |
15728 | ||
9f95a23c | 15729 | client_lock.unlock(); |
7c673cae | 15730 | _finish_init(); |
f67539c2 | 15731 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
15732 | |
15733 | return 0; | |
15734 | } | |
15735 | ||
15736 | void StandaloneClient::shutdown() | |
15737 | { | |
15738 | Client::shutdown(); | |
15739 | objecter->shutdown(); | |
15740 | monclient->shutdown(); | |
15741 | } |