]>
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 | ||
2a845540 TL |
179 | // running average and standard deviation -- presented in |
180 | // Donald Knuth's TAoCP, Volume II. | |
181 | double calc_average(double old_avg, double value, uint64_t count) { | |
182 | double new_avg; | |
183 | if (count == 1) { | |
184 | new_avg = value; | |
185 | } else { | |
186 | new_avg = old_avg + ((value - old_avg) / count); | |
187 | } | |
188 | ||
189 | return new_avg; | |
190 | } | |
191 | ||
192 | double calc_sq_sum(double old_sq_sum, double old_mean, double new_mean, | |
193 | double value, uint64_t count) { | |
194 | double new_sq_sum; | |
195 | if (count == 1) { | |
196 | new_sq_sum = 0.0; | |
197 | } else { | |
198 | new_sq_sum = old_sq_sum + (value - old_mean)*(value - new_mean); | |
199 | } | |
200 | ||
201 | return new_sq_sum; | |
202 | } | |
7c673cae FG |
203 | |
204 | // ------------- | |
205 | ||
206 | Client::CommandHook::CommandHook(Client *client) : | |
207 | m_client(client) | |
208 | { | |
209 | } | |
210 | ||
9f95a23c TL |
211 | int Client::CommandHook::call( |
212 | std::string_view command, | |
213 | const cmdmap_t& cmdmap, | |
214 | Formatter *f, | |
215 | std::ostream& errss, | |
216 | bufferlist& out) | |
7c673cae | 217 | { |
7c673cae | 218 | f->open_object_section("result"); |
9f95a23c | 219 | { |
f67539c2 | 220 | std::scoped_lock l{m_client->client_lock}; |
9f95a23c TL |
221 | if (command == "mds_requests") |
222 | m_client->dump_mds_requests(f); | |
adb31ebb TL |
223 | else if (command == "mds_sessions") { |
224 | bool cap_dump = false; | |
225 | cmd_getval(cmdmap, "cap_dump", cap_dump); | |
226 | m_client->dump_mds_sessions(f, cap_dump); | |
227 | } else if (command == "dump_cache") | |
9f95a23c TL |
228 | m_client->dump_cache(f); |
229 | else if (command == "kick_stale_sessions") | |
230 | m_client->_kick_stale_sessions(); | |
231 | else if (command == "status") | |
232 | m_client->dump_status(f); | |
233 | else | |
234 | ceph_abort_msg("bad command registered"); | |
235 | } | |
7c673cae | 236 | f->close_section(); |
9f95a23c | 237 | return 0; |
7c673cae FG |
238 | } |
239 | ||
240 | ||
241 | // ------------- | |
242 | ||
b3b6e05e TL |
243 | int Client::get_fd_inode(int fd, InodeRef *in) { |
244 | int r = 0; | |
245 | if (fd == CEPHFS_AT_FDCWD) { | |
246 | *in = cwd; | |
247 | } else { | |
248 | Fh *f = get_filehandle(fd); | |
249 | if (!f) { | |
250 | r = -CEPHFS_EBADF; | |
251 | } else { | |
252 | *in = f->inode; | |
253 | } | |
254 | } | |
255 | return r; | |
256 | } | |
257 | ||
7c673cae FG |
258 | dir_result_t::dir_result_t(Inode *in, const UserPerm& perms) |
259 | : inode(in), offset(0), next_offset(2), | |
260 | release_count(0), ordered_count(0), cache_index(0), start_shared_gen(0), | |
261 | perms(perms) | |
262 | { } | |
263 | ||
264 | void Client::_reset_faked_inos() | |
265 | { | |
266 | ino_t start = 1024; | |
267 | free_faked_inos.clear(); | |
268 | free_faked_inos.insert(start, (uint32_t)-1 - start + 1); | |
269 | last_used_faked_ino = 0; | |
11fdf7f2 | 270 | last_used_faked_root = 0; |
f67539c2 TL |
271 | #ifdef _WIN32 |
272 | // On Windows, sizeof(ino_t) is just 2. Despite that, most "native" | |
273 | // Windows structures, including Dokan ones, are using 64B identifiers. | |
274 | _use_faked_inos = false; | |
275 | #else | |
7c673cae | 276 | _use_faked_inos = sizeof(ino_t) < 8 || cct->_conf->client_use_faked_inos; |
f67539c2 | 277 | #endif |
7c673cae FG |
278 | } |
279 | ||
280 | void Client::_assign_faked_ino(Inode *in) | |
281 | { | |
11fdf7f2 TL |
282 | if (0 == last_used_faked_ino) |
283 | last_used_faked_ino = last_used_faked_ino + 2048; // start(1024)~2048 reserved for _assign_faked_root | |
7c673cae FG |
284 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
285 | if (it == free_faked_inos.end() && last_used_faked_ino > 0) { | |
11fdf7f2 | 286 | last_used_faked_ino = 2048; |
7c673cae FG |
287 | it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
288 | } | |
11fdf7f2 | 289 | ceph_assert(it != free_faked_inos.end()); |
7c673cae | 290 | if (last_used_faked_ino < it.get_start()) { |
11fdf7f2 | 291 | ceph_assert(it.get_len() > 0); |
7c673cae FG |
292 | last_used_faked_ino = it.get_start(); |
293 | } else { | |
294 | ++last_used_faked_ino; | |
11fdf7f2 | 295 | ceph_assert(it.get_start() + it.get_len() > last_used_faked_ino); |
7c673cae FG |
296 | } |
297 | in->faked_ino = last_used_faked_ino; | |
298 | free_faked_inos.erase(in->faked_ino); | |
299 | faked_ino_map[in->faked_ino] = in->vino(); | |
300 | } | |
301 | ||
11fdf7f2 TL |
302 | /* |
303 | * In the faked mode, if you export multiple subdirectories, | |
304 | * you will see that the inode numbers of the exported subdirectories | |
305 | * are the same. so we distinguish the mount point by reserving | |
306 | * the "fake ids" between "1024~2048" and combining the last | |
307 | * 10bits(0x3ff) of the "root inodes". | |
308 | */ | |
309 | void Client::_assign_faked_root(Inode *in) | |
310 | { | |
311 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
312 | if (it == free_faked_inos.end() && last_used_faked_root > 0) { | |
313 | last_used_faked_root = 0; | |
314 | it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
315 | } | |
20effc67 | 316 | ceph_assert(it != free_faked_inos.end()); |
11fdf7f2 TL |
317 | vinodeno_t inode_info = in->vino(); |
318 | uint64_t inode_num = (uint64_t)inode_info.ino; | |
319 | ldout(cct, 10) << "inode_num " << inode_num << "inode_num & 0x3ff=" << (inode_num & 0x3ff)<< dendl; | |
320 | last_used_faked_root = it.get_start() + (inode_num & 0x3ff); // 0x3ff mask and get_start will not exceed 2048 | |
20effc67 | 321 | ceph_assert(it.get_start() + it.get_len() > last_used_faked_root); |
11fdf7f2 TL |
322 | |
323 | in->faked_ino = last_used_faked_root; | |
324 | free_faked_inos.erase(in->faked_ino); | |
325 | faked_ino_map[in->faked_ino] = in->vino(); | |
326 | } | |
327 | ||
7c673cae FG |
328 | void Client::_release_faked_ino(Inode *in) |
329 | { | |
330 | free_faked_inos.insert(in->faked_ino); | |
331 | faked_ino_map.erase(in->faked_ino); | |
332 | } | |
333 | ||
334 | vinodeno_t Client::_map_faked_ino(ino_t ino) | |
335 | { | |
336 | vinodeno_t vino; | |
337 | if (ino == 1) | |
338 | vino = root->vino(); | |
339 | else if (faked_ino_map.count(ino)) | |
340 | vino = faked_ino_map[ino]; | |
341 | else | |
342 | vino = vinodeno_t(0, CEPH_NOSNAP); | |
11fdf7f2 | 343 | ldout(cct, 10) << __func__ << " " << ino << " -> " << vino << dendl; |
7c673cae FG |
344 | return vino; |
345 | } | |
346 | ||
347 | vinodeno_t Client::map_faked_ino(ino_t ino) | |
348 | { | |
f67539c2 | 349 | std::scoped_lock lock(client_lock); |
7c673cae FG |
350 | return _map_faked_ino(ino); |
351 | } | |
352 | ||
353 | // cons/des | |
354 | ||
355 | Client::Client(Messenger *m, MonClient *mc, Objecter *objecter_) | |
f67539c2 TL |
356 | : Dispatcher(m->cct->get()), |
357 | timer(m->cct, timer_lock, false), | |
11fdf7f2 TL |
358 | messenger(m), |
359 | monclient(mc), | |
360 | objecter(objecter_), | |
361 | whoami(mc->get_global_id()), | |
f67539c2 TL |
362 | mount_state(CLIENT_UNMOUNTED, "Client::mountstate_lock"), |
363 | initialize_state(CLIENT_NEW, "Client::initstate_lock"), | |
364 | cct_deleter{m->cct, [](CephContext *p) {p->put();}}, | |
7c673cae FG |
365 | async_ino_invalidator(m->cct), |
366 | async_dentry_invalidator(m->cct), | |
367 | interrupt_finisher(m->cct), | |
368 | remount_finisher(m->cct), | |
e306af50 | 369 | async_ino_releasor(m->cct), |
7c673cae | 370 | objecter_finisher(m->cct), |
11fdf7f2 TL |
371 | m_command_hook(this), |
372 | fscid(0) | |
7c673cae FG |
373 | { |
374 | _reset_faked_inos(); | |
7c673cae | 375 | |
7c673cae FG |
376 | user_id = cct->_conf->client_mount_uid; |
377 | group_id = cct->_conf->client_mount_gid; | |
92f5a8d4 TL |
378 | fuse_default_permissions = cct->_conf.get_val<bool>( |
379 | "fuse_default_permissions"); | |
7c673cae | 380 | |
33c7a0ef TL |
381 | _collect_and_send_global_metrics = cct->_conf.get_val<bool>( |
382 | "client_collect_and_send_global_metrics"); | |
383 | ||
2a845540 TL |
384 | mount_timeout = cct->_conf.get_val<std::chrono::seconds>( |
385 | "client_mount_timeout"); | |
386 | ||
387 | caps_release_delay = cct->_conf.get_val<std::chrono::seconds>( | |
388 | "client_caps_release_delay"); | |
389 | ||
7c673cae FG |
390 | if (cct->_conf->client_acl_type == "posix_acl") |
391 | acl_type = POSIX_ACL; | |
392 | ||
7c673cae FG |
393 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
394 | ||
395 | // file handles | |
396 | free_fd_set.insert(10, 1<<30); | |
397 | ||
398 | mdsmap.reset(new MDSMap); | |
399 | ||
400 | // osd interfaces | |
401 | writeback_handler.reset(new ObjecterWriteback(objecter, &objecter_finisher, | |
402 | &client_lock)); | |
403 | objectcacher.reset(new ObjectCacher(cct, "libcephfs", *writeback_handler, client_lock, | |
404 | client_flush_set_callback, // all commit callback | |
405 | (void*)this, | |
406 | cct->_conf->client_oc_size, | |
407 | cct->_conf->client_oc_max_objects, | |
408 | cct->_conf->client_oc_max_dirty, | |
409 | cct->_conf->client_oc_target_dirty, | |
410 | cct->_conf->client_oc_max_dirty_age, | |
411 | true)); | |
7c673cae FG |
412 | } |
413 | ||
414 | ||
415 | Client::~Client() | |
416 | { | |
9f95a23c | 417 | ceph_assert(ceph_mutex_is_not_locked(client_lock)); |
7c673cae | 418 | |
f67539c2 TL |
419 | // If the task is crashed or aborted and doesn't |
420 | // get any chance to run the umount and shutdow. | |
421 | { | |
422 | std::scoped_lock l{client_lock}; | |
423 | tick_thread_stopped = true; | |
424 | upkeep_cond.notify_one(); | |
425 | } | |
426 | ||
427 | if (upkeeper.joinable()) | |
428 | upkeeper.join(); | |
429 | ||
31f18b77 FG |
430 | // It is necessary to hold client_lock, because any inode destruction |
431 | // may call into ObjectCacher, which asserts that it's lock (which is | |
432 | // client_lock) is held. | |
f67539c2 | 433 | std::scoped_lock l{client_lock}; |
7c673cae FG |
434 | tear_down_cache(); |
435 | } | |
436 | ||
437 | void Client::tear_down_cache() | |
438 | { | |
439 | // fd's | |
f67539c2 TL |
440 | for (auto &[fd, fh] : fd_map) { |
441 | ldout(cct, 1) << __func__ << " forcing close of fh " << fd << " ino " << fh->inode->ino << dendl; | |
7c673cae FG |
442 | _release_fh(fh); |
443 | } | |
444 | fd_map.clear(); | |
445 | ||
446 | while (!opened_dirs.empty()) { | |
447 | dir_result_t *dirp = *opened_dirs.begin(); | |
11fdf7f2 | 448 | ldout(cct, 1) << __func__ << " forcing close of dir " << dirp << " ino " << dirp->inode->ino << dendl; |
7c673cae FG |
449 | _closedir(dirp); |
450 | } | |
451 | ||
452 | // caps! | |
453 | // *** FIXME *** | |
454 | ||
455 | // empty lru | |
7c673cae | 456 | trim_cache(); |
11fdf7f2 | 457 | ceph_assert(lru.lru_get_size() == 0); |
7c673cae FG |
458 | |
459 | // close root ino | |
11fdf7f2 | 460 | ceph_assert(inode_map.size() <= 1 + root_parents.size()); |
7c673cae | 461 | if (root && inode_map.size() == 1 + root_parents.size()) { |
b3b6e05e | 462 | root.reset(); |
7c673cae FG |
463 | } |
464 | ||
11fdf7f2 | 465 | ceph_assert(inode_map.empty()); |
7c673cae FG |
466 | } |
467 | ||
468 | inodeno_t Client::get_root_ino() | |
469 | { | |
f67539c2 | 470 | std::scoped_lock l(client_lock); |
7c673cae FG |
471 | if (use_faked_inos()) |
472 | return root->faked_ino; | |
473 | else | |
474 | return root->ino; | |
475 | } | |
476 | ||
477 | Inode *Client::get_root() | |
478 | { | |
f67539c2 | 479 | std::scoped_lock l(client_lock); |
7c673cae | 480 | root->ll_get(); |
b3b6e05e | 481 | return root.get(); |
7c673cae FG |
482 | } |
483 | ||
484 | ||
485 | // debug crapola | |
486 | ||
487 | void Client::dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected) | |
488 | { | |
489 | filepath path; | |
490 | in->make_long_path(path); | |
491 | ldout(cct, 1) << "dump_inode: " | |
492 | << (disconnected ? "DISCONNECTED ":"") | |
493 | << "inode " << in->ino | |
494 | << " " << path | |
b3b6e05e | 495 | << " ref " << in->get_nref() |
f67539c2 | 496 | << " " << *in << dendl; |
7c673cae FG |
497 | |
498 | if (f) { | |
499 | f->open_object_section("inode"); | |
500 | f->dump_stream("path") << path; | |
501 | if (disconnected) | |
502 | f->dump_int("disconnected", 1); | |
503 | in->dump(f); | |
504 | f->close_section(); | |
505 | } | |
506 | ||
507 | did.insert(in); | |
508 | if (in->dir) { | |
509 | ldout(cct, 1) << " dir " << in->dir << " size " << in->dir->dentries.size() << dendl; | |
510 | for (ceph::unordered_map<string, Dentry*>::iterator it = in->dir->dentries.begin(); | |
511 | it != in->dir->dentries.end(); | |
512 | ++it) { | |
513 | ldout(cct, 1) << " " << in->ino << " dn " << it->first << " " << it->second << " ref " << it->second->ref << dendl; | |
514 | if (f) { | |
515 | f->open_object_section("dentry"); | |
516 | it->second->dump(f); | |
517 | f->close_section(); | |
518 | } | |
519 | if (it->second->inode) | |
520 | dump_inode(f, it->second->inode.get(), did, false); | |
521 | } | |
522 | } | |
523 | } | |
524 | ||
525 | void Client::dump_cache(Formatter *f) | |
526 | { | |
527 | set<Inode*> did; | |
528 | ||
11fdf7f2 | 529 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
530 | |
531 | if (f) | |
532 | f->open_array_section("cache"); | |
533 | ||
534 | if (root) | |
b3b6e05e | 535 | dump_inode(f, root.get(), did, true); |
7c673cae FG |
536 | |
537 | // make a second pass to catch anything disconnected | |
538 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
539 | it != inode_map.end(); | |
540 | ++it) { | |
541 | if (did.count(it->second)) | |
542 | continue; | |
543 | dump_inode(f, it->second, did, true); | |
544 | } | |
545 | ||
546 | if (f) | |
547 | f->close_section(); | |
548 | } | |
549 | ||
550 | void Client::dump_status(Formatter *f) | |
551 | { | |
9f95a23c | 552 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
553 | |
554 | ldout(cct, 1) << __func__ << dendl; | |
555 | ||
556 | const epoch_t osd_epoch | |
557 | = objecter->with_osdmap(std::mem_fn(&OSDMap::get_epoch)); | |
558 | ||
559 | if (f) { | |
560 | f->open_object_section("metadata"); | |
561 | for (const auto& kv : metadata) | |
562 | f->dump_string(kv.first.c_str(), kv.second); | |
563 | f->close_section(); | |
564 | ||
565 | f->dump_int("dentry_count", lru.lru_get_size()); | |
566 | f->dump_int("dentry_pinned_count", lru.lru_get_num_pinned()); | |
567 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 568 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 | 569 | f->dump_object("inst", inst); |
11fdf7f2 TL |
570 | f->dump_object("addr", inst.addr); |
571 | f->dump_stream("inst_str") << inst.name << " " << inst.addr.get_legacy_str(); | |
572 | f->dump_string("addr_str", inst.addr.get_legacy_str()); | |
7c673cae FG |
573 | f->dump_int("inode_count", inode_map.size()); |
574 | f->dump_int("mds_epoch", mdsmap->get_epoch()); | |
575 | f->dump_int("osd_epoch", osd_epoch); | |
576 | f->dump_int("osd_epoch_barrier", cap_epoch_barrier); | |
f67539c2 | 577 | f->dump_bool("blocklisted", blocklisted); |
adb31ebb | 578 | f->dump_string("fs_name", mdsmap->get_fs_name()); |
7c673cae FG |
579 | } |
580 | } | |
581 | ||
e306af50 | 582 | void Client::_pre_init() |
7c673cae FG |
583 | { |
584 | timer.init(); | |
e306af50 TL |
585 | |
586 | objecter_finisher.start(); | |
587 | filer.reset(new Filer(objecter, &objecter_finisher)); | |
e306af50 | 588 | |
7c673cae | 589 | objectcacher->start(); |
e306af50 TL |
590 | } |
591 | ||
592 | int Client::init() | |
593 | { | |
f67539c2 TL |
594 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
595 | ceph_assert(iref_writer.is_first_writer()); | |
596 | ||
e306af50 | 597 | _pre_init(); |
9f95a23c | 598 | { |
f67539c2 | 599 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
600 | messenger->add_dispatcher_tail(this); |
601 | } | |
7c673cae | 602 | _finish_init(); |
f67539c2 | 603 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
604 | return 0; |
605 | } | |
606 | ||
607 | void Client::_finish_init() | |
608 | { | |
9f95a23c | 609 | { |
f67539c2 | 610 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
611 | // logger |
612 | PerfCountersBuilder plb(cct, "client", l_c_first, l_c_last); | |
613 | plb.add_time_avg(l_c_reply, "reply", "Latency of receiving a reply on metadata request"); | |
614 | plb.add_time_avg(l_c_lat, "lat", "Latency of processing a metadata request"); | |
615 | plb.add_time_avg(l_c_wrlat, "wrlat", "Latency of a file data write operation"); | |
616 | plb.add_time_avg(l_c_read, "rdlat", "Latency of a file data read operation"); | |
617 | plb.add_time_avg(l_c_fsync, "fsync", "Latency of a file sync operation"); | |
2a845540 TL |
618 | // average, standard deviation mds/r/w/ latencies |
619 | plb.add_time(l_c_md_avg, "mdavg", "Average latency for processing metadata requests"); | |
620 | plb.add_u64(l_c_md_sqsum, "mdsqsum", "Sum of squares (to calculate variability/stdev) for metadata requests"); | |
621 | plb.add_u64(l_c_md_ops, "mdops", "Total metadata IO operations"); | |
622 | plb.add_time(l_c_rd_avg, "readavg", "Average latency for processing read requests"); | |
623 | plb.add_u64(l_c_rd_sqsum, "readsqsum", "Sum of squares ((to calculate variability/stdev) for read requests"); | |
624 | plb.add_u64(l_c_rd_ops, "rdops", "Total read IO operations"); | |
625 | plb.add_time(l_c_wr_avg, "writeavg", "Average latency for processing write requests"); | |
626 | plb.add_u64(l_c_wr_sqsum, "writesqsum", "Sum of squares ((to calculate variability/stdev) for write requests"); | |
627 | plb.add_u64(l_c_wr_ops, "rdops", "Total write IO operations"); | |
9f95a23c TL |
628 | logger.reset(plb.create_perf_counters()); |
629 | cct->get_perfcounters_collection()->add(logger.get()); | |
630 | } | |
7c673cae | 631 | |
11fdf7f2 | 632 | cct->_conf.add_observer(this); |
7c673cae FG |
633 | |
634 | AdminSocket* admin_socket = cct->get_admin_socket(); | |
635 | int ret = admin_socket->register_command("mds_requests", | |
7c673cae FG |
636 | &m_command_hook, |
637 | "show in-progress mds requests"); | |
638 | if (ret < 0) { | |
639 | lderr(cct) << "error registering admin socket command: " | |
640 | << cpp_strerror(-ret) << dendl; | |
641 | } | |
adb31ebb TL |
642 | ret = admin_socket->register_command("mds_sessions " |
643 | "name=cap_dump,type=CephBool,req=false", | |
7c673cae FG |
644 | &m_command_hook, |
645 | "show mds session state"); | |
646 | if (ret < 0) { | |
647 | lderr(cct) << "error registering admin socket command: " | |
648 | << cpp_strerror(-ret) << dendl; | |
649 | } | |
650 | ret = admin_socket->register_command("dump_cache", | |
7c673cae FG |
651 | &m_command_hook, |
652 | "show in-memory metadata cache contents"); | |
653 | if (ret < 0) { | |
654 | lderr(cct) << "error registering admin socket command: " | |
655 | << cpp_strerror(-ret) << dendl; | |
656 | } | |
657 | ret = admin_socket->register_command("kick_stale_sessions", | |
7c673cae FG |
658 | &m_command_hook, |
659 | "kick sessions that were remote reset"); | |
660 | if (ret < 0) { | |
661 | lderr(cct) << "error registering admin socket command: " | |
662 | << cpp_strerror(-ret) << dendl; | |
663 | } | |
664 | ret = admin_socket->register_command("status", | |
7c673cae FG |
665 | &m_command_hook, |
666 | "show overall client status"); | |
667 | if (ret < 0) { | |
668 | lderr(cct) << "error registering admin socket command: " | |
669 | << cpp_strerror(-ret) << dendl; | |
670 | } | |
7c673cae FG |
671 | } |
672 | ||
673 | void Client::shutdown() | |
674 | { | |
11fdf7f2 | 675 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
676 | |
677 | // If we were not mounted, but were being used for sending | |
678 | // MDS commands, we may have sessions that need closing. | |
9f95a23c | 679 | { |
f67539c2 TL |
680 | std::scoped_lock l{client_lock}; |
681 | ||
682 | // To make sure the tick thread will be stoppped before | |
683 | // destructing the Client, just in case like the _mount() | |
684 | // failed but didn't not get a chance to stop the tick | |
685 | // thread | |
686 | tick_thread_stopped = true; | |
687 | upkeep_cond.notify_one(); | |
688 | ||
9f95a23c TL |
689 | _close_sessions(); |
690 | } | |
11fdf7f2 | 691 | cct->_conf.remove_observer(this); |
7c673cae | 692 | |
11fdf7f2 | 693 | cct->get_admin_socket()->unregister_commands(&m_command_hook); |
7c673cae FG |
694 | |
695 | if (ino_invalidate_cb) { | |
696 | ldout(cct, 10) << "shutdown stopping cache invalidator finisher" << dendl; | |
697 | async_ino_invalidator.wait_for_empty(); | |
698 | async_ino_invalidator.stop(); | |
699 | } | |
700 | ||
701 | if (dentry_invalidate_cb) { | |
702 | ldout(cct, 10) << "shutdown stopping dentry invalidator finisher" << dendl; | |
703 | async_dentry_invalidator.wait_for_empty(); | |
704 | async_dentry_invalidator.stop(); | |
705 | } | |
706 | ||
707 | if (switch_interrupt_cb) { | |
708 | ldout(cct, 10) << "shutdown stopping interrupt finisher" << dendl; | |
709 | interrupt_finisher.wait_for_empty(); | |
710 | interrupt_finisher.stop(); | |
711 | } | |
712 | ||
713 | if (remount_cb) { | |
714 | ldout(cct, 10) << "shutdown stopping remount finisher" << dendl; | |
715 | remount_finisher.wait_for_empty(); | |
716 | remount_finisher.stop(); | |
717 | } | |
718 | ||
e306af50 TL |
719 | if (ino_release_cb) { |
720 | ldout(cct, 10) << "shutdown stopping inode release finisher" << dendl; | |
721 | async_ino_releasor.wait_for_empty(); | |
722 | async_ino_releasor.stop(); | |
723 | } | |
724 | ||
7c673cae | 725 | objectcacher->stop(); // outside of client_lock! this does a join. |
f67539c2 TL |
726 | |
727 | /* | |
728 | * We are shuting down the client. | |
729 | * | |
730 | * Just declare the state to CLIENT_NEW to block and fail any | |
731 | * new comming "reader" and then try to wait all the in-flight | |
732 | * "readers" to finish. | |
733 | */ | |
734 | RWRef_t iref_writer(initialize_state, CLIENT_NEW, false); | |
735 | if (!iref_writer.is_first_writer()) | |
736 | return; | |
737 | iref_writer.wait_readers_done(); | |
738 | ||
9f95a23c | 739 | { |
f67539c2 | 740 | std::scoped_lock l(timer_lock); |
9f95a23c TL |
741 | timer.shutdown(); |
742 | } | |
f67539c2 | 743 | |
7c673cae FG |
744 | objecter_finisher.wait_for_empty(); |
745 | objecter_finisher.stop(); | |
746 | ||
747 | if (logger) { | |
748 | cct->get_perfcounters_collection()->remove(logger.get()); | |
749 | logger.reset(); | |
750 | } | |
751 | } | |
752 | ||
2a845540 TL |
753 | void Client::update_io_stat_metadata(utime_t latency) { |
754 | auto lat_nsec = latency.to_nsec(); | |
755 | // old values are used to compute new ones | |
756 | auto o_avg = logger->tget(l_c_md_avg).to_nsec(); | |
757 | auto o_sqsum = logger->get(l_c_md_sqsum); | |
758 | ||
759 | auto n_avg = calc_average(o_avg, lat_nsec, nr_metadata_request); | |
760 | auto n_sqsum = calc_sq_sum(o_sqsum, o_avg, n_avg, lat_nsec, | |
761 | nr_metadata_request); | |
762 | ||
763 | logger->tinc(l_c_lat, latency); | |
764 | logger->tinc(l_c_reply, latency); | |
765 | ||
766 | utime_t avg; | |
767 | avg.set_from_double(n_avg / 1000000000); | |
768 | logger->tset(l_c_md_avg, avg); | |
769 | logger->set(l_c_md_sqsum, n_sqsum); | |
770 | logger->set(l_c_md_ops, nr_metadata_request); | |
771 | } | |
772 | ||
773 | void Client::update_io_stat_read(utime_t latency) { | |
774 | auto lat_nsec = latency.to_nsec(); | |
775 | // old values are used to compute new ones | |
776 | auto o_avg = logger->tget(l_c_rd_avg).to_nsec(); | |
777 | auto o_sqsum = logger->get(l_c_rd_sqsum); | |
778 | ||
779 | auto n_avg = calc_average(o_avg, lat_nsec, nr_read_request); | |
780 | auto n_sqsum = calc_sq_sum(o_sqsum, o_avg, n_avg, lat_nsec, | |
781 | nr_read_request); | |
782 | ||
783 | logger->tinc(l_c_read, latency); | |
784 | ||
785 | utime_t avg; | |
786 | avg.set_from_double(n_avg / 1000000000); | |
787 | logger->tset(l_c_rd_avg, avg); | |
788 | logger->set(l_c_rd_sqsum, n_sqsum); | |
789 | logger->set(l_c_rd_ops, nr_read_request); | |
790 | } | |
791 | ||
792 | void Client::update_io_stat_write(utime_t latency) { | |
793 | auto lat_nsec = latency.to_nsec(); | |
794 | // old values are used to compute new ones | |
795 | auto o_avg = logger->tget(l_c_wr_avg).to_nsec(); | |
796 | auto o_sqsum = logger->get(l_c_wr_sqsum); | |
797 | ||
798 | auto n_avg = calc_average(o_avg, lat_nsec, nr_write_request); | |
799 | auto n_sqsum = calc_sq_sum(o_sqsum, o_avg, n_avg, lat_nsec, | |
800 | nr_write_request); | |
801 | ||
802 | logger->tinc(l_c_wrlat, latency); | |
803 | ||
804 | utime_t avg; | |
805 | avg.set_from_double(n_avg / 1000000000); | |
806 | logger->tset(l_c_wr_avg, avg); | |
807 | logger->set(l_c_wr_sqsum, n_sqsum); | |
808 | logger->set(l_c_wr_ops, nr_write_request); | |
809 | } | |
7c673cae FG |
810 | |
811 | // =================== | |
812 | // metadata cache stuff | |
813 | ||
814 | void Client::trim_cache(bool trim_kernel_dcache) | |
815 | { | |
181888fb FG |
816 | uint64_t max = cct->_conf->client_cache_size; |
817 | ldout(cct, 20) << "trim_cache size " << lru.lru_get_size() << " max " << max << dendl; | |
7c673cae FG |
818 | unsigned last = 0; |
819 | while (lru.lru_get_size() != last) { | |
820 | last = lru.lru_get_size(); | |
821 | ||
f67539c2 | 822 | if (!is_unmounting() && lru.lru_get_size() <= max) break; |
7c673cae FG |
823 | |
824 | // trim! | |
31f18b77 | 825 | Dentry *dn = static_cast<Dentry*>(lru.lru_get_next_expire()); |
7c673cae FG |
826 | if (!dn) |
827 | break; // done | |
f67539c2 | 828 | |
7c673cae FG |
829 | trim_dentry(dn); |
830 | } | |
831 | ||
181888fb | 832 | if (trim_kernel_dcache && lru.lru_get_size() > max) |
7c673cae FG |
833 | _invalidate_kernel_dcache(); |
834 | ||
835 | // hose root? | |
b3b6e05e | 836 | if (lru.lru_get_size() == 0 && root && root->get_nref() == 1 && inode_map.size() == 1 + root_parents.size()) { |
7c673cae | 837 | ldout(cct, 15) << "trim_cache trimmed root " << root << dendl; |
b3b6e05e | 838 | root.reset(); |
7c673cae FG |
839 | } |
840 | } | |
841 | ||
842 | void Client::trim_cache_for_reconnect(MetaSession *s) | |
843 | { | |
844 | mds_rank_t mds = s->mds_num; | |
11fdf7f2 | 845 | ldout(cct, 20) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
846 | |
847 | int trimmed = 0; | |
848 | list<Dentry*> skipped; | |
849 | while (lru.lru_get_size() > 0) { | |
850 | Dentry *dn = static_cast<Dentry*>(lru.lru_expire()); | |
851 | if (!dn) | |
852 | break; | |
853 | ||
854 | if ((dn->inode && dn->inode->caps.count(mds)) || | |
855 | dn->dir->parent_inode->caps.count(mds)) { | |
856 | trim_dentry(dn); | |
857 | trimmed++; | |
858 | } else | |
859 | skipped.push_back(dn); | |
860 | } | |
861 | ||
862 | for(list<Dentry*>::iterator p = skipped.begin(); p != skipped.end(); ++p) | |
863 | lru.lru_insert_mid(*p); | |
864 | ||
11fdf7f2 | 865 | ldout(cct, 20) << __func__ << " mds." << mds |
7c673cae FG |
866 | << " trimmed " << trimmed << " dentries" << dendl; |
867 | ||
868 | if (s->caps.size() > 0) | |
869 | _invalidate_kernel_dcache(); | |
870 | } | |
871 | ||
872 | void Client::trim_dentry(Dentry *dn) | |
873 | { | |
874 | ldout(cct, 15) << "trim_dentry unlinking dn " << dn->name | |
11fdf7f2 TL |
875 | << " in dir " |
876 | << std::hex << dn->dir->parent_inode->ino << std::dec | |
7c673cae FG |
877 | << dendl; |
878 | if (dn->inode) { | |
879 | Inode *diri = dn->dir->parent_inode; | |
7c673cae FG |
880 | clear_dir_complete_and_ordered(diri, true); |
881 | } | |
882 | unlink(dn, false, false); // drop dir, drop dentry | |
883 | } | |
884 | ||
885 | ||
1adf2230 AA |
886 | void Client::update_inode_file_size(Inode *in, int issued, uint64_t size, |
887 | uint64_t truncate_seq, uint64_t truncate_size) | |
7c673cae | 888 | { |
7c673cae FG |
889 | uint64_t prior_size = in->size; |
890 | ||
7c673cae FG |
891 | if (truncate_seq > in->truncate_seq || |
892 | (truncate_seq == in->truncate_seq && size > in->size)) { | |
893 | ldout(cct, 10) << "size " << in->size << " -> " << size << dendl; | |
894 | in->size = size; | |
895 | in->reported_size = size; | |
896 | if (truncate_seq != in->truncate_seq) { | |
897 | ldout(cct, 10) << "truncate_seq " << in->truncate_seq << " -> " | |
898 | << truncate_seq << dendl; | |
899 | in->truncate_seq = truncate_seq; | |
900 | in->oset.truncate_seq = truncate_seq; | |
901 | ||
902 | // truncate cached file data | |
903 | if (prior_size > size) { | |
904 | _invalidate_inode_cache(in, truncate_size, prior_size - truncate_size); | |
905 | } | |
906 | } | |
907 | ||
908 | // truncate inline data | |
909 | if (in->inline_version < CEPH_INLINE_NONE) { | |
910 | uint32_t len = in->inline_data.length(); | |
911 | if (size < len) | |
912 | in->inline_data.splice(size, len - size); | |
913 | } | |
914 | } | |
915 | if (truncate_seq >= in->truncate_seq && | |
916 | in->truncate_size != truncate_size) { | |
917 | if (in->is_file()) { | |
918 | ldout(cct, 10) << "truncate_size " << in->truncate_size << " -> " | |
919 | << truncate_size << dendl; | |
920 | in->truncate_size = truncate_size; | |
921 | in->oset.truncate_size = truncate_size; | |
922 | } else { | |
923 | ldout(cct, 0) << "Hmmm, truncate_seq && truncate_size changed on non-file inode!" << dendl; | |
924 | } | |
925 | } | |
1adf2230 AA |
926 | } |
927 | ||
928 | void Client::update_inode_file_time(Inode *in, int issued, uint64_t time_warp_seq, | |
929 | utime_t ctime, utime_t mtime, utime_t atime) | |
930 | { | |
931 | ldout(cct, 10) << __func__ << " " << *in << " " << ccap_string(issued) | |
932 | << " ctime " << ctime << " mtime " << mtime << dendl; | |
933 | ||
934 | if (time_warp_seq > in->time_warp_seq) | |
935 | ldout(cct, 10) << " mds time_warp_seq " << time_warp_seq | |
936 | << " is higher than local time_warp_seq " | |
937 | << in->time_warp_seq << dendl; | |
938 | ||
939 | int warn = false; | |
7c673cae FG |
940 | // be careful with size, mtime, atime |
941 | if (issued & (CEPH_CAP_FILE_EXCL| | |
942 | CEPH_CAP_FILE_WR| | |
943 | CEPH_CAP_FILE_BUFFER| | |
944 | CEPH_CAP_AUTH_EXCL| | |
945 | CEPH_CAP_XATTR_EXCL)) { | |
946 | ldout(cct, 30) << "Yay have enough caps to look at our times" << dendl; | |
947 | if (ctime > in->ctime) | |
948 | in->ctime = ctime; | |
949 | if (time_warp_seq > in->time_warp_seq) { | |
7c673cae FG |
950 | //the mds updated times, so take those! |
951 | in->mtime = mtime; | |
952 | in->atime = atime; | |
953 | in->time_warp_seq = time_warp_seq; | |
954 | } else if (time_warp_seq == in->time_warp_seq) { | |
955 | //take max times | |
956 | if (mtime > in->mtime) | |
957 | in->mtime = mtime; | |
958 | if (atime > in->atime) | |
959 | in->atime = atime; | |
960 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
961 | //ignore mds values as we have a higher seq | |
962 | } else warn = true; | |
963 | } else { | |
964 | ldout(cct, 30) << "Don't have enough caps, just taking mds' time values" << dendl; | |
965 | if (time_warp_seq >= in->time_warp_seq) { | |
966 | in->ctime = ctime; | |
967 | in->mtime = mtime; | |
968 | in->atime = atime; | |
969 | in->time_warp_seq = time_warp_seq; | |
970 | } else warn = true; | |
971 | } | |
972 | if (warn) { | |
973 | ldout(cct, 0) << "WARNING: " << *in << " mds time_warp_seq " | |
974 | << time_warp_seq << " is lower than local time_warp_seq " | |
975 | << in->time_warp_seq | |
976 | << dendl; | |
977 | } | |
978 | } | |
979 | ||
980 | void Client::_fragmap_remove_non_leaves(Inode *in) | |
981 | { | |
982 | for (map<frag_t,int>::iterator p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
983 | if (!in->dirfragtree.is_leaf(p->first)) | |
984 | in->fragmap.erase(p++); | |
985 | else | |
986 | ++p; | |
987 | } | |
988 | ||
989 | void Client::_fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds) | |
990 | { | |
991 | for (auto p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
992 | if (p->second == mds) | |
993 | in->fragmap.erase(p++); | |
994 | else | |
995 | ++p; | |
996 | } | |
997 | ||
998 | Inode * Client::add_update_inode(InodeStat *st, utime_t from, | |
999 | MetaSession *session, | |
1000 | const UserPerm& request_perms) | |
1001 | { | |
1002 | Inode *in; | |
1003 | bool was_new = false; | |
1004 | if (inode_map.count(st->vino)) { | |
1005 | in = inode_map[st->vino]; | |
11fdf7f2 | 1006 | ldout(cct, 12) << __func__ << " had " << *in << " caps " << ccap_string(st->cap.caps) << dendl; |
7c673cae FG |
1007 | } else { |
1008 | in = new Inode(this, st->vino, &st->layout); | |
1009 | inode_map[st->vino] = in; | |
1010 | ||
1011 | if (use_faked_inos()) | |
1012 | _assign_faked_ino(in); | |
1013 | ||
1014 | if (!root) { | |
1015 | root = in; | |
11fdf7f2 | 1016 | if (use_faked_inos()) |
b3b6e05e | 1017 | _assign_faked_root(root.get()); |
7c673cae FG |
1018 | root_ancestor = in; |
1019 | cwd = root; | |
f67539c2 | 1020 | } else if (is_mounting()) { |
7c673cae FG |
1021 | root_parents[root_ancestor] = in; |
1022 | root_ancestor = in; | |
1023 | } | |
1024 | ||
1025 | // immutable bits | |
1026 | in->ino = st->vino.ino; | |
1027 | in->snapid = st->vino.snapid; | |
1028 | in->mode = st->mode & S_IFMT; | |
1029 | was_new = true; | |
1030 | } | |
1031 | ||
1032 | in->rdev = st->rdev; | |
1033 | if (in->is_symlink()) | |
1034 | in->symlink = st->symlink; | |
1035 | ||
7c673cae | 1036 | // only update inode if mds info is strictly newer, or it is the same and projected (odd). |
1adf2230 AA |
1037 | bool new_version = false; |
1038 | if (in->version == 0 || | |
1039 | ((st->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
1040 | (in->version & ~1) < st->version)) | |
1041 | new_version = true; | |
7c673cae | 1042 | |
1adf2230 AA |
1043 | int issued; |
1044 | in->caps_issued(&issued); | |
1045 | issued |= in->caps_dirty(); | |
1046 | int new_issued = ~issued & (int)st->cap.caps; | |
7c673cae | 1047 | |
1adf2230 AA |
1048 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
1049 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
1050 | in->mode = st->mode; | |
1051 | in->uid = st->uid; | |
1052 | in->gid = st->gid; | |
1053 | in->btime = st->btime; | |
81eedcae | 1054 | in->snap_btime = st->snap_btime; |
f67539c2 | 1055 | in->snap_metadata = st->snap_metadata; |
1adf2230 | 1056 | } |
7c673cae | 1057 | |
1adf2230 AA |
1058 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
1059 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
1060 | in->nlink = st->nlink; | |
1061 | } | |
7c673cae | 1062 | |
1adf2230 AA |
1063 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
1064 | update_inode_file_time(in, issued, st->time_warp_seq, | |
1065 | st->ctime, st->mtime, st->atime); | |
1066 | } | |
7c673cae | 1067 | |
1adf2230 AA |
1068 | if (new_version || |
1069 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7c673cae | 1070 | in->layout = st->layout; |
1adf2230 AA |
1071 | update_inode_file_size(in, issued, st->size, st->truncate_seq, st->truncate_size); |
1072 | } | |
7c673cae | 1073 | |
1adf2230 AA |
1074 | if (in->is_dir()) { |
1075 | if (new_version || (new_issued & CEPH_CAP_FILE_SHARED)) { | |
1076 | in->dirstat = st->dirstat; | |
1077 | } | |
1078 | // dir_layout/rstat/quota are not tracked by capability, update them only if | |
1079 | // the inode stat is from auth mds | |
1080 | if (new_version || (st->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
7c673cae FG |
1081 | in->dir_layout = st->dir_layout; |
1082 | ldout(cct, 20) << " dir hash is " << (int)in->dir_layout.dl_dir_hash << dendl; | |
1adf2230 AA |
1083 | in->rstat = st->rstat; |
1084 | in->quota = st->quota; | |
11fdf7f2 | 1085 | in->dir_pin = st->dir_pin; |
1adf2230 AA |
1086 | } |
1087 | // move me if/when version reflects fragtree changes. | |
1088 | if (in->dirfragtree != st->dirfragtree) { | |
1089 | in->dirfragtree = st->dirfragtree; | |
1090 | _fragmap_remove_non_leaves(in); | |
7c673cae | 1091 | } |
7c673cae FG |
1092 | } |
1093 | ||
1094 | if ((in->xattr_version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && | |
1095 | st->xattrbl.length() && | |
1096 | st->xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
1097 | auto p = st->xattrbl.cbegin(); |
1098 | decode(in->xattrs, p); | |
7c673cae FG |
1099 | in->xattr_version = st->xattr_version; |
1100 | } | |
1101 | ||
1adf2230 AA |
1102 | if (st->inline_version > in->inline_version) { |
1103 | in->inline_data = st->inline_data; | |
1104 | in->inline_version = st->inline_version; | |
7c673cae FG |
1105 | } |
1106 | ||
1adf2230 AA |
1107 | /* always take a newer change attr */ |
1108 | if (st->change_attr > in->change_attr) | |
1109 | in->change_attr = st->change_attr; | |
1110 | ||
1111 | if (st->version > in->version) | |
1112 | in->version = st->version; | |
1113 | ||
1114 | if (was_new) | |
1115 | ldout(cct, 12) << __func__ << " adding " << *in << " caps " << ccap_string(st->cap.caps) << dendl; | |
1116 | ||
1117 | if (!st->cap.caps) | |
1118 | return in; // as with readdir returning indoes in different snaprealms (no caps!) | |
1119 | ||
7c673cae | 1120 | if (in->snapid == CEPH_NOSNAP) { |
a8e16298 TL |
1121 | add_update_cap(in, session, st->cap.cap_id, st->cap.caps, st->cap.wanted, |
1122 | st->cap.seq, st->cap.mseq, inodeno_t(st->cap.realm), | |
1123 | st->cap.flags, request_perms); | |
28e407b8 | 1124 | if (in->auth_cap && in->auth_cap->session == session) { |
7c673cae | 1125 | in->max_size = st->max_size; |
28e407b8 AA |
1126 | in->rstat = st->rstat; |
1127 | } | |
7c673cae | 1128 | |
1adf2230 AA |
1129 | // setting I_COMPLETE needs to happen after adding the cap |
1130 | if (in->is_dir() && | |
1131 | (st->cap.caps & CEPH_CAP_FILE_SHARED) && | |
1132 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
1133 | in->dirstat.nfiles == 0 && | |
1134 | in->dirstat.nsubdirs == 0) { | |
1135 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on empty dir " << *in << dendl; | |
1136 | in->flags |= I_COMPLETE | I_DIR_ORDERED; | |
1137 | if (in->dir) { | |
1138 | ldout(cct, 10) << " dir is open on empty dir " << in->ino << " with " | |
1139 | << in->dir->dentries.size() << " entries, marking all dentries null" << dendl; | |
1140 | in->dir->readdir_cache.clear(); | |
1141 | for (const auto& p : in->dir->dentries) { | |
1142 | unlink(p.second, true, true); // keep dir, keep dentry | |
1143 | } | |
1144 | if (in->dir->dentries.empty()) | |
1145 | close_dir(in->dir); | |
7c673cae | 1146 | } |
7c673cae | 1147 | } |
1adf2230 AA |
1148 | } else { |
1149 | in->snap_caps |= st->cap.caps; | |
7c673cae FG |
1150 | } |
1151 | ||
f67539c2 | 1152 | in->fscrypt = st->fscrypt; |
7c673cae FG |
1153 | return in; |
1154 | } | |
1155 | ||
1156 | ||
1157 | /* | |
1158 | * insert_dentry_inode - insert + link a single dentry + inode into the metadata cache. | |
1159 | */ | |
1160 | Dentry *Client::insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease, | |
1161 | Inode *in, utime_t from, MetaSession *session, | |
1162 | Dentry *old_dentry) | |
1163 | { | |
1164 | Dentry *dn = NULL; | |
1165 | if (dir->dentries.count(dname)) | |
1166 | dn = dir->dentries[dname]; | |
1167 | ||
11fdf7f2 | 1168 | ldout(cct, 12) << __func__ << " '" << dname << "' vino " << in->vino() |
7c673cae FG |
1169 | << " in dir " << dir->parent_inode->vino() << " dn " << dn |
1170 | << dendl; | |
1171 | ||
1172 | if (dn && dn->inode) { | |
1173 | if (dn->inode->vino() == in->vino()) { | |
1174 | touch_dn(dn); | |
1175 | ldout(cct, 12) << " had dentry " << dname | |
1176 | << " with correct vino " << dn->inode->vino() | |
1177 | << dendl; | |
1178 | } else { | |
1179 | ldout(cct, 12) << " had dentry " << dname | |
1180 | << " with WRONG vino " << dn->inode->vino() | |
1181 | << dendl; | |
1182 | unlink(dn, true, true); // keep dir, keep dentry | |
1183 | } | |
1184 | } | |
1185 | ||
1186 | if (!dn || !dn->inode) { | |
1187 | InodeRef tmp_ref(in); | |
1188 | if (old_dentry) { | |
1189 | if (old_dentry->dir != dir) { | |
1190 | Inode *old_diri = old_dentry->dir->parent_inode; | |
7c673cae FG |
1191 | clear_dir_complete_and_ordered(old_diri, false); |
1192 | } | |
1193 | unlink(old_dentry, dir == old_dentry->dir, false); // drop dentry, keep dir open if its the same dir | |
1194 | } | |
1195 | Inode *diri = dir->parent_inode; | |
7c673cae FG |
1196 | clear_dir_complete_and_ordered(diri, false); |
1197 | dn = link(dir, dname, in, dn); | |
1198 | } | |
1199 | ||
1200 | update_dentry_lease(dn, dlease, from, session); | |
1201 | return dn; | |
1202 | } | |
1203 | ||
1204 | void Client::update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session) | |
1205 | { | |
1206 | utime_t dttl = from; | |
1207 | dttl += (float)dlease->duration_ms / 1000.0; | |
f67539c2 TL |
1208 | |
1209 | ldout(cct, 15) << __func__ << " " << *dn << " " << *dlease << " from " << from << dendl; | |
7c673cae | 1210 | |
11fdf7f2 | 1211 | ceph_assert(dn); |
7c673cae | 1212 | |
9f95a23c | 1213 | if (dlease->mask & CEPH_LEASE_VALID) { |
7c673cae FG |
1214 | if (dttl > dn->lease_ttl) { |
1215 | ldout(cct, 10) << "got dentry lease on " << dn->name | |
1216 | << " dur " << dlease->duration_ms << "ms ttl " << dttl << dendl; | |
1217 | dn->lease_ttl = dttl; | |
1218 | dn->lease_mds = session->mds_num; | |
1219 | dn->lease_seq = dlease->seq; | |
1220 | dn->lease_gen = session->cap_gen; | |
1221 | } | |
1222 | } | |
1223 | dn->cap_shared_gen = dn->dir->parent_inode->shared_gen; | |
f91f0fd5 TL |
1224 | if (dlease->mask & CEPH_LEASE_PRIMARY_LINK) |
1225 | dn->mark_primary(); | |
f67539c2 | 1226 | dn->alternate_name = std::move(dlease->alternate_name); |
7c673cae FG |
1227 | } |
1228 | ||
1229 | ||
1230 | /* | |
1231 | * update MDS location cache for a single inode | |
1232 | */ | |
522d829b | 1233 | void Client::update_dir_dist(Inode *in, DirStat *dst, mds_rank_t from) |
7c673cae FG |
1234 | { |
1235 | // auth | |
1236 | ldout(cct, 20) << "got dirfrag map for " << in->ino << " frag " << dst->frag << " to mds " << dst->auth << dendl; | |
1237 | if (dst->auth >= 0) { | |
1238 | in->fragmap[dst->frag] = dst->auth; | |
1239 | } else { | |
1240 | in->fragmap.erase(dst->frag); | |
1241 | } | |
1242 | if (!in->dirfragtree.is_leaf(dst->frag)) { | |
1243 | in->dirfragtree.force_to_leaf(cct, dst->frag); | |
1244 | _fragmap_remove_non_leaves(in); | |
1245 | } | |
1246 | ||
522d829b TL |
1247 | // replicated, only update from auth mds reply |
1248 | if (from == dst->auth) { | |
1249 | in->dir_replicated = !dst->dist.empty(); | |
1250 | if (!dst->dist.empty()) | |
1251 | in->frag_repmap[dst->frag].assign(dst->dist.begin(), dst->dist.end()) ; | |
1252 | else | |
1253 | in->frag_repmap.erase(dst->frag); | |
1254 | } | |
7c673cae FG |
1255 | } |
1256 | ||
1257 | void Client::clear_dir_complete_and_ordered(Inode *diri, bool complete) | |
1258 | { | |
f91f0fd5 TL |
1259 | if (complete) |
1260 | diri->dir_release_count++; | |
1261 | else | |
1262 | diri->dir_ordered_count++; | |
7c673cae FG |
1263 | if (diri->flags & I_COMPLETE) { |
1264 | if (complete) { | |
1265 | ldout(cct, 10) << " clearing (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
1266 | diri->flags &= ~(I_COMPLETE | I_DIR_ORDERED); | |
1267 | } else { | |
1268 | if (diri->flags & I_DIR_ORDERED) { | |
1269 | ldout(cct, 10) << " clearing I_DIR_ORDERED on " << *diri << dendl; | |
1270 | diri->flags &= ~I_DIR_ORDERED; | |
1271 | } | |
1272 | } | |
1273 | if (diri->dir) | |
1274 | diri->dir->readdir_cache.clear(); | |
1275 | } | |
1276 | } | |
1277 | ||
1278 | /* | |
1279 | * insert results from readdir or lssnap into the metadata cache. | |
1280 | */ | |
1281 | void Client::insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri) { | |
1282 | ||
11fdf7f2 | 1283 | auto& reply = request->reply; |
7c673cae | 1284 | ConnectionRef con = request->reply->get_connection(); |
11fdf7f2 TL |
1285 | uint64_t features; |
1286 | if(session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1287 | features = (uint64_t)-1; | |
1288 | } | |
1289 | else { | |
1290 | features = con->get_features(); | |
1291 | } | |
7c673cae FG |
1292 | |
1293 | dir_result_t *dirp = request->dirp; | |
11fdf7f2 | 1294 | ceph_assert(dirp); |
7c673cae FG |
1295 | |
1296 | // the extra buffer list is only set for readdir and lssnap replies | |
11fdf7f2 | 1297 | auto p = reply->get_extra_bl().cbegin(); |
7c673cae FG |
1298 | if (!p.end()) { |
1299 | // snapdir? | |
1300 | if (request->head.op == CEPH_MDS_OP_LSSNAP) { | |
11fdf7f2 | 1301 | ceph_assert(diri); |
7c673cae FG |
1302 | diri = open_snapdir(diri); |
1303 | } | |
1304 | ||
1305 | // only open dir if we're actually adding stuff to it! | |
1306 | Dir *dir = diri->open_dir(); | |
11fdf7f2 | 1307 | ceph_assert(dir); |
7c673cae FG |
1308 | |
1309 | // dirstat | |
11fdf7f2 | 1310 | DirStat dst(p, features); |
7c673cae FG |
1311 | __u32 numdn; |
1312 | __u16 flags; | |
11fdf7f2 TL |
1313 | decode(numdn, p); |
1314 | decode(flags, p); | |
7c673cae FG |
1315 | |
1316 | bool end = ((unsigned)flags & CEPH_READDIR_FRAG_END); | |
1317 | bool hash_order = ((unsigned)flags & CEPH_READDIR_HASH_ORDER); | |
1318 | ||
1319 | frag_t fg = (unsigned)request->head.args.readdir.frag; | |
1320 | unsigned readdir_offset = dirp->next_offset; | |
1321 | string readdir_start = dirp->last_name; | |
11fdf7f2 | 1322 | ceph_assert(!readdir_start.empty() || readdir_offset == 2); |
7c673cae FG |
1323 | |
1324 | unsigned last_hash = 0; | |
1325 | if (hash_order) { | |
1326 | if (!readdir_start.empty()) { | |
1327 | last_hash = ceph_frag_value(diri->hash_dentry_name(readdir_start)); | |
1328 | } else if (flags & CEPH_READDIR_OFFSET_HASH) { | |
1329 | /* mds understands offset_hash */ | |
1330 | last_hash = (unsigned)request->head.args.readdir.offset_hash; | |
1331 | } | |
1332 | } | |
1333 | ||
1334 | if (fg != dst.frag) { | |
1335 | ldout(cct, 10) << "insert_trace got new frag " << fg << " -> " << dst.frag << dendl; | |
1336 | fg = dst.frag; | |
1337 | if (!hash_order) { | |
1338 | readdir_offset = 2; | |
1339 | readdir_start.clear(); | |
1340 | dirp->offset = dir_result_t::make_fpos(fg, readdir_offset, false); | |
1341 | } | |
1342 | } | |
1343 | ||
1344 | ldout(cct, 10) << __func__ << " " << numdn << " readdir items, end=" << end | |
1345 | << ", hash_order=" << hash_order | |
1346 | << ", readdir_start " << readdir_start | |
1347 | << ", last_hash " << last_hash | |
1348 | << ", next_offset " << readdir_offset << dendl; | |
1349 | ||
1350 | if (diri->snapid != CEPH_SNAPDIR && | |
1351 | fg.is_leftmost() && readdir_offset == 2 && | |
1352 | !(hash_order && last_hash)) { | |
1353 | dirp->release_count = diri->dir_release_count; | |
1354 | dirp->ordered_count = diri->dir_ordered_count; | |
1355 | dirp->start_shared_gen = diri->shared_gen; | |
1356 | dirp->cache_index = 0; | |
1357 | } | |
1358 | ||
1359 | dirp->buffer_frag = fg; | |
1360 | ||
1361 | _readdir_drop_dirp_buffer(dirp); | |
1362 | dirp->buffer.reserve(numdn); | |
1363 | ||
1364 | string dname; | |
1365 | LeaseStat dlease; | |
1366 | for (unsigned i=0; i<numdn; i++) { | |
11fdf7f2 TL |
1367 | decode(dname, p); |
1368 | dlease.decode(p, features); | |
7c673cae FG |
1369 | InodeStat ist(p, features); |
1370 | ||
1371 | ldout(cct, 15) << "" << i << ": '" << dname << "'" << dendl; | |
1372 | ||
1373 | Inode *in = add_update_inode(&ist, request->sent_stamp, session, | |
1374 | request->perms); | |
1375 | Dentry *dn; | |
1376 | if (diri->dir->dentries.count(dname)) { | |
1377 | Dentry *olddn = diri->dir->dentries[dname]; | |
1378 | if (olddn->inode != in) { | |
1379 | // replace incorrect dentry | |
1380 | unlink(olddn, true, true); // keep dir, dentry | |
1381 | dn = link(dir, dname, in, olddn); | |
11fdf7f2 | 1382 | ceph_assert(dn == olddn); |
7c673cae FG |
1383 | } else { |
1384 | // keep existing dn | |
1385 | dn = olddn; | |
1386 | touch_dn(dn); | |
1387 | } | |
1388 | } else { | |
1389 | // new dn | |
1390 | dn = link(dir, dname, in, NULL); | |
1391 | } | |
f67539c2 | 1392 | dn->alternate_name = std::move(dlease.alternate_name); |
7c673cae FG |
1393 | |
1394 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1395 | if (hash_order) { | |
1396 | unsigned hash = ceph_frag_value(diri->hash_dentry_name(dname)); | |
1397 | if (hash != last_hash) | |
1398 | readdir_offset = 2; | |
1399 | last_hash = hash; | |
1400 | dn->offset = dir_result_t::make_fpos(hash, readdir_offset++, true); | |
1401 | } else { | |
1402 | dn->offset = dir_result_t::make_fpos(fg, readdir_offset++, false); | |
1403 | } | |
1404 | // add to readdir cache | |
1405 | if (dirp->release_count == diri->dir_release_count && | |
1406 | dirp->ordered_count == diri->dir_ordered_count && | |
1407 | dirp->start_shared_gen == diri->shared_gen) { | |
1408 | if (dirp->cache_index == dir->readdir_cache.size()) { | |
1409 | if (i == 0) { | |
11fdf7f2 | 1410 | ceph_assert(!dirp->inode->is_complete_and_ordered()); |
7c673cae FG |
1411 | dir->readdir_cache.reserve(dirp->cache_index + numdn); |
1412 | } | |
1413 | dir->readdir_cache.push_back(dn); | |
1414 | } else if (dirp->cache_index < dir->readdir_cache.size()) { | |
1415 | if (dirp->inode->is_complete_and_ordered()) | |
11fdf7f2 | 1416 | ceph_assert(dir->readdir_cache[dirp->cache_index] == dn); |
7c673cae FG |
1417 | else |
1418 | dir->readdir_cache[dirp->cache_index] = dn; | |
1419 | } else { | |
11fdf7f2 | 1420 | ceph_abort_msg("unexpected readdir buffer idx"); |
7c673cae FG |
1421 | } |
1422 | dirp->cache_index++; | |
1423 | } | |
1424 | // add to cached result list | |
f67539c2 | 1425 | dirp->buffer.push_back(dir_result_t::dentry(dn->offset, dname, dn->alternate_name, in)); |
7c673cae FG |
1426 | ldout(cct, 15) << __func__ << " " << hex << dn->offset << dec << ": '" << dname << "' -> " << in->ino << dendl; |
1427 | } | |
1428 | ||
1429 | if (numdn > 0) | |
1430 | dirp->last_name = dname; | |
1431 | if (end) | |
1432 | dirp->next_offset = 2; | |
1433 | else | |
1434 | dirp->next_offset = readdir_offset; | |
1435 | ||
1436 | if (dir->is_empty()) | |
1437 | close_dir(dir); | |
1438 | } | |
1439 | } | |
1440 | ||
1441 | /** insert_trace | |
1442 | * | |
1443 | * insert a trace from a MDS reply into the cache. | |
1444 | */ | |
1445 | Inode* Client::insert_trace(MetaRequest *request, MetaSession *session) | |
1446 | { | |
11fdf7f2 | 1447 | auto& reply = request->reply; |
7c673cae FG |
1448 | int op = request->get_op(); |
1449 | ||
1450 | ldout(cct, 10) << "insert_trace from " << request->sent_stamp << " mds." << session->mds_num | |
1451 | << " is_target=" << (int)reply->head.is_target | |
1452 | << " is_dentry=" << (int)reply->head.is_dentry | |
1453 | << dendl; | |
1454 | ||
11fdf7f2 | 1455 | auto p = reply->get_trace_bl().cbegin(); |
7c673cae FG |
1456 | if (request->got_unsafe) { |
1457 | ldout(cct, 10) << "insert_trace -- already got unsafe; ignoring" << dendl; | |
11fdf7f2 | 1458 | ceph_assert(p.end()); |
7c673cae FG |
1459 | return NULL; |
1460 | } | |
1461 | ||
1462 | if (p.end()) { | |
1463 | ldout(cct, 10) << "insert_trace -- no trace" << dendl; | |
1464 | ||
1465 | Dentry *d = request->dentry(); | |
1466 | if (d) { | |
1467 | Inode *diri = d->dir->parent_inode; | |
7c673cae FG |
1468 | clear_dir_complete_and_ordered(diri, true); |
1469 | } | |
1470 | ||
1471 | if (d && reply->get_result() == 0) { | |
1472 | if (op == CEPH_MDS_OP_RENAME) { | |
1473 | // rename | |
1474 | Dentry *od = request->old_dentry(); | |
1475 | ldout(cct, 10) << " unlinking rename src dn " << od << " for traceless reply" << dendl; | |
11fdf7f2 | 1476 | ceph_assert(od); |
7c673cae FG |
1477 | unlink(od, true, true); // keep dir, dentry |
1478 | } else if (op == CEPH_MDS_OP_RMDIR || | |
1479 | op == CEPH_MDS_OP_UNLINK) { | |
1480 | // unlink, rmdir | |
1481 | ldout(cct, 10) << " unlinking unlink/rmdir dn " << d << " for traceless reply" << dendl; | |
1482 | unlink(d, true, true); // keep dir, dentry | |
1483 | } | |
1484 | } | |
1485 | return NULL; | |
1486 | } | |
1487 | ||
1488 | ConnectionRef con = request->reply->get_connection(); | |
11fdf7f2 TL |
1489 | uint64_t features; |
1490 | if (session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1491 | features = (uint64_t)-1; | |
1492 | } | |
1493 | else { | |
1494 | features = con->get_features(); | |
1495 | } | |
7c673cae FG |
1496 | ldout(cct, 10) << " features 0x" << hex << features << dec << dendl; |
1497 | ||
1498 | // snap trace | |
1499 | SnapRealm *realm = NULL; | |
1500 | if (reply->snapbl.length()) | |
1501 | update_snap_trace(reply->snapbl, &realm); | |
1502 | ||
1503 | ldout(cct, 10) << " hrm " | |
1504 | << " is_target=" << (int)reply->head.is_target | |
1505 | << " is_dentry=" << (int)reply->head.is_dentry | |
1506 | << dendl; | |
1507 | ||
1508 | InodeStat dirst; | |
1509 | DirStat dst; | |
1510 | string dname; | |
1511 | LeaseStat dlease; | |
1512 | InodeStat ist; | |
1513 | ||
1514 | if (reply->head.is_dentry) { | |
1515 | dirst.decode(p, features); | |
11fdf7f2 TL |
1516 | dst.decode(p, features); |
1517 | decode(dname, p); | |
1518 | dlease.decode(p, features); | |
7c673cae FG |
1519 | } |
1520 | ||
1521 | Inode *in = 0; | |
1522 | if (reply->head.is_target) { | |
1523 | ist.decode(p, features); | |
1524 | if (cct->_conf->client_debug_getattr_caps) { | |
1525 | unsigned wanted = 0; | |
1526 | if (op == CEPH_MDS_OP_GETATTR || op == CEPH_MDS_OP_LOOKUP) | |
1527 | wanted = request->head.args.getattr.mask; | |
1528 | else if (op == CEPH_MDS_OP_OPEN || op == CEPH_MDS_OP_CREATE) | |
1529 | wanted = request->head.args.open.mask; | |
1530 | ||
1531 | if ((wanted & CEPH_CAP_XATTR_SHARED) && | |
1532 | !(ist.xattr_version > 0 && ist.xattrbl.length() > 0)) | |
11fdf7f2 | 1533 | ceph_abort_msg("MDS reply does not contain xattrs"); |
7c673cae FG |
1534 | } |
1535 | ||
1536 | in = add_update_inode(&ist, request->sent_stamp, session, | |
1537 | request->perms); | |
1538 | } | |
1539 | ||
1540 | Inode *diri = NULL; | |
1541 | if (reply->head.is_dentry) { | |
1542 | diri = add_update_inode(&dirst, request->sent_stamp, session, | |
1543 | request->perms); | |
522d829b TL |
1544 | mds_rank_t from_mds = mds_rank_t(reply->get_source().num()); |
1545 | update_dir_dist(diri, &dst, from_mds); // dir stat info is attached to .. | |
7c673cae FG |
1546 | |
1547 | if (in) { | |
1548 | Dir *dir = diri->open_dir(); | |
1549 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session, | |
1550 | (op == CEPH_MDS_OP_RENAME) ? request->old_dentry() : NULL); | |
1551 | } else { | |
1552 | Dentry *dn = NULL; | |
1553 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1554 | dn = diri->dir->dentries[dname]; | |
1555 | if (dn->inode) { | |
7c673cae FG |
1556 | clear_dir_complete_and_ordered(diri, false); |
1557 | unlink(dn, true, true); // keep dir, dentry | |
1558 | } | |
1559 | } | |
1560 | if (dlease.duration_ms > 0) { | |
1561 | if (!dn) { | |
1562 | Dir *dir = diri->open_dir(); | |
1563 | dn = link(dir, dname, NULL, NULL); | |
1564 | } | |
1565 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1566 | } | |
1567 | } | |
1568 | } else if (op == CEPH_MDS_OP_LOOKUPSNAP || | |
1569 | op == CEPH_MDS_OP_MKSNAP) { | |
1570 | ldout(cct, 10) << " faking snap lookup weirdness" << dendl; | |
1571 | // fake it for snap lookup | |
1572 | vinodeno_t vino = ist.vino; | |
1573 | vino.snapid = CEPH_SNAPDIR; | |
11fdf7f2 | 1574 | ceph_assert(inode_map.count(vino)); |
7c673cae FG |
1575 | diri = inode_map[vino]; |
1576 | ||
1577 | string dname = request->path.last_dentry(); | |
1578 | ||
1579 | LeaseStat dlease; | |
1580 | dlease.duration_ms = 0; | |
1581 | ||
1582 | if (in) { | |
1583 | Dir *dir = diri->open_dir(); | |
1584 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session); | |
1585 | } else { | |
1586 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1587 | Dentry *dn = diri->dir->dentries[dname]; | |
1588 | if (dn->inode) | |
1589 | unlink(dn, true, true); // keep dir, dentry | |
1590 | } | |
1591 | } | |
1592 | } | |
1593 | ||
1594 | if (in) { | |
1595 | if (op == CEPH_MDS_OP_READDIR || | |
1596 | op == CEPH_MDS_OP_LSSNAP) { | |
1597 | insert_readdir_results(request, session, in); | |
1598 | } else if (op == CEPH_MDS_OP_LOOKUPNAME) { | |
1599 | // hack: return parent inode instead | |
1600 | in = diri; | |
1601 | } | |
1602 | ||
1603 | if (request->dentry() == NULL && in != request->inode()) { | |
1604 | // pin the target inode if its parent dentry is not pinned | |
1605 | request->set_other_inode(in); | |
1606 | } | |
1607 | } | |
1608 | ||
1609 | if (realm) | |
1610 | put_snap_realm(realm); | |
1611 | ||
1612 | request->target = in; | |
1613 | return in; | |
1614 | } | |
1615 | ||
1616 | // ------- | |
1617 | ||
1618 | mds_rank_t Client::choose_target_mds(MetaRequest *req, Inode** phash_diri) | |
1619 | { | |
1620 | mds_rank_t mds = MDS_RANK_NONE; | |
1621 | __u32 hash = 0; | |
1622 | bool is_hash = false; | |
2a845540 | 1623 | int issued = 0; |
7c673cae FG |
1624 | |
1625 | Inode *in = NULL; | |
1626 | Dentry *de = NULL; | |
7c673cae FG |
1627 | |
1628 | if (req->resend_mds >= 0) { | |
1629 | mds = req->resend_mds; | |
1630 | req->resend_mds = -1; | |
11fdf7f2 | 1631 | ldout(cct, 10) << __func__ << " resend_mds specified as mds." << mds << dendl; |
7c673cae FG |
1632 | goto out; |
1633 | } | |
1634 | ||
1635 | if (cct->_conf->client_use_random_mds) | |
1636 | goto random_mds; | |
1637 | ||
1638 | in = req->inode(); | |
1639 | de = req->dentry(); | |
1640 | if (in) { | |
11fdf7f2 | 1641 | ldout(cct, 20) << __func__ << " starting with req->inode " << *in << dendl; |
7c673cae FG |
1642 | if (req->path.depth()) { |
1643 | hash = in->hash_dentry_name(req->path[0]); | |
11fdf7f2 | 1644 | ldout(cct, 20) << __func__ << " inode dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1645 | << " on " << req->path[0] |
1646 | << " => " << hash << dendl; | |
1647 | is_hash = true; | |
1648 | } | |
1649 | } else if (de) { | |
1650 | if (de->inode) { | |
1651 | in = de->inode.get(); | |
11fdf7f2 | 1652 | ldout(cct, 20) << __func__ << " starting with req->dentry inode " << *in << dendl; |
7c673cae FG |
1653 | } else { |
1654 | in = de->dir->parent_inode; | |
1655 | hash = in->hash_dentry_name(de->name); | |
11fdf7f2 | 1656 | ldout(cct, 20) << __func__ << " dentry dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1657 | << " on " << de->name |
1658 | << " => " << hash << dendl; | |
1659 | is_hash = true; | |
1660 | } | |
1661 | } | |
1662 | if (in) { | |
1663 | if (in->snapid != CEPH_NOSNAP) { | |
11fdf7f2 | 1664 | ldout(cct, 10) << __func__ << " " << *in << " is snapped, using nonsnap parent" << dendl; |
7c673cae FG |
1665 | while (in->snapid != CEPH_NOSNAP) { |
1666 | if (in->snapid == CEPH_SNAPDIR) | |
1667 | in = in->snapdir_parent.get(); | |
11fdf7f2 | 1668 | else if (!in->dentries.empty()) |
7c673cae FG |
1669 | /* In most cases there will only be one dentry, so getting it |
1670 | * will be the correct action. If there are multiple hard links, | |
1671 | * I think the MDS should be able to redirect as needed*/ | |
1672 | in = in->get_first_parent()->dir->parent_inode; | |
1673 | else { | |
1674 | ldout(cct, 10) << "got unlinked inode, can't look at parent" << dendl; | |
1675 | break; | |
1676 | } | |
1677 | } | |
1678 | is_hash = false; | |
1679 | } | |
1680 | ||
11fdf7f2 | 1681 | ldout(cct, 20) << __func__ << " " << *in << " is_hash=" << is_hash |
7c673cae FG |
1682 | << " hash=" << hash << dendl; |
1683 | ||
2a845540 TL |
1684 | if (req->get_op() == CEPH_MDS_OP_GETATTR) |
1685 | issued = req->inode()->caps_issued(); | |
1686 | ||
f67539c2 | 1687 | if (is_hash && S_ISDIR(in->mode) && (!in->fragmap.empty() || !in->frag_repmap.empty())) { |
7c673cae | 1688 | frag_t fg = in->dirfragtree[hash]; |
2a845540 | 1689 | if (!req->auth_is_best(issued)) { |
f67539c2 TL |
1690 | auto repmapit = in->frag_repmap.find(fg); |
1691 | if (repmapit != in->frag_repmap.end()) { | |
1692 | auto& repmap = repmapit->second; | |
1693 | auto r = ceph::util::generate_random_number<uint64_t>(0, repmap.size()-1); | |
1694 | mds = repmap.at(r); | |
1695 | } | |
1696 | } else if (in->fragmap.count(fg)) { | |
7c673cae FG |
1697 | mds = in->fragmap[fg]; |
1698 | if (phash_diri) | |
1699 | *phash_diri = in; | |
91327a77 | 1700 | } else if (in->auth_cap) { |
f67539c2 | 1701 | req->send_to_auth = true; |
91327a77 AA |
1702 | mds = in->auth_cap->session->mds_num; |
1703 | } | |
1704 | if (mds >= 0) { | |
11fdf7f2 | 1705 | ldout(cct, 10) << __func__ << " from dirfragtree hash" << dendl; |
7c673cae FG |
1706 | goto out; |
1707 | } | |
1708 | } | |
1709 | ||
2a845540 | 1710 | if (in->auth_cap && req->auth_is_best(issued)) { |
11fdf7f2 TL |
1711 | mds = in->auth_cap->session->mds_num; |
1712 | } else if (!in->caps.empty()) { | |
1713 | mds = in->caps.begin()->second.session->mds_num; | |
1714 | } else { | |
7c673cae | 1715 | goto random_mds; |
11fdf7f2 TL |
1716 | } |
1717 | ldout(cct, 10) << __func__ << " from caps on inode " << *in << dendl; | |
7c673cae FG |
1718 | |
1719 | goto out; | |
1720 | } | |
1721 | ||
1722 | random_mds: | |
1723 | if (mds < 0) { | |
1724 | mds = _get_random_up_mds(); | |
1725 | ldout(cct, 10) << "did not get mds through better means, so chose random mds " << mds << dendl; | |
1726 | } | |
1727 | ||
1728 | out: | |
1729 | ldout(cct, 20) << "mds is " << mds << dendl; | |
1730 | return mds; | |
1731 | } | |
1732 | ||
7c673cae FG |
1733 | void Client::connect_mds_targets(mds_rank_t mds) |
1734 | { | |
11fdf7f2 TL |
1735 | ldout(cct, 10) << __func__ << " for mds." << mds << dendl; |
1736 | ceph_assert(mds_sessions.count(mds)); | |
7c673cae | 1737 | const MDSMap::mds_info_t& info = mdsmap->get_mds_info(mds); |
f67539c2 TL |
1738 | for (const auto &rank : info.export_targets) { |
1739 | if (mds_sessions.count(rank) == 0 && | |
1740 | mdsmap->is_clientreplay_or_active_or_stopping(rank)) { | |
7c673cae | 1741 | ldout(cct, 10) << "check_mds_sessions opening mds." << mds |
f67539c2 TL |
1742 | << " export target mds." << rank << dendl; |
1743 | _open_mds_session(rank); | |
7c673cae FG |
1744 | } |
1745 | } | |
1746 | } | |
1747 | ||
adb31ebb | 1748 | void Client::dump_mds_sessions(Formatter *f, bool cap_dump) |
7c673cae FG |
1749 | { |
1750 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 1751 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 AA |
1752 | f->dump_object("inst", inst); |
1753 | f->dump_stream("inst_str") << inst; | |
1754 | f->dump_stream("addr_str") << inst.addr; | |
7c673cae | 1755 | f->open_array_section("sessions"); |
11fdf7f2 | 1756 | for (const auto &p : mds_sessions) { |
7c673cae | 1757 | f->open_object_section("session"); |
20effc67 | 1758 | p.second->dump(f, cap_dump); |
7c673cae FG |
1759 | f->close_section(); |
1760 | } | |
1761 | f->close_section(); | |
1762 | f->dump_int("mdsmap_epoch", mdsmap->get_epoch()); | |
1763 | } | |
f67539c2 | 1764 | |
7c673cae FG |
1765 | void Client::dump_mds_requests(Formatter *f) |
1766 | { | |
1767 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
1768 | p != mds_requests.end(); | |
1769 | ++p) { | |
1770 | f->open_object_section("request"); | |
1771 | p->second->dump(f); | |
1772 | f->close_section(); | |
1773 | } | |
1774 | } | |
1775 | ||
9f95a23c | 1776 | int Client::verify_reply_trace(int r, MetaSession *session, |
11fdf7f2 | 1777 | MetaRequest *request, const MConstRef<MClientReply>& reply, |
7c673cae FG |
1778 | InodeRef *ptarget, bool *pcreated, |
1779 | const UserPerm& perms) | |
1780 | { | |
1781 | // check whether this request actually did the create, and set created flag | |
1782 | bufferlist extra_bl; | |
1783 | inodeno_t created_ino; | |
1784 | bool got_created_ino = false; | |
1785 | ceph::unordered_map<vinodeno_t, Inode*>::iterator p; | |
1786 | ||
11fdf7f2 | 1787 | extra_bl = reply->get_extra_bl(); |
7c673cae | 1788 | if (extra_bl.length() >= 8) { |
9f95a23c TL |
1789 | if (session->mds_features.test(CEPHFS_FEATURE_DELEG_INO)) { |
1790 | struct openc_response_t ocres; | |
1791 | ||
1792 | decode(ocres, extra_bl); | |
1793 | created_ino = ocres.created_ino; | |
1794 | /* | |
1795 | * The userland cephfs client doesn't have a way to do an async create | |
1796 | * (yet), so just discard delegated_inos for now. Eventually we should | |
1797 | * store them and use them in create calls, even if they are synchronous, | |
1798 | * if only for testing purposes. | |
1799 | */ | |
1800 | ldout(cct, 10) << "delegated_inos: " << ocres.delegated_inos << dendl; | |
1801 | } else { | |
1802 | // u64 containing number of created ino | |
1803 | decode(created_ino, extra_bl); | |
1804 | } | |
7c673cae | 1805 | ldout(cct, 10) << "make_request created ino " << created_ino << dendl; |
9f95a23c | 1806 | got_created_ino = true; |
7c673cae FG |
1807 | } |
1808 | ||
1809 | if (pcreated) | |
1810 | *pcreated = got_created_ino; | |
1811 | ||
1812 | if (request->target) { | |
1813 | *ptarget = request->target; | |
1814 | ldout(cct, 20) << "make_request target is " << *ptarget->get() << dendl; | |
1815 | } else { | |
1816 | if (got_created_ino && (p = inode_map.find(vinodeno_t(created_ino, CEPH_NOSNAP))) != inode_map.end()) { | |
1817 | (*ptarget) = p->second; | |
1818 | ldout(cct, 20) << "make_request created, target is " << *ptarget->get() << dendl; | |
1819 | } else { | |
1820 | // we got a traceless reply, and need to look up what we just | |
1821 | // created. for now, do this by name. someday, do this by the | |
1822 | // ino... which we know! FIXME. | |
1823 | InodeRef target; | |
1824 | Dentry *d = request->dentry(); | |
1825 | if (d) { | |
1826 | if (d->dir) { | |
1827 | ldout(cct, 10) << "make_request got traceless reply, looking up #" | |
1828 | << d->dir->parent_inode->ino << "/" << d->name | |
1829 | << " got_ino " << got_created_ino | |
1830 | << " ino " << created_ino | |
1831 | << dendl; | |
1832 | r = _do_lookup(d->dir->parent_inode, d->name, request->regetattr_mask, | |
1833 | &target, perms); | |
1834 | } else { | |
1835 | // if the dentry is not linked, just do our best. see #5021. | |
11fdf7f2 | 1836 | ceph_abort_msg("how did this happen? i want logs!"); |
7c673cae FG |
1837 | } |
1838 | } else { | |
1839 | Inode *in = request->inode(); | |
1840 | ldout(cct, 10) << "make_request got traceless reply, forcing getattr on #" | |
1841 | << in->ino << dendl; | |
1842 | r = _getattr(in, request->regetattr_mask, perms, true); | |
1843 | target = in; | |
1844 | } | |
1845 | if (r >= 0) { | |
1846 | // verify ino returned in reply and trace_dist are the same | |
1847 | if (got_created_ino && | |
1848 | created_ino.val != target->ino.val) { | |
1849 | ldout(cct, 5) << "create got ino " << created_ino << " but then failed on lookup; EINTR?" << dendl; | |
f67539c2 | 1850 | r = -CEPHFS_EINTR; |
7c673cae FG |
1851 | } |
1852 | if (ptarget) | |
1853 | ptarget->swap(target); | |
1854 | } | |
1855 | } | |
1856 | } | |
1857 | ||
1858 | return r; | |
1859 | } | |
1860 | ||
1861 | ||
1862 | /** | |
1863 | * make a request | |
1864 | * | |
1865 | * Blocking helper to make an MDS request. | |
1866 | * | |
1867 | * If the ptarget flag is set, behavior changes slightly: the caller | |
1868 | * expects to get a pointer to the inode we are creating or operating | |
1869 | * on. As a result, we will follow up any traceless mutation reply | |
1870 | * with a getattr or lookup to transparently handle a traceless reply | |
1871 | * from the MDS (as when the MDS restarts and the client has to replay | |
1872 | * a request). | |
1873 | * | |
1874 | * @param request the MetaRequest to execute | |
1875 | * @param perms The user uid/gid to execute as (eventually, full group lists?) | |
1876 | * @param ptarget [optional] address to store a pointer to the target inode we want to create or operate on | |
1877 | * @param pcreated [optional; required if ptarget] where to store a bool of whether our create atomically created a file | |
1878 | * @param use_mds [optional] prefer a specific mds (-1 for default) | |
1879 | * @param pdirbl [optional; disallowed if ptarget] where to pass extra reply payload to the caller | |
1880 | */ | |
1881 | int Client::make_request(MetaRequest *request, | |
1882 | const UserPerm& perms, | |
1883 | InodeRef *ptarget, bool *pcreated, | |
1884 | mds_rank_t use_mds, | |
1885 | bufferlist *pdirbl) | |
1886 | { | |
1887 | int r = 0; | |
1888 | ||
1889 | // assign a unique tid | |
1890 | ceph_tid_t tid = ++last_tid; | |
1891 | request->set_tid(tid); | |
1892 | ||
1893 | // and timestamp | |
1894 | request->op_stamp = ceph_clock_now(); | |
2a845540 | 1895 | request->created = ceph::coarse_mono_clock::now(); |
7c673cae FG |
1896 | |
1897 | // make note | |
1898 | mds_requests[tid] = request->get(); | |
1899 | if (oldest_tid == 0 && request->get_op() != CEPH_MDS_OP_SETFILELOCK) | |
1900 | oldest_tid = tid; | |
1901 | ||
1902 | request->set_caller_perms(perms); | |
1903 | ||
1904 | if (cct->_conf->client_inject_fixed_oldest_tid) { | |
1905 | ldout(cct, 20) << __func__ << " injecting fixed oldest_client_tid(1)" << dendl; | |
1906 | request->set_oldest_client_tid(1); | |
1907 | } else { | |
1908 | request->set_oldest_client_tid(oldest_tid); | |
1909 | } | |
1910 | ||
1911 | // hack target mds? | |
1912 | if (use_mds >= 0) | |
1913 | request->resend_mds = use_mds; | |
1914 | ||
20effc67 | 1915 | MetaSessionRef session = NULL; |
7c673cae FG |
1916 | while (1) { |
1917 | if (request->aborted()) | |
1918 | break; | |
1919 | ||
f67539c2 TL |
1920 | if (blocklisted) { |
1921 | request->abort(-CEPHFS_EBLOCKLISTED); | |
31f18b77 FG |
1922 | break; |
1923 | } | |
1924 | ||
7c673cae | 1925 | // set up wait cond |
9f95a23c | 1926 | ceph::condition_variable caller_cond; |
7c673cae FG |
1927 | request->caller_cond = &caller_cond; |
1928 | ||
1929 | // choose mds | |
1930 | Inode *hash_diri = NULL; | |
1931 | mds_rank_t mds = choose_target_mds(request, &hash_diri); | |
1932 | int mds_state = (mds == MDS_RANK_NONE) ? MDSMap::STATE_NULL : mdsmap->get_state(mds); | |
1933 | if (mds_state != MDSMap::STATE_ACTIVE && mds_state != MDSMap::STATE_STOPPING) { | |
1934 | if (mds_state == MDSMap::STATE_NULL && mds >= mdsmap->get_max_mds()) { | |
1935 | if (hash_diri) { | |
1936 | ldout(cct, 10) << " target mds." << mds << " has stopped, remove it from fragmap" << dendl; | |
1937 | _fragmap_remove_stopped_mds(hash_diri, mds); | |
1938 | } else { | |
1939 | ldout(cct, 10) << " target mds." << mds << " has stopped, trying a random mds" << dendl; | |
1940 | request->resend_mds = _get_random_up_mds(); | |
1941 | } | |
1942 | } else { | |
1943 | ldout(cct, 10) << " target mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
1944 | wait_on_list(waiting_for_mdsmap); | |
1945 | } | |
1946 | continue; | |
1947 | } | |
1948 | ||
1949 | // open a session? | |
7c673cae FG |
1950 | if (!have_open_session(mds)) { |
1951 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 1952 | if (session->state == MetaSession::STATE_REJECTED) { |
f67539c2 | 1953 | request->abort(-CEPHFS_EPERM); |
f6b5b4d7 TL |
1954 | break; |
1955 | } | |
7c673cae FG |
1956 | // wait |
1957 | if (session->state == MetaSession::STATE_OPENING) { | |
1958 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
1959 | wait_on_context_list(session->waiting_for_open); | |
7c673cae FG |
1960 | continue; |
1961 | } | |
1962 | ||
1963 | if (!have_open_session(mds)) | |
1964 | continue; | |
1965 | } else { | |
20effc67 | 1966 | session = mds_sessions.at(mds); |
7c673cae FG |
1967 | } |
1968 | ||
1969 | // send request. | |
20effc67 | 1970 | send_request(request, session.get()); |
7c673cae FG |
1971 | |
1972 | // wait for signal | |
1973 | ldout(cct, 20) << "awaiting reply|forward|kick on " << &caller_cond << dendl; | |
1974 | request->kick = false; | |
9f95a23c TL |
1975 | std::unique_lock l{client_lock, std::adopt_lock}; |
1976 | caller_cond.wait(l, [request] { | |
1977 | return (request->reply || // reply | |
1978 | request->resend_mds >= 0 || // forward | |
1979 | request->kick); | |
1980 | }); | |
1981 | l.release(); | |
1982 | request->caller_cond = nullptr; | |
7c673cae FG |
1983 | |
1984 | // did we get a reply? | |
1985 | if (request->reply) | |
1986 | break; | |
1987 | } | |
1988 | ||
1989 | if (!request->reply) { | |
11fdf7f2 TL |
1990 | ceph_assert(request->aborted()); |
1991 | ceph_assert(!request->got_unsafe); | |
7c673cae FG |
1992 | r = request->get_abort_code(); |
1993 | request->item.remove_myself(); | |
1994 | unregister_request(request); | |
11fdf7f2 | 1995 | put_request(request); |
7c673cae FG |
1996 | return r; |
1997 | } | |
1998 | ||
1999 | // got it! | |
11fdf7f2 | 2000 | auto reply = std::move(request->reply); |
7c673cae FG |
2001 | r = reply->get_result(); |
2002 | if (r >= 0) | |
2003 | request->success = true; | |
2004 | ||
2005 | // kick dispatcher (we've got it!) | |
11fdf7f2 | 2006 | ceph_assert(request->dispatch_cond); |
9f95a23c | 2007 | request->dispatch_cond->notify_all(); |
7c673cae FG |
2008 | ldout(cct, 20) << "sendrecv kickback on tid " << tid << " " << request->dispatch_cond << dendl; |
2009 | request->dispatch_cond = 0; | |
2010 | ||
2011 | if (r >= 0 && ptarget) | |
20effc67 | 2012 | r = verify_reply_trace(r, session.get(), request, reply, ptarget, pcreated, perms); |
7c673cae FG |
2013 | |
2014 | if (pdirbl) | |
11fdf7f2 | 2015 | *pdirbl = reply->get_extra_bl(); |
7c673cae FG |
2016 | |
2017 | // -- log times -- | |
2018 | utime_t lat = ceph_clock_now(); | |
2019 | lat -= request->sent_stamp; | |
2020 | ldout(cct, 20) << "lat " << lat << dendl; | |
2a845540 TL |
2021 | |
2022 | ++nr_metadata_request; | |
2023 | update_io_stat_metadata(lat); | |
7c673cae FG |
2024 | |
2025 | put_request(request); | |
7c673cae FG |
2026 | return r; |
2027 | } | |
2028 | ||
2029 | void Client::unregister_request(MetaRequest *req) | |
2030 | { | |
2031 | mds_requests.erase(req->tid); | |
2032 | if (req->tid == oldest_tid) { | |
2033 | map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.upper_bound(oldest_tid); | |
2034 | while (true) { | |
2035 | if (p == mds_requests.end()) { | |
2036 | oldest_tid = 0; | |
2037 | break; | |
2038 | } | |
2039 | if (p->second->get_op() != CEPH_MDS_OP_SETFILELOCK) { | |
2040 | oldest_tid = p->first; | |
2041 | break; | |
2042 | } | |
2043 | ++p; | |
2044 | } | |
2045 | } | |
2046 | put_request(req); | |
2047 | } | |
2048 | ||
2049 | void Client::put_request(MetaRequest *request) | |
2050 | { | |
2051 | if (request->_put()) { | |
2052 | int op = -1; | |
2053 | if (request->success) | |
2054 | op = request->get_op(); | |
2055 | InodeRef other_in; | |
2056 | request->take_other_inode(&other_in); | |
2057 | delete request; | |
2058 | ||
2059 | if (other_in && | |
2060 | (op == CEPH_MDS_OP_RMDIR || | |
2061 | op == CEPH_MDS_OP_RENAME || | |
2062 | op == CEPH_MDS_OP_RMSNAP)) { | |
2063 | _try_to_trim_inode(other_in.get(), false); | |
2064 | } | |
2065 | } | |
2066 | } | |
2067 | ||
2068 | int Client::encode_inode_release(Inode *in, MetaRequest *req, | |
2069 | mds_rank_t mds, int drop, | |
2070 | int unless, int force) | |
2071 | { | |
11fdf7f2 | 2072 | ldout(cct, 20) << __func__ << " enter(in:" << *in << ", req:" << req |
f67539c2 | 2073 | << " mds:" << mds << ", drop:" << ccap_string(drop) << ", unless:" << ccap_string(unless) |
1911f103 | 2074 | << ", force:" << force << ")" << dendl; |
7c673cae | 2075 | int released = 0; |
11fdf7f2 TL |
2076 | auto it = in->caps.find(mds); |
2077 | if (it != in->caps.end()) { | |
2078 | Cap &cap = it->second; | |
7c673cae | 2079 | drop &= ~(in->dirty_caps | get_caps_used(in)); |
11fdf7f2 TL |
2080 | if ((drop & cap.issued) && |
2081 | !(unless & cap.issued)) { | |
1911f103 | 2082 | ldout(cct, 25) << "dropping caps " << ccap_string(drop) << dendl; |
11fdf7f2 TL |
2083 | cap.issued &= ~drop; |
2084 | cap.implemented &= ~drop; | |
7c673cae | 2085 | released = 1; |
7c673cae FG |
2086 | } else { |
2087 | released = force; | |
2088 | } | |
2089 | if (released) { | |
1911f103 TL |
2090 | cap.wanted = in->caps_wanted(); |
2091 | if (&cap == in->auth_cap && | |
2092 | !(cap.wanted & CEPH_CAP_ANY_FILE_WR)) { | |
2093 | in->requested_max_size = 0; | |
2094 | ldout(cct, 25) << "reset requested_max_size due to not wanting any file write cap" << dendl; | |
2095 | } | |
7c673cae FG |
2096 | ceph_mds_request_release rel; |
2097 | rel.ino = in->ino; | |
11fdf7f2 TL |
2098 | rel.cap_id = cap.cap_id; |
2099 | rel.seq = cap.seq; | |
2100 | rel.issue_seq = cap.issue_seq; | |
2101 | rel.mseq = cap.mseq; | |
2102 | rel.caps = cap.implemented; | |
2103 | rel.wanted = cap.wanted; | |
7c673cae FG |
2104 | rel.dname_len = 0; |
2105 | rel.dname_seq = 0; | |
2106 | req->cap_releases.push_back(MClientRequest::Release(rel,"")); | |
2107 | } | |
2108 | } | |
11fdf7f2 | 2109 | ldout(cct, 25) << __func__ << " exit(in:" << *in << ") released:" |
7c673cae FG |
2110 | << released << dendl; |
2111 | return released; | |
2112 | } | |
2113 | ||
2114 | void Client::encode_dentry_release(Dentry *dn, MetaRequest *req, | |
2115 | mds_rank_t mds, int drop, int unless) | |
2116 | { | |
11fdf7f2 | 2117 | ldout(cct, 20) << __func__ << " enter(dn:" |
7c673cae FG |
2118 | << dn << ")" << dendl; |
2119 | int released = 0; | |
2120 | if (dn->dir) | |
2121 | released = encode_inode_release(dn->dir->parent_inode, req, | |
2122 | mds, drop, unless, 1); | |
2123 | if (released && dn->lease_mds == mds) { | |
2124 | ldout(cct, 25) << "preemptively releasing dn to mds" << dendl; | |
11fdf7f2 | 2125 | auto& rel = req->cap_releases.back(); |
7c673cae FG |
2126 | rel.item.dname_len = dn->name.length(); |
2127 | rel.item.dname_seq = dn->lease_seq; | |
2128 | rel.dname = dn->name; | |
adb31ebb | 2129 | dn->lease_mds = -1; |
7c673cae | 2130 | } |
11fdf7f2 | 2131 | ldout(cct, 25) << __func__ << " exit(dn:" |
7c673cae FG |
2132 | << dn << ")" << dendl; |
2133 | } | |
2134 | ||
2135 | ||
2136 | /* | |
2137 | * This requires the MClientRequest *request member to be set. | |
2138 | * It will error out horribly without one. | |
2139 | * Additionally, if you set any *drop member, you'd better have | |
2140 | * set the corresponding dentry! | |
2141 | */ | |
2142 | void Client::encode_cap_releases(MetaRequest *req, mds_rank_t mds) | |
2143 | { | |
11fdf7f2 | 2144 | ldout(cct, 20) << __func__ << " enter (req: " |
7c673cae FG |
2145 | << req << ", mds: " << mds << ")" << dendl; |
2146 | if (req->inode_drop && req->inode()) | |
2147 | encode_inode_release(req->inode(), req, | |
2148 | mds, req->inode_drop, | |
2149 | req->inode_unless); | |
2150 | ||
2151 | if (req->old_inode_drop && req->old_inode()) | |
2152 | encode_inode_release(req->old_inode(), req, | |
2153 | mds, req->old_inode_drop, | |
2154 | req->old_inode_unless); | |
2155 | if (req->other_inode_drop && req->other_inode()) | |
2156 | encode_inode_release(req->other_inode(), req, | |
2157 | mds, req->other_inode_drop, | |
2158 | req->other_inode_unless); | |
2159 | ||
2160 | if (req->dentry_drop && req->dentry()) | |
2161 | encode_dentry_release(req->dentry(), req, | |
2162 | mds, req->dentry_drop, | |
2163 | req->dentry_unless); | |
2164 | ||
2165 | if (req->old_dentry_drop && req->old_dentry()) | |
2166 | encode_dentry_release(req->old_dentry(), req, | |
2167 | mds, req->old_dentry_drop, | |
2168 | req->old_dentry_unless); | |
11fdf7f2 | 2169 | ldout(cct, 25) << __func__ << " exit (req: " |
7c673cae FG |
2170 | << req << ", mds " << mds <<dendl; |
2171 | } | |
2172 | ||
2173 | bool Client::have_open_session(mds_rank_t mds) | |
2174 | { | |
11fdf7f2 TL |
2175 | const auto &it = mds_sessions.find(mds); |
2176 | return it != mds_sessions.end() && | |
20effc67 TL |
2177 | (it->second->state == MetaSession::STATE_OPEN || |
2178 | it->second->state == MetaSession::STATE_STALE); | |
7c673cae FG |
2179 | } |
2180 | ||
20effc67 | 2181 | MetaSessionRef Client::_get_mds_session(mds_rank_t mds, Connection *con) |
7c673cae | 2182 | { |
11fdf7f2 | 2183 | const auto &it = mds_sessions.find(mds); |
20effc67 | 2184 | if (it == mds_sessions.end() || it->second->con != con) { |
7c673cae | 2185 | return NULL; |
11fdf7f2 | 2186 | } else { |
20effc67 | 2187 | return it->second; |
11fdf7f2 | 2188 | } |
7c673cae FG |
2189 | } |
2190 | ||
20effc67 | 2191 | MetaSessionRef Client::_get_or_open_mds_session(mds_rank_t mds) |
7c673cae | 2192 | { |
11fdf7f2 | 2193 | auto it = mds_sessions.find(mds); |
20effc67 | 2194 | return it == mds_sessions.end() ? _open_mds_session(mds) : it->second; |
7c673cae FG |
2195 | } |
2196 | ||
2197 | /** | |
2198 | * Populate a map of strings with client-identifying metadata, | |
2199 | * such as the hostname. Call this once at initialization. | |
2200 | */ | |
2201 | void Client::populate_metadata(const std::string &mount_root) | |
2202 | { | |
2203 | // Hostname | |
f67539c2 TL |
2204 | #ifdef _WIN32 |
2205 | // TODO: move this to compat.h | |
2206 | char hostname[64]; | |
2207 | DWORD hostname_sz = 64; | |
2208 | GetComputerNameA(hostname, &hostname_sz); | |
2209 | metadata["hostname"] = hostname; | |
2210 | #else | |
7c673cae FG |
2211 | struct utsname u; |
2212 | int r = uname(&u); | |
2213 | if (r >= 0) { | |
2214 | metadata["hostname"] = u.nodename; | |
2215 | ldout(cct, 20) << __func__ << " read hostname '" << u.nodename << "'" << dendl; | |
2216 | } else { | |
2217 | ldout(cct, 1) << __func__ << " failed to read hostname (" << cpp_strerror(r) << ")" << dendl; | |
2218 | } | |
f67539c2 | 2219 | #endif |
7c673cae FG |
2220 | |
2221 | metadata["pid"] = stringify(getpid()); | |
2222 | ||
2223 | // Ceph entity id (the '0' in "client.0") | |
2224 | metadata["entity_id"] = cct->_conf->name.get_id(); | |
2225 | ||
2226 | // Our mount position | |
2227 | if (!mount_root.empty()) { | |
2228 | metadata["root"] = mount_root; | |
2229 | } | |
2230 | ||
2231 | // Ceph version | |
2232 | metadata["ceph_version"] = pretty_version_to_str(); | |
2233 | metadata["ceph_sha1"] = git_version_to_str(); | |
2234 | ||
2235 | // Apply any metadata from the user's configured overrides | |
2236 | std::vector<std::string> tokens; | |
2237 | get_str_vec(cct->_conf->client_metadata, ",", tokens); | |
2238 | for (const auto &i : tokens) { | |
2239 | auto eqpos = i.find("="); | |
2240 | // Throw out anything that isn't of the form "<str>=<str>" | |
2241 | if (eqpos == 0 || eqpos == std::string::npos || eqpos == i.size()) { | |
2242 | lderr(cct) << "Invalid metadata keyval pair: '" << i << "'" << dendl; | |
2243 | continue; | |
2244 | } | |
2245 | metadata[i.substr(0, eqpos)] = i.substr(eqpos + 1); | |
2246 | } | |
2247 | } | |
2248 | ||
2249 | /** | |
2250 | * Optionally add or override client metadata fields. | |
2251 | */ | |
2252 | void Client::update_metadata(std::string const &k, std::string const &v) | |
2253 | { | |
f67539c2 TL |
2254 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2255 | ceph_assert(iref_reader.is_state_satisfied()); | |
2256 | ||
2257 | std::scoped_lock l(client_lock); | |
7c673cae | 2258 | |
11fdf7f2 TL |
2259 | auto it = metadata.find(k); |
2260 | if (it != metadata.end()) { | |
7c673cae | 2261 | ldout(cct, 1) << __func__ << " warning, overriding metadata field '" << k |
11fdf7f2 | 2262 | << "' from '" << it->second << "' to '" << v << "'" << dendl; |
7c673cae FG |
2263 | } |
2264 | ||
2265 | metadata[k] = v; | |
2266 | } | |
2267 | ||
20effc67 | 2268 | MetaSessionRef Client::_open_mds_session(mds_rank_t mds) |
7c673cae | 2269 | { |
11fdf7f2 TL |
2270 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
2271 | auto addrs = mdsmap->get_addrs(mds); | |
2272 | auto em = mds_sessions.emplace(std::piecewise_construct, | |
2273 | std::forward_as_tuple(mds), | |
20effc67 | 2274 | std::forward_as_tuple(new MetaSession(mds, messenger->connect_to_mds(addrs), addrs))); |
11fdf7f2 | 2275 | ceph_assert(em.second); /* not already present */ |
20effc67 | 2276 | auto session = em.first->second; |
7c673cae | 2277 | |
9f95a23c | 2278 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_OPEN); |
11fdf7f2 TL |
2279 | m->metadata = metadata; |
2280 | m->supported_features = feature_bitset_t(CEPHFS_FEATURES_CLIENT_SUPPORTED); | |
f67539c2 | 2281 | m->metric_spec = feature_bitset_t(CEPHFS_METRIC_FEATURES_ALL); |
11fdf7f2 | 2282 | session->con->send_message2(std::move(m)); |
7c673cae FG |
2283 | return session; |
2284 | } | |
2285 | ||
2286 | void Client::_close_mds_session(MetaSession *s) | |
2287 | { | |
11fdf7f2 | 2288 | ldout(cct, 2) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
7c673cae | 2289 | s->state = MetaSession::STATE_CLOSING; |
9f95a23c | 2290 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
2291 | } |
2292 | ||
f6b5b4d7 | 2293 | void Client::_closed_mds_session(MetaSession *s, int err, bool rejected) |
7c673cae | 2294 | { |
11fdf7f2 | 2295 | ldout(cct, 5) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
f6b5b4d7 TL |
2296 | if (rejected && s->state != MetaSession::STATE_CLOSING) |
2297 | s->state = MetaSession::STATE_REJECTED; | |
2298 | else | |
2299 | s->state = MetaSession::STATE_CLOSED; | |
7c673cae FG |
2300 | s->con->mark_down(); |
2301 | signal_context_list(s->waiting_for_open); | |
9f95a23c | 2302 | mount_cond.notify_all(); |
f6b5b4d7 | 2303 | remove_session_caps(s, err); |
7c673cae | 2304 | kick_requests_closed(s); |
f6b5b4d7 TL |
2305 | mds_ranks_closing.erase(s->mds_num); |
2306 | if (s->state == MetaSession::STATE_CLOSED) | |
2307 | mds_sessions.erase(s->mds_num); | |
7c673cae FG |
2308 | } |
2309 | ||
11fdf7f2 | 2310 | void Client::handle_client_session(const MConstRef<MClientSession>& m) |
7c673cae FG |
2311 | { |
2312 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
11fdf7f2 | 2313 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << from << dendl; |
7c673cae | 2314 | |
f67539c2 | 2315 | std::scoped_lock cl(client_lock); |
20effc67 | 2316 | auto session = _get_mds_session(from, m->get_connection().get()); |
7c673cae FG |
2317 | if (!session) { |
2318 | ldout(cct, 10) << " discarding session message from sessionless mds " << m->get_source_inst() << dendl; | |
7c673cae FG |
2319 | return; |
2320 | } | |
2321 | ||
2322 | switch (m->get_op()) { | |
2323 | case CEPH_SESSION_OPEN: | |
11fdf7f2 TL |
2324 | { |
2325 | feature_bitset_t missing_features(CEPHFS_FEATURES_CLIENT_REQUIRED); | |
2326 | missing_features -= m->supported_features; | |
2327 | if (!missing_features.empty()) { | |
2328 | lderr(cct) << "mds." << from << " lacks required features '" | |
2329 | << missing_features << "', closing session " << dendl; | |
20effc67 TL |
2330 | _close_mds_session(session.get()); |
2331 | _closed_mds_session(session.get(), -CEPHFS_EPERM, true); | |
11fdf7f2 TL |
2332 | break; |
2333 | } | |
2334 | session->mds_features = std::move(m->supported_features); | |
33c7a0ef | 2335 | session->mds_metric_flags = std::move(m->metric_spec.metric_flags); |
11fdf7f2 | 2336 | |
20effc67 | 2337 | renew_caps(session.get()); |
11fdf7f2 | 2338 | session->state = MetaSession::STATE_OPEN; |
f67539c2 | 2339 | if (is_unmounting()) |
9f95a23c | 2340 | mount_cond.notify_all(); |
11fdf7f2 TL |
2341 | else |
2342 | connect_mds_targets(from); | |
2343 | signal_context_list(session->waiting_for_open); | |
2344 | break; | |
2345 | } | |
7c673cae FG |
2346 | |
2347 | case CEPH_SESSION_CLOSE: | |
20effc67 | 2348 | _closed_mds_session(session.get()); |
7c673cae FG |
2349 | break; |
2350 | ||
2351 | case CEPH_SESSION_RENEWCAPS: | |
2352 | if (session->cap_renew_seq == m->get_seq()) { | |
a8e16298 | 2353 | bool was_stale = ceph_clock_now() >= session->cap_ttl; |
7c673cae FG |
2354 | session->cap_ttl = |
2355 | session->last_cap_renew_request + mdsmap->get_session_timeout(); | |
a8e16298 | 2356 | if (was_stale) |
20effc67 | 2357 | wake_up_session_caps(session.get(), false); |
7c673cae FG |
2358 | } |
2359 | break; | |
2360 | ||
2361 | case CEPH_SESSION_STALE: | |
28e407b8 AA |
2362 | // invalidate session caps/leases |
2363 | session->cap_gen++; | |
2364 | session->cap_ttl = ceph_clock_now(); | |
2365 | session->cap_ttl -= 1; | |
20effc67 | 2366 | renew_caps(session.get()); |
7c673cae FG |
2367 | break; |
2368 | ||
2369 | case CEPH_SESSION_RECALL_STATE: | |
f67539c2 TL |
2370 | /* |
2371 | * Call the renew caps and flush cap releases just before | |
2372 | * triming the caps in case the tick() won't get a chance | |
2373 | * to run them, which could cause the client to be blocklisted | |
2374 | * and MDS daemons trying to recall the caps again and | |
2375 | * again. | |
2376 | * | |
2377 | * In most cases it will do nothing, and the new cap releases | |
2378 | * added by trim_caps() followed will be deferred flushing | |
2379 | * by tick(). | |
2380 | */ | |
2381 | renew_and_flush_cap_releases(); | |
20effc67 | 2382 | trim_caps(session.get(), m->get_max_caps()); |
7c673cae FG |
2383 | break; |
2384 | ||
2385 | case CEPH_SESSION_FLUSHMSG: | |
a8e16298 | 2386 | /* flush cap release */ |
11fdf7f2 TL |
2387 | if (auto& m = session->release; m) { |
2388 | session->con->send_message2(std::move(m)); | |
a8e16298 | 2389 | } |
9f95a23c | 2390 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_FLUSHMSG_ACK, m->get_seq())); |
7c673cae FG |
2391 | break; |
2392 | ||
2393 | case CEPH_SESSION_FORCE_RO: | |
20effc67 | 2394 | force_session_readonly(session.get()); |
7c673cae FG |
2395 | break; |
2396 | ||
2397 | case CEPH_SESSION_REJECT: | |
11fdf7f2 TL |
2398 | { |
2399 | std::string_view error_str; | |
2400 | auto it = m->metadata.find("error_string"); | |
2401 | if (it != m->metadata.end()) | |
2402 | error_str = it->second; | |
2403 | else | |
2404 | error_str = "unknown error"; | |
2405 | lderr(cct) << "mds." << from << " rejected us (" << error_str << ")" << dendl; | |
7c673cae | 2406 | |
20effc67 | 2407 | _closed_mds_session(session.get(), -CEPHFS_EPERM, true); |
11fdf7f2 | 2408 | } |
7c673cae FG |
2409 | break; |
2410 | ||
2411 | default: | |
2412 | ceph_abort(); | |
2413 | } | |
7c673cae FG |
2414 | } |
2415 | ||
2416 | bool Client::_any_stale_sessions() const | |
2417 | { | |
9f95a23c | 2418 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae | 2419 | |
11fdf7f2 | 2420 | for (const auto &p : mds_sessions) { |
20effc67 | 2421 | if (p.second->state == MetaSession::STATE_STALE) { |
7c673cae FG |
2422 | return true; |
2423 | } | |
2424 | } | |
2425 | ||
2426 | return false; | |
2427 | } | |
2428 | ||
2429 | void Client::_kick_stale_sessions() | |
2430 | { | |
11fdf7f2 | 2431 | ldout(cct, 1) << __func__ << dendl; |
7c673cae | 2432 | |
11fdf7f2 | 2433 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
20effc67 TL |
2434 | auto s = it->second; |
2435 | if (s->state == MetaSession::STATE_REJECTED) { | |
2436 | mds_sessions.erase(it->first); | |
f6b5b4d7 TL |
2437 | continue; |
2438 | } | |
20effc67 TL |
2439 | if (s->state == MetaSession::STATE_STALE) |
2440 | _closed_mds_session(s.get()); | |
7c673cae FG |
2441 | } |
2442 | } | |
2443 | ||
2444 | void Client::send_request(MetaRequest *request, MetaSession *session, | |
2445 | bool drop_cap_releases) | |
2446 | { | |
2447 | // make the request | |
2448 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2449 | ldout(cct, 10) << __func__ << " rebuilding request " << request->get_tid() |
7c673cae | 2450 | << " for mds." << mds << dendl; |
11fdf7f2 | 2451 | auto r = build_client_request(request); |
7c673cae FG |
2452 | if (request->dentry()) { |
2453 | r->set_dentry_wanted(); | |
2454 | } | |
2455 | if (request->got_unsafe) { | |
2456 | r->set_replayed_op(); | |
2457 | if (request->target) | |
2458 | r->head.ino = request->target->ino; | |
2459 | } else { | |
2460 | encode_cap_releases(request, mds); | |
2461 | if (drop_cap_releases) // we haven't send cap reconnect yet, drop cap releases | |
2462 | request->cap_releases.clear(); | |
2463 | else | |
2464 | r->releases.swap(request->cap_releases); | |
2465 | } | |
2466 | r->set_mdsmap_epoch(mdsmap->get_epoch()); | |
2467 | if (r->head.op == CEPH_MDS_OP_SETXATTR) { | |
2468 | objecter->with_osdmap([r](const OSDMap& o) { | |
2469 | r->set_osdmap_epoch(o.get_epoch()); | |
2470 | }); | |
2471 | } | |
2472 | ||
2473 | if (request->mds == -1) { | |
2474 | request->sent_stamp = ceph_clock_now(); | |
11fdf7f2 | 2475 | ldout(cct, 20) << __func__ << " set sent_stamp to " << request->sent_stamp << dendl; |
7c673cae FG |
2476 | } |
2477 | request->mds = mds; | |
2478 | ||
2479 | Inode *in = request->inode(); | |
11fdf7f2 TL |
2480 | if (in) { |
2481 | auto it = in->caps.find(mds); | |
2482 | if (it != in->caps.end()) { | |
2483 | request->sent_on_mseq = it->second.mseq; | |
2484 | } | |
2485 | } | |
7c673cae FG |
2486 | |
2487 | session->requests.push_back(&request->item); | |
2488 | ||
11fdf7f2 TL |
2489 | ldout(cct, 10) << __func__ << " " << *r << " to mds." << mds << dendl; |
2490 | session->con->send_message2(std::move(r)); | |
7c673cae FG |
2491 | } |
2492 | ||
9f95a23c | 2493 | ref_t<MClientRequest> Client::build_client_request(MetaRequest *request) |
7c673cae | 2494 | { |
9f95a23c | 2495 | auto req = make_message<MClientRequest>(request->get_op()); |
7c673cae FG |
2496 | req->set_tid(request->tid); |
2497 | req->set_stamp(request->op_stamp); | |
2498 | memcpy(&req->head, &request->head, sizeof(ceph_mds_request_head)); | |
2499 | ||
2500 | // if the filepath's haven't been set, set them! | |
2501 | if (request->path.empty()) { | |
2502 | Inode *in = request->inode(); | |
2503 | Dentry *de = request->dentry(); | |
2504 | if (in) | |
2505 | in->make_nosnap_relative_path(request->path); | |
2506 | else if (de) { | |
2507 | if (de->inode) | |
2508 | de->inode->make_nosnap_relative_path(request->path); | |
2509 | else if (de->dir) { | |
2510 | de->dir->parent_inode->make_nosnap_relative_path(request->path); | |
2511 | request->path.push_dentry(de->name); | |
2512 | } | |
2513 | else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2514 | << " No path, inode, or appropriately-endowed dentry given!" | |
2515 | << dendl; | |
2516 | } else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2517 | << " No path, inode, or dentry given!" | |
2518 | << dendl; | |
2519 | } | |
2520 | req->set_filepath(request->get_filepath()); | |
2521 | req->set_filepath2(request->get_filepath2()); | |
f67539c2 | 2522 | req->set_alternate_name(request->alternate_name); |
7c673cae FG |
2523 | req->set_data(request->data); |
2524 | req->set_retry_attempt(request->retry_attempt++); | |
2525 | req->head.num_fwd = request->num_fwd; | |
2526 | const gid_t *_gids; | |
2527 | int gid_count = request->perms.get_gids(&_gids); | |
2528 | req->set_gid_list(gid_count, _gids); | |
2529 | return req; | |
2530 | } | |
2531 | ||
2532 | ||
2533 | ||
11fdf7f2 | 2534 | void Client::handle_client_request_forward(const MConstRef<MClientRequestForward>& fwd) |
7c673cae FG |
2535 | { |
2536 | mds_rank_t mds = mds_rank_t(fwd->get_source().num()); | |
f67539c2 TL |
2537 | |
2538 | std::scoped_lock cl(client_lock); | |
20effc67 | 2539 | auto session = _get_mds_session(mds, fwd->get_connection().get()); |
7c673cae | 2540 | if (!session) { |
7c673cae FG |
2541 | return; |
2542 | } | |
2543 | ceph_tid_t tid = fwd->get_tid(); | |
2544 | ||
2545 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2546 | ldout(cct, 10) << __func__ << " no pending request on tid " << tid << dendl; |
7c673cae FG |
2547 | return; |
2548 | } | |
2549 | ||
2550 | MetaRequest *request = mds_requests[tid]; | |
11fdf7f2 | 2551 | ceph_assert(request); |
7c673cae | 2552 | |
33c7a0ef TL |
2553 | /* |
2554 | * The type of 'num_fwd' in ceph 'MClientRequestForward' | |
2555 | * is 'int32_t', while in 'ceph_mds_request_head' the | |
2556 | * type is '__u8'. So in case the request bounces between | |
2557 | * MDSes exceeding 256 times, the client will get stuck. | |
2558 | * | |
2559 | * In this case it's ususally a bug in MDS and continue | |
2560 | * bouncing the request makes no sense. | |
2561 | * | |
2562 | * In future this could be fixed in ceph code, so avoid | |
2563 | * using the hardcode here. | |
2564 | */ | |
2565 | int max_fwd = sizeof(((struct ceph_mds_request_head*)0)->num_fwd); | |
2566 | max_fwd = 1 << (max_fwd * CHAR_BIT) - 1; | |
2567 | auto num_fwd = fwd->get_num_fwd(); | |
2568 | if (num_fwd <= request->num_fwd || num_fwd >= max_fwd) { | |
2569 | if (request->num_fwd >= max_fwd || num_fwd >= max_fwd) { | |
2570 | request->abort(-EMULTIHOP); | |
2571 | request->caller_cond->notify_all(); | |
2572 | ldout(cct, 1) << __func__ << " tid " << tid << " seq overflow" | |
2573 | << ", abort it" << dendl; | |
2574 | } else { | |
2575 | ldout(cct, 10) << __func__ << " tid " << tid | |
2576 | << " old fwd seq " << fwd->get_num_fwd() | |
2577 | << " <= req fwd " << request->num_fwd | |
2578 | << ", ignore it" << dendl; | |
2579 | } | |
2580 | return; | |
2581 | } | |
2582 | ||
7c673cae FG |
2583 | // reset retry counter |
2584 | request->retry_attempt = 0; | |
2585 | ||
2586 | // request not forwarded, or dest mds has no session. | |
2587 | // resend. | |
11fdf7f2 | 2588 | ldout(cct, 10) << __func__ << " tid " << tid |
7c673cae FG |
2589 | << " fwd " << fwd->get_num_fwd() |
2590 | << " to mds." << fwd->get_dest_mds() | |
2591 | << ", resending to " << fwd->get_dest_mds() | |
2592 | << dendl; | |
2593 | ||
2594 | request->mds = -1; | |
2595 | request->item.remove_myself(); | |
33c7a0ef | 2596 | request->num_fwd = num_fwd; |
7c673cae | 2597 | request->resend_mds = fwd->get_dest_mds(); |
9f95a23c | 2598 | request->caller_cond->notify_all(); |
7c673cae FG |
2599 | } |
2600 | ||
2601 | bool Client::is_dir_operation(MetaRequest *req) | |
2602 | { | |
2603 | int op = req->get_op(); | |
2604 | if (op == CEPH_MDS_OP_MKNOD || op == CEPH_MDS_OP_LINK || | |
2605 | op == CEPH_MDS_OP_UNLINK || op == CEPH_MDS_OP_RENAME || | |
2606 | op == CEPH_MDS_OP_MKDIR || op == CEPH_MDS_OP_RMDIR || | |
2607 | op == CEPH_MDS_OP_SYMLINK || op == CEPH_MDS_OP_CREATE) | |
2608 | return true; | |
2609 | return false; | |
2610 | } | |
2611 | ||
11fdf7f2 | 2612 | void Client::handle_client_reply(const MConstRef<MClientReply>& reply) |
7c673cae FG |
2613 | { |
2614 | mds_rank_t mds_num = mds_rank_t(reply->get_source().num()); | |
f67539c2 TL |
2615 | |
2616 | std::scoped_lock cl(client_lock); | |
20effc67 | 2617 | auto session = _get_mds_session(mds_num, reply->get_connection().get()); |
7c673cae | 2618 | if (!session) { |
7c673cae FG |
2619 | return; |
2620 | } | |
2621 | ||
2622 | ceph_tid_t tid = reply->get_tid(); | |
2623 | bool is_safe = reply->is_safe(); | |
2624 | ||
2625 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2626 | lderr(cct) << __func__ << " no pending request on tid " << tid |
7c673cae | 2627 | << " safe is:" << is_safe << dendl; |
7c673cae FG |
2628 | return; |
2629 | } | |
2630 | MetaRequest *request = mds_requests.at(tid); | |
2631 | ||
11fdf7f2 | 2632 | ldout(cct, 20) << __func__ << " got a reply. Safe:" << is_safe |
7c673cae FG |
2633 | << " tid " << tid << dendl; |
2634 | ||
2635 | if (request->got_unsafe && !is_safe) { | |
2636 | //duplicate response | |
2637 | ldout(cct, 0) << "got a duplicate reply on tid " << tid << " from mds " | |
2638 | << mds_num << " safe:" << is_safe << dendl; | |
7c673cae FG |
2639 | return; |
2640 | } | |
2641 | ||
11fdf7f2 | 2642 | ceph_assert(!request->reply); |
7c673cae | 2643 | request->reply = reply; |
20effc67 | 2644 | insert_trace(request, session.get()); |
7c673cae FG |
2645 | |
2646 | // Handle unsafe reply | |
2647 | if (!is_safe) { | |
2648 | request->got_unsafe = true; | |
2649 | session->unsafe_requests.push_back(&request->unsafe_item); | |
2650 | if (is_dir_operation(request)) { | |
2651 | Inode *dir = request->inode(); | |
11fdf7f2 | 2652 | ceph_assert(dir); |
7c673cae FG |
2653 | dir->unsafe_ops.push_back(&request->unsafe_dir_item); |
2654 | } | |
2655 | if (request->target) { | |
2656 | InodeRef &in = request->target; | |
2657 | in->unsafe_ops.push_back(&request->unsafe_target_item); | |
2658 | } | |
2659 | } | |
2660 | ||
2661 | // Only signal the caller once (on the first reply): | |
2662 | // Either its an unsafe reply, or its a safe reply and no unsafe reply was sent. | |
2663 | if (!is_safe || !request->got_unsafe) { | |
9f95a23c | 2664 | ceph::condition_variable cond; |
7c673cae FG |
2665 | request->dispatch_cond = &cond; |
2666 | ||
2667 | // wake up waiter | |
11fdf7f2 | 2668 | ldout(cct, 20) << __func__ << " signalling caller " << (void*)request->caller_cond << dendl; |
9f95a23c | 2669 | request->caller_cond->notify_all(); |
7c673cae FG |
2670 | |
2671 | // wake for kick back | |
9f95a23c TL |
2672 | std::unique_lock l{client_lock, std::adopt_lock}; |
2673 | cond.wait(l, [tid, request, &cond, this] { | |
2674 | if (request->dispatch_cond) { | |
2675 | ldout(cct, 20) << "handle_client_reply awaiting kickback on tid " | |
2676 | << tid << " " << &cond << dendl; | |
2677 | } | |
2678 | return !request->dispatch_cond; | |
2679 | }); | |
2680 | l.release(); | |
7c673cae FG |
2681 | } |
2682 | ||
2683 | if (is_safe) { | |
2684 | // the filesystem change is committed to disk | |
2685 | // we're done, clean up | |
2686 | if (request->got_unsafe) { | |
2687 | request->unsafe_item.remove_myself(); | |
2688 | request->unsafe_dir_item.remove_myself(); | |
2689 | request->unsafe_target_item.remove_myself(); | |
2690 | signal_cond_list(request->waitfor_safe); | |
2691 | } | |
2692 | request->item.remove_myself(); | |
2693 | unregister_request(request); | |
2694 | } | |
f67539c2 | 2695 | if (is_unmounting()) |
9f95a23c | 2696 | mount_cond.notify_all(); |
7c673cae FG |
2697 | } |
2698 | ||
2699 | void Client::_handle_full_flag(int64_t pool) | |
2700 | { | |
2701 | ldout(cct, 1) << __func__ << ": FULL: cancelling outstanding operations " | |
2702 | << "on " << pool << dendl; | |
f67539c2 | 2703 | // Cancel all outstanding ops in this pool with -CEPHFS_ENOSPC: it is necessary |
7c673cae FG |
2704 | // to do this rather than blocking, because otherwise when we fill up we |
2705 | // potentially lock caps forever on files with dirty pages, and we need | |
2706 | // to be able to release those caps to the MDS so that it can delete files | |
2707 | // and free up space. | |
f67539c2 | 2708 | epoch_t cancelled_epoch = objecter->op_cancel_writes(-CEPHFS_ENOSPC, pool); |
7c673cae FG |
2709 | |
2710 | // For all inodes with layouts in this pool and a pending flush write op | |
2711 | // (i.e. one of the ones we will cancel), we've got to purge_set their data | |
2712 | // from ObjectCacher so that it doesn't re-issue the write in response to | |
2713 | // the ENOSPC error. | |
2714 | // Fortunately since we're cancelling everything in a given pool, we don't | |
2715 | // need to know which ops belong to which ObjectSet, we can just blow all | |
2716 | // the un-flushed cached data away and mark any dirty inodes' async_err | |
f67539c2 | 2717 | // field with -CEPHFS_ENOSPC as long as we're sure all the ops we cancelled were |
7c673cae FG |
2718 | // affecting this pool, and all the objectsets we're purging were also |
2719 | // in this pool. | |
2720 | for (unordered_map<vinodeno_t,Inode*>::iterator i = inode_map.begin(); | |
2721 | i != inode_map.end(); ++i) | |
2722 | { | |
2723 | Inode *inode = i->second; | |
2724 | if (inode->oset.dirty_or_tx | |
2725 | && (pool == -1 || inode->layout.pool_id == pool)) { | |
2726 | ldout(cct, 4) << __func__ << ": FULL: inode 0x" << std::hex << i->first << std::dec | |
2727 | << " has dirty objects, purging and setting ENOSPC" << dendl; | |
2728 | objectcacher->purge_set(&inode->oset); | |
f67539c2 | 2729 | inode->set_async_err(-CEPHFS_ENOSPC); |
7c673cae FG |
2730 | } |
2731 | } | |
2732 | ||
2733 | if (cancelled_epoch != (epoch_t)-1) { | |
2734 | set_cap_epoch_barrier(cancelled_epoch); | |
2735 | } | |
2736 | } | |
2737 | ||
11fdf7f2 | 2738 | void Client::handle_osd_map(const MConstRef<MOSDMap>& m) |
7c673cae | 2739 | { |
f67539c2 | 2740 | std::scoped_lock cl(client_lock); |
31f18b77 | 2741 | |
11fdf7f2 | 2742 | const auto myaddrs = messenger->get_myaddrs(); |
33c7a0ef | 2743 | bool new_blocklist = objecter->with_osdmap( |
11fdf7f2 | 2744 | [&](const OSDMap& o) { |
33c7a0ef | 2745 | return o.is_blocklisted(myaddrs); |
11fdf7f2 | 2746 | }); |
33c7a0ef TL |
2747 | |
2748 | if (new_blocklist && !blocklisted) { | |
31f18b77 FG |
2749 | auto epoch = objecter->with_osdmap([](const OSDMap &o){ |
2750 | return o.get_epoch(); | |
2751 | }); | |
f67539c2 TL |
2752 | lderr(cct) << "I was blocklisted at osd epoch " << epoch << dendl; |
2753 | blocklisted = true; | |
31f18b77 | 2754 | |
f67539c2 | 2755 | _abort_mds_sessions(-CEPHFS_EBLOCKLISTED); |
31f18b77 FG |
2756 | |
2757 | // Since we know all our OSD ops will fail, cancel them all preemtively, | |
2758 | // so that on an unhealthy cluster we can umount promptly even if e.g. | |
2759 | // some PGs were inaccessible. | |
f67539c2 TL |
2760 | objecter->op_cancel_writes(-CEPHFS_EBLOCKLISTED); |
2761 | ||
2762 | } | |
31f18b77 | 2763 | |
f67539c2 TL |
2764 | if (blocklisted) { |
2765 | // Handle case where we were blocklisted but no longer are | |
2766 | blocklisted = objecter->with_osdmap([myaddrs](const OSDMap &o){ | |
2767 | return o.is_blocklisted(myaddrs);}); | |
31f18b77 FG |
2768 | } |
2769 | ||
f67539c2 TL |
2770 | // Always subscribe to next osdmap for blocklisted client |
2771 | // until this client is not blocklisted. | |
2772 | if (blocklisted) { | |
f64942e4 AA |
2773 | objecter->maybe_request_map(); |
2774 | } | |
2775 | ||
7c673cae FG |
2776 | if (objecter->osdmap_full_flag()) { |
2777 | _handle_full_flag(-1); | |
2778 | } else { | |
2779 | // Accumulate local list of full pools so that I can drop | |
2780 | // the objecter lock before re-entering objecter in | |
2781 | // cancel_writes | |
2782 | std::vector<int64_t> full_pools; | |
2783 | ||
2784 | objecter->with_osdmap([&full_pools](const OSDMap &o) { | |
2785 | for (const auto& kv : o.get_pools()) { | |
2786 | if (kv.second.has_flag(pg_pool_t::FLAG_FULL)) { | |
2787 | full_pools.push_back(kv.first); | |
2788 | } | |
2789 | } | |
2790 | }); | |
2791 | ||
2792 | for (auto p : full_pools) | |
2793 | _handle_full_flag(p); | |
2794 | ||
2795 | // Subscribe to subsequent maps to watch for the full flag going | |
2796 | // away. For the global full flag objecter does this for us, but | |
2797 | // it pays no attention to the per-pool full flag so in this branch | |
2798 | // we do it ourselves. | |
2799 | if (!full_pools.empty()) { | |
2800 | objecter->maybe_request_map(); | |
2801 | } | |
2802 | } | |
7c673cae FG |
2803 | } |
2804 | ||
2805 | ||
2806 | // ------------------------ | |
2807 | // incoming messages | |
2808 | ||
2809 | ||
11fdf7f2 | 2810 | bool Client::ms_dispatch2(const MessageRef &m) |
7c673cae | 2811 | { |
f67539c2 TL |
2812 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2813 | if (!iref_reader.is_state_satisfied()) { | |
7c673cae | 2814 | ldout(cct, 10) << "inactive, discarding " << *m << dendl; |
7c673cae FG |
2815 | return true; |
2816 | } | |
2817 | ||
2818 | switch (m->get_type()) { | |
2819 | // mounting and mds sessions | |
2820 | case CEPH_MSG_MDS_MAP: | |
9f95a23c | 2821 | handle_mds_map(ref_cast<MMDSMap>(m)); |
7c673cae FG |
2822 | break; |
2823 | case CEPH_MSG_FS_MAP: | |
9f95a23c | 2824 | handle_fs_map(ref_cast<MFSMap>(m)); |
7c673cae FG |
2825 | break; |
2826 | case CEPH_MSG_FS_MAP_USER: | |
9f95a23c | 2827 | handle_fs_map_user(ref_cast<MFSMapUser>(m)); |
7c673cae FG |
2828 | break; |
2829 | case CEPH_MSG_CLIENT_SESSION: | |
9f95a23c | 2830 | handle_client_session(ref_cast<MClientSession>(m)); |
7c673cae FG |
2831 | break; |
2832 | ||
2833 | case CEPH_MSG_OSD_MAP: | |
9f95a23c | 2834 | handle_osd_map(ref_cast<MOSDMap>(m)); |
7c673cae FG |
2835 | break; |
2836 | ||
2837 | // requests | |
2838 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: | |
9f95a23c | 2839 | handle_client_request_forward(ref_cast<MClientRequestForward>(m)); |
7c673cae FG |
2840 | break; |
2841 | case CEPH_MSG_CLIENT_REPLY: | |
9f95a23c | 2842 | handle_client_reply(ref_cast<MClientReply>(m)); |
11fdf7f2 TL |
2843 | break; |
2844 | ||
2845 | // reclaim reply | |
2846 | case CEPH_MSG_CLIENT_RECLAIM_REPLY: | |
9f95a23c | 2847 | handle_client_reclaim_reply(ref_cast<MClientReclaimReply>(m)); |
7c673cae FG |
2848 | break; |
2849 | ||
2850 | case CEPH_MSG_CLIENT_SNAP: | |
9f95a23c | 2851 | handle_snap(ref_cast<MClientSnap>(m)); |
7c673cae FG |
2852 | break; |
2853 | case CEPH_MSG_CLIENT_CAPS: | |
9f95a23c | 2854 | handle_caps(ref_cast<MClientCaps>(m)); |
7c673cae FG |
2855 | break; |
2856 | case CEPH_MSG_CLIENT_LEASE: | |
9f95a23c | 2857 | handle_lease(ref_cast<MClientLease>(m)); |
7c673cae FG |
2858 | break; |
2859 | case MSG_COMMAND_REPLY: | |
2860 | if (m->get_source().type() == CEPH_ENTITY_TYPE_MDS) { | |
9f95a23c | 2861 | handle_command_reply(ref_cast<MCommandReply>(m)); |
7c673cae FG |
2862 | } else { |
2863 | return false; | |
2864 | } | |
2865 | break; | |
2866 | case CEPH_MSG_CLIENT_QUOTA: | |
9f95a23c | 2867 | handle_quota(ref_cast<MClientQuota>(m)); |
7c673cae FG |
2868 | break; |
2869 | ||
2870 | default: | |
2871 | return false; | |
2872 | } | |
2873 | ||
2874 | // unmounting? | |
f67539c2 TL |
2875 | std::scoped_lock cl(client_lock); |
2876 | if (is_unmounting()) { | |
7c673cae FG |
2877 | ldout(cct, 10) << "unmounting: trim pass, size was " << lru.lru_get_size() |
2878 | << "+" << inode_map.size() << dendl; | |
f67539c2 | 2879 | uint64_t size = lru.lru_get_size() + inode_map.size(); |
7c673cae | 2880 | trim_cache(); |
f67539c2 | 2881 | if (size > lru.lru_get_size() + inode_map.size()) { |
7c673cae | 2882 | ldout(cct, 10) << "unmounting: trim pass, cache shrank, poking unmount()" << dendl; |
9f95a23c | 2883 | mount_cond.notify_all(); |
7c673cae FG |
2884 | } else { |
2885 | ldout(cct, 10) << "unmounting: trim pass, size still " << lru.lru_get_size() | |
2886 | << "+" << inode_map.size() << dendl; | |
2887 | } | |
2888 | } | |
2889 | ||
2890 | return true; | |
2891 | } | |
2892 | ||
11fdf7f2 | 2893 | void Client::handle_fs_map(const MConstRef<MFSMap>& m) |
7c673cae | 2894 | { |
f67539c2 | 2895 | std::scoped_lock cl(client_lock); |
7c673cae | 2896 | fsmap.reset(new FSMap(m->get_fsmap())); |
7c673cae FG |
2897 | |
2898 | signal_cond_list(waiting_for_fsmap); | |
2899 | ||
2900 | monclient->sub_got("fsmap", fsmap->get_epoch()); | |
2901 | } | |
2902 | ||
11fdf7f2 | 2903 | void Client::handle_fs_map_user(const MConstRef<MFSMapUser>& m) |
7c673cae | 2904 | { |
f67539c2 | 2905 | std::scoped_lock cl(client_lock); |
7c673cae FG |
2906 | fsmap_user.reset(new FSMapUser); |
2907 | *fsmap_user = m->get_fsmap(); | |
7c673cae FG |
2908 | |
2909 | monclient->sub_got("fsmap.user", fsmap_user->get_epoch()); | |
2910 | signal_cond_list(waiting_for_fsmap); | |
2911 | } | |
2912 | ||
f67539c2 TL |
2913 | // Cancel all the commands for missing or laggy GIDs |
2914 | void Client::cancel_commands(const MDSMap& newmap) | |
7c673cae | 2915 | { |
f67539c2 | 2916 | std::vector<ceph_tid_t> cancel_ops; |
7c673cae | 2917 | |
f67539c2 | 2918 | std::scoped_lock cmd_lock(command_lock); |
7c673cae | 2919 | auto &commands = command_table.get_commands(); |
f67539c2 | 2920 | for (const auto &[tid, op] : commands) { |
7c673cae | 2921 | const mds_gid_t op_mds_gid = op.mds_gid; |
f67539c2 TL |
2922 | if (newmap.is_dne_gid(op_mds_gid) || newmap.is_laggy_gid(op_mds_gid)) { |
2923 | ldout(cct, 1) << __func__ << ": cancelling command op " << tid << dendl; | |
2924 | cancel_ops.push_back(tid); | |
7c673cae FG |
2925 | if (op.outs) { |
2926 | std::ostringstream ss; | |
2927 | ss << "MDS " << op_mds_gid << " went away"; | |
2928 | *(op.outs) = ss.str(); | |
2929 | } | |
f67539c2 TL |
2930 | /* |
2931 | * No need to make the con->mark_down under | |
2932 | * client_lock here, because the con will | |
2933 | * has its own lock. | |
2934 | */ | |
7c673cae | 2935 | op.con->mark_down(); |
f67539c2 TL |
2936 | if (op.on_finish) |
2937 | op.on_finish->complete(-CEPHFS_ETIMEDOUT); | |
7c673cae FG |
2938 | } |
2939 | } | |
2940 | ||
f67539c2 TL |
2941 | for (const auto &tid : cancel_ops) |
2942 | command_table.erase(tid); | |
2943 | } | |
2944 | ||
2945 | void Client::handle_mds_map(const MConstRef<MMDSMap>& m) | |
2946 | { | |
2947 | std::unique_lock cl(client_lock); | |
2948 | if (m->get_epoch() <= mdsmap->get_epoch()) { | |
2949 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() | |
2950 | << " is identical to or older than our " | |
2951 | << mdsmap->get_epoch() << dendl; | |
2952 | return; | |
7c673cae FG |
2953 | } |
2954 | ||
f67539c2 TL |
2955 | cl.unlock(); |
2956 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() << dendl; | |
2957 | std::unique_ptr<MDSMap> _mdsmap(new MDSMap); | |
2958 | _mdsmap->decode(m->get_encoded()); | |
2959 | cancel_commands(*_mdsmap.get()); | |
2960 | cl.lock(); | |
2961 | ||
2962 | _mdsmap.swap(mdsmap); | |
2963 | ||
7c673cae | 2964 | // reset session |
11fdf7f2 | 2965 | for (auto p = mds_sessions.begin(); p != mds_sessions.end(); ) { |
7c673cae | 2966 | mds_rank_t mds = p->first; |
20effc67 | 2967 | MetaSessionRef session = p->second; |
7c673cae FG |
2968 | ++p; |
2969 | ||
f67539c2 | 2970 | int oldstate = _mdsmap->get_state(mds); |
7c673cae FG |
2971 | int newstate = mdsmap->get_state(mds); |
2972 | if (!mdsmap->is_up(mds)) { | |
2973 | session->con->mark_down(); | |
11fdf7f2 | 2974 | } else if (mdsmap->get_addrs(mds) != session->addrs) { |
f67539c2 TL |
2975 | auto old_inc = _mdsmap->get_incarnation(mds); |
2976 | auto new_inc = mdsmap->get_incarnation(mds); | |
f64942e4 AA |
2977 | if (old_inc != new_inc) { |
2978 | ldout(cct, 1) << "mds incarnation changed from " | |
2979 | << old_inc << " to " << new_inc << dendl; | |
2980 | oldstate = MDSMap::STATE_NULL; | |
2981 | } | |
7c673cae | 2982 | session->con->mark_down(); |
11fdf7f2 | 2983 | session->addrs = mdsmap->get_addrs(mds); |
7c673cae FG |
2984 | // When new MDS starts to take over, notify kernel to trim unused entries |
2985 | // in its dcache/icache. Hopefully, the kernel will release some unused | |
2986 | // inodes before the new MDS enters reconnect state. | |
20effc67 | 2987 | trim_cache_for_reconnect(session.get()); |
7c673cae FG |
2988 | } else if (oldstate == newstate) |
2989 | continue; // no change | |
f67539c2 | 2990 | |
7c673cae FG |
2991 | session->mds_state = newstate; |
2992 | if (newstate == MDSMap::STATE_RECONNECT) { | |
11fdf7f2 | 2993 | session->con = messenger->connect_to_mds(session->addrs); |
20effc67 | 2994 | send_reconnect(session.get()); |
81eedcae TL |
2995 | } else if (newstate > MDSMap::STATE_RECONNECT) { |
2996 | if (oldstate < MDSMap::STATE_RECONNECT) { | |
2997 | ldout(cct, 1) << "we may miss the MDSMap::RECONNECT, close mds session ... " << dendl; | |
20effc67 | 2998 | _closed_mds_session(session.get()); |
81eedcae TL |
2999 | continue; |
3000 | } | |
3001 | if (newstate >= MDSMap::STATE_ACTIVE) { | |
3002 | if (oldstate < MDSMap::STATE_ACTIVE) { | |
3003 | // kick new requests | |
20effc67 TL |
3004 | kick_requests(session.get()); |
3005 | kick_flushing_caps(session.get()); | |
81eedcae | 3006 | signal_context_list(session->waiting_for_open); |
20effc67 | 3007 | wake_up_session_caps(session.get(), true); |
81eedcae TL |
3008 | } |
3009 | connect_mds_targets(mds); | |
7c673cae | 3010 | } |
7c673cae FG |
3011 | } else if (newstate == MDSMap::STATE_NULL && |
3012 | mds >= mdsmap->get_max_mds()) { | |
20effc67 | 3013 | _closed_mds_session(session.get()); |
7c673cae FG |
3014 | } |
3015 | } | |
3016 | ||
3017 | // kick any waiting threads | |
3018 | signal_cond_list(waiting_for_mdsmap); | |
3019 | ||
7c673cae FG |
3020 | monclient->sub_got("mdsmap", mdsmap->get_epoch()); |
3021 | } | |
3022 | ||
3023 | void Client::send_reconnect(MetaSession *session) | |
3024 | { | |
3025 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 3026 | ldout(cct, 10) << __func__ << " to mds." << mds << dendl; |
7c673cae FG |
3027 | |
3028 | // trim unused caps to reduce MDS's cache rejoin time | |
3029 | trim_cache_for_reconnect(session); | |
3030 | ||
3031 | session->readonly = false; | |
3032 | ||
11fdf7f2 | 3033 | session->release.reset(); |
7c673cae FG |
3034 | |
3035 | // reset my cap seq number | |
3036 | session->seq = 0; | |
3037 | //connect to the mds' offload targets | |
3038 | connect_mds_targets(mds); | |
3039 | //make sure unsafe requests get saved | |
3040 | resend_unsafe_requests(session); | |
3041 | ||
11fdf7f2 TL |
3042 | early_kick_flushing_caps(session); |
3043 | ||
9f95a23c | 3044 | auto m = make_message<MClientReconnect>(); |
11fdf7f2 | 3045 | bool allow_multi = session->mds_features.test(CEPHFS_FEATURE_MULTI_RECONNECT); |
7c673cae FG |
3046 | |
3047 | // i have an open session. | |
3048 | ceph::unordered_set<inodeno_t> did_snaprealm; | |
3049 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator p = inode_map.begin(); | |
3050 | p != inode_map.end(); | |
3051 | ++p) { | |
3052 | Inode *in = p->second; | |
11fdf7f2 TL |
3053 | auto it = in->caps.find(mds); |
3054 | if (it != in->caps.end()) { | |
3055 | if (allow_multi && | |
9f95a23c TL |
3056 | m->get_approx_size() >= |
3057 | static_cast<size_t>((std::numeric_limits<int>::max() >> 1))) { | |
11fdf7f2 TL |
3058 | m->mark_more(); |
3059 | session->con->send_message2(std::move(m)); | |
3060 | ||
9f95a23c | 3061 | m = make_message<MClientReconnect>(); |
11fdf7f2 TL |
3062 | } |
3063 | ||
3064 | Cap &cap = it->second; | |
7c673cae | 3065 | ldout(cct, 10) << " caps on " << p->first |
11fdf7f2 | 3066 | << " " << ccap_string(cap.issued) |
7c673cae FG |
3067 | << " wants " << ccap_string(in->caps_wanted()) |
3068 | << dendl; | |
3069 | filepath path; | |
f91f0fd5 | 3070 | in->make_short_path(path); |
7c673cae FG |
3071 | ldout(cct, 10) << " path " << path << dendl; |
3072 | ||
3073 | bufferlist flockbl; | |
3074 | _encode_filelocks(in, flockbl); | |
3075 | ||
11fdf7f2 TL |
3076 | cap.seq = 0; // reset seq. |
3077 | cap.issue_seq = 0; // reset seq. | |
3078 | cap.mseq = 0; // reset seq. | |
3079 | // cap gen should catch up with session cap_gen | |
3080 | if (cap.gen < session->cap_gen) { | |
3081 | cap.gen = session->cap_gen; | |
3082 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
3083 | } else { | |
3084 | cap.issued = cap.implemented; | |
3085 | } | |
7c673cae FG |
3086 | snapid_t snap_follows = 0; |
3087 | if (!in->cap_snaps.empty()) | |
3088 | snap_follows = in->cap_snaps.begin()->first; | |
3089 | ||
3090 | m->add_cap(p->first.ino, | |
11fdf7f2 | 3091 | cap.cap_id, |
7c673cae FG |
3092 | path.get_ino(), path.get_path(), // ino |
3093 | in->caps_wanted(), // wanted | |
11fdf7f2 | 3094 | cap.issued, // issued |
7c673cae FG |
3095 | in->snaprealm->ino, |
3096 | snap_follows, | |
3097 | flockbl); | |
3098 | ||
3099 | if (did_snaprealm.count(in->snaprealm->ino) == 0) { | |
3100 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
3101 | m->add_snaprealm(in->snaprealm->ino, in->snaprealm->seq, in->snaprealm->parent); | |
3102 | did_snaprealm.insert(in->snaprealm->ino); | |
3103 | } | |
3104 | } | |
3105 | } | |
3106 | ||
11fdf7f2 TL |
3107 | if (!allow_multi) |
3108 | m->set_encoding_version(0); // use connection features to choose encoding | |
3109 | session->con->send_message2(std::move(m)); | |
7c673cae | 3110 | |
9f95a23c | 3111 | mount_cond.notify_all(); |
11fdf7f2 TL |
3112 | |
3113 | if (session->reclaim_state == MetaSession::RECLAIMING) | |
3114 | signal_cond_list(waiting_for_reclaim); | |
7c673cae FG |
3115 | } |
3116 | ||
3117 | ||
3118 | void Client::kick_requests(MetaSession *session) | |
3119 | { | |
11fdf7f2 | 3120 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3121 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3122 | p != mds_requests.end(); | |
3123 | ++p) { | |
31f18b77 FG |
3124 | MetaRequest *req = p->second; |
3125 | if (req->got_unsafe) | |
3126 | continue; | |
3127 | if (req->aborted()) { | |
3128 | if (req->caller_cond) { | |
3129 | req->kick = true; | |
9f95a23c | 3130 | req->caller_cond->notify_all(); |
31f18b77 | 3131 | } |
7c673cae | 3132 | continue; |
31f18b77 FG |
3133 | } |
3134 | if (req->retry_attempt > 0) | |
7c673cae | 3135 | continue; // new requests only |
31f18b77 | 3136 | if (req->mds == session->mds_num) { |
7c673cae FG |
3137 | send_request(p->second, session); |
3138 | } | |
3139 | } | |
3140 | } | |
3141 | ||
3142 | void Client::resend_unsafe_requests(MetaSession *session) | |
3143 | { | |
3144 | for (xlist<MetaRequest*>::iterator iter = session->unsafe_requests.begin(); | |
3145 | !iter.end(); | |
3146 | ++iter) | |
3147 | send_request(*iter, session); | |
3148 | ||
3149 | // also re-send old requests when MDS enters reconnect stage. So that MDS can | |
3150 | // process completed requests in clientreplay stage. | |
3151 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
3152 | p != mds_requests.end(); | |
3153 | ++p) { | |
3154 | MetaRequest *req = p->second; | |
3155 | if (req->got_unsafe) | |
3156 | continue; | |
31f18b77 FG |
3157 | if (req->aborted()) |
3158 | continue; | |
7c673cae FG |
3159 | if (req->retry_attempt == 0) |
3160 | continue; // old requests only | |
3161 | if (req->mds == session->mds_num) | |
3162 | send_request(req, session, true); | |
3163 | } | |
3164 | } | |
3165 | ||
3166 | void Client::wait_unsafe_requests() | |
3167 | { | |
3168 | list<MetaRequest*> last_unsafe_reqs; | |
11fdf7f2 | 3169 | for (const auto &p : mds_sessions) { |
20effc67 TL |
3170 | const auto s = p.second; |
3171 | if (!s->unsafe_requests.empty()) { | |
3172 | MetaRequest *req = s->unsafe_requests.back(); | |
7c673cae FG |
3173 | req->get(); |
3174 | last_unsafe_reqs.push_back(req); | |
3175 | } | |
3176 | } | |
3177 | ||
3178 | for (list<MetaRequest*>::iterator p = last_unsafe_reqs.begin(); | |
3179 | p != last_unsafe_reqs.end(); | |
3180 | ++p) { | |
3181 | MetaRequest *req = *p; | |
3182 | if (req->unsafe_item.is_on_list()) | |
3183 | wait_on_list(req->waitfor_safe); | |
3184 | put_request(req); | |
3185 | } | |
3186 | } | |
3187 | ||
3188 | void Client::kick_requests_closed(MetaSession *session) | |
3189 | { | |
11fdf7f2 | 3190 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3191 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3192 | p != mds_requests.end(); ) { | |
3193 | MetaRequest *req = p->second; | |
3194 | ++p; | |
3195 | if (req->mds == session->mds_num) { | |
3196 | if (req->caller_cond) { | |
3197 | req->kick = true; | |
9f95a23c | 3198 | req->caller_cond->notify_all(); |
7c673cae FG |
3199 | } |
3200 | req->item.remove_myself(); | |
3201 | if (req->got_unsafe) { | |
11fdf7f2 | 3202 | lderr(cct) << __func__ << " removing unsafe request " << req->get_tid() << dendl; |
7c673cae | 3203 | req->unsafe_item.remove_myself(); |
eafe8130 TL |
3204 | if (is_dir_operation(req)) { |
3205 | Inode *dir = req->inode(); | |
20effc67 | 3206 | ceph_assert(dir); |
f67539c2 | 3207 | dir->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3208 | lderr(cct) << "kick_requests_closed drop req of inode(dir) : " |
3209 | << dir->ino << " " << req->get_tid() << dendl; | |
3210 | req->unsafe_dir_item.remove_myself(); | |
3211 | } | |
3212 | if (req->target) { | |
3213 | InodeRef &in = req->target; | |
f67539c2 | 3214 | in->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3215 | lderr(cct) << "kick_requests_closed drop req of inode : " |
3216 | << in->ino << " " << req->get_tid() << dendl; | |
3217 | req->unsafe_target_item.remove_myself(); | |
3218 | } | |
7c673cae FG |
3219 | signal_cond_list(req->waitfor_safe); |
3220 | unregister_request(req); | |
3221 | } | |
3222 | } | |
3223 | } | |
11fdf7f2 TL |
3224 | ceph_assert(session->requests.empty()); |
3225 | ceph_assert(session->unsafe_requests.empty()); | |
7c673cae FG |
3226 | } |
3227 | ||
3228 | ||
3229 | ||
3230 | ||
3231 | /************ | |
3232 | * leases | |
3233 | */ | |
3234 | ||
3235 | void Client::got_mds_push(MetaSession *s) | |
3236 | { | |
3237 | s->seq++; | |
3238 | ldout(cct, 10) << " mds." << s->mds_num << " seq now " << s->seq << dendl; | |
3239 | if (s->state == MetaSession::STATE_CLOSING) { | |
9f95a23c | 3240 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
3241 | } |
3242 | } | |
3243 | ||
11fdf7f2 | 3244 | void Client::handle_lease(const MConstRef<MClientLease>& m) |
7c673cae | 3245 | { |
11fdf7f2 | 3246 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 3247 | |
11fdf7f2 | 3248 | ceph_assert(m->get_action() == CEPH_MDS_LEASE_REVOKE); |
7c673cae | 3249 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
3250 | |
3251 | std::scoped_lock cl(client_lock); | |
20effc67 | 3252 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 3253 | if (!session) { |
7c673cae FG |
3254 | return; |
3255 | } | |
3256 | ||
20effc67 | 3257 | got_mds_push(session.get()); |
7c673cae FG |
3258 | |
3259 | ceph_seq_t seq = m->get_seq(); | |
3260 | ||
3261 | Inode *in; | |
3262 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); | |
3263 | if (inode_map.count(vino) == 0) { | |
3264 | ldout(cct, 10) << " don't have vino " << vino << dendl; | |
3265 | goto revoke; | |
3266 | } | |
3267 | in = inode_map[vino]; | |
3268 | ||
9f95a23c | 3269 | if (m->get_mask() & CEPH_LEASE_VALID) { |
7c673cae FG |
3270 | if (!in->dir || in->dir->dentries.count(m->dname) == 0) { |
3271 | ldout(cct, 10) << " don't have dir|dentry " << m->get_ino() << "/" << m->dname <<dendl; | |
3272 | goto revoke; | |
3273 | } | |
3274 | Dentry *dn = in->dir->dentries[m->dname]; | |
3275 | ldout(cct, 10) << " revoked DN lease on " << dn << dendl; | |
3276 | dn->lease_mds = -1; | |
3277 | } | |
3278 | ||
3279 | revoke: | |
11fdf7f2 | 3280 | { |
9f95a23c TL |
3281 | auto reply = make_message<MClientLease>(CEPH_MDS_LEASE_RELEASE, seq, |
3282 | m->get_mask(), m->get_ino(), | |
3283 | m->get_first(), m->get_last(), m->dname); | |
11fdf7f2 TL |
3284 | m->get_connection()->send_message2(std::move(reply)); |
3285 | } | |
7c673cae FG |
3286 | } |
3287 | ||
f67539c2 | 3288 | void Client::_put_inode(Inode *in, int n) |
7c673cae | 3289 | { |
f67539c2 TL |
3290 | ldout(cct, 10) << __func__ << " on " << *in << " n = " << n << dendl; |
3291 | ||
b3b6e05e TL |
3292 | int left = in->get_nref(); |
3293 | ceph_assert(left >= n + 1); | |
3294 | in->iput(n); | |
3295 | left -= n; | |
3296 | if (left == 1) { // the last one will be held by the inode_map | |
7c673cae FG |
3297 | // release any caps |
3298 | remove_all_caps(in); | |
3299 | ||
11fdf7f2 | 3300 | ldout(cct, 10) << __func__ << " deleting " << *in << dendl; |
7c673cae | 3301 | bool unclean = objectcacher->release_set(&in->oset); |
11fdf7f2 | 3302 | ceph_assert(!unclean); |
7c673cae FG |
3303 | inode_map.erase(in->vino()); |
3304 | if (use_faked_inos()) | |
3305 | _release_faked_ino(in); | |
3306 | ||
b3b6e05e | 3307 | if (root == nullptr) { |
7c673cae FG |
3308 | root_ancestor = 0; |
3309 | while (!root_parents.empty()) | |
3310 | root_parents.erase(root_parents.begin()); | |
3311 | } | |
3312 | ||
b3b6e05e | 3313 | in->iput(); |
7c673cae FG |
3314 | } |
3315 | } | |
3316 | ||
f67539c2 TL |
3317 | void Client::delay_put_inodes(bool wakeup) |
3318 | { | |
3319 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
3320 | ||
3321 | std::map<Inode*,int> release; | |
3322 | { | |
3323 | std::scoped_lock dl(delay_i_lock); | |
3324 | release.swap(delay_i_release); | |
3325 | } | |
3326 | ||
3327 | if (release.empty()) | |
3328 | return; | |
3329 | ||
3330 | for (auto &[in, cnt] : release) | |
3331 | _put_inode(in, cnt); | |
3332 | ||
3333 | if (wakeup) | |
3334 | mount_cond.notify_all(); | |
3335 | } | |
3336 | ||
3337 | void Client::put_inode(Inode *in, int n) | |
3338 | { | |
3339 | ldout(cct, 20) << __func__ << " on " << *in << " n = " << n << dendl; | |
3340 | ||
3341 | std::scoped_lock dl(delay_i_lock); | |
3342 | delay_i_release[in] += n; | |
3343 | } | |
3344 | ||
7c673cae FG |
3345 | void Client::close_dir(Dir *dir) |
3346 | { | |
3347 | Inode *in = dir->parent_inode; | |
11fdf7f2 TL |
3348 | ldout(cct, 15) << __func__ << " dir " << dir << " on " << in << dendl; |
3349 | ceph_assert(dir->is_empty()); | |
3350 | ceph_assert(in->dir == dir); | |
3351 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked | |
3352 | if (!in->dentries.empty()) | |
7c673cae FG |
3353 | in->get_first_parent()->put(); // unpin dentry |
3354 | ||
3355 | delete in->dir; | |
3356 | in->dir = 0; | |
3357 | put_inode(in); // unpin inode | |
3358 | } | |
3359 | ||
3360 | /** | |
3361 | * Don't call this with in==NULL, use get_or_create for that | |
3362 | * leave dn set to default NULL unless you're trying to add | |
3363 | * a new inode to a pre-created Dentry | |
3364 | */ | |
3365 | Dentry* Client::link(Dir *dir, const string& name, Inode *in, Dentry *dn) | |
3366 | { | |
3367 | if (!dn) { | |
3368 | // create a new Dentry | |
11fdf7f2 TL |
3369 | dn = new Dentry(dir, name); |
3370 | ||
7c673cae FG |
3371 | lru.lru_insert_mid(dn); // mid or top? |
3372 | ||
3373 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in | |
3374 | << " dn " << dn << " (new dn)" << dendl; | |
3375 | } else { | |
11fdf7f2 | 3376 | ceph_assert(!dn->inode); |
7c673cae FG |
3377 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in |
3378 | << " dn " << dn << " (old dn)" << dendl; | |
3379 | } | |
3380 | ||
3381 | if (in) { // link to inode | |
11fdf7f2 | 3382 | InodeRef tmp_ref; |
7c673cae | 3383 | // only one parent for directories! |
11fdf7f2 TL |
3384 | if (in->is_dir() && !in->dentries.empty()) { |
3385 | tmp_ref = in; // prevent unlink below from freeing the inode. | |
7c673cae | 3386 | Dentry *olddn = in->get_first_parent(); |
11fdf7f2 | 3387 | ceph_assert(olddn->dir != dir || olddn->name != name); |
7c673cae | 3388 | Inode *old_diri = olddn->dir->parent_inode; |
7c673cae FG |
3389 | clear_dir_complete_and_ordered(old_diri, true); |
3390 | unlink(olddn, true, true); // keep dir, dentry | |
3391 | } | |
3392 | ||
11fdf7f2 | 3393 | dn->link(in); |
f67539c2 | 3394 | inc_dentry_nr(); |
11fdf7f2 | 3395 | ldout(cct, 20) << "link inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3396 | } |
3397 | ||
3398 | return dn; | |
3399 | } | |
3400 | ||
3401 | void Client::unlink(Dentry *dn, bool keepdir, bool keepdentry) | |
3402 | { | |
11fdf7f2 | 3403 | InodeRef in(dn->inode); |
7c673cae FG |
3404 | ldout(cct, 15) << "unlink dir " << dn->dir->parent_inode << " '" << dn->name << "' dn " << dn |
3405 | << " inode " << dn->inode << dendl; | |
3406 | ||
3407 | // unlink from inode | |
11fdf7f2 TL |
3408 | if (dn->inode) { |
3409 | dn->unlink(); | |
f67539c2 | 3410 | dec_dentry_nr(); |
11fdf7f2 | 3411 | ldout(cct, 20) << "unlink inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3412 | } |
3413 | ||
3414 | if (keepdentry) { | |
3415 | dn->lease_mds = -1; | |
3416 | } else { | |
3417 | ldout(cct, 15) << "unlink removing '" << dn->name << "' dn " << dn << dendl; | |
3418 | ||
3419 | // unlink from dir | |
11fdf7f2 TL |
3420 | Dir *dir = dn->dir; |
3421 | dn->detach(); | |
7c673cae FG |
3422 | |
3423 | // delete den | |
3424 | lru.lru_remove(dn); | |
3425 | dn->put(); | |
11fdf7f2 TL |
3426 | |
3427 | if (dir->is_empty() && !keepdir) | |
3428 | close_dir(dir); | |
7c673cae FG |
3429 | } |
3430 | } | |
3431 | ||
3432 | /** | |
3433 | * For asynchronous flushes, check for errors from the IO and | |
3434 | * update the inode if necessary | |
3435 | */ | |
3436 | class C_Client_FlushComplete : public Context { | |
3437 | private: | |
3438 | Client *client; | |
3439 | InodeRef inode; | |
3440 | public: | |
3441 | C_Client_FlushComplete(Client *c, Inode *in) : client(c), inode(in) { } | |
3442 | void finish(int r) override { | |
9f95a23c | 3443 | ceph_assert(ceph_mutex_is_locked_by_me(client->client_lock)); |
7c673cae FG |
3444 | if (r != 0) { |
3445 | client_t const whoami = client->whoami; // For the benefit of ldout prefix | |
3446 | ldout(client->cct, 1) << "I/O error from flush on inode " << inode | |
3447 | << " 0x" << std::hex << inode->ino << std::dec | |
3448 | << ": " << r << "(" << cpp_strerror(r) << ")" << dendl; | |
3449 | inode->set_async_err(r); | |
3450 | } | |
3451 | } | |
3452 | }; | |
3453 | ||
3454 | ||
3455 | /**** | |
3456 | * caps | |
3457 | */ | |
3458 | ||
3459 | void Client::get_cap_ref(Inode *in, int cap) | |
3460 | { | |
3461 | if ((cap & CEPH_CAP_FILE_BUFFER) && | |
3462 | in->cap_refs[CEPH_CAP_FILE_BUFFER] == 0) { | |
11fdf7f2 | 3463 | ldout(cct, 5) << __func__ << " got first FILE_BUFFER ref on " << *in << dendl; |
b3b6e05e | 3464 | in->iget(); |
7c673cae FG |
3465 | } |
3466 | if ((cap & CEPH_CAP_FILE_CACHE) && | |
3467 | in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
11fdf7f2 | 3468 | ldout(cct, 5) << __func__ << " got first FILE_CACHE ref on " << *in << dendl; |
b3b6e05e | 3469 | in->iget(); |
7c673cae FG |
3470 | } |
3471 | in->get_cap_ref(cap); | |
3472 | } | |
3473 | ||
3474 | void Client::put_cap_ref(Inode *in, int cap) | |
3475 | { | |
3476 | int last = in->put_cap_ref(cap); | |
3477 | if (last) { | |
3478 | int put_nref = 0; | |
3479 | int drop = last & ~in->caps_issued(); | |
3480 | if (in->snapid == CEPH_NOSNAP) { | |
f67539c2 | 3481 | if ((last & (CEPH_CAP_FILE_WR | CEPH_CAP_FILE_BUFFER)) && |
7c673cae FG |
3482 | !in->cap_snaps.empty() && |
3483 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3484 | ldout(cct, 10) << __func__ << " finishing pending cap_snap on " << *in << dendl; |
7c673cae FG |
3485 | in->cap_snaps.rbegin()->second.writing = 0; |
3486 | finish_cap_snap(in, in->cap_snaps.rbegin()->second, get_caps_used(in)); | |
3487 | signal_cond_list(in->waitfor_caps); // wake up blocked sync writers | |
3488 | } | |
3489 | if (last & CEPH_CAP_FILE_BUFFER) { | |
3490 | for (auto &p : in->cap_snaps) | |
3491 | p.second.dirty_data = 0; | |
3492 | signal_cond_list(in->waitfor_commit); | |
11fdf7f2 | 3493 | ldout(cct, 5) << __func__ << " dropped last FILE_BUFFER ref on " << *in << dendl; |
7c673cae FG |
3494 | ++put_nref; |
3495 | } | |
3496 | } | |
3497 | if (last & CEPH_CAP_FILE_CACHE) { | |
11fdf7f2 | 3498 | ldout(cct, 5) << __func__ << " dropped last FILE_CACHE ref on " << *in << dendl; |
7c673cae FG |
3499 | ++put_nref; |
3500 | } | |
3501 | if (drop) | |
3502 | check_caps(in, 0); | |
3503 | if (put_nref) | |
3504 | put_inode(in, put_nref); | |
3505 | } | |
3506 | } | |
3507 | ||
f67539c2 TL |
3508 | // get caps for a given file handle -- the inode should have @need caps |
3509 | // issued by the mds and @want caps not revoked (or not under revocation). | |
3510 | // this routine blocks till the cap requirement is satisfied. also account | |
3511 | // (track) for capability hit when required (when cap requirement succeedes). | |
f6b5b4d7 | 3512 | int Client::get_caps(Fh *fh, int need, int want, int *phave, loff_t endoff) |
7c673cae | 3513 | { |
f6b5b4d7 TL |
3514 | Inode *in = fh->inode.get(); |
3515 | ||
7c673cae FG |
3516 | int r = check_pool_perm(in, need); |
3517 | if (r < 0) | |
3518 | return r; | |
3519 | ||
3520 | while (1) { | |
3521 | int file_wanted = in->caps_file_wanted(); | |
3522 | if ((file_wanted & need) != need) { | |
3523 | ldout(cct, 10) << "get_caps " << *in << " need " << ccap_string(need) | |
3524 | << " file_wanted " << ccap_string(file_wanted) << ", EBADF " | |
3525 | << dendl; | |
f67539c2 | 3526 | return -CEPHFS_EBADF; |
7c673cae FG |
3527 | } |
3528 | ||
f6b5b4d7 | 3529 | if ((fh->mode & CEPH_FILE_MODE_WR) && fh->gen != fd_gen) |
f67539c2 | 3530 | return -CEPHFS_EBADF; |
f6b5b4d7 TL |
3531 | |
3532 | if ((in->flags & I_ERROR_FILELOCK) && fh->has_any_filelocks()) | |
f67539c2 | 3533 | return -CEPHFS_EIO; |
f6b5b4d7 | 3534 | |
7c673cae FG |
3535 | int implemented; |
3536 | int have = in->caps_issued(&implemented); | |
3537 | ||
3538 | bool waitfor_caps = false; | |
3539 | bool waitfor_commit = false; | |
3540 | ||
3541 | if (have & need & CEPH_CAP_FILE_WR) { | |
1911f103 TL |
3542 | if (endoff > 0) { |
3543 | if ((endoff >= (loff_t)in->max_size || | |
3544 | endoff > (loff_t)(in->size << 1)) && | |
3545 | endoff > (loff_t)in->wanted_max_size) { | |
3546 | ldout(cct, 10) << "wanted_max_size " << in->wanted_max_size << " -> " << endoff << dendl; | |
3547 | in->wanted_max_size = endoff; | |
3548 | } | |
3549 | if (in->wanted_max_size > in->max_size && | |
3550 | in->wanted_max_size > in->requested_max_size) | |
3551 | check_caps(in, 0); | |
7c673cae FG |
3552 | } |
3553 | ||
3554 | if (endoff >= 0 && endoff > (loff_t)in->max_size) { | |
3555 | ldout(cct, 10) << "waiting on max_size, endoff " << endoff << " max_size " << in->max_size << " on " << *in << dendl; | |
3556 | waitfor_caps = true; | |
3557 | } | |
3558 | if (!in->cap_snaps.empty()) { | |
3559 | if (in->cap_snaps.rbegin()->second.writing) { | |
3560 | ldout(cct, 10) << "waiting on cap_snap write to complete" << dendl; | |
3561 | waitfor_caps = true; | |
3562 | } | |
3563 | for (auto &p : in->cap_snaps) { | |
3564 | if (p.second.dirty_data) { | |
3565 | waitfor_commit = true; | |
3566 | break; | |
3567 | } | |
3568 | } | |
3569 | if (waitfor_commit) { | |
3570 | _flush(in, new C_Client_FlushComplete(this, in)); | |
3571 | ldout(cct, 10) << "waiting for WRBUFFER to get dropped" << dendl; | |
3572 | } | |
3573 | } | |
3574 | } | |
3575 | ||
3576 | if (!waitfor_caps && !waitfor_commit) { | |
3577 | if ((have & need) == need) { | |
7c673cae FG |
3578 | int revoking = implemented & ~have; |
3579 | ldout(cct, 10) << "get_caps " << *in << " have " << ccap_string(have) | |
3580 | << " need " << ccap_string(need) << " want " << ccap_string(want) | |
c07f9fc5 | 3581 | << " revoking " << ccap_string(revoking) |
7c673cae | 3582 | << dendl; |
c07f9fc5 | 3583 | if ((revoking & want) == 0) { |
7c673cae FG |
3584 | *phave = need | (have & want); |
3585 | in->get_cap_ref(need); | |
f67539c2 | 3586 | cap_hit(); |
7c673cae FG |
3587 | return 0; |
3588 | } | |
3589 | } | |
3590 | ldout(cct, 10) << "waiting for caps " << *in << " need " << ccap_string(need) << " want " << ccap_string(want) << dendl; | |
3591 | waitfor_caps = true; | |
3592 | } | |
3593 | ||
3594 | if ((need & CEPH_CAP_FILE_WR) && in->auth_cap && | |
3595 | in->auth_cap->session->readonly) | |
f67539c2 | 3596 | return -CEPHFS_EROFS; |
7c673cae FG |
3597 | |
3598 | if (in->flags & I_CAP_DROPPED) { | |
3599 | int mds_wanted = in->caps_mds_wanted(); | |
3600 | if ((mds_wanted & need) != need) { | |
3601 | int ret = _renew_caps(in); | |
3602 | if (ret < 0) | |
3603 | return ret; | |
3604 | continue; | |
3605 | } | |
a8e16298 | 3606 | if (!(file_wanted & ~mds_wanted)) |
7c673cae | 3607 | in->flags &= ~I_CAP_DROPPED; |
7c673cae FG |
3608 | } |
3609 | ||
3610 | if (waitfor_caps) | |
3611 | wait_on_list(in->waitfor_caps); | |
3612 | else if (waitfor_commit) | |
3613 | wait_on_list(in->waitfor_commit); | |
3614 | } | |
3615 | } | |
3616 | ||
3617 | int Client::get_caps_used(Inode *in) | |
3618 | { | |
3619 | unsigned used = in->caps_used(); | |
3620 | if (!(used & CEPH_CAP_FILE_CACHE) && | |
3621 | !objectcacher->set_is_empty(&in->oset)) | |
3622 | used |= CEPH_CAP_FILE_CACHE; | |
3623 | return used; | |
3624 | } | |
3625 | ||
3626 | void Client::cap_delay_requeue(Inode *in) | |
3627 | { | |
11fdf7f2 | 3628 | ldout(cct, 10) << __func__ << " on " << *in << dendl; |
2a845540 TL |
3629 | |
3630 | in->hold_caps_until = ceph::coarse_mono_clock::now() + caps_release_delay; | |
28e407b8 | 3631 | delayed_list.push_back(&in->delay_cap_item); |
7c673cae FG |
3632 | } |
3633 | ||
3634 | void Client::send_cap(Inode *in, MetaSession *session, Cap *cap, | |
eafe8130 | 3635 | int flags, int used, int want, int retain, |
7c673cae FG |
3636 | int flush, ceph_tid_t flush_tid) |
3637 | { | |
3638 | int held = cap->issued | cap->implemented; | |
3639 | int revoking = cap->implemented & ~cap->issued; | |
3640 | retain &= ~revoking; | |
3641 | int dropping = cap->issued & ~retain; | |
3642 | int op = CEPH_CAP_OP_UPDATE; | |
3643 | ||
11fdf7f2 | 3644 | ldout(cct, 10) << __func__ << " " << *in |
7c673cae | 3645 | << " mds." << session->mds_num << " seq " << cap->seq |
7c673cae FG |
3646 | << " used " << ccap_string(used) |
3647 | << " want " << ccap_string(want) | |
3648 | << " flush " << ccap_string(flush) | |
3649 | << " retain " << ccap_string(retain) | |
3650 | << " held "<< ccap_string(held) | |
3651 | << " revoking " << ccap_string(revoking) | |
3652 | << " dropping " << ccap_string(dropping) | |
3653 | << dendl; | |
3654 | ||
3655 | if (cct->_conf->client_inject_release_failure && revoking) { | |
3656 | const int would_have_issued = cap->issued & retain; | |
3657 | const int would_have_implemented = cap->implemented & (cap->issued | used); | |
3658 | // Simulated bug: | |
3659 | // - tell the server we think issued is whatever they issued plus whatever we implemented | |
3660 | // - leave what we have implemented in place | |
3661 | ldout(cct, 20) << __func__ << " injecting failure to release caps" << dendl; | |
3662 | cap->issued = cap->issued | cap->implemented; | |
3663 | ||
3664 | // Make an exception for revoking xattr caps: we are injecting | |
3665 | // failure to release other caps, but allow xattr because client | |
3666 | // will block on xattr ops if it can't release these to MDS (#9800) | |
3667 | const int xattr_mask = CEPH_CAP_XATTR_SHARED | CEPH_CAP_XATTR_EXCL; | |
3668 | cap->issued ^= xattr_mask & revoking; | |
3669 | cap->implemented ^= xattr_mask & revoking; | |
3670 | ||
3671 | ldout(cct, 20) << __func__ << " issued " << ccap_string(cap->issued) << " vs " << ccap_string(would_have_issued) << dendl; | |
3672 | ldout(cct, 20) << __func__ << " implemented " << ccap_string(cap->implemented) << " vs " << ccap_string(would_have_implemented) << dendl; | |
3673 | } else { | |
3674 | // Normal behaviour | |
3675 | cap->issued &= retain; | |
3676 | cap->implemented &= cap->issued | used; | |
3677 | } | |
3678 | ||
3679 | snapid_t follows = 0; | |
3680 | ||
3681 | if (flush) | |
3682 | follows = in->snaprealm->get_snap_context().seq; | |
20effc67 | 3683 | |
9f95a23c | 3684 | auto m = make_message<MClientCaps>(op, |
7c673cae FG |
3685 | in->ino, |
3686 | 0, | |
3687 | cap->cap_id, cap->seq, | |
3688 | cap->implemented, | |
3689 | want, | |
3690 | flush, | |
3691 | cap->mseq, | |
3692 | cap_epoch_barrier); | |
3693 | m->caller_uid = in->cap_dirtier_uid; | |
3694 | m->caller_gid = in->cap_dirtier_gid; | |
3695 | ||
3696 | m->head.issue_seq = cap->issue_seq; | |
3697 | m->set_tid(flush_tid); | |
3698 | ||
3699 | m->head.uid = in->uid; | |
3700 | m->head.gid = in->gid; | |
3701 | m->head.mode = in->mode; | |
20effc67 | 3702 | |
7c673cae | 3703 | m->head.nlink = in->nlink; |
20effc67 | 3704 | |
7c673cae | 3705 | if (flush & CEPH_CAP_XATTR_EXCL) { |
11fdf7f2 | 3706 | encode(in->xattrs, m->xattrbl); |
7c673cae FG |
3707 | m->head.xattr_version = in->xattr_version; |
3708 | } | |
20effc67 | 3709 | |
7c673cae FG |
3710 | m->size = in->size; |
3711 | m->max_size = in->max_size; | |
3712 | m->truncate_seq = in->truncate_seq; | |
3713 | m->truncate_size = in->truncate_size; | |
3714 | m->mtime = in->mtime; | |
3715 | m->atime = in->atime; | |
3716 | m->ctime = in->ctime; | |
3717 | m->btime = in->btime; | |
3718 | m->time_warp_seq = in->time_warp_seq; | |
3719 | m->change_attr = in->change_attr; | |
eafe8130 TL |
3720 | |
3721 | if (!(flags & MClientCaps::FLAG_PENDING_CAPSNAP) && | |
3722 | !in->cap_snaps.empty() && | |
3723 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3724 | flags |= MClientCaps::FLAG_PENDING_CAPSNAP; | |
3725 | m->flags = flags; | |
3726 | ||
7c673cae FG |
3727 | if (flush & CEPH_CAP_FILE_WR) { |
3728 | m->inline_version = in->inline_version; | |
3729 | m->inline_data = in->inline_data; | |
3730 | } | |
3731 | ||
3732 | in->reported_size = in->size; | |
3733 | m->set_snap_follows(follows); | |
3734 | cap->wanted = want; | |
3735 | if (cap == in->auth_cap) { | |
1911f103 TL |
3736 | if (want & CEPH_CAP_ANY_FILE_WR) { |
3737 | m->set_max_size(in->wanted_max_size); | |
3738 | in->requested_max_size = in->wanted_max_size; | |
3739 | ldout(cct, 15) << "auth cap, requesting max_size " << in->requested_max_size << dendl; | |
3740 | } else { | |
3741 | in->requested_max_size = 0; | |
3742 | ldout(cct, 15) << "auth cap, reset requested_max_size due to not wanting any file write cap" << dendl; | |
3743 | } | |
7c673cae FG |
3744 | } |
3745 | ||
3746 | if (!session->flushing_caps_tids.empty()) | |
3747 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3748 | ||
11fdf7f2 | 3749 | session->con->send_message2(std::move(m)); |
7c673cae FG |
3750 | } |
3751 | ||
31f18b77 FG |
3752 | static bool is_max_size_approaching(Inode *in) |
3753 | { | |
3754 | /* mds will adjust max size according to the reported size */ | |
3755 | if (in->flushing_caps & CEPH_CAP_FILE_WR) | |
3756 | return false; | |
3757 | if (in->size >= in->max_size) | |
3758 | return true; | |
3759 | /* half of previous max_size increment has been used */ | |
3760 | if (in->max_size > in->reported_size && | |
3761 | (in->size << 1) >= in->max_size + in->reported_size) | |
3762 | return true; | |
3763 | return false; | |
3764 | } | |
7c673cae | 3765 | |
11fdf7f2 TL |
3766 | static int adjust_caps_used_for_lazyio(int used, int issued, int implemented) |
3767 | { | |
3768 | if (!(used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER))) | |
3769 | return used; | |
3770 | if (!(implemented & CEPH_CAP_FILE_LAZYIO)) | |
3771 | return used; | |
3772 | ||
3773 | if (issued & CEPH_CAP_FILE_LAZYIO) { | |
3774 | if (!(issued & CEPH_CAP_FILE_CACHE)) { | |
3775 | used &= ~CEPH_CAP_FILE_CACHE; | |
3776 | used |= CEPH_CAP_FILE_LAZYIO; | |
3777 | } | |
3778 | if (!(issued & CEPH_CAP_FILE_BUFFER)) { | |
3779 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3780 | used |= CEPH_CAP_FILE_LAZYIO; | |
3781 | } | |
3782 | } else { | |
3783 | if (!(implemented & CEPH_CAP_FILE_CACHE)) { | |
3784 | used &= ~CEPH_CAP_FILE_CACHE; | |
3785 | used |= CEPH_CAP_FILE_LAZYIO; | |
3786 | } | |
3787 | if (!(implemented & CEPH_CAP_FILE_BUFFER)) { | |
3788 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3789 | used |= CEPH_CAP_FILE_LAZYIO; | |
3790 | } | |
3791 | } | |
3792 | return used; | |
3793 | } | |
3794 | ||
7c673cae FG |
3795 | /** |
3796 | * check_caps | |
3797 | * | |
3798 | * Examine currently used and wanted versus held caps. Release, flush or ack | |
3799 | * revoked caps to the MDS as appropriate. | |
3800 | * | |
3801 | * @param in the inode to check | |
3802 | * @param flags flags to apply to cap check | |
3803 | */ | |
3804 | void Client::check_caps(Inode *in, unsigned flags) | |
3805 | { | |
3806 | unsigned wanted = in->caps_wanted(); | |
3807 | unsigned used = get_caps_used(in); | |
3808 | unsigned cap_used; | |
3809 | ||
7c673cae FG |
3810 | int implemented; |
3811 | int issued = in->caps_issued(&implemented); | |
3812 | int revoking = implemented & ~issued; | |
3813 | ||
11fdf7f2 TL |
3814 | int orig_used = used; |
3815 | used = adjust_caps_used_for_lazyio(used, issued, implemented); | |
3816 | ||
7c673cae | 3817 | int retain = wanted | used | CEPH_CAP_PIN; |
f67539c2 | 3818 | if (!is_unmounting() && in->nlink > 0) { |
a8e16298 | 3819 | if (wanted) { |
7c673cae | 3820 | retain |= CEPH_CAP_ANY; |
a8e16298 TL |
3821 | } else if (in->is_dir() && |
3822 | (issued & CEPH_CAP_FILE_SHARED) && | |
3823 | (in->flags & I_COMPLETE)) { | |
3824 | // we do this here because we don't want to drop to Fs (and then | |
3825 | // drop the Fs if we do a create!) if that alone makes us send lookups | |
3826 | // to the MDS. Doing it in in->caps_wanted() has knock-on effects elsewhere | |
3827 | wanted = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; | |
3828 | retain |= wanted; | |
3829 | } else { | |
7c673cae | 3830 | retain |= CEPH_CAP_ANY_SHARED; |
a8e16298 TL |
3831 | // keep RD only if we didn't have the file open RW, |
3832 | // because then the mds would revoke it anyway to | |
3833 | // journal max_size=0. | |
3834 | if (in->max_size == 0) | |
3835 | retain |= CEPH_CAP_ANY_RD; | |
3836 | } | |
7c673cae FG |
3837 | } |
3838 | ||
11fdf7f2 | 3839 | ldout(cct, 10) << __func__ << " on " << *in |
7c673cae FG |
3840 | << " wanted " << ccap_string(wanted) |
3841 | << " used " << ccap_string(used) | |
3842 | << " issued " << ccap_string(issued) | |
3843 | << " revoking " << ccap_string(revoking) | |
3844 | << " flags=" << flags | |
3845 | << dendl; | |
3846 | ||
3847 | if (in->snapid != CEPH_NOSNAP) | |
3848 | return; //snap caps last forever, can't write | |
3849 | ||
3850 | if (in->caps.empty()) | |
3851 | return; // guard if at end of func | |
3852 | ||
11fdf7f2 TL |
3853 | if (!(orig_used & CEPH_CAP_FILE_BUFFER) && |
3854 | (revoking & used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
94b18763 | 3855 | if (_release(in)) |
11fdf7f2 | 3856 | used &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
94b18763 | 3857 | } |
7c673cae | 3858 | |
20effc67 TL |
3859 | for (auto &[mds, cap] : in->caps) { |
3860 | auto session = mds_sessions.at(mds); | |
7c673cae FG |
3861 | |
3862 | cap_used = used; | |
11fdf7f2 | 3863 | if (in->auth_cap && &cap != in->auth_cap) |
7c673cae FG |
3864 | cap_used &= ~in->auth_cap->issued; |
3865 | ||
11fdf7f2 | 3866 | revoking = cap.implemented & ~cap.issued; |
20effc67 | 3867 | |
7c673cae | 3868 | ldout(cct, 10) << " cap mds." << mds |
11fdf7f2 TL |
3869 | << " issued " << ccap_string(cap.issued) |
3870 | << " implemented " << ccap_string(cap.implemented) | |
7c673cae FG |
3871 | << " revoking " << ccap_string(revoking) << dendl; |
3872 | ||
3873 | if (in->wanted_max_size > in->max_size && | |
3874 | in->wanted_max_size > in->requested_max_size && | |
11fdf7f2 | 3875 | &cap == in->auth_cap) |
7c673cae FG |
3876 | goto ack; |
3877 | ||
3878 | /* approaching file_max? */ | |
11fdf7f2 TL |
3879 | if ((cap.issued & CEPH_CAP_FILE_WR) && |
3880 | &cap == in->auth_cap && | |
31f18b77 | 3881 | is_max_size_approaching(in)) { |
7c673cae | 3882 | ldout(cct, 10) << "size " << in->size << " approaching max_size " << in->max_size |
31f18b77 | 3883 | << ", reported " << in->reported_size << dendl; |
7c673cae FG |
3884 | goto ack; |
3885 | } | |
3886 | ||
3887 | /* completed revocation? */ | |
3888 | if (revoking && (revoking & cap_used) == 0) { | |
11fdf7f2 | 3889 | ldout(cct, 10) << "completed revocation of " << ccap_string(cap.implemented & ~cap.issued) << dendl; |
7c673cae FG |
3890 | goto ack; |
3891 | } | |
3892 | ||
3893 | /* want more caps from mds? */ | |
11fdf7f2 | 3894 | if (wanted & ~(cap.wanted | cap.issued)) |
7c673cae FG |
3895 | goto ack; |
3896 | ||
f67539c2 | 3897 | if (!revoking && is_unmounting() && (cap_used == 0)) |
7c673cae FG |
3898 | goto ack; |
3899 | ||
11fdf7f2 | 3900 | if ((cap.issued & ~retain) == 0 && // and we don't have anything we wouldn't like |
a8e16298 | 3901 | !in->dirty_caps) // and we have no dirty caps |
7c673cae FG |
3902 | continue; |
3903 | ||
11fdf7f2 | 3904 | if (!(flags & CHECK_CAPS_NODELAY)) { |
7c673cae | 3905 | ldout(cct, 10) << "delaying cap release" << dendl; |
11fdf7f2 | 3906 | cap_delay_requeue(in); |
7c673cae FG |
3907 | continue; |
3908 | } | |
3909 | ||
3910 | ack: | |
eafe8130 TL |
3911 | if (&cap == in->auth_cap) { |
3912 | if (in->flags & I_KICK_FLUSH) { | |
3913 | ldout(cct, 20) << " reflushing caps (check_caps) on " << *in | |
3914 | << " to mds." << mds << dendl; | |
20effc67 | 3915 | kick_flushing_caps(in, session.get()); |
eafe8130 TL |
3916 | } |
3917 | if (!in->cap_snaps.empty() && | |
3918 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3919 | flush_snaps(in); | |
7c673cae FG |
3920 | } |
3921 | ||
3922 | int flushing; | |
e306af50 | 3923 | int msg_flags = 0; |
7c673cae | 3924 | ceph_tid_t flush_tid; |
11fdf7f2 | 3925 | if (in->auth_cap == &cap && in->dirty_caps) { |
7c673cae | 3926 | flushing = mark_caps_flushing(in, &flush_tid); |
e306af50 TL |
3927 | if (flags & CHECK_CAPS_SYNCHRONOUS) |
3928 | msg_flags |= MClientCaps::FLAG_SYNC; | |
7c673cae FG |
3929 | } else { |
3930 | flushing = 0; | |
3931 | flush_tid = 0; | |
3932 | } | |
3933 | ||
20effc67 TL |
3934 | in->delay_cap_item.remove_myself(); |
3935 | send_cap(in, session.get(), &cap, msg_flags, cap_used, wanted, retain, | |
eafe8130 | 3936 | flushing, flush_tid); |
7c673cae FG |
3937 | } |
3938 | } | |
3939 | ||
3940 | ||
3941 | void Client::queue_cap_snap(Inode *in, SnapContext& old_snapc) | |
3942 | { | |
3943 | int used = get_caps_used(in); | |
3944 | int dirty = in->caps_dirty(); | |
11fdf7f2 | 3945 | ldout(cct, 10) << __func__ << " " << *in << " snapc " << old_snapc << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3946 | |
3947 | if (in->cap_snaps.size() && | |
3948 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3949 | ldout(cct, 10) << __func__ << " already have pending cap_snap on " << *in << dendl; |
7c673cae FG |
3950 | return; |
3951 | } else if (in->caps_dirty() || | |
3952 | (used & CEPH_CAP_FILE_WR) || | |
3953 | (dirty & CEPH_CAP_ANY_WR)) { | |
3954 | const auto &capsnapem = in->cap_snaps.emplace(std::piecewise_construct, std::make_tuple(old_snapc.seq), std::make_tuple(in)); | |
11fdf7f2 | 3955 | ceph_assert(capsnapem.second); /* element inserted */ |
7c673cae FG |
3956 | CapSnap &capsnap = capsnapem.first->second; |
3957 | capsnap.context = old_snapc; | |
3958 | capsnap.issued = in->caps_issued(); | |
3959 | capsnap.dirty = in->caps_dirty(); | |
f67539c2 | 3960 | |
7c673cae | 3961 | capsnap.dirty_data = (used & CEPH_CAP_FILE_BUFFER); |
f67539c2 | 3962 | |
7c673cae FG |
3963 | capsnap.uid = in->uid; |
3964 | capsnap.gid = in->gid; | |
3965 | capsnap.mode = in->mode; | |
3966 | capsnap.btime = in->btime; | |
3967 | capsnap.xattrs = in->xattrs; | |
3968 | capsnap.xattr_version = in->xattr_version; | |
11fdf7f2 TL |
3969 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; |
3970 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
f67539c2 | 3971 | |
7c673cae | 3972 | if (used & CEPH_CAP_FILE_WR) { |
11fdf7f2 | 3973 | ldout(cct, 10) << __func__ << " WR used on " << *in << dendl; |
7c673cae FG |
3974 | capsnap.writing = 1; |
3975 | } else { | |
3976 | finish_cap_snap(in, capsnap, used); | |
3977 | } | |
3978 | } else { | |
11fdf7f2 | 3979 | ldout(cct, 10) << __func__ << " not dirty|writing on " << *in << dendl; |
7c673cae FG |
3980 | } |
3981 | } | |
3982 | ||
3983 | void Client::finish_cap_snap(Inode *in, CapSnap &capsnap, int used) | |
3984 | { | |
11fdf7f2 | 3985 | ldout(cct, 10) << __func__ << " " << *in << " capsnap " << (void *)&capsnap << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3986 | capsnap.size = in->size; |
3987 | capsnap.mtime = in->mtime; | |
3988 | capsnap.atime = in->atime; | |
3989 | capsnap.ctime = in->ctime; | |
3990 | capsnap.time_warp_seq = in->time_warp_seq; | |
3991 | capsnap.change_attr = in->change_attr; | |
7c673cae FG |
3992 | capsnap.dirty |= in->caps_dirty(); |
3993 | ||
11fdf7f2 TL |
3994 | /* Only reset it if it wasn't set before */ |
3995 | if (capsnap.cap_dirtier_uid == -1) { | |
3996 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; | |
3997 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
3998 | } | |
3999 | ||
7c673cae FG |
4000 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { |
4001 | capsnap.inline_data = in->inline_data; | |
4002 | capsnap.inline_version = in->inline_version; | |
4003 | } | |
4004 | ||
4005 | if (used & CEPH_CAP_FILE_BUFFER) { | |
f67539c2 | 4006 | capsnap.writing = 1; |
11fdf7f2 | 4007 | ldout(cct, 10) << __func__ << " " << *in << " cap_snap " << &capsnap << " used " << used |
7c673cae FG |
4008 | << " WRBUFFER, delaying" << dendl; |
4009 | } else { | |
4010 | capsnap.dirty_data = 0; | |
4011 | flush_snaps(in); | |
4012 | } | |
4013 | } | |
4014 | ||
eafe8130 TL |
4015 | void Client::send_flush_snap(Inode *in, MetaSession *session, |
4016 | snapid_t follows, CapSnap& capsnap) | |
4017 | { | |
9f95a23c TL |
4018 | auto m = make_message<MClientCaps>(CEPH_CAP_OP_FLUSHSNAP, |
4019 | in->ino, in->snaprealm->ino, 0, | |
4020 | in->auth_cap->mseq, cap_epoch_barrier); | |
eafe8130 TL |
4021 | m->caller_uid = capsnap.cap_dirtier_uid; |
4022 | m->caller_gid = capsnap.cap_dirtier_gid; | |
4023 | ||
4024 | m->set_client_tid(capsnap.flush_tid); | |
4025 | m->head.snap_follows = follows; | |
4026 | ||
4027 | m->head.caps = capsnap.issued; | |
4028 | m->head.dirty = capsnap.dirty; | |
4029 | ||
4030 | m->head.uid = capsnap.uid; | |
4031 | m->head.gid = capsnap.gid; | |
4032 | m->head.mode = capsnap.mode; | |
4033 | m->btime = capsnap.btime; | |
4034 | ||
4035 | m->size = capsnap.size; | |
4036 | ||
4037 | m->head.xattr_version = capsnap.xattr_version; | |
4038 | encode(capsnap.xattrs, m->xattrbl); | |
4039 | ||
4040 | m->ctime = capsnap.ctime; | |
4041 | m->btime = capsnap.btime; | |
4042 | m->mtime = capsnap.mtime; | |
4043 | m->atime = capsnap.atime; | |
4044 | m->time_warp_seq = capsnap.time_warp_seq; | |
4045 | m->change_attr = capsnap.change_attr; | |
4046 | ||
4047 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { | |
4048 | m->inline_version = in->inline_version; | |
4049 | m->inline_data = in->inline_data; | |
4050 | } | |
4051 | ||
4052 | ceph_assert(!session->flushing_caps_tids.empty()); | |
4053 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
4054 | ||
4055 | session->con->send_message2(std::move(m)); | |
4056 | } | |
4057 | ||
4058 | void Client::flush_snaps(Inode *in) | |
7c673cae | 4059 | { |
eafe8130 | 4060 | ldout(cct, 10) << "flush_snaps on " << *in << dendl; |
11fdf7f2 | 4061 | ceph_assert(in->cap_snaps.size()); |
7c673cae FG |
4062 | |
4063 | // pick auth mds | |
11fdf7f2 | 4064 | ceph_assert(in->auth_cap); |
7c673cae | 4065 | MetaSession *session = in->auth_cap->session; |
7c673cae FG |
4066 | |
4067 | for (auto &p : in->cap_snaps) { | |
4068 | CapSnap &capsnap = p.second; | |
eafe8130 TL |
4069 | // only do new flush |
4070 | if (capsnap.flush_tid > 0) | |
4071 | continue; | |
7c673cae FG |
4072 | |
4073 | ldout(cct, 10) << "flush_snaps mds." << session->mds_num | |
4074 | << " follows " << p.first | |
4075 | << " size " << capsnap.size | |
4076 | << " mtime " << capsnap.mtime | |
4077 | << " dirty_data=" << capsnap.dirty_data | |
4078 | << " writing=" << capsnap.writing | |
4079 | << " on " << *in << dendl; | |
4080 | if (capsnap.dirty_data || capsnap.writing) | |
eafe8130 | 4081 | break; |
f67539c2 | 4082 | |
eafe8130 TL |
4083 | capsnap.flush_tid = ++last_flush_tid; |
4084 | session->flushing_caps_tids.insert(capsnap.flush_tid); | |
4085 | in->flushing_cap_tids[capsnap.flush_tid] = 0; | |
4086 | if (!in->flushing_cap_item.is_on_list()) | |
4087 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
7c673cae | 4088 | |
eafe8130 | 4089 | send_flush_snap(in, session, p.first, capsnap); |
7c673cae FG |
4090 | } |
4091 | } | |
4092 | ||
9f95a23c | 4093 | void Client::wait_on_list(list<ceph::condition_variable*>& ls) |
7c673cae | 4094 | { |
9f95a23c | 4095 | ceph::condition_variable cond; |
7c673cae | 4096 | ls.push_back(&cond); |
9f95a23c TL |
4097 | std::unique_lock l{client_lock, std::adopt_lock}; |
4098 | cond.wait(l); | |
4099 | l.release(); | |
7c673cae FG |
4100 | ls.remove(&cond); |
4101 | } | |
4102 | ||
9f95a23c | 4103 | void Client::signal_cond_list(list<ceph::condition_variable*>& ls) |
7c673cae | 4104 | { |
9f95a23c TL |
4105 | for (auto cond : ls) { |
4106 | cond->notify_all(); | |
4107 | } | |
7c673cae FG |
4108 | } |
4109 | ||
4110 | void Client::wait_on_context_list(list<Context*>& ls) | |
4111 | { | |
9f95a23c | 4112 | ceph::condition_variable cond; |
7c673cae FG |
4113 | bool done = false; |
4114 | int r; | |
9f95a23c TL |
4115 | ls.push_back(new C_Cond(cond, &done, &r)); |
4116 | std::unique_lock l{client_lock, std::adopt_lock}; | |
4117 | cond.wait(l, [&done] { return done;}); | |
4118 | l.release(); | |
7c673cae FG |
4119 | } |
4120 | ||
4121 | void Client::signal_context_list(list<Context*>& ls) | |
4122 | { | |
4123 | while (!ls.empty()) { | |
4124 | ls.front()->complete(0); | |
4125 | ls.pop_front(); | |
4126 | } | |
4127 | } | |
4128 | ||
a8e16298 | 4129 | void Client::wake_up_session_caps(MetaSession *s, bool reconnect) |
7c673cae | 4130 | { |
11fdf7f2 TL |
4131 | for (const auto &cap : s->caps) { |
4132 | auto &in = cap->inode; | |
a8e16298 | 4133 | if (reconnect) { |
11fdf7f2 TL |
4134 | in.requested_max_size = 0; |
4135 | in.wanted_max_size = 0; | |
a8e16298 TL |
4136 | } else { |
4137 | if (cap->gen < s->cap_gen) { | |
4138 | // mds did not re-issue stale cap. | |
4139 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
4140 | // make sure mds knows what we want. | |
11fdf7f2 TL |
4141 | if (in.caps_file_wanted() & ~cap->wanted) |
4142 | in.flags |= I_CAP_DROPPED; | |
a8e16298 TL |
4143 | } |
4144 | } | |
11fdf7f2 | 4145 | signal_cond_list(in.waitfor_caps); |
7c673cae FG |
4146 | } |
4147 | } | |
4148 | ||
4149 | ||
4150 | // flush dirty data (from objectcache) | |
4151 | ||
4152 | class C_Client_CacheInvalidate : public Context { | |
4153 | private: | |
4154 | Client *client; | |
4155 | vinodeno_t ino; | |
4156 | int64_t offset, length; | |
4157 | public: | |
4158 | C_Client_CacheInvalidate(Client *c, Inode *in, int64_t off, int64_t len) : | |
4159 | client(c), offset(off), length(len) { | |
4160 | if (client->use_faked_inos()) | |
4161 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4162 | else | |
4163 | ino = in->vino(); | |
4164 | } | |
4165 | void finish(int r) override { | |
4166 | // _async_invalidate takes the lock when it needs to, call this back from outside of lock. | |
9f95a23c | 4167 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
4168 | client->_async_invalidate(ino, offset, length); |
4169 | } | |
4170 | }; | |
4171 | ||
4172 | void Client::_async_invalidate(vinodeno_t ino, int64_t off, int64_t len) | |
4173 | { | |
f67539c2 TL |
4174 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4175 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4176 | return; |
f67539c2 | 4177 | |
11fdf7f2 | 4178 | ldout(cct, 10) << __func__ << " " << ino << " " << off << "~" << len << dendl; |
7c673cae FG |
4179 | ino_invalidate_cb(callback_handle, ino, off, len); |
4180 | } | |
4181 | ||
4182 | void Client::_schedule_invalidate_callback(Inode *in, int64_t off, int64_t len) { | |
4183 | ||
4184 | if (ino_invalidate_cb) | |
4185 | // we queue the invalidate, which calls the callback and decrements the ref | |
4186 | async_ino_invalidator.queue(new C_Client_CacheInvalidate(this, in, off, len)); | |
4187 | } | |
4188 | ||
4189 | void Client::_invalidate_inode_cache(Inode *in) | |
4190 | { | |
11fdf7f2 | 4191 | ldout(cct, 10) << __func__ << " " << *in << dendl; |
7c673cae FG |
4192 | |
4193 | // invalidate our userspace inode cache | |
94b18763 | 4194 | if (cct->_conf->client_oc) { |
7c673cae | 4195 | objectcacher->release_set(&in->oset); |
94b18763 FG |
4196 | if (!objectcacher->set_is_empty(&in->oset)) |
4197 | lderr(cct) << "failed to invalidate cache for " << *in << dendl; | |
4198 | } | |
7c673cae FG |
4199 | |
4200 | _schedule_invalidate_callback(in, 0, 0); | |
4201 | } | |
4202 | ||
4203 | void Client::_invalidate_inode_cache(Inode *in, int64_t off, int64_t len) | |
4204 | { | |
11fdf7f2 | 4205 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
4206 | |
4207 | // invalidate our userspace inode cache | |
4208 | if (cct->_conf->client_oc) { | |
4209 | vector<ObjectExtent> ls; | |
4210 | Striper::file_to_extents(cct, in->ino, &in->layout, off, len, in->truncate_size, ls); | |
28e407b8 | 4211 | objectcacher->discard_writeback(&in->oset, ls, nullptr); |
7c673cae FG |
4212 | } |
4213 | ||
4214 | _schedule_invalidate_callback(in, off, len); | |
4215 | } | |
4216 | ||
4217 | bool Client::_release(Inode *in) | |
4218 | { | |
4219 | ldout(cct, 20) << "_release " << *in << dendl; | |
4220 | if (in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
4221 | _invalidate_inode_cache(in); | |
4222 | return true; | |
4223 | } | |
4224 | return false; | |
4225 | } | |
4226 | ||
4227 | bool Client::_flush(Inode *in, Context *onfinish) | |
4228 | { | |
4229 | ldout(cct, 10) << "_flush " << *in << dendl; | |
4230 | ||
4231 | if (!in->oset.dirty_or_tx) { | |
4232 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4233 | onfinish->complete(0); | |
4234 | return true; | |
4235 | } | |
4236 | ||
4237 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
1adf2230 | 4238 | ldout(cct, 8) << __func__ << ": FULL, purging for ENOSPC" << dendl; |
7c673cae FG |
4239 | objectcacher->purge_set(&in->oset); |
4240 | if (onfinish) { | |
f67539c2 | 4241 | onfinish->complete(-CEPHFS_ENOSPC); |
7c673cae FG |
4242 | } |
4243 | return true; | |
4244 | } | |
4245 | ||
4246 | return objectcacher->flush_set(&in->oset, onfinish); | |
4247 | } | |
4248 | ||
4249 | void Client::_flush_range(Inode *in, int64_t offset, uint64_t size) | |
4250 | { | |
f67539c2 | 4251 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
4252 | if (!in->oset.dirty_or_tx) { |
4253 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4254 | return; | |
4255 | } | |
4256 | ||
11fdf7f2 | 4257 | C_SaferCond onflush("Client::_flush_range flock"); |
7c673cae | 4258 | bool ret = objectcacher->file_flush(&in->oset, &in->layout, in->snaprealm->get_snap_context(), |
11fdf7f2 | 4259 | offset, size, &onflush); |
7c673cae FG |
4260 | if (!ret) { |
4261 | // wait for flush | |
9f95a23c | 4262 | client_lock.unlock(); |
11fdf7f2 | 4263 | onflush.wait(); |
9f95a23c | 4264 | client_lock.lock(); |
7c673cae FG |
4265 | } |
4266 | } | |
4267 | ||
4268 | void Client::flush_set_callback(ObjectCacher::ObjectSet *oset) | |
4269 | { | |
f67539c2 TL |
4270 | // std::scoped_lock l(client_lock); |
4271 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); // will be called via dispatch() -> objecter -> ... | |
7c673cae | 4272 | Inode *in = static_cast<Inode *>(oset->parent); |
11fdf7f2 | 4273 | ceph_assert(in); |
7c673cae FG |
4274 | _flushed(in); |
4275 | } | |
4276 | ||
4277 | void Client::_flushed(Inode *in) | |
4278 | { | |
4279 | ldout(cct, 10) << "_flushed " << *in << dendl; | |
4280 | ||
4281 | put_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
4282 | } | |
4283 | ||
4284 | ||
4285 | ||
4286 | // checks common to add_update_cap, handle_cap_grant | |
11fdf7f2 | 4287 | void Client::check_cap_issue(Inode *in, unsigned issued) |
7c673cae FG |
4288 | { |
4289 | unsigned had = in->caps_issued(); | |
4290 | ||
4291 | if ((issued & CEPH_CAP_FILE_CACHE) && | |
4292 | !(had & CEPH_CAP_FILE_CACHE)) | |
4293 | in->cache_gen++; | |
4294 | ||
f91f0fd5 TL |
4295 | if ((issued & CEPH_CAP_FILE_SHARED) != |
4296 | (had & CEPH_CAP_FILE_SHARED)) { | |
4297 | if (issued & CEPH_CAP_FILE_SHARED) | |
4298 | in->shared_gen++; | |
7c673cae FG |
4299 | if (in->is_dir()) |
4300 | clear_dir_complete_and_ordered(in, true); | |
4301 | } | |
4302 | } | |
4303 | ||
4304 | void Client::add_update_cap(Inode *in, MetaSession *mds_session, uint64_t cap_id, | |
a8e16298 TL |
4305 | unsigned issued, unsigned wanted, unsigned seq, unsigned mseq, |
4306 | inodeno_t realm, int flags, const UserPerm& cap_perms) | |
7c673cae | 4307 | { |
11fdf7f2 TL |
4308 | if (!in->is_any_caps()) { |
4309 | ceph_assert(in->snaprealm == 0); | |
4310 | in->snaprealm = get_snap_realm(realm); | |
4311 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4312 | ldout(cct, 15) << __func__ << " first one, opened snaprealm " << in->snaprealm << dendl; | |
4313 | } else { | |
4314 | ceph_assert(in->snaprealm); | |
4315 | if ((flags & CEPH_CAP_FLAG_AUTH) && | |
4316 | realm != inodeno_t(-1) && in->snaprealm->ino != realm) { | |
4317 | in->snaprealm_item.remove_myself(); | |
4318 | auto oldrealm = in->snaprealm; | |
4319 | in->snaprealm = get_snap_realm(realm); | |
4320 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4321 | put_snap_realm(oldrealm); | |
4322 | } | |
4323 | } | |
4324 | ||
7c673cae | 4325 | mds_rank_t mds = mds_session->mds_num; |
11fdf7f2 TL |
4326 | const auto &capem = in->caps.emplace(std::piecewise_construct, std::forward_as_tuple(mds), std::forward_as_tuple(*in, mds_session)); |
4327 | Cap &cap = capem.first->second; | |
4328 | if (!capem.second) { | |
4329 | if (cap.gen < mds_session->cap_gen) | |
4330 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
7c673cae FG |
4331 | |
4332 | /* | |
4333 | * auth mds of the inode changed. we received the cap export | |
4334 | * message, but still haven't received the cap import message. | |
4335 | * handle_cap_export() updated the new auth MDS' cap. | |
4336 | * | |
4337 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
4338 | * a message that was send before the cap import message. So | |
4339 | * don't remove caps. | |
4340 | */ | |
11fdf7f2 | 4341 | if (ceph_seq_cmp(seq, cap.seq) <= 0) { |
92f5a8d4 TL |
4342 | if (&cap != in->auth_cap) |
4343 | ldout(cct, 0) << "WARNING: " << "inode " << *in << " caps on mds." << mds << " != auth_cap." << dendl; | |
4344 | ||
11fdf7f2 TL |
4345 | ceph_assert(cap.cap_id == cap_id); |
4346 | seq = cap.seq; | |
4347 | mseq = cap.mseq; | |
4348 | issued |= cap.issued; | |
7c673cae FG |
4349 | flags |= CEPH_CAP_FLAG_AUTH; |
4350 | } | |
f67539c2 TL |
4351 | } else { |
4352 | inc_pinned_icaps(); | |
7c673cae FG |
4353 | } |
4354 | ||
11fdf7f2 | 4355 | check_cap_issue(in, issued); |
7c673cae FG |
4356 | |
4357 | if (flags & CEPH_CAP_FLAG_AUTH) { | |
11fdf7f2 | 4358 | if (in->auth_cap != &cap && |
7c673cae FG |
4359 | (!in->auth_cap || ceph_seq_cmp(in->auth_cap->mseq, mseq) < 0)) { |
4360 | if (in->auth_cap && in->flushing_cap_item.is_on_list()) { | |
11fdf7f2 | 4361 | ldout(cct, 10) << __func__ << " changing auth cap: " |
7c673cae FG |
4362 | << "add myself to new auth MDS' flushing caps list" << dendl; |
4363 | adjust_session_flushing_caps(in, in->auth_cap->session, mds_session); | |
4364 | } | |
11fdf7f2 | 4365 | in->auth_cap = ∩ |
7c673cae FG |
4366 | } |
4367 | } | |
4368 | ||
11fdf7f2 TL |
4369 | unsigned old_caps = cap.issued; |
4370 | cap.cap_id = cap_id; | |
4371 | cap.issued = issued; | |
4372 | cap.implemented |= issued; | |
4373 | if (ceph_seq_cmp(mseq, cap.mseq) > 0) | |
4374 | cap.wanted = wanted; | |
a8e16298 | 4375 | else |
11fdf7f2 TL |
4376 | cap.wanted |= wanted; |
4377 | cap.seq = seq; | |
4378 | cap.issue_seq = seq; | |
4379 | cap.mseq = mseq; | |
4380 | cap.gen = mds_session->cap_gen; | |
4381 | cap.latest_perms = cap_perms; | |
4382 | ldout(cct, 10) << __func__ << " issued " << ccap_string(old_caps) << " -> " << ccap_string(cap.issued) | |
4383 | << " from mds." << mds | |
4384 | << " on " << *in | |
4385 | << dendl; | |
4386 | ||
4387 | if ((issued & ~old_caps) && in->auth_cap == &cap) { | |
7c673cae | 4388 | // non-auth MDS is revoking the newly grant caps ? |
11fdf7f2 TL |
4389 | for (auto &p : in->caps) { |
4390 | if (&p.second == &cap) | |
7c673cae | 4391 | continue; |
11fdf7f2 | 4392 | if (p.second.implemented & ~p.second.issued & issued) { |
7c673cae FG |
4393 | check_caps(in, CHECK_CAPS_NODELAY); |
4394 | break; | |
4395 | } | |
4396 | } | |
4397 | } | |
4398 | ||
4399 | if (issued & ~old_caps) | |
4400 | signal_cond_list(in->waitfor_caps); | |
4401 | } | |
4402 | ||
4403 | void Client::remove_cap(Cap *cap, bool queue_release) | |
4404 | { | |
11fdf7f2 | 4405 | auto &in = cap->inode; |
7c673cae FG |
4406 | MetaSession *session = cap->session; |
4407 | mds_rank_t mds = cap->session->mds_num; | |
4408 | ||
11fdf7f2 | 4409 | ldout(cct, 10) << __func__ << " mds." << mds << " on " << in << dendl; |
7c673cae FG |
4410 | |
4411 | if (queue_release) { | |
4412 | session->enqueue_cap_release( | |
11fdf7f2 | 4413 | in.ino, |
7c673cae FG |
4414 | cap->cap_id, |
4415 | cap->issue_seq, | |
4416 | cap->mseq, | |
4417 | cap_epoch_barrier); | |
f67539c2 TL |
4418 | } else { |
4419 | dec_pinned_icaps(); | |
7c673cae FG |
4420 | } |
4421 | ||
f67539c2 | 4422 | |
11fdf7f2 TL |
4423 | if (in.auth_cap == cap) { |
4424 | if (in.flushing_cap_item.is_on_list()) { | |
7c673cae | 4425 | ldout(cct, 10) << " removing myself from flushing_cap list" << dendl; |
11fdf7f2 | 4426 | in.flushing_cap_item.remove_myself(); |
7c673cae | 4427 | } |
11fdf7f2 | 4428 | in.auth_cap = NULL; |
7c673cae | 4429 | } |
11fdf7f2 TL |
4430 | size_t n = in.caps.erase(mds); |
4431 | ceph_assert(n == 1); | |
7c673cae FG |
4432 | cap = nullptr; |
4433 | ||
11fdf7f2 TL |
4434 | if (!in.is_any_caps()) { |
4435 | ldout(cct, 15) << __func__ << " last one, closing snaprealm " << in.snaprealm << dendl; | |
4436 | in.snaprealm_item.remove_myself(); | |
4437 | put_snap_realm(in.snaprealm); | |
4438 | in.snaprealm = 0; | |
7c673cae FG |
4439 | } |
4440 | } | |
4441 | ||
4442 | void Client::remove_all_caps(Inode *in) | |
4443 | { | |
4444 | while (!in->caps.empty()) | |
11fdf7f2 | 4445 | remove_cap(&in->caps.begin()->second, true); |
7c673cae FG |
4446 | } |
4447 | ||
f6b5b4d7 | 4448 | void Client::remove_session_caps(MetaSession *s, int err) |
7c673cae | 4449 | { |
11fdf7f2 | 4450 | ldout(cct, 10) << __func__ << " mds." << s->mds_num << dendl; |
7c673cae FG |
4451 | |
4452 | while (s->caps.size()) { | |
4453 | Cap *cap = *s->caps.begin(); | |
11fdf7f2 | 4454 | InodeRef in(&cap->inode); |
eafe8130 | 4455 | bool dirty_caps = false; |
7c673cae | 4456 | if (in->auth_cap == cap) { |
7c673cae FG |
4457 | dirty_caps = in->dirty_caps | in->flushing_caps; |
4458 | in->wanted_max_size = 0; | |
4459 | in->requested_max_size = 0; | |
f6b5b4d7 TL |
4460 | if (in->has_any_filelocks()) |
4461 | in->flags |= I_ERROR_FILELOCK; | |
7c673cae | 4462 | } |
f6b5b4d7 | 4463 | auto caps = cap->implemented; |
a8e16298 TL |
4464 | if (cap->wanted | cap->issued) |
4465 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 4466 | remove_cap(cap, false); |
eafe8130 | 4467 | in->cap_snaps.clear(); |
7c673cae | 4468 | if (dirty_caps) { |
11fdf7f2 | 4469 | lderr(cct) << __func__ << " still has dirty|flushing caps on " << *in << dendl; |
7c673cae FG |
4470 | if (in->flushing_caps) { |
4471 | num_flushing_caps--; | |
4472 | in->flushing_cap_tids.clear(); | |
4473 | } | |
4474 | in->flushing_caps = 0; | |
28e407b8 | 4475 | in->mark_caps_clean(); |
11fdf7f2 | 4476 | put_inode(in.get()); |
7c673cae | 4477 | } |
f6b5b4d7 TL |
4478 | caps &= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER; |
4479 | if (caps && !in->caps_issued_mask(caps, true)) { | |
f67539c2 | 4480 | if (err == -CEPHFS_EBLOCKLISTED) { |
f6b5b4d7 TL |
4481 | if (in->oset.dirty_or_tx) { |
4482 | lderr(cct) << __func__ << " still has dirty data on " << *in << dendl; | |
4483 | in->set_async_err(err); | |
4484 | } | |
4485 | objectcacher->purge_set(&in->oset); | |
4486 | } else { | |
4487 | objectcacher->release_set(&in->oset); | |
4488 | } | |
4489 | _schedule_invalidate_callback(in.get(), 0, 0); | |
4490 | } | |
4491 | ||
a8e16298 | 4492 | signal_cond_list(in->waitfor_caps); |
7c673cae FG |
4493 | } |
4494 | s->flushing_caps_tids.clear(); | |
9f95a23c | 4495 | sync_cond.notify_all(); |
7c673cae FG |
4496 | } |
4497 | ||
1d09f67e | 4498 | std::pair<int, bool> Client::_do_remount(bool retry_on_error) |
b32b8144 | 4499 | { |
adb31ebb | 4500 | uint64_t max_retries = cct->_conf.get_val<uint64_t>("mds_max_retries_on_remount_failure"); |
1d09f67e | 4501 | bool abort_on_failure = false; |
91327a77 | 4502 | |
b32b8144 FG |
4503 | errno = 0; |
4504 | int r = remount_cb(callback_handle); | |
91327a77 AA |
4505 | if (r == 0) { |
4506 | retries_on_invalidate = 0; | |
4507 | } else { | |
b32b8144 FG |
4508 | int e = errno; |
4509 | client_t whoami = get_nodeid(); | |
4510 | if (r == -1) { | |
4511 | lderr(cct) << | |
4512 | "failed to remount (to trim kernel dentries): " | |
4513 | "errno = " << e << " (" << strerror(e) << ")" << dendl; | |
4514 | } else { | |
4515 | lderr(cct) << | |
4516 | "failed to remount (to trim kernel dentries): " | |
4517 | "return code = " << r << dendl; | |
4518 | } | |
91327a77 | 4519 | bool should_abort = |
11fdf7f2 TL |
4520 | (cct->_conf.get_val<bool>("client_die_on_failed_remount") || |
4521 | cct->_conf.get_val<bool>("client_die_on_failed_dentry_invalidate")) && | |
91327a77 | 4522 | !(retry_on_error && (++retries_on_invalidate < max_retries)); |
f67539c2 | 4523 | if (should_abort && !is_unmounting()) { |
b32b8144 | 4524 | lderr(cct) << "failed to remount for kernel dentry trimming; quitting!" << dendl; |
1d09f67e | 4525 | abort_on_failure = true; |
b32b8144 FG |
4526 | } |
4527 | } | |
1d09f67e | 4528 | return std::make_pair(r, abort_on_failure); |
b32b8144 FG |
4529 | } |
4530 | ||
7c673cae FG |
4531 | class C_Client_Remount : public Context { |
4532 | private: | |
4533 | Client *client; | |
4534 | public: | |
4535 | explicit C_Client_Remount(Client *c) : client(c) {} | |
4536 | void finish(int r) override { | |
11fdf7f2 | 4537 | ceph_assert(r == 0); |
91327a77 | 4538 | client->_do_remount(true); |
7c673cae FG |
4539 | } |
4540 | }; | |
4541 | ||
4542 | void Client::_invalidate_kernel_dcache() | |
4543 | { | |
f67539c2 TL |
4544 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4545 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4546 | return; |
f67539c2 | 4547 | |
94b18763 FG |
4548 | if (can_invalidate_dentries) { |
4549 | if (dentry_invalidate_cb && root->dir) { | |
4550 | for (ceph::unordered_map<string, Dentry*>::iterator p = root->dir->dentries.begin(); | |
4551 | p != root->dir->dentries.end(); | |
4552 | ++p) { | |
4553 | if (p->second->inode) | |
4554 | _schedule_invalidate_dentry_callback(p->second, false); | |
4555 | } | |
7c673cae FG |
4556 | } |
4557 | } else if (remount_cb) { | |
4558 | // Hacky: | |
4559 | // when remounting a file system, linux kernel trims all unused dentries in the fs | |
4560 | remount_finisher.queue(new C_Client_Remount(this)); | |
4561 | } | |
4562 | } | |
4563 | ||
91327a77 AA |
4564 | void Client::_trim_negative_child_dentries(InodeRef& in) |
4565 | { | |
4566 | if (!in->is_dir()) | |
4567 | return; | |
4568 | ||
4569 | Dir* dir = in->dir; | |
4570 | if (dir && dir->dentries.size() == dir->num_null_dentries) { | |
4571 | for (auto p = dir->dentries.begin(); p != dir->dentries.end(); ) { | |
4572 | Dentry *dn = p->second; | |
4573 | ++p; | |
11fdf7f2 | 4574 | ceph_assert(!dn->inode); |
91327a77 AA |
4575 | if (dn->lru_is_expireable()) |
4576 | unlink(dn, true, false); // keep dir, drop dentry | |
4577 | } | |
4578 | if (dir->dentries.empty()) { | |
4579 | close_dir(dir); | |
4580 | } | |
4581 | } | |
4582 | ||
4583 | if (in->flags & I_SNAPDIR_OPEN) { | |
4584 | InodeRef snapdir = open_snapdir(in.get()); | |
4585 | _trim_negative_child_dentries(snapdir); | |
4586 | } | |
4587 | } | |
4588 | ||
e306af50 TL |
4589 | class C_Client_CacheRelease : public Context { |
4590 | private: | |
4591 | Client *client; | |
4592 | vinodeno_t ino; | |
4593 | public: | |
4594 | C_Client_CacheRelease(Client *c, Inode *in) : | |
4595 | client(c) { | |
4596 | if (client->use_faked_inos()) | |
4597 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4598 | else | |
4599 | ino = in->vino(); | |
4600 | } | |
4601 | void finish(int r) override { | |
4602 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); | |
4603 | client->_async_inode_release(ino); | |
4604 | } | |
4605 | }; | |
4606 | ||
4607 | void Client::_async_inode_release(vinodeno_t ino) | |
4608 | { | |
f67539c2 TL |
4609 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4610 | if (!mref_reader.is_state_satisfied()) | |
e306af50 | 4611 | return; |
f67539c2 | 4612 | |
e306af50 TL |
4613 | ldout(cct, 10) << __func__ << " " << ino << dendl; |
4614 | ino_release_cb(callback_handle, ino); | |
4615 | } | |
4616 | ||
4617 | void Client::_schedule_ino_release_callback(Inode *in) { | |
4618 | ||
4619 | if (ino_release_cb) | |
4620 | // we queue the invalidate, which calls the callback and decrements the ref | |
4621 | async_ino_releasor.queue(new C_Client_CacheRelease(this, in)); | |
4622 | } | |
4623 | ||
28e407b8 | 4624 | void Client::trim_caps(MetaSession *s, uint64_t max) |
7c673cae FG |
4625 | { |
4626 | mds_rank_t mds = s->mds_num; | |
28e407b8 | 4627 | size_t caps_size = s->caps.size(); |
11fdf7f2 | 4628 | ldout(cct, 10) << __func__ << " mds." << mds << " max " << max |
7c673cae FG |
4629 | << " caps " << caps_size << dendl; |
4630 | ||
28e407b8 AA |
4631 | uint64_t trimmed = 0; |
4632 | auto p = s->caps.begin(); | |
4633 | std::set<Dentry *> to_trim; /* this avoids caps other than the one we're | |
4634 | * looking at from getting deleted during traversal. */ | |
7c673cae FG |
4635 | while ((caps_size - trimmed) > max && !p.end()) { |
4636 | Cap *cap = *p; | |
11fdf7f2 | 4637 | InodeRef in(&cap->inode); |
7c673cae FG |
4638 | |
4639 | // Increment p early because it will be invalidated if cap | |
4640 | // is deleted inside remove_cap | |
4641 | ++p; | |
4642 | ||
4643 | if (in->caps.size() > 1 && cap != in->auth_cap) { | |
4644 | int mine = cap->issued | cap->implemented; | |
4645 | int oissued = in->auth_cap ? in->auth_cap->issued : 0; | |
4646 | // disposable non-auth cap | |
b32b8144 | 4647 | if (!(get_caps_used(in.get()) & ~oissued & mine)) { |
7c673cae | 4648 | ldout(cct, 20) << " removing unused, unneeded non-auth cap on " << *in << dendl; |
28e407b8 | 4649 | cap = (remove_cap(cap, true), nullptr); |
7c673cae FG |
4650 | trimmed++; |
4651 | } | |
4652 | } else { | |
4653 | ldout(cct, 20) << " trying to trim dentries for " << *in << dendl; | |
91327a77 | 4654 | _trim_negative_child_dentries(in); |
7c673cae | 4655 | bool all = true; |
11fdf7f2 TL |
4656 | auto q = in->dentries.begin(); |
4657 | while (q != in->dentries.end()) { | |
4658 | Dentry *dn = *q; | |
4659 | ++q; | |
7c673cae FG |
4660 | if (dn->lru_is_expireable()) { |
4661 | if (can_invalidate_dentries && | |
b3b6e05e | 4662 | dn->dir->parent_inode->ino == CEPH_INO_ROOT) { |
7c673cae FG |
4663 | // Only issue one of these per DN for inodes in root: handle |
4664 | // others more efficiently by calling for root-child DNs at | |
4665 | // the end of this function. | |
4666 | _schedule_invalidate_dentry_callback(dn, true); | |
4667 | } | |
28e407b8 AA |
4668 | ldout(cct, 20) << " queueing dentry for trimming: " << dn->name << dendl; |
4669 | to_trim.insert(dn); | |
7c673cae FG |
4670 | } else { |
4671 | ldout(cct, 20) << " not expirable: " << dn->name << dendl; | |
4672 | all = false; | |
4673 | } | |
4674 | } | |
b3b6e05e | 4675 | if (in->ll_ref == 1 && in->ino != CEPH_INO_ROOT) { |
f91f0fd5 TL |
4676 | _schedule_ino_release_callback(in.get()); |
4677 | } | |
b3b6e05e | 4678 | if (all && in->ino != CEPH_INO_ROOT) { |
7c673cae FG |
4679 | ldout(cct, 20) << __func__ << " counting as trimmed: " << *in << dendl; |
4680 | trimmed++; | |
4681 | } | |
4682 | } | |
4683 | } | |
28e407b8 AA |
4684 | ldout(cct, 20) << " trimming queued dentries: " << dendl; |
4685 | for (const auto &dn : to_trim) { | |
4686 | trim_dentry(dn); | |
4687 | } | |
4688 | to_trim.clear(); | |
7c673cae | 4689 | |
b32b8144 | 4690 | caps_size = s->caps.size(); |
11fdf7f2 | 4691 | if (caps_size > (size_t)max) |
7c673cae FG |
4692 | _invalidate_kernel_dcache(); |
4693 | } | |
4694 | ||
4695 | void Client::force_session_readonly(MetaSession *s) | |
4696 | { | |
4697 | s->readonly = true; | |
4698 | for (xlist<Cap*>::iterator p = s->caps.begin(); !p.end(); ++p) { | |
11fdf7f2 TL |
4699 | auto &in = (*p)->inode; |
4700 | if (in.caps_wanted() & CEPH_CAP_FILE_WR) | |
4701 | signal_cond_list(in.waitfor_caps); | |
7c673cae FG |
4702 | } |
4703 | } | |
4704 | ||
7c673cae FG |
4705 | int Client::mark_caps_flushing(Inode *in, ceph_tid_t* ptid) |
4706 | { | |
4707 | MetaSession *session = in->auth_cap->session; | |
4708 | ||
4709 | int flushing = in->dirty_caps; | |
11fdf7f2 | 4710 | ceph_assert(flushing); |
7c673cae FG |
4711 | |
4712 | ceph_tid_t flush_tid = ++last_flush_tid; | |
4713 | in->flushing_cap_tids[flush_tid] = flushing; | |
4714 | ||
4715 | if (!in->flushing_caps) { | |
11fdf7f2 | 4716 | ldout(cct, 10) << __func__ << " " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4717 | num_flushing_caps++; |
4718 | } else { | |
11fdf7f2 | 4719 | ldout(cct, 10) << __func__ << " (more) " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4720 | } |
4721 | ||
4722 | in->flushing_caps |= flushing; | |
28e407b8 | 4723 | in->mark_caps_clean(); |
7c673cae FG |
4724 | |
4725 | if (!in->flushing_cap_item.is_on_list()) | |
4726 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
4727 | session->flushing_caps_tids.insert(flush_tid); | |
4728 | ||
4729 | *ptid = flush_tid; | |
4730 | return flushing; | |
4731 | } | |
4732 | ||
4733 | void Client::adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s) | |
4734 | { | |
4735 | for (auto &p : in->cap_snaps) { | |
4736 | CapSnap &capsnap = p.second; | |
4737 | if (capsnap.flush_tid > 0) { | |
4738 | old_s->flushing_caps_tids.erase(capsnap.flush_tid); | |
4739 | new_s->flushing_caps_tids.insert(capsnap.flush_tid); | |
4740 | } | |
4741 | } | |
4742 | for (map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4743 | it != in->flushing_cap_tids.end(); | |
4744 | ++it) { | |
4745 | old_s->flushing_caps_tids.erase(it->first); | |
4746 | new_s->flushing_caps_tids.insert(it->first); | |
4747 | } | |
4748 | new_s->flushing_caps.push_back(&in->flushing_cap_item); | |
4749 | } | |
4750 | ||
4751 | /* | |
20effc67 TL |
4752 | * Flush all the dirty caps back to the MDS. Because the callers |
4753 | * generally wait on the result of this function (syncfs and umount | |
4754 | * cases), we set CHECK_CAPS_SYNCHRONOUS on the last check_caps call. | |
7c673cae FG |
4755 | */ |
4756 | void Client::flush_caps_sync() | |
4757 | { | |
4758 | ldout(cct, 10) << __func__ << dendl; | |
20effc67 TL |
4759 | for (auto &q : mds_sessions) { |
4760 | auto s = q.second; | |
4761 | xlist<Inode*>::iterator p = s->dirty_list.begin(); | |
4762 | while (!p.end()) { | |
4763 | unsigned flags = CHECK_CAPS_NODELAY; | |
4764 | Inode *in = *p; | |
7c673cae | 4765 | |
20effc67 TL |
4766 | ++p; |
4767 | if (p.end()) | |
4768 | flags |= CHECK_CAPS_SYNCHRONOUS; | |
4769 | check_caps(in, flags); | |
4770 | } | |
7c673cae FG |
4771 | } |
4772 | } | |
4773 | ||
7c673cae FG |
4774 | void Client::wait_sync_caps(Inode *in, ceph_tid_t want) |
4775 | { | |
4776 | while (in->flushing_caps) { | |
4777 | map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
11fdf7f2 | 4778 | ceph_assert(it != in->flushing_cap_tids.end()); |
7c673cae FG |
4779 | if (it->first > want) |
4780 | break; | |
11fdf7f2 | 4781 | ldout(cct, 10) << __func__ << " on " << *in << " flushing " |
7c673cae FG |
4782 | << ccap_string(it->second) << " want " << want |
4783 | << " last " << it->first << dendl; | |
4784 | wait_on_list(in->waitfor_caps); | |
4785 | } | |
4786 | } | |
4787 | ||
4788 | void Client::wait_sync_caps(ceph_tid_t want) | |
4789 | { | |
4790 | retry: | |
11fdf7f2 | 4791 | ldout(cct, 10) << __func__ << " want " << want << " (last is " << last_flush_tid << ", " |
7c673cae | 4792 | << num_flushing_caps << " total flushing)" << dendl; |
11fdf7f2 | 4793 | for (auto &p : mds_sessions) { |
20effc67 | 4794 | auto s = p.second; |
7c673cae FG |
4795 | if (s->flushing_caps_tids.empty()) |
4796 | continue; | |
4797 | ceph_tid_t oldest_tid = *s->flushing_caps_tids.begin(); | |
4798 | if (oldest_tid <= want) { | |
11fdf7f2 | 4799 | ldout(cct, 10) << " waiting on mds." << p.first << " tid " << oldest_tid |
7c673cae | 4800 | << " (want " << want << ")" << dendl; |
9f95a23c TL |
4801 | std::unique_lock l{client_lock, std::adopt_lock}; |
4802 | sync_cond.wait(l); | |
4803 | l.release(); | |
7c673cae FG |
4804 | goto retry; |
4805 | } | |
4806 | } | |
4807 | } | |
4808 | ||
eafe8130 TL |
4809 | void Client::kick_flushing_caps(Inode *in, MetaSession *session) |
4810 | { | |
4811 | in->flags &= ~I_KICK_FLUSH; | |
4812 | ||
4813 | Cap *cap = in->auth_cap; | |
4814 | ceph_assert(cap->session == session); | |
4815 | ||
4816 | ceph_tid_t last_snap_flush = 0; | |
4817 | for (auto p = in->flushing_cap_tids.rbegin(); | |
4818 | p != in->flushing_cap_tids.rend(); | |
4819 | ++p) { | |
4820 | if (!p->second) { | |
4821 | last_snap_flush = p->first; | |
4822 | break; | |
4823 | } | |
4824 | } | |
4825 | ||
4826 | int wanted = in->caps_wanted(); | |
4827 | int used = get_caps_used(in) | in->caps_dirty(); | |
4828 | auto it = in->cap_snaps.begin(); | |
4829 | for (auto& p : in->flushing_cap_tids) { | |
4830 | if (p.second) { | |
4831 | int msg_flags = p.first < last_snap_flush ? MClientCaps::FLAG_PENDING_CAPSNAP : 0; | |
4832 | send_cap(in, session, cap, msg_flags, used, wanted, (cap->issued | cap->implemented), | |
4833 | p.second, p.first); | |
4834 | } else { | |
4835 | ceph_assert(it != in->cap_snaps.end()); | |
4836 | ceph_assert(it->second.flush_tid == p.first); | |
4837 | send_flush_snap(in, session, it->first, it->second); | |
4838 | ++it; | |
4839 | } | |
4840 | } | |
4841 | } | |
4842 | ||
7c673cae FG |
4843 | void Client::kick_flushing_caps(MetaSession *session) |
4844 | { | |
4845 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 4846 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
4847 | |
4848 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { | |
4849 | Inode *in = *p; | |
eafe8130 TL |
4850 | if (in->flags & I_KICK_FLUSH) { |
4851 | ldout(cct, 20) << " reflushing caps on " << *in << " to mds." << mds << dendl; | |
4852 | kick_flushing_caps(in, session); | |
4853 | } | |
7c673cae | 4854 | } |
7c673cae FG |
4855 | } |
4856 | ||
4857 | void Client::early_kick_flushing_caps(MetaSession *session) | |
4858 | { | |
7c673cae FG |
4859 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { |
4860 | Inode *in = *p; | |
11fdf7f2 TL |
4861 | Cap *cap = in->auth_cap; |
4862 | ceph_assert(cap); | |
7c673cae FG |
4863 | |
4864 | // if flushing caps were revoked, we re-send the cap flush in client reconnect | |
4865 | // stage. This guarantees that MDS processes the cap flush message before issuing | |
4866 | // the flushing caps to other client. | |
eafe8130 TL |
4867 | if ((in->flushing_caps & in->auth_cap->issued) == in->flushing_caps) { |
4868 | in->flags |= I_KICK_FLUSH; | |
7c673cae | 4869 | continue; |
eafe8130 | 4870 | } |
7c673cae FG |
4871 | |
4872 | ldout(cct, 20) << " reflushing caps (early_kick) on " << *in | |
4873 | << " to mds." << session->mds_num << dendl; | |
11fdf7f2 TL |
4874 | // send_reconnect() also will reset these sequence numbers. make sure |
4875 | // sequence numbers in cap flush message match later reconnect message. | |
4876 | cap->seq = 0; | |
4877 | cap->issue_seq = 0; | |
4878 | cap->mseq = 0; | |
4879 | cap->issued = cap->implemented; | |
4880 | ||
eafe8130 | 4881 | kick_flushing_caps(in, session); |
7c673cae FG |
4882 | } |
4883 | } | |
4884 | ||
7c673cae FG |
4885 | void Client::invalidate_snaprealm_and_children(SnapRealm *realm) |
4886 | { | |
4887 | list<SnapRealm*> q; | |
4888 | q.push_back(realm); | |
4889 | ||
4890 | while (!q.empty()) { | |
4891 | realm = q.front(); | |
4892 | q.pop_front(); | |
4893 | ||
11fdf7f2 | 4894 | ldout(cct, 10) << __func__ << " " << *realm << dendl; |
7c673cae FG |
4895 | realm->invalidate_cache(); |
4896 | ||
4897 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4898 | p != realm->pchildren.end(); | |
4899 | ++p) | |
4900 | q.push_back(*p); | |
4901 | } | |
4902 | } | |
4903 | ||
4904 | SnapRealm *Client::get_snap_realm(inodeno_t r) | |
4905 | { | |
4906 | SnapRealm *realm = snap_realms[r]; | |
2a845540 TL |
4907 | |
4908 | ldout(cct, 20) << __func__ << " " << r << " " << realm << ", nref was " | |
4909 | << (realm ? realm->nref : 0) << dendl; | |
4910 | if (!realm) { | |
7c673cae | 4911 | snap_realms[r] = realm = new SnapRealm(r); |
2a845540 TL |
4912 | |
4913 | // Do not release the global snaprealm until unmounting. | |
4914 | if (r == CEPH_INO_GLOBAL_SNAPREALM) | |
4915 | realm->nref++; | |
4916 | } | |
4917 | ||
7c673cae | 4918 | realm->nref++; |
2a845540 TL |
4919 | ldout(cct, 20) << __func__ << " " << r << " " << realm << ", nref now is " |
4920 | << realm->nref << dendl; | |
7c673cae FG |
4921 | return realm; |
4922 | } | |
4923 | ||
4924 | SnapRealm *Client::get_snap_realm_maybe(inodeno_t r) | |
4925 | { | |
4926 | if (snap_realms.count(r) == 0) { | |
11fdf7f2 | 4927 | ldout(cct, 20) << __func__ << " " << r << " fail" << dendl; |
7c673cae FG |
4928 | return NULL; |
4929 | } | |
4930 | SnapRealm *realm = snap_realms[r]; | |
11fdf7f2 | 4931 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4932 | realm->nref++; |
4933 | return realm; | |
4934 | } | |
4935 | ||
4936 | void Client::put_snap_realm(SnapRealm *realm) | |
4937 | { | |
11fdf7f2 | 4938 | ldout(cct, 20) << __func__ << " " << realm->ino << " " << realm |
7c673cae FG |
4939 | << " " << realm->nref << " -> " << (realm->nref - 1) << dendl; |
4940 | if (--realm->nref == 0) { | |
4941 | snap_realms.erase(realm->ino); | |
4942 | if (realm->pparent) { | |
4943 | realm->pparent->pchildren.erase(realm); | |
4944 | put_snap_realm(realm->pparent); | |
4945 | } | |
4946 | delete realm; | |
4947 | } | |
4948 | } | |
4949 | ||
4950 | bool Client::adjust_realm_parent(SnapRealm *realm, inodeno_t parent) | |
4951 | { | |
4952 | if (realm->parent != parent) { | |
11fdf7f2 | 4953 | ldout(cct, 10) << __func__ << " " << *realm |
7c673cae FG |
4954 | << " " << realm->parent << " -> " << parent << dendl; |
4955 | realm->parent = parent; | |
4956 | if (realm->pparent) { | |
4957 | realm->pparent->pchildren.erase(realm); | |
4958 | put_snap_realm(realm->pparent); | |
4959 | } | |
4960 | realm->pparent = get_snap_realm(parent); | |
4961 | realm->pparent->pchildren.insert(realm); | |
4962 | return true; | |
4963 | } | |
4964 | return false; | |
4965 | } | |
4966 | ||
4967 | static bool has_new_snaps(const SnapContext& old_snapc, | |
4968 | const SnapContext& new_snapc) | |
4969 | { | |
4970 | return !new_snapc.snaps.empty() && new_snapc.snaps[0] > old_snapc.seq; | |
4971 | } | |
4972 | ||
4973 | ||
11fdf7f2 | 4974 | void Client::update_snap_trace(const bufferlist& bl, SnapRealm **realm_ret, bool flush) |
7c673cae FG |
4975 | { |
4976 | SnapRealm *first_realm = NULL; | |
11fdf7f2 | 4977 | ldout(cct, 10) << __func__ << " len " << bl.length() << dendl; |
7c673cae FG |
4978 | |
4979 | map<SnapRealm*, SnapContext> dirty_realms; | |
4980 | ||
11fdf7f2 | 4981 | auto p = bl.cbegin(); |
7c673cae FG |
4982 | while (!p.end()) { |
4983 | SnapRealmInfo info; | |
11fdf7f2 | 4984 | decode(info, p); |
7c673cae FG |
4985 | SnapRealm *realm = get_snap_realm(info.ino()); |
4986 | ||
4987 | bool invalidate = false; | |
4988 | ||
4989 | if (info.seq() > realm->seq) { | |
11fdf7f2 | 4990 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() << " > " << realm->seq |
7c673cae FG |
4991 | << dendl; |
4992 | ||
4993 | if (flush) { | |
4994 | // writeback any dirty caps _before_ updating snap list (i.e. with old snap info) | |
4995 | // flush me + children | |
4996 | list<SnapRealm*> q; | |
4997 | q.push_back(realm); | |
4998 | while (!q.empty()) { | |
4999 | SnapRealm *realm = q.front(); | |
5000 | q.pop_front(); | |
5001 | ||
5002 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
5003 | p != realm->pchildren.end(); | |
5004 | ++p) | |
5005 | q.push_back(*p); | |
5006 | ||
5007 | if (dirty_realms.count(realm) == 0) { | |
5008 | realm->nref++; | |
5009 | dirty_realms[realm] = realm->get_snap_context(); | |
5010 | } | |
5011 | } | |
5012 | } | |
5013 | ||
5014 | // update | |
5015 | realm->seq = info.seq(); | |
5016 | realm->created = info.created(); | |
5017 | realm->parent_since = info.parent_since(); | |
5018 | realm->prior_parent_snaps = info.prior_parent_snaps; | |
5019 | realm->my_snaps = info.my_snaps; | |
5020 | invalidate = true; | |
5021 | } | |
5022 | ||
5023 | // _always_ verify parent | |
5024 | if (adjust_realm_parent(realm, info.parent())) | |
5025 | invalidate = true; | |
5026 | ||
5027 | if (invalidate) { | |
5028 | invalidate_snaprealm_and_children(realm); | |
11fdf7f2 | 5029 | ldout(cct, 15) << __func__ << " " << *realm << " self|parent updated" << dendl; |
7c673cae FG |
5030 | ldout(cct, 15) << " snapc " << realm->get_snap_context() << dendl; |
5031 | } else { | |
11fdf7f2 | 5032 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() |
7c673cae FG |
5033 | << " <= " << realm->seq << " and same parent, SKIPPING" << dendl; |
5034 | } | |
f67539c2 | 5035 | |
7c673cae FG |
5036 | if (!first_realm) |
5037 | first_realm = realm; | |
5038 | else | |
5039 | put_snap_realm(realm); | |
5040 | } | |
5041 | ||
f67539c2 | 5042 | for (auto &[realm, snapc] : dirty_realms) { |
7c673cae | 5043 | // if there are new snaps ? |
f67539c2 | 5044 | if (has_new_snaps(snapc, realm->get_snap_context())) { |
7c673cae | 5045 | ldout(cct, 10) << " flushing caps on " << *realm << dendl; |
f67539c2 TL |
5046 | for (auto&& in : realm->inodes_with_caps) { |
5047 | queue_cap_snap(in, snapc); | |
7c673cae FG |
5048 | } |
5049 | } else { | |
5050 | ldout(cct, 10) << " no new snap on " << *realm << dendl; | |
5051 | } | |
5052 | put_snap_realm(realm); | |
5053 | } | |
5054 | ||
5055 | if (realm_ret) | |
5056 | *realm_ret = first_realm; | |
5057 | else | |
5058 | put_snap_realm(first_realm); | |
5059 | } | |
5060 | ||
11fdf7f2 | 5061 | void Client::handle_snap(const MConstRef<MClientSnap>& m) |
7c673cae | 5062 | { |
11fdf7f2 | 5063 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 5064 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
5065 | |
5066 | std::scoped_lock cl(client_lock); | |
20effc67 | 5067 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 5068 | if (!session) { |
7c673cae FG |
5069 | return; |
5070 | } | |
5071 | ||
20effc67 | 5072 | got_mds_push(session.get()); |
7c673cae FG |
5073 | |
5074 | map<Inode*, SnapContext> to_move; | |
5075 | SnapRealm *realm = 0; | |
5076 | ||
5077 | if (m->head.op == CEPH_SNAP_OP_SPLIT) { | |
11fdf7f2 | 5078 | ceph_assert(m->head.split); |
7c673cae | 5079 | SnapRealmInfo info; |
11fdf7f2 TL |
5080 | auto p = m->bl.cbegin(); |
5081 | decode(info, p); | |
5082 | ceph_assert(info.ino() == m->head.split); | |
7c673cae FG |
5083 | |
5084 | // flush, then move, ino's. | |
5085 | realm = get_snap_realm(info.ino()); | |
5086 | ldout(cct, 10) << " splitting off " << *realm << dendl; | |
11fdf7f2 TL |
5087 | for (auto& ino : m->split_inos) { |
5088 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
7c673cae FG |
5089 | if (inode_map.count(vino)) { |
5090 | Inode *in = inode_map[vino]; | |
5091 | if (!in->snaprealm || in->snaprealm == realm) | |
5092 | continue; | |
5093 | if (in->snaprealm->created > info.created()) { | |
5094 | ldout(cct, 10) << " NOT moving " << *in << " from _newer_ realm " | |
5095 | << *in->snaprealm << dendl; | |
5096 | continue; | |
5097 | } | |
5098 | ldout(cct, 10) << " moving " << *in << " from " << *in->snaprealm << dendl; | |
5099 | ||
5100 | ||
5101 | in->snaprealm_item.remove_myself(); | |
5102 | to_move[in] = in->snaprealm->get_snap_context(); | |
5103 | put_snap_realm(in->snaprealm); | |
5104 | } | |
5105 | } | |
5106 | ||
5107 | // move child snaprealms, too | |
11fdf7f2 TL |
5108 | for (auto& child_realm : m->split_realms) { |
5109 | ldout(cct, 10) << "adjusting snaprealm " << child_realm << " parent" << dendl; | |
5110 | SnapRealm *child = get_snap_realm_maybe(child_realm); | |
7c673cae FG |
5111 | if (!child) |
5112 | continue; | |
5113 | adjust_realm_parent(child, realm->ino); | |
5114 | put_snap_realm(child); | |
5115 | } | |
5116 | } | |
5117 | ||
5118 | update_snap_trace(m->bl, NULL, m->head.op != CEPH_SNAP_OP_DESTROY); | |
5119 | ||
5120 | if (realm) { | |
5121 | for (auto p = to_move.begin(); p != to_move.end(); ++p) { | |
5122 | Inode *in = p->first; | |
5123 | in->snaprealm = realm; | |
5124 | realm->inodes_with_caps.push_back(&in->snaprealm_item); | |
5125 | realm->nref++; | |
5126 | // queue for snap writeback | |
5127 | if (has_new_snaps(p->second, realm->get_snap_context())) | |
5128 | queue_cap_snap(in, p->second); | |
5129 | } | |
5130 | put_snap_realm(realm); | |
5131 | } | |
7c673cae FG |
5132 | } |
5133 | ||
11fdf7f2 | 5134 | void Client::handle_quota(const MConstRef<MClientQuota>& m) |
7c673cae FG |
5135 | { |
5136 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5137 | |
5138 | std::scoped_lock cl(client_lock); | |
20effc67 | 5139 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 5140 | if (!session) { |
7c673cae FG |
5141 | return; |
5142 | } | |
5143 | ||
20effc67 | 5144 | got_mds_push(session.get()); |
7c673cae | 5145 | |
11fdf7f2 | 5146 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << mds << dendl; |
7c673cae FG |
5147 | |
5148 | vinodeno_t vino(m->ino, CEPH_NOSNAP); | |
5149 | if (inode_map.count(vino)) { | |
5150 | Inode *in = NULL; | |
5151 | in = inode_map[vino]; | |
5152 | ||
5153 | if (in) { | |
5154 | in->quota = m->quota; | |
5155 | in->rstat = m->rstat; | |
5156 | } | |
5157 | } | |
7c673cae FG |
5158 | } |
5159 | ||
11fdf7f2 | 5160 | void Client::handle_caps(const MConstRef<MClientCaps>& m) |
7c673cae FG |
5161 | { |
5162 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5163 | |
5164 | std::scoped_lock cl(client_lock); | |
20effc67 | 5165 | auto session = _get_mds_session(mds, m->get_connection().get()); |
7c673cae | 5166 | if (!session) { |
7c673cae FG |
5167 | return; |
5168 | } | |
5169 | ||
5170 | if (m->osd_epoch_barrier && !objecter->have_map(m->osd_epoch_barrier)) { | |
5171 | // Pause RADOS operations until we see the required epoch | |
5172 | objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
5173 | } | |
5174 | ||
5175 | if (m->osd_epoch_barrier > cap_epoch_barrier) { | |
5176 | // Record the barrier so that we will transmit it to MDS when releasing | |
5177 | set_cap_epoch_barrier(m->osd_epoch_barrier); | |
5178 | } | |
5179 | ||
20effc67 | 5180 | got_mds_push(session.get()); |
7c673cae | 5181 | |
11fdf7f2 | 5182 | Inode *in; |
7c673cae | 5183 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); |
11fdf7f2 TL |
5184 | if (auto it = inode_map.find(vino); it != inode_map.end()) { |
5185 | in = it->second; | |
5186 | } else { | |
7c673cae | 5187 | if (m->get_op() == CEPH_CAP_OP_IMPORT) { |
11fdf7f2 | 5188 | ldout(cct, 5) << __func__ << " don't have vino " << vino << " on IMPORT, immediately releasing" << dendl; |
7c673cae FG |
5189 | session->enqueue_cap_release( |
5190 | m->get_ino(), | |
5191 | m->get_cap_id(), | |
5192 | m->get_seq(), | |
5193 | m->get_mseq(), | |
5194 | cap_epoch_barrier); | |
5195 | } else { | |
11fdf7f2 | 5196 | ldout(cct, 5) << __func__ << " don't have vino " << vino << ", dropping" << dendl; |
7c673cae | 5197 | } |
7c673cae FG |
5198 | |
5199 | // in case the mds is waiting on e.g. a revocation | |
5200 | flush_cap_releases(); | |
5201 | return; | |
5202 | } | |
5203 | ||
5204 | switch (m->get_op()) { | |
20effc67 TL |
5205 | case CEPH_CAP_OP_EXPORT: return handle_cap_export(session.get(), in, m); |
5206 | case CEPH_CAP_OP_FLUSHSNAP_ACK: return handle_cap_flushsnap_ack(session.get(), in, m); | |
5207 | case CEPH_CAP_OP_IMPORT: /* no return */ handle_cap_import(session.get(), in, m); | |
7c673cae FG |
5208 | } |
5209 | ||
11fdf7f2 TL |
5210 | if (auto it = in->caps.find(mds); it != in->caps.end()) { |
5211 | Cap &cap = in->caps.at(mds); | |
7c673cae | 5212 | |
11fdf7f2 | 5213 | switch (m->get_op()) { |
20effc67 | 5214 | case CEPH_CAP_OP_TRUNC: return handle_cap_trunc(session.get(), in, m); |
11fdf7f2 TL |
5215 | case CEPH_CAP_OP_IMPORT: |
5216 | case CEPH_CAP_OP_REVOKE: | |
20effc67 TL |
5217 | case CEPH_CAP_OP_GRANT: return handle_cap_grant(session.get(), in, &cap, m); |
5218 | case CEPH_CAP_OP_FLUSH_ACK: return handle_cap_flush_ack(session.get(), in, &cap, m); | |
11fdf7f2 TL |
5219 | } |
5220 | } else { | |
5221 | ldout(cct, 5) << __func__ << " don't have " << *in << " cap on mds." << mds << dendl; | |
5222 | return; | |
7c673cae FG |
5223 | } |
5224 | } | |
5225 | ||
11fdf7f2 | 5226 | void Client::handle_cap_import(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5227 | { |
5228 | mds_rank_t mds = session->mds_num; | |
5229 | ||
11fdf7f2 | 5230 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5231 | << " IMPORT from mds." << mds << dendl; |
5232 | ||
5233 | const mds_rank_t peer_mds = mds_rank_t(m->peer.mds); | |
5234 | Cap *cap = NULL; | |
5235 | UserPerm cap_perms; | |
11fdf7f2 TL |
5236 | if (auto it = in->caps.find(peer_mds); m->peer.cap_id && it != in->caps.end()) { |
5237 | cap = &it->second; | |
5238 | cap_perms = cap->latest_perms; | |
7c673cae FG |
5239 | } |
5240 | ||
5241 | // add/update it | |
5242 | SnapRealm *realm = NULL; | |
5243 | update_snap_trace(m->snapbl, &realm); | |
5244 | ||
1911f103 TL |
5245 | int issued = m->get_caps(); |
5246 | int wanted = m->get_wanted(); | |
7c673cae | 5247 | add_update_cap(in, session, m->get_cap_id(), |
1911f103 | 5248 | issued, wanted, m->get_seq(), m->get_mseq(), |
a8e16298 | 5249 | m->get_realm(), CEPH_CAP_FLAG_AUTH, cap_perms); |
7c673cae FG |
5250 | |
5251 | if (cap && cap->cap_id == m->peer.cap_id) { | |
5252 | remove_cap(cap, (m->peer.flags & CEPH_CAP_FLAG_RELEASE)); | |
5253 | } | |
5254 | ||
5255 | if (realm) | |
5256 | put_snap_realm(realm); | |
5257 | ||
eafe8130 | 5258 | if (in->auth_cap && in->auth_cap->session == session) { |
1911f103 TL |
5259 | if (!(wanted & CEPH_CAP_ANY_FILE_WR) || |
5260 | in->requested_max_size > m->get_max_size()) { | |
5261 | in->requested_max_size = 0; | |
5262 | ldout(cct, 15) << "reset requested_max_size after cap import" << dendl; | |
5263 | } | |
7c673cae | 5264 | // reflush any/all caps (if we are now the auth_cap) |
eafe8130 | 5265 | kick_flushing_caps(in, session); |
7c673cae FG |
5266 | } |
5267 | } | |
5268 | ||
11fdf7f2 | 5269 | void Client::handle_cap_export(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5270 | { |
5271 | mds_rank_t mds = session->mds_num; | |
5272 | ||
11fdf7f2 | 5273 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5274 | << " EXPORT from mds." << mds << dendl; |
5275 | ||
11fdf7f2 TL |
5276 | auto it = in->caps.find(mds); |
5277 | if (it != in->caps.end()) { | |
5278 | Cap &cap = it->second; | |
5279 | if (cap.cap_id == m->get_cap_id()) { | |
5280 | if (m->peer.cap_id) { | |
5281 | const auto peer_mds = mds_rank_t(m->peer.mds); | |
20effc67 | 5282 | auto tsession = _get_or_open_mds_session(peer_mds); |
11fdf7f2 TL |
5283 | auto it = in->caps.find(peer_mds); |
5284 | if (it != in->caps.end()) { | |
5285 | Cap &tcap = it->second; | |
5286 | if (tcap.cap_id == m->peer.cap_id && | |
5287 | ceph_seq_cmp(tcap.seq, m->peer.seq) < 0) { | |
5288 | tcap.cap_id = m->peer.cap_id; | |
5289 | tcap.seq = m->peer.seq - 1; | |
5290 | tcap.issue_seq = tcap.seq; | |
5291 | tcap.issued |= cap.issued; | |
5292 | tcap.implemented |= cap.issued; | |
5293 | if (&cap == in->auth_cap) | |
5294 | in->auth_cap = &tcap; | |
5295 | if (in->auth_cap == &tcap && in->flushing_cap_item.is_on_list()) | |
20effc67 | 5296 | adjust_session_flushing_caps(in, session, tsession.get()); |
11fdf7f2 TL |
5297 | } |
5298 | } else { | |
20effc67 | 5299 | add_update_cap(in, tsession.get(), m->peer.cap_id, cap.issued, 0, |
11fdf7f2 TL |
5300 | m->peer.seq - 1, m->peer.mseq, (uint64_t)-1, |
5301 | &cap == in->auth_cap ? CEPH_CAP_FLAG_AUTH : 0, | |
5302 | cap.latest_perms); | |
5303 | } | |
7c673cae | 5304 | } else { |
11fdf7f2 TL |
5305 | if (cap.wanted | cap.issued) |
5306 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 5307 | } |
7c673cae | 5308 | |
11fdf7f2 TL |
5309 | remove_cap(&cap, false); |
5310 | } | |
7c673cae | 5311 | } |
7c673cae FG |
5312 | } |
5313 | ||
11fdf7f2 | 5314 | void Client::handle_cap_trunc(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5315 | { |
5316 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 5317 | ceph_assert(in->caps.count(mds)); |
7c673cae | 5318 | |
11fdf7f2 | 5319 | ldout(cct, 10) << __func__ << " on ino " << *in |
7c673cae FG |
5320 | << " size " << in->size << " -> " << m->get_size() |
5321 | << dendl; | |
5322 | ||
1adf2230 AA |
5323 | int issued; |
5324 | in->caps_issued(&issued); | |
5325 | issued |= in->caps_dirty(); | |
5326 | update_inode_file_size(in, issued, m->get_size(), | |
5327 | m->get_truncate_seq(), m->get_truncate_size()); | |
7c673cae FG |
5328 | } |
5329 | ||
11fdf7f2 | 5330 | void Client::handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5331 | { |
5332 | ceph_tid_t flush_ack_tid = m->get_client_tid(); | |
5333 | int dirty = m->get_dirty(); | |
5334 | int cleaned = 0; | |
5335 | int flushed = 0; | |
5336 | ||
11fdf7f2 TL |
5337 | auto it = in->flushing_cap_tids.begin(); |
5338 | if (it->first < flush_ack_tid) { | |
5339 | ldout(cct, 0) << __func__ << " mds." << session->mds_num | |
5340 | << " got unexpected flush ack tid " << flush_ack_tid | |
5341 | << " expected is " << it->first << dendl; | |
5342 | } | |
5343 | for (; it != in->flushing_cap_tids.end(); ) { | |
eafe8130 TL |
5344 | if (!it->second) { |
5345 | // cap snap | |
5346 | ++it; | |
5347 | continue; | |
5348 | } | |
7c673cae FG |
5349 | if (it->first == flush_ack_tid) |
5350 | cleaned = it->second; | |
5351 | if (it->first <= flush_ack_tid) { | |
5352 | session->flushing_caps_tids.erase(it->first); | |
5353 | in->flushing_cap_tids.erase(it++); | |
5354 | ++flushed; | |
5355 | continue; | |
5356 | } | |
5357 | cleaned &= ~it->second; | |
5358 | if (!cleaned) | |
5359 | break; | |
5360 | ++it; | |
5361 | } | |
5362 | ||
11fdf7f2 | 5363 | ldout(cct, 5) << __func__ << " mds." << session->mds_num |
7c673cae FG |
5364 | << " cleaned " << ccap_string(cleaned) << " on " << *in |
5365 | << " with " << ccap_string(dirty) << dendl; | |
5366 | ||
5367 | if (flushed) { | |
5368 | signal_cond_list(in->waitfor_caps); | |
5369 | if (session->flushing_caps_tids.empty() || | |
5370 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5371 | sync_cond.notify_all(); |
7c673cae FG |
5372 | } |
5373 | ||
5374 | if (!dirty) { | |
5375 | in->cap_dirtier_uid = -1; | |
5376 | in->cap_dirtier_gid = -1; | |
5377 | } | |
5378 | ||
5379 | if (!cleaned) { | |
5380 | ldout(cct, 10) << " tid " << m->get_client_tid() << " != any cap bit tids" << dendl; | |
5381 | } else { | |
5382 | if (in->flushing_caps) { | |
5383 | ldout(cct, 5) << " flushing_caps " << ccap_string(in->flushing_caps) | |
5384 | << " -> " << ccap_string(in->flushing_caps & ~cleaned) << dendl; | |
5385 | in->flushing_caps &= ~cleaned; | |
5386 | if (in->flushing_caps == 0) { | |
5387 | ldout(cct, 10) << " " << *in << " !flushing" << dendl; | |
5388 | num_flushing_caps--; | |
eafe8130 | 5389 | if (in->flushing_cap_tids.empty()) |
7c673cae FG |
5390 | in->flushing_cap_item.remove_myself(); |
5391 | } | |
5392 | if (!in->caps_dirty()) | |
5393 | put_inode(in); | |
5394 | } | |
5395 | } | |
7c673cae FG |
5396 | } |
5397 | ||
5398 | ||
11fdf7f2 | 5399 | void Client::handle_cap_flushsnap_ack(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae | 5400 | { |
eafe8130 | 5401 | ceph_tid_t flush_ack_tid = m->get_client_tid(); |
7c673cae | 5402 | mds_rank_t mds = session->mds_num; |
11fdf7f2 | 5403 | ceph_assert(in->caps.count(mds)); |
7c673cae FG |
5404 | snapid_t follows = m->get_snap_follows(); |
5405 | ||
11fdf7f2 TL |
5406 | if (auto it = in->cap_snaps.find(follows); it != in->cap_snaps.end()) { |
5407 | auto& capsnap = it->second; | |
eafe8130 TL |
5408 | if (flush_ack_tid != capsnap.flush_tid) { |
5409 | ldout(cct, 10) << " tid " << flush_ack_tid << " != " << capsnap.flush_tid << dendl; | |
7c673cae | 5410 | } else { |
eafe8130 | 5411 | InodeRef tmp_ref(in); |
11fdf7f2 | 5412 | ldout(cct, 5) << __func__ << " mds." << mds << " flushed snap follows " << follows |
7c673cae | 5413 | << " on " << *in << dendl; |
7c673cae | 5414 | session->flushing_caps_tids.erase(capsnap.flush_tid); |
eafe8130 TL |
5415 | in->flushing_cap_tids.erase(capsnap.flush_tid); |
5416 | if (in->flushing_caps == 0 && in->flushing_cap_tids.empty()) | |
5417 | in->flushing_cap_item.remove_myself(); | |
11fdf7f2 | 5418 | in->cap_snaps.erase(it); |
eafe8130 TL |
5419 | |
5420 | signal_cond_list(in->waitfor_caps); | |
5421 | if (session->flushing_caps_tids.empty() || | |
5422 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5423 | sync_cond.notify_all(); |
7c673cae FG |
5424 | } |
5425 | } else { | |
11fdf7f2 | 5426 | ldout(cct, 5) << __func__ << " DUP(?) mds." << mds << " flushed snap follows " << follows |
7c673cae FG |
5427 | << " on " << *in << dendl; |
5428 | // we may not have it if we send multiple FLUSHSNAP requests and (got multiple FLUSHEDSNAPs back) | |
5429 | } | |
7c673cae FG |
5430 | } |
5431 | ||
5432 | class C_Client_DentryInvalidate : public Context { | |
5433 | private: | |
5434 | Client *client; | |
5435 | vinodeno_t dirino; | |
5436 | vinodeno_t ino; | |
5437 | string name; | |
5438 | public: | |
5439 | C_Client_DentryInvalidate(Client *c, Dentry *dn, bool del) : | |
5440 | client(c), name(dn->name) { | |
5441 | if (client->use_faked_inos()) { | |
5442 | dirino.ino = dn->dir->parent_inode->faked_ino; | |
5443 | if (del) | |
5444 | ino.ino = dn->inode->faked_ino; | |
5445 | } else { | |
5446 | dirino = dn->dir->parent_inode->vino(); | |
5447 | if (del) | |
5448 | ino = dn->inode->vino(); | |
5449 | } | |
5450 | if (!del) | |
5451 | ino.ino = inodeno_t(); | |
5452 | } | |
5453 | void finish(int r) override { | |
5454 | // _async_dentry_invalidate is responsible for its own locking | |
9f95a23c | 5455 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
5456 | client->_async_dentry_invalidate(dirino, ino, name); |
5457 | } | |
5458 | }; | |
5459 | ||
5460 | void Client::_async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name) | |
5461 | { | |
f67539c2 TL |
5462 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
5463 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 5464 | return; |
f67539c2 | 5465 | |
11fdf7f2 | 5466 | ldout(cct, 10) << __func__ << " '" << name << "' ino " << ino |
7c673cae | 5467 | << " in dir " << dirino << dendl; |
e306af50 | 5468 | dentry_invalidate_cb(callback_handle, dirino, ino, name.c_str(), name.length()); |
7c673cae FG |
5469 | } |
5470 | ||
5471 | void Client::_schedule_invalidate_dentry_callback(Dentry *dn, bool del) | |
5472 | { | |
5473 | if (dentry_invalidate_cb && dn->inode->ll_ref > 0) | |
5474 | async_dentry_invalidator.queue(new C_Client_DentryInvalidate(this, dn, del)); | |
5475 | } | |
5476 | ||
5477 | void Client::_try_to_trim_inode(Inode *in, bool sched_inval) | |
5478 | { | |
b3b6e05e | 5479 | int ref = in->get_nref(); |
494da23a | 5480 | ldout(cct, 5) << __func__ << " in " << *in <<dendl; |
7c673cae FG |
5481 | |
5482 | if (in->dir && !in->dir->dentries.empty()) { | |
5483 | for (auto p = in->dir->dentries.begin(); | |
5484 | p != in->dir->dentries.end(); ) { | |
5485 | Dentry *dn = p->second; | |
5486 | ++p; | |
5487 | /* rmsnap removes whole subtree, need trim inodes recursively. | |
5488 | * we don't need to invalidate dentries recursively. because | |
5489 | * invalidating a directory dentry effectively invalidate | |
5490 | * whole subtree */ | |
5491 | if (in->snapid != CEPH_NOSNAP && dn->inode && dn->inode->is_dir()) | |
5492 | _try_to_trim_inode(dn->inode.get(), false); | |
5493 | ||
5494 | if (dn->lru_is_expireable()) | |
5495 | unlink(dn, true, false); // keep dir, drop dentry | |
5496 | } | |
5497 | if (in->dir->dentries.empty()) { | |
5498 | close_dir(in->dir); | |
5499 | --ref; | |
5500 | } | |
5501 | } | |
5502 | ||
b3b6e05e | 5503 | if (ref > 1 && (in->flags & I_SNAPDIR_OPEN)) { |
7c673cae FG |
5504 | InodeRef snapdir = open_snapdir(in); |
5505 | _try_to_trim_inode(snapdir.get(), false); | |
5506 | --ref; | |
5507 | } | |
5508 | ||
b3b6e05e | 5509 | if (ref > 1) { |
11fdf7f2 TL |
5510 | auto q = in->dentries.begin(); |
5511 | while (q != in->dentries.end()) { | |
5512 | Dentry *dn = *q; | |
5513 | ++q; | |
494da23a TL |
5514 | if( in->ll_ref > 0 && sched_inval) { |
5515 | // FIXME: we play lots of unlink/link tricks when handling MDS replies, | |
5516 | // so in->dentries doesn't always reflect the state of kernel's dcache. | |
5517 | _schedule_invalidate_dentry_callback(dn, true); | |
5518 | } | |
7c673cae FG |
5519 | unlink(dn, true, true); |
5520 | } | |
5521 | } | |
5522 | } | |
5523 | ||
11fdf7f2 | 5524 | void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5525 | { |
5526 | mds_rank_t mds = session->mds_num; | |
5527 | int used = get_caps_used(in); | |
5528 | int wanted = in->caps_wanted(); | |
a4b75251 | 5529 | int flags = 0; |
7c673cae | 5530 | |
a8e16298 TL |
5531 | const unsigned new_caps = m->get_caps(); |
5532 | const bool was_stale = session->cap_gen > cap->gen; | |
11fdf7f2 | 5533 | ldout(cct, 5) << __func__ << " on in " << m->get_ino() |
7c673cae FG |
5534 | << " mds." << mds << " seq " << m->get_seq() |
5535 | << " caps now " << ccap_string(new_caps) | |
a8e16298 | 5536 | << " was " << ccap_string(cap->issued) |
92f5a8d4 | 5537 | << (was_stale ? " (stale)" : "") << dendl; |
a8e16298 TL |
5538 | |
5539 | if (was_stale) | |
5540 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
7c673cae | 5541 | cap->seq = m->get_seq(); |
28e407b8 | 5542 | cap->gen = session->cap_gen; |
7c673cae | 5543 | |
11fdf7f2 | 5544 | check_cap_issue(in, new_caps); |
a8e16298 | 5545 | |
7c673cae | 5546 | // update inode |
1adf2230 AA |
5547 | int issued; |
5548 | in->caps_issued(&issued); | |
5549 | issued |= in->caps_dirty(); | |
7c673cae | 5550 | |
1adf2230 AA |
5551 | if ((new_caps & CEPH_CAP_AUTH_SHARED) && |
5552 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
5553 | in->mode = m->head.mode; |
5554 | in->uid = m->head.uid; | |
5555 | in->gid = m->head.gid; | |
5556 | in->btime = m->btime; | |
5557 | } | |
5558 | bool deleted_inode = false; | |
1adf2230 AA |
5559 | if ((new_caps & CEPH_CAP_LINK_SHARED) && |
5560 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
7c673cae | 5561 | in->nlink = m->head.nlink; |
20effc67 | 5562 | if (in->nlink == 0) |
7c673cae FG |
5563 | deleted_inode = true; |
5564 | } | |
1adf2230 | 5565 | if (!(issued & CEPH_CAP_XATTR_EXCL) && |
7c673cae FG |
5566 | m->xattrbl.length() && |
5567 | m->head.xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
5568 | auto p = m->xattrbl.cbegin(); |
5569 | decode(in->xattrs, p); | |
7c673cae FG |
5570 | in->xattr_version = m->head.xattr_version; |
5571 | } | |
28e407b8 AA |
5572 | |
5573 | if ((new_caps & CEPH_CAP_FILE_SHARED) && m->dirstat_is_valid()) { | |
5574 | in->dirstat.nfiles = m->get_nfiles(); | |
5575 | in->dirstat.nsubdirs = m->get_nsubdirs(); | |
5576 | } | |
5577 | ||
1adf2230 AA |
5578 | if (new_caps & CEPH_CAP_ANY_RD) { |
5579 | update_inode_file_time(in, issued, m->get_time_warp_seq(), | |
5580 | m->get_ctime(), m->get_mtime(), m->get_atime()); | |
5581 | } | |
5582 | ||
5583 | if (new_caps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { | |
5584 | in->layout = m->get_layout(); | |
5585 | update_inode_file_size(in, issued, m->get_size(), | |
5586 | m->get_truncate_seq(), m->get_truncate_size()); | |
5587 | } | |
5588 | ||
5589 | if (m->inline_version > in->inline_version) { | |
5590 | in->inline_data = m->inline_data; | |
5591 | in->inline_version = m->inline_version; | |
5592 | } | |
5593 | ||
5594 | /* always take a newer change attr */ | |
5595 | if (m->get_change_attr() > in->change_attr) | |
5596 | in->change_attr = m->get_change_attr(); | |
7c673cae FG |
5597 | |
5598 | // max_size | |
5599 | if (cap == in->auth_cap && | |
1adf2230 AA |
5600 | (new_caps & CEPH_CAP_ANY_FILE_WR) && |
5601 | (m->get_max_size() != in->max_size)) { | |
7c673cae FG |
5602 | ldout(cct, 10) << "max_size " << in->max_size << " -> " << m->get_max_size() << dendl; |
5603 | in->max_size = m->get_max_size(); | |
5604 | if (in->max_size > in->wanted_max_size) { | |
5605 | in->wanted_max_size = 0; | |
5606 | in->requested_max_size = 0; | |
5607 | } | |
5608 | } | |
5609 | ||
5610 | bool check = false; | |
a8e16298 TL |
5611 | if ((was_stale || m->get_op() == CEPH_CAP_OP_IMPORT) && |
5612 | (wanted & ~(cap->wanted | new_caps))) { | |
5613 | // If mds is importing cap, prior cap messages that update 'wanted' | |
5614 | // may get dropped by mds (migrate seq mismatch). | |
5615 | // | |
5616 | // We don't send cap message to update 'wanted' if what we want are | |
5617 | // already issued. If mds revokes caps, cap message that releases caps | |
5618 | // also tells mds what we want. But if caps got revoked by mds forcedly | |
5619 | // (session stale). We may haven't told mds what we want. | |
7c673cae | 5620 | check = true; |
a8e16298 | 5621 | } |
7c673cae | 5622 | |
7c673cae FG |
5623 | |
5624 | // update caps | |
a8e16298 | 5625 | auto revoked = cap->issued & ~new_caps; |
b32b8144 FG |
5626 | if (revoked) { |
5627 | ldout(cct, 10) << " revocation of " << ccap_string(revoked) << dendl; | |
7c673cae FG |
5628 | cap->issued = new_caps; |
5629 | cap->implemented |= new_caps; | |
5630 | ||
b32b8144 FG |
5631 | // recall delegations if we're losing caps necessary for them |
5632 | if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_RD)) | |
5633 | in->recall_deleg(false); | |
5634 | else if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_WR)) | |
5635 | in->recall_deleg(true); | |
5636 | ||
11fdf7f2 TL |
5637 | used = adjust_caps_used_for_lazyio(used, cap->issued, cap->implemented); |
5638 | if ((used & revoked & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO)) && | |
28e407b8 | 5639 | !_flush(in, new C_Client_FlushComplete(this, in))) { |
7c673cae | 5640 | // waitin' for flush |
11fdf7f2 | 5641 | } else if (used & revoked & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) { |
a4b75251 TL |
5642 | if (_release(in)) { |
5643 | check = true; | |
5644 | flags = CHECK_CAPS_NODELAY; | |
5645 | } | |
7c673cae FG |
5646 | } else { |
5647 | cap->wanted = 0; // don't let check_caps skip sending a response to MDS | |
5648 | check = true; | |
a4b75251 | 5649 | flags = CHECK_CAPS_NODELAY; |
7c673cae | 5650 | } |
a8e16298 TL |
5651 | } else if (cap->issued == new_caps) { |
5652 | ldout(cct, 10) << " caps unchanged at " << ccap_string(cap->issued) << dendl; | |
7c673cae | 5653 | } else { |
a8e16298 | 5654 | ldout(cct, 10) << " grant, new caps are " << ccap_string(new_caps & ~cap->issued) << dendl; |
7c673cae FG |
5655 | cap->issued = new_caps; |
5656 | cap->implemented |= new_caps; | |
5657 | ||
5658 | if (cap == in->auth_cap) { | |
5659 | // non-auth MDS is revoking the newly grant caps ? | |
11fdf7f2 TL |
5660 | for (const auto &p : in->caps) { |
5661 | if (&p.second == cap) | |
7c673cae | 5662 | continue; |
11fdf7f2 | 5663 | if (p.second.implemented & ~p.second.issued & new_caps) { |
7c673cae FG |
5664 | check = true; |
5665 | break; | |
5666 | } | |
5667 | } | |
5668 | } | |
5669 | } | |
5670 | ||
5671 | if (check) | |
a4b75251 | 5672 | check_caps(in, flags); |
7c673cae FG |
5673 | |
5674 | // wake up waiters | |
5675 | if (new_caps) | |
5676 | signal_cond_list(in->waitfor_caps); | |
5677 | ||
5678 | // may drop inode's last ref | |
5679 | if (deleted_inode) | |
5680 | _try_to_trim_inode(in, true); | |
7c673cae FG |
5681 | } |
5682 | ||
7c673cae FG |
5683 | int Client::inode_permission(Inode *in, const UserPerm& perms, unsigned want) |
5684 | { | |
b3b6e05e | 5685 | if (perms.uid() == 0) { |
2a845540 TL |
5686 | // For directories, DACs are overridable. |
5687 | // For files, Read/write DACs are always overridable but executable DACs are | |
5688 | // overridable when there is at least one exec bit set | |
5689 | if(!S_ISDIR(in->mode) && (want & MAY_EXEC) && !(in->mode & S_IXUGO)) | |
b3b6e05e | 5690 | return -CEPHFS_EACCES; |
7c673cae | 5691 | return 0; |
b3b6e05e | 5692 | } |
7c673cae FG |
5693 | |
5694 | if (perms.uid() != in->uid && (in->mode & S_IRWXG)) { | |
5695 | int ret = _posix_acl_permission(in, perms, want); | |
f67539c2 | 5696 | if (ret != -CEPHFS_EAGAIN) |
7c673cae FG |
5697 | return ret; |
5698 | } | |
5699 | ||
5700 | // check permissions before doing anything else | |
5701 | if (!in->check_mode(perms, want)) | |
f67539c2 | 5702 | return -CEPHFS_EACCES; |
7c673cae FG |
5703 | return 0; |
5704 | } | |
5705 | ||
5706 | int Client::xattr_permission(Inode *in, const char *name, unsigned want, | |
5707 | const UserPerm& perms) | |
5708 | { | |
5709 | int r = _getattr_for_perm(in, perms); | |
5710 | if (r < 0) | |
5711 | goto out; | |
5712 | ||
5713 | r = 0; | |
5714 | if (strncmp(name, "system.", 7) == 0) { | |
5715 | if ((want & MAY_WRITE) && (perms.uid() != 0 && perms.uid() != in->uid)) | |
f67539c2 | 5716 | r = -CEPHFS_EPERM; |
7c673cae FG |
5717 | } else { |
5718 | r = inode_permission(in, perms, want); | |
5719 | } | |
5720 | out: | |
1adf2230 | 5721 | ldout(cct, 5) << __func__ << " " << in << " = " << r << dendl; |
7c673cae FG |
5722 | return r; |
5723 | } | |
5724 | ||
20effc67 | 5725 | std::ostream& operator<<(std::ostream &out, const UserPerm& perm) { |
7c673cae FG |
5726 | out << "UserPerm(uid: " << perm.uid() << ", gid: " << perm.gid() << ")"; |
5727 | return out; | |
5728 | } | |
5729 | ||
5730 | int Client::may_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
5731 | const UserPerm& perms) | |
5732 | { | |
181888fb | 5733 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5734 | int r = _getattr_for_perm(in, perms); |
5735 | if (r < 0) | |
5736 | goto out; | |
5737 | ||
5738 | if (mask & CEPH_SETATTR_SIZE) { | |
5739 | r = inode_permission(in, perms, MAY_WRITE); | |
5740 | if (r < 0) | |
5741 | goto out; | |
5742 | } | |
5743 | ||
f67539c2 | 5744 | r = -CEPHFS_EPERM; |
7c673cae FG |
5745 | if (mask & CEPH_SETATTR_UID) { |
5746 | if (perms.uid() != 0 && (perms.uid() != in->uid || stx->stx_uid != in->uid)) | |
5747 | goto out; | |
5748 | } | |
5749 | if (mask & CEPH_SETATTR_GID) { | |
5750 | if (perms.uid() != 0 && (perms.uid() != in->uid || | |
5751 | (!perms.gid_in_groups(stx->stx_gid) && stx->stx_gid != in->gid))) | |
5752 | goto out; | |
5753 | } | |
5754 | ||
5755 | if (mask & CEPH_SETATTR_MODE) { | |
5756 | if (perms.uid() != 0 && perms.uid() != in->uid) | |
5757 | goto out; | |
5758 | ||
5759 | gid_t i_gid = (mask & CEPH_SETATTR_GID) ? stx->stx_gid : in->gid; | |
5760 | if (perms.uid() != 0 && !perms.gid_in_groups(i_gid)) | |
5761 | stx->stx_mode &= ~S_ISGID; | |
5762 | } | |
5763 | ||
5764 | if (mask & (CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME | | |
5765 | CEPH_SETATTR_MTIME | CEPH_SETATTR_ATIME)) { | |
5766 | if (perms.uid() != 0 && perms.uid() != in->uid) { | |
5767 | int check_mask = CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME; | |
5768 | if (!(mask & CEPH_SETATTR_MTIME_NOW)) | |
5769 | check_mask |= CEPH_SETATTR_MTIME; | |
5770 | if (!(mask & CEPH_SETATTR_ATIME_NOW)) | |
5771 | check_mask |= CEPH_SETATTR_ATIME; | |
5772 | if (check_mask & mask) { | |
5773 | goto out; | |
5774 | } else { | |
5775 | r = inode_permission(in, perms, MAY_WRITE); | |
5776 | if (r < 0) | |
5777 | goto out; | |
5778 | } | |
5779 | } | |
5780 | } | |
5781 | r = 0; | |
5782 | out: | |
5783 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5784 | return r; | |
5785 | } | |
5786 | ||
5787 | int Client::may_open(Inode *in, int flags, const UserPerm& perms) | |
5788 | { | |
181888fb | 5789 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5790 | unsigned want = 0; |
5791 | ||
5792 | if ((flags & O_ACCMODE) == O_WRONLY) | |
5793 | want = MAY_WRITE; | |
5794 | else if ((flags & O_ACCMODE) == O_RDWR) | |
5795 | want = MAY_READ | MAY_WRITE; | |
5796 | else if ((flags & O_ACCMODE) == O_RDONLY) | |
5797 | want = MAY_READ; | |
5798 | if (flags & O_TRUNC) | |
5799 | want |= MAY_WRITE; | |
5800 | ||
5801 | int r = 0; | |
5802 | switch (in->mode & S_IFMT) { | |
5803 | case S_IFLNK: | |
f67539c2 | 5804 | r = -CEPHFS_ELOOP; |
7c673cae FG |
5805 | goto out; |
5806 | case S_IFDIR: | |
5807 | if (want & MAY_WRITE) { | |
f67539c2 | 5808 | r = -CEPHFS_EISDIR; |
7c673cae FG |
5809 | goto out; |
5810 | } | |
5811 | break; | |
5812 | } | |
5813 | ||
5814 | r = _getattr_for_perm(in, perms); | |
5815 | if (r < 0) | |
5816 | goto out; | |
5817 | ||
5818 | r = inode_permission(in, perms, want); | |
5819 | out: | |
5820 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5821 | return r; | |
5822 | } | |
5823 | ||
5824 | int Client::may_lookup(Inode *dir, const UserPerm& perms) | |
5825 | { | |
181888fb | 5826 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5827 | int r = _getattr_for_perm(dir, perms); |
5828 | if (r < 0) | |
5829 | goto out; | |
5830 | ||
5831 | r = inode_permission(dir, perms, MAY_EXEC); | |
5832 | out: | |
5833 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5834 | return r; | |
5835 | } | |
5836 | ||
5837 | int Client::may_create(Inode *dir, const UserPerm& perms) | |
5838 | { | |
181888fb | 5839 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5840 | int r = _getattr_for_perm(dir, perms); |
5841 | if (r < 0) | |
5842 | goto out; | |
5843 | ||
5844 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5845 | out: | |
5846 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5847 | return r; | |
5848 | } | |
5849 | ||
5850 | int Client::may_delete(Inode *dir, const char *name, const UserPerm& perms) | |
5851 | { | |
181888fb | 5852 | ldout(cct, 20) << __func__ << " " << *dir << "; " << "; name " << name << "; " << perms << dendl; |
7c673cae FG |
5853 | int r = _getattr_for_perm(dir, perms); |
5854 | if (r < 0) | |
5855 | goto out; | |
5856 | ||
5857 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5858 | if (r < 0) | |
5859 | goto out; | |
5860 | ||
f67539c2 | 5861 | /* 'name == NULL' means rmsnap w/o permission checks */ |
7c673cae FG |
5862 | if (perms.uid() != 0 && name && (dir->mode & S_ISVTX)) { |
5863 | InodeRef otherin; | |
5864 | r = _lookup(dir, name, CEPH_CAP_AUTH_SHARED, &otherin, perms); | |
5865 | if (r < 0) | |
5866 | goto out; | |
5867 | if (dir->uid != perms.uid() && otherin->uid != perms.uid()) | |
f67539c2 | 5868 | r = -CEPHFS_EPERM; |
7c673cae FG |
5869 | } |
5870 | out: | |
5871 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5872 | return r; | |
5873 | } | |
5874 | ||
f67539c2 TL |
5875 | int Client::may_delete(const char *relpath, const UserPerm& perms) { |
5876 | ldout(cct, 20) << __func__ << " " << relpath << "; " << perms << dendl; | |
5877 | ||
5878 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
5879 | if (!mref_reader.is_state_satisfied()) | |
5880 | return -ENOTCONN; | |
5881 | ||
5882 | filepath path(relpath); | |
5883 | string name = path.last_dentry(); | |
5884 | path.pop_dentry(); | |
5885 | InodeRef dir; | |
5886 | ||
5887 | std::scoped_lock lock(client_lock); | |
5888 | int r = path_walk(path, &dir, perms); | |
5889 | if (r < 0) | |
5890 | return r; | |
5891 | if (cct->_conf->client_permissions) { | |
5892 | int r = may_delete(dir.get(), name.c_str(), perms); | |
5893 | if (r < 0) | |
5894 | return r; | |
5895 | } | |
5896 | ||
5897 | return 0; | |
5898 | } | |
5899 | ||
7c673cae FG |
5900 | int Client::may_hardlink(Inode *in, const UserPerm& perms) |
5901 | { | |
181888fb | 5902 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5903 | int r = _getattr_for_perm(in, perms); |
5904 | if (r < 0) | |
5905 | goto out; | |
5906 | ||
5907 | if (perms.uid() == 0 || perms.uid() == in->uid) { | |
5908 | r = 0; | |
5909 | goto out; | |
5910 | } | |
5911 | ||
f67539c2 | 5912 | r = -CEPHFS_EPERM; |
7c673cae FG |
5913 | if (!S_ISREG(in->mode)) |
5914 | goto out; | |
5915 | ||
5916 | if (in->mode & S_ISUID) | |
5917 | goto out; | |
5918 | ||
5919 | if ((in->mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
5920 | goto out; | |
5921 | ||
5922 | r = inode_permission(in, perms, MAY_READ | MAY_WRITE); | |
5923 | out: | |
5924 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5925 | return r; | |
5926 | } | |
5927 | ||
5928 | int Client::_getattr_for_perm(Inode *in, const UserPerm& perms) | |
5929 | { | |
5930 | int mask = CEPH_STAT_CAP_MODE; | |
5931 | bool force = false; | |
5932 | if (acl_type != NO_ACL) { | |
5933 | mask |= CEPH_STAT_CAP_XATTR; | |
5934 | force = in->xattr_version == 0; | |
5935 | } | |
5936 | return _getattr(in, mask, perms, force); | |
5937 | } | |
5938 | ||
5939 | vinodeno_t Client::_get_vino(Inode *in) | |
5940 | { | |
5941 | /* The caller must hold the client lock */ | |
5942 | return vinodeno_t(in->ino, in->snapid); | |
5943 | } | |
5944 | ||
7c673cae FG |
5945 | /** |
5946 | * Resolve an MDS spec to a list of MDS daemon GIDs. | |
5947 | * | |
5948 | * The spec is a string representing a GID, rank, filesystem:rank, or name/id. | |
5949 | * It may be '*' in which case it matches all GIDs. | |
5950 | * | |
5951 | * If no error is returned, the `targets` vector will be populated with at least | |
5952 | * one MDS. | |
5953 | */ | |
5954 | int Client::resolve_mds( | |
5955 | const std::string &mds_spec, | |
5956 | std::vector<mds_gid_t> *targets) | |
5957 | { | |
11fdf7f2 TL |
5958 | ceph_assert(fsmap); |
5959 | ceph_assert(targets != nullptr); | |
7c673cae FG |
5960 | |
5961 | mds_role_t role; | |
f67539c2 TL |
5962 | CachedStackStringStream css; |
5963 | int role_r = fsmap->parse_role(mds_spec, &role, *css); | |
7c673cae FG |
5964 | if (role_r == 0) { |
5965 | // We got a role, resolve it to a GID | |
f67539c2 TL |
5966 | auto& info = fsmap->get_filesystem(role.fscid)->mds_map.get_info(role.rank); |
5967 | ldout(cct, 10) << __func__ << ": resolved " << mds_spec << " to role '" | |
5968 | << role << "' aka " << info.human_name() << dendl; | |
5969 | targets->push_back(info.global_id); | |
7c673cae FG |
5970 | return 0; |
5971 | } | |
5972 | ||
5973 | std::string strtol_err; | |
5974 | long long rank_or_gid = strict_strtoll(mds_spec.c_str(), 10, &strtol_err); | |
5975 | if (strtol_err.empty()) { | |
5976 | // It is a possible GID | |
5977 | const mds_gid_t mds_gid = mds_gid_t(rank_or_gid); | |
5978 | if (fsmap->gid_exists(mds_gid)) { | |
f67539c2 TL |
5979 | auto& info = fsmap->get_info_gid(mds_gid); |
5980 | ldout(cct, 10) << __func__ << ": validated gid " << mds_gid << " aka " | |
5981 | << info.human_name() << dendl; | |
7c673cae | 5982 | targets->push_back(mds_gid); |
f67539c2 | 5983 | return 0; |
7c673cae | 5984 | } else { |
f67539c2 | 5985 | lderr(cct) << __func__ << ": gid " << mds_gid << " not in MDS map" |
7c673cae | 5986 | << dendl; |
f67539c2 TL |
5987 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
5988 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5989 | } |
5990 | } else if (mds_spec == "*") { | |
5991 | // It is a wildcard: use all MDSs | |
f67539c2 | 5992 | const auto& mds_info = fsmap->get_mds_info(); |
7c673cae | 5993 | |
f67539c2 | 5994 | ldout(cct, 10) << __func__ << ": resolving `*' to all MDS daemons" << dendl; |
7c673cae | 5995 | if (mds_info.empty()) { |
f67539c2 TL |
5996 | lderr(cct) << __func__ << ": no MDS daemons found" << dendl; |
5997 | lderr(cct) << "FSMap: " << *fsmap << dendl; | |
5998 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5999 | } |
6000 | ||
f67539c2 TL |
6001 | for (const auto& [gid, info] : mds_info) { |
6002 | ldout(cct, 10) << __func__ << ": appending " << info.human_name() << " to targets" << dendl; | |
6003 | targets->push_back(gid); | |
7c673cae | 6004 | } |
f67539c2 | 6005 | return 0; |
7c673cae FG |
6006 | } else { |
6007 | // It did not parse as an integer, it is not a wildcard, it must be a name | |
6008 | const mds_gid_t mds_gid = fsmap->find_mds_gid_by_name(mds_spec); | |
6009 | if (mds_gid == 0) { | |
f67539c2 | 6010 | lderr(cct) << __func__ << ": no MDS daemons found by name `" << mds_spec << "'" << dendl; |
7c673cae | 6011 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
f67539c2 | 6012 | return -CEPHFS_ENOENT; |
7c673cae | 6013 | } else { |
f67539c2 TL |
6014 | auto& info = fsmap->get_info_gid(mds_gid); |
6015 | ldout(cct, 10) << __func__ << ": resolved name '" << mds_spec | |
6016 | << "' to " << info.human_name() << dendl; | |
7c673cae FG |
6017 | targets->push_back(mds_gid); |
6018 | } | |
f67539c2 | 6019 | return 0; |
7c673cae | 6020 | } |
7c673cae FG |
6021 | } |
6022 | ||
6023 | ||
6024 | /** | |
6025 | * Authenticate with mon and establish global ID | |
6026 | */ | |
6027 | int Client::authenticate() | |
6028 | { | |
9f95a23c | 6029 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
6030 | |
6031 | if (monclient->is_authenticated()) { | |
6032 | return 0; | |
6033 | } | |
6034 | ||
9f95a23c | 6035 | client_lock.unlock(); |
2a845540 | 6036 | int r = monclient->authenticate(std::chrono::duration<double>(mount_timeout).count()); |
9f95a23c | 6037 | client_lock.lock(); |
7c673cae FG |
6038 | if (r < 0) { |
6039 | return r; | |
6040 | } | |
6041 | ||
6042 | whoami = monclient->get_global_id(); | |
6043 | messenger->set_myname(entity_name_t::CLIENT(whoami.v)); | |
6044 | ||
6045 | return 0; | |
6046 | } | |
6047 | ||
6048 | int Client::fetch_fsmap(bool user) | |
6049 | { | |
f67539c2 TL |
6050 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
6051 | ||
7c673cae FG |
6052 | // Retrieve FSMap to enable looking up daemon addresses. We need FSMap |
6053 | // rather than MDSMap because no one MDSMap contains all the daemons, and | |
6054 | // a `tell` can address any daemon. | |
6055 | version_t fsmap_latest; | |
f67539c2 | 6056 | bs::error_code ec; |
7c673cae | 6057 | do { |
9f95a23c | 6058 | client_lock.unlock(); |
f67539c2 TL |
6059 | std::tie(fsmap_latest, std::ignore) = |
6060 | monclient->get_version("fsmap", ca::use_blocked[ec]); | |
9f95a23c | 6061 | client_lock.lock(); |
f67539c2 | 6062 | } while (ec == bs::errc::resource_unavailable_try_again); |
7c673cae | 6063 | |
f67539c2 TL |
6064 | if (ec) { |
6065 | lderr(cct) << "Failed to learn FSMap version: " << ec << dendl; | |
6066 | return ceph::from_error_code(ec); | |
7c673cae FG |
6067 | } |
6068 | ||
6069 | ldout(cct, 10) << __func__ << " learned FSMap version " << fsmap_latest << dendl; | |
6070 | ||
6071 | if (user) { | |
6072 | if (!fsmap_user || fsmap_user->get_epoch() < fsmap_latest) { | |
6073 | monclient->sub_want("fsmap.user", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
6074 | monclient->renew_subs(); | |
6075 | wait_on_list(waiting_for_fsmap); | |
6076 | } | |
11fdf7f2 TL |
6077 | ceph_assert(fsmap_user); |
6078 | ceph_assert(fsmap_user->get_epoch() >= fsmap_latest); | |
7c673cae FG |
6079 | } else { |
6080 | if (!fsmap || fsmap->get_epoch() < fsmap_latest) { | |
6081 | monclient->sub_want("fsmap", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
6082 | monclient->renew_subs(); | |
6083 | wait_on_list(waiting_for_fsmap); | |
6084 | } | |
11fdf7f2 TL |
6085 | ceph_assert(fsmap); |
6086 | ceph_assert(fsmap->get_epoch() >= fsmap_latest); | |
7c673cae FG |
6087 | } |
6088 | ldout(cct, 10) << __func__ << " finished waiting for FSMap version " | |
6089 | << fsmap_latest << dendl; | |
6090 | return 0; | |
6091 | } | |
6092 | ||
6093 | /** | |
6094 | * | |
6095 | * @mds_spec one of ID, rank, GID, "*" | |
6096 | * | |
6097 | */ | |
6098 | int Client::mds_command( | |
6099 | const std::string &mds_spec, | |
6100 | const vector<string>& cmd, | |
6101 | const bufferlist& inbl, | |
6102 | bufferlist *outbl, | |
6103 | string *outs, | |
6104 | Context *onfinish) | |
6105 | { | |
f67539c2 TL |
6106 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
6107 | if (!iref_reader.is_state_satisfied()) | |
6108 | return -CEPHFS_ENOTCONN; | |
7c673cae | 6109 | |
f67539c2 | 6110 | std::unique_lock cl(client_lock); |
7c673cae FG |
6111 | |
6112 | int r; | |
6113 | r = authenticate(); | |
6114 | if (r < 0) { | |
6115 | return r; | |
6116 | } | |
6117 | ||
6118 | r = fetch_fsmap(false); | |
6119 | if (r < 0) { | |
6120 | return r; | |
6121 | } | |
6122 | ||
6123 | // Look up MDS target(s) of the command | |
6124 | std::vector<mds_gid_t> targets; | |
6125 | r = resolve_mds(mds_spec, &targets); | |
6126 | if (r < 0) { | |
6127 | return r; | |
6128 | } | |
6129 | ||
6130 | // If daemons are laggy, we won't send them commands. If all | |
6131 | // are laggy then we fail. | |
6132 | std::vector<mds_gid_t> non_laggy; | |
f67539c2 | 6133 | for (const auto& gid : targets) { |
7c673cae FG |
6134 | const auto info = fsmap->get_info_gid(gid); |
6135 | if (!info.laggy()) { | |
6136 | non_laggy.push_back(gid); | |
6137 | } | |
6138 | } | |
6139 | if (non_laggy.size() == 0) { | |
6140 | *outs = "All targeted MDS daemons are laggy"; | |
f67539c2 | 6141 | return -CEPHFS_ENOENT; |
7c673cae FG |
6142 | } |
6143 | ||
6144 | if (metadata.empty()) { | |
6145 | // We are called on an unmounted client, so metadata | |
6146 | // won't be initialized yet. | |
6147 | populate_metadata(""); | |
6148 | } | |
6149 | ||
6150 | // Send commands to targets | |
6151 | C_GatherBuilder gather(cct, onfinish); | |
f67539c2 | 6152 | for (const auto& target_gid : non_laggy) { |
7c673cae FG |
6153 | const auto info = fsmap->get_info_gid(target_gid); |
6154 | ||
6155 | // Open a connection to the target MDS | |
11fdf7f2 | 6156 | ConnectionRef conn = messenger->connect_to_mds(info.get_addrs()); |
7c673cae | 6157 | |
f67539c2 TL |
6158 | cl.unlock(); |
6159 | { | |
6160 | std::scoped_lock cmd_lock(command_lock); | |
6161 | // Generate MDSCommandOp state | |
6162 | auto &op = command_table.start_command(); | |
7c673cae | 6163 | |
f67539c2 TL |
6164 | op.on_finish = gather.new_sub(); |
6165 | op.cmd = cmd; | |
6166 | op.outbl = outbl; | |
6167 | op.outs = outs; | |
6168 | op.inbl = inbl; | |
6169 | op.mds_gid = target_gid; | |
6170 | op.con = conn; | |
7c673cae | 6171 | |
f67539c2 TL |
6172 | ldout(cct, 4) << __func__ << ": new command op to " << target_gid |
6173 | << " tid=" << op.tid << cmd << dendl; | |
7c673cae | 6174 | |
f67539c2 TL |
6175 | // Construct and send MCommand |
6176 | MessageRef m = op.get_message(monclient->get_fsid()); | |
6177 | conn->send_message2(std::move(m)); | |
6178 | } | |
6179 | cl.lock(); | |
7c673cae FG |
6180 | } |
6181 | gather.activate(); | |
6182 | ||
6183 | return 0; | |
6184 | } | |
6185 | ||
11fdf7f2 | 6186 | void Client::handle_command_reply(const MConstRef<MCommandReply>& m) |
7c673cae FG |
6187 | { |
6188 | ceph_tid_t const tid = m->get_tid(); | |
6189 | ||
6190 | ldout(cct, 10) << __func__ << ": tid=" << m->get_tid() << dendl; | |
6191 | ||
f67539c2 | 6192 | std::scoped_lock cmd_lock(command_lock); |
7c673cae FG |
6193 | if (!command_table.exists(tid)) { |
6194 | ldout(cct, 1) << __func__ << ": unknown tid " << tid << ", dropping" << dendl; | |
7c673cae FG |
6195 | return; |
6196 | } | |
6197 | ||
6198 | auto &op = command_table.get_command(tid); | |
6199 | if (op.outbl) { | |
11fdf7f2 | 6200 | *op.outbl = m->get_data(); |
7c673cae FG |
6201 | } |
6202 | if (op.outs) { | |
6203 | *op.outs = m->rs; | |
6204 | } | |
6205 | ||
6206 | if (op.on_finish) { | |
6207 | op.on_finish->complete(m->r); | |
6208 | } | |
6209 | ||
6210 | command_table.erase(tid); | |
7c673cae FG |
6211 | } |
6212 | ||
6213 | // ------------------- | |
6214 | // MOUNT | |
6215 | ||
11fdf7f2 | 6216 | int Client::subscribe_mdsmap(const std::string &fs_name) |
7c673cae | 6217 | { |
7c673cae FG |
6218 | int r = authenticate(); |
6219 | if (r < 0) { | |
6220 | lderr(cct) << "authentication failed: " << cpp_strerror(r) << dendl; | |
6221 | return r; | |
6222 | } | |
6223 | ||
11fdf7f2 TL |
6224 | std::string resolved_fs_name; |
6225 | if (fs_name.empty()) { | |
9f95a23c TL |
6226 | resolved_fs_name = cct->_conf.get_val<std::string>("client_fs"); |
6227 | if (resolved_fs_name.empty()) | |
6228 | // Try the backwards compatibility fs name option | |
6229 | resolved_fs_name = cct->_conf.get_val<std::string>("client_mds_namespace"); | |
11fdf7f2 TL |
6230 | } else { |
6231 | resolved_fs_name = fs_name; | |
6232 | } | |
6233 | ||
7c673cae | 6234 | std::string want = "mdsmap"; |
11fdf7f2 | 6235 | if (!resolved_fs_name.empty()) { |
7c673cae FG |
6236 | r = fetch_fsmap(true); |
6237 | if (r < 0) | |
6238 | return r; | |
11fdf7f2 TL |
6239 | fscid = fsmap_user->get_fs_cid(resolved_fs_name); |
6240 | if (fscid == FS_CLUSTER_ID_NONE) { | |
f67539c2 | 6241 | return -CEPHFS_ENOENT; |
11fdf7f2 | 6242 | } |
7c673cae FG |
6243 | |
6244 | std::ostringstream oss; | |
11fdf7f2 | 6245 | oss << want << "." << fscid; |
7c673cae FG |
6246 | want = oss.str(); |
6247 | } | |
6248 | ldout(cct, 10) << "Subscribing to map '" << want << "'" << dendl; | |
6249 | ||
6250 | monclient->sub_want(want, 0, 0); | |
6251 | monclient->renew_subs(); | |
6252 | ||
11fdf7f2 TL |
6253 | return 0; |
6254 | } | |
6255 | ||
6256 | int Client::mount(const std::string &mount_root, const UserPerm& perms, | |
6257 | bool require_mds, const std::string &fs_name) | |
6258 | { | |
f67539c2 | 6259 | ceph_assert(is_initialized()); |
11fdf7f2 | 6260 | |
f67539c2 TL |
6261 | /* |
6262 | * To make sure that the _unmount() must wait until the mount() | |
6263 | * is done. | |
6264 | */ | |
6265 | RWRef_t mref_writer(mount_state, CLIENT_MOUNTING, false); | |
6266 | if (!mref_writer.is_first_writer()) // already mounting or mounted | |
11fdf7f2 | 6267 | return 0; |
11fdf7f2 | 6268 | |
f67539c2 | 6269 | std::unique_lock cl(client_lock); |
11fdf7f2 TL |
6270 | |
6271 | int r = subscribe_mdsmap(fs_name); | |
6272 | if (r < 0) { | |
6273 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
6274 | return r; | |
6275 | } | |
6276 | ||
f67539c2 TL |
6277 | start_tick_thread(); // start tick thread |
6278 | ||
7c673cae FG |
6279 | if (require_mds) { |
6280 | while (1) { | |
6281 | auto availability = mdsmap->is_cluster_available(); | |
6282 | if (availability == MDSMap::STUCK_UNAVAILABLE) { | |
6283 | // Error out | |
6284 | ldout(cct, 10) << "mds cluster unavailable: epoch=" << mdsmap->get_epoch() << dendl; | |
6285 | return CEPH_FUSE_NO_MDS_UP; | |
6286 | } else if (availability == MDSMap::AVAILABLE) { | |
6287 | // Continue to mount | |
6288 | break; | |
6289 | } else if (availability == MDSMap::TRANSIENT_UNAVAILABLE) { | |
6290 | // Else, wait. MDSMonitor will update the map to bring | |
6291 | // us to a conclusion eventually. | |
6292 | wait_on_list(waiting_for_mdsmap); | |
6293 | } else { | |
6294 | // Unexpected value! | |
6295 | ceph_abort(); | |
6296 | } | |
6297 | } | |
6298 | } | |
6299 | ||
6300 | populate_metadata(mount_root.empty() ? "/" : mount_root); | |
6301 | ||
6302 | filepath fp(CEPH_INO_ROOT); | |
6303 | if (!mount_root.empty()) { | |
6304 | fp = filepath(mount_root.c_str()); | |
6305 | } | |
6306 | while (true) { | |
6307 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
6308 | req->set_filepath(fp); | |
6309 | req->head.args.getattr.mask = CEPH_STAT_CAP_INODE_ALL; | |
6310 | int res = make_request(req, perms); | |
6311 | if (res < 0) { | |
f67539c2 | 6312 | if (res == -CEPHFS_EACCES && root) { |
7c673cae FG |
6313 | ldout(cct, 1) << __func__ << " EACCES on parent of mount point; quotas may not work" << dendl; |
6314 | break; | |
6315 | } | |
6316 | return res; | |
6317 | } | |
6318 | ||
6319 | if (fp.depth()) | |
6320 | fp.pop_dentry(); | |
6321 | else | |
6322 | break; | |
6323 | } | |
6324 | ||
11fdf7f2 | 6325 | ceph_assert(root); |
b3b6e05e | 6326 | _ll_get(root.get()); |
7c673cae | 6327 | |
7c673cae FG |
6328 | // trace? |
6329 | if (!cct->_conf->client_trace.empty()) { | |
6330 | traceout.open(cct->_conf->client_trace.c_str()); | |
6331 | if (traceout.is_open()) { | |
6332 | ldout(cct, 1) << "opened trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6333 | } else { | |
6334 | ldout(cct, 1) << "FAILED to open trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6335 | } | |
6336 | } | |
6337 | ||
6338 | /* | |
6339 | ldout(cct, 3) << "op: // client trace data structs" << dendl; | |
6340 | ldout(cct, 3) << "op: struct stat st;" << dendl; | |
6341 | ldout(cct, 3) << "op: struct utimbuf utim;" << dendl; | |
6342 | ldout(cct, 3) << "op: int readlinkbuf_len = 1000;" << dendl; | |
6343 | ldout(cct, 3) << "op: char readlinkbuf[readlinkbuf_len];" << dendl; | |
6344 | ldout(cct, 3) << "op: map<string, inode_t*> dir_contents;" << dendl; | |
6345 | ldout(cct, 3) << "op: map<int, int> open_files;" << dendl; | |
6346 | ldout(cct, 3) << "op: int fd;" << dendl; | |
6347 | */ | |
f67539c2 TL |
6348 | |
6349 | mref_writer.update_state(CLIENT_MOUNTED); | |
7c673cae FG |
6350 | return 0; |
6351 | } | |
6352 | ||
6353 | // UNMOUNT | |
6354 | ||
6355 | void Client::_close_sessions() | |
6356 | { | |
f6b5b4d7 | 6357 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
20effc67 | 6358 | if (it->second->state == MetaSession::STATE_REJECTED) |
f6b5b4d7 TL |
6359 | mds_sessions.erase(it++); |
6360 | else | |
6361 | ++it; | |
6362 | } | |
6363 | ||
7c673cae FG |
6364 | while (!mds_sessions.empty()) { |
6365 | // send session closes! | |
11fdf7f2 | 6366 | for (auto &p : mds_sessions) { |
20effc67 TL |
6367 | if (p.second->state != MetaSession::STATE_CLOSING) { |
6368 | _close_mds_session(p.second.get()); | |
f6b5b4d7 | 6369 | mds_ranks_closing.insert(p.first); |
7c673cae FG |
6370 | } |
6371 | } | |
6372 | ||
6373 | // wait for sessions to close | |
f6b5b4d7 TL |
6374 | double timo = cct->_conf.get_val<std::chrono::seconds>("client_shutdown_timeout").count(); |
6375 | ldout(cct, 2) << "waiting for " << mds_ranks_closing.size() << " mds session(s) to close (timeout: " | |
6376 | << timo << "s)" << dendl; | |
9f95a23c | 6377 | std::unique_lock l{client_lock, std::adopt_lock}; |
f6b5b4d7 TL |
6378 | if (!timo) { |
6379 | mount_cond.wait(l); | |
6380 | } else if (!mount_cond.wait_for(l, ceph::make_timespan(timo), [this] { return mds_ranks_closing.empty(); })) { | |
6381 | ldout(cct, 1) << mds_ranks_closing.size() << " mds(s) did not respond to session close -- timing out." << dendl; | |
6382 | while (!mds_ranks_closing.empty()) { | |
6383 | auto session = mds_sessions.at(*mds_ranks_closing.begin()); | |
6384 | // this prunes entry from mds_sessions and mds_ranks_closing | |
20effc67 | 6385 | _closed_mds_session(session.get(), -CEPHFS_ETIMEDOUT); |
f6b5b4d7 TL |
6386 | } |
6387 | } | |
6388 | ||
6389 | mds_ranks_closing.clear(); | |
9f95a23c | 6390 | l.release(); |
7c673cae FG |
6391 | } |
6392 | } | |
6393 | ||
522d829b TL |
6394 | void Client::flush_mdlog_sync(Inode *in) |
6395 | { | |
6396 | if (in->unsafe_ops.empty()) { | |
6397 | return; | |
6398 | } | |
6399 | ||
6400 | std::set<mds_rank_t> anchor; | |
6401 | for (auto &&p : in->unsafe_ops) { | |
6402 | anchor.emplace(p->mds); | |
6403 | } | |
6404 | if (in->auth_cap) { | |
6405 | anchor.emplace(in->auth_cap->session->mds_num); | |
6406 | } | |
6407 | ||
6408 | for (auto &rank : anchor) { | |
6409 | auto session = &mds_sessions.at(rank); | |
20effc67 | 6410 | flush_mdlog(session->get()); |
522d829b TL |
6411 | } |
6412 | } | |
6413 | ||
31f18b77 FG |
6414 | void Client::flush_mdlog_sync() |
6415 | { | |
522d829b | 6416 | if (mds_requests.empty()) |
31f18b77 | 6417 | return; |
11fdf7f2 | 6418 | for (auto &p : mds_sessions) { |
20effc67 | 6419 | flush_mdlog(p.second.get()); |
31f18b77 FG |
6420 | } |
6421 | } | |
6422 | ||
6423 | void Client::flush_mdlog(MetaSession *session) | |
6424 | { | |
6425 | // Only send this to Luminous or newer MDS daemons, older daemons | |
6426 | // will crash if they see an unknown CEPH_SESSION_* value in this msg. | |
6427 | const uint64_t features = session->con->get_features(); | |
6428 | if (HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
9f95a23c | 6429 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_FLUSH_MDLOG); |
11fdf7f2 | 6430 | session->con->send_message2(std::move(m)); |
31f18b77 FG |
6431 | } |
6432 | } | |
6433 | ||
6434 | ||
11fdf7f2 TL |
6435 | void Client::_abort_mds_sessions(int err) |
6436 | { | |
6437 | for (auto p = mds_requests.begin(); p != mds_requests.end(); ) { | |
6438 | auto req = p->second; | |
6439 | ++p; | |
6440 | // unsafe requests will be removed during close session below. | |
6441 | if (req->got_unsafe) | |
6442 | continue; | |
6443 | ||
6444 | req->abort(err); | |
6445 | if (req->caller_cond) { | |
6446 | req->kick = true; | |
9f95a23c | 6447 | req->caller_cond->notify_all(); |
11fdf7f2 TL |
6448 | } |
6449 | } | |
6450 | ||
6451 | // Process aborts on any requests that were on this waitlist. | |
6452 | // Any requests that were on a waiting_for_open session waitlist | |
6453 | // will get kicked during close session below. | |
6454 | signal_cond_list(waiting_for_mdsmap); | |
6455 | ||
6456 | // Force-close all sessions | |
6457 | while(!mds_sessions.empty()) { | |
20effc67 TL |
6458 | auto session = mds_sessions.begin()->second; |
6459 | _closed_mds_session(session.get(), err); | |
11fdf7f2 TL |
6460 | } |
6461 | } | |
6462 | ||
6463 | void Client::_unmount(bool abort) | |
7c673cae | 6464 | { |
f67539c2 TL |
6465 | /* |
6466 | * We are unmounting the client. | |
6467 | * | |
6468 | * Just declare the state to STATE_UNMOUNTING to block and fail | |
6469 | * any new comming "reader" and then try to wait all the in-flight | |
6470 | * "readers" to finish. | |
6471 | */ | |
6472 | RWRef_t mref_writer(mount_state, CLIENT_UNMOUNTING, false); | |
6473 | if (!mref_writer.is_first_writer()) | |
181888fb | 6474 | return; |
f67539c2 | 6475 | mref_writer.wait_readers_done(); |
7c673cae | 6476 | |
f67539c2 TL |
6477 | std::unique_lock lock{client_lock}; |
6478 | ||
6479 | if (abort || blocklisted) { | |
6480 | ldout(cct, 2) << "unmounting (" << (abort ? "abort)" : "blocklisted)") << dendl; | |
11fdf7f2 TL |
6481 | } else { |
6482 | ldout(cct, 2) << "unmounting" << dendl; | |
6483 | } | |
7c673cae | 6484 | |
b32b8144 FG |
6485 | deleg_timeout = 0; |
6486 | ||
11fdf7f2 | 6487 | if (abort) { |
f67539c2 | 6488 | mount_aborted = true; |
11fdf7f2 | 6489 | // Abort all mds sessions |
f67539c2 | 6490 | _abort_mds_sessions(-CEPHFS_ENOTCONN); |
11fdf7f2 | 6491 | |
f67539c2 | 6492 | objecter->op_cancel_writes(-CEPHFS_ENOTCONN); |
11fdf7f2 TL |
6493 | } else { |
6494 | // flush the mdlog for pending requests, if any | |
6495 | flush_mdlog_sync(); | |
6496 | } | |
6497 | ||
9f95a23c TL |
6498 | mount_cond.wait(lock, [this] { |
6499 | if (!mds_requests.empty()) { | |
6500 | ldout(cct, 10) << "waiting on " << mds_requests.size() << " requests" | |
6501 | << dendl; | |
6502 | } | |
6503 | return mds_requests.empty(); | |
6504 | }); | |
7c673cae FG |
6505 | |
6506 | cwd.reset(); | |
b3b6e05e | 6507 | root.reset(); |
7c673cae FG |
6508 | |
6509 | // clean up any unclosed files | |
6510 | while (!fd_map.empty()) { | |
6511 | Fh *fh = fd_map.begin()->second; | |
6512 | fd_map.erase(fd_map.begin()); | |
6513 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *fh->inode << dendl; | |
6514 | _release_fh(fh); | |
6515 | } | |
6516 | ||
6517 | while (!ll_unclosed_fh_set.empty()) { | |
6518 | set<Fh*>::iterator it = ll_unclosed_fh_set.begin(); | |
6519 | Fh *fh = *it; | |
6520 | ll_unclosed_fh_set.erase(fh); | |
6521 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *(fh->inode) << dendl; | |
6522 | _release_fh(fh); | |
6523 | } | |
6524 | ||
6525 | while (!opened_dirs.empty()) { | |
6526 | dir_result_t *dirp = *opened_dirs.begin(); | |
6527 | ldout(cct, 0) << " destroyed lost open dir " << dirp << " on " << *dirp->inode << dendl; | |
6528 | _closedir(dirp); | |
6529 | } | |
6530 | ||
6531 | _ll_drop_pins(); | |
6532 | ||
7c673cae FG |
6533 | if (cct->_conf->client_oc) { |
6534 | // flush/release all buffered data | |
11fdf7f2 TL |
6535 | std::list<InodeRef> anchor; |
6536 | for (auto& p : inode_map) { | |
6537 | Inode *in = p.second; | |
7c673cae | 6538 | if (!in) { |
11fdf7f2 TL |
6539 | ldout(cct, 0) << "null inode_map entry ino " << p.first << dendl; |
6540 | ceph_assert(in); | |
7c673cae | 6541 | } |
11fdf7f2 TL |
6542 | |
6543 | // prevent inode from getting freed | |
6544 | anchor.emplace_back(in); | |
6545 | ||
f67539c2 | 6546 | if (abort || blocklisted) { |
11fdf7f2 TL |
6547 | objectcacher->purge_set(&in->oset); |
6548 | } else if (!in->caps.empty()) { | |
7c673cae FG |
6549 | _release(in); |
6550 | _flush(in, new C_Client_FlushComplete(this, in)); | |
6551 | } | |
6552 | } | |
6553 | } | |
6554 | ||
f67539c2 | 6555 | if (abort || blocklisted) { |
20effc67 TL |
6556 | for (auto &q : mds_sessions) { |
6557 | auto s = q.second; | |
6558 | for (auto p = s->dirty_list.begin(); !p.end(); ) { | |
6559 | Inode *in = *p; | |
6560 | ++p; | |
6561 | if (in->dirty_caps) { | |
6562 | ldout(cct, 0) << " drop dirty caps on " << *in << dendl; | |
6563 | in->mark_caps_clean(); | |
6564 | put_inode(in); | |
6565 | } | |
11fdf7f2 TL |
6566 | } |
6567 | } | |
6568 | } else { | |
6569 | flush_caps_sync(); | |
6570 | wait_sync_caps(last_flush_tid); | |
6571 | } | |
7c673cae FG |
6572 | |
6573 | // empty lru cache | |
7c673cae FG |
6574 | trim_cache(); |
6575 | ||
f67539c2 TL |
6576 | delay_put_inodes(); |
6577 | ||
7c673cae FG |
6578 | while (lru.lru_get_size() > 0 || |
6579 | !inode_map.empty()) { | |
6580 | ldout(cct, 2) << "cache still has " << lru.lru_get_size() | |
6581 | << "+" << inode_map.size() << " items" | |
6582 | << ", waiting (for caps to release?)" | |
6583 | << dendl; | |
f67539c2 | 6584 | |
9f95a23c TL |
6585 | if (auto r = mount_cond.wait_for(lock, ceph::make_timespan(5)); |
6586 | r == std::cv_status::timeout) { | |
7c673cae FG |
6587 | dump_cache(NULL); |
6588 | } | |
6589 | } | |
11fdf7f2 TL |
6590 | ceph_assert(lru.lru_get_size() == 0); |
6591 | ceph_assert(inode_map.empty()); | |
7c673cae FG |
6592 | |
6593 | // stop tracing | |
6594 | if (!cct->_conf->client_trace.empty()) { | |
6595 | ldout(cct, 1) << "closing trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6596 | traceout.close(); | |
6597 | } | |
6598 | ||
f67539c2 TL |
6599 | // stop the tick thread |
6600 | tick_thread_stopped = true; | |
6601 | upkeep_cond.notify_one(); | |
6602 | ||
7c673cae FG |
6603 | _close_sessions(); |
6604 | ||
2a845540 TL |
6605 | // release the global snapshot realm |
6606 | SnapRealm *global_realm = snap_realms[CEPH_INO_GLOBAL_SNAPREALM]; | |
6607 | if (global_realm) { | |
6608 | ceph_assert(global_realm->nref == 1); | |
6609 | put_snap_realm(global_realm); | |
6610 | } | |
6611 | ||
f67539c2 | 6612 | mref_writer.update_state(CLIENT_UNMOUNTED); |
7c673cae FG |
6613 | |
6614 | ldout(cct, 2) << "unmounted." << dendl; | |
6615 | } | |
6616 | ||
b32b8144 FG |
6617 | void Client::unmount() |
6618 | { | |
11fdf7f2 TL |
6619 | _unmount(false); |
6620 | } | |
6621 | ||
6622 | void Client::abort_conn() | |
6623 | { | |
11fdf7f2 | 6624 | _unmount(true); |
b32b8144 FG |
6625 | } |
6626 | ||
7c673cae FG |
6627 | void Client::flush_cap_releases() |
6628 | { | |
f67539c2 TL |
6629 | uint64_t nr_caps = 0; |
6630 | ||
7c673cae | 6631 | // send any cap releases |
11fdf7f2 | 6632 | for (auto &p : mds_sessions) { |
20effc67 TL |
6633 | auto session = p.second; |
6634 | if (session->release && mdsmap->is_clientreplay_or_active_or_stopping( | |
11fdf7f2 | 6635 | p.first)) { |
20effc67 | 6636 | nr_caps += session->release->caps.size(); |
7c673cae FG |
6637 | if (cct->_conf->client_inject_release_failure) { |
6638 | ldout(cct, 20) << __func__ << " injecting failure to send cap release message" << dendl; | |
7c673cae | 6639 | } else { |
20effc67 | 6640 | session->con->send_message2(std::move(session->release)); |
7c673cae | 6641 | } |
20effc67 | 6642 | session->release.reset(); |
7c673cae FG |
6643 | } |
6644 | } | |
f67539c2 TL |
6645 | |
6646 | if (nr_caps > 0) { | |
6647 | dec_pinned_icaps(nr_caps); | |
6648 | } | |
7c673cae FG |
6649 | } |
6650 | ||
f67539c2 | 6651 | void Client::renew_and_flush_cap_releases() |
7c673cae | 6652 | { |
f67539c2 TL |
6653 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
6654 | ||
6655 | if (!mount_aborted && mdsmap->get_epoch()) { | |
6656 | // renew caps? | |
2a845540 TL |
6657 | auto el = ceph::coarse_mono_clock::now() - last_cap_renew; |
6658 | if (unlikely(utime_t(el) > mdsmap->get_session_timeout() / 3.0)) | |
f67539c2 TL |
6659 | renew_caps(); |
6660 | ||
6661 | flush_cap_releases(); | |
7c673cae | 6662 | } |
f67539c2 TL |
6663 | } |
6664 | ||
6665 | void Client::tick() | |
6666 | { | |
6667 | ldout(cct, 20) << "tick" << dendl; | |
7c673cae | 6668 | |
2a845540 | 6669 | auto now = ceph::coarse_mono_clock::now(); |
7c673cae | 6670 | |
f67539c2 TL |
6671 | /* |
6672 | * If the mount() is not finished | |
6673 | */ | |
6674 | if (is_mounting() && !mds_requests.empty()) { | |
7c673cae | 6675 | MetaRequest *req = mds_requests.begin()->second; |
f67539c2 | 6676 | |
2a845540 | 6677 | if (req->created + mount_timeout < now) { |
f67539c2 | 6678 | req->abort(-CEPHFS_ETIMEDOUT); |
7c673cae | 6679 | if (req->caller_cond) { |
f67539c2 TL |
6680 | req->kick = true; |
6681 | req->caller_cond->notify_all(); | |
7c673cae FG |
6682 | } |
6683 | signal_cond_list(waiting_for_mdsmap); | |
11fdf7f2 | 6684 | for (auto &p : mds_sessions) { |
20effc67 | 6685 | signal_context_list(p.second->waiting_for_open); |
11fdf7f2 | 6686 | } |
7c673cae FG |
6687 | } |
6688 | } | |
6689 | ||
f67539c2 | 6690 | renew_and_flush_cap_releases(); |
7c673cae FG |
6691 | |
6692 | // delayed caps | |
28e407b8 | 6693 | xlist<Inode*>::iterator p = delayed_list.begin(); |
7c673cae FG |
6694 | while (!p.end()) { |
6695 | Inode *in = *p; | |
6696 | ++p; | |
f67539c2 | 6697 | if (!mount_aborted && in->hold_caps_until > now) |
7c673cae | 6698 | break; |
28e407b8 | 6699 | delayed_list.pop_front(); |
f67539c2 TL |
6700 | if (!mount_aborted) |
6701 | check_caps(in, CHECK_CAPS_NODELAY); | |
7c673cae FG |
6702 | } |
6703 | ||
f67539c2 TL |
6704 | if (!mount_aborted) |
6705 | collect_and_send_metrics(); | |
6706 | ||
6707 | delay_put_inodes(is_unmounting()); | |
7c673cae | 6708 | trim_cache(true); |
f6b5b4d7 | 6709 | |
f67539c2 | 6710 | if (blocklisted && (is_mounted() || is_unmounting()) && |
2a845540 | 6711 | last_auto_reconnect + std::chrono::seconds(30 * 60) < now && |
f6b5b4d7 TL |
6712 | cct->_conf.get_val<bool>("client_reconnect_stale")) { |
6713 | messenger->client_reset(); | |
6714 | fd_gen++; // invalidate open files | |
f67539c2 | 6715 | blocklisted = false; |
f6b5b4d7 TL |
6716 | _kick_stale_sessions(); |
6717 | last_auto_reconnect = now; | |
6718 | } | |
7c673cae FG |
6719 | } |
6720 | ||
f67539c2 TL |
6721 | void Client::start_tick_thread() |
6722 | { | |
6723 | upkeeper = std::thread([this]() { | |
6724 | using time = ceph::coarse_mono_time; | |
6725 | using sec = std::chrono::seconds; | |
6726 | ||
6727 | auto last_tick = time::min(); | |
6728 | ||
6729 | std::unique_lock cl(client_lock); | |
6730 | while (!tick_thread_stopped) { | |
6731 | auto now = clock::now(); | |
6732 | auto since = now - last_tick; | |
6733 | ||
6734 | auto t_interval = clock::duration(cct->_conf.get_val<sec>("client_tick_interval")); | |
6735 | auto d_interval = clock::duration(cct->_conf.get_val<sec>("client_debug_inject_tick_delay")); | |
6736 | ||
6737 | auto interval = std::max(t_interval, d_interval); | |
6738 | if (likely(since >= interval*.90)) { | |
6739 | tick(); | |
6740 | last_tick = clock::now(); | |
6741 | } else { | |
6742 | interval -= since; | |
6743 | } | |
6744 | ||
6745 | ldout(cct, 20) << "upkeep thread waiting interval " << interval << dendl; | |
6746 | if (!tick_thread_stopped) | |
6747 | upkeep_cond.wait_for(cl, interval); | |
6748 | } | |
6749 | }); | |
6750 | } | |
6751 | ||
6752 | void Client::collect_and_send_metrics() { | |
6753 | ldout(cct, 20) << __func__ << dendl; | |
6754 | ||
6755 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6756 | ||
6757 | // right now, we only track and send global metrics. its sufficient | |
6758 | // to send these metrics to MDS rank0. | |
6759 | collect_and_send_global_metrics(); | |
6760 | } | |
6761 | ||
6762 | void Client::collect_and_send_global_metrics() { | |
6763 | ldout(cct, 20) << __func__ << dendl; | |
6764 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6765 | ||
6766 | if (!have_open_session((mds_rank_t)0)) { | |
6767 | ldout(cct, 5) << __func__ << ": no session with rank=0 -- not sending metric" | |
6768 | << dendl; | |
6769 | return; | |
6770 | } | |
6771 | auto session = _get_or_open_mds_session((mds_rank_t)0); | |
6772 | if (!session->mds_features.test(CEPHFS_FEATURE_METRIC_COLLECT)) { | |
6773 | ldout(cct, 5) << __func__ << ": rank=0 does not support metrics" << dendl; | |
6774 | return; | |
6775 | } | |
6776 | ||
6777 | ClientMetricMessage metric; | |
6778 | std::vector<ClientMetricMessage> message; | |
6779 | ||
6780 | // read latency | |
33c7a0ef TL |
6781 | if (_collect_and_send_global_metrics || |
6782 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_READ_LATENCY)) { | |
2a845540 TL |
6783 | metric = ClientMetricMessage(ReadLatencyPayload(logger->tget(l_c_read), |
6784 | logger->tget(l_c_rd_avg), | |
6785 | logger->get(l_c_rd_sqsum), | |
6786 | nr_read_request)); | |
33c7a0ef TL |
6787 | message.push_back(metric); |
6788 | } | |
f67539c2 TL |
6789 | |
6790 | // write latency | |
33c7a0ef TL |
6791 | if (_collect_and_send_global_metrics || |
6792 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_WRITE_LATENCY)) { | |
2a845540 TL |
6793 | metric = ClientMetricMessage(WriteLatencyPayload(logger->tget(l_c_wrlat), |
6794 | logger->tget(l_c_wr_avg), | |
6795 | logger->get(l_c_wr_sqsum), | |
6796 | nr_write_request)); | |
33c7a0ef TL |
6797 | message.push_back(metric); |
6798 | } | |
f67539c2 TL |
6799 | |
6800 | // metadata latency | |
33c7a0ef TL |
6801 | if (_collect_and_send_global_metrics || |
6802 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_METADATA_LATENCY)) { | |
2a845540 TL |
6803 | metric = ClientMetricMessage(MetadataLatencyPayload(logger->tget(l_c_lat), |
6804 | logger->tget(l_c_md_avg), | |
6805 | logger->get(l_c_md_sqsum), | |
6806 | nr_metadata_request)); | |
33c7a0ef TL |
6807 | message.push_back(metric); |
6808 | } | |
f67539c2 TL |
6809 | |
6810 | // cap hit ratio -- nr_caps is unused right now | |
33c7a0ef TL |
6811 | if (_collect_and_send_global_metrics || |
6812 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_CAP_INFO)) { | |
6813 | auto [cap_hits, cap_misses] = get_cap_hit_rates(); | |
6814 | metric = ClientMetricMessage(CapInfoPayload(cap_hits, cap_misses, 0)); | |
6815 | message.push_back(metric); | |
6816 | } | |
f67539c2 TL |
6817 | |
6818 | // dentry lease hit ratio | |
33c7a0ef TL |
6819 | if (_collect_and_send_global_metrics || |
6820 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_DENTRY_LEASE)) { | |
6821 | auto [dlease_hits, dlease_misses, nr] = get_dlease_hit_rates(); | |
6822 | metric = ClientMetricMessage(DentryLeasePayload(dlease_hits, dlease_misses, nr)); | |
6823 | message.push_back(metric); | |
6824 | } | |
f67539c2 TL |
6825 | |
6826 | // opened files | |
33c7a0ef TL |
6827 | if (_collect_and_send_global_metrics || |
6828 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_OPENED_FILES)) { | |
f67539c2 TL |
6829 | auto [opened_files, total_inodes] = get_opened_files_rates(); |
6830 | metric = ClientMetricMessage(OpenedFilesPayload(opened_files, total_inodes)); | |
33c7a0ef | 6831 | message.push_back(metric); |
f67539c2 | 6832 | } |
f67539c2 TL |
6833 | |
6834 | // pinned i_caps | |
33c7a0ef TL |
6835 | if (_collect_and_send_global_metrics || |
6836 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_PINNED_ICAPS)) { | |
f67539c2 TL |
6837 | auto [pinned_icaps, total_inodes] = get_pinned_icaps_rates(); |
6838 | metric = ClientMetricMessage(PinnedIcapsPayload(pinned_icaps, total_inodes)); | |
33c7a0ef | 6839 | message.push_back(metric); |
f67539c2 | 6840 | } |
f67539c2 TL |
6841 | |
6842 | // opened inodes | |
33c7a0ef TL |
6843 | if (_collect_and_send_global_metrics || |
6844 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_OPENED_INODES)) { | |
f67539c2 TL |
6845 | auto [opened_inodes, total_inodes] = get_opened_inodes_rates(); |
6846 | metric = ClientMetricMessage(OpenedInodesPayload(opened_inodes, total_inodes)); | |
33c7a0ef | 6847 | message.push_back(metric); |
f67539c2 | 6848 | } |
f67539c2 | 6849 | |
a4b75251 | 6850 | // read io sizes |
33c7a0ef TL |
6851 | if (_collect_and_send_global_metrics || |
6852 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_READ_IO_SIZES)) { | |
6853 | metric = ClientMetricMessage(ReadIoSizesPayload(total_read_ops, | |
6854 | total_read_size)); | |
6855 | message.push_back(metric); | |
6856 | } | |
a4b75251 TL |
6857 | |
6858 | // write io sizes | |
33c7a0ef TL |
6859 | if (_collect_and_send_global_metrics || |
6860 | session->mds_metric_flags.test(CLIENT_METRIC_TYPE_WRITE_IO_SIZES)) { | |
6861 | metric = ClientMetricMessage(WriteIoSizesPayload(total_write_ops, | |
6862 | total_write_size)); | |
6863 | message.push_back(metric); | |
6864 | } | |
a4b75251 | 6865 | |
f67539c2 TL |
6866 | session->con->send_message2(make_message<MClientMetrics>(std::move(message))); |
6867 | } | |
6868 | ||
7c673cae FG |
6869 | void Client::renew_caps() |
6870 | { | |
6871 | ldout(cct, 10) << "renew_caps()" << dendl; | |
2a845540 | 6872 | last_cap_renew = ceph::coarse_mono_clock::now(); |
7c673cae | 6873 | |
11fdf7f2 TL |
6874 | for (auto &p : mds_sessions) { |
6875 | ldout(cct, 15) << "renew_caps requesting from mds." << p.first << dendl; | |
6876 | if (mdsmap->get_state(p.first) >= MDSMap::STATE_REJOIN) | |
20effc67 | 6877 | renew_caps(p.second.get()); |
7c673cae FG |
6878 | } |
6879 | } | |
6880 | ||
6881 | void Client::renew_caps(MetaSession *session) | |
6882 | { | |
6883 | ldout(cct, 10) << "renew_caps mds." << session->mds_num << dendl; | |
6884 | session->last_cap_renew_request = ceph_clock_now(); | |
6885 | uint64_t seq = ++session->cap_renew_seq; | |
9f95a23c | 6886 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_RENEWCAPS, seq)); |
7c673cae FG |
6887 | } |
6888 | ||
6889 | ||
6890 | // =============================================================== | |
6891 | // high level (POSIXy) interface | |
6892 | ||
6893 | int Client::_do_lookup(Inode *dir, const string& name, int mask, | |
6894 | InodeRef *target, const UserPerm& perms) | |
6895 | { | |
6896 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
6897 | MetaRequest *req = new MetaRequest(op); | |
6898 | filepath path; | |
6899 | dir->make_nosnap_relative_path(path); | |
6900 | path.push_dentry(name); | |
6901 | req->set_filepath(path); | |
6902 | req->set_inode(dir); | |
6903 | if (cct->_conf->client_debug_getattr_caps && op == CEPH_MDS_OP_LOOKUP) | |
6904 | mask |= DEBUG_GETATTR_CAPS; | |
6905 | req->head.args.getattr.mask = mask; | |
6906 | ||
11fdf7f2 | 6907 | ldout(cct, 10) << __func__ << " on " << path << dendl; |
7c673cae FG |
6908 | |
6909 | int r = make_request(req, perms, target); | |
11fdf7f2 | 6910 | ldout(cct, 10) << __func__ << " res is " << r << dendl; |
7c673cae FG |
6911 | return r; |
6912 | } | |
6913 | ||
f67539c2 TL |
6914 | bool Client::_dentry_valid(const Dentry *dn) |
6915 | { | |
6916 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6917 | ||
6918 | // is dn lease valid? | |
6919 | utime_t now = ceph_clock_now(); | |
6920 | if (dn->lease_mds >= 0 && dn->lease_ttl > now && | |
6921 | mds_sessions.count(dn->lease_mds)) { | |
20effc67 TL |
6922 | auto s = mds_sessions.at(dn->lease_mds); |
6923 | if (s->cap_ttl > now && s->cap_gen == dn->lease_gen) { | |
f67539c2 TL |
6924 | dlease_hit(); |
6925 | return true; | |
6926 | } | |
6927 | ||
20effc67 | 6928 | ldout(cct, 20) << " bad lease, cap_ttl " << s->cap_ttl << ", cap_gen " << s->cap_gen |
f67539c2 TL |
6929 | << " vs lease_gen " << dn->lease_gen << dendl; |
6930 | } | |
6931 | ||
6932 | dlease_miss(); | |
6933 | return false; | |
6934 | } | |
6935 | ||
7c673cae | 6936 | int Client::_lookup(Inode *dir, const string& dname, int mask, InodeRef *target, |
f67539c2 | 6937 | const UserPerm& perms, std::string* alternate_name) |
7c673cae FG |
6938 | { |
6939 | int r = 0; | |
6940 | Dentry *dn = NULL; | |
f67539c2 | 6941 | bool did_lookup_request = false; |
f91f0fd5 TL |
6942 | // can only request shared caps |
6943 | mask &= CEPH_CAP_ANY_SHARED | CEPH_STAT_RSTAT; | |
7c673cae | 6944 | |
7c673cae | 6945 | if (dname == "..") { |
11fdf7f2 TL |
6946 | if (dir->dentries.empty()) { |
6947 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); | |
6948 | filepath path(dir->ino); | |
6949 | req->set_filepath(path); | |
6950 | ||
6951 | InodeRef tmptarget; | |
6952 | int r = make_request(req, perms, &tmptarget, NULL, rand() % mdsmap->get_num_in_mds()); | |
6953 | ||
6954 | if (r == 0) { | |
f91f0fd5 | 6955 | *target = std::move(tmptarget); |
11fdf7f2 TL |
6956 | ldout(cct, 8) << __func__ << " found target " << (*target)->ino << dendl; |
6957 | } else { | |
6958 | *target = dir; | |
6959 | } | |
6960 | } | |
7c673cae FG |
6961 | else |
6962 | *target = dir->get_first_parent()->dir->parent_inode; //dirs can't be hard-linked | |
6963 | goto done; | |
6964 | } | |
6965 | ||
6966 | if (dname == ".") { | |
6967 | *target = dir; | |
6968 | goto done; | |
6969 | } | |
6970 | ||
11fdf7f2 | 6971 | if (!dir->is_dir()) { |
f67539c2 | 6972 | r = -CEPHFS_ENOTDIR; |
11fdf7f2 TL |
6973 | goto done; |
6974 | } | |
6975 | ||
7c673cae | 6976 | if (dname.length() > NAME_MAX) { |
f67539c2 | 6977 | r = -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
6978 | goto done; |
6979 | } | |
6980 | ||
6981 | if (dname == cct->_conf->client_snapdir && | |
6982 | dir->snapid == CEPH_NOSNAP) { | |
6983 | *target = open_snapdir(dir); | |
6984 | goto done; | |
6985 | } | |
6986 | ||
f67539c2 | 6987 | relookup: |
7c673cae FG |
6988 | if (dir->dir && |
6989 | dir->dir->dentries.count(dname)) { | |
6990 | dn = dir->dir->dentries[dname]; | |
6991 | ||
f67539c2 TL |
6992 | ldout(cct, 20) << __func__ << " have " << *dn << " from mds." << dn->lease_mds |
6993 | << " ttl " << dn->lease_ttl << " seq " << dn->lease_seq << dendl; | |
7c673cae | 6994 | |
94b18763 | 6995 | if (!dn->inode || dn->inode->caps_issued_mask(mask, true)) { |
f67539c2 TL |
6996 | if (_dentry_valid(dn)) { |
6997 | // touch this mds's dir cap too, even though we don't _explicitly_ use it here, to | |
6998 | // make trim_caps() behave. | |
6999 | dir->try_touch_cap(dn->lease_mds); | |
7000 | goto hit_dn; | |
7c673cae | 7001 | } |
92f5a8d4 | 7002 | // dir shared caps? |
94b18763 | 7003 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 7004 | if (dn->cap_shared_gen == dir->shared_gen && |
94b18763 | 7005 | (!dn->inode || dn->inode->caps_issued_mask(mask, true))) |
7c673cae FG |
7006 | goto hit_dn; |
7007 | if (!dn->inode && (dir->flags & I_COMPLETE)) { | |
11fdf7f2 | 7008 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " |
7c673cae | 7009 | << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 7010 | return -CEPHFS_ENOENT; |
7c673cae FG |
7011 | } |
7012 | } | |
7013 | } else { | |
7014 | ldout(cct, 20) << " no cap on " << dn->inode->vino() << dendl; | |
7015 | } | |
7016 | } else { | |
7017 | // can we conclude ENOENT locally? | |
94b18763 | 7018 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true) && |
7c673cae | 7019 | (dir->flags & I_COMPLETE)) { |
11fdf7f2 | 7020 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 7021 | return -CEPHFS_ENOENT; |
7c673cae FG |
7022 | } |
7023 | } | |
7024 | ||
f67539c2 TL |
7025 | if (did_lookup_request) { |
7026 | r = 0; | |
7027 | goto done; | |
7028 | } | |
7c673cae | 7029 | r = _do_lookup(dir, dname, mask, target, perms); |
f67539c2 TL |
7030 | did_lookup_request = true; |
7031 | if (r == 0) { | |
7032 | /* complete lookup to get dentry for alternate_name */ | |
7033 | goto relookup; | |
7034 | } else { | |
7035 | goto done; | |
7036 | } | |
7037 | ||
7038 | hit_dn: | |
7039 | if (dn->inode) { | |
7c673cae | 7040 | *target = dn->inode; |
f67539c2 TL |
7041 | if (alternate_name) |
7042 | *alternate_name = dn->alternate_name; | |
7c673cae | 7043 | } else { |
f67539c2 | 7044 | r = -CEPHFS_ENOENT; |
7c673cae FG |
7045 | } |
7046 | touch_dn(dn); | |
f67539c2 | 7047 | goto done; |
7c673cae FG |
7048 | |
7049 | done: | |
7050 | if (r < 0) | |
11fdf7f2 | 7051 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << r << dendl; |
7c673cae | 7052 | else |
11fdf7f2 | 7053 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << **target << dendl; |
7c673cae FG |
7054 | return r; |
7055 | } | |
7056 | ||
7057 | int Client::get_or_create(Inode *dir, const char* name, | |
7058 | Dentry **pdn, bool expect_null) | |
7059 | { | |
7060 | // lookup | |
11fdf7f2 | 7061 | ldout(cct, 20) << __func__ << " " << *dir << " name " << name << dendl; |
7c673cae FG |
7062 | dir->open_dir(); |
7063 | if (dir->dir->dentries.count(name)) { | |
7064 | Dentry *dn = dir->dir->dentries[name]; | |
f67539c2 TL |
7065 | if (_dentry_valid(dn)) { |
7066 | if (expect_null) | |
7067 | return -CEPHFS_EEXIST; | |
7c673cae FG |
7068 | } |
7069 | *pdn = dn; | |
7070 | } else { | |
7071 | // otherwise link up a new one | |
7072 | *pdn = link(dir->dir, name, NULL, NULL); | |
7073 | } | |
7074 | ||
7075 | // success | |
7076 | return 0; | |
7077 | } | |
7078 | ||
f67539c2 TL |
7079 | int Client::walk(std::string_view path, walk_dentry_result* wdr, const UserPerm& perms, bool followsym) |
7080 | { | |
7081 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
7082 | if (!mref_reader.is_state_satisfied()) | |
7083 | return -CEPHFS_ENOTCONN; | |
7084 | ||
7085 | ldout(cct, 10) << __func__ << ": " << path << dendl; | |
7086 | ||
7087 | std::scoped_lock lock(client_lock); | |
7088 | ||
7089 | return path_walk(path, wdr, perms, followsym); | |
7090 | } | |
7091 | ||
7c673cae | 7092 | int Client::path_walk(const filepath& origpath, InodeRef *end, |
b3b6e05e | 7093 | const UserPerm& perms, bool followsym, int mask, InodeRef dirinode) |
f67539c2 TL |
7094 | { |
7095 | walk_dentry_result wdr; | |
b3b6e05e | 7096 | int rc = path_walk(origpath, &wdr, perms, followsym, mask, dirinode); |
f67539c2 TL |
7097 | *end = std::move(wdr.in); |
7098 | return rc; | |
7099 | } | |
7100 | ||
b3b6e05e TL |
7101 | int Client::path_walk(const filepath& origpath, walk_dentry_result* result, const UserPerm& perms, |
7102 | bool followsym, int mask, InodeRef dirinode) | |
7c673cae FG |
7103 | { |
7104 | filepath path = origpath; | |
7105 | InodeRef cur; | |
f67539c2 | 7106 | std::string alternate_name; |
7c673cae FG |
7107 | if (origpath.absolute()) |
7108 | cur = root; | |
b3b6e05e | 7109 | else if (!dirinode) |
7c673cae | 7110 | cur = cwd; |
b3b6e05e TL |
7111 | else { |
7112 | cur = dirinode; | |
7113 | } | |
11fdf7f2 | 7114 | ceph_assert(cur); |
7c673cae | 7115 | |
b3b6e05e | 7116 | ldout(cct, 20) << __func__ << " cur=" << *cur << dendl; |
11fdf7f2 | 7117 | ldout(cct, 10) << __func__ << " " << path << dendl; |
7c673cae FG |
7118 | |
7119 | int symlinks = 0; | |
7120 | ||
7121 | unsigned i=0; | |
7122 | while (i < path.depth() && cur) { | |
7123 | int caps = 0; | |
7124 | const string &dname = path[i]; | |
7125 | ldout(cct, 10) << " " << i << " " << *cur << " " << dname << dendl; | |
7126 | ldout(cct, 20) << " (path is " << path << ")" << dendl; | |
7127 | InodeRef next; | |
7128 | if (cct->_conf->client_permissions) { | |
7129 | int r = may_lookup(cur.get(), perms); | |
7130 | if (r < 0) | |
7131 | return r; | |
7132 | caps = CEPH_CAP_AUTH_SHARED; | |
7133 | } | |
7134 | ||
7135 | /* Get extra requested caps on the last component */ | |
7136 | if (i == (path.depth() - 1)) | |
7137 | caps |= mask; | |
f67539c2 | 7138 | int r = _lookup(cur.get(), dname, caps, &next, perms, &alternate_name); |
7c673cae FG |
7139 | if (r < 0) |
7140 | return r; | |
7141 | // only follow trailing symlink if followsym. always follow | |
7142 | // 'directory' symlinks. | |
7143 | if (next && next->is_symlink()) { | |
7144 | symlinks++; | |
7145 | ldout(cct, 20) << " symlink count " << symlinks << ", value is '" << next->symlink << "'" << dendl; | |
7146 | if (symlinks > MAXSYMLINKS) { | |
f67539c2 | 7147 | return -CEPHFS_ELOOP; |
7c673cae FG |
7148 | } |
7149 | ||
7150 | if (i < path.depth() - 1) { | |
7151 | // dir symlink | |
7152 | // replace consumed components of path with symlink dir target | |
7153 | filepath resolved(next->symlink.c_str()); | |
7154 | resolved.append(path.postfixpath(i + 1)); | |
7155 | path = resolved; | |
7156 | i = 0; | |
7157 | if (next->symlink[0] == '/') { | |
7158 | cur = root; | |
7159 | } | |
7160 | continue; | |
7161 | } else if (followsym) { | |
7162 | if (next->symlink[0] == '/') { | |
7163 | path = next->symlink.c_str(); | |
7164 | i = 0; | |
7165 | // reset position | |
7166 | cur = root; | |
7167 | } else { | |
7168 | filepath more(next->symlink.c_str()); | |
7169 | // we need to remove the symlink component from off of the path | |
7170 | // before adding the target that the symlink points to. remain | |
7171 | // at the same position in the path. | |
7172 | path.pop_dentry(); | |
7173 | path.append(more); | |
7174 | } | |
7175 | continue; | |
7176 | } | |
7177 | } | |
7178 | cur.swap(next); | |
7179 | i++; | |
7180 | } | |
7181 | if (!cur) | |
f67539c2 TL |
7182 | return -CEPHFS_ENOENT; |
7183 | if (result) { | |
7184 | result->in = std::move(cur); | |
7185 | result->alternate_name = std::move(alternate_name); | |
7186 | } | |
7c673cae FG |
7187 | return 0; |
7188 | } | |
7189 | ||
7190 | ||
7191 | // namespace ops | |
7192 | ||
f67539c2 | 7193 | int Client::link(const char *relexisting, const char *relpath, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7194 | { |
f67539c2 TL |
7195 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7196 | if (!mref_reader.is_state_satisfied()) | |
7197 | return -CEPHFS_ENOTCONN; | |
7198 | ||
7c673cae FG |
7199 | tout(cct) << "link" << std::endl; |
7200 | tout(cct) << relexisting << std::endl; | |
7201 | tout(cct) << relpath << std::endl; | |
7202 | ||
7203 | filepath existing(relexisting); | |
7204 | ||
7205 | InodeRef in, dir; | |
f67539c2 TL |
7206 | |
7207 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7208 | int r = path_walk(existing, &in, perm, true); |
7209 | if (r < 0) | |
7210 | return r; | |
7211 | if (std::string(relpath) == "/") { | |
f67539c2 | 7212 | r = -CEPHFS_EEXIST; |
7c673cae FG |
7213 | return r; |
7214 | } | |
7215 | filepath path(relpath); | |
7216 | string name = path.last_dentry(); | |
7217 | path.pop_dentry(); | |
7218 | ||
7219 | r = path_walk(path, &dir, perm, true); | |
7220 | if (r < 0) | |
7221 | return r; | |
7222 | if (cct->_conf->client_permissions) { | |
7223 | if (S_ISDIR(in->mode)) { | |
f67539c2 | 7224 | r = -CEPHFS_EPERM; |
7c673cae FG |
7225 | return r; |
7226 | } | |
7227 | r = may_hardlink(in.get(), perm); | |
7228 | if (r < 0) | |
7229 | return r; | |
7230 | r = may_create(dir.get(), perm); | |
7231 | if (r < 0) | |
7232 | return r; | |
7233 | } | |
f67539c2 | 7234 | r = _link(in.get(), dir.get(), name.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7235 | return r; |
7236 | } | |
7237 | ||
7238 | int Client::unlink(const char *relpath, const UserPerm& perm) | |
b3b6e05e TL |
7239 | { |
7240 | return unlinkat(CEPHFS_AT_FDCWD, relpath, 0, perm); | |
7241 | } | |
7242 | ||
7243 | int Client::unlinkat(int dirfd, const char *relpath, int flags, const UserPerm& perm) | |
7c673cae | 7244 | { |
f67539c2 | 7245 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7246 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7247 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7248 | } |
f67539c2 | 7249 | |
11fdf7f2 | 7250 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7251 | tout(cct) << dirfd << std::endl; |
7c673cae | 7252 | tout(cct) << relpath << std::endl; |
b3b6e05e | 7253 | tout(cct) << flags << std::endl; |
7c673cae | 7254 | |
b3b6e05e TL |
7255 | if (std::string(relpath) == "/") { |
7256 | return flags & AT_REMOVEDIR ? -CEPHFS_EBUSY : -CEPHFS_EISDIR; | |
7257 | } | |
7c673cae FG |
7258 | |
7259 | filepath path(relpath); | |
7260 | string name = path.last_dentry(); | |
7261 | path.pop_dentry(); | |
7262 | InodeRef dir; | |
f67539c2 TL |
7263 | |
7264 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7265 | |
7266 | InodeRef dirinode; | |
7267 | int r = get_fd_inode(dirfd, &dirinode); | |
7268 | if (r < 0) { | |
7269 | return r; | |
7270 | } | |
7271 | ||
7272 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7273 | if (r < 0) { | |
7c673cae | 7274 | return r; |
b3b6e05e | 7275 | } |
7c673cae FG |
7276 | if (cct->_conf->client_permissions) { |
7277 | r = may_delete(dir.get(), name.c_str(), perm); | |
b3b6e05e | 7278 | if (r < 0) { |
7c673cae | 7279 | return r; |
b3b6e05e | 7280 | } |
7c673cae | 7281 | } |
b3b6e05e TL |
7282 | if (flags & AT_REMOVEDIR) { |
7283 | r = _rmdir(dir.get(), name.c_str(), perm); | |
7284 | } else { | |
7285 | r = _unlink(dir.get(), name.c_str(), perm); | |
7286 | } | |
7287 | return r; | |
7c673cae FG |
7288 | } |
7289 | ||
f67539c2 | 7290 | int Client::rename(const char *relfrom, const char *relto, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7291 | { |
f67539c2 TL |
7292 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7293 | if (!mref_reader.is_state_satisfied()) | |
7294 | return -CEPHFS_ENOTCONN; | |
7295 | ||
11fdf7f2 | 7296 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7297 | tout(cct) << relfrom << std::endl; |
7298 | tout(cct) << relto << std::endl; | |
7299 | ||
7300 | if (std::string(relfrom) == "/" || std::string(relto) == "/") | |
f67539c2 | 7301 | return -CEPHFS_EBUSY; |
7c673cae FG |
7302 | |
7303 | filepath from(relfrom); | |
7304 | filepath to(relto); | |
7305 | string fromname = from.last_dentry(); | |
7306 | from.pop_dentry(); | |
7307 | string toname = to.last_dentry(); | |
7308 | to.pop_dentry(); | |
7309 | ||
7310 | InodeRef fromdir, todir; | |
f67539c2 TL |
7311 | |
7312 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7313 | int r = path_walk(from, &fromdir, perm); |
7314 | if (r < 0) | |
7315 | goto out; | |
7316 | r = path_walk(to, &todir, perm); | |
7317 | if (r < 0) | |
7318 | goto out; | |
7319 | ||
7320 | if (cct->_conf->client_permissions) { | |
7321 | int r = may_delete(fromdir.get(), fromname.c_str(), perm); | |
7322 | if (r < 0) | |
7323 | return r; | |
7324 | r = may_delete(todir.get(), toname.c_str(), perm); | |
f67539c2 | 7325 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
7326 | return r; |
7327 | } | |
f67539c2 | 7328 | r = _rename(fromdir.get(), fromname.c_str(), todir.get(), toname.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7329 | out: |
7330 | return r; | |
7331 | } | |
7332 | ||
7333 | // dirs | |
7334 | ||
f67539c2 | 7335 | int Client::mkdir(const char *relpath, mode_t mode, const UserPerm& perm, std::string alternate_name) |
b3b6e05e TL |
7336 | { |
7337 | return mkdirat(CEPHFS_AT_FDCWD, relpath, mode, perm, alternate_name); | |
7338 | } | |
7339 | ||
7340 | int Client::mkdirat(int dirfd, const char *relpath, mode_t mode, const UserPerm& perm, | |
7341 | std::string alternate_name) | |
7c673cae | 7342 | { |
f67539c2 TL |
7343 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7344 | if (!mref_reader.is_state_satisfied()) | |
7345 | return -CEPHFS_ENOTCONN; | |
7346 | ||
11fdf7f2 | 7347 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7348 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7349 | tout(cct) << relpath << std::endl; |
7350 | tout(cct) << mode << std::endl; | |
11fdf7f2 | 7351 | ldout(cct, 10) << __func__ << ": " << relpath << dendl; |
7c673cae | 7352 | |
b3b6e05e | 7353 | if (std::string(relpath) == "/") { |
f67539c2 | 7354 | return -CEPHFS_EEXIST; |
b3b6e05e | 7355 | } |
7c673cae FG |
7356 | |
7357 | filepath path(relpath); | |
7358 | string name = path.last_dentry(); | |
7359 | path.pop_dentry(); | |
7360 | InodeRef dir; | |
f67539c2 TL |
7361 | |
7362 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7363 | |
7364 | InodeRef dirinode; | |
7365 | int r = get_fd_inode(dirfd, &dirinode); | |
7366 | if (r < 0) { | |
7c673cae | 7367 | return r; |
b3b6e05e TL |
7368 | } |
7369 | ||
7370 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7371 | if (r < 0) { | |
7372 | return r; | |
7373 | } | |
7c673cae FG |
7374 | if (cct->_conf->client_permissions) { |
7375 | r = may_create(dir.get(), perm); | |
b3b6e05e | 7376 | if (r < 0) { |
7c673cae | 7377 | return r; |
b3b6e05e | 7378 | } |
7c673cae | 7379 | } |
f67539c2 | 7380 | return _mkdir(dir.get(), name.c_str(), mode, perm, 0, {}, std::move(alternate_name)); |
7c673cae FG |
7381 | } |
7382 | ||
7383 | int Client::mkdirs(const char *relpath, mode_t mode, const UserPerm& perms) | |
7384 | { | |
f67539c2 TL |
7385 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7386 | if (!mref_reader.is_state_satisfied()) | |
7387 | return -CEPHFS_ENOTCONN; | |
7388 | ||
7c673cae | 7389 | ldout(cct, 10) << "Client::mkdirs " << relpath << dendl; |
11fdf7f2 | 7390 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7391 | tout(cct) << relpath << std::endl; |
7392 | tout(cct) << mode << std::endl; | |
7393 | ||
7394 | //get through existing parts of path | |
7395 | filepath path(relpath); | |
7396 | unsigned int i; | |
7397 | int r = 0, caps = 0; | |
7398 | InodeRef cur, next; | |
f67539c2 TL |
7399 | |
7400 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7401 | cur = cwd; |
7402 | for (i=0; i<path.depth(); ++i) { | |
7403 | if (cct->_conf->client_permissions) { | |
7404 | r = may_lookup(cur.get(), perms); | |
7405 | if (r < 0) | |
7406 | break; | |
7407 | caps = CEPH_CAP_AUTH_SHARED; | |
7408 | } | |
7409 | r = _lookup(cur.get(), path[i].c_str(), caps, &next, perms); | |
7410 | if (r < 0) | |
7411 | break; | |
7412 | cur.swap(next); | |
7413 | } | |
f67539c2 | 7414 | if (r!=-CEPHFS_ENOENT) return r; |
11fdf7f2 | 7415 | ldout(cct, 20) << __func__ << " got through " << i << " directories on path " << relpath << dendl; |
7c673cae FG |
7416 | //make new directory at each level |
7417 | for (; i<path.depth(); ++i) { | |
7418 | if (cct->_conf->client_permissions) { | |
7419 | r = may_create(cur.get(), perms); | |
7420 | if (r < 0) | |
7421 | return r; | |
7422 | } | |
7423 | //make new dir | |
7424 | r = _mkdir(cur.get(), path[i].c_str(), mode, perms, &next); | |
c07f9fc5 | 7425 | |
7c673cae | 7426 | //check proper creation/existence |
f67539c2 | 7427 | if(-CEPHFS_EEXIST == r && i < path.depth() - 1) { |
c07f9fc5 FG |
7428 | r = _lookup(cur.get(), path[i].c_str(), CEPH_CAP_AUTH_SHARED, &next, perms); |
7429 | } | |
7430 | if (r < 0) | |
7431 | return r; | |
7c673cae FG |
7432 | //move to new dir and continue |
7433 | cur.swap(next); | |
11fdf7f2 | 7434 | ldout(cct, 20) << __func__ << ": successfully created directory " |
7c673cae FG |
7435 | << filepath(cur->ino).get_path() << dendl; |
7436 | } | |
7437 | return 0; | |
7438 | } | |
7439 | ||
7440 | int Client::rmdir(const char *relpath, const UserPerm& perms) | |
7441 | { | |
b3b6e05e | 7442 | return unlinkat(CEPHFS_AT_FDCWD, relpath, AT_REMOVEDIR, perms); |
7c673cae FG |
7443 | } |
7444 | ||
7445 | int Client::mknod(const char *relpath, mode_t mode, const UserPerm& perms, dev_t rdev) | |
f67539c2 TL |
7446 | { |
7447 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
7448 | if (!mref_reader.is_state_satisfied()) | |
7449 | return -CEPHFS_ENOTCONN; | |
7450 | ||
11fdf7f2 | 7451 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7452 | tout(cct) << relpath << std::endl; |
7453 | tout(cct) << mode << std::endl; | |
7454 | tout(cct) << rdev << std::endl; | |
7455 | ||
7456 | if (std::string(relpath) == "/") | |
f67539c2 | 7457 | return -CEPHFS_EEXIST; |
7c673cae FG |
7458 | |
7459 | filepath path(relpath); | |
7460 | string name = path.last_dentry(); | |
7461 | path.pop_dentry(); | |
7462 | InodeRef dir; | |
f67539c2 TL |
7463 | |
7464 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7465 | int r = path_walk(path, &dir, perms); |
7466 | if (r < 0) | |
7467 | return r; | |
7468 | if (cct->_conf->client_permissions) { | |
7469 | int r = may_create(dir.get(), perms); | |
7470 | if (r < 0) | |
7471 | return r; | |
7472 | } | |
7473 | return _mknod(dir.get(), name.c_str(), mode, rdev, perms); | |
7474 | } | |
7475 | ||
7476 | // symlinks | |
7477 | ||
f67539c2 | 7478 | int Client::symlink(const char *target, const char *relpath, const UserPerm& perms, std::string alternate_name) |
b3b6e05e TL |
7479 | { |
7480 | return symlinkat(target, CEPHFS_AT_FDCWD, relpath, perms, alternate_name); | |
7481 | } | |
7482 | ||
7483 | int Client::symlinkat(const char *target, int dirfd, const char *relpath, const UserPerm& perms, | |
7484 | std::string alternate_name) | |
7c673cae | 7485 | { |
f67539c2 | 7486 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7487 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7488 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7489 | } |
f67539c2 | 7490 | |
11fdf7f2 | 7491 | tout(cct) << __func__ << std::endl; |
7c673cae | 7492 | tout(cct) << target << std::endl; |
b3b6e05e | 7493 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7494 | tout(cct) << relpath << std::endl; |
7495 | ||
b3b6e05e | 7496 | if (std::string(relpath) == "/") { |
f67539c2 | 7497 | return -CEPHFS_EEXIST; |
b3b6e05e | 7498 | } |
7c673cae FG |
7499 | |
7500 | filepath path(relpath); | |
7501 | string name = path.last_dentry(); | |
7502 | path.pop_dentry(); | |
7503 | InodeRef dir; | |
f67539c2 TL |
7504 | |
7505 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7506 | |
7507 | InodeRef dirinode; | |
7508 | int r = get_fd_inode(dirfd, &dirinode); | |
7509 | if (r < 0) { | |
7c673cae | 7510 | return r; |
b3b6e05e TL |
7511 | } |
7512 | r = path_walk(path, &dir, perms, true, 0, dirinode); | |
7513 | if (r < 0) { | |
7514 | return r; | |
7515 | } | |
7c673cae FG |
7516 | if (cct->_conf->client_permissions) { |
7517 | int r = may_create(dir.get(), perms); | |
b3b6e05e | 7518 | if (r < 0) { |
7c673cae | 7519 | return r; |
b3b6e05e | 7520 | } |
7c673cae | 7521 | } |
f67539c2 | 7522 | return _symlink(dir.get(), name.c_str(), target, perms, std::move(alternate_name)); |
7c673cae FG |
7523 | } |
7524 | ||
7525 | int Client::readlink(const char *relpath, char *buf, loff_t size, const UserPerm& perms) | |
7526 | { | |
b3b6e05e TL |
7527 | return readlinkat(CEPHFS_AT_FDCWD, relpath, buf, size, perms); |
7528 | } | |
7529 | ||
7530 | int Client::readlinkat(int dirfd, const char *relpath, char *buf, loff_t size, const UserPerm& perms) { | |
f67539c2 | 7531 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7532 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7533 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7534 | } |
f67539c2 | 7535 | |
11fdf7f2 | 7536 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7537 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7538 | tout(cct) << relpath << std::endl; |
7539 | ||
b3b6e05e | 7540 | InodeRef dirinode; |
f67539c2 | 7541 | std::scoped_lock lock(client_lock); |
b3b6e05e TL |
7542 | int r = get_fd_inode(dirfd, &dirinode); |
7543 | if (r < 0) { | |
7c673cae | 7544 | return r; |
b3b6e05e TL |
7545 | } |
7546 | ||
7547 | InodeRef in; | |
7548 | filepath path(relpath); | |
7549 | r = path_walk(path, &in, perms, false, 0, dirinode); | |
7550 | if (r < 0) { | |
7551 | return r; | |
7552 | } | |
7c673cae FG |
7553 | |
7554 | return _readlink(in.get(), buf, size); | |
7555 | } | |
7556 | ||
7557 | int Client::_readlink(Inode *in, char *buf, size_t size) | |
7558 | { | |
7559 | if (!in->is_symlink()) | |
f67539c2 | 7560 | return -CEPHFS_EINVAL; |
7c673cae FG |
7561 | |
7562 | // copy into buf (at most size bytes) | |
7563 | int r = in->symlink.length(); | |
7564 | if (r > (int)size) | |
7565 | r = size; | |
7566 | memcpy(buf, in->symlink.c_str(), r); | |
7567 | return r; | |
7568 | } | |
7569 | ||
7570 | ||
7571 | // inode stuff | |
7572 | ||
7573 | int Client::_getattr(Inode *in, int mask, const UserPerm& perms, bool force) | |
7574 | { | |
94b18763 | 7575 | bool yes = in->caps_issued_mask(mask, true); |
7c673cae | 7576 | |
11fdf7f2 | 7577 | ldout(cct, 10) << __func__ << " mask " << ccap_string(mask) << " issued=" << yes << dendl; |
7c673cae FG |
7578 | if (yes && !force) |
7579 | return 0; | |
7580 | ||
7581 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
7582 | filepath path; | |
7583 | in->make_nosnap_relative_path(path); | |
7584 | req->set_filepath(path); | |
7585 | req->set_inode(in); | |
7586 | req->head.args.getattr.mask = mask; | |
7587 | ||
7588 | int res = make_request(req, perms); | |
11fdf7f2 | 7589 | ldout(cct, 10) << __func__ << " result=" << res << dendl; |
7c673cae FG |
7590 | return res; |
7591 | } | |
7592 | ||
1d09f67e TL |
7593 | int Client::_getvxattr( |
7594 | Inode *in, | |
7595 | const UserPerm& perms, | |
7596 | const char *xattr_name, | |
7597 | ssize_t size, | |
7598 | void *value, | |
7599 | mds_rank_t rank) | |
7600 | { | |
7601 | if (!xattr_name || strlen(xattr_name) <= 0 || strlen(xattr_name) > 255) { | |
7602 | return -CEPHFS_ENODATA; | |
7603 | } | |
7604 | ||
7605 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETVXATTR); | |
7606 | filepath path; | |
7607 | in->make_nosnap_relative_path(path); | |
7608 | req->set_filepath(path); | |
7609 | req->set_inode(in); | |
7610 | req->set_string2(xattr_name); | |
7611 | ||
7612 | bufferlist bl; | |
7613 | int res = make_request(req, perms, nullptr, nullptr, rank, &bl); | |
7614 | ldout(cct, 10) << __func__ << " result=" << res << dendl; | |
7615 | ||
7616 | if (res < 0) { | |
7617 | return res; | |
7618 | } | |
7619 | ||
7620 | std::string buf; | |
7621 | auto p = bl.cbegin(); | |
7622 | ||
7623 | DECODE_START(1, p); | |
7624 | decode(buf, p); | |
7625 | DECODE_FINISH(p); | |
7626 | ||
7627 | ssize_t len = buf.length(); | |
7628 | ||
7629 | res = len; // refer to man getxattr(2) for output buffer size == 0 | |
7630 | ||
7631 | if (size > 0) { | |
7632 | if (len > size) { | |
7633 | res = -CEPHFS_ERANGE; // insufficient output buffer space | |
7634 | } else { | |
7635 | memcpy(value, buf.c_str(), len); | |
7636 | } | |
7637 | } | |
7638 | return res; | |
7639 | } | |
7640 | ||
7c673cae FG |
7641 | int Client::_do_setattr(Inode *in, struct ceph_statx *stx, int mask, |
7642 | const UserPerm& perms, InodeRef *inp) | |
7643 | { | |
7644 | int issued = in->caps_issued(); | |
20effc67 TL |
7645 | union ceph_mds_request_args args; |
7646 | bool kill_sguid = false; | |
7647 | int inode_drop = 0; | |
7c673cae | 7648 | |
11fdf7f2 | 7649 | ldout(cct, 10) << __func__ << " mask " << mask << " issued " << |
7c673cae FG |
7650 | ccap_string(issued) << dendl; |
7651 | ||
7652 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 7653 | return -CEPHFS_EROFS; |
7c673cae FG |
7654 | } |
7655 | if ((mask & CEPH_SETATTR_SIZE) && | |
f67539c2 TL |
7656 | (uint64_t)stx->stx_size > in->size && |
7657 | is_quota_bytes_exceeded(in, (uint64_t)stx->stx_size - in->size, | |
7c673cae | 7658 | perms)) { |
f67539c2 | 7659 | return -CEPHFS_EDQUOT; |
7c673cae FG |
7660 | } |
7661 | ||
20effc67 TL |
7662 | memset(&args, 0, sizeof(args)); |
7663 | ||
7c673cae FG |
7664 | // make the change locally? |
7665 | if ((in->cap_dirtier_uid >= 0 && perms.uid() != in->cap_dirtier_uid) || | |
7666 | (in->cap_dirtier_gid >= 0 && perms.gid() != in->cap_dirtier_gid)) { | |
7667 | ldout(cct, 10) << __func__ << " caller " << perms.uid() << ":" << perms.gid() | |
7668 | << " != cap dirtier " << in->cap_dirtier_uid << ":" | |
7669 | << in->cap_dirtier_gid << ", forcing sync setattr" | |
7670 | << dendl; | |
7671 | /* | |
7672 | * This works because we implicitly flush the caps as part of the | |
7673 | * request, so the cap update check will happen with the writeback | |
7674 | * cap context, and then the setattr check will happen with the | |
7675 | * caller's context. | |
7676 | * | |
7677 | * In reality this pattern is likely pretty rare (different users | |
7678 | * setattr'ing the same file). If that turns out not to be the | |
7679 | * case later, we can build a more complex pipelined cap writeback | |
7680 | * infrastructure... | |
7681 | */ | |
20effc67 | 7682 | mask |= CEPH_SETATTR_CTIME; |
7c673cae FG |
7683 | } |
7684 | ||
7685 | if (!mask) { | |
7686 | // caller just needs us to bump the ctime | |
7687 | in->ctime = ceph_clock_now(); | |
7688 | in->cap_dirtier_uid = perms.uid(); | |
7689 | in->cap_dirtier_gid = perms.gid(); | |
7690 | if (issued & CEPH_CAP_AUTH_EXCL) | |
28e407b8 | 7691 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7692 | else if (issued & CEPH_CAP_FILE_EXCL) |
28e407b8 | 7693 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
7c673cae | 7694 | else if (issued & CEPH_CAP_XATTR_EXCL) |
28e407b8 | 7695 | in->mark_caps_dirty(CEPH_CAP_XATTR_EXCL); |
7c673cae FG |
7696 | else |
7697 | mask |= CEPH_SETATTR_CTIME; | |
7698 | } | |
7699 | ||
7700 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
20effc67 | 7701 | kill_sguid = mask & (CEPH_SETATTR_SIZE|CEPH_SETATTR_KILL_SGUID); |
7c673cae FG |
7702 | |
7703 | mask &= ~CEPH_SETATTR_KILL_SGUID; | |
20effc67 TL |
7704 | } else if (mask & CEPH_SETATTR_SIZE) { |
7705 | /* If we don't have Ax, then we must ask the server to clear them on truncate */ | |
7706 | mask |= CEPH_SETATTR_KILL_SGUID; | |
7707 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7708 | } | |
7709 | ||
7710 | if (mask & CEPH_SETATTR_UID) { | |
7711 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
7c673cae | 7712 | |
20effc67 | 7713 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { |
7c673cae FG |
7714 | in->ctime = ceph_clock_now(); |
7715 | in->cap_dirtier_uid = perms.uid(); | |
7716 | in->cap_dirtier_gid = perms.gid(); | |
7717 | in->uid = stx->stx_uid; | |
28e407b8 | 7718 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7719 | mask &= ~CEPH_SETATTR_UID; |
7720 | kill_sguid = true; | |
20effc67 TL |
7721 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7722 | in->uid != stx->stx_uid) { | |
7723 | args.setattr.uid = stx->stx_uid; | |
7724 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7725 | } else { | |
7726 | mask &= ~CEPH_SETATTR_UID; | |
7c673cae | 7727 | } |
20effc67 TL |
7728 | } |
7729 | ||
7730 | if (mask & CEPH_SETATTR_GID) { | |
7731 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
7732 | ||
7733 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
7734 | in->ctime = ceph_clock_now(); |
7735 | in->cap_dirtier_uid = perms.uid(); | |
7736 | in->cap_dirtier_gid = perms.gid(); | |
7737 | in->gid = stx->stx_gid; | |
28e407b8 | 7738 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7739 | mask &= ~CEPH_SETATTR_GID; |
7740 | kill_sguid = true; | |
20effc67 TL |
7741 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7742 | in->gid != stx->stx_gid) { | |
7743 | args.setattr.gid = stx->stx_gid; | |
7744 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7745 | } else { | |
7746 | mask &= ~CEPH_SETATTR_GID; | |
7c673cae | 7747 | } |
20effc67 | 7748 | } |
7c673cae | 7749 | |
20effc67 TL |
7750 | if (mask & CEPH_SETATTR_MODE) { |
7751 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
7752 | ||
7753 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
7754 | in->ctime = ceph_clock_now(); |
7755 | in->cap_dirtier_uid = perms.uid(); | |
7756 | in->cap_dirtier_gid = perms.gid(); | |
7757 | in->mode = (in->mode & ~07777) | (stx->stx_mode & 07777); | |
28e407b8 | 7758 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7759 | mask &= ~CEPH_SETATTR_MODE; |
20effc67 TL |
7760 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7761 | in->mode != stx->stx_mode) { | |
7762 | args.setattr.mode = stx->stx_mode; | |
7763 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7764 | } else { | |
7765 | mask &= ~CEPH_SETATTR_MODE; | |
7c673cae | 7766 | } |
20effc67 TL |
7767 | } else if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL) && |
7768 | kill_sguid && S_ISREG(in->mode) && | |
7769 | (in->mode & (S_IXUSR|S_IXGRP|S_IXOTH))) { | |
7770 | /* Must squash the any setuid/setgid bits with an ownership change */ | |
7771 | in->mode &= ~(S_ISUID|S_ISGID); | |
7772 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); | |
7773 | } | |
7774 | ||
7775 | if (mask & CEPH_SETATTR_BTIME) { | |
7776 | ldout(cct,10) << "changing btime to " << in->btime << dendl; | |
7c673cae | 7777 | |
20effc67 | 7778 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { |
7c673cae FG |
7779 | in->ctime = ceph_clock_now(); |
7780 | in->cap_dirtier_uid = perms.uid(); | |
7781 | in->cap_dirtier_gid = perms.gid(); | |
7782 | in->btime = utime_t(stx->stx_btime); | |
28e407b8 | 7783 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7784 | mask &= ~CEPH_SETATTR_BTIME; |
20effc67 TL |
7785 | } else if (!in->caps_issued_mask(CEPH_CAP_AUTH_SHARED) || |
7786 | in->btime != utime_t(stx->stx_btime)) { | |
7787 | args.setattr.btime = utime_t(stx->stx_btime); | |
7788 | inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7789 | } else { | |
7790 | mask &= ~CEPH_SETATTR_BTIME; | |
7791 | } | |
7792 | } | |
7793 | ||
7794 | if (mask & CEPH_SETATTR_SIZE) { | |
7795 | if ((uint64_t)stx->stx_size >= mdsmap->get_max_filesize()) { | |
7796 | //too big! | |
7797 | ldout(cct,10) << "unable to set size to " << stx->stx_size << ". Too large!" << dendl; | |
7798 | return -CEPHFS_EFBIG; | |
7799 | } | |
7800 | ||
7801 | ldout(cct,10) << "changing size to " << stx->stx_size << dendl; | |
7802 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL) && | |
7803 | !(mask & CEPH_SETATTR_KILL_SGUID) && | |
7804 | stx->stx_size >= in->size) { | |
7805 | if (stx->stx_size > in->size) { | |
7806 | in->size = in->reported_size = stx->stx_size; | |
7807 | in->cap_dirtier_uid = perms.uid(); | |
7808 | in->cap_dirtier_gid = perms.gid(); | |
7809 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); | |
7810 | mask &= ~(CEPH_SETATTR_SIZE); | |
7811 | mask |= CEPH_SETATTR_MTIME; | |
7812 | } else { | |
7813 | // ignore it when size doesn't change | |
7814 | mask &= ~(CEPH_SETATTR_SIZE); | |
7815 | } | |
7816 | } else { | |
7817 | args.setattr.size = stx->stx_size; | |
7818 | inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
7819 | CEPH_CAP_FILE_WR; | |
7820 | } | |
7821 | } | |
7822 | ||
7823 | if (mask & CEPH_SETATTR_MTIME) { | |
7824 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
7825 | in->mtime = utime_t(stx->stx_mtime); | |
7826 | in->ctime = ceph_clock_now(); | |
7827 | in->cap_dirtier_uid = perms.uid(); | |
7828 | in->cap_dirtier_gid = perms.gid(); | |
7829 | in->time_warp_seq++; | |
7830 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); | |
7831 | mask &= ~CEPH_SETATTR_MTIME; | |
7832 | } else if (in->caps_issued_mask(CEPH_CAP_FILE_WR) && | |
7833 | utime_t(stx->stx_mtime) > in->mtime) { | |
7834 | in->mtime = utime_t(stx->stx_mtime); | |
7835 | in->ctime = ceph_clock_now(); | |
7836 | in->cap_dirtier_uid = perms.uid(); | |
7837 | in->cap_dirtier_gid = perms.gid(); | |
7838 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); | |
7839 | mask &= ~CEPH_SETATTR_MTIME; | |
7840 | } else if (!in->caps_issued_mask(CEPH_CAP_FILE_SHARED) || | |
7841 | in->mtime != utime_t(stx->stx_mtime)) { | |
7842 | args.setattr.mtime = utime_t(stx->stx_mtime); | |
7843 | inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
7844 | CEPH_CAP_FILE_WR; | |
7845 | } else { | |
7846 | mask &= ~CEPH_SETATTR_MTIME; | |
7c673cae | 7847 | } |
7c673cae FG |
7848 | } |
7849 | ||
20effc67 TL |
7850 | if (mask & CEPH_SETATTR_ATIME) { |
7851 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
7852 | in->atime = utime_t(stx->stx_atime); | |
7c673cae FG |
7853 | in->ctime = ceph_clock_now(); |
7854 | in->cap_dirtier_uid = perms.uid(); | |
7855 | in->cap_dirtier_gid = perms.gid(); | |
7856 | in->time_warp_seq++; | |
28e407b8 | 7857 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
20effc67 TL |
7858 | mask &= ~CEPH_SETATTR_ATIME; |
7859 | } else if (in->caps_issued_mask(CEPH_CAP_FILE_WR) && | |
7860 | utime_t(stx->stx_atime) > in->atime) { | |
7861 | in->atime = utime_t(stx->stx_atime); | |
7862 | in->ctime = ceph_clock_now(); | |
7863 | in->cap_dirtier_uid = perms.uid(); | |
7864 | in->cap_dirtier_gid = perms.gid(); | |
7865 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); | |
7866 | mask &= ~CEPH_SETATTR_ATIME; | |
7867 | } else if (!in->caps_issued_mask(CEPH_CAP_FILE_SHARED) || | |
7868 | in->atime != utime_t(stx->stx_atime)) { | |
7869 | args.setattr.atime = utime_t(stx->stx_atime); | |
7870 | inode_drop |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
7871 | CEPH_CAP_FILE_WR; | |
7872 | } else { | |
7873 | mask &= ~CEPH_SETATTR_ATIME; | |
7c673cae FG |
7874 | } |
7875 | } | |
20effc67 | 7876 | |
7c673cae FG |
7877 | if (!mask) { |
7878 | in->change_attr++; | |
7879 | return 0; | |
7880 | } | |
7881 | ||
7c673cae FG |
7882 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETATTR); |
7883 | ||
7884 | filepath path; | |
7885 | ||
7886 | in->make_nosnap_relative_path(path); | |
7887 | req->set_filepath(path); | |
7888 | req->set_inode(in); | |
7889 | ||
20effc67 TL |
7890 | req->head.args = args; |
7891 | req->inode_drop = inode_drop; | |
7c673cae | 7892 | req->head.args.setattr.mask = mask; |
7c673cae FG |
7893 | req->regetattr_mask = mask; |
7894 | ||
7895 | int res = make_request(req, perms, inp); | |
7896 | ldout(cct, 10) << "_setattr result=" << res << dendl; | |
7897 | return res; | |
7898 | } | |
7899 | ||
7900 | /* Note that we only care about attrs that setattr cares about */ | |
7901 | void Client::stat_to_statx(struct stat *st, struct ceph_statx *stx) | |
7902 | { | |
7903 | stx->stx_size = st->st_size; | |
7904 | stx->stx_mode = st->st_mode; | |
7905 | stx->stx_uid = st->st_uid; | |
7906 | stx->stx_gid = st->st_gid; | |
11fdf7f2 TL |
7907 | #ifdef __APPLE__ |
7908 | stx->stx_mtime = st->st_mtimespec; | |
7909 | stx->stx_atime = st->st_atimespec; | |
f67539c2 TL |
7910 | #elif __WIN32 |
7911 | stx->stx_mtime.tv_sec = st->st_mtime; | |
7912 | stx->stx_atime.tv_sec = st->st_atime; | |
11fdf7f2 | 7913 | #else |
7c673cae FG |
7914 | stx->stx_mtime = st->st_mtim; |
7915 | stx->stx_atime = st->st_atim; | |
11fdf7f2 | 7916 | #endif |
7c673cae FG |
7917 | } |
7918 | ||
7919 | int Client::__setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
7920 | const UserPerm& perms, InodeRef *inp) | |
7921 | { | |
7922 | int ret = _do_setattr(in, stx, mask, perms, inp); | |
7923 | if (ret < 0) | |
7924 | return ret; | |
7925 | if (mask & CEPH_SETATTR_MODE) | |
7926 | ret = _posix_acl_chmod(in, stx->stx_mode, perms); | |
7927 | return ret; | |
7928 | } | |
7929 | ||
7930 | int Client::_setattrx(InodeRef &in, struct ceph_statx *stx, int mask, | |
7931 | const UserPerm& perms) | |
7932 | { | |
7933 | mask &= (CEPH_SETATTR_MODE | CEPH_SETATTR_UID | | |
7934 | CEPH_SETATTR_GID | CEPH_SETATTR_MTIME | | |
7935 | CEPH_SETATTR_ATIME | CEPH_SETATTR_SIZE | | |
7936 | CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME); | |
7937 | if (cct->_conf->client_permissions) { | |
7938 | int r = may_setattr(in.get(), stx, mask, perms); | |
7939 | if (r < 0) | |
7940 | return r; | |
7941 | } | |
7942 | return __setattrx(in.get(), stx, mask, perms); | |
7943 | } | |
7944 | ||
7945 | int Client::_setattr(InodeRef &in, struct stat *attr, int mask, | |
7946 | const UserPerm& perms) | |
7947 | { | |
7948 | struct ceph_statx stx; | |
7949 | ||
7950 | stat_to_statx(attr, &stx); | |
7951 | mask &= ~CEPH_SETATTR_BTIME; | |
181888fb FG |
7952 | |
7953 | if ((mask & CEPH_SETATTR_UID) && attr->st_uid == static_cast<uid_t>(-1)) { | |
7954 | mask &= ~CEPH_SETATTR_UID; | |
7955 | } | |
7956 | if ((mask & CEPH_SETATTR_GID) && attr->st_gid == static_cast<uid_t>(-1)) { | |
7957 | mask &= ~CEPH_SETATTR_GID; | |
7958 | } | |
7959 | ||
7c673cae FG |
7960 | return _setattrx(in, &stx, mask, perms); |
7961 | } | |
7962 | ||
7963 | int Client::setattr(const char *relpath, struct stat *attr, int mask, | |
7964 | const UserPerm& perms) | |
7965 | { | |
f67539c2 TL |
7966 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7967 | if (!mref_reader.is_state_satisfied()) | |
7968 | return -CEPHFS_ENOTCONN; | |
7969 | ||
11fdf7f2 | 7970 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7971 | tout(cct) << relpath << std::endl; |
7972 | tout(cct) << mask << std::endl; | |
7973 | ||
7974 | filepath path(relpath); | |
7975 | InodeRef in; | |
f67539c2 TL |
7976 | |
7977 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7978 | int r = path_walk(path, &in, perms); |
7979 | if (r < 0) | |
7980 | return r; | |
7981 | return _setattr(in, attr, mask, perms); | |
7982 | } | |
7983 | ||
7984 | int Client::setattrx(const char *relpath, struct ceph_statx *stx, int mask, | |
7985 | const UserPerm& perms, int flags) | |
7986 | { | |
f67539c2 TL |
7987 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7988 | if (!mref_reader.is_state_satisfied()) | |
7989 | return -CEPHFS_ENOTCONN; | |
7990 | ||
11fdf7f2 | 7991 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7992 | tout(cct) << relpath << std::endl; |
7993 | tout(cct) << mask << std::endl; | |
7994 | ||
7995 | filepath path(relpath); | |
7996 | InodeRef in; | |
f67539c2 TL |
7997 | |
7998 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7999 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW)); |
8000 | if (r < 0) | |
8001 | return r; | |
8002 | return _setattrx(in, stx, mask, perms); | |
8003 | } | |
8004 | ||
8005 | int Client::fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms) | |
8006 | { | |
f67539c2 TL |
8007 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8008 | if (!mref_reader.is_state_satisfied()) | |
8009 | return -CEPHFS_ENOTCONN; | |
8010 | ||
11fdf7f2 | 8011 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8012 | tout(cct) << fd << std::endl; |
8013 | tout(cct) << mask << std::endl; | |
8014 | ||
f67539c2 | 8015 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8016 | Fh *f = get_filehandle(fd); |
8017 | if (!f) | |
f67539c2 | 8018 | return -CEPHFS_EBADF; |
7c673cae FG |
8019 | #if defined(__linux__) && defined(O_PATH) |
8020 | if (f->flags & O_PATH) | |
f67539c2 | 8021 | return -CEPHFS_EBADF; |
7c673cae FG |
8022 | #endif |
8023 | return _setattr(f->inode, attr, mask, perms); | |
8024 | } | |
8025 | ||
8026 | int Client::fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms) | |
8027 | { | |
f67539c2 TL |
8028 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8029 | if (!mref_reader.is_state_satisfied()) | |
8030 | return -CEPHFS_ENOTCONN; | |
8031 | ||
11fdf7f2 | 8032 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8033 | tout(cct) << fd << std::endl; |
8034 | tout(cct) << mask << std::endl; | |
8035 | ||
f67539c2 | 8036 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8037 | Fh *f = get_filehandle(fd); |
8038 | if (!f) | |
f67539c2 | 8039 | return -CEPHFS_EBADF; |
7c673cae FG |
8040 | #if defined(__linux__) && defined(O_PATH) |
8041 | if (f->flags & O_PATH) | |
f67539c2 | 8042 | return -CEPHFS_EBADF; |
7c673cae FG |
8043 | #endif |
8044 | return _setattrx(f->inode, stx, mask, perms); | |
8045 | } | |
8046 | ||
8047 | int Client::stat(const char *relpath, struct stat *stbuf, const UserPerm& perms, | |
8048 | frag_info_t *dirstat, int mask) | |
8049 | { | |
f67539c2 TL |
8050 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8051 | if (!mref_reader.is_state_satisfied()) | |
8052 | return -CEPHFS_ENOTCONN; | |
8053 | ||
11fdf7f2 | 8054 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
8055 | tout(cct) << "stat" << std::endl; |
8056 | tout(cct) << relpath << std::endl; | |
181888fb | 8057 | |
7c673cae FG |
8058 | filepath path(relpath); |
8059 | InodeRef in; | |
f67539c2 TL |
8060 | |
8061 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8062 | int r = path_walk(path, &in, perms, true, mask); |
8063 | if (r < 0) | |
8064 | return r; | |
8065 | r = _getattr(in, mask, perms); | |
8066 | if (r < 0) { | |
11fdf7f2 | 8067 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
8068 | return r; |
8069 | } | |
8070 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 8071 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
8072 | return r; |
8073 | } | |
8074 | ||
8075 | unsigned Client::statx_to_mask(unsigned int flags, unsigned int want) | |
8076 | { | |
8077 | unsigned mask = 0; | |
8078 | ||
2a845540 TL |
8079 | /* The AT_STATX_FORCE_SYNC is always in higher priority than AT_STATX_DONT_SYNC. */ |
8080 | if ((flags & AT_STATX_SYNC_TYPE) == AT_STATX_DONT_SYNC) | |
7c673cae FG |
8081 | goto out; |
8082 | ||
2a845540 | 8083 | /* Always set PIN to distinguish from AT_STATX_DONT_SYNC case */ |
7c673cae FG |
8084 | mask |= CEPH_CAP_PIN; |
8085 | if (want & (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
8086 | mask |= CEPH_CAP_AUTH_SHARED; | |
8087 | if (want & (CEPH_STATX_NLINK|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
8088 | mask |= CEPH_CAP_LINK_SHARED; | |
adb31ebb | 8089 | 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 |
8090 | mask |= CEPH_CAP_FILE_SHARED; |
8091 | if (want & (CEPH_STATX_VERSION|CEPH_STATX_CTIME)) | |
8092 | mask |= CEPH_CAP_XATTR_SHARED; | |
8093 | out: | |
8094 | return mask; | |
8095 | } | |
8096 | ||
8097 | int Client::statx(const char *relpath, struct ceph_statx *stx, | |
8098 | const UserPerm& perms, | |
8099 | unsigned int want, unsigned int flags) | |
8100 | { | |
b3b6e05e | 8101 | return statxat(CEPHFS_AT_FDCWD, relpath, stx, perms, want, flags); |
7c673cae FG |
8102 | } |
8103 | ||
8104 | int Client::lstat(const char *relpath, struct stat *stbuf, | |
8105 | const UserPerm& perms, frag_info_t *dirstat, int mask) | |
8106 | { | |
f67539c2 TL |
8107 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8108 | if (!mref_reader.is_state_satisfied()) | |
8109 | return -CEPHFS_ENOTCONN; | |
8110 | ||
11fdf7f2 | 8111 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
11fdf7f2 | 8112 | tout(cct) << __func__ << std::endl; |
7c673cae | 8113 | tout(cct) << relpath << std::endl; |
181888fb | 8114 | |
7c673cae FG |
8115 | filepath path(relpath); |
8116 | InodeRef in; | |
f67539c2 TL |
8117 | |
8118 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8119 | // don't follow symlinks |
8120 | int r = path_walk(path, &in, perms, false, mask); | |
8121 | if (r < 0) | |
8122 | return r; | |
8123 | r = _getattr(in, mask, perms); | |
8124 | if (r < 0) { | |
11fdf7f2 | 8125 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
8126 | return r; |
8127 | } | |
8128 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 8129 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
8130 | return r; |
8131 | } | |
8132 | ||
8133 | int Client::fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat, nest_info_t *rstat) | |
8134 | { | |
11fdf7f2 | 8135 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
8136 | << " mode 0" << oct << in->mode << dec |
8137 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
8138 | memset(st, 0, sizeof(struct stat)); | |
8139 | if (use_faked_inos()) | |
8140 | st->st_ino = in->faked_ino; | |
8141 | else | |
8142 | st->st_ino = in->ino; | |
8143 | st->st_dev = in->snapid; | |
8144 | st->st_mode = in->mode; | |
8145 | st->st_rdev = in->rdev; | |
28e407b8 AA |
8146 | if (in->is_dir()) { |
8147 | switch (in->nlink) { | |
8148 | case 0: | |
8149 | st->st_nlink = 0; /* dir is unlinked */ | |
8150 | break; | |
8151 | case 1: | |
8152 | st->st_nlink = 1 /* parent dentry */ | |
8153 | + 1 /* <dir>/. */ | |
8154 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
8155 | break; | |
8156 | default: | |
8157 | ceph_abort(); | |
8158 | } | |
8159 | } else { | |
8160 | st->st_nlink = in->nlink; | |
8161 | } | |
7c673cae FG |
8162 | st->st_uid = in->uid; |
8163 | st->st_gid = in->gid; | |
8164 | if (in->ctime > in->mtime) { | |
8165 | stat_set_ctime_sec(st, in->ctime.sec()); | |
8166 | stat_set_ctime_nsec(st, in->ctime.nsec()); | |
8167 | } else { | |
8168 | stat_set_ctime_sec(st, in->mtime.sec()); | |
8169 | stat_set_ctime_nsec(st, in->mtime.nsec()); | |
8170 | } | |
8171 | stat_set_atime_sec(st, in->atime.sec()); | |
8172 | stat_set_atime_nsec(st, in->atime.nsec()); | |
8173 | stat_set_mtime_sec(st, in->mtime.sec()); | |
8174 | stat_set_mtime_nsec(st, in->mtime.nsec()); | |
8175 | if (in->is_dir()) { | |
8176 | if (cct->_conf->client_dirsize_rbytes) | |
8177 | st->st_size = in->rstat.rbytes; | |
8178 | else | |
8179 | st->st_size = in->dirstat.size(); | |
f67539c2 TL |
8180 | // The Windows "stat" structure provides just a subset of the fields that are |
8181 | // available on Linux. | |
8182 | #ifndef _WIN32 | |
7c673cae | 8183 | st->st_blocks = 1; |
f67539c2 | 8184 | #endif |
7c673cae FG |
8185 | } else { |
8186 | st->st_size = in->size; | |
f67539c2 | 8187 | #ifndef _WIN32 |
7c673cae | 8188 | st->st_blocks = (in->size + 511) >> 9; |
f67539c2 | 8189 | #endif |
7c673cae | 8190 | } |
f67539c2 | 8191 | #ifndef _WIN32 |
11fdf7f2 | 8192 | st->st_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
f67539c2 | 8193 | #endif |
7c673cae FG |
8194 | |
8195 | if (dirstat) | |
8196 | *dirstat = in->dirstat; | |
8197 | if (rstat) | |
8198 | *rstat = in->rstat; | |
8199 | ||
8200 | return in->caps_issued(); | |
8201 | } | |
8202 | ||
8203 | void Client::fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx) | |
8204 | { | |
11fdf7f2 | 8205 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
8206 | << " mode 0" << oct << in->mode << dec |
8207 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
8208 | memset(stx, 0, sizeof(struct ceph_statx)); | |
8209 | ||
8210 | /* | |
2a845540 | 8211 | * If mask is 0, then the caller set AT_STATX_DONT_SYNC. Reset the mask |
7c673cae FG |
8212 | * so that all bits are set. |
8213 | */ | |
8214 | if (!mask) | |
8215 | mask = ~0; | |
8216 | ||
8217 | /* These are always considered to be available */ | |
8218 | stx->stx_dev = in->snapid; | |
11fdf7f2 | 8219 | stx->stx_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
7c673cae FG |
8220 | |
8221 | /* Type bits are always set, even when CEPH_STATX_MODE is not */ | |
8222 | stx->stx_mode = S_IFMT & in->mode; | |
8223 | stx->stx_ino = use_faked_inos() ? in->faked_ino : (ino_t)in->ino; | |
8224 | stx->stx_rdev = in->rdev; | |
8225 | stx->stx_mask |= (CEPH_STATX_INO|CEPH_STATX_RDEV); | |
8226 | ||
8227 | if (mask & CEPH_CAP_AUTH_SHARED) { | |
8228 | stx->stx_uid = in->uid; | |
8229 | stx->stx_gid = in->gid; | |
8230 | stx->stx_mode = in->mode; | |
8231 | in->btime.to_timespec(&stx->stx_btime); | |
8232 | stx->stx_mask |= (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME); | |
8233 | } | |
8234 | ||
8235 | if (mask & CEPH_CAP_LINK_SHARED) { | |
28e407b8 AA |
8236 | if (in->is_dir()) { |
8237 | switch (in->nlink) { | |
8238 | case 0: | |
8239 | stx->stx_nlink = 0; /* dir is unlinked */ | |
8240 | break; | |
8241 | case 1: | |
8242 | stx->stx_nlink = 1 /* parent dentry */ | |
8243 | + 1 /* <dir>/. */ | |
8244 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
8245 | break; | |
8246 | default: | |
8247 | ceph_abort(); | |
8248 | } | |
8249 | } else { | |
8250 | stx->stx_nlink = in->nlink; | |
8251 | } | |
7c673cae FG |
8252 | stx->stx_mask |= CEPH_STATX_NLINK; |
8253 | } | |
8254 | ||
8255 | if (mask & CEPH_CAP_FILE_SHARED) { | |
8256 | ||
8257 | in->atime.to_timespec(&stx->stx_atime); | |
8258 | in->mtime.to_timespec(&stx->stx_mtime); | |
8259 | ||
8260 | if (in->is_dir()) { | |
8261 | if (cct->_conf->client_dirsize_rbytes) | |
8262 | stx->stx_size = in->rstat.rbytes; | |
8263 | else | |
8264 | stx->stx_size = in->dirstat.size(); | |
8265 | stx->stx_blocks = 1; | |
8266 | } else { | |
8267 | stx->stx_size = in->size; | |
8268 | stx->stx_blocks = (in->size + 511) >> 9; | |
8269 | } | |
8270 | stx->stx_mask |= (CEPH_STATX_ATIME|CEPH_STATX_MTIME| | |
8271 | CEPH_STATX_SIZE|CEPH_STATX_BLOCKS); | |
8272 | } | |
8273 | ||
8274 | /* Change time and change_attr both require all shared caps to view */ | |
8275 | if ((mask & CEPH_STAT_CAP_INODE_ALL) == CEPH_STAT_CAP_INODE_ALL) { | |
8276 | stx->stx_version = in->change_attr; | |
8277 | if (in->ctime > in->mtime) | |
8278 | in->ctime.to_timespec(&stx->stx_ctime); | |
8279 | else | |
8280 | in->mtime.to_timespec(&stx->stx_ctime); | |
8281 | stx->stx_mask |= (CEPH_STATX_CTIME|CEPH_STATX_VERSION); | |
8282 | } | |
8283 | ||
8284 | } | |
8285 | ||
8286 | void Client::touch_dn(Dentry *dn) | |
8287 | { | |
8288 | lru.lru_touch(dn); | |
8289 | } | |
8290 | ||
8291 | int Client::chmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
8292 | { | |
b3b6e05e | 8293 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, 0, perms); |
7c673cae FG |
8294 | } |
8295 | ||
8296 | int Client::fchmod(int fd, mode_t mode, const UserPerm& perms) | |
8297 | { | |
f67539c2 TL |
8298 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8299 | if (!mref_reader.is_state_satisfied()) | |
8300 | return -CEPHFS_ENOTCONN; | |
8301 | ||
11fdf7f2 | 8302 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8303 | tout(cct) << fd << std::endl; |
8304 | tout(cct) << mode << std::endl; | |
181888fb | 8305 | |
f67539c2 | 8306 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8307 | Fh *f = get_filehandle(fd); |
8308 | if (!f) | |
f67539c2 | 8309 | return -CEPHFS_EBADF; |
7c673cae FG |
8310 | #if defined(__linux__) && defined(O_PATH) |
8311 | if (f->flags & O_PATH) | |
f67539c2 | 8312 | return -CEPHFS_EBADF; |
7c673cae FG |
8313 | #endif |
8314 | struct stat attr; | |
8315 | attr.st_mode = mode; | |
8316 | return _setattr(f->inode, &attr, CEPH_SETATTR_MODE, perms); | |
8317 | } | |
8318 | ||
b3b6e05e TL |
8319 | int Client::chmodat(int dirfd, const char *relpath, mode_t mode, int flags, |
8320 | const UserPerm& perms) { | |
f67539c2 | 8321 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8322 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8323 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8324 | } |
f67539c2 | 8325 | |
11fdf7f2 | 8326 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8327 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8328 | tout(cct) << relpath << std::endl; |
8329 | tout(cct) << mode << std::endl; | |
b3b6e05e | 8330 | tout(cct) << flags << std::endl; |
181888fb | 8331 | |
7c673cae FG |
8332 | filepath path(relpath); |
8333 | InodeRef in; | |
b3b6e05e | 8334 | InodeRef dirinode; |
f67539c2 TL |
8335 | |
8336 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8337 | int r = get_fd_inode(dirfd, &dirinode); |
8338 | if (r < 0) { | |
8339 | return r; | |
8340 | } | |
8341 | ||
8342 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8343 | if (r < 0) { | |
7c673cae | 8344 | return r; |
b3b6e05e | 8345 | } |
7c673cae FG |
8346 | struct stat attr; |
8347 | attr.st_mode = mode; | |
8348 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
8349 | } | |
8350 | ||
b3b6e05e TL |
8351 | int Client::lchmod(const char *relpath, mode_t mode, const UserPerm& perms) |
8352 | { | |
8353 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, AT_SYMLINK_NOFOLLOW, perms); | |
8354 | } | |
8355 | ||
7c673cae FG |
8356 | int Client::chown(const char *relpath, uid_t new_uid, gid_t new_gid, |
8357 | const UserPerm& perms) | |
8358 | { | |
b3b6e05e | 8359 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, 0, perms); |
7c673cae FG |
8360 | } |
8361 | ||
8362 | int Client::fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms) | |
8363 | { | |
f67539c2 TL |
8364 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8365 | if (!mref_reader.is_state_satisfied()) | |
8366 | return -CEPHFS_ENOTCONN; | |
8367 | ||
11fdf7f2 | 8368 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8369 | tout(cct) << fd << std::endl; |
8370 | tout(cct) << new_uid << std::endl; | |
8371 | tout(cct) << new_gid << std::endl; | |
181888fb | 8372 | |
f67539c2 | 8373 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8374 | Fh *f = get_filehandle(fd); |
8375 | if (!f) | |
f67539c2 | 8376 | return -CEPHFS_EBADF; |
7c673cae FG |
8377 | #if defined(__linux__) && defined(O_PATH) |
8378 | if (f->flags & O_PATH) | |
f67539c2 | 8379 | return -CEPHFS_EBADF; |
7c673cae FG |
8380 | #endif |
8381 | struct stat attr; | |
8382 | attr.st_uid = new_uid; | |
8383 | attr.st_gid = new_gid; | |
8384 | int mask = 0; | |
8385 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
8386 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
8387 | return _setattr(f->inode, &attr, mask, perms); | |
8388 | } | |
8389 | ||
8390 | int Client::lchown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
8391 | const UserPerm& perms) | |
8392 | { | |
b3b6e05e TL |
8393 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, AT_SYMLINK_NOFOLLOW, perms); |
8394 | } | |
8395 | ||
8396 | int Client::chownat(int dirfd, const char *relpath, uid_t new_uid, gid_t new_gid, | |
8397 | int flags, const UserPerm& perms) { | |
f67539c2 | 8398 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8399 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8400 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8401 | } |
f67539c2 | 8402 | |
11fdf7f2 | 8403 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8404 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8405 | tout(cct) << relpath << std::endl; |
8406 | tout(cct) << new_uid << std::endl; | |
8407 | tout(cct) << new_gid << std::endl; | |
b3b6e05e | 8408 | tout(cct) << flags << std::endl; |
181888fb | 8409 | |
7c673cae FG |
8410 | filepath path(relpath); |
8411 | InodeRef in; | |
b3b6e05e | 8412 | InodeRef dirinode; |
f67539c2 TL |
8413 | |
8414 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8415 | int r = get_fd_inode(dirfd, &dirinode); |
8416 | if (r < 0) { | |
7c673cae | 8417 | return r; |
b3b6e05e TL |
8418 | } |
8419 | ||
8420 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8421 | if (r < 0) { | |
8422 | return r; | |
8423 | } | |
7c673cae FG |
8424 | struct stat attr; |
8425 | attr.st_uid = new_uid; | |
8426 | attr.st_gid = new_gid; | |
b3b6e05e | 8427 | return _setattr(in, &attr, CEPH_SETATTR_UID|CEPH_SETATTR_GID, perms); |
7c673cae FG |
8428 | } |
8429 | ||
11fdf7f2 TL |
8430 | static void attr_set_atime_and_mtime(struct stat *attr, |
8431 | const utime_t &atime, | |
8432 | const utime_t &mtime) | |
8433 | { | |
8434 | stat_set_atime_sec(attr, atime.tv.tv_sec); | |
8435 | stat_set_atime_nsec(attr, atime.tv.tv_nsec); | |
8436 | stat_set_mtime_sec(attr, mtime.tv.tv_sec); | |
8437 | stat_set_mtime_nsec(attr, mtime.tv.tv_nsec); | |
8438 | } | |
8439 | ||
8440 | // for [l]utime() invoke the timeval variant as the timespec | |
8441 | // variant are not yet implemented. for futime[s](), invoke | |
8442 | // the timespec variant. | |
7c673cae FG |
8443 | int Client::utime(const char *relpath, struct utimbuf *buf, |
8444 | const UserPerm& perms) | |
8445 | { | |
11fdf7f2 TL |
8446 | struct timeval tv[2]; |
8447 | tv[0].tv_sec = buf->actime; | |
8448 | tv[0].tv_usec = 0; | |
8449 | tv[1].tv_sec = buf->modtime; | |
8450 | tv[1].tv_usec = 0; | |
8451 | ||
8452 | return utimes(relpath, tv, perms); | |
8453 | } | |
8454 | ||
8455 | int Client::lutime(const char *relpath, struct utimbuf *buf, | |
8456 | const UserPerm& perms) | |
8457 | { | |
8458 | struct timeval tv[2]; | |
8459 | tv[0].tv_sec = buf->actime; | |
8460 | tv[0].tv_usec = 0; | |
8461 | tv[1].tv_sec = buf->modtime; | |
8462 | tv[1].tv_usec = 0; | |
8463 | ||
8464 | return lutimes(relpath, tv, perms); | |
8465 | } | |
8466 | ||
8467 | int Client::futime(int fd, struct utimbuf *buf, const UserPerm& perms) | |
8468 | { | |
8469 | struct timespec ts[2]; | |
8470 | ts[0].tv_sec = buf->actime; | |
8471 | ts[0].tv_nsec = 0; | |
8472 | ts[1].tv_sec = buf->modtime; | |
8473 | ts[1].tv_nsec = 0; | |
8474 | ||
8475 | return futimens(fd, ts, perms); | |
8476 | } | |
8477 | ||
8478 | int Client::utimes(const char *relpath, struct timeval times[2], | |
8479 | const UserPerm& perms) | |
8480 | { | |
f67539c2 TL |
8481 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8482 | if (!mref_reader.is_state_satisfied()) | |
8483 | return -CEPHFS_ENOTCONN; | |
8484 | ||
11fdf7f2 | 8485 | tout(cct) << __func__ << std::endl; |
7c673cae | 8486 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8487 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8488 | << std::endl; | |
8489 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8490 | << std::endl; | |
181888fb | 8491 | |
7c673cae FG |
8492 | filepath path(relpath); |
8493 | InodeRef in; | |
f67539c2 TL |
8494 | |
8495 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8496 | int r = path_walk(path, &in, perms); |
8497 | if (r < 0) | |
8498 | return r; | |
8499 | struct stat attr; | |
11fdf7f2 TL |
8500 | utime_t atime(times[0]); |
8501 | utime_t mtime(times[1]); | |
8502 | ||
8503 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8504 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8505 | } | |
8506 | ||
11fdf7f2 TL |
8507 | int Client::lutimes(const char *relpath, struct timeval times[2], |
8508 | const UserPerm& perms) | |
7c673cae | 8509 | { |
f67539c2 TL |
8510 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8511 | if (!mref_reader.is_state_satisfied()) | |
8512 | return -CEPHFS_ENOTCONN; | |
8513 | ||
11fdf7f2 | 8514 | tout(cct) << __func__ << std::endl; |
7c673cae | 8515 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8516 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8517 | << std::endl; | |
8518 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8519 | << std::endl; | |
181888fb | 8520 | |
7c673cae FG |
8521 | filepath path(relpath); |
8522 | InodeRef in; | |
f67539c2 TL |
8523 | |
8524 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8525 | int r = path_walk(path, &in, perms, false); |
8526 | if (r < 0) | |
8527 | return r; | |
8528 | struct stat attr; | |
11fdf7f2 TL |
8529 | utime_t atime(times[0]); |
8530 | utime_t mtime(times[1]); | |
8531 | ||
8532 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8533 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8534 | } | |
8535 | ||
11fdf7f2 TL |
8536 | int Client::futimes(int fd, struct timeval times[2], const UserPerm& perms) |
8537 | { | |
8538 | struct timespec ts[2]; | |
8539 | ts[0].tv_sec = times[0].tv_sec; | |
8540 | ts[0].tv_nsec = times[0].tv_usec * 1000; | |
8541 | ts[1].tv_sec = times[1].tv_sec; | |
8542 | ts[1].tv_nsec = times[1].tv_usec * 1000; | |
8543 | ||
8544 | return futimens(fd, ts, perms); | |
8545 | } | |
8546 | ||
8547 | int Client::futimens(int fd, struct timespec times[2], const UserPerm& perms) | |
8548 | { | |
f67539c2 TL |
8549 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8550 | if (!mref_reader.is_state_satisfied()) | |
8551 | return -CEPHFS_ENOTCONN; | |
8552 | ||
11fdf7f2 TL |
8553 | tout(cct) << __func__ << std::endl; |
8554 | tout(cct) << fd << std::endl; | |
8555 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8556 | << std::endl; | |
8557 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8558 | << std::endl; | |
8559 | ||
f67539c2 | 8560 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
8561 | Fh *f = get_filehandle(fd); |
8562 | if (!f) | |
f67539c2 | 8563 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8564 | #if defined(__linux__) && defined(O_PATH) |
8565 | if (f->flags & O_PATH) | |
f67539c2 | 8566 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8567 | #endif |
8568 | struct stat attr; | |
8569 | utime_t atime(times[0]); | |
8570 | utime_t mtime(times[1]); | |
8571 | ||
8572 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8573 | return _setattr(f->inode, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8574 | } | |
8575 | ||
b3b6e05e TL |
8576 | int Client::utimensat(int dirfd, const char *relpath, struct timespec times[2], int flags, |
8577 | const UserPerm& perms) { | |
8578 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8579 | if (!mref_reader.is_state_satisfied()) { | |
8580 | return -CEPHFS_ENOTCONN; | |
8581 | } | |
8582 | ||
8583 | tout(cct) << __func__ << std::endl; | |
8584 | tout(cct) << dirfd << std::endl; | |
8585 | tout(cct) << relpath << std::endl; | |
8586 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8587 | << std::endl; | |
8588 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8589 | << std::endl; | |
8590 | tout(cct) << flags << std::endl; | |
8591 | ||
8592 | filepath path(relpath); | |
8593 | InodeRef in; | |
8594 | InodeRef dirinode; | |
8595 | ||
8596 | std::scoped_lock lock(client_lock); | |
8597 | int r = get_fd_inode(dirfd, &dirinode); | |
8598 | if (r < 0) { | |
8599 | return r; | |
8600 | } | |
8601 | ||
8602 | #if defined(__linux__) && defined(O_PATH) | |
8603 | if (flags & O_PATH) { | |
8604 | return -CEPHFS_EBADF; | |
8605 | } | |
8606 | #endif | |
8607 | ||
8608 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8609 | if (r < 0) { | |
8610 | return r; | |
8611 | } | |
8612 | struct stat attr; | |
8613 | utime_t atime(times[0]); | |
8614 | utime_t mtime(times[1]); | |
8615 | ||
8616 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8617 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8618 | } | |
8619 | ||
7c673cae FG |
8620 | int Client::flock(int fd, int operation, uint64_t owner) |
8621 | { | |
f67539c2 TL |
8622 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8623 | if (!mref_reader.is_state_satisfied()) | |
8624 | return -CEPHFS_ENOTCONN; | |
8625 | ||
11fdf7f2 | 8626 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8627 | tout(cct) << fd << std::endl; |
8628 | tout(cct) << operation << std::endl; | |
8629 | tout(cct) << owner << std::endl; | |
181888fb | 8630 | |
f67539c2 | 8631 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8632 | Fh *f = get_filehandle(fd); |
8633 | if (!f) | |
f67539c2 | 8634 | return -CEPHFS_EBADF; |
7c673cae FG |
8635 | |
8636 | return _flock(f, operation, owner); | |
8637 | } | |
8638 | ||
8639 | int Client::opendir(const char *relpath, dir_result_t **dirpp, const UserPerm& perms) | |
8640 | { | |
f67539c2 TL |
8641 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8642 | if (!mref_reader.is_state_satisfied()) | |
8643 | return -CEPHFS_ENOTCONN; | |
8644 | ||
11fdf7f2 | 8645 | tout(cct) << __func__ << std::endl; |
7c673cae | 8646 | tout(cct) << relpath << std::endl; |
181888fb | 8647 | |
7c673cae FG |
8648 | filepath path(relpath); |
8649 | InodeRef in; | |
f67539c2 TL |
8650 | |
8651 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8652 | int r = path_walk(path, &in, perms, true); |
8653 | if (r < 0) | |
8654 | return r; | |
8655 | if (cct->_conf->client_permissions) { | |
8656 | int r = may_open(in.get(), O_RDONLY, perms); | |
8657 | if (r < 0) | |
8658 | return r; | |
8659 | } | |
8660 | r = _opendir(in.get(), dirpp, perms); | |
8661 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
f67539c2 TL |
8662 | if (r != -CEPHFS_ENOTDIR) |
8663 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
7c673cae FG |
8664 | return r; |
8665 | } | |
8666 | ||
b3b6e05e TL |
8667 | int Client::fdopendir(int dirfd, dir_result_t **dirpp, const UserPerm &perms) { |
8668 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8669 | if (!mref_reader.is_state_satisfied()) { | |
8670 | return -CEPHFS_ENOTCONN; | |
8671 | } | |
8672 | ||
8673 | tout(cct) << __func__ << std::endl; | |
8674 | tout(cct) << dirfd << std::endl; | |
8675 | ||
8676 | InodeRef dirinode; | |
8677 | std::scoped_lock locker(client_lock); | |
8678 | int r = get_fd_inode(dirfd, &dirinode); | |
8679 | if (r < 0) { | |
8680 | return r; | |
8681 | } | |
8682 | ||
8683 | if (cct->_conf->client_permissions) { | |
8684 | r = may_open(dirinode.get(), O_RDONLY, perms); | |
8685 | if (r < 0) { | |
8686 | return r; | |
8687 | } | |
8688 | } | |
8689 | r = _opendir(dirinode.get(), dirpp, perms); | |
8690 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
8691 | if (r != -CEPHFS_ENOTDIR) { | |
8692 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
8693 | } | |
8694 | return r; | |
8695 | } | |
8696 | ||
7c673cae FG |
8697 | int Client::_opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms) |
8698 | { | |
8699 | if (!in->is_dir()) | |
f67539c2 | 8700 | return -CEPHFS_ENOTDIR; |
7c673cae FG |
8701 | *dirpp = new dir_result_t(in, perms); |
8702 | opened_dirs.insert(*dirpp); | |
11fdf7f2 | 8703 | ldout(cct, 8) << __func__ << "(" << in->ino << ") = " << 0 << " (" << *dirpp << ")" << dendl; |
7c673cae FG |
8704 | return 0; |
8705 | } | |
8706 | ||
8707 | ||
8708 | int Client::closedir(dir_result_t *dir) | |
8709 | { | |
11fdf7f2 | 8710 | tout(cct) << __func__ << std::endl; |
f67539c2 | 8711 | tout(cct) << (uintptr_t)dir << std::endl; |
7c673cae | 8712 | |
11fdf7f2 | 8713 | ldout(cct, 3) << __func__ << "(" << dir << ") = 0" << dendl; |
f67539c2 | 8714 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8715 | _closedir(dir); |
8716 | return 0; | |
8717 | } | |
8718 | ||
8719 | void Client::_closedir(dir_result_t *dirp) | |
8720 | { | |
11fdf7f2 | 8721 | ldout(cct, 10) << __func__ << "(" << dirp << ")" << dendl; |
f67539c2 | 8722 | |
7c673cae | 8723 | if (dirp->inode) { |
11fdf7f2 | 8724 | ldout(cct, 10) << __func__ << " detaching inode " << dirp->inode << dendl; |
7c673cae FG |
8725 | dirp->inode.reset(); |
8726 | } | |
8727 | _readdir_drop_dirp_buffer(dirp); | |
8728 | opened_dirs.erase(dirp); | |
8729 | delete dirp; | |
8730 | } | |
8731 | ||
8732 | void Client::rewinddir(dir_result_t *dirp) | |
8733 | { | |
11fdf7f2 | 8734 | ldout(cct, 3) << __func__ << "(" << dirp << ")" << dendl; |
181888fb | 8735 | |
f67539c2 TL |
8736 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8737 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8738 | return; |
8739 | ||
f67539c2 | 8740 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8741 | dir_result_t *d = static_cast<dir_result_t*>(dirp); |
8742 | _readdir_drop_dirp_buffer(d); | |
8743 | d->reset(); | |
8744 | } | |
8745 | ||
8746 | loff_t Client::telldir(dir_result_t *dirp) | |
8747 | { | |
8748 | dir_result_t *d = static_cast<dir_result_t*>(dirp); | |
11fdf7f2 | 8749 | ldout(cct, 3) << __func__ << "(" << dirp << ") = " << d->offset << dendl; |
7c673cae FG |
8750 | return d->offset; |
8751 | } | |
8752 | ||
8753 | void Client::seekdir(dir_result_t *dirp, loff_t offset) | |
8754 | { | |
11fdf7f2 | 8755 | ldout(cct, 3) << __func__ << "(" << dirp << ", " << offset << ")" << dendl; |
7c673cae | 8756 | |
f67539c2 TL |
8757 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8758 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8759 | return; |
8760 | ||
f67539c2 TL |
8761 | std::scoped_lock lock(client_lock); |
8762 | ||
7c673cae FG |
8763 | if (offset == dirp->offset) |
8764 | return; | |
8765 | ||
8766 | if (offset > dirp->offset) | |
8767 | dirp->release_count = 0; // bump if we do a forward seek | |
8768 | else | |
8769 | dirp->ordered_count = 0; // disable filling readdir cache | |
8770 | ||
8771 | if (dirp->hash_order()) { | |
8772 | if (dirp->offset > offset) { | |
8773 | _readdir_drop_dirp_buffer(dirp); | |
8774 | dirp->reset(); | |
8775 | } | |
8776 | } else { | |
8777 | if (offset == 0 || | |
8778 | dirp->buffer_frag != frag_t(dir_result_t::fpos_high(offset)) || | |
8779 | dirp->offset_low() > dir_result_t::fpos_low(offset)) { | |
8780 | _readdir_drop_dirp_buffer(dirp); | |
8781 | dirp->reset(); | |
8782 | } | |
8783 | } | |
8784 | ||
8785 | dirp->offset = offset; | |
8786 | } | |
8787 | ||
8788 | ||
8789 | //struct dirent { | |
8790 | // ino_t d_ino; /* inode number */ | |
8791 | // off_t d_off; /* offset to the next dirent */ | |
8792 | // unsigned short d_reclen; /* length of this record */ | |
8793 | // unsigned char d_type; /* type of file */ | |
8794 | // char d_name[256]; /* filename */ | |
8795 | //}; | |
8796 | void Client::fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off) | |
8797 | { | |
8798 | strncpy(de->d_name, name, 255); | |
8799 | de->d_name[255] = '\0'; | |
f67539c2 | 8800 | #if !defined(__CYGWIN__) && !(defined(_WIN32)) |
7c673cae | 8801 | de->d_ino = ino; |
11fdf7f2 | 8802 | #if !defined(__APPLE__) && !defined(__FreeBSD__) |
7c673cae FG |
8803 | de->d_off = next_off; |
8804 | #endif | |
8805 | de->d_reclen = 1; | |
8806 | de->d_type = IFTODT(type); | |
11fdf7f2 | 8807 | ldout(cct, 10) << __func__ << " '" << de->d_name << "' -> " << inodeno_t(de->d_ino) |
7c673cae FG |
8808 | << " type " << (int)de->d_type << " w/ next_off " << hex << next_off << dec << dendl; |
8809 | #endif | |
8810 | } | |
8811 | ||
8812 | void Client::_readdir_next_frag(dir_result_t *dirp) | |
8813 | { | |
8814 | frag_t fg = dirp->buffer_frag; | |
8815 | ||
8816 | if (fg.is_rightmost()) { | |
11fdf7f2 | 8817 | ldout(cct, 10) << __func__ << " advance from " << fg << " to END" << dendl; |
7c673cae FG |
8818 | dirp->set_end(); |
8819 | return; | |
8820 | } | |
8821 | ||
8822 | // advance | |
8823 | fg = fg.next(); | |
11fdf7f2 | 8824 | ldout(cct, 10) << __func__ << " advance from " << dirp->buffer_frag << " to " << fg << dendl; |
7c673cae FG |
8825 | |
8826 | if (dirp->hash_order()) { | |
8827 | // keep last_name | |
8828 | int64_t new_offset = dir_result_t::make_fpos(fg.value(), 2, true); | |
8829 | if (dirp->offset < new_offset) // don't decrease offset | |
8830 | dirp->offset = new_offset; | |
8831 | } else { | |
8832 | dirp->last_name.clear(); | |
8833 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); | |
8834 | _readdir_rechoose_frag(dirp); | |
8835 | } | |
8836 | } | |
8837 | ||
8838 | void Client::_readdir_rechoose_frag(dir_result_t *dirp) | |
8839 | { | |
11fdf7f2 | 8840 | ceph_assert(dirp->inode); |
7c673cae FG |
8841 | |
8842 | if (dirp->hash_order()) | |
8843 | return; | |
8844 | ||
8845 | frag_t cur = frag_t(dirp->offset_high()); | |
8846 | frag_t fg = dirp->inode->dirfragtree[cur.value()]; | |
8847 | if (fg != cur) { | |
11fdf7f2 | 8848 | ldout(cct, 10) << __func__ << " frag " << cur << " maps to " << fg << dendl; |
7c673cae FG |
8849 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); |
8850 | dirp->last_name.clear(); | |
8851 | dirp->next_offset = 2; | |
8852 | } | |
8853 | } | |
8854 | ||
8855 | void Client::_readdir_drop_dirp_buffer(dir_result_t *dirp) | |
8856 | { | |
11fdf7f2 | 8857 | ldout(cct, 10) << __func__ << " " << dirp << dendl; |
7c673cae FG |
8858 | dirp->buffer.clear(); |
8859 | } | |
8860 | ||
8861 | int Client::_readdir_get_frag(dir_result_t *dirp) | |
8862 | { | |
11fdf7f2 TL |
8863 | ceph_assert(dirp); |
8864 | ceph_assert(dirp->inode); | |
7c673cae FG |
8865 | |
8866 | // get the current frag. | |
8867 | frag_t fg; | |
8868 | if (dirp->hash_order()) | |
8869 | fg = dirp->inode->dirfragtree[dirp->offset_high()]; | |
8870 | else | |
8871 | fg = frag_t(dirp->offset_high()); | |
8872 | ||
11fdf7f2 | 8873 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " fg " << fg |
7c673cae FG |
8874 | << " offset " << hex << dirp->offset << dec << dendl; |
8875 | ||
8876 | int op = CEPH_MDS_OP_READDIR; | |
8877 | if (dirp->inode && dirp->inode->snapid == CEPH_SNAPDIR) | |
8878 | op = CEPH_MDS_OP_LSSNAP; | |
8879 | ||
8880 | InodeRef& diri = dirp->inode; | |
8881 | ||
8882 | MetaRequest *req = new MetaRequest(op); | |
8883 | filepath path; | |
8884 | diri->make_nosnap_relative_path(path); | |
8885 | req->set_filepath(path); | |
8886 | req->set_inode(diri.get()); | |
8887 | req->head.args.readdir.frag = fg; | |
8888 | req->head.args.readdir.flags = CEPH_READDIR_REPLY_BITFLAGS; | |
8889 | if (dirp->last_name.length()) { | |
94b18763 | 8890 | req->path2.set_path(dirp->last_name); |
7c673cae FG |
8891 | } else if (dirp->hash_order()) { |
8892 | req->head.args.readdir.offset_hash = dirp->offset_high(); | |
8893 | } | |
8894 | req->dirp = dirp; | |
8895 | ||
8896 | bufferlist dirbl; | |
8897 | int res = make_request(req, dirp->perms, NULL, NULL, -1, &dirbl); | |
8898 | ||
f67539c2 | 8899 | if (res == -CEPHFS_EAGAIN) { |
11fdf7f2 | 8900 | ldout(cct, 10) << __func__ << " got EAGAIN, retrying" << dendl; |
7c673cae FG |
8901 | _readdir_rechoose_frag(dirp); |
8902 | return _readdir_get_frag(dirp); | |
8903 | } | |
8904 | ||
8905 | if (res == 0) { | |
11fdf7f2 | 8906 | ldout(cct, 10) << __func__ << " " << dirp << " got frag " << dirp->buffer_frag |
7c673cae FG |
8907 | << " size " << dirp->buffer.size() << dendl; |
8908 | } else { | |
11fdf7f2 | 8909 | ldout(cct, 10) << __func__ << " got error " << res << ", setting end flag" << dendl; |
7c673cae FG |
8910 | dirp->set_end(); |
8911 | } | |
8912 | ||
8913 | return res; | |
8914 | } | |
8915 | ||
8916 | struct dentry_off_lt { | |
8917 | bool operator()(const Dentry* dn, int64_t off) const { | |
8918 | return dir_result_t::fpos_cmp(dn->offset, off) < 0; | |
8919 | } | |
8920 | }; | |
8921 | ||
8922 | int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, | |
8923 | int caps, bool getref) | |
8924 | { | |
f67539c2 | 8925 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11fdf7f2 | 8926 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino |
7c673cae FG |
8927 | << " last_name " << dirp->last_name << " offset " << hex << dirp->offset << dec |
8928 | << dendl; | |
8929 | Dir *dir = dirp->inode->dir; | |
8930 | ||
8931 | if (!dir) { | |
8932 | ldout(cct, 10) << " dir is empty" << dendl; | |
8933 | dirp->set_end(); | |
8934 | return 0; | |
8935 | } | |
8936 | ||
8937 | vector<Dentry*>::iterator pd = std::lower_bound(dir->readdir_cache.begin(), | |
8938 | dir->readdir_cache.end(), | |
8939 | dirp->offset, dentry_off_lt()); | |
8940 | ||
8941 | string dn_name; | |
8942 | while (true) { | |
adb31ebb | 8943 | int mask = caps; |
7c673cae | 8944 | if (!dirp->inode->is_complete_and_ordered()) |
f67539c2 | 8945 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8946 | if (pd == dir->readdir_cache.end()) |
8947 | break; | |
8948 | Dentry *dn = *pd; | |
8949 | if (dn->inode == NULL) { | |
8950 | ldout(cct, 15) << " skipping null '" << dn->name << "'" << dendl; | |
8951 | ++pd; | |
8952 | continue; | |
8953 | } | |
8954 | if (dn->cap_shared_gen != dir->parent_inode->shared_gen) { | |
8955 | ldout(cct, 15) << " skipping mismatch shared gen '" << dn->name << "'" << dendl; | |
8956 | ++pd; | |
8957 | continue; | |
8958 | } | |
8959 | ||
92f5a8d4 | 8960 | int idx = pd - dir->readdir_cache.begin(); |
adb31ebb TL |
8961 | if (dn->inode->is_dir()) { |
8962 | mask |= CEPH_STAT_RSTAT; | |
8963 | } | |
8964 | int r = _getattr(dn->inode, mask, dirp->perms); | |
7c673cae FG |
8965 | if (r < 0) |
8966 | return r; | |
92f5a8d4 TL |
8967 | |
8968 | // the content of readdir_cache may change after _getattr(), so pd may be invalid iterator | |
8969 | pd = dir->readdir_cache.begin() + idx; | |
8970 | if (pd >= dir->readdir_cache.end() || *pd != dn) | |
f67539c2 | 8971 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8972 | |
8973 | struct ceph_statx stx; | |
8974 | struct dirent de; | |
8975 | fill_statx(dn->inode, caps, &stx); | |
8976 | ||
8977 | uint64_t next_off = dn->offset + 1; | |
eafe8130 | 8978 | fill_dirent(&de, dn->name.c_str(), stx.stx_mode, stx.stx_ino, next_off); |
7c673cae FG |
8979 | ++pd; |
8980 | if (pd == dir->readdir_cache.end()) | |
8981 | next_off = dir_result_t::END; | |
8982 | ||
8983 | Inode *in = NULL; | |
7c673cae FG |
8984 | if (getref) { |
8985 | in = dn->inode.get(); | |
8986 | _ll_get(in); | |
8987 | } | |
8988 | ||
8989 | dn_name = dn->name; // fill in name while we have lock | |
8990 | ||
9f95a23c | 8991 | client_lock.unlock(); |
7c673cae | 8992 | r = cb(p, &de, &stx, next_off, in); // _next_ offset |
9f95a23c | 8993 | client_lock.lock(); |
7c673cae FG |
8994 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << dn->offset << dec |
8995 | << " = " << r << dendl; | |
8996 | if (r < 0) { | |
8997 | return r; | |
8998 | } | |
8999 | ||
9000 | dirp->offset = next_off; | |
9001 | if (dirp->at_end()) | |
9002 | dirp->next_offset = 2; | |
9003 | else | |
9004 | dirp->next_offset = dirp->offset_low(); | |
9005 | dirp->last_name = dn_name; // we successfully returned this one; update! | |
28e407b8 | 9006 | dirp->release_count = 0; // last_name no longer match cache index |
7c673cae FG |
9007 | if (r > 0) |
9008 | return r; | |
9009 | } | |
9010 | ||
11fdf7f2 | 9011 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " at end" << dendl; |
7c673cae FG |
9012 | dirp->set_end(); |
9013 | return 0; | |
9014 | } | |
9015 | ||
9016 | int Client::readdir_r_cb(dir_result_t *d, add_dirent_cb_t cb, void *p, | |
9017 | unsigned want, unsigned flags, bool getref) | |
9018 | { | |
9019 | int caps = statx_to_mask(flags, want); | |
9020 | ||
f67539c2 TL |
9021 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9022 | if (!mref_reader.is_state_satisfied()) | |
9023 | return -CEPHFS_ENOTCONN; | |
7c673cae | 9024 | |
f67539c2 | 9025 | std::unique_lock cl(client_lock); |
181888fb | 9026 | |
7c673cae FG |
9027 | dir_result_t *dirp = static_cast<dir_result_t*>(d); |
9028 | ||
11fdf7f2 | 9029 | ldout(cct, 10) << __func__ << " " << *dirp->inode << " offset " << hex << dirp->offset |
7c673cae FG |
9030 | << dec << " at_end=" << dirp->at_end() |
9031 | << " hash_order=" << dirp->hash_order() << dendl; | |
9032 | ||
9033 | struct dirent de; | |
9034 | struct ceph_statx stx; | |
9035 | memset(&de, 0, sizeof(de)); | |
9036 | memset(&stx, 0, sizeof(stx)); | |
9037 | ||
9038 | InodeRef& diri = dirp->inode; | |
9039 | ||
9040 | if (dirp->at_end()) | |
9041 | return 0; | |
9042 | ||
9043 | if (dirp->offset == 0) { | |
9044 | ldout(cct, 15) << " including ." << dendl; | |
11fdf7f2 | 9045 | ceph_assert(diri->dentries.size() < 2); // can't have multiple hard-links to a dir |
7c673cae FG |
9046 | uint64_t next_off = 1; |
9047 | ||
9048 | int r; | |
adb31ebb | 9049 | r = _getattr(diri, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
9050 | if (r < 0) |
9051 | return r; | |
9052 | ||
9053 | fill_statx(diri, caps, &stx); | |
9054 | fill_dirent(&de, ".", S_IFDIR, stx.stx_ino, next_off); | |
9055 | ||
9056 | Inode *inode = NULL; | |
9057 | if (getref) { | |
9058 | inode = diri.get(); | |
9059 | _ll_get(inode); | |
9060 | } | |
9061 | ||
f67539c2 | 9062 | cl.unlock(); |
7c673cae | 9063 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 9064 | cl.lock(); |
7c673cae FG |
9065 | if (r < 0) |
9066 | return r; | |
9067 | ||
9068 | dirp->offset = next_off; | |
9069 | if (r > 0) | |
9070 | return r; | |
9071 | } | |
9072 | if (dirp->offset == 1) { | |
9073 | ldout(cct, 15) << " including .." << dendl; | |
9074 | uint64_t next_off = 2; | |
9075 | InodeRef in; | |
11fdf7f2 | 9076 | if (diri->dentries.empty()) |
7c673cae FG |
9077 | in = diri; |
9078 | else | |
94b18763 | 9079 | in = diri->get_first_parent()->dir->parent_inode; |
7c673cae FG |
9080 | |
9081 | int r; | |
adb31ebb | 9082 | r = _getattr(in, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
9083 | if (r < 0) |
9084 | return r; | |
9085 | ||
9086 | fill_statx(in, caps, &stx); | |
9087 | fill_dirent(&de, "..", S_IFDIR, stx.stx_ino, next_off); | |
9088 | ||
9089 | Inode *inode = NULL; | |
9090 | if (getref) { | |
9091 | inode = in.get(); | |
9092 | _ll_get(inode); | |
9093 | } | |
9094 | ||
f67539c2 | 9095 | cl.unlock(); |
7c673cae | 9096 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 9097 | cl.lock(); |
7c673cae FG |
9098 | if (r < 0) |
9099 | return r; | |
9100 | ||
9101 | dirp->offset = next_off; | |
9102 | if (r > 0) | |
9103 | return r; | |
9104 | } | |
9105 | ||
9106 | // can we read from our cache? | |
9107 | ldout(cct, 10) << "offset " << hex << dirp->offset << dec | |
9108 | << " snapid " << dirp->inode->snapid << " (complete && ordered) " | |
9109 | << dirp->inode->is_complete_and_ordered() | |
9110 | << " issued " << ccap_string(dirp->inode->caps_issued()) | |
9111 | << dendl; | |
9112 | if (dirp->inode->snapid != CEPH_SNAPDIR && | |
9113 | dirp->inode->is_complete_and_ordered() && | |
94b18763 | 9114 | dirp->inode->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 9115 | int err = _readdir_cache_cb(dirp, cb, p, caps, getref); |
f67539c2 | 9116 | if (err != -CEPHFS_EAGAIN) |
7c673cae FG |
9117 | return err; |
9118 | } | |
9119 | ||
9120 | while (1) { | |
9121 | if (dirp->at_end()) | |
9122 | return 0; | |
9123 | ||
9124 | bool check_caps = true; | |
9125 | if (!dirp->is_cached()) { | |
9126 | int r = _readdir_get_frag(dirp); | |
9127 | if (r) | |
9128 | return r; | |
9129 | // _readdir_get_frag () may updates dirp->offset if the replied dirfrag is | |
9130 | // different than the requested one. (our dirfragtree was outdated) | |
9131 | check_caps = false; | |
9132 | } | |
9133 | frag_t fg = dirp->buffer_frag; | |
9134 | ||
9135 | ldout(cct, 10) << "frag " << fg << " buffer size " << dirp->buffer.size() | |
9136 | << " offset " << hex << dirp->offset << dendl; | |
9137 | ||
9138 | for (auto it = std::lower_bound(dirp->buffer.begin(), dirp->buffer.end(), | |
9139 | dirp->offset, dir_result_t::dentry_off_lt()); | |
9140 | it != dirp->buffer.end(); | |
9141 | ++it) { | |
9142 | dir_result_t::dentry &entry = *it; | |
9143 | ||
9144 | uint64_t next_off = entry.offset + 1; | |
9145 | ||
9146 | int r; | |
9147 | if (check_caps) { | |
adb31ebb TL |
9148 | int mask = caps; |
9149 | if(entry.inode->is_dir()){ | |
9150 | mask |= CEPH_STAT_RSTAT; | |
9151 | } | |
9152 | r = _getattr(entry.inode, mask, dirp->perms); | |
7c673cae FG |
9153 | if (r < 0) |
9154 | return r; | |
9155 | } | |
9156 | ||
9157 | fill_statx(entry.inode, caps, &stx); | |
9158 | fill_dirent(&de, entry.name.c_str(), stx.stx_mode, stx.stx_ino, next_off); | |
9159 | ||
9160 | Inode *inode = NULL; | |
9161 | if (getref) { | |
9162 | inode = entry.inode.get(); | |
9163 | _ll_get(inode); | |
9164 | } | |
9165 | ||
f67539c2 | 9166 | cl.unlock(); |
7c673cae | 9167 | r = cb(p, &de, &stx, next_off, inode); // _next_ offset |
f67539c2 | 9168 | cl.lock(); |
7c673cae FG |
9169 | |
9170 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << next_off - 1 << dec | |
9171 | << " = " << r << dendl; | |
9172 | if (r < 0) | |
9173 | return r; | |
9174 | ||
9175 | dirp->offset = next_off; | |
9176 | if (r > 0) | |
9177 | return r; | |
9178 | } | |
9179 | ||
9180 | if (dirp->next_offset > 2) { | |
9181 | ldout(cct, 10) << " fetching next chunk of this frag" << dendl; | |
9182 | _readdir_drop_dirp_buffer(dirp); | |
9183 | continue; // more! | |
9184 | } | |
9185 | ||
9186 | if (!fg.is_rightmost()) { | |
9187 | // next frag! | |
9188 | _readdir_next_frag(dirp); | |
9189 | continue; | |
9190 | } | |
9191 | ||
9192 | if (diri->shared_gen == dirp->start_shared_gen && | |
9193 | diri->dir_release_count == dirp->release_count) { | |
9194 | if (diri->dir_ordered_count == dirp->ordered_count) { | |
9195 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
9196 | if (diri->dir) { | |
11fdf7f2 | 9197 | ceph_assert(diri->dir->readdir_cache.size() >= dirp->cache_index); |
7c673cae FG |
9198 | diri->dir->readdir_cache.resize(dirp->cache_index); |
9199 | } | |
9200 | diri->flags |= I_COMPLETE | I_DIR_ORDERED; | |
9201 | } else { | |
9202 | ldout(cct, 10) << " marking I_COMPLETE on " << *diri << dendl; | |
9203 | diri->flags |= I_COMPLETE; | |
9204 | } | |
9205 | } | |
9206 | ||
9207 | dirp->set_end(); | |
9208 | return 0; | |
9209 | } | |
9210 | ceph_abort(); | |
9211 | return 0; | |
9212 | } | |
9213 | ||
9214 | ||
9215 | int Client::readdir_r(dir_result_t *d, struct dirent *de) | |
9216 | { | |
9217 | return readdirplus_r(d, de, 0, 0, 0, NULL); | |
9218 | } | |
9219 | ||
9220 | /* | |
9221 | * readdirplus_r | |
9222 | * | |
9223 | * returns | |
9224 | * 1 if we got a dirent | |
9225 | * 0 for end of directory | |
9226 | * <0 on error | |
9227 | */ | |
9228 | ||
9229 | struct single_readdir { | |
9230 | struct dirent *de; | |
9231 | struct ceph_statx *stx; | |
9232 | Inode *inode; | |
9233 | bool full; | |
9234 | }; | |
9235 | ||
9236 | static int _readdir_single_dirent_cb(void *p, struct dirent *de, | |
9237 | struct ceph_statx *stx, off_t off, | |
9238 | Inode *in) | |
9239 | { | |
9240 | single_readdir *c = static_cast<single_readdir *>(p); | |
9241 | ||
9242 | if (c->full) | |
9243 | return -1; // already filled this dirent | |
9244 | ||
9245 | *c->de = *de; | |
9246 | if (c->stx) | |
9247 | *c->stx = *stx; | |
9248 | c->inode = in; | |
9249 | c->full = true; | |
9250 | return 1; | |
9251 | } | |
9252 | ||
9253 | struct dirent *Client::readdir(dir_result_t *d) | |
9254 | { | |
9255 | int ret; | |
f91f0fd5 | 9256 | auto& de = d->de; |
7c673cae FG |
9257 | single_readdir sr; |
9258 | sr.de = &de; | |
9259 | sr.stx = NULL; | |
9260 | sr.inode = NULL; | |
9261 | sr.full = false; | |
9262 | ||
9263 | // our callback fills the dirent and sets sr.full=true on first | |
9264 | // call, and returns -1 the second time around. | |
9265 | ret = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr); | |
9266 | if (ret < -1) { | |
9267 | errno = -ret; // this sucks. | |
9268 | return (dirent *) NULL; | |
9269 | } | |
9270 | if (sr.full) { | |
9271 | return &de; | |
9272 | } | |
9273 | return (dirent *) NULL; | |
9274 | } | |
9275 | ||
9276 | int Client::readdirplus_r(dir_result_t *d, struct dirent *de, | |
9277 | struct ceph_statx *stx, unsigned want, | |
9278 | unsigned flags, Inode **out) | |
9279 | { | |
9280 | single_readdir sr; | |
9281 | sr.de = de; | |
9282 | sr.stx = stx; | |
9283 | sr.inode = NULL; | |
9284 | sr.full = false; | |
9285 | ||
9286 | // our callback fills the dirent and sets sr.full=true on first | |
9287 | // call, and returns -1 the second time around. | |
9288 | int r = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr, want, flags, out); | |
9289 | if (r < -1) | |
9290 | return r; | |
9291 | if (out) | |
9292 | *out = sr.inode; | |
9293 | if (sr.full) | |
9294 | return 1; | |
9295 | return 0; | |
9296 | } | |
9297 | ||
9298 | ||
9299 | /* getdents */ | |
9300 | struct getdents_result { | |
9301 | char *buf; | |
9302 | int buflen; | |
9303 | int pos; | |
9304 | bool fullent; | |
9305 | }; | |
9306 | ||
9307 | static int _readdir_getdent_cb(void *p, struct dirent *de, | |
9308 | struct ceph_statx *stx, off_t off, Inode *in) | |
9309 | { | |
9310 | struct getdents_result *c = static_cast<getdents_result *>(p); | |
9311 | ||
9312 | int dlen; | |
9313 | if (c->fullent) | |
9314 | dlen = sizeof(*de); | |
9315 | else | |
9316 | dlen = strlen(de->d_name) + 1; | |
9317 | ||
9318 | if (c->pos + dlen > c->buflen) | |
9319 | return -1; // doesn't fit | |
9320 | ||
9321 | if (c->fullent) { | |
9322 | memcpy(c->buf + c->pos, de, sizeof(*de)); | |
9323 | } else { | |
9324 | memcpy(c->buf + c->pos, de->d_name, dlen); | |
9325 | } | |
9326 | c->pos += dlen; | |
9327 | return 0; | |
9328 | } | |
9329 | ||
9330 | int Client::_getdents(dir_result_t *dir, char *buf, int buflen, bool fullent) | |
9331 | { | |
9332 | getdents_result gr; | |
9333 | gr.buf = buf; | |
9334 | gr.buflen = buflen; | |
9335 | gr.fullent = fullent; | |
9336 | gr.pos = 0; | |
9337 | ||
9338 | int r = readdir_r_cb(dir, _readdir_getdent_cb, (void *)&gr); | |
9339 | ||
9340 | if (r < 0) { // some error | |
9341 | if (r == -1) { // buffer ran out of space | |
9342 | if (gr.pos) { // but we got some entries already! | |
9343 | return gr.pos; | |
9344 | } // or we need a larger buffer | |
f67539c2 | 9345 | return -CEPHFS_ERANGE; |
7c673cae FG |
9346 | } else { // actual error, return it |
9347 | return r; | |
9348 | } | |
9349 | } | |
9350 | return gr.pos; | |
9351 | } | |
9352 | ||
9353 | ||
9354 | /* getdir */ | |
9355 | struct getdir_result { | |
9356 | list<string> *contents; | |
9357 | int num; | |
9358 | }; | |
9359 | ||
9360 | static int _getdir_cb(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in) | |
9361 | { | |
9362 | getdir_result *r = static_cast<getdir_result *>(p); | |
9363 | ||
9364 | r->contents->push_back(de->d_name); | |
9365 | r->num++; | |
9366 | return 0; | |
9367 | } | |
9368 | ||
9369 | int Client::getdir(const char *relpath, list<string>& contents, | |
9370 | const UserPerm& perms) | |
9371 | { | |
9372 | ldout(cct, 3) << "getdir(" << relpath << ")" << dendl; | |
f67539c2 TL |
9373 | tout(cct) << "getdir" << std::endl; |
9374 | tout(cct) << relpath << std::endl; | |
7c673cae FG |
9375 | |
9376 | dir_result_t *d; | |
9377 | int r = opendir(relpath, &d, perms); | |
9378 | if (r < 0) | |
9379 | return r; | |
9380 | ||
9381 | getdir_result gr; | |
9382 | gr.contents = &contents; | |
9383 | gr.num = 0; | |
9384 | r = readdir_r_cb(d, _getdir_cb, (void *)&gr); | |
9385 | ||
9386 | closedir(d); | |
9387 | ||
9388 | if (r < 0) | |
9389 | return r; | |
9390 | return gr.num; | |
9391 | } | |
9392 | ||
9393 | ||
9394 | /****** file i/o **********/ | |
f67539c2 | 9395 | |
b3b6e05e | 9396 | // common parts for open and openat. call with client_lock locked. |
20effc67 | 9397 | int Client::create_and_open(int dirfd, const char *relpath, int flags, |
b3b6e05e TL |
9398 | const UserPerm& perms, mode_t mode, int stripe_unit, |
9399 | int stripe_count, int object_size, const char *data_pool, | |
9400 | std::string alternate_name) { | |
9401 | ceph_assert(ceph_mutex_is_locked(client_lock)); | |
f91f0fd5 | 9402 | int cflags = ceph_flags_sys2wire(flags); |
f91f0fd5 | 9403 | tout(cct) << cflags << std::endl; |
7c673cae FG |
9404 | |
9405 | Fh *fh = NULL; | |
9406 | ||
9407 | #if defined(__linux__) && defined(O_PATH) | |
9408 | /* When the O_PATH is being specified, others flags than O_DIRECTORY | |
9409 | * and O_NOFOLLOW are ignored. Please refer do_entry_open() function | |
9410 | * in kernel (fs/open.c). */ | |
9411 | if (flags & O_PATH) | |
9412 | flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; | |
9413 | #endif | |
9414 | ||
9415 | filepath path(relpath); | |
9416 | InodeRef in; | |
9417 | bool created = false; | |
9418 | /* O_CREATE with O_EXCL enforces O_NOFOLLOW. */ | |
9419 | bool followsym = !((flags & O_NOFOLLOW) || ((flags & O_CREAT) && (flags & O_EXCL))); | |
f91f0fd5 TL |
9420 | int mask = ceph_caps_for_mode(ceph_flags_to_mode(cflags)); |
9421 | ||
b3b6e05e | 9422 | InodeRef dirinode = nullptr; |
20effc67 TL |
9423 | int r = get_fd_inode(dirfd, &dirinode); |
9424 | if (r < 0) { | |
9425 | return r; | |
b3b6e05e | 9426 | } |
7c673cae | 9427 | |
20effc67 | 9428 | r = path_walk(path, &in, perms, followsym, mask, dirinode); |
7c673cae | 9429 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) |
f67539c2 | 9430 | return -CEPHFS_EEXIST; |
7c673cae FG |
9431 | |
9432 | #if defined(__linux__) && defined(O_PATH) | |
9433 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW) && !(flags & O_PATH)) | |
9434 | #else | |
b3b6e05e | 9435 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW)) |
7c673cae | 9436 | #endif |
f67539c2 | 9437 | return -CEPHFS_ELOOP; |
7c673cae | 9438 | |
f67539c2 | 9439 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
7c673cae FG |
9440 | filepath dirpath = path; |
9441 | string dname = dirpath.last_dentry(); | |
9442 | dirpath.pop_dentry(); | |
9443 | InodeRef dir; | |
9444 | r = path_walk(dirpath, &dir, perms, true, | |
b3b6e05e TL |
9445 | cct->_conf->client_permissions ? CEPH_CAP_AUTH_SHARED : 0, dirinode); |
9446 | if (r < 0) { | |
7c673cae | 9447 | goto out; |
b3b6e05e | 9448 | } |
7c673cae FG |
9449 | if (cct->_conf->client_permissions) { |
9450 | r = may_create(dir.get(), perms); | |
9451 | if (r < 0) | |
b3b6e05e | 9452 | goto out; |
7c673cae FG |
9453 | } |
9454 | r = _create(dir.get(), dname.c_str(), flags, mode, &in, &fh, stripe_unit, | |
f67539c2 TL |
9455 | stripe_count, object_size, data_pool, &created, perms, |
9456 | std::move(alternate_name)); | |
7c673cae FG |
9457 | } |
9458 | if (r < 0) | |
9459 | goto out; | |
9460 | ||
9461 | if (!created) { | |
9462 | // posix says we can only check permissions of existing files | |
9463 | if (cct->_conf->client_permissions) { | |
9464 | r = may_open(in.get(), flags, perms); | |
9465 | if (r < 0) | |
b3b6e05e | 9466 | goto out; |
7c673cae FG |
9467 | } |
9468 | } | |
9469 | ||
9470 | if (!fh) | |
9471 | r = _open(in.get(), flags, mode, &fh, perms); | |
9472 | if (r >= 0) { | |
9473 | // allocate a integer file descriptor | |
11fdf7f2 | 9474 | ceph_assert(fh); |
7c673cae | 9475 | r = get_fd(); |
11fdf7f2 | 9476 | ceph_assert(fd_map.count(r) == 0); |
7c673cae FG |
9477 | fd_map[r] = fh; |
9478 | } | |
9479 | ||
9480 | out: | |
b3b6e05e TL |
9481 | return r; |
9482 | } | |
9483 | ||
9484 | int Client::open(const char *relpath, int flags, const UserPerm& perms, | |
9485 | mode_t mode, int stripe_unit, int stripe_count, | |
9486 | int object_size, const char *data_pool, std::string alternate_name) | |
9487 | { | |
9488 | return openat(CEPHFS_AT_FDCWD, relpath, flags, perms, mode, stripe_unit, | |
9489 | stripe_count, object_size, data_pool, alternate_name); | |
9490 | } | |
9491 | ||
b3b6e05e TL |
9492 | int Client::openat(int dirfd, const char *relpath, int flags, const UserPerm& perms, |
9493 | mode_t mode, int stripe_unit, int stripe_count, int object_size, | |
9494 | const char *data_pool, std::string alternate_name) { | |
9495 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9496 | if (!mref_reader.is_state_satisfied()) { | |
9497 | return -CEPHFS_ENOTCONN; | |
9498 | } | |
9499 | ||
9500 | ldout(cct, 3) << "openat enter(" << relpath << ")" << dendl; | |
9501 | tout(cct) << dirfd << std::endl; | |
9502 | tout(cct) << relpath << std::endl; | |
9503 | tout(cct) << flags << std::endl; | |
9504 | tout(cct) << mode << std::endl; | |
9505 | ||
9506 | std::scoped_lock locker(client_lock); | |
9507 | int r = create_and_open(dirfd, relpath, flags, perms, mode, stripe_unit, stripe_count, | |
9508 | object_size, data_pool, alternate_name); | |
9509 | ||
7c673cae | 9510 | tout(cct) << r << std::endl; |
b3b6e05e | 9511 | ldout(cct, 3) << "openat exit(" << relpath << ")" << dendl; |
7c673cae FG |
9512 | return r; |
9513 | } | |
9514 | ||
7c673cae FG |
9515 | int Client::lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name, |
9516 | const UserPerm& perms) | |
9517 | { | |
11fdf7f2 | 9518 | ldout(cct, 3) << __func__ << " enter(" << ino << ", #" << dirino << "/" << name << ")" << dendl; |
7c673cae | 9519 | |
f67539c2 TL |
9520 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9521 | if (!mref_reader.is_state_satisfied()) | |
9522 | return -CEPHFS_ENOTCONN; | |
181888fb | 9523 | |
f67539c2 | 9524 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9525 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPHASH); |
9526 | filepath path(ino); | |
9527 | req->set_filepath(path); | |
9528 | ||
9529 | uint32_t h = ceph_str_hash(CEPH_STR_HASH_RJENKINS, name, strlen(name)); | |
9530 | char f[30]; | |
9531 | sprintf(f, "%u", h); | |
9532 | filepath path2(dirino); | |
9533 | path2.push_dentry(string(f)); | |
9534 | req->set_filepath2(path2); | |
9535 | ||
9536 | int r = make_request(req, perms, NULL, NULL, | |
9537 | rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9538 | ldout(cct, 3) << __func__ << " exit(" << ino << ", #" << dirino << "/" << name << ") = " << r << dendl; |
7c673cae FG |
9539 | return r; |
9540 | } | |
9541 | ||
9542 | ||
9543 | /** | |
9544 | * Load inode into local cache. | |
9545 | * | |
9546 | * If inode pointer is non-NULL, and take a reference on | |
9547 | * the resulting Inode object in one operation, so that caller | |
9548 | * can safely assume inode will still be there after return. | |
9549 | */ | |
f67539c2 | 9550 | int Client::_lookup_vino(vinodeno_t vino, const UserPerm& perms, Inode **inode) |
7c673cae | 9551 | { |
f67539c2 | 9552 | ldout(cct, 8) << __func__ << " enter(" << vino << ")" << dendl; |
7c673cae | 9553 | |
f67539c2 TL |
9554 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9555 | if (!mref_reader.is_state_satisfied()) | |
9556 | return -CEPHFS_ENOTCONN; | |
181888fb | 9557 | |
b3b6e05e TL |
9558 | if (is_reserved_vino(vino)) |
9559 | return -CEPHFS_ESTALE; | |
9560 | ||
7c673cae | 9561 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPINO); |
f67539c2 | 9562 | filepath path(vino.ino); |
7c673cae FG |
9563 | req->set_filepath(path); |
9564 | ||
f67539c2 TL |
9565 | /* |
9566 | * The MDS expects either a "real" snapid here or 0. The special value | |
9567 | * carveouts for the snapid are all at the end of the range so we can | |
9568 | * just look for any snapid below this value. | |
9569 | */ | |
9570 | if (vino.snapid < CEPH_NOSNAP) | |
9571 | req->head.args.lookupino.snapid = vino.snapid; | |
9572 | ||
7c673cae FG |
9573 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); |
9574 | if (r == 0 && inode != NULL) { | |
7c673cae | 9575 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
11fdf7f2 | 9576 | ceph_assert(p != inode_map.end()); |
7c673cae FG |
9577 | *inode = p->second; |
9578 | _ll_get(*inode); | |
9579 | } | |
f67539c2 | 9580 | ldout(cct, 8) << __func__ << " exit(" << vino << ") = " << r << dendl; |
7c673cae FG |
9581 | return r; |
9582 | } | |
9583 | ||
1adf2230 AA |
9584 | int Client::lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode) |
9585 | { | |
f67539c2 TL |
9586 | vinodeno_t vino(ino, CEPH_NOSNAP); |
9587 | std::scoped_lock lock(client_lock); | |
9588 | return _lookup_vino(vino, perms, inode); | |
1adf2230 | 9589 | } |
7c673cae FG |
9590 | |
9591 | /** | |
9592 | * Find the parent inode of `ino` and insert it into | |
9593 | * our cache. Conditionally also set `parent` to a referenced | |
9594 | * Inode* if caller provides non-NULL value. | |
9595 | */ | |
1adf2230 | 9596 | int Client::_lookup_parent(Inode *ino, const UserPerm& perms, Inode **parent) |
7c673cae | 9597 | { |
11fdf7f2 | 9598 | ldout(cct, 8) << __func__ << " enter(" << ino->ino << ")" << dendl; |
7c673cae | 9599 | |
7c673cae FG |
9600 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); |
9601 | filepath path(ino->ino); | |
9602 | req->set_filepath(path); | |
9603 | ||
9604 | InodeRef target; | |
9605 | int r = make_request(req, perms, &target, NULL, rand() % mdsmap->get_num_in_mds()); | |
9606 | // Give caller a reference to the parent ino if they provided a pointer. | |
9607 | if (parent != NULL) { | |
9608 | if (r == 0) { | |
9609 | *parent = target.get(); | |
9610 | _ll_get(*parent); | |
11fdf7f2 | 9611 | ldout(cct, 8) << __func__ << " found parent " << (*parent)->ino << dendl; |
7c673cae FG |
9612 | } else { |
9613 | *parent = NULL; | |
9614 | } | |
9615 | } | |
11fdf7f2 | 9616 | ldout(cct, 8) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9617 | return r; |
9618 | } | |
9619 | ||
7c673cae FG |
9620 | /** |
9621 | * Populate the parent dentry for `ino`, provided it is | |
9622 | * a child of `parent`. | |
9623 | */ | |
1adf2230 | 9624 | int Client::_lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
7c673cae | 9625 | { |
11fdf7f2 TL |
9626 | ceph_assert(parent->is_dir()); |
9627 | ldout(cct, 3) << __func__ << " enter(" << ino->ino << ")" << dendl; | |
7c673cae | 9628 | |
f67539c2 TL |
9629 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9630 | if (!mref_reader.is_state_satisfied()) | |
9631 | return -CEPHFS_ENOTCONN; | |
181888fb | 9632 | |
7c673cae FG |
9633 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
9634 | req->set_filepath2(filepath(parent->ino)); | |
9635 | req->set_filepath(filepath(ino->ino)); | |
9636 | req->set_inode(ino); | |
9637 | ||
9638 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9639 | ldout(cct, 3) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9640 | return r; |
9641 | } | |
9642 | ||
1adf2230 AA |
9643 | int Client::lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
9644 | { | |
f67539c2 | 9645 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
9646 | return _lookup_name(ino, parent, perms); |
9647 | } | |
7c673cae | 9648 | |
11fdf7f2 | 9649 | Fh *Client::_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms) |
7c673cae | 9650 | { |
11fdf7f2 | 9651 | ceph_assert(in); |
f6b5b4d7 | 9652 | Fh *f = new Fh(in, flags, cmode, fd_gen, perms); |
7c673cae | 9653 | |
11fdf7f2 | 9654 | ldout(cct, 10) << __func__ << " " << in->ino << " mode " << cmode << dendl; |
7c673cae FG |
9655 | |
9656 | if (in->snapid != CEPH_NOSNAP) { | |
9657 | in->snap_cap_refs++; | |
9658 | ldout(cct, 5) << "open success, fh is " << f << " combined IMMUTABLE SNAP caps " | |
9659 | << ccap_string(in->caps_issued()) << dendl; | |
9660 | } | |
9661 | ||
11fdf7f2 | 9662 | const auto& conf = cct->_conf; |
7c673cae FG |
9663 | f->readahead.set_trigger_requests(1); |
9664 | f->readahead.set_min_readahead_size(conf->client_readahead_min); | |
9665 | uint64_t max_readahead = Readahead::NO_LIMIT; | |
9666 | if (conf->client_readahead_max_bytes) { | |
11fdf7f2 | 9667 | max_readahead = std::min(max_readahead, (uint64_t)conf->client_readahead_max_bytes); |
7c673cae FG |
9668 | } |
9669 | if (conf->client_readahead_max_periods) { | |
11fdf7f2 | 9670 | max_readahead = std::min(max_readahead, in->layout.get_period()*(uint64_t)conf->client_readahead_max_periods); |
7c673cae FG |
9671 | } |
9672 | f->readahead.set_max_readahead_size(max_readahead); | |
9673 | vector<uint64_t> alignments; | |
9674 | alignments.push_back(in->layout.get_period()); | |
9675 | alignments.push_back(in->layout.stripe_unit); | |
9676 | f->readahead.set_alignments(alignments); | |
9677 | ||
9678 | return f; | |
9679 | } | |
9680 | ||
9681 | int Client::_release_fh(Fh *f) | |
9682 | { | |
9683 | //ldout(cct, 3) << "op: client->close(open_files[ " << fh << " ]);" << dendl; | |
9684 | //ldout(cct, 3) << "op: open_files.erase( " << fh << " );" << dendl; | |
9685 | Inode *in = f->inode.get(); | |
11fdf7f2 | 9686 | ldout(cct, 8) << __func__ << " " << f << " mode " << f->mode << " on " << *in << dendl; |
7c673cae | 9687 | |
b32b8144 FG |
9688 | in->unset_deleg(f); |
9689 | ||
7c673cae FG |
9690 | if (in->snapid == CEPH_NOSNAP) { |
9691 | if (in->put_open_ref(f->mode)) { | |
9692 | _flush(in, new C_Client_FlushComplete(this, in)); | |
9693 | check_caps(in, 0); | |
9694 | } | |
9695 | } else { | |
11fdf7f2 | 9696 | ceph_assert(in->snap_cap_refs > 0); |
7c673cae FG |
9697 | in->snap_cap_refs--; |
9698 | } | |
9699 | ||
9700 | _release_filelocks(f); | |
9701 | ||
9702 | // Finally, read any async err (i.e. from flushes) | |
9703 | int err = f->take_async_err(); | |
9704 | if (err != 0) { | |
11fdf7f2 | 9705 | ldout(cct, 1) << __func__ << " " << f << " on inode " << *in << " caught async_err = " |
7c673cae FG |
9706 | << cpp_strerror(err) << dendl; |
9707 | } else { | |
11fdf7f2 | 9708 | ldout(cct, 10) << __func__ << " " << f << " on inode " << *in << " no async_err state" << dendl; |
7c673cae FG |
9709 | } |
9710 | ||
9711 | _put_fh(f); | |
9712 | ||
9713 | return err; | |
9714 | } | |
9715 | ||
9716 | void Client::_put_fh(Fh *f) | |
9717 | { | |
9718 | int left = f->put(); | |
9719 | if (!left) { | |
9720 | delete f; | |
9721 | } | |
9722 | } | |
9723 | ||
9724 | int Client::_open(Inode *in, int flags, mode_t mode, Fh **fhp, | |
9725 | const UserPerm& perms) | |
9726 | { | |
9727 | if (in->snapid != CEPH_NOSNAP && | |
9728 | (flags & (O_WRONLY | O_RDWR | O_CREAT | O_TRUNC | O_APPEND))) { | |
f67539c2 | 9729 | return -CEPHFS_EROFS; |
7c673cae FG |
9730 | } |
9731 | ||
9732 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
9733 | int cflags = ceph_flags_sys2wire(flags); |
9734 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
9735 | cflags |= CEPH_O_LAZY; | |
9736 | ||
9737 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
9738 | int want = ceph_caps_for_mode(cmode); |
9739 | int result = 0; | |
9740 | ||
9741 | in->get_open_ref(cmode); // make note of pending open, since it effects _wanted_ caps. | |
9742 | ||
b32b8144 | 9743 | if ((flags & O_TRUNC) == 0 && in->caps_issued_mask(want)) { |
7c673cae FG |
9744 | // update wanted? |
9745 | check_caps(in, CHECK_CAPS_NODELAY); | |
9746 | } else { | |
b32b8144 | 9747 | |
7c673cae FG |
9748 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); |
9749 | filepath path; | |
9750 | in->make_nosnap_relative_path(path); | |
9751 | req->set_filepath(path); | |
11fdf7f2 | 9752 | req->head.args.open.flags = cflags & ~CEPH_O_CREAT; |
7c673cae FG |
9753 | req->head.args.open.mode = mode; |
9754 | req->head.args.open.pool = -1; | |
9755 | if (cct->_conf->client_debug_getattr_caps) | |
9756 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9757 | else | |
9758 | req->head.args.open.mask = 0; | |
9759 | req->head.args.open.old_size = in->size; // for O_TRUNC | |
9760 | req->set_inode(in); | |
9761 | result = make_request(req, perms); | |
b32b8144 FG |
9762 | |
9763 | /* | |
9764 | * NFS expects that delegations will be broken on a conflicting open, | |
9765 | * not just when there is actual conflicting access to the file. SMB leases | |
9766 | * and oplocks also have similar semantics. | |
9767 | * | |
9768 | * Ensure that clients that have delegations enabled will wait on minimal | |
9769 | * caps during open, just to ensure that other clients holding delegations | |
9770 | * return theirs first. | |
9771 | */ | |
9772 | if (deleg_timeout && result == 0) { | |
9773 | int need = 0, have; | |
9774 | ||
9775 | if (cmode & CEPH_FILE_MODE_WR) | |
9776 | need |= CEPH_CAP_FILE_WR; | |
9777 | if (cmode & CEPH_FILE_MODE_RD) | |
9778 | need |= CEPH_CAP_FILE_RD; | |
9779 | ||
f6b5b4d7 TL |
9780 | Fh fh(in, flags, cmode, fd_gen, perms); |
9781 | result = get_caps(&fh, need, want, &have, -1); | |
b32b8144 | 9782 | if (result < 0) { |
1adf2230 | 9783 | ldout(cct, 8) << "Unable to get caps after open of inode " << *in << |
b32b8144 FG |
9784 | " . Denying open: " << |
9785 | cpp_strerror(result) << dendl; | |
b32b8144 FG |
9786 | } else { |
9787 | put_cap_ref(in, need); | |
9788 | } | |
9789 | } | |
7c673cae FG |
9790 | } |
9791 | ||
9792 | // success? | |
9793 | if (result >= 0) { | |
9794 | if (fhp) | |
9795 | *fhp = _create_fh(in, flags, cmode, perms); | |
9796 | } else { | |
9797 | in->put_open_ref(cmode); | |
9798 | } | |
9799 | ||
9800 | trim_cache(); | |
9801 | ||
9802 | return result; | |
9803 | } | |
9804 | ||
9805 | int Client::_renew_caps(Inode *in) | |
9806 | { | |
9807 | int wanted = in->caps_file_wanted(); | |
9808 | if (in->is_any_caps() && | |
9809 | ((wanted & CEPH_CAP_ANY_WR) == 0 || in->auth_cap)) { | |
9810 | check_caps(in, CHECK_CAPS_NODELAY); | |
9811 | return 0; | |
9812 | } | |
9813 | ||
9814 | int flags = 0; | |
9815 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
9816 | flags = O_RDWR; | |
9817 | else if (wanted & CEPH_CAP_FILE_RD) | |
9818 | flags = O_RDONLY; | |
9819 | else if (wanted & CEPH_CAP_FILE_WR) | |
9820 | flags = O_WRONLY; | |
9821 | ||
9822 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); | |
9823 | filepath path; | |
9824 | in->make_nosnap_relative_path(path); | |
9825 | req->set_filepath(path); | |
9826 | req->head.args.open.flags = flags; | |
9827 | req->head.args.open.pool = -1; | |
9828 | if (cct->_conf->client_debug_getattr_caps) | |
9829 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9830 | else | |
9831 | req->head.args.open.mask = 0; | |
9832 | req->set_inode(in); | |
9833 | ||
9834 | // duplicate in case Cap goes away; not sure if that race is a concern? | |
9835 | const UserPerm *pperm = in->get_best_perms(); | |
9836 | UserPerm perms; | |
9837 | if (pperm != NULL) | |
9838 | perms = *pperm; | |
9839 | int ret = make_request(req, perms); | |
9840 | return ret; | |
9841 | } | |
9842 | ||
b3b6e05e | 9843 | int Client::_close(int fd) |
7c673cae FG |
9844 | { |
9845 | ldout(cct, 3) << "close enter(" << fd << ")" << dendl; | |
7c673cae FG |
9846 | tout(cct) << "close" << std::endl; |
9847 | tout(cct) << fd << std::endl; | |
9848 | ||
9849 | Fh *fh = get_filehandle(fd); | |
9850 | if (!fh) | |
f67539c2 | 9851 | return -CEPHFS_EBADF; |
7c673cae FG |
9852 | int err = _release_fh(fh); |
9853 | fd_map.erase(fd); | |
9854 | put_fd(fd); | |
9855 | ldout(cct, 3) << "close exit(" << fd << ")" << dendl; | |
9856 | return err; | |
9857 | } | |
9858 | ||
b3b6e05e TL |
9859 | int Client::close(int fd) { |
9860 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9861 | if (!mref_reader.is_state_satisfied()) | |
9862 | return -CEPHFS_ENOTCONN; | |
9863 | ||
9864 | std::scoped_lock lock(client_lock); | |
9865 | return _close(fd); | |
9866 | } | |
7c673cae FG |
9867 | |
9868 | // ------------ | |
9869 | // read, write | |
9870 | ||
9871 | loff_t Client::lseek(int fd, loff_t offset, int whence) | |
9872 | { | |
f67539c2 TL |
9873 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9874 | if (!mref_reader.is_state_satisfied()) | |
9875 | return -CEPHFS_ENOTCONN; | |
9876 | ||
7c673cae FG |
9877 | tout(cct) << "lseek" << std::endl; |
9878 | tout(cct) << fd << std::endl; | |
9879 | tout(cct) << offset << std::endl; | |
9880 | tout(cct) << whence << std::endl; | |
9881 | ||
f67539c2 | 9882 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9883 | Fh *f = get_filehandle(fd); |
9884 | if (!f) | |
f67539c2 | 9885 | return -CEPHFS_EBADF; |
7c673cae FG |
9886 | #if defined(__linux__) && defined(O_PATH) |
9887 | if (f->flags & O_PATH) | |
f67539c2 | 9888 | return -CEPHFS_EBADF; |
7c673cae FG |
9889 | #endif |
9890 | return _lseek(f, offset, whence); | |
9891 | } | |
9892 | ||
9893 | loff_t Client::_lseek(Fh *f, loff_t offset, int whence) | |
9894 | { | |
9895 | Inode *in = f->inode.get(); | |
9f95a23c | 9896 | bool whence_check = false; |
11fdf7f2 | 9897 | loff_t pos = -1; |
7c673cae | 9898 | |
9f95a23c TL |
9899 | switch (whence) { |
9900 | case SEEK_END: | |
9901 | whence_check = true; | |
9902 | break; | |
9903 | ||
9904 | #ifdef SEEK_DATA | |
9905 | case SEEK_DATA: | |
9906 | whence_check = true; | |
9907 | break; | |
9908 | #endif | |
9909 | ||
9910 | #ifdef SEEK_HOLE | |
9911 | case SEEK_HOLE: | |
9912 | whence_check = true; | |
9913 | break; | |
9914 | #endif | |
9915 | } | |
9916 | ||
9917 | if (whence_check) { | |
9918 | int r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
9919 | if (r < 0) | |
92f5a8d4 | 9920 | return r; |
92f5a8d4 TL |
9921 | } |
9922 | ||
7c673cae FG |
9923 | switch (whence) { |
9924 | case SEEK_SET: | |
11fdf7f2 | 9925 | pos = offset; |
7c673cae FG |
9926 | break; |
9927 | ||
9928 | case SEEK_CUR: | |
92f5a8d4 | 9929 | pos = f->pos + offset; |
7c673cae FG |
9930 | break; |
9931 | ||
9932 | case SEEK_END: | |
11fdf7f2 | 9933 | pos = in->size + offset; |
7c673cae FG |
9934 | break; |
9935 | ||
9f95a23c | 9936 | #ifdef SEEK_DATA |
92f5a8d4 | 9937 | case SEEK_DATA: |
9f95a23c | 9938 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9939 | return -CEPHFS_ENXIO; |
92f5a8d4 TL |
9940 | pos = offset; |
9941 | break; | |
9f95a23c | 9942 | #endif |
92f5a8d4 | 9943 | |
9f95a23c | 9944 | #ifdef SEEK_HOLE |
92f5a8d4 | 9945 | case SEEK_HOLE: |
9f95a23c | 9946 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9947 | return -CEPHFS_ENXIO; |
9f95a23c | 9948 | pos = in->size; |
92f5a8d4 | 9949 | break; |
9f95a23c | 9950 | #endif |
92f5a8d4 | 9951 | |
7c673cae | 9952 | default: |
92f5a8d4 | 9953 | ldout(cct, 1) << __func__ << ": invalid whence value " << whence << dendl; |
f67539c2 | 9954 | return -CEPHFS_EINVAL; |
7c673cae FG |
9955 | } |
9956 | ||
11fdf7f2 | 9957 | if (pos < 0) { |
f67539c2 | 9958 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
9959 | } else { |
9960 | f->pos = pos; | |
9961 | } | |
9962 | ||
1adf2230 | 9963 | ldout(cct, 8) << "_lseek(" << f << ", " << offset << ", " << whence << ") = " << f->pos << dendl; |
7c673cae FG |
9964 | return f->pos; |
9965 | } | |
9966 | ||
9967 | ||
9968 | void Client::lock_fh_pos(Fh *f) | |
9969 | { | |
11fdf7f2 | 9970 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae FG |
9971 | |
9972 | if (f->pos_locked || !f->pos_waiters.empty()) { | |
9f95a23c | 9973 | ceph::condition_variable cond; |
7c673cae | 9974 | f->pos_waiters.push_back(&cond); |
11fdf7f2 | 9975 | ldout(cct, 10) << __func__ << " BLOCKING on " << f << dendl; |
9f95a23c TL |
9976 | std::unique_lock l{client_lock, std::adopt_lock}; |
9977 | cond.wait(l, [f, me=&cond] { | |
9978 | return !f->pos_locked && f->pos_waiters.front() == me; | |
9979 | }); | |
9980 | l.release(); | |
11fdf7f2 TL |
9981 | ldout(cct, 10) << __func__ << " UNBLOCKING on " << f << dendl; |
9982 | ceph_assert(f->pos_waiters.front() == &cond); | |
7c673cae FG |
9983 | f->pos_waiters.pop_front(); |
9984 | } | |
9985 | ||
9986 | f->pos_locked = true; | |
9987 | } | |
9988 | ||
9989 | void Client::unlock_fh_pos(Fh *f) | |
9990 | { | |
f67539c2 TL |
9991 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9992 | ||
11fdf7f2 | 9993 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae | 9994 | f->pos_locked = false; |
f67539c2 TL |
9995 | if (!f->pos_waiters.empty()) { |
9996 | // only wake up the oldest waiter | |
9997 | auto cond = f->pos_waiters.front(); | |
9998 | cond->notify_one(); | |
9999 | } | |
7c673cae FG |
10000 | } |
10001 | ||
10002 | int Client::uninline_data(Inode *in, Context *onfinish) | |
10003 | { | |
10004 | if (!in->inline_data.length()) { | |
10005 | onfinish->complete(0); | |
10006 | return 0; | |
10007 | } | |
10008 | ||
10009 | char oid_buf[32]; | |
10010 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (long long unsigned)in->ino); | |
10011 | object_t oid = oid_buf; | |
10012 | ||
10013 | ObjectOperation create_ops; | |
10014 | create_ops.create(false); | |
10015 | ||
10016 | objecter->mutate(oid, | |
10017 | OSDMap::file_to_object_locator(in->layout), | |
10018 | create_ops, | |
10019 | in->snaprealm->get_snap_context(), | |
10020 | ceph::real_clock::now(), | |
10021 | 0, | |
10022 | NULL); | |
10023 | ||
10024 | bufferlist inline_version_bl; | |
11fdf7f2 | 10025 | encode(in->inline_version, inline_version_bl); |
7c673cae FG |
10026 | |
10027 | ObjectOperation uninline_ops; | |
10028 | uninline_ops.cmpxattr("inline_version", | |
10029 | CEPH_OSD_CMPXATTR_OP_GT, | |
10030 | CEPH_OSD_CMPXATTR_MODE_U64, | |
10031 | inline_version_bl); | |
10032 | bufferlist inline_data = in->inline_data; | |
10033 | uninline_ops.write(0, inline_data, in->truncate_size, in->truncate_seq); | |
10034 | uninline_ops.setxattr("inline_version", stringify(in->inline_version)); | |
10035 | ||
10036 | objecter->mutate(oid, | |
10037 | OSDMap::file_to_object_locator(in->layout), | |
10038 | uninline_ops, | |
10039 | in->snaprealm->get_snap_context(), | |
10040 | ceph::real_clock::now(), | |
10041 | 0, | |
10042 | onfinish); | |
10043 | ||
10044 | return 0; | |
10045 | } | |
10046 | ||
10047 | // | |
10048 | ||
10049 | // blocking osd interface | |
10050 | ||
10051 | int Client::read(int fd, char *buf, loff_t size, loff_t offset) | |
10052 | { | |
f67539c2 TL |
10053 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10054 | if (!mref_reader.is_state_satisfied()) | |
10055 | return -CEPHFS_ENOTCONN; | |
10056 | ||
7c673cae FG |
10057 | tout(cct) << "read" << std::endl; |
10058 | tout(cct) << fd << std::endl; | |
10059 | tout(cct) << size << std::endl; | |
10060 | tout(cct) << offset << std::endl; | |
10061 | ||
f67539c2 | 10062 | std::unique_lock lock(client_lock); |
7c673cae FG |
10063 | Fh *f = get_filehandle(fd); |
10064 | if (!f) | |
f67539c2 | 10065 | return -CEPHFS_EBADF; |
7c673cae FG |
10066 | #if defined(__linux__) && defined(O_PATH) |
10067 | if (f->flags & O_PATH) | |
f67539c2 | 10068 | return -CEPHFS_EBADF; |
7c673cae FG |
10069 | #endif |
10070 | bufferlist bl; | |
11fdf7f2 TL |
10071 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
10072 | size = std::min(size, (loff_t)INT_MAX); | |
7c673cae FG |
10073 | int r = _read(f, offset, size, &bl); |
10074 | ldout(cct, 3) << "read(" << fd << ", " << (void*)buf << ", " << size << ", " << offset << ") = " << r << dendl; | |
10075 | if (r >= 0) { | |
f6b5b4d7 | 10076 | lock.unlock(); |
9f95a23c | 10077 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
10078 | r = bl.length(); |
10079 | } | |
10080 | return r; | |
10081 | } | |
10082 | ||
10083 | int Client::preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset) | |
10084 | { | |
10085 | if (iovcnt < 0) | |
f67539c2 | 10086 | return -CEPHFS_EINVAL; |
7c673cae FG |
10087 | return _preadv_pwritev(fd, iov, iovcnt, offset, false); |
10088 | } | |
10089 | ||
11fdf7f2 | 10090 | int64_t Client::_read(Fh *f, int64_t offset, uint64_t size, bufferlist *bl) |
7c673cae | 10091 | { |
f67539c2 TL |
10092 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10093 | ||
11fdf7f2 TL |
10094 | int want, have = 0; |
10095 | bool movepos = false; | |
10096 | std::unique_ptr<C_SaferCond> onuninline; | |
adb31ebb | 10097 | int64_t rc = 0; |
11fdf7f2 | 10098 | const auto& conf = cct->_conf; |
7c673cae | 10099 | Inode *in = f->inode.get(); |
11fdf7f2 TL |
10100 | utime_t lat; |
10101 | utime_t start = ceph_clock_now(); | |
7c673cae FG |
10102 | |
10103 | if ((f->mode & CEPH_FILE_MODE_RD) == 0) | |
f67539c2 | 10104 | return -CEPHFS_EBADF; |
7c673cae FG |
10105 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
10106 | ||
7c673cae FG |
10107 | if (offset < 0) { |
10108 | lock_fh_pos(f); | |
10109 | offset = f->pos; | |
10110 | movepos = true; | |
10111 | } | |
10112 | loff_t start_pos = offset; | |
10113 | ||
10114 | if (in->inline_version == 0) { | |
adb31ebb | 10115 | auto r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); |
c07f9fc5 | 10116 | if (r < 0) { |
adb31ebb | 10117 | rc = r; |
11fdf7f2 | 10118 | goto done; |
c07f9fc5 | 10119 | } |
11fdf7f2 | 10120 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
10121 | } |
10122 | ||
10123 | retry: | |
11fdf7f2 TL |
10124 | if (f->mode & CEPH_FILE_MODE_LAZY) |
10125 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
10126 | else | |
10127 | want = CEPH_CAP_FILE_CACHE; | |
adb31ebb TL |
10128 | { |
10129 | auto r = get_caps(f, CEPH_CAP_FILE_RD, want, &have, -1); | |
10130 | if (r < 0) { | |
10131 | rc = r; | |
10132 | goto done; | |
10133 | } | |
c07f9fc5 | 10134 | } |
7c673cae | 10135 | if (f->flags & O_DIRECT) |
11fdf7f2 | 10136 | have &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
10137 | |
10138 | if (in->inline_version < CEPH_INLINE_NONE) { | |
10139 | if (!(have & CEPH_CAP_FILE_CACHE)) { | |
11fdf7f2 TL |
10140 | onuninline.reset(new C_SaferCond("Client::_read_uninline_data flock")); |
10141 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
10142 | } else { |
10143 | uint32_t len = in->inline_data.length(); | |
7c673cae FG |
10144 | uint64_t endoff = offset + size; |
10145 | if (endoff > in->size) | |
10146 | endoff = in->size; | |
10147 | ||
10148 | if (offset < len) { | |
10149 | if (endoff <= len) { | |
10150 | bl->substr_of(in->inline_data, offset, endoff - offset); | |
10151 | } else { | |
10152 | bl->substr_of(in->inline_data, offset, len - offset); | |
10153 | bl->append_zero(endoff - len); | |
10154 | } | |
adb31ebb | 10155 | rc = endoff - offset; |
7c673cae FG |
10156 | } else if ((uint64_t)offset < endoff) { |
10157 | bl->append_zero(endoff - offset); | |
adb31ebb | 10158 | rc = endoff - offset; |
11fdf7f2 | 10159 | } else { |
adb31ebb | 10160 | rc = 0; |
7c673cae | 10161 | } |
7c673cae FG |
10162 | goto success; |
10163 | } | |
10164 | } | |
10165 | ||
10166 | if (!conf->client_debug_force_sync_read && | |
11fdf7f2 TL |
10167 | conf->client_oc && |
10168 | (have & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
10169 | |
10170 | if (f->flags & O_RSYNC) { | |
10171 | _flush_range(in, offset, size); | |
10172 | } | |
adb31ebb TL |
10173 | rc = _read_async(f, offset, size, bl); |
10174 | if (rc < 0) | |
7c673cae FG |
10175 | goto done; |
10176 | } else { | |
10177 | if (f->flags & O_DIRECT) | |
10178 | _flush_range(in, offset, size); | |
10179 | ||
10180 | bool checkeof = false; | |
adb31ebb TL |
10181 | rc = _read_sync(f, offset, size, bl, &checkeof); |
10182 | if (rc < 0) | |
7c673cae FG |
10183 | goto done; |
10184 | if (checkeof) { | |
adb31ebb TL |
10185 | offset += rc; |
10186 | size -= rc; | |
7c673cae FG |
10187 | |
10188 | put_cap_ref(in, CEPH_CAP_FILE_RD); | |
10189 | have = 0; | |
10190 | // reverify size | |
adb31ebb TL |
10191 | { |
10192 | auto r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
10193 | if (r < 0) { | |
10194 | rc = r; | |
10195 | goto done; | |
10196 | } | |
10197 | } | |
7c673cae FG |
10198 | |
10199 | // eof? short read. | |
10200 | if ((uint64_t)offset < in->size) | |
10201 | goto retry; | |
10202 | } | |
10203 | } | |
10204 | ||
10205 | success: | |
adb31ebb | 10206 | ceph_assert(rc >= 0); |
a4b75251 | 10207 | update_read_io_size(bl->length()); |
7c673cae FG |
10208 | if (movepos) { |
10209 | // adjust fd pos | |
adb31ebb | 10210 | f->pos = start_pos + rc; |
7c673cae | 10211 | } |
11fdf7f2 TL |
10212 | |
10213 | lat = ceph_clock_now(); | |
10214 | lat -= start; | |
2a845540 TL |
10215 | |
10216 | ++nr_read_request; | |
10217 | update_io_stat_read(lat); | |
7c673cae FG |
10218 | |
10219 | done: | |
10220 | // done! | |
11fdf7f2 | 10221 | |
7c673cae | 10222 | if (onuninline) { |
9f95a23c | 10223 | client_lock.unlock(); |
11fdf7f2 | 10224 | int ret = onuninline->wait(); |
9f95a23c | 10225 | client_lock.lock(); |
f67539c2 | 10226 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
10227 | in->inline_data.clear(); |
10228 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 10229 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10230 | check_caps(in, 0); |
10231 | } else | |
adb31ebb | 10232 | rc = ret; |
7c673cae | 10233 | } |
11fdf7f2 | 10234 | if (have) { |
7c673cae | 10235 | put_cap_ref(in, CEPH_CAP_FILE_RD); |
11fdf7f2 TL |
10236 | } |
10237 | if (movepos) { | |
10238 | unlock_fh_pos(f); | |
10239 | } | |
adb31ebb | 10240 | return rc; |
7c673cae FG |
10241 | } |
10242 | ||
10243 | Client::C_Readahead::C_Readahead(Client *c, Fh *f) : | |
10244 | client(c), f(f) { | |
10245 | f->get(); | |
10246 | f->readahead.inc_pending(); | |
10247 | } | |
10248 | ||
10249 | Client::C_Readahead::~C_Readahead() { | |
10250 | f->readahead.dec_pending(); | |
10251 | client->_put_fh(f); | |
10252 | } | |
10253 | ||
10254 | void Client::C_Readahead::finish(int r) { | |
10255 | lgeneric_subdout(client->cct, client, 20) << "client." << client->get_nodeid() << " " << "C_Readahead on " << f->inode << dendl; | |
10256 | client->put_cap_ref(f->inode.get(), CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
a4b75251 TL |
10257 | if (r > 0) { |
10258 | client->update_read_io_size(r); | |
10259 | } | |
7c673cae FG |
10260 | } |
10261 | ||
10262 | int Client::_read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl) | |
10263 | { | |
f67539c2 TL |
10264 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10265 | ||
11fdf7f2 | 10266 | const auto& conf = cct->_conf; |
7c673cae FG |
10267 | Inode *in = f->inode.get(); |
10268 | ||
11fdf7f2 | 10269 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
10270 | |
10271 | // trim read based on file size? | |
10272 | if (off >= in->size) | |
10273 | return 0; | |
10274 | if (len == 0) | |
10275 | return 0; | |
10276 | if (off + len > in->size) { | |
10277 | len = in->size - off; | |
10278 | } | |
10279 | ||
10280 | ldout(cct, 10) << " min_bytes=" << f->readahead.get_min_readahead_size() | |
10281 | << " max_bytes=" << f->readahead.get_max_readahead_size() | |
10282 | << " max_periods=" << conf->client_readahead_max_periods << dendl; | |
10283 | ||
10284 | // read (and possibly block) | |
11fdf7f2 TL |
10285 | int r = 0; |
10286 | C_SaferCond onfinish("Client::_read_async flock"); | |
7c673cae | 10287 | r = objectcacher->file_read(&in->oset, &in->layout, in->snapid, |
11fdf7f2 | 10288 | off, len, bl, 0, &onfinish); |
7c673cae FG |
10289 | if (r == 0) { |
10290 | get_cap_ref(in, CEPH_CAP_FILE_CACHE); | |
9f95a23c | 10291 | client_lock.unlock(); |
11fdf7f2 | 10292 | r = onfinish.wait(); |
9f95a23c | 10293 | client_lock.lock(); |
7c673cae | 10294 | put_cap_ref(in, CEPH_CAP_FILE_CACHE); |
a4b75251 | 10295 | update_read_io_size(bl->length()); |
7c673cae FG |
10296 | } |
10297 | ||
10298 | if(f->readahead.get_min_readahead_size() > 0) { | |
10299 | pair<uint64_t, uint64_t> readahead_extent = f->readahead.update(off, len, in->size); | |
10300 | if (readahead_extent.second > 0) { | |
10301 | ldout(cct, 20) << "readahead " << readahead_extent.first << "~" << readahead_extent.second | |
10302 | << " (caller wants " << off << "~" << len << ")" << dendl; | |
10303 | Context *onfinish2 = new C_Readahead(this, f); | |
10304 | int r2 = objectcacher->file_read(&in->oset, &in->layout, in->snapid, | |
10305 | readahead_extent.first, readahead_extent.second, | |
10306 | NULL, 0, onfinish2); | |
10307 | if (r2 == 0) { | |
10308 | ldout(cct, 20) << "readahead initiated, c " << onfinish2 << dendl; | |
10309 | get_cap_ref(in, CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
10310 | } else { | |
10311 | ldout(cct, 20) << "readahead was no-op, already cached" << dendl; | |
10312 | delete onfinish2; | |
10313 | } | |
10314 | } | |
10315 | } | |
10316 | ||
10317 | return r; | |
10318 | } | |
10319 | ||
10320 | int Client::_read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, | |
10321 | bool *checkeof) | |
10322 | { | |
f67539c2 TL |
10323 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10324 | ||
7c673cae FG |
10325 | Inode *in = f->inode.get(); |
10326 | uint64_t pos = off; | |
10327 | int left = len; | |
10328 | int read = 0; | |
10329 | ||
11fdf7f2 | 10330 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae | 10331 | |
f67539c2 TL |
10332 | // 0 success, 1 continue and < 0 error happen. |
10333 | auto wait_and_copy = [&](C_SaferCond &onfinish, bufferlist &tbl, int wanted) { | |
11fdf7f2 | 10334 | int r = onfinish.wait(); |
7c673cae FG |
10335 | |
10336 | // if we get ENOENT from OSD, assume 0 bytes returned | |
f67539c2 | 10337 | if (r == -CEPHFS_ENOENT) |
7c673cae FG |
10338 | r = 0; |
10339 | if (r < 0) | |
10340 | return r; | |
f67539c2 | 10341 | |
7c673cae FG |
10342 | if (tbl.length()) { |
10343 | r = tbl.length(); | |
10344 | ||
10345 | read += r; | |
10346 | pos += r; | |
10347 | left -= r; | |
10348 | bl->claim_append(tbl); | |
10349 | } | |
10350 | // short read? | |
10351 | if (r >= 0 && r < wanted) { | |
10352 | if (pos < in->size) { | |
10353 | // zero up to known EOF | |
10354 | int64_t some = in->size - pos; | |
10355 | if (some > left) | |
10356 | some = left; | |
11fdf7f2 TL |
10357 | auto z = buffer::ptr_node::create(some); |
10358 | z->zero(); | |
10359 | bl->push_back(std::move(z)); | |
7c673cae FG |
10360 | read += some; |
10361 | pos += some; | |
10362 | left -= some; | |
10363 | if (left == 0) | |
f67539c2 | 10364 | return 0; |
7c673cae FG |
10365 | } |
10366 | ||
10367 | *checkeof = true; | |
f67539c2 | 10368 | return 0; |
7c673cae | 10369 | } |
f67539c2 TL |
10370 | return 1; |
10371 | }; | |
7c673cae | 10372 | |
f67539c2 TL |
10373 | while (left > 0) { |
10374 | C_SaferCond onfinish("Client::_read_sync flock"); | |
10375 | bufferlist tbl; | |
7c673cae | 10376 | |
f67539c2 TL |
10377 | int wanted = left; |
10378 | filer->read_trunc(in->ino, &in->layout, in->snapid, | |
10379 | pos, left, &tbl, 0, | |
10380 | in->truncate_size, in->truncate_seq, | |
10381 | &onfinish); | |
10382 | client_lock.unlock(); | |
10383 | int r = wait_and_copy(onfinish, tbl, wanted); | |
10384 | client_lock.lock(); | |
10385 | if (!r) | |
10386 | return read; | |
10387 | if (r < 0) | |
10388 | return r; | |
7c673cae | 10389 | } |
f67539c2 | 10390 | return read; |
7c673cae FG |
10391 | } |
10392 | ||
10393 | int Client::write(int fd, const char *buf, loff_t size, loff_t offset) | |
10394 | { | |
f67539c2 TL |
10395 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10396 | if (!mref_reader.is_state_satisfied()) | |
10397 | return -CEPHFS_ENOTCONN; | |
10398 | ||
7c673cae FG |
10399 | tout(cct) << "write" << std::endl; |
10400 | tout(cct) << fd << std::endl; | |
10401 | tout(cct) << size << std::endl; | |
10402 | tout(cct) << offset << std::endl; | |
10403 | ||
f67539c2 | 10404 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10405 | Fh *fh = get_filehandle(fd); |
10406 | if (!fh) | |
f67539c2 | 10407 | return -CEPHFS_EBADF; |
7c673cae FG |
10408 | #if defined(__linux__) && defined(O_PATH) |
10409 | if (fh->flags & O_PATH) | |
f67539c2 | 10410 | return -CEPHFS_EBADF; |
7c673cae | 10411 | #endif |
11fdf7f2 TL |
10412 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
10413 | size = std::min(size, (loff_t)INT_MAX); | |
10414 | int r = _write(fh, offset, size, buf, NULL, false); | |
7c673cae FG |
10415 | ldout(cct, 3) << "write(" << fd << ", \"...\", " << size << ", " << offset << ") = " << r << dendl; |
10416 | return r; | |
10417 | } | |
10418 | ||
10419 | int Client::pwritev(int fd, const struct iovec *iov, int iovcnt, int64_t offset) | |
10420 | { | |
10421 | if (iovcnt < 0) | |
f67539c2 | 10422 | return -CEPHFS_EINVAL; |
7c673cae FG |
10423 | return _preadv_pwritev(fd, iov, iovcnt, offset, true); |
10424 | } | |
10425 | ||
11fdf7f2 | 10426 | int64_t Client::_preadv_pwritev_locked(Fh *fh, const struct iovec *iov, |
20effc67 TL |
10427 | unsigned iovcnt, int64_t offset, |
10428 | bool write, bool clamp_to_int) | |
7c673cae | 10429 | { |
20effc67 TL |
10430 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10431 | ||
7c673cae FG |
10432 | #if defined(__linux__) && defined(O_PATH) |
10433 | if (fh->flags & O_PATH) | |
f67539c2 | 10434 | return -CEPHFS_EBADF; |
7c673cae FG |
10435 | #endif |
10436 | loff_t totallen = 0; | |
10437 | for (unsigned i = 0; i < iovcnt; i++) { | |
10438 | totallen += iov[i].iov_len; | |
10439 | } | |
11fdf7f2 TL |
10440 | |
10441 | /* | |
10442 | * Some of the API functions take 64-bit size values, but only return | |
10443 | * 32-bit signed integers. Clamp the I/O sizes in those functions so that | |
10444 | * we don't do I/Os larger than the values we can return. | |
10445 | */ | |
10446 | if (clamp_to_int) { | |
10447 | totallen = std::min(totallen, (loff_t)INT_MAX); | |
10448 | } | |
7c673cae | 10449 | if (write) { |
11fdf7f2 TL |
10450 | int64_t w = _write(fh, offset, totallen, NULL, iov, iovcnt); |
10451 | ldout(cct, 3) << "pwritev(" << fh << ", \"...\", " << totallen << ", " << offset << ") = " << w << dendl; | |
7c673cae FG |
10452 | return w; |
10453 | } else { | |
10454 | bufferlist bl; | |
11fdf7f2 TL |
10455 | int64_t r = _read(fh, offset, totallen, &bl); |
10456 | ldout(cct, 3) << "preadv(" << fh << ", " << offset << ") = " << r << dendl; | |
7c673cae FG |
10457 | if (r <= 0) |
10458 | return r; | |
10459 | ||
20effc67 | 10460 | client_lock.unlock(); |
9f95a23c | 10461 | auto iter = bl.cbegin(); |
7c673cae FG |
10462 | for (unsigned j = 0, resid = r; j < iovcnt && resid > 0; j++) { |
10463 | /* | |
f67539c2 TL |
10464 | * This piece of code aims to handle the case that bufferlist |
10465 | * does not have enough data to fill in the iov | |
7c673cae | 10466 | */ |
9f95a23c TL |
10467 | const auto round_size = std::min<unsigned>(resid, iov[j].iov_len); |
10468 | iter.copy(round_size, reinterpret_cast<char*>(iov[j].iov_base)); | |
10469 | resid -= round_size; | |
10470 | /* iter is self-updating */ | |
7c673cae | 10471 | } |
20effc67 | 10472 | client_lock.lock(); |
f67539c2 | 10473 | return r; |
7c673cae FG |
10474 | } |
10475 | } | |
10476 | ||
11fdf7f2 TL |
10477 | int Client::_preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write) |
10478 | { | |
f67539c2 TL |
10479 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10480 | if (!mref_reader.is_state_satisfied()) | |
10481 | return -CEPHFS_ENOTCONN; | |
10482 | ||
11fdf7f2 TL |
10483 | tout(cct) << fd << std::endl; |
10484 | tout(cct) << offset << std::endl; | |
10485 | ||
20effc67 | 10486 | std::scoped_lock cl(client_lock); |
11fdf7f2 TL |
10487 | Fh *fh = get_filehandle(fd); |
10488 | if (!fh) | |
f67539c2 | 10489 | return -CEPHFS_EBADF; |
20effc67 | 10490 | return _preadv_pwritev_locked(fh, iov, iovcnt, offset, write, true); |
11fdf7f2 TL |
10491 | } |
10492 | ||
10493 | int64_t Client::_write(Fh *f, int64_t offset, uint64_t size, const char *buf, | |
10494 | const struct iovec *iov, int iovcnt) | |
7c673cae | 10495 | { |
f67539c2 TL |
10496 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10497 | ||
f64942e4 | 10498 | uint64_t fpos = 0; |
2a845540 | 10499 | Inode *in = f->inode.get(); |
f64942e4 | 10500 | |
2a845540 TL |
10501 | if ( (uint64_t)(offset+size) > mdsmap->get_max_filesize() && //exceeds config |
10502 | (uint64_t)(offset+size) > in->size ) { //exceeds filesize | |
10503 | return -CEPHFS_EFBIG; | |
10504 | } | |
7c673cae | 10505 | //ldout(cct, 7) << "write fh " << fh << " size " << size << " offset " << offset << dendl; |
7c673cae FG |
10506 | |
10507 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
f67539c2 | 10508 | return -CEPHFS_ENOSPC; |
7c673cae FG |
10509 | } |
10510 | ||
11fdf7f2 | 10511 | ceph_assert(in->snapid == CEPH_NOSNAP); |
7c673cae FG |
10512 | |
10513 | // was Fh opened as writeable? | |
10514 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 10515 | return -CEPHFS_EBADF; |
7c673cae | 10516 | |
7c673cae FG |
10517 | // use/adjust fd pos? |
10518 | if (offset < 0) { | |
10519 | lock_fh_pos(f); | |
10520 | /* | |
10521 | * FIXME: this is racy in that we may block _after_ this point waiting for caps, and size may | |
10522 | * change out from under us. | |
10523 | */ | |
10524 | if (f->flags & O_APPEND) { | |
9f95a23c | 10525 | auto r = _lseek(f, 0, SEEK_END); |
7c673cae FG |
10526 | if (r < 0) { |
10527 | unlock_fh_pos(f); | |
10528 | return r; | |
10529 | } | |
10530 | } | |
10531 | offset = f->pos; | |
f64942e4 | 10532 | fpos = offset+size; |
7c673cae FG |
10533 | unlock_fh_pos(f); |
10534 | } | |
10535 | ||
11fdf7f2 TL |
10536 | // check quota |
10537 | uint64_t endoff = offset + size; | |
10538 | if (endoff > in->size && is_quota_bytes_exceeded(in, endoff - in->size, | |
10539 | f->actor_perms)) { | |
f67539c2 | 10540 | return -CEPHFS_EDQUOT; |
11fdf7f2 TL |
10541 | } |
10542 | ||
7c673cae FG |
10543 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
10544 | ||
10545 | ldout(cct, 10) << "cur file size is " << in->size << dendl; | |
10546 | ||
10547 | // time it. | |
10548 | utime_t start = ceph_clock_now(); | |
10549 | ||
10550 | if (in->inline_version == 0) { | |
10551 | int r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); | |
10552 | if (r < 0) | |
10553 | return r; | |
11fdf7f2 | 10554 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
10555 | } |
10556 | ||
10557 | // copy into fresh buffer (since our write may be resub, async) | |
10558 | bufferlist bl; | |
10559 | if (buf) { | |
10560 | if (size > 0) | |
10561 | bl.append(buf, size); | |
10562 | } else if (iov){ | |
10563 | for (int i = 0; i < iovcnt; i++) { | |
10564 | if (iov[i].iov_len > 0) { | |
10565 | bl.append((const char *)iov[i].iov_base, iov[i].iov_len); | |
10566 | } | |
10567 | } | |
10568 | } | |
10569 | ||
10570 | utime_t lat; | |
10571 | uint64_t totalwritten; | |
11fdf7f2 TL |
10572 | int want, have; |
10573 | if (f->mode & CEPH_FILE_MODE_LAZY) | |
10574 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
10575 | else | |
10576 | want = CEPH_CAP_FILE_BUFFER; | |
f6b5b4d7 | 10577 | int r = get_caps(f, CEPH_CAP_FILE_WR|CEPH_CAP_AUTH_SHARED, want, &have, endoff); |
7c673cae FG |
10578 | if (r < 0) |
10579 | return r; | |
10580 | ||
10581 | /* clear the setuid/setgid bits, if any */ | |
181888fb | 10582 | if (unlikely(in->mode & (S_ISUID|S_ISGID)) && size > 0) { |
7c673cae FG |
10583 | struct ceph_statx stx = { 0 }; |
10584 | ||
10585 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10586 | r = __setattrx(in, &stx, CEPH_SETATTR_KILL_SGUID, f->actor_perms); | |
10587 | if (r < 0) | |
10588 | return r; | |
10589 | } else { | |
10590 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10591 | } | |
10592 | ||
10593 | if (f->flags & O_DIRECT) | |
11fdf7f2 | 10594 | have &= ~(CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
10595 | |
10596 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
10597 | ||
11fdf7f2 TL |
10598 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
10599 | ||
7c673cae FG |
10600 | if (in->inline_version < CEPH_INLINE_NONE) { |
10601 | if (endoff > cct->_conf->client_max_inline_size || | |
10602 | endoff > CEPH_INLINE_MAX_SIZE || | |
10603 | !(have & CEPH_CAP_FILE_BUFFER)) { | |
11fdf7f2 TL |
10604 | onuninline.reset(new C_SaferCond("Client::_write_uninline_data flock")); |
10605 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
10606 | } else { |
10607 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10608 | ||
10609 | uint32_t len = in->inline_data.length(); | |
10610 | ||
10611 | if (endoff < len) | |
9f95a23c | 10612 | in->inline_data.begin(endoff).copy(len - endoff, bl); // XXX |
7c673cae FG |
10613 | |
10614 | if (offset < len) | |
10615 | in->inline_data.splice(offset, len - offset); | |
10616 | else if (offset > len) | |
10617 | in->inline_data.append_zero(offset - len); | |
10618 | ||
10619 | in->inline_data.append(bl); | |
10620 | in->inline_version++; | |
10621 | ||
10622 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10623 | ||
10624 | goto success; | |
10625 | } | |
10626 | } | |
10627 | ||
11fdf7f2 TL |
10628 | if (cct->_conf->client_oc && |
10629 | (have & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
10630 | // do buffered write |
10631 | if (!in->oset.dirty_or_tx) | |
10632 | get_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
10633 | ||
10634 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10635 | ||
10636 | // async, caching, non-blocking. | |
10637 | r = objectcacher->file_write(&in->oset, &in->layout, | |
10638 | in->snaprealm->get_snap_context(), | |
10639 | offset, size, bl, ceph::real_clock::now(), | |
10640 | 0); | |
10641 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10642 | ||
10643 | if (r < 0) | |
10644 | goto done; | |
10645 | ||
10646 | // flush cached write if O_SYNC is set on file fh | |
10647 | // O_DSYNC == O_SYNC on linux < 2.6.33 | |
10648 | // O_SYNC = __O_SYNC | O_DSYNC on linux >= 2.6.33 | |
10649 | if ((f->flags & O_SYNC) || (f->flags & O_DSYNC)) { | |
10650 | _flush_range(in, offset, size); | |
10651 | } | |
10652 | } else { | |
10653 | if (f->flags & O_DIRECT) | |
10654 | _flush_range(in, offset, size); | |
10655 | ||
10656 | // simple, non-atomic sync write | |
11fdf7f2 | 10657 | C_SaferCond onfinish("Client::_write flock"); |
f67539c2 | 10658 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
10659 | |
10660 | filer->write_trunc(in->ino, &in->layout, in->snaprealm->get_snap_context(), | |
10661 | offset, size, bl, ceph::real_clock::now(), 0, | |
10662 | in->truncate_size, in->truncate_seq, | |
11fdf7f2 | 10663 | &onfinish); |
9f95a23c | 10664 | client_lock.unlock(); |
f6b5b4d7 | 10665 | r = onfinish.wait(); |
9f95a23c | 10666 | client_lock.lock(); |
f67539c2 | 10667 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
f6b5b4d7 TL |
10668 | if (r < 0) |
10669 | goto done; | |
7c673cae FG |
10670 | } |
10671 | ||
10672 | // if we get here, write was successful, update client metadata | |
10673 | success: | |
a4b75251 | 10674 | update_write_io_size(size); |
7c673cae FG |
10675 | // time |
10676 | lat = ceph_clock_now(); | |
10677 | lat -= start; | |
2a845540 TL |
10678 | |
10679 | ++nr_write_request; | |
10680 | update_io_stat_write(lat); | |
7c673cae | 10681 | |
f64942e4 AA |
10682 | if (fpos) { |
10683 | lock_fh_pos(f); | |
10684 | f->pos = fpos; | |
10685 | unlock_fh_pos(f); | |
10686 | } | |
7c673cae | 10687 | totalwritten = size; |
11fdf7f2 | 10688 | r = (int64_t)totalwritten; |
7c673cae FG |
10689 | |
10690 | // extend file? | |
10691 | if (totalwritten + offset > in->size) { | |
10692 | in->size = totalwritten + offset; | |
28e407b8 | 10693 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 10694 | |
11fdf7f2 | 10695 | if (is_quota_bytes_approaching(in, f->actor_perms)) { |
7c673cae | 10696 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
10697 | } else if (is_max_size_approaching(in)) { |
10698 | check_caps(in, 0); | |
7c673cae FG |
10699 | } |
10700 | ||
10701 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", extending file size" << dendl; | |
10702 | } else { | |
10703 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", leaving file size at " << in->size << dendl; | |
10704 | } | |
10705 | ||
10706 | // mtime | |
91327a77 | 10707 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 10708 | in->change_attr++; |
28e407b8 | 10709 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10710 | |
10711 | done: | |
10712 | ||
11fdf7f2 | 10713 | if (nullptr != onuninline) { |
9f95a23c | 10714 | client_lock.unlock(); |
11fdf7f2 | 10715 | int uninline_ret = onuninline->wait(); |
9f95a23c | 10716 | client_lock.lock(); |
7c673cae | 10717 | |
f67539c2 | 10718 | if (uninline_ret >= 0 || uninline_ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
10719 | in->inline_data.clear(); |
10720 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 10721 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10722 | check_caps(in, 0); |
10723 | } else | |
10724 | r = uninline_ret; | |
10725 | } | |
10726 | ||
10727 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
10728 | return r; | |
10729 | } | |
10730 | ||
10731 | int Client::_flush(Fh *f) | |
10732 | { | |
10733 | Inode *in = f->inode.get(); | |
10734 | int err = f->take_async_err(); | |
10735 | if (err != 0) { | |
10736 | ldout(cct, 1) << __func__ << ": " << f << " on inode " << *in << " caught async_err = " | |
10737 | << cpp_strerror(err) << dendl; | |
10738 | } else { | |
10739 | ldout(cct, 10) << __func__ << ": " << f << " on inode " << *in << " no async_err state" << dendl; | |
10740 | } | |
10741 | ||
10742 | return err; | |
10743 | } | |
10744 | ||
10745 | int Client::truncate(const char *relpath, loff_t length, const UserPerm& perms) | |
10746 | { | |
10747 | struct ceph_statx stx; | |
10748 | stx.stx_size = length; | |
10749 | return setattrx(relpath, &stx, CEPH_SETATTR_SIZE, perms); | |
10750 | } | |
10751 | ||
10752 | int Client::ftruncate(int fd, loff_t length, const UserPerm& perms) | |
10753 | { | |
f67539c2 TL |
10754 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10755 | if (!mref_reader.is_state_satisfied()) | |
10756 | return -CEPHFS_ENOTCONN; | |
10757 | ||
11fdf7f2 | 10758 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
10759 | tout(cct) << fd << std::endl; |
10760 | tout(cct) << length << std::endl; | |
10761 | ||
f67539c2 | 10762 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10763 | Fh *f = get_filehandle(fd); |
10764 | if (!f) | |
f67539c2 | 10765 | return -CEPHFS_EBADF; |
7c673cae FG |
10766 | #if defined(__linux__) && defined(O_PATH) |
10767 | if (f->flags & O_PATH) | |
f67539c2 | 10768 | return -CEPHFS_EBADF; |
7c673cae | 10769 | #endif |
adb31ebb | 10770 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) |
f67539c2 | 10771 | return -CEPHFS_EBADF; |
7c673cae FG |
10772 | struct stat attr; |
10773 | attr.st_size = length; | |
10774 | return _setattr(f->inode, &attr, CEPH_SETATTR_SIZE, perms); | |
10775 | } | |
10776 | ||
10777 | int Client::fsync(int fd, bool syncdataonly) | |
10778 | { | |
f67539c2 TL |
10779 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10780 | if (!mref_reader.is_state_satisfied()) | |
10781 | return -CEPHFS_ENOTCONN; | |
10782 | ||
7c673cae FG |
10783 | tout(cct) << "fsync" << std::endl; |
10784 | tout(cct) << fd << std::endl; | |
10785 | tout(cct) << syncdataonly << std::endl; | |
10786 | ||
f67539c2 | 10787 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10788 | Fh *f = get_filehandle(fd); |
10789 | if (!f) | |
f67539c2 | 10790 | return -CEPHFS_EBADF; |
7c673cae FG |
10791 | #if defined(__linux__) && defined(O_PATH) |
10792 | if (f->flags & O_PATH) | |
f67539c2 | 10793 | return -CEPHFS_EBADF; |
7c673cae FG |
10794 | #endif |
10795 | int r = _fsync(f, syncdataonly); | |
10796 | if (r == 0) { | |
10797 | // The IOs in this fsync were okay, but maybe something happened | |
10798 | // in the background that we shoudl be reporting? | |
10799 | r = f->take_async_err(); | |
1adf2230 | 10800 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly |
7c673cae FG |
10801 | << ") = 0, async_err = " << r << dendl; |
10802 | } else { | |
10803 | // Assume that an error we encountered during fsync, even reported | |
10804 | // synchronously, would also have applied the error to the Fh, and we | |
10805 | // should clear it here to avoid returning the same error again on next | |
10806 | // call. | |
1adf2230 | 10807 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly << ") = " |
7c673cae FG |
10808 | << r << dendl; |
10809 | f->take_async_err(); | |
10810 | } | |
10811 | return r; | |
10812 | } | |
10813 | ||
10814 | int Client::_fsync(Inode *in, bool syncdataonly) | |
10815 | { | |
f67539c2 TL |
10816 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10817 | ||
7c673cae | 10818 | int r = 0; |
11fdf7f2 | 10819 | std::unique_ptr<C_SaferCond> object_cacher_completion = nullptr; |
7c673cae FG |
10820 | ceph_tid_t flush_tid = 0; |
10821 | InodeRef tmp_ref; | |
11fdf7f2 TL |
10822 | utime_t lat; |
10823 | utime_t start = ceph_clock_now(); | |
7c673cae | 10824 | |
1adf2230 | 10825 | ldout(cct, 8) << "_fsync on " << *in << " " << (syncdataonly ? "(dataonly)":"(data+metadata)") << dendl; |
7c673cae FG |
10826 | |
10827 | if (cct->_conf->client_oc) { | |
11fdf7f2 TL |
10828 | object_cacher_completion.reset(new C_SaferCond("Client::_fsync::lock")); |
10829 | tmp_ref = in; // take a reference; C_SaferCond doesn't and _flush won't either | |
10830 | _flush(in, object_cacher_completion.get()); | |
7c673cae FG |
10831 | ldout(cct, 15) << "using return-valued form of _fsync" << dendl; |
10832 | } | |
10833 | ||
10834 | if (!syncdataonly && in->dirty_caps) { | |
10835 | check_caps(in, CHECK_CAPS_NODELAY|CHECK_CAPS_SYNCHRONOUS); | |
10836 | if (in->flushing_caps) | |
10837 | flush_tid = last_flush_tid; | |
10838 | } else ldout(cct, 10) << "no metadata needs to commit" << dendl; | |
10839 | ||
10840 | if (!syncdataonly && !in->unsafe_ops.empty()) { | |
522d829b | 10841 | flush_mdlog_sync(in); |
28e407b8 | 10842 | |
7c673cae FG |
10843 | MetaRequest *req = in->unsafe_ops.back(); |
10844 | ldout(cct, 15) << "waiting on unsafe requests, last tid " << req->get_tid() << dendl; | |
10845 | ||
10846 | req->get(); | |
10847 | wait_on_list(req->waitfor_safe); | |
10848 | put_request(req); | |
10849 | } | |
10850 | ||
11fdf7f2 | 10851 | if (nullptr != object_cacher_completion) { // wait on a real reply instead of guessing |
9f95a23c | 10852 | client_lock.unlock(); |
7c673cae | 10853 | ldout(cct, 15) << "waiting on data to flush" << dendl; |
11fdf7f2 | 10854 | r = object_cacher_completion->wait(); |
9f95a23c | 10855 | client_lock.lock(); |
7c673cae FG |
10856 | ldout(cct, 15) << "got " << r << " from flush writeback" << dendl; |
10857 | } else { | |
10858 | // FIXME: this can starve | |
10859 | while (in->cap_refs[CEPH_CAP_FILE_BUFFER] > 0) { | |
10860 | ldout(cct, 10) << "ino " << in->ino << " has " << in->cap_refs[CEPH_CAP_FILE_BUFFER] | |
10861 | << " uncommitted, waiting" << dendl; | |
10862 | wait_on_list(in->waitfor_commit); | |
10863 | } | |
10864 | } | |
10865 | ||
10866 | if (!r) { | |
10867 | if (flush_tid > 0) | |
10868 | wait_sync_caps(in, flush_tid); | |
10869 | ||
10870 | ldout(cct, 10) << "ino " << in->ino << " has no uncommitted writes" << dendl; | |
10871 | } else { | |
1adf2230 | 10872 | ldout(cct, 8) << "ino " << in->ino << " failed to commit to disk! " |
7c673cae FG |
10873 | << cpp_strerror(-r) << dendl; |
10874 | } | |
11fdf7f2 TL |
10875 | |
10876 | lat = ceph_clock_now(); | |
10877 | lat -= start; | |
10878 | logger->tinc(l_c_fsync, lat); | |
7c673cae FG |
10879 | |
10880 | return r; | |
10881 | } | |
10882 | ||
10883 | int Client::_fsync(Fh *f, bool syncdataonly) | |
10884 | { | |
1adf2230 | 10885 | ldout(cct, 8) << "_fsync(" << f << ", " << (syncdataonly ? "dataonly)":"data+metadata)") << dendl; |
7c673cae FG |
10886 | return _fsync(f->inode.get(), syncdataonly); |
10887 | } | |
10888 | ||
10889 | int Client::fstat(int fd, struct stat *stbuf, const UserPerm& perms, int mask) | |
10890 | { | |
f67539c2 TL |
10891 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10892 | if (!mref_reader.is_state_satisfied()) | |
10893 | return -CEPHFS_ENOTCONN; | |
10894 | ||
7c673cae FG |
10895 | tout(cct) << "fstat mask " << hex << mask << dec << std::endl; |
10896 | tout(cct) << fd << std::endl; | |
10897 | ||
f67539c2 | 10898 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10899 | Fh *f = get_filehandle(fd); |
10900 | if (!f) | |
f67539c2 | 10901 | return -CEPHFS_EBADF; |
7c673cae FG |
10902 | int r = _getattr(f->inode, mask, perms); |
10903 | if (r < 0) | |
10904 | return r; | |
10905 | fill_stat(f->inode, stbuf, NULL); | |
1adf2230 | 10906 | ldout(cct, 5) << "fstat(" << fd << ", " << stbuf << ") = " << r << dendl; |
7c673cae FG |
10907 | return r; |
10908 | } | |
10909 | ||
10910 | int Client::fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms, | |
10911 | unsigned int want, unsigned int flags) | |
10912 | { | |
f67539c2 TL |
10913 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10914 | if (!mref_reader.is_state_satisfied()) | |
10915 | return -CEPHFS_ENOTCONN; | |
10916 | ||
7c673cae FG |
10917 | tout(cct) << "fstatx flags " << hex << flags << " want " << want << dec << std::endl; |
10918 | tout(cct) << fd << std::endl; | |
10919 | ||
f67539c2 | 10920 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10921 | Fh *f = get_filehandle(fd); |
10922 | if (!f) | |
f67539c2 | 10923 | return -CEPHFS_EBADF; |
7c673cae FG |
10924 | |
10925 | unsigned mask = statx_to_mask(flags, want); | |
10926 | ||
10927 | int r = 0; | |
b3b6e05e | 10928 | if (mask) { |
7c673cae FG |
10929 | r = _getattr(f->inode, mask, perms); |
10930 | if (r < 0) { | |
10931 | ldout(cct, 3) << "fstatx exit on error!" << dendl; | |
10932 | return r; | |
10933 | } | |
10934 | } | |
10935 | ||
10936 | fill_statx(f->inode, mask, stx); | |
10937 | ldout(cct, 3) << "fstatx(" << fd << ", " << stx << ") = " << r << dendl; | |
10938 | return r; | |
10939 | } | |
10940 | ||
b3b6e05e TL |
10941 | int Client::statxat(int dirfd, const char *relpath, |
10942 | struct ceph_statx *stx, const UserPerm& perms, | |
10943 | unsigned int want, unsigned int flags) { | |
10944 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
10945 | if (!mref_reader.is_state_satisfied()) { | |
10946 | return -CEPHFS_ENOTCONN; | |
10947 | } | |
10948 | ||
10949 | tout(cct) << __func__ << " flags " << hex << flags << " want " << want << dec << std::endl; | |
10950 | tout(cct) << dirfd << std::endl; | |
10951 | tout(cct) << relpath << std::endl; | |
10952 | ||
10953 | unsigned mask = statx_to_mask(flags, want); | |
10954 | ||
10955 | InodeRef dirinode; | |
10956 | std::scoped_lock lock(client_lock); | |
10957 | int r = get_fd_inode(dirfd, &dirinode); | |
10958 | if (r < 0) { | |
10959 | return r; | |
10960 | } | |
10961 | ||
10962 | InodeRef in; | |
10963 | filepath path(relpath); | |
10964 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask, dirinode); | |
10965 | if (r < 0) { | |
10966 | return r; | |
10967 | } | |
10968 | r = _getattr(in, mask, perms); | |
10969 | if (r < 0) { | |
10970 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; | |
10971 | return r; | |
10972 | } | |
10973 | ||
10974 | fill_statx(in, mask, stx); | |
10975 | ldout(cct, 3) << __func__ << " dirfd" << dirfd << ", r= " << r << dendl; | |
10976 | return r; | |
10977 | } | |
10978 | ||
7c673cae FG |
10979 | // not written yet, but i want to link! |
10980 | ||
10981 | int Client::chdir(const char *relpath, std::string &new_cwd, | |
10982 | const UserPerm& perms) | |
10983 | { | |
f67539c2 TL |
10984 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10985 | if (!mref_reader.is_state_satisfied()) | |
10986 | return -CEPHFS_ENOTCONN; | |
10987 | ||
7c673cae FG |
10988 | tout(cct) << "chdir" << std::endl; |
10989 | tout(cct) << relpath << std::endl; | |
181888fb | 10990 | |
7c673cae FG |
10991 | filepath path(relpath); |
10992 | InodeRef in; | |
f67539c2 TL |
10993 | |
10994 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
10995 | int r = path_walk(path, &in, perms); |
10996 | if (r < 0) | |
10997 | return r; | |
92f5a8d4 TL |
10998 | |
10999 | if (!(in.get()->is_dir())) | |
f67539c2 | 11000 | return -CEPHFS_ENOTDIR; |
92f5a8d4 | 11001 | |
7c673cae FG |
11002 | if (cwd != in) |
11003 | cwd.swap(in); | |
11004 | ldout(cct, 3) << "chdir(" << relpath << ") cwd now " << cwd->ino << dendl; | |
11005 | ||
b5b8bbf5 | 11006 | _getcwd(new_cwd, perms); |
7c673cae FG |
11007 | return 0; |
11008 | } | |
11009 | ||
b5b8bbf5 | 11010 | void Client::_getcwd(string& dir, const UserPerm& perms) |
7c673cae FG |
11011 | { |
11012 | filepath path; | |
11fdf7f2 | 11013 | ldout(cct, 10) << __func__ << " " << *cwd << dendl; |
7c673cae FG |
11014 | |
11015 | Inode *in = cwd.get(); | |
b3b6e05e | 11016 | while (in != root.get()) { |
11fdf7f2 | 11017 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked |
7c673cae FG |
11018 | |
11019 | // A cwd or ancester is unlinked | |
11fdf7f2 | 11020 | if (in->dentries.empty()) { |
7c673cae FG |
11021 | return; |
11022 | } | |
11023 | ||
11024 | Dentry *dn = in->get_first_parent(); | |
11025 | ||
11026 | ||
11027 | if (!dn) { | |
11028 | // look it up | |
11fdf7f2 | 11029 | ldout(cct, 10) << __func__ << " looking up parent for " << *in << dendl; |
7c673cae FG |
11030 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
11031 | filepath path(in->ino); | |
11032 | req->set_filepath(path); | |
11033 | req->set_inode(in); | |
11034 | int res = make_request(req, perms); | |
11035 | if (res < 0) | |
11036 | break; | |
11037 | ||
11038 | // start over | |
11039 | path = filepath(); | |
11040 | in = cwd.get(); | |
11041 | continue; | |
11042 | } | |
11043 | path.push_front_dentry(dn->name); | |
11044 | in = dn->dir->parent_inode; | |
11045 | } | |
11046 | dir = "/"; | |
11047 | dir += path.get_path(); | |
11048 | } | |
11049 | ||
b5b8bbf5 FG |
11050 | void Client::getcwd(string& dir, const UserPerm& perms) |
11051 | { | |
f67539c2 TL |
11052 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11053 | if (!mref_reader.is_state_satisfied()) | |
11054 | return; | |
11055 | ||
11056 | std::scoped_lock l(client_lock); | |
11057 | ||
11058 | _getcwd(dir, perms); | |
b5b8bbf5 FG |
11059 | } |
11060 | ||
7c673cae FG |
11061 | int Client::statfs(const char *path, struct statvfs *stbuf, |
11062 | const UserPerm& perms) | |
11063 | { | |
f67539c2 TL |
11064 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11065 | if (!mref_reader.is_state_satisfied()) | |
11066 | return -CEPHFS_ENOTCONN; | |
11067 | ||
11fdf7f2 | 11068 | tout(cct) << __func__ << std::endl; |
91327a77 | 11069 | unsigned long int total_files_on_fs; |
7c673cae FG |
11070 | |
11071 | ceph_statfs stats; | |
11072 | C_SaferCond cond; | |
d2e6a577 | 11073 | |
f67539c2 | 11074 | std::unique_lock lock(client_lock); |
d2e6a577 FG |
11075 | const vector<int64_t> &data_pools = mdsmap->get_data_pools(); |
11076 | if (data_pools.size() == 1) { | |
11077 | objecter->get_fs_stats(stats, data_pools[0], &cond); | |
11078 | } else { | |
20effc67 | 11079 | objecter->get_fs_stats(stats, std::optional<int64_t>(), &cond); |
d2e6a577 | 11080 | } |
7c673cae | 11081 | |
f67539c2 | 11082 | lock.unlock(); |
7c673cae | 11083 | int rval = cond.wait(); |
f67539c2 TL |
11084 | lock.lock(); |
11085 | ||
20effc67 | 11086 | ceph_assert(root); |
91327a77 | 11087 | total_files_on_fs = root->rstat.rfiles + root->rstat.rsubdirs; |
7c673cae FG |
11088 | |
11089 | if (rval < 0) { | |
11090 | ldout(cct, 1) << "underlying call to statfs returned error: " | |
11091 | << cpp_strerror(rval) | |
11092 | << dendl; | |
11093 | return rval; | |
11094 | } | |
11095 | ||
11096 | memset(stbuf, 0, sizeof(*stbuf)); | |
11097 | ||
11098 | /* | |
11099 | * we're going to set a block size of 4MB so we can represent larger | |
11100 | * FSes without overflowing. Additionally convert the space | |
11101 | * measurements from KB to bytes while making them in terms of | |
11102 | * blocks. We use 4MB only because it is big enough, and because it | |
11103 | * actually *is* the (ceph) default block size. | |
11104 | */ | |
11105 | const int CEPH_BLOCK_SHIFT = 22; | |
11106 | stbuf->f_frsize = 1 << CEPH_BLOCK_SHIFT; | |
11107 | stbuf->f_bsize = 1 << CEPH_BLOCK_SHIFT; | |
91327a77 | 11108 | stbuf->f_files = total_files_on_fs; |
f67539c2 | 11109 | stbuf->f_ffree = -1; |
7c673cae FG |
11110 | stbuf->f_favail = -1; |
11111 | stbuf->f_fsid = -1; // ?? | |
11112 | stbuf->f_flag = 0; // ?? | |
11113 | stbuf->f_namemax = NAME_MAX; | |
11114 | ||
11115 | // Usually quota_root will == root_ancestor, but if the mount root has no | |
11116 | // quota but we can see a parent of it that does have a quota, we'll | |
11117 | // respect that one instead. | |
11fdf7f2 | 11118 | ceph_assert(root != nullptr); |
b3b6e05e | 11119 | InodeRef quota_root = root->quota.is_enable() ? root : get_quota_root(root.get(), perms); |
7c673cae | 11120 | |
2a845540 TL |
11121 | // get_quota_root should always give us something if client quotas are |
11122 | // enabled | |
11123 | ceph_assert(cct->_conf.get_val<bool>("client_quota") == false || quota_root != nullptr); | |
7c673cae FG |
11124 | |
11125 | if (quota_root && cct->_conf->client_quota_df && quota_root->quota.max_bytes) { | |
11126 | ||
11127 | // Skip the getattr if any sessions are stale, as we don't want to | |
11128 | // block `df` if this client has e.g. been evicted, or if the MDS cluster | |
11129 | // is unhealthy. | |
11130 | if (!_any_stale_sessions()) { | |
11131 | int r = _getattr(quota_root, 0, perms, true); | |
11132 | if (r != 0) { | |
11133 | // Ignore return value: error getting latest inode metadata is not a good | |
11134 | // reason to break "df". | |
11135 | lderr(cct) << "Error in getattr on quota root 0x" | |
11136 | << std::hex << quota_root->ino << std::dec | |
11137 | << " statfs result may be outdated" << dendl; | |
11138 | } | |
11139 | } | |
11140 | ||
11141 | // Special case: if there is a size quota set on the Inode acting | |
11142 | // as the root for this client mount, then report the quota status | |
11143 | // as the filesystem statistics. | |
11144 | const fsblkcnt_t total = quota_root->quota.max_bytes >> CEPH_BLOCK_SHIFT; | |
11145 | const fsblkcnt_t used = quota_root->rstat.rbytes >> CEPH_BLOCK_SHIFT; | |
31f18b77 FG |
11146 | // It is possible for a quota to be exceeded: arithmetic here must |
11147 | // handle case where used > total. | |
11148 | const fsblkcnt_t free = total > used ? total - used : 0; | |
7c673cae FG |
11149 | |
11150 | stbuf->f_blocks = total; | |
11151 | stbuf->f_bfree = free; | |
11152 | stbuf->f_bavail = free; | |
11153 | } else { | |
d2e6a577 | 11154 | // General case: report the cluster statistics returned from RADOS. Because |
7c673cae FG |
11155 | // multiple pools may be used without one filesystem namespace via |
11156 | // layouts, this is the most correct thing we can do. | |
11157 | stbuf->f_blocks = stats.kb >> (CEPH_BLOCK_SHIFT - 10); | |
11158 | stbuf->f_bfree = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
11159 | stbuf->f_bavail = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
11160 | } | |
11161 | ||
11162 | return rval; | |
11163 | } | |
11164 | ||
11165 | int Client::_do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep, | |
11166 | struct flock *fl, uint64_t owner, bool removing) | |
11167 | { | |
11fdf7f2 | 11168 | ldout(cct, 10) << __func__ << " ino " << in->ino |
7c673cae FG |
11169 | << (lock_type == CEPH_LOCK_FCNTL ? " fcntl" : " flock") |
11170 | << " type " << fl->l_type << " owner " << owner | |
11171 | << " " << fl->l_start << "~" << fl->l_len << dendl; | |
11172 | ||
f6b5b4d7 | 11173 | if (in->flags & I_ERROR_FILELOCK) |
f67539c2 | 11174 | return -CEPHFS_EIO; |
f6b5b4d7 | 11175 | |
7c673cae FG |
11176 | int lock_cmd; |
11177 | if (F_RDLCK == fl->l_type) | |
11178 | lock_cmd = CEPH_LOCK_SHARED; | |
11179 | else if (F_WRLCK == fl->l_type) | |
11180 | lock_cmd = CEPH_LOCK_EXCL; | |
11181 | else if (F_UNLCK == fl->l_type) | |
11182 | lock_cmd = CEPH_LOCK_UNLOCK; | |
11183 | else | |
f67539c2 | 11184 | return -CEPHFS_EIO; |
7c673cae FG |
11185 | |
11186 | if (op != CEPH_MDS_OP_SETFILELOCK || lock_cmd == CEPH_LOCK_UNLOCK) | |
11187 | sleep = 0; | |
11188 | ||
11189 | /* | |
11190 | * Set the most significant bit, so that MDS knows the 'owner' | |
11191 | * is sufficient to identify the owner of lock. (old code uses | |
11192 | * both 'owner' and 'pid') | |
11193 | */ | |
11194 | owner |= (1ULL << 63); | |
11195 | ||
11196 | MetaRequest *req = new MetaRequest(op); | |
11197 | filepath path; | |
11198 | in->make_nosnap_relative_path(path); | |
11199 | req->set_filepath(path); | |
11200 | req->set_inode(in); | |
11201 | ||
11202 | req->head.args.filelock_change.rule = lock_type; | |
11203 | req->head.args.filelock_change.type = lock_cmd; | |
11204 | req->head.args.filelock_change.owner = owner; | |
11205 | req->head.args.filelock_change.pid = fl->l_pid; | |
11206 | req->head.args.filelock_change.start = fl->l_start; | |
11207 | req->head.args.filelock_change.length = fl->l_len; | |
11208 | req->head.args.filelock_change.wait = sleep; | |
11209 | ||
11210 | int ret; | |
11211 | bufferlist bl; | |
11212 | ||
11213 | if (sleep && switch_interrupt_cb) { | |
11214 | // enable interrupt | |
11215 | switch_interrupt_cb(callback_handle, req->get()); | |
11216 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
7c673cae FG |
11217 | // disable interrupt |
11218 | switch_interrupt_cb(callback_handle, NULL); | |
31f18b77 FG |
11219 | if (ret == 0 && req->aborted()) { |
11220 | // effect of this lock request has been revoked by the 'lock intr' request | |
11221 | ret = req->get_abort_code(); | |
11222 | } | |
7c673cae FG |
11223 | put_request(req); |
11224 | } else { | |
11225 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
11226 | } | |
11227 | ||
11228 | if (ret == 0) { | |
11229 | if (op == CEPH_MDS_OP_GETFILELOCK) { | |
11230 | ceph_filelock filelock; | |
11fdf7f2 TL |
11231 | auto p = bl.cbegin(); |
11232 | decode(filelock, p); | |
7c673cae FG |
11233 | |
11234 | if (CEPH_LOCK_SHARED == filelock.type) | |
11235 | fl->l_type = F_RDLCK; | |
11236 | else if (CEPH_LOCK_EXCL == filelock.type) | |
11237 | fl->l_type = F_WRLCK; | |
11238 | else | |
11239 | fl->l_type = F_UNLCK; | |
11240 | ||
11241 | fl->l_whence = SEEK_SET; | |
11242 | fl->l_start = filelock.start; | |
11243 | fl->l_len = filelock.length; | |
11244 | fl->l_pid = filelock.pid; | |
11245 | } else if (op == CEPH_MDS_OP_SETFILELOCK) { | |
11246 | ceph_lock_state_t *lock_state; | |
11247 | if (lock_type == CEPH_LOCK_FCNTL) { | |
11248 | if (!in->fcntl_locks) | |
11fdf7f2 TL |
11249 | in->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
11250 | lock_state = in->fcntl_locks.get(); | |
7c673cae FG |
11251 | } else if (lock_type == CEPH_LOCK_FLOCK) { |
11252 | if (!in->flock_locks) | |
11fdf7f2 TL |
11253 | in->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
11254 | lock_state = in->flock_locks.get(); | |
7c673cae FG |
11255 | } else { |
11256 | ceph_abort(); | |
f67539c2 | 11257 | return -CEPHFS_EINVAL; |
7c673cae FG |
11258 | } |
11259 | _update_lock_state(fl, owner, lock_state); | |
11260 | ||
11261 | if (!removing) { | |
11262 | if (lock_type == CEPH_LOCK_FCNTL) { | |
11263 | if (!fh->fcntl_locks) | |
11fdf7f2 TL |
11264 | fh->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
11265 | lock_state = fh->fcntl_locks.get(); | |
7c673cae FG |
11266 | } else { |
11267 | if (!fh->flock_locks) | |
11fdf7f2 TL |
11268 | fh->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
11269 | lock_state = fh->flock_locks.get(); | |
7c673cae FG |
11270 | } |
11271 | _update_lock_state(fl, owner, lock_state); | |
11272 | } | |
11273 | } else | |
11274 | ceph_abort(); | |
11275 | } | |
11276 | return ret; | |
11277 | } | |
11278 | ||
11279 | int Client::_interrupt_filelock(MetaRequest *req) | |
11280 | { | |
31f18b77 FG |
11281 | // Set abort code, but do not kick. The abort code prevents the request |
11282 | // from being re-sent. | |
f67539c2 | 11283 | req->abort(-CEPHFS_EINTR); |
31f18b77 FG |
11284 | if (req->mds < 0) |
11285 | return 0; // haven't sent the request | |
11286 | ||
7c673cae FG |
11287 | Inode *in = req->inode(); |
11288 | ||
11289 | int lock_type; | |
11290 | if (req->head.args.filelock_change.rule == CEPH_LOCK_FLOCK) | |
11291 | lock_type = CEPH_LOCK_FLOCK_INTR; | |
11292 | else if (req->head.args.filelock_change.rule == CEPH_LOCK_FCNTL) | |
11293 | lock_type = CEPH_LOCK_FCNTL_INTR; | |
11294 | else { | |
11295 | ceph_abort(); | |
f67539c2 | 11296 | return -CEPHFS_EINVAL; |
7c673cae FG |
11297 | } |
11298 | ||
11299 | MetaRequest *intr_req = new MetaRequest(CEPH_MDS_OP_SETFILELOCK); | |
11300 | filepath path; | |
11301 | in->make_nosnap_relative_path(path); | |
11302 | intr_req->set_filepath(path); | |
11303 | intr_req->set_inode(in); | |
11304 | intr_req->head.args.filelock_change = req->head.args.filelock_change; | |
11305 | intr_req->head.args.filelock_change.rule = lock_type; | |
11306 | intr_req->head.args.filelock_change.type = CEPH_LOCK_UNLOCK; | |
11307 | ||
11308 | UserPerm perms(req->get_uid(), req->get_gid()); | |
11309 | return make_request(intr_req, perms, NULL, NULL, -1); | |
11310 | } | |
11311 | ||
11312 | void Client::_encode_filelocks(Inode *in, bufferlist& bl) | |
11313 | { | |
11314 | if (!in->fcntl_locks && !in->flock_locks) | |
11315 | return; | |
11316 | ||
11317 | unsigned nr_fcntl_locks = in->fcntl_locks ? in->fcntl_locks->held_locks.size() : 0; | |
11fdf7f2 | 11318 | encode(nr_fcntl_locks, bl); |
7c673cae | 11319 | if (nr_fcntl_locks) { |
11fdf7f2 | 11320 | auto &lock_state = in->fcntl_locks; |
20effc67 | 11321 | for(auto p = lock_state->held_locks.begin(); |
7c673cae FG |
11322 | p != lock_state->held_locks.end(); |
11323 | ++p) | |
11fdf7f2 | 11324 | encode(p->second, bl); |
7c673cae FG |
11325 | } |
11326 | ||
11327 | unsigned nr_flock_locks = in->flock_locks ? in->flock_locks->held_locks.size() : 0; | |
11fdf7f2 | 11328 | encode(nr_flock_locks, bl); |
7c673cae | 11329 | if (nr_flock_locks) { |
11fdf7f2 | 11330 | auto &lock_state = in->flock_locks; |
20effc67 | 11331 | for(auto p = lock_state->held_locks.begin(); |
7c673cae FG |
11332 | p != lock_state->held_locks.end(); |
11333 | ++p) | |
11fdf7f2 | 11334 | encode(p->second, bl); |
7c673cae FG |
11335 | } |
11336 | ||
11fdf7f2 | 11337 | ldout(cct, 10) << __func__ << " ino " << in->ino << ", " << nr_fcntl_locks |
7c673cae FG |
11338 | << " fcntl locks, " << nr_flock_locks << " flock locks" << dendl; |
11339 | } | |
11340 | ||
11341 | void Client::_release_filelocks(Fh *fh) | |
11342 | { | |
11343 | if (!fh->fcntl_locks && !fh->flock_locks) | |
11344 | return; | |
11345 | ||
11346 | Inode *in = fh->inode.get(); | |
11fdf7f2 | 11347 | ldout(cct, 10) << __func__ << " " << fh << " ino " << in->ino << dendl; |
7c673cae | 11348 | |
f6b5b4d7 TL |
11349 | list<ceph_filelock> activated_locks; |
11350 | ||
7c673cae FG |
11351 | list<pair<int, ceph_filelock> > to_release; |
11352 | ||
11353 | if (fh->fcntl_locks) { | |
11fdf7f2 | 11354 | auto &lock_state = fh->fcntl_locks; |
f6b5b4d7 TL |
11355 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11356 | auto q = p++; | |
11357 | if (in->flags & I_ERROR_FILELOCK) { | |
11358 | lock_state->remove_lock(q->second, activated_locks); | |
11359 | } else { | |
11360 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FCNTL, q->second)); | |
11361 | } | |
11362 | } | |
11fdf7f2 | 11363 | lock_state.reset(); |
7c673cae FG |
11364 | } |
11365 | if (fh->flock_locks) { | |
11fdf7f2 | 11366 | auto &lock_state = fh->flock_locks; |
f6b5b4d7 TL |
11367 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11368 | auto q = p++; | |
11369 | if (in->flags & I_ERROR_FILELOCK) { | |
11370 | lock_state->remove_lock(q->second, activated_locks); | |
11371 | } else { | |
11372 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FLOCK, q->second)); | |
11373 | } | |
11374 | } | |
11fdf7f2 | 11375 | lock_state.reset(); |
7c673cae FG |
11376 | } |
11377 | ||
f6b5b4d7 TL |
11378 | if ((in->flags & I_ERROR_FILELOCK) && !in->has_any_filelocks()) |
11379 | in->flags &= ~I_ERROR_FILELOCK; | |
7c673cae | 11380 | |
f6b5b4d7 | 11381 | if (to_release.empty()) |
11fdf7f2 TL |
11382 | return; |
11383 | ||
7c673cae FG |
11384 | struct flock fl; |
11385 | memset(&fl, 0, sizeof(fl)); | |
11386 | fl.l_whence = SEEK_SET; | |
11387 | fl.l_type = F_UNLCK; | |
11388 | ||
11389 | for (list<pair<int, ceph_filelock> >::iterator p = to_release.begin(); | |
11390 | p != to_release.end(); | |
11391 | ++p) { | |
11392 | fl.l_start = p->second.start; | |
11393 | fl.l_len = p->second.length; | |
11394 | fl.l_pid = p->second.pid; | |
11395 | _do_filelock(in, fh, p->first, CEPH_MDS_OP_SETFILELOCK, 0, &fl, | |
11396 | p->second.owner, true); | |
11397 | } | |
11398 | } | |
11399 | ||
11400 | void Client::_update_lock_state(struct flock *fl, uint64_t owner, | |
11401 | ceph_lock_state_t *lock_state) | |
11402 | { | |
11403 | int lock_cmd; | |
11404 | if (F_RDLCK == fl->l_type) | |
11405 | lock_cmd = CEPH_LOCK_SHARED; | |
11406 | else if (F_WRLCK == fl->l_type) | |
11407 | lock_cmd = CEPH_LOCK_EXCL; | |
11408 | else | |
11409 | lock_cmd = CEPH_LOCK_UNLOCK;; | |
11410 | ||
11411 | ceph_filelock filelock; | |
11412 | filelock.start = fl->l_start; | |
11413 | filelock.length = fl->l_len; | |
11414 | filelock.client = 0; | |
11415 | // see comment in _do_filelock() | |
11416 | filelock.owner = owner | (1ULL << 63); | |
11417 | filelock.pid = fl->l_pid; | |
11418 | filelock.type = lock_cmd; | |
11419 | ||
11420 | if (filelock.type == CEPH_LOCK_UNLOCK) { | |
11421 | list<ceph_filelock> activated_locks; | |
11422 | lock_state->remove_lock(filelock, activated_locks); | |
11423 | } else { | |
11424 | bool r = lock_state->add_lock(filelock, false, false, NULL); | |
11fdf7f2 | 11425 | ceph_assert(r); |
7c673cae FG |
11426 | } |
11427 | } | |
11428 | ||
11429 | int Client::_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
11430 | { | |
11431 | Inode *in = fh->inode.get(); | |
11432 | ldout(cct, 10) << "_getlk " << fh << " ino " << in->ino << dendl; | |
11433 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_GETFILELOCK, 0, fl, owner); | |
11434 | return ret; | |
11435 | } | |
11436 | ||
11437 | int Client::_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
11438 | { | |
11439 | Inode *in = fh->inode.get(); | |
11440 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << dendl; | |
11441 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_SETFILELOCK, sleep, fl, owner); | |
11442 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11443 | return ret; | |
11444 | } | |
11445 | ||
11446 | int Client::_flock(Fh *fh, int cmd, uint64_t owner) | |
11447 | { | |
11448 | Inode *in = fh->inode.get(); | |
11449 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << dendl; | |
11450 | ||
11451 | int sleep = !(cmd & LOCK_NB); | |
11452 | cmd &= ~LOCK_NB; | |
11453 | ||
11454 | int type; | |
11455 | switch (cmd) { | |
11456 | case LOCK_SH: | |
11457 | type = F_RDLCK; | |
11458 | break; | |
11459 | case LOCK_EX: | |
11460 | type = F_WRLCK; | |
11461 | break; | |
11462 | case LOCK_UN: | |
11463 | type = F_UNLCK; | |
11464 | break; | |
11465 | default: | |
f67539c2 | 11466 | return -CEPHFS_EINVAL; |
7c673cae FG |
11467 | } |
11468 | ||
11469 | struct flock fl; | |
11470 | memset(&fl, 0, sizeof(fl)); | |
11471 | fl.l_type = type; | |
11472 | fl.l_whence = SEEK_SET; | |
11473 | ||
11474 | int ret = _do_filelock(in, fh, CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK, sleep, &fl, owner); | |
11475 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11476 | return ret; | |
11477 | } | |
11478 | ||
f67539c2 TL |
11479 | int Client::get_snap_info(const char *path, const UserPerm &perms, SnapInfo *snap_info) { |
11480 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11481 | if (!mref_reader.is_state_satisfied()) { | |
11482 | return -CEPHFS_ENOTCONN; | |
11483 | } | |
11484 | ||
20effc67 | 11485 | std::scoped_lock lock(client_lock); |
f67539c2 TL |
11486 | InodeRef in; |
11487 | int r = Client::path_walk(path, &in, perms, true); | |
11488 | if (r < 0) { | |
11489 | return r; | |
11490 | } | |
11491 | ||
11492 | if (in->snapid == CEPH_NOSNAP) { | |
11493 | return -CEPHFS_EINVAL; | |
11494 | } | |
11495 | ||
11496 | snap_info->id = in->snapid; | |
11497 | snap_info->metadata = in->snap_metadata; | |
11498 | return 0; | |
11499 | } | |
11500 | ||
11501 | int Client::ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms) | |
11502 | { | |
11503 | /* Since the only thing this does is wrap a call to statfs, and | |
11504 | statfs takes a lock, it doesn't seem we have a need to split it | |
11505 | out. */ | |
7c673cae FG |
11506 | return statfs(0, stbuf, perms); |
11507 | } | |
11508 | ||
20effc67 | 11509 | void Client::_ll_register_callbacks(struct ceph_client_callback_args *args) |
7c673cae FG |
11510 | { |
11511 | if (!args) | |
11512 | return; | |
20effc67 | 11513 | |
11fdf7f2 | 11514 | ldout(cct, 10) << __func__ << " cb " << args->handle |
7c673cae FG |
11515 | << " invalidate_ino_cb " << args->ino_cb |
11516 | << " invalidate_dentry_cb " << args->dentry_cb | |
7c673cae FG |
11517 | << " switch_interrupt_cb " << args->switch_intr_cb |
11518 | << " remount_cb " << args->remount_cb | |
11519 | << dendl; | |
11520 | callback_handle = args->handle; | |
11521 | if (args->ino_cb) { | |
11522 | ino_invalidate_cb = args->ino_cb; | |
11523 | async_ino_invalidator.start(); | |
11524 | } | |
11525 | if (args->dentry_cb) { | |
11526 | dentry_invalidate_cb = args->dentry_cb; | |
11527 | async_dentry_invalidator.start(); | |
11528 | } | |
11529 | if (args->switch_intr_cb) { | |
11530 | switch_interrupt_cb = args->switch_intr_cb; | |
11531 | interrupt_finisher.start(); | |
11532 | } | |
11533 | if (args->remount_cb) { | |
11534 | remount_cb = args->remount_cb; | |
11535 | remount_finisher.start(); | |
11536 | } | |
e306af50 TL |
11537 | if (args->ino_release_cb) { |
11538 | ino_release_cb = args->ino_release_cb; | |
11539 | async_ino_releasor.start(); | |
11540 | } | |
11541 | if (args->umask_cb) | |
11542 | umask_cb = args->umask_cb; | |
7c673cae FG |
11543 | } |
11544 | ||
20effc67 TL |
11545 | // This is deprecated, use ll_register_callbacks2() instead. |
11546 | void Client::ll_register_callbacks(struct ceph_client_callback_args *args) | |
11547 | { | |
11548 | ceph_assert(!is_mounting() && !is_mounted() && !is_unmounting()); | |
11549 | ||
11550 | _ll_register_callbacks(args); | |
11551 | } | |
11552 | ||
11553 | int Client::ll_register_callbacks2(struct ceph_client_callback_args *args) | |
11554 | { | |
11555 | if (is_mounting() || is_mounted() || is_unmounting()) | |
11556 | return -CEPHFS_EBUSY; | |
11557 | ||
11558 | _ll_register_callbacks(args); | |
11559 | return 0; | |
11560 | } | |
11561 | ||
1d09f67e | 11562 | std::pair<int, bool> Client::test_dentry_handling(bool can_invalidate) |
7c673cae | 11563 | { |
1d09f67e | 11564 | std::pair <int, bool> r(0, false); |
7c673cae | 11565 | |
f67539c2 TL |
11566 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
11567 | if (!iref_reader.is_state_satisfied()) | |
1d09f67e | 11568 | return std::make_pair(-CEPHFS_ENOTCONN, false); |
f67539c2 | 11569 | |
7c673cae FG |
11570 | can_invalidate_dentries = can_invalidate; |
11571 | ||
11572 | if (can_invalidate_dentries) { | |
11fdf7f2 | 11573 | ceph_assert(dentry_invalidate_cb); |
7c673cae | 11574 | ldout(cct, 1) << "using dentry_invalidate_cb" << dendl; |
11fdf7f2 TL |
11575 | } else { |
11576 | ceph_assert(remount_cb); | |
7c673cae | 11577 | ldout(cct, 1) << "using remount_cb" << dendl; |
91327a77 | 11578 | r = _do_remount(false); |
b32b8144 | 11579 | } |
11fdf7f2 | 11580 | |
7c673cae FG |
11581 | return r; |
11582 | } | |
11583 | ||
11584 | int Client::_sync_fs() | |
11585 | { | |
f67539c2 TL |
11586 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11587 | ||
11fdf7f2 | 11588 | ldout(cct, 10) << __func__ << dendl; |
7c673cae FG |
11589 | |
11590 | // flush file data | |
11fdf7f2 TL |
11591 | std::unique_ptr<C_SaferCond> cond = nullptr; |
11592 | if (cct->_conf->client_oc) { | |
11593 | cond.reset(new C_SaferCond("Client::_sync_fs:lock")); | |
11594 | objectcacher->flush_all(cond.get()); | |
11595 | } | |
7c673cae FG |
11596 | |
11597 | // flush caps | |
11598 | flush_caps_sync(); | |
11599 | ceph_tid_t flush_tid = last_flush_tid; | |
11600 | ||
11601 | // wait for unsafe mds requests | |
11602 | wait_unsafe_requests(); | |
11603 | ||
11604 | wait_sync_caps(flush_tid); | |
11605 | ||
11fdf7f2 | 11606 | if (nullptr != cond) { |
9f95a23c | 11607 | client_lock.unlock(); |
11fdf7f2 TL |
11608 | ldout(cct, 15) << __func__ << " waiting on data to flush" << dendl; |
11609 | cond->wait(); | |
11610 | ldout(cct, 15) << __func__ << " flush finished" << dendl; | |
9f95a23c | 11611 | client_lock.lock(); |
7c673cae FG |
11612 | } |
11613 | ||
11614 | return 0; | |
11615 | } | |
11616 | ||
11617 | int Client::sync_fs() | |
11618 | { | |
f67539c2 TL |
11619 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11620 | if (!mref_reader.is_state_satisfied()) | |
11621 | return -CEPHFS_ENOTCONN; | |
181888fb | 11622 | |
f67539c2 | 11623 | std::scoped_lock l(client_lock); |
181888fb | 11624 | |
7c673cae FG |
11625 | return _sync_fs(); |
11626 | } | |
11627 | ||
11628 | int64_t Client::drop_caches() | |
11629 | { | |
f67539c2 | 11630 | std::scoped_lock l(client_lock); |
7c673cae FG |
11631 | return objectcacher->release_all(); |
11632 | } | |
11633 | ||
11fdf7f2 TL |
11634 | int Client::_lazyio(Fh *fh, int enable) |
11635 | { | |
11636 | Inode *in = fh->inode.get(); | |
11637 | ldout(cct, 20) << __func__ << " " << *in << " " << !!enable << dendl; | |
11638 | ||
11639 | if (!!(fh->mode & CEPH_FILE_MODE_LAZY) == !!enable) | |
11640 | return 0; | |
11641 | ||
11642 | int orig_mode = fh->mode; | |
11643 | if (enable) { | |
11644 | fh->mode |= CEPH_FILE_MODE_LAZY; | |
11645 | in->get_open_ref(fh->mode); | |
11646 | in->put_open_ref(orig_mode); | |
11647 | check_caps(in, CHECK_CAPS_NODELAY); | |
11648 | } else { | |
11649 | fh->mode &= ~CEPH_FILE_MODE_LAZY; | |
11650 | in->get_open_ref(fh->mode); | |
11651 | in->put_open_ref(orig_mode); | |
11652 | check_caps(in, 0); | |
11653 | } | |
11654 | ||
11655 | return 0; | |
11656 | } | |
11657 | ||
11658 | int Client::lazyio(int fd, int enable) | |
11659 | { | |
f67539c2 | 11660 | std::scoped_lock l(client_lock); |
11fdf7f2 TL |
11661 | Fh *f = get_filehandle(fd); |
11662 | if (!f) | |
f67539c2 | 11663 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
11664 | |
11665 | return _lazyio(f, enable); | |
11666 | } | |
11667 | ||
11668 | int Client::ll_lazyio(Fh *fh, int enable) | |
11669 | { | |
11fdf7f2 TL |
11670 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << !!enable << dendl; |
11671 | tout(cct) << __func__ << std::endl; | |
11672 | ||
f67539c2 | 11673 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
11674 | return _lazyio(fh, enable); |
11675 | } | |
7c673cae | 11676 | |
92f5a8d4 | 11677 | int Client::lazyio_propagate(int fd, loff_t offset, size_t count) |
7c673cae | 11678 | { |
f67539c2 | 11679 | std::scoped_lock l(client_lock); |
92f5a8d4 | 11680 | ldout(cct, 3) << "op: client->lazyio_propagate(" << fd |
7c673cae FG |
11681 | << ", " << offset << ", " << count << ")" << dendl; |
11682 | ||
11683 | Fh *f = get_filehandle(fd); | |
11684 | if (!f) | |
f67539c2 | 11685 | return -CEPHFS_EBADF; |
7c673cae FG |
11686 | |
11687 | // for now | |
11688 | _fsync(f, true); | |
11689 | ||
11690 | return 0; | |
11691 | } | |
11692 | ||
11693 | int Client::lazyio_synchronize(int fd, loff_t offset, size_t count) | |
11694 | { | |
f67539c2 | 11695 | std::scoped_lock l(client_lock); |
7c673cae FG |
11696 | ldout(cct, 3) << "op: client->lazyio_synchronize(" << fd |
11697 | << ", " << offset << ", " << count << ")" << dendl; | |
11698 | ||
11699 | Fh *f = get_filehandle(fd); | |
11700 | if (!f) | |
f67539c2 | 11701 | return -CEPHFS_EBADF; |
7c673cae FG |
11702 | Inode *in = f->inode.get(); |
11703 | ||
11704 | _fsync(f, true); | |
92f5a8d4 TL |
11705 | if (_release(in)) { |
11706 | int r =_getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
11707 | if (r < 0) | |
11708 | return r; | |
11709 | } | |
7c673cae FG |
11710 | return 0; |
11711 | } | |
11712 | ||
11713 | ||
11714 | // ============================= | |
11715 | // snaps | |
11716 | ||
f67539c2 TL |
11717 | int Client::mksnap(const char *relpath, const char *name, const UserPerm& perm, |
11718 | mode_t mode, const std::map<std::string, std::string> &metadata) | |
7c673cae | 11719 | { |
f67539c2 TL |
11720 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11721 | if (!mref_reader.is_state_satisfied()) | |
11722 | return -CEPHFS_ENOTCONN; | |
181888fb | 11723 | |
f67539c2 | 11724 | std::scoped_lock l(client_lock); |
181888fb | 11725 | |
7c673cae FG |
11726 | filepath path(relpath); |
11727 | InodeRef in; | |
11728 | int r = path_walk(path, &in, perm); | |
11729 | if (r < 0) | |
11730 | return r; | |
11731 | if (cct->_conf->client_permissions) { | |
11732 | r = may_create(in.get(), perm); | |
11733 | if (r < 0) | |
11734 | return r; | |
11735 | } | |
11736 | Inode *snapdir = open_snapdir(in.get()); | |
f67539c2 | 11737 | return _mkdir(snapdir, name, mode, perm, nullptr, metadata); |
7c673cae | 11738 | } |
181888fb | 11739 | |
f67539c2 | 11740 | int Client::rmsnap(const char *relpath, const char *name, const UserPerm& perms, bool check_perms) |
7c673cae | 11741 | { |
f67539c2 TL |
11742 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11743 | if (!mref_reader.is_state_satisfied()) | |
11744 | return -CEPHFS_ENOTCONN; | |
181888fb | 11745 | |
f67539c2 | 11746 | std::scoped_lock l(client_lock); |
181888fb | 11747 | |
7c673cae FG |
11748 | filepath path(relpath); |
11749 | InodeRef in; | |
11750 | int r = path_walk(path, &in, perms); | |
11751 | if (r < 0) | |
11752 | return r; | |
f67539c2 | 11753 | Inode *snapdir = open_snapdir(in.get()); |
7c673cae | 11754 | if (cct->_conf->client_permissions) { |
f67539c2 | 11755 | r = may_delete(snapdir, check_perms ? name : NULL, perms); |
7c673cae FG |
11756 | if (r < 0) |
11757 | return r; | |
11758 | } | |
7c673cae FG |
11759 | return _rmdir(snapdir, name, perms); |
11760 | } | |
11761 | ||
11762 | // ============================= | |
11763 | // expose caps | |
11764 | ||
f67539c2 TL |
11765 | int Client::get_caps_issued(int fd) |
11766 | { | |
11767 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11768 | if (!mref_reader.is_state_satisfied()) | |
11769 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11770 | |
f67539c2 | 11771 | std::scoped_lock lock(client_lock); |
181888fb | 11772 | |
7c673cae FG |
11773 | Fh *f = get_filehandle(fd); |
11774 | if (!f) | |
f67539c2 | 11775 | return -CEPHFS_EBADF; |
7c673cae FG |
11776 | |
11777 | return f->inode->caps_issued(); | |
11778 | } | |
11779 | ||
11780 | int Client::get_caps_issued(const char *path, const UserPerm& perms) | |
11781 | { | |
f67539c2 TL |
11782 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11783 | if (!mref_reader.is_state_satisfied()) | |
11784 | return -CEPHFS_ENOTCONN; | |
181888fb | 11785 | |
f67539c2 | 11786 | std::scoped_lock lock(client_lock); |
181888fb | 11787 | |
7c673cae FG |
11788 | filepath p(path); |
11789 | InodeRef in; | |
11790 | int r = path_walk(p, &in, perms, true); | |
11791 | if (r < 0) | |
11792 | return r; | |
11793 | return in->caps_issued(); | |
11794 | } | |
11795 | ||
11796 | // ========================================= | |
11797 | // low level | |
11798 | ||
11799 | Inode *Client::open_snapdir(Inode *diri) | |
11800 | { | |
11801 | Inode *in; | |
11802 | vinodeno_t vino(diri->ino, CEPH_SNAPDIR); | |
11803 | if (!inode_map.count(vino)) { | |
11804 | in = new Inode(this, vino, &diri->layout); | |
11805 | ||
11806 | in->ino = diri->ino; | |
11807 | in->snapid = CEPH_SNAPDIR; | |
11808 | in->mode = diri->mode; | |
11809 | in->uid = diri->uid; | |
11810 | in->gid = diri->gid; | |
494da23a | 11811 | in->nlink = 1; |
7c673cae FG |
11812 | in->mtime = diri->mtime; |
11813 | in->ctime = diri->ctime; | |
11814 | in->btime = diri->btime; | |
f6b5b4d7 | 11815 | in->atime = diri->atime; |
7c673cae FG |
11816 | in->size = diri->size; |
11817 | in->change_attr = diri->change_attr; | |
11818 | ||
11819 | in->dirfragtree.clear(); | |
11820 | in->snapdir_parent = diri; | |
11821 | diri->flags |= I_SNAPDIR_OPEN; | |
11822 | inode_map[vino] = in; | |
11823 | if (use_faked_inos()) | |
11824 | _assign_faked_ino(in); | |
11825 | ldout(cct, 10) << "open_snapdir created snapshot inode " << *in << dendl; | |
11826 | } else { | |
11827 | in = inode_map[vino]; | |
11828 | ldout(cct, 10) << "open_snapdir had snapshot inode " << *in << dendl; | |
11829 | } | |
11830 | return in; | |
11831 | } | |
11832 | ||
11833 | int Client::ll_lookup(Inode *parent, const char *name, struct stat *attr, | |
11834 | Inode **out, const UserPerm& perms) | |
11835 | { | |
f67539c2 TL |
11836 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11837 | if (!mref_reader.is_state_satisfied()) | |
11838 | return -CEPHFS_ENOTCONN; | |
11839 | ||
31f18b77 | 11840 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 TL |
11841 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
11842 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11843 | tout(cct) << name << std::endl; |
11844 | ||
f67539c2 | 11845 | std::scoped_lock lock(client_lock); |
181888fb | 11846 | |
7c673cae | 11847 | int r = 0; |
11fdf7f2 TL |
11848 | if (!fuse_default_permissions) { |
11849 | if (strcmp(name, ".") && strcmp(name, "..")) { | |
11850 | r = may_lookup(parent, perms); | |
11851 | if (r < 0) | |
11852 | return r; | |
11853 | } | |
7c673cae FG |
11854 | } |
11855 | ||
11856 | string dname(name); | |
11857 | InodeRef in; | |
11858 | ||
11859 | r = _lookup(parent, dname, CEPH_STAT_CAP_INODE_ALL, &in, perms); | |
11860 | if (r < 0) { | |
11861 | attr->st_ino = 0; | |
11862 | goto out; | |
11863 | } | |
11864 | ||
11fdf7f2 | 11865 | ceph_assert(in); |
7c673cae FG |
11866 | fill_stat(in, attr); |
11867 | _ll_get(in.get()); | |
11868 | ||
11869 | out: | |
11fdf7f2 | 11870 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11871 | << " -> " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; |
11872 | tout(cct) << attr->st_ino << std::endl; | |
11873 | *out = in.get(); | |
11874 | return r; | |
11875 | } | |
11876 | ||
f67539c2 TL |
11877 | int Client::ll_lookup_vino( |
11878 | vinodeno_t vino, | |
1adf2230 AA |
11879 | const UserPerm& perms, |
11880 | Inode **inode) | |
11881 | { | |
81eedcae | 11882 | ceph_assert(inode != NULL); |
f67539c2 TL |
11883 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11884 | if (!mref_reader.is_state_satisfied()) | |
11885 | return -CEPHFS_ENOTCONN; | |
81eedcae | 11886 | |
b3b6e05e TL |
11887 | if (is_reserved_vino(vino)) |
11888 | return -CEPHFS_ESTALE; | |
11889 | ||
f67539c2 TL |
11890 | std::scoped_lock lock(client_lock); |
11891 | ldout(cct, 3) << __func__ << " " << vino << dendl; | |
1adf2230 | 11892 | |
f67539c2 TL |
11893 | // Check the cache first |
11894 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
11895 | if (p != inode_map.end()) { | |
11896 | *inode = p->second; | |
11897 | _ll_get(*inode); | |
81eedcae TL |
11898 | return 0; |
11899 | } | |
11900 | ||
f67539c2 | 11901 | uint64_t snapid = vino.snapid; |
81eedcae | 11902 | |
f67539c2 TL |
11903 | // for snapdir, find the non-snapped dir inode |
11904 | if (snapid == CEPH_SNAPDIR) | |
11905 | vino.snapid = CEPH_NOSNAP; | |
11906 | ||
11907 | int r = _lookup_vino(vino, perms, inode); | |
11908 | if (r) | |
1adf2230 | 11909 | return r; |
f67539c2 | 11910 | ceph_assert(*inode != NULL); |
81eedcae | 11911 | |
f67539c2 TL |
11912 | if (snapid == CEPH_SNAPDIR) { |
11913 | Inode *tmp = *inode; | |
1adf2230 | 11914 | |
f67539c2 TL |
11915 | // open the snapdir and put the inode ref |
11916 | *inode = open_snapdir(tmp); | |
11917 | _ll_forget(tmp, 1); | |
11918 | _ll_get(*inode); | |
1adf2230 | 11919 | } |
1adf2230 AA |
11920 | return 0; |
11921 | } | |
11922 | ||
f67539c2 TL |
11923 | int Client::ll_lookup_inode( |
11924 | struct inodeno_t ino, | |
11925 | const UserPerm& perms, | |
11926 | Inode **inode) | |
11927 | { | |
11928 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
11929 | return ll_lookup_vino(vino, perms, inode); | |
11930 | } | |
11931 | ||
7c673cae FG |
11932 | int Client::ll_lookupx(Inode *parent, const char *name, Inode **out, |
11933 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
11934 | const UserPerm& perms) | |
11935 | { | |
f67539c2 TL |
11936 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11937 | if (!mref_reader.is_state_satisfied()) | |
11938 | return -CEPHFS_ENOTCONN; | |
11939 | ||
31f18b77 | 11940 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 | 11941 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
7c673cae FG |
11942 | tout(cct) << "ll_lookupx" << std::endl; |
11943 | tout(cct) << name << std::endl; | |
11944 | ||
f67539c2 | 11945 | std::scoped_lock lock(client_lock); |
181888fb | 11946 | |
7c673cae | 11947 | int r = 0; |
11fdf7f2 | 11948 | if (!fuse_default_permissions) { |
7c673cae FG |
11949 | r = may_lookup(parent, perms); |
11950 | if (r < 0) | |
11951 | return r; | |
11952 | } | |
11953 | ||
11954 | string dname(name); | |
11955 | InodeRef in; | |
11956 | ||
11957 | unsigned mask = statx_to_mask(flags, want); | |
11958 | r = _lookup(parent, dname, mask, &in, perms); | |
11959 | if (r < 0) { | |
11960 | stx->stx_ino = 0; | |
11961 | stx->stx_mask = 0; | |
11962 | } else { | |
11fdf7f2 | 11963 | ceph_assert(in); |
7c673cae FG |
11964 | fill_statx(in, mask, stx); |
11965 | _ll_get(in.get()); | |
11966 | } | |
11967 | ||
11fdf7f2 | 11968 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11969 | << " -> " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; |
11970 | tout(cct) << stx->stx_ino << std::endl; | |
11971 | *out = in.get(); | |
11972 | return r; | |
11973 | } | |
11974 | ||
11975 | int Client::ll_walk(const char* name, Inode **out, struct ceph_statx *stx, | |
11976 | unsigned int want, unsigned int flags, const UserPerm& perms) | |
11977 | { | |
f67539c2 TL |
11978 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11979 | if (!mref_reader.is_state_satisfied()) | |
11980 | return -CEPHFS_ENOTCONN; | |
181888fb | 11981 | |
7c673cae FG |
11982 | filepath fp(name, 0); |
11983 | InodeRef in; | |
11984 | int rc; | |
11985 | unsigned mask = statx_to_mask(flags, want); | |
11986 | ||
11fdf7f2 TL |
11987 | ldout(cct, 3) << __func__ << " " << name << dendl; |
11988 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11989 | tout(cct) << name << std::endl; |
11990 | ||
f67539c2 | 11991 | std::scoped_lock lock(client_lock); |
7c673cae FG |
11992 | rc = path_walk(fp, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); |
11993 | if (rc < 0) { | |
11994 | /* zero out mask, just in case... */ | |
11995 | stx->stx_mask = 0; | |
11996 | stx->stx_ino = 0; | |
11997 | *out = NULL; | |
11998 | return rc; | |
11999 | } else { | |
11fdf7f2 | 12000 | ceph_assert(in); |
7c673cae FG |
12001 | fill_statx(in, mask, stx); |
12002 | _ll_get(in.get()); | |
12003 | *out = in.get(); | |
12004 | return 0; | |
12005 | } | |
12006 | } | |
12007 | ||
12008 | void Client::_ll_get(Inode *in) | |
12009 | { | |
12010 | if (in->ll_ref == 0) { | |
b3b6e05e | 12011 | in->iget(); |
11fdf7f2 TL |
12012 | if (in->is_dir() && !in->dentries.empty()) { |
12013 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
12014 | in->get_first_parent()->get(); // pin dentry |
12015 | } | |
11fdf7f2 TL |
12016 | if (in->snapid != CEPH_NOSNAP) |
12017 | ll_snap_ref[in->snapid]++; | |
7c673cae FG |
12018 | } |
12019 | in->ll_get(); | |
11fdf7f2 | 12020 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " -> " << in->ll_ref << dendl; |
7c673cae FG |
12021 | } |
12022 | ||
494da23a | 12023 | int Client::_ll_put(Inode *in, uint64_t num) |
7c673cae FG |
12024 | { |
12025 | in->ll_put(num); | |
11fdf7f2 | 12026 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " " << num << " -> " << in->ll_ref << dendl; |
7c673cae | 12027 | if (in->ll_ref == 0) { |
11fdf7f2 TL |
12028 | if (in->is_dir() && !in->dentries.empty()) { |
12029 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
12030 | in->get_first_parent()->put(); // unpin dentry |
12031 | } | |
11fdf7f2 TL |
12032 | if (in->snapid != CEPH_NOSNAP) { |
12033 | auto p = ll_snap_ref.find(in->snapid); | |
12034 | ceph_assert(p != ll_snap_ref.end()); | |
12035 | ceph_assert(p->second > 0); | |
12036 | if (--p->second == 0) | |
12037 | ll_snap_ref.erase(p); | |
12038 | } | |
7c673cae FG |
12039 | put_inode(in); |
12040 | return 0; | |
12041 | } else { | |
12042 | return in->ll_ref; | |
12043 | } | |
12044 | } | |
12045 | ||
12046 | void Client::_ll_drop_pins() | |
12047 | { | |
11fdf7f2 | 12048 | ldout(cct, 10) << __func__ << dendl; |
1adf2230 | 12049 | std::set<InodeRef> to_be_put; //this set will be deconstructed item by item when exit |
7c673cae FG |
12050 | ceph::unordered_map<vinodeno_t, Inode*>::iterator next; |
12051 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
12052 | it != inode_map.end(); | |
12053 | it = next) { | |
12054 | Inode *in = it->second; | |
12055 | next = it; | |
12056 | ++next; | |
1adf2230 AA |
12057 | if (in->ll_ref){ |
12058 | to_be_put.insert(in); | |
7c673cae | 12059 | _ll_put(in, in->ll_ref); |
1adf2230 | 12060 | } |
7c673cae FG |
12061 | } |
12062 | } | |
12063 | ||
494da23a | 12064 | bool Client::_ll_forget(Inode *in, uint64_t count) |
7c673cae | 12065 | { |
11fdf7f2 | 12066 | inodeno_t ino = in->ino; |
7c673cae | 12067 | |
11fdf7f2 TL |
12068 | ldout(cct, 8) << __func__ << " " << ino << " " << count << dendl; |
12069 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12070 | tout(cct) << ino.val << std::endl; |
12071 | tout(cct) << count << std::endl; | |
12072 | ||
181888fb | 12073 | // Ignore forget if we're no longer mounted |
f67539c2 TL |
12074 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12075 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
12076 | return true; |
12077 | ||
7c673cae FG |
12078 | if (ino == 1) return true; // ignore forget on root. |
12079 | ||
12080 | bool last = false; | |
12081 | if (in->ll_ref < count) { | |
12082 | ldout(cct, 1) << "WARNING: ll_forget on " << ino << " " << count | |
12083 | << ", which only has ll_ref=" << in->ll_ref << dendl; | |
12084 | _ll_put(in, in->ll_ref); | |
12085 | last = true; | |
12086 | } else { | |
12087 | if (_ll_put(in, count) == 0) | |
12088 | last = true; | |
12089 | } | |
12090 | ||
12091 | return last; | |
12092 | } | |
12093 | ||
494da23a | 12094 | bool Client::ll_forget(Inode *in, uint64_t count) |
1adf2230 | 12095 | { |
f67539c2 | 12096 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
12097 | return _ll_forget(in, count); |
12098 | } | |
12099 | ||
7c673cae FG |
12100 | bool Client::ll_put(Inode *in) |
12101 | { | |
12102 | /* ll_forget already takes the lock */ | |
12103 | return ll_forget(in, 1); | |
12104 | } | |
12105 | ||
11fdf7f2 TL |
12106 | int Client::ll_get_snap_ref(snapid_t snap) |
12107 | { | |
f67539c2 | 12108 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
12109 | auto p = ll_snap_ref.find(snap); |
12110 | if (p != ll_snap_ref.end()) | |
12111 | return p->second; | |
12112 | return 0; | |
12113 | } | |
12114 | ||
7c673cae FG |
12115 | snapid_t Client::ll_get_snapid(Inode *in) |
12116 | { | |
f67539c2 | 12117 | std::scoped_lock lock(client_lock); |
7c673cae FG |
12118 | return in->snapid; |
12119 | } | |
12120 | ||
12121 | Inode *Client::ll_get_inode(ino_t ino) | |
12122 | { | |
f67539c2 TL |
12123 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12124 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
12125 | return NULL; |
12126 | ||
f67539c2 TL |
12127 | std::scoped_lock lock(client_lock); |
12128 | ||
7c673cae FG |
12129 | vinodeno_t vino = _map_faked_ino(ino); |
12130 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
12131 | if (p == inode_map.end()) | |
12132 | return NULL; | |
12133 | Inode *in = p->second; | |
12134 | _ll_get(in); | |
12135 | return in; | |
12136 | } | |
12137 | ||
12138 | Inode *Client::ll_get_inode(vinodeno_t vino) | |
12139 | { | |
f67539c2 TL |
12140 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12141 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
12142 | return NULL; |
12143 | ||
b3b6e05e TL |
12144 | if (is_reserved_vino(vino)) |
12145 | return NULL; | |
12146 | ||
f67539c2 TL |
12147 | std::scoped_lock lock(client_lock); |
12148 | ||
7c673cae FG |
12149 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
12150 | if (p == inode_map.end()) | |
12151 | return NULL; | |
12152 | Inode *in = p->second; | |
12153 | _ll_get(in); | |
12154 | return in; | |
12155 | } | |
12156 | ||
12157 | int Client::_ll_getattr(Inode *in, int caps, const UserPerm& perms) | |
12158 | { | |
12159 | vinodeno_t vino = _get_vino(in); | |
12160 | ||
11fdf7f2 TL |
12161 | ldout(cct, 8) << __func__ << " " << vino << dendl; |
12162 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12163 | tout(cct) << vino.ino.val << std::endl; |
12164 | ||
12165 | if (vino.snapid < CEPH_NOSNAP) | |
12166 | return 0; | |
12167 | else | |
12168 | return _getattr(in, caps, perms); | |
12169 | } | |
12170 | ||
12171 | int Client::ll_getattr(Inode *in, struct stat *attr, const UserPerm& perms) | |
12172 | { | |
f67539c2 TL |
12173 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12174 | if (!mref_reader.is_state_satisfied()) | |
12175 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12176 | |
f67539c2 | 12177 | std::scoped_lock lock(client_lock); |
181888fb | 12178 | |
7c673cae FG |
12179 | int res = _ll_getattr(in, CEPH_STAT_CAP_INODE_ALL, perms); |
12180 | ||
12181 | if (res == 0) | |
12182 | fill_stat(in, attr); | |
11fdf7f2 | 12183 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12184 | return res; |
12185 | } | |
12186 | ||
12187 | int Client::ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want, | |
12188 | unsigned int flags, const UserPerm& perms) | |
12189 | { | |
f67539c2 TL |
12190 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12191 | if (!mref_reader.is_state_satisfied()) | |
12192 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12193 | |
f67539c2 | 12194 | std::scoped_lock lock(client_lock); |
181888fb | 12195 | |
7c673cae FG |
12196 | int res = 0; |
12197 | unsigned mask = statx_to_mask(flags, want); | |
12198 | ||
94b18763 | 12199 | if (mask && !in->caps_issued_mask(mask, true)) |
7c673cae FG |
12200 | res = _ll_getattr(in, mask, perms); |
12201 | ||
12202 | if (res == 0) | |
12203 | fill_statx(in, mask, stx); | |
11fdf7f2 | 12204 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12205 | return res; |
12206 | } | |
12207 | ||
12208 | int Client::_ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
12209 | const UserPerm& perms, InodeRef *inp) | |
12210 | { | |
12211 | vinodeno_t vino = _get_vino(in); | |
12212 | ||
11fdf7f2 | 12213 | ldout(cct, 8) << __func__ << " " << vino << " mask " << hex << mask << dec |
7c673cae | 12214 | << dendl; |
11fdf7f2 | 12215 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
12216 | tout(cct) << vino.ino.val << std::endl; |
12217 | tout(cct) << stx->stx_mode << std::endl; | |
12218 | tout(cct) << stx->stx_uid << std::endl; | |
12219 | tout(cct) << stx->stx_gid << std::endl; | |
12220 | tout(cct) << stx->stx_size << std::endl; | |
12221 | tout(cct) << stx->stx_mtime << std::endl; | |
12222 | tout(cct) << stx->stx_atime << std::endl; | |
12223 | tout(cct) << stx->stx_btime << std::endl; | |
12224 | tout(cct) << mask << std::endl; | |
12225 | ||
11fdf7f2 | 12226 | if (!fuse_default_permissions) { |
7c673cae FG |
12227 | int res = may_setattr(in, stx, mask, perms); |
12228 | if (res < 0) | |
12229 | return res; | |
12230 | } | |
12231 | ||
12232 | mask &= ~(CEPH_SETATTR_MTIME_NOW | CEPH_SETATTR_ATIME_NOW); | |
12233 | ||
12234 | return __setattrx(in, stx, mask, perms, inp); | |
12235 | } | |
12236 | ||
12237 | int Client::ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
12238 | const UserPerm& perms) | |
12239 | { | |
f67539c2 TL |
12240 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12241 | if (!mref_reader.is_state_satisfied()) | |
12242 | return -CEPHFS_ENOTCONN; | |
181888fb | 12243 | |
f67539c2 | 12244 | std::scoped_lock lock(client_lock); |
181888fb | 12245 | |
7c673cae FG |
12246 | InodeRef target(in); |
12247 | int res = _ll_setattrx(in, stx, mask, perms, &target); | |
12248 | if (res == 0) { | |
11fdf7f2 | 12249 | ceph_assert(in == target.get()); |
7c673cae FG |
12250 | fill_statx(in, in->caps_issued(), stx); |
12251 | } | |
12252 | ||
11fdf7f2 | 12253 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12254 | return res; |
12255 | } | |
12256 | ||
12257 | int Client::ll_setattr(Inode *in, struct stat *attr, int mask, | |
12258 | const UserPerm& perms) | |
12259 | { | |
12260 | struct ceph_statx stx; | |
12261 | stat_to_statx(attr, &stx); | |
12262 | ||
f67539c2 TL |
12263 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12264 | if (!mref_reader.is_state_satisfied()) | |
12265 | return -CEPHFS_ENOTCONN; | |
181888fb | 12266 | |
f67539c2 | 12267 | std::scoped_lock lock(client_lock); |
181888fb | 12268 | |
7c673cae FG |
12269 | InodeRef target(in); |
12270 | int res = _ll_setattrx(in, &stx, mask, perms, &target); | |
12271 | if (res == 0) { | |
11fdf7f2 | 12272 | ceph_assert(in == target.get()); |
7c673cae FG |
12273 | fill_stat(in, attr); |
12274 | } | |
12275 | ||
11fdf7f2 | 12276 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
12277 | return res; |
12278 | } | |
12279 | ||
12280 | ||
12281 | // ---------- | |
12282 | // xattrs | |
12283 | ||
12284 | int Client::getxattr(const char *path, const char *name, void *value, size_t size, | |
12285 | const UserPerm& perms) | |
12286 | { | |
f67539c2 TL |
12287 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12288 | if (!mref_reader.is_state_satisfied()) | |
12289 | return -CEPHFS_ENOTCONN; | |
181888fb | 12290 | |
f67539c2 | 12291 | std::scoped_lock lock(client_lock); |
181888fb | 12292 | |
7c673cae FG |
12293 | InodeRef in; |
12294 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12295 | if (r < 0) | |
12296 | return r; | |
12297 | return _getxattr(in, name, value, size, perms); | |
12298 | } | |
12299 | ||
12300 | int Client::lgetxattr(const char *path, const char *name, void *value, size_t size, | |
12301 | const UserPerm& perms) | |
12302 | { | |
f67539c2 TL |
12303 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12304 | if (!mref_reader.is_state_satisfied()) | |
12305 | return -CEPHFS_ENOTCONN; | |
181888fb | 12306 | |
f67539c2 | 12307 | std::scoped_lock lock(client_lock); |
181888fb | 12308 | |
7c673cae FG |
12309 | InodeRef in; |
12310 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12311 | if (r < 0) | |
12312 | return r; | |
12313 | return _getxattr(in, name, value, size, perms); | |
12314 | } | |
12315 | ||
12316 | int Client::fgetxattr(int fd, const char *name, void *value, size_t size, | |
12317 | const UserPerm& perms) | |
12318 | { | |
f67539c2 TL |
12319 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12320 | if (!mref_reader.is_state_satisfied()) | |
12321 | return -CEPHFS_ENOTCONN; | |
181888fb | 12322 | |
f67539c2 | 12323 | std::scoped_lock lock(client_lock); |
181888fb | 12324 | |
7c673cae FG |
12325 | Fh *f = get_filehandle(fd); |
12326 | if (!f) | |
f67539c2 | 12327 | return -CEPHFS_EBADF; |
7c673cae FG |
12328 | return _getxattr(f->inode, name, value, size, perms); |
12329 | } | |
12330 | ||
12331 | int Client::listxattr(const char *path, char *list, size_t size, | |
12332 | const UserPerm& perms) | |
12333 | { | |
f67539c2 TL |
12334 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12335 | if (!mref_reader.is_state_satisfied()) | |
12336 | return -CEPHFS_ENOTCONN; | |
181888fb | 12337 | |
f67539c2 | 12338 | std::scoped_lock lock(client_lock); |
181888fb | 12339 | |
7c673cae FG |
12340 | InodeRef in; |
12341 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12342 | if (r < 0) | |
12343 | return r; | |
12344 | return Client::_listxattr(in.get(), list, size, perms); | |
12345 | } | |
12346 | ||
12347 | int Client::llistxattr(const char *path, char *list, size_t size, | |
12348 | const UserPerm& perms) | |
12349 | { | |
f67539c2 TL |
12350 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12351 | if (!mref_reader.is_state_satisfied()) | |
12352 | return -CEPHFS_ENOTCONN; | |
181888fb | 12353 | |
f67539c2 | 12354 | std::scoped_lock lock(client_lock); |
181888fb | 12355 | |
7c673cae FG |
12356 | InodeRef in; |
12357 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12358 | if (r < 0) | |
12359 | return r; | |
12360 | return Client::_listxattr(in.get(), list, size, perms); | |
12361 | } | |
12362 | ||
12363 | int Client::flistxattr(int fd, char *list, size_t size, const UserPerm& perms) | |
12364 | { | |
f67539c2 TL |
12365 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12366 | if (!mref_reader.is_state_satisfied()) | |
12367 | return -CEPHFS_ENOTCONN; | |
181888fb | 12368 | |
f67539c2 | 12369 | std::scoped_lock lock(client_lock); |
181888fb | 12370 | |
7c673cae FG |
12371 | Fh *f = get_filehandle(fd); |
12372 | if (!f) | |
f67539c2 | 12373 | return -CEPHFS_EBADF; |
7c673cae FG |
12374 | return Client::_listxattr(f->inode.get(), list, size, perms); |
12375 | } | |
12376 | ||
12377 | int Client::removexattr(const char *path, const char *name, | |
12378 | const UserPerm& perms) | |
12379 | { | |
f67539c2 TL |
12380 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12381 | if (!mref_reader.is_state_satisfied()) | |
12382 | return -CEPHFS_ENOTCONN; | |
181888fb | 12383 | |
f67539c2 | 12384 | std::scoped_lock lock(client_lock); |
181888fb | 12385 | |
7c673cae FG |
12386 | InodeRef in; |
12387 | int r = Client::path_walk(path, &in, perms, true); | |
12388 | if (r < 0) | |
12389 | return r; | |
12390 | return _removexattr(in, name, perms); | |
12391 | } | |
12392 | ||
12393 | int Client::lremovexattr(const char *path, const char *name, | |
12394 | const UserPerm& perms) | |
12395 | { | |
f67539c2 TL |
12396 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12397 | if (!mref_reader.is_state_satisfied()) | |
12398 | return -CEPHFS_ENOTCONN; | |
181888fb | 12399 | |
f67539c2 | 12400 | std::scoped_lock lock(client_lock); |
181888fb | 12401 | |
7c673cae FG |
12402 | InodeRef in; |
12403 | int r = Client::path_walk(path, &in, perms, false); | |
12404 | if (r < 0) | |
12405 | return r; | |
12406 | return _removexattr(in, name, perms); | |
12407 | } | |
12408 | ||
12409 | int Client::fremovexattr(int fd, const char *name, const UserPerm& perms) | |
12410 | { | |
f67539c2 TL |
12411 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12412 | if (!mref_reader.is_state_satisfied()) | |
12413 | return -CEPHFS_ENOTCONN; | |
181888fb | 12414 | |
f67539c2 | 12415 | std::scoped_lock lock(client_lock); |
181888fb | 12416 | |
7c673cae FG |
12417 | Fh *f = get_filehandle(fd); |
12418 | if (!f) | |
f67539c2 | 12419 | return -CEPHFS_EBADF; |
7c673cae FG |
12420 | return _removexattr(f->inode, name, perms); |
12421 | } | |
12422 | ||
12423 | int Client::setxattr(const char *path, const char *name, const void *value, | |
12424 | size_t size, int flags, const UserPerm& perms) | |
12425 | { | |
f67539c2 TL |
12426 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12427 | if (!mref_reader.is_state_satisfied()) | |
12428 | return -CEPHFS_ENOTCONN; | |
12429 | ||
7c673cae FG |
12430 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
12431 | ||
f67539c2 | 12432 | std::scoped_lock lock(client_lock); |
181888fb | 12433 | |
7c673cae FG |
12434 | InodeRef in; |
12435 | int r = Client::path_walk(path, &in, perms, true); | |
12436 | if (r < 0) | |
12437 | return r; | |
12438 | return _setxattr(in, name, value, size, flags, perms); | |
12439 | } | |
12440 | ||
12441 | int Client::lsetxattr(const char *path, const char *name, const void *value, | |
12442 | size_t size, int flags, const UserPerm& perms) | |
12443 | { | |
f67539c2 TL |
12444 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12445 | if (!mref_reader.is_state_satisfied()) | |
12446 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12447 | |
f67539c2 | 12448 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12449 | |
f67539c2 | 12450 | std::scoped_lock lock(client_lock); |
181888fb | 12451 | |
7c673cae FG |
12452 | InodeRef in; |
12453 | int r = Client::path_walk(path, &in, perms, false); | |
12454 | if (r < 0) | |
12455 | return r; | |
12456 | return _setxattr(in, name, value, size, flags, perms); | |
12457 | } | |
12458 | ||
12459 | int Client::fsetxattr(int fd, const char *name, const void *value, size_t size, | |
12460 | int flags, const UserPerm& perms) | |
12461 | { | |
f67539c2 TL |
12462 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12463 | if (!mref_reader.is_state_satisfied()) | |
12464 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12465 | |
f67539c2 | 12466 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12467 | |
f67539c2 | 12468 | std::scoped_lock lock(client_lock); |
181888fb | 12469 | |
7c673cae FG |
12470 | Fh *f = get_filehandle(fd); |
12471 | if (!f) | |
f67539c2 | 12472 | return -CEPHFS_EBADF; |
7c673cae FG |
12473 | return _setxattr(f->inode, name, value, size, flags, perms); |
12474 | } | |
12475 | ||
12476 | int Client::_getxattr(Inode *in, const char *name, void *value, size_t size, | |
12477 | const UserPerm& perms) | |
12478 | { | |
12479 | int r; | |
1d09f67e | 12480 | const VXattr *vxattr = nullptr; |
7c673cae | 12481 | |
1d09f67e | 12482 | vxattr = _match_vxattr(in, name); |
7c673cae | 12483 | if (vxattr) { |
f67539c2 | 12484 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12485 | |
12486 | // Do a force getattr to get the latest quota before returning | |
12487 | // a value to userspace. | |
28e407b8 AA |
12488 | int flags = 0; |
12489 | if (vxattr->flags & VXATTR_RSTAT) { | |
12490 | flags |= CEPH_STAT_RSTAT; | |
12491 | } | |
adb31ebb TL |
12492 | if (vxattr->flags & VXATTR_DIRSTAT) { |
12493 | flags |= CEPH_CAP_FILE_SHARED; | |
12494 | } | |
f67539c2 | 12495 | r = _getattr(in, flags | CEPH_STAT_CAP_XATTR, perms, true); |
7c673cae FG |
12496 | if (r != 0) { |
12497 | // Error from getattr! | |
12498 | return r; | |
12499 | } | |
12500 | ||
12501 | // call pointer-to-member function | |
12502 | char buf[256]; | |
12503 | if (!(vxattr->exists_cb && !(this->*(vxattr->exists_cb))(in))) { | |
12504 | r = (this->*(vxattr->getxattr_cb))(in, buf, sizeof(buf)); | |
12505 | } else { | |
f67539c2 | 12506 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12507 | } |
12508 | ||
12509 | if (size != 0) { | |
12510 | if (r > (int)size) { | |
f67539c2 | 12511 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12512 | } else if (r > 0) { |
12513 | memcpy(value, buf, r); | |
12514 | } | |
12515 | } | |
12516 | goto out; | |
12517 | } | |
12518 | ||
1d09f67e TL |
12519 | if (!strncmp(name, "ceph.", 5)) { |
12520 | r = _getvxattr(in, perms, name, size, value, MDS_RANK_NONE); | |
12521 | goto out; | |
12522 | } | |
12523 | ||
7c673cae | 12524 | if (acl_type == NO_ACL && !strncmp(name, "system.", 7)) { |
f67539c2 | 12525 | r = -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12526 | goto out; |
12527 | } | |
12528 | ||
12529 | r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
12530 | if (r == 0) { | |
12531 | string n(name); | |
f67539c2 | 12532 | r = -CEPHFS_ENODATA; |
1d09f67e | 12533 | if (in->xattrs.count(n)) { |
7c673cae FG |
12534 | r = in->xattrs[n].length(); |
12535 | if (r > 0 && size != 0) { | |
12536 | if (size >= (unsigned)r) | |
12537 | memcpy(value, in->xattrs[n].c_str(), r); | |
12538 | else | |
f67539c2 | 12539 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12540 | } |
12541 | } | |
12542 | } | |
12543 | out: | |
1adf2230 | 12544 | ldout(cct, 8) << "_getxattr(" << in->ino << ", \"" << name << "\", " << size << ") = " << r << dendl; |
7c673cae FG |
12545 | return r; |
12546 | } | |
12547 | ||
12548 | int Client::_getxattr(InodeRef &in, const char *name, void *value, size_t size, | |
12549 | const UserPerm& perms) | |
12550 | { | |
12551 | if (cct->_conf->client_permissions) { | |
12552 | int r = xattr_permission(in.get(), name, MAY_READ, perms); | |
12553 | if (r < 0) | |
12554 | return r; | |
12555 | } | |
12556 | return _getxattr(in.get(), name, value, size, perms); | |
12557 | } | |
12558 | ||
12559 | int Client::ll_getxattr(Inode *in, const char *name, void *value, | |
12560 | size_t size, const UserPerm& perms) | |
12561 | { | |
f67539c2 TL |
12562 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12563 | if (!mref_reader.is_state_satisfied()) | |
12564 | return -CEPHFS_ENOTCONN; | |
181888fb | 12565 | |
7c673cae FG |
12566 | vinodeno_t vino = _get_vino(in); |
12567 | ||
11fdf7f2 TL |
12568 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12569 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12570 | tout(cct) << vino.ino.val << std::endl; |
12571 | tout(cct) << name << std::endl; | |
12572 | ||
f67539c2 | 12573 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12574 | if (!fuse_default_permissions) { |
7c673cae FG |
12575 | int r = xattr_permission(in, name, MAY_READ, perms); |
12576 | if (r < 0) | |
12577 | return r; | |
12578 | } | |
12579 | ||
12580 | return _getxattr(in, name, value, size, perms); | |
12581 | } | |
12582 | ||
12583 | int Client::_listxattr(Inode *in, char *name, size_t size, | |
12584 | const UserPerm& perms) | |
12585 | { | |
81eedcae | 12586 | bool len_only = (size == 0); |
7c673cae | 12587 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); |
81eedcae TL |
12588 | if (r != 0) { |
12589 | goto out; | |
12590 | } | |
7c673cae | 12591 | |
81eedcae | 12592 | r = 0; |
f67539c2 TL |
12593 | for ([[maybe_unused]] const auto &[xattr_name, xattr_value_bl] : in->xattrs) { |
12594 | if (xattr_name.rfind("ceph.", 0) == 0) { | |
12595 | continue; | |
12596 | } | |
12597 | ||
12598 | size_t this_len = xattr_name.length() + 1; | |
81eedcae TL |
12599 | r += this_len; |
12600 | if (len_only) | |
12601 | continue; | |
7c673cae | 12602 | |
81eedcae | 12603 | if (this_len > size) { |
f67539c2 | 12604 | r = -CEPHFS_ERANGE; |
81eedcae TL |
12605 | goto out; |
12606 | } | |
12607 | ||
f67539c2 | 12608 | memcpy(name, xattr_name.c_str(), this_len); |
81eedcae TL |
12609 | name += this_len; |
12610 | size -= this_len; | |
12611 | } | |
81eedcae | 12612 | out: |
11fdf7f2 | 12613 | ldout(cct, 8) << __func__ << "(" << in->ino << ", " << size << ") = " << r << dendl; |
7c673cae FG |
12614 | return r; |
12615 | } | |
12616 | ||
12617 | int Client::ll_listxattr(Inode *in, char *names, size_t size, | |
12618 | const UserPerm& perms) | |
12619 | { | |
f67539c2 TL |
12620 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12621 | if (!mref_reader.is_state_satisfied()) | |
12622 | return -CEPHFS_ENOTCONN; | |
181888fb | 12623 | |
7c673cae FG |
12624 | vinodeno_t vino = _get_vino(in); |
12625 | ||
11fdf7f2 TL |
12626 | ldout(cct, 3) << __func__ << " " << vino << " size " << size << dendl; |
12627 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12628 | tout(cct) << vino.ino.val << std::endl; |
12629 | tout(cct) << size << std::endl; | |
12630 | ||
f67539c2 | 12631 | std::scoped_lock lock(client_lock); |
7c673cae FG |
12632 | return _listxattr(in, names, size, perms); |
12633 | } | |
12634 | ||
12635 | int Client::_do_setxattr(Inode *in, const char *name, const void *value, | |
12636 | size_t size, int flags, const UserPerm& perms) | |
12637 | { | |
12638 | ||
12639 | int xattr_flags = 0; | |
12640 | if (!value) | |
12641 | xattr_flags |= CEPH_XATTR_REMOVE; | |
12642 | if (flags & XATTR_CREATE) | |
12643 | xattr_flags |= CEPH_XATTR_CREATE; | |
12644 | if (flags & XATTR_REPLACE) | |
12645 | xattr_flags |= CEPH_XATTR_REPLACE; | |
12646 | ||
12647 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETXATTR); | |
12648 | filepath path; | |
12649 | in->make_nosnap_relative_path(path); | |
12650 | req->set_filepath(path); | |
12651 | req->set_string2(name); | |
12652 | req->set_inode(in); | |
12653 | req->head.args.setxattr.flags = xattr_flags; | |
12654 | ||
12655 | bufferlist bl; | |
20effc67 | 12656 | ceph_assert(value || size == 0); |
7c673cae FG |
12657 | bl.append((const char*)value, size); |
12658 | req->set_data(bl); | |
12659 | ||
12660 | int res = make_request(req, perms); | |
12661 | ||
12662 | trim_cache(); | |
11fdf7f2 | 12663 | ldout(cct, 3) << __func__ << "(" << in->ino << ", \"" << name << "\") = " << |
7c673cae FG |
12664 | res << dendl; |
12665 | return res; | |
12666 | } | |
12667 | ||
12668 | int Client::_setxattr(Inode *in, const char *name, const void *value, | |
12669 | size_t size, int flags, const UserPerm& perms) | |
12670 | { | |
12671 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12672 | return -CEPHFS_EROFS; |
7c673cae FG |
12673 | } |
12674 | ||
f6b5b4d7 TL |
12675 | if (size == 0) { |
12676 | value = ""; | |
12677 | } else if (value == NULL) { | |
f67539c2 | 12678 | return -CEPHFS_EINVAL; |
f6b5b4d7 TL |
12679 | } |
12680 | ||
7c673cae FG |
12681 | bool posix_acl_xattr = false; |
12682 | if (acl_type == POSIX_ACL) | |
12683 | posix_acl_xattr = !strncmp(name, "system.", 7); | |
12684 | ||
12685 | if (strncmp(name, "user.", 5) && | |
12686 | strncmp(name, "security.", 9) && | |
12687 | strncmp(name, "trusted.", 8) && | |
12688 | strncmp(name, "ceph.", 5) && | |
12689 | !posix_acl_xattr) | |
f67539c2 | 12690 | return -CEPHFS_EOPNOTSUPP; |
7c673cae | 12691 | |
11fdf7f2 TL |
12692 | bool check_realm = false; |
12693 | ||
7c673cae FG |
12694 | if (posix_acl_xattr) { |
12695 | if (!strcmp(name, ACL_EA_ACCESS)) { | |
12696 | mode_t new_mode = in->mode; | |
12697 | if (value) { | |
12698 | int ret = posix_acl_equiv_mode(value, size, &new_mode); | |
12699 | if (ret < 0) | |
12700 | return ret; | |
12701 | if (ret == 0) { | |
12702 | value = NULL; | |
12703 | size = 0; | |
12704 | } | |
12705 | if (new_mode != in->mode) { | |
12706 | struct ceph_statx stx; | |
12707 | stx.stx_mode = new_mode; | |
12708 | ret = _do_setattr(in, &stx, CEPH_SETATTR_MODE, perms, NULL); | |
12709 | if (ret < 0) | |
12710 | return ret; | |
12711 | } | |
12712 | } | |
12713 | } else if (!strcmp(name, ACL_EA_DEFAULT)) { | |
12714 | if (value) { | |
12715 | if (!S_ISDIR(in->mode)) | |
f67539c2 | 12716 | return -CEPHFS_EACCES; |
7c673cae FG |
12717 | int ret = posix_acl_check(value, size); |
12718 | if (ret < 0) | |
f67539c2 | 12719 | return -CEPHFS_EINVAL; |
7c673cae FG |
12720 | if (ret == 0) { |
12721 | value = NULL; | |
12722 | size = 0; | |
12723 | } | |
12724 | } | |
12725 | } else { | |
f67539c2 | 12726 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12727 | } |
12728 | } else { | |
12729 | const VXattr *vxattr = _match_vxattr(in, name); | |
11fdf7f2 TL |
12730 | if (vxattr) { |
12731 | if (vxattr->readonly) | |
f67539c2 | 12732 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12733 | if (vxattr->name.compare(0, 10, "ceph.quota") == 0 && value) |
12734 | check_realm = true; | |
12735 | } | |
7c673cae FG |
12736 | } |
12737 | ||
11fdf7f2 TL |
12738 | int ret = _do_setxattr(in, name, value, size, flags, perms); |
12739 | if (ret >= 0 && check_realm) { | |
12740 | // check if snaprealm was created for quota inode | |
12741 | if (in->quota.is_enable() && | |
12742 | !(in->snaprealm && in->snaprealm->ino == in->ino)) | |
f67539c2 | 12743 | ret = -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12744 | } |
12745 | ||
12746 | return ret; | |
7c673cae FG |
12747 | } |
12748 | ||
12749 | int Client::_setxattr(InodeRef &in, const char *name, const void *value, | |
12750 | size_t size, int flags, const UserPerm& perms) | |
12751 | { | |
12752 | if (cct->_conf->client_permissions) { | |
12753 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12754 | if (r < 0) | |
12755 | return r; | |
12756 | } | |
12757 | return _setxattr(in.get(), name, value, size, flags, perms); | |
12758 | } | |
12759 | ||
12760 | int Client::_setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap) | |
12761 | { | |
12762 | string tmp; | |
12763 | if (name == "layout") { | |
12764 | string::iterator begin = value.begin(); | |
12765 | string::iterator end = value.end(); | |
12766 | keys_and_values<string::iterator> p; // create instance of parser | |
12767 | std::map<string, string> m; // map to receive results | |
12768 | if (!qi::parse(begin, end, p, m)) { // returns true if successful | |
f67539c2 | 12769 | return -CEPHFS_EINVAL; |
7c673cae FG |
12770 | } |
12771 | if (begin != end) | |
f67539c2 | 12772 | return -CEPHFS_EINVAL; |
7c673cae FG |
12773 | for (map<string,string>::iterator q = m.begin(); q != m.end(); ++q) { |
12774 | if (q->first == "pool") { | |
12775 | tmp = q->second; | |
12776 | break; | |
12777 | } | |
12778 | } | |
12779 | } else if (name == "layout.pool") { | |
12780 | tmp = value; | |
12781 | } | |
12782 | ||
12783 | if (tmp.length()) { | |
12784 | int64_t pool; | |
12785 | try { | |
12786 | pool = boost::lexical_cast<unsigned>(tmp); | |
12787 | if (!osdmap->have_pg_pool(pool)) | |
f67539c2 | 12788 | return -CEPHFS_ENOENT; |
7c673cae FG |
12789 | } catch (boost::bad_lexical_cast const&) { |
12790 | pool = osdmap->lookup_pg_pool_name(tmp); | |
12791 | if (pool < 0) { | |
f67539c2 | 12792 | return -CEPHFS_ENOENT; |
7c673cae FG |
12793 | } |
12794 | } | |
12795 | } | |
12796 | ||
12797 | return 0; | |
12798 | } | |
12799 | ||
12800 | void Client::_setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t size) | |
12801 | { | |
12802 | // For setting pool of layout, MetaRequest need osdmap epoch. | |
12803 | // There is a race which create a new data pool but client and mds both don't have. | |
12804 | // Make client got the latest osdmap which make mds quickly judge whether get newer osdmap. | |
f67539c2 | 12805 | ldout(cct, 15) << __func__ << ": name = " << name << dendl; |
7c673cae FG |
12806 | if (strcmp(name, "ceph.file.layout.pool") == 0 || strcmp(name, "ceph.dir.layout.pool") == 0 || |
12807 | strcmp(name, "ceph.file.layout") == 0 || strcmp(name, "ceph.dir.layout") == 0) { | |
12808 | string rest(strstr(name, "layout")); | |
12809 | string v((const char*)value, size); | |
12810 | int r = objecter->with_osdmap([&](const OSDMap& o) { | |
12811 | return _setxattr_check_data_pool(rest, v, &o); | |
12812 | }); | |
12813 | ||
f67539c2 TL |
12814 | if (r == -CEPHFS_ENOENT) { |
12815 | bs::error_code ec; | |
12816 | ldout(cct, 20) << __func__ << ": waiting for latest osdmap" << dendl; | |
12817 | objecter->wait_for_latest_osdmap(ca::use_blocked[ec]); | |
12818 | ldout(cct, 20) << __func__ << ": got latest osdmap: " << ec << dendl; | |
7c673cae FG |
12819 | } |
12820 | } | |
12821 | } | |
12822 | ||
12823 | int Client::ll_setxattr(Inode *in, const char *name, const void *value, | |
12824 | size_t size, int flags, const UserPerm& perms) | |
12825 | { | |
f67539c2 TL |
12826 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12827 | if (!mref_reader.is_state_satisfied()) | |
12828 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12829 | |
f67539c2 | 12830 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12831 | |
7c673cae FG |
12832 | vinodeno_t vino = _get_vino(in); |
12833 | ||
11fdf7f2 TL |
12834 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12835 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12836 | tout(cct) << vino.ino.val << std::endl; |
12837 | tout(cct) << name << std::endl; | |
12838 | ||
f67539c2 | 12839 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12840 | if (!fuse_default_permissions) { |
7c673cae FG |
12841 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12842 | if (r < 0) | |
12843 | return r; | |
12844 | } | |
12845 | return _setxattr(in, name, value, size, flags, perms); | |
12846 | } | |
12847 | ||
12848 | int Client::_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12849 | { | |
12850 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12851 | return -CEPHFS_EROFS; |
7c673cae FG |
12852 | } |
12853 | ||
12854 | // same xattrs supported by kernel client | |
12855 | if (strncmp(name, "user.", 5) && | |
12856 | strncmp(name, "system.", 7) && | |
12857 | strncmp(name, "security.", 9) && | |
12858 | strncmp(name, "trusted.", 8) && | |
12859 | strncmp(name, "ceph.", 5)) | |
f67539c2 | 12860 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12861 | |
12862 | const VXattr *vxattr = _match_vxattr(in, name); | |
12863 | if (vxattr && vxattr->readonly) | |
f67539c2 | 12864 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12865 | |
12866 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_RMXATTR); | |
12867 | filepath path; | |
12868 | in->make_nosnap_relative_path(path); | |
12869 | req->set_filepath(path); | |
12870 | req->set_filepath2(name); | |
12871 | req->set_inode(in); | |
12872 | ||
12873 | int res = make_request(req, perms); | |
12874 | ||
12875 | trim_cache(); | |
1adf2230 | 12876 | ldout(cct, 8) << "_removexattr(" << in->ino << ", \"" << name << "\") = " << res << dendl; |
7c673cae FG |
12877 | return res; |
12878 | } | |
12879 | ||
12880 | int Client::_removexattr(InodeRef &in, const char *name, const UserPerm& perms) | |
12881 | { | |
12882 | if (cct->_conf->client_permissions) { | |
12883 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12884 | if (r < 0) | |
12885 | return r; | |
12886 | } | |
12887 | return _removexattr(in.get(), name, perms); | |
12888 | } | |
12889 | ||
12890 | int Client::ll_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12891 | { | |
f67539c2 TL |
12892 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12893 | if (!mref_reader.is_state_satisfied()) | |
12894 | return -CEPHFS_ENOTCONN; | |
181888fb | 12895 | |
7c673cae FG |
12896 | vinodeno_t vino = _get_vino(in); |
12897 | ||
12898 | ldout(cct, 3) << "ll_removexattr " << vino << " " << name << dendl; | |
12899 | tout(cct) << "ll_removexattr" << std::endl; | |
12900 | tout(cct) << vino.ino.val << std::endl; | |
12901 | tout(cct) << name << std::endl; | |
12902 | ||
f67539c2 | 12903 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12904 | if (!fuse_default_permissions) { |
7c673cae FG |
12905 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12906 | if (r < 0) | |
12907 | return r; | |
12908 | } | |
12909 | ||
12910 | return _removexattr(in, name, perms); | |
12911 | } | |
12912 | ||
12913 | bool Client::_vxattrcb_quota_exists(Inode *in) | |
12914 | { | |
11fdf7f2 | 12915 | return in->quota.is_enable() && |
f6b5b4d7 TL |
12916 | (in->snapid != CEPH_NOSNAP || |
12917 | (in->snaprealm && in->snaprealm->ino == in->ino)); | |
7c673cae FG |
12918 | } |
12919 | size_t Client::_vxattrcb_quota(Inode *in, char *val, size_t size) | |
12920 | { | |
12921 | return snprintf(val, size, | |
12922 | "max_bytes=%lld max_files=%lld", | |
12923 | (long long int)in->quota.max_bytes, | |
12924 | (long long int)in->quota.max_files); | |
12925 | } | |
12926 | size_t Client::_vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size) | |
12927 | { | |
12928 | return snprintf(val, size, "%lld", (long long int)in->quota.max_bytes); | |
12929 | } | |
12930 | size_t Client::_vxattrcb_quota_max_files(Inode *in, char *val, size_t size) | |
12931 | { | |
12932 | return snprintf(val, size, "%lld", (long long int)in->quota.max_files); | |
12933 | } | |
12934 | ||
12935 | bool Client::_vxattrcb_layout_exists(Inode *in) | |
12936 | { | |
12937 | return in->layout != file_layout_t(); | |
12938 | } | |
12939 | size_t Client::_vxattrcb_layout(Inode *in, char *val, size_t size) | |
12940 | { | |
12941 | int r = snprintf(val, size, | |
11fdf7f2 | 12942 | "stripe_unit=%llu stripe_count=%llu object_size=%llu pool=", |
7c673cae FG |
12943 | (unsigned long long)in->layout.stripe_unit, |
12944 | (unsigned long long)in->layout.stripe_count, | |
12945 | (unsigned long long)in->layout.object_size); | |
12946 | objecter->with_osdmap([&](const OSDMap& o) { | |
12947 | if (o.have_pg_pool(in->layout.pool_id)) | |
12948 | r += snprintf(val + r, size - r, "%s", | |
12949 | o.get_pool_name(in->layout.pool_id).c_str()); | |
12950 | else | |
12951 | r += snprintf(val + r, size - r, "%" PRIu64, | |
12952 | (uint64_t)in->layout.pool_id); | |
12953 | }); | |
12954 | if (in->layout.pool_ns.length()) | |
12955 | r += snprintf(val + r, size - r, " pool_namespace=%s", | |
12956 | in->layout.pool_ns.c_str()); | |
12957 | return r; | |
12958 | } | |
12959 | size_t Client::_vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size) | |
12960 | { | |
11fdf7f2 | 12961 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_unit); |
7c673cae FG |
12962 | } |
12963 | size_t Client::_vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size) | |
12964 | { | |
11fdf7f2 | 12965 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_count); |
7c673cae FG |
12966 | } |
12967 | size_t Client::_vxattrcb_layout_object_size(Inode *in, char *val, size_t size) | |
12968 | { | |
11fdf7f2 | 12969 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.object_size); |
7c673cae FG |
12970 | } |
12971 | size_t Client::_vxattrcb_layout_pool(Inode *in, char *val, size_t size) | |
12972 | { | |
12973 | size_t r; | |
12974 | objecter->with_osdmap([&](const OSDMap& o) { | |
12975 | if (o.have_pg_pool(in->layout.pool_id)) | |
12976 | r = snprintf(val, size, "%s", o.get_pool_name( | |
12977 | in->layout.pool_id).c_str()); | |
12978 | else | |
12979 | r = snprintf(val, size, "%" PRIu64, (uint64_t)in->layout.pool_id); | |
12980 | }); | |
12981 | return r; | |
12982 | } | |
12983 | size_t Client::_vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size) | |
12984 | { | |
12985 | return snprintf(val, size, "%s", in->layout.pool_ns.c_str()); | |
12986 | } | |
12987 | size_t Client::_vxattrcb_dir_entries(Inode *in, char *val, size_t size) | |
12988 | { | |
11fdf7f2 | 12989 | return snprintf(val, size, "%llu", (unsigned long long)(in->dirstat.nfiles + in->dirstat.nsubdirs)); |
7c673cae FG |
12990 | } |
12991 | size_t Client::_vxattrcb_dir_files(Inode *in, char *val, size_t size) | |
12992 | { | |
11fdf7f2 | 12993 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nfiles); |
7c673cae FG |
12994 | } |
12995 | size_t Client::_vxattrcb_dir_subdirs(Inode *in, char *val, size_t size) | |
12996 | { | |
11fdf7f2 | 12997 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nsubdirs); |
7c673cae FG |
12998 | } |
12999 | size_t Client::_vxattrcb_dir_rentries(Inode *in, char *val, size_t size) | |
13000 | { | |
11fdf7f2 | 13001 | return snprintf(val, size, "%llu", (unsigned long long)(in->rstat.rfiles + in->rstat.rsubdirs)); |
7c673cae FG |
13002 | } |
13003 | size_t Client::_vxattrcb_dir_rfiles(Inode *in, char *val, size_t size) | |
13004 | { | |
11fdf7f2 | 13005 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rfiles); |
7c673cae FG |
13006 | } |
13007 | size_t Client::_vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size) | |
13008 | { | |
11fdf7f2 | 13009 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsubdirs); |
7c673cae | 13010 | } |
f67539c2 TL |
13011 | size_t Client::_vxattrcb_dir_rsnaps(Inode *in, char *val, size_t size) |
13012 | { | |
13013 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsnaps); | |
13014 | } | |
7c673cae FG |
13015 | size_t Client::_vxattrcb_dir_rbytes(Inode *in, char *val, size_t size) |
13016 | { | |
11fdf7f2 | 13017 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rbytes); |
7c673cae FG |
13018 | } |
13019 | size_t Client::_vxattrcb_dir_rctime(Inode *in, char *val, size_t size) | |
13020 | { | |
81eedcae | 13021 | return snprintf(val, size, "%ld.%09ld", (long)in->rstat.rctime.sec(), |
7c673cae FG |
13022 | (long)in->rstat.rctime.nsec()); |
13023 | } | |
11fdf7f2 TL |
13024 | bool Client::_vxattrcb_dir_pin_exists(Inode *in) |
13025 | { | |
f67539c2 | 13026 | return in->dir_pin != -CEPHFS_ENODATA; |
11fdf7f2 TL |
13027 | } |
13028 | size_t Client::_vxattrcb_dir_pin(Inode *in, char *val, size_t size) | |
13029 | { | |
13030 | return snprintf(val, size, "%ld", (long)in->dir_pin); | |
13031 | } | |
7c673cae | 13032 | |
81eedcae TL |
13033 | bool Client::_vxattrcb_snap_btime_exists(Inode *in) |
13034 | { | |
13035 | return !in->snap_btime.is_zero(); | |
13036 | } | |
13037 | ||
13038 | size_t Client::_vxattrcb_snap_btime(Inode *in, char *val, size_t size) | |
13039 | { | |
13040 | return snprintf(val, size, "%llu.%09lu", | |
13041 | (long long unsigned)in->snap_btime.sec(), | |
13042 | (long unsigned)in->snap_btime.nsec()); | |
13043 | } | |
13044 | ||
20effc67 TL |
13045 | size_t Client::_vxattrcb_caps(Inode *in, char *val, size_t size) |
13046 | { | |
13047 | int issued; | |
13048 | ||
13049 | in->caps_issued(&issued); | |
13050 | return snprintf(val, size, "%s/0x%x", ccap_string(issued).c_str(), issued); | |
13051 | } | |
13052 | ||
f67539c2 TL |
13053 | bool Client::_vxattrcb_mirror_info_exists(Inode *in) |
13054 | { | |
13055 | // checking one of the xattrs would suffice | |
13056 | return in->xattrs.count("ceph.mirror.info.cluster_id") != 0; | |
13057 | } | |
13058 | ||
13059 | size_t Client::_vxattrcb_mirror_info(Inode *in, char *val, size_t size) | |
13060 | { | |
13061 | return snprintf(val, size, "cluster_id=%.*s fs_id=%.*s", | |
13062 | in->xattrs["ceph.mirror.info.cluster_id"].length(), | |
13063 | in->xattrs["ceph.mirror.info.cluster_id"].c_str(), | |
13064 | in->xattrs["ceph.mirror.info.fs_id"].length(), | |
13065 | in->xattrs["ceph.mirror.info.fs_id"].c_str()); | |
13066 | } | |
13067 | ||
adb31ebb TL |
13068 | size_t Client::_vxattrcb_cluster_fsid(Inode *in, char *val, size_t size) |
13069 | { | |
13070 | return snprintf(val, size, "%s", monclient->get_fsid().to_string().c_str()); | |
13071 | } | |
13072 | ||
13073 | size_t Client::_vxattrcb_client_id(Inode *in, char *val, size_t size) | |
13074 | { | |
13075 | auto name = messenger->get_myname(); | |
20effc67 | 13076 | return snprintf(val, size, "%s%" PRId64, name.type_str(), name.num()); |
adb31ebb TL |
13077 | } |
13078 | ||
7c673cae FG |
13079 | #define CEPH_XATTR_NAME(_type, _name) "ceph." #_type "." #_name |
13080 | #define CEPH_XATTR_NAME2(_type, _name, _name2) "ceph." #_type "." #_name "." #_name2 | |
13081 | ||
adb31ebb | 13082 | #define XATTR_NAME_CEPH(_type, _name, _flags) \ |
28e407b8 AA |
13083 | { \ |
13084 | name: CEPH_XATTR_NAME(_type, _name), \ | |
13085 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
13086 | readonly: true, \ | |
28e407b8 AA |
13087 | exists_cb: NULL, \ |
13088 | flags: _flags, \ | |
7c673cae FG |
13089 | } |
13090 | #define XATTR_LAYOUT_FIELD(_type, _name, _field) \ | |
13091 | { \ | |
13092 | name: CEPH_XATTR_NAME2(_type, _name, _field), \ | |
13093 | getxattr_cb: &Client::_vxattrcb_ ## _name ## _ ## _field, \ | |
13094 | readonly: false, \ | |
7c673cae | 13095 | exists_cb: &Client::_vxattrcb_layout_exists, \ |
28e407b8 | 13096 | flags: 0, \ |
7c673cae FG |
13097 | } |
13098 | #define XATTR_QUOTA_FIELD(_type, _name) \ | |
13099 | { \ | |
13100 | name: CEPH_XATTR_NAME(_type, _name), \ | |
13101 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
13102 | readonly: false, \ | |
7c673cae | 13103 | exists_cb: &Client::_vxattrcb_quota_exists, \ |
28e407b8 | 13104 | flags: 0, \ |
7c673cae FG |
13105 | } |
13106 | ||
13107 | const Client::VXattr Client::_dir_vxattrs[] = { | |
13108 | { | |
13109 | name: "ceph.dir.layout", | |
13110 | getxattr_cb: &Client::_vxattrcb_layout, | |
13111 | readonly: false, | |
7c673cae | 13112 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 13113 | flags: 0, |
7c673cae | 13114 | }, |
1d09f67e TL |
13115 | // FIXME |
13116 | // Delete the following dir layout field definitions for release "S" | |
7c673cae FG |
13117 | XATTR_LAYOUT_FIELD(dir, layout, stripe_unit), |
13118 | XATTR_LAYOUT_FIELD(dir, layout, stripe_count), | |
13119 | XATTR_LAYOUT_FIELD(dir, layout, object_size), | |
13120 | XATTR_LAYOUT_FIELD(dir, layout, pool), | |
13121 | XATTR_LAYOUT_FIELD(dir, layout, pool_namespace), | |
adb31ebb TL |
13122 | XATTR_NAME_CEPH(dir, entries, VXATTR_DIRSTAT), |
13123 | XATTR_NAME_CEPH(dir, files, VXATTR_DIRSTAT), | |
13124 | XATTR_NAME_CEPH(dir, subdirs, VXATTR_DIRSTAT), | |
13125 | XATTR_NAME_CEPH(dir, rentries, VXATTR_RSTAT), | |
13126 | XATTR_NAME_CEPH(dir, rfiles, VXATTR_RSTAT), | |
13127 | XATTR_NAME_CEPH(dir, rsubdirs, VXATTR_RSTAT), | |
f67539c2 | 13128 | XATTR_NAME_CEPH(dir, rsnaps, VXATTR_RSTAT), |
adb31ebb TL |
13129 | XATTR_NAME_CEPH(dir, rbytes, VXATTR_RSTAT), |
13130 | XATTR_NAME_CEPH(dir, rctime, VXATTR_RSTAT), | |
7c673cae FG |
13131 | { |
13132 | name: "ceph.quota", | |
13133 | getxattr_cb: &Client::_vxattrcb_quota, | |
13134 | readonly: false, | |
7c673cae | 13135 | exists_cb: &Client::_vxattrcb_quota_exists, |
28e407b8 | 13136 | flags: 0, |
7c673cae FG |
13137 | }, |
13138 | XATTR_QUOTA_FIELD(quota, max_bytes), | |
13139 | XATTR_QUOTA_FIELD(quota, max_files), | |
1d09f67e TL |
13140 | // FIXME |
13141 | // Delete the following dir pin field definitions for release "S" | |
11fdf7f2 TL |
13142 | { |
13143 | name: "ceph.dir.pin", | |
13144 | getxattr_cb: &Client::_vxattrcb_dir_pin, | |
13145 | readonly: false, | |
11fdf7f2 TL |
13146 | exists_cb: &Client::_vxattrcb_dir_pin_exists, |
13147 | flags: 0, | |
13148 | }, | |
81eedcae TL |
13149 | { |
13150 | name: "ceph.snap.btime", | |
13151 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
13152 | readonly: true, | |
81eedcae TL |
13153 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
13154 | flags: 0, | |
13155 | }, | |
f67539c2 TL |
13156 | { |
13157 | name: "ceph.mirror.info", | |
13158 | getxattr_cb: &Client::_vxattrcb_mirror_info, | |
13159 | readonly: false, | |
13160 | exists_cb: &Client::_vxattrcb_mirror_info_exists, | |
13161 | flags: 0, | |
13162 | }, | |
20effc67 TL |
13163 | { |
13164 | name: "ceph.caps", | |
13165 | getxattr_cb: &Client::_vxattrcb_caps, | |
13166 | readonly: true, | |
13167 | exists_cb: NULL, | |
13168 | flags: 0, | |
13169 | }, | |
7c673cae FG |
13170 | { name: "" } /* Required table terminator */ |
13171 | }; | |
13172 | ||
13173 | const Client::VXattr Client::_file_vxattrs[] = { | |
13174 | { | |
13175 | name: "ceph.file.layout", | |
13176 | getxattr_cb: &Client::_vxattrcb_layout, | |
13177 | readonly: false, | |
7c673cae | 13178 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 13179 | flags: 0, |
7c673cae FG |
13180 | }, |
13181 | XATTR_LAYOUT_FIELD(file, layout, stripe_unit), | |
13182 | XATTR_LAYOUT_FIELD(file, layout, stripe_count), | |
13183 | XATTR_LAYOUT_FIELD(file, layout, object_size), | |
13184 | XATTR_LAYOUT_FIELD(file, layout, pool), | |
13185 | XATTR_LAYOUT_FIELD(file, layout, pool_namespace), | |
81eedcae TL |
13186 | { |
13187 | name: "ceph.snap.btime", | |
13188 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
13189 | readonly: true, | |
81eedcae TL |
13190 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
13191 | flags: 0, | |
13192 | }, | |
20effc67 TL |
13193 | { |
13194 | name: "ceph.caps", | |
13195 | getxattr_cb: &Client::_vxattrcb_caps, | |
13196 | readonly: true, | |
13197 | exists_cb: NULL, | |
13198 | flags: 0, | |
13199 | }, | |
7c673cae FG |
13200 | { name: "" } /* Required table terminator */ |
13201 | }; | |
13202 | ||
adb31ebb TL |
13203 | const Client::VXattr Client::_common_vxattrs[] = { |
13204 | { | |
13205 | name: "ceph.cluster_fsid", | |
13206 | getxattr_cb: &Client::_vxattrcb_cluster_fsid, | |
13207 | readonly: true, | |
13208 | exists_cb: nullptr, | |
13209 | flags: 0, | |
13210 | }, | |
13211 | { | |
13212 | name: "ceph.client_id", | |
13213 | getxattr_cb: &Client::_vxattrcb_client_id, | |
13214 | readonly: true, | |
13215 | exists_cb: nullptr, | |
13216 | flags: 0, | |
13217 | }, | |
13218 | { name: "" } /* Required table terminator */ | |
13219 | }; | |
13220 | ||
7c673cae FG |
13221 | const Client::VXattr *Client::_get_vxattrs(Inode *in) |
13222 | { | |
13223 | if (in->is_dir()) | |
13224 | return _dir_vxattrs; | |
13225 | else if (in->is_file()) | |
13226 | return _file_vxattrs; | |
13227 | return NULL; | |
13228 | } | |
13229 | ||
13230 | const Client::VXattr *Client::_match_vxattr(Inode *in, const char *name) | |
13231 | { | |
13232 | if (strncmp(name, "ceph.", 5) == 0) { | |
13233 | const VXattr *vxattr = _get_vxattrs(in); | |
13234 | if (vxattr) { | |
13235 | while (!vxattr->name.empty()) { | |
13236 | if (vxattr->name == name) | |
13237 | return vxattr; | |
13238 | vxattr++; | |
13239 | } | |
13240 | } | |
adb31ebb TL |
13241 | |
13242 | // for common vxattrs | |
13243 | vxattr = _common_vxattrs; | |
13244 | while (!vxattr->name.empty()) { | |
13245 | if (vxattr->name == name) | |
13246 | return vxattr; | |
13247 | vxattr++; | |
13248 | } | |
7c673cae | 13249 | } |
adb31ebb | 13250 | |
7c673cae FG |
13251 | return NULL; |
13252 | } | |
13253 | ||
7c673cae FG |
13254 | int Client::ll_readlink(Inode *in, char *buf, size_t buflen, const UserPerm& perms) |
13255 | { | |
f67539c2 TL |
13256 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13257 | if (!mref_reader.is_state_satisfied()) | |
13258 | return -CEPHFS_ENOTCONN; | |
181888fb | 13259 | |
7c673cae FG |
13260 | vinodeno_t vino = _get_vino(in); |
13261 | ||
13262 | ldout(cct, 3) << "ll_readlink " << vino << dendl; | |
13263 | tout(cct) << "ll_readlink" << std::endl; | |
13264 | tout(cct) << vino.ino.val << std::endl; | |
13265 | ||
f67539c2 | 13266 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
13267 | for (auto dn : in->dentries) { |
13268 | touch_dn(dn); | |
7c673cae FG |
13269 | } |
13270 | ||
13271 | int r = _readlink(in, buf, buflen); // FIXME: no permission checking! | |
13272 | ldout(cct, 3) << "ll_readlink " << vino << " = " << r << dendl; | |
13273 | return r; | |
13274 | } | |
13275 | ||
13276 | int Client::_mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev, | |
13277 | const UserPerm& perms, InodeRef *inp) | |
13278 | { | |
1adf2230 | 13279 | ldout(cct, 8) << "_mknod(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
13280 | << mode << dec << ", " << rdev << ", uid " << perms.uid() |
13281 | << ", gid " << perms.gid() << ")" << dendl; | |
13282 | ||
13283 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13284 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13285 | |
13286 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13287 | return -CEPHFS_EROFS; |
7c673cae FG |
13288 | } |
13289 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13290 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13291 | } |
13292 | ||
13293 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_MKNOD); | |
13294 | ||
13295 | filepath path; | |
13296 | dir->make_nosnap_relative_path(path); | |
13297 | path.push_dentry(name); | |
13298 | req->set_filepath(path); | |
13299 | req->set_inode(dir); | |
13300 | req->head.args.mknod.rdev = rdev; | |
13301 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13302 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13303 | ||
13304 | bufferlist xattrs_bl; | |
13305 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
13306 | if (res < 0) | |
13307 | goto fail; | |
13308 | req->head.args.mknod.mode = mode; | |
13309 | if (xattrs_bl.length() > 0) | |
13310 | req->set_data(xattrs_bl); | |
13311 | ||
13312 | Dentry *de; | |
13313 | res = get_or_create(dir, name, &de); | |
13314 | if (res < 0) | |
13315 | goto fail; | |
13316 | req->set_dentry(de); | |
13317 | ||
13318 | res = make_request(req, perms, inp); | |
13319 | ||
13320 | trim_cache(); | |
13321 | ||
1adf2230 | 13322 | ldout(cct, 8) << "mknod(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13323 | return res; |
13324 | ||
13325 | fail: | |
13326 | put_request(req); | |
13327 | return res; | |
13328 | } | |
13329 | ||
13330 | int Client::ll_mknod(Inode *parent, const char *name, mode_t mode, | |
13331 | dev_t rdev, struct stat *attr, Inode **out, | |
13332 | const UserPerm& perms) | |
13333 | { | |
f67539c2 TL |
13334 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13335 | if (!mref_reader.is_state_satisfied()) | |
13336 | return -CEPHFS_ENOTCONN; | |
181888fb | 13337 | |
7c673cae FG |
13338 | vinodeno_t vparent = _get_vino(parent); |
13339 | ||
13340 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name << dendl; | |
13341 | tout(cct) << "ll_mknod" << std::endl; | |
13342 | tout(cct) << vparent.ino.val << std::endl; | |
13343 | tout(cct) << name << std::endl; | |
13344 | tout(cct) << mode << std::endl; | |
13345 | tout(cct) << rdev << std::endl; | |
13346 | ||
f67539c2 | 13347 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 13348 | if (!fuse_default_permissions) { |
7c673cae FG |
13349 | int r = may_create(parent, perms); |
13350 | if (r < 0) | |
13351 | return r; | |
13352 | } | |
13353 | ||
13354 | InodeRef in; | |
13355 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13356 | if (r == 0) { | |
13357 | fill_stat(in, attr); | |
13358 | _ll_get(in.get()); | |
13359 | } | |
13360 | tout(cct) << attr->st_ino << std::endl; | |
13361 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name | |
13362 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13363 | *out = in.get(); | |
13364 | return r; | |
13365 | } | |
13366 | ||
13367 | int Client::ll_mknodx(Inode *parent, const char *name, mode_t mode, | |
13368 | dev_t rdev, Inode **out, | |
13369 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13370 | const UserPerm& perms) | |
13371 | { | |
f67539c2 TL |
13372 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13373 | if (!mref_reader.is_state_satisfied()) | |
13374 | return -CEPHFS_ENOTCONN; | |
7c673cae | 13375 | |
f67539c2 | 13376 | unsigned caps = statx_to_mask(flags, want); |
181888fb | 13377 | |
7c673cae FG |
13378 | vinodeno_t vparent = _get_vino(parent); |
13379 | ||
13380 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name << dendl; | |
13381 | tout(cct) << "ll_mknodx" << std::endl; | |
13382 | tout(cct) << vparent.ino.val << std::endl; | |
13383 | tout(cct) << name << std::endl; | |
13384 | tout(cct) << mode << std::endl; | |
13385 | tout(cct) << rdev << std::endl; | |
13386 | ||
f67539c2 TL |
13387 | std::scoped_lock lock(client_lock); |
13388 | ||
11fdf7f2 | 13389 | if (!fuse_default_permissions) { |
7c673cae FG |
13390 | int r = may_create(parent, perms); |
13391 | if (r < 0) | |
13392 | return r; | |
13393 | } | |
13394 | ||
13395 | InodeRef in; | |
13396 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13397 | if (r == 0) { | |
13398 | fill_statx(in, caps, stx); | |
13399 | _ll_get(in.get()); | |
13400 | } | |
13401 | tout(cct) << stx->stx_ino << std::endl; | |
13402 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name | |
13403 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13404 | *out = in.get(); | |
13405 | return r; | |
13406 | } | |
13407 | ||
13408 | int Client::_create(Inode *dir, const char *name, int flags, mode_t mode, | |
13409 | InodeRef *inp, Fh **fhp, int stripe_unit, int stripe_count, | |
13410 | int object_size, const char *data_pool, bool *created, | |
f67539c2 | 13411 | const UserPerm& perms, std::string alternate_name) |
7c673cae | 13412 | { |
1adf2230 | 13413 | ldout(cct, 8) << "_create(" << dir->ino << " " << name << ", 0" << oct << |
7c673cae FG |
13414 | mode << dec << ")" << dendl; |
13415 | ||
13416 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13417 | return -CEPHFS_ENAMETOOLONG; |
7c673cae | 13418 | if (dir->snapid != CEPH_NOSNAP) { |
f67539c2 | 13419 | return -CEPHFS_EROFS; |
7c673cae FG |
13420 | } |
13421 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13422 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13423 | } |
13424 | ||
13425 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
13426 | int cflags = ceph_flags_sys2wire(flags); |
13427 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
13428 | cflags |= CEPH_O_LAZY; | |
13429 | ||
13430 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
13431 | |
13432 | int64_t pool_id = -1; | |
13433 | if (data_pool && *data_pool) { | |
13434 | pool_id = objecter->with_osdmap( | |
13435 | std::mem_fn(&OSDMap::lookup_pg_pool_name), data_pool); | |
13436 | if (pool_id < 0) | |
f67539c2 | 13437 | return -CEPHFS_EINVAL; |
7c673cae | 13438 | if (pool_id > 0xffffffffll) |
f67539c2 | 13439 | return -CEPHFS_ERANGE; // bummer! |
7c673cae FG |
13440 | } |
13441 | ||
13442 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_CREATE); | |
13443 | ||
13444 | filepath path; | |
13445 | dir->make_nosnap_relative_path(path); | |
13446 | path.push_dentry(name); | |
13447 | req->set_filepath(path); | |
f67539c2 | 13448 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae | 13449 | req->set_inode(dir); |
11fdf7f2 | 13450 | req->head.args.open.flags = cflags | CEPH_O_CREAT; |
7c673cae FG |
13451 | |
13452 | req->head.args.open.stripe_unit = stripe_unit; | |
13453 | req->head.args.open.stripe_count = stripe_count; | |
13454 | req->head.args.open.object_size = object_size; | |
13455 | if (cct->_conf->client_debug_getattr_caps) | |
13456 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
13457 | else | |
13458 | req->head.args.open.mask = 0; | |
13459 | req->head.args.open.pool = pool_id; | |
13460 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13461 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13462 | ||
13463 | mode |= S_IFREG; | |
13464 | bufferlist xattrs_bl; | |
13465 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
13466 | if (res < 0) | |
13467 | goto fail; | |
13468 | req->head.args.open.mode = mode; | |
13469 | if (xattrs_bl.length() > 0) | |
13470 | req->set_data(xattrs_bl); | |
13471 | ||
13472 | Dentry *de; | |
13473 | res = get_or_create(dir, name, &de); | |
13474 | if (res < 0) | |
13475 | goto fail; | |
13476 | req->set_dentry(de); | |
13477 | ||
13478 | res = make_request(req, perms, inp, created); | |
13479 | if (res < 0) { | |
13480 | goto reply_error; | |
13481 | } | |
13482 | ||
13483 | /* If the caller passed a value in fhp, do the open */ | |
13484 | if(fhp) { | |
13485 | (*inp)->get_open_ref(cmode); | |
13486 | *fhp = _create_fh(inp->get(), flags, cmode, perms); | |
13487 | } | |
13488 | ||
13489 | reply_error: | |
13490 | trim_cache(); | |
13491 | ||
1adf2230 | 13492 | ldout(cct, 8) << "create(" << path << ", 0" << oct << mode << dec |
7c673cae FG |
13493 | << " layout " << stripe_unit |
13494 | << ' ' << stripe_count | |
13495 | << ' ' << object_size | |
13496 | <<") = " << res << dendl; | |
13497 | return res; | |
13498 | ||
13499 | fail: | |
13500 | put_request(req); | |
13501 | return res; | |
13502 | } | |
13503 | ||
7c673cae | 13504 | int Client::_mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm, |
f67539c2 TL |
13505 | InodeRef *inp, const std::map<std::string, std::string> &metadata, |
13506 | std::string alternate_name) | |
7c673cae | 13507 | { |
1adf2230 | 13508 | ldout(cct, 8) << "_mkdir(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
13509 | << mode << dec << ", uid " << perm.uid() |
13510 | << ", gid " << perm.gid() << ")" << dendl; | |
13511 | ||
13512 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13513 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13514 | |
13515 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13516 | return -CEPHFS_EROFS; |
7c673cae FG |
13517 | } |
13518 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 13519 | return -CEPHFS_EDQUOT; |
7c673cae | 13520 | } |
f67539c2 TL |
13521 | |
13522 | bool is_snap_op = dir->snapid == CEPH_SNAPDIR; | |
13523 | MetaRequest *req = new MetaRequest(is_snap_op ? | |
7c673cae FG |
13524 | CEPH_MDS_OP_MKSNAP : CEPH_MDS_OP_MKDIR); |
13525 | ||
13526 | filepath path; | |
13527 | dir->make_nosnap_relative_path(path); | |
13528 | path.push_dentry(name); | |
13529 | req->set_filepath(path); | |
13530 | req->set_inode(dir); | |
13531 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13532 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
f67539c2 | 13533 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13534 | |
13535 | mode |= S_IFDIR; | |
f67539c2 TL |
13536 | bufferlist bl; |
13537 | int res = _posix_acl_create(dir, &mode, bl, perm); | |
7c673cae FG |
13538 | if (res < 0) |
13539 | goto fail; | |
13540 | req->head.args.mkdir.mode = mode; | |
f67539c2 TL |
13541 | if (is_snap_op) { |
13542 | SnapPayload payload; | |
13543 | // clear the bufferlist that may have been populated by the call | |
13544 | // to _posix_acl_create(). MDS mksnap does not make use of it. | |
13545 | // So, reuse it to pass metadata payload. | |
13546 | bl.clear(); | |
13547 | payload.metadata = metadata; | |
13548 | encode(payload, bl); | |
13549 | } | |
13550 | if (bl.length() > 0) { | |
13551 | req->set_data(bl); | |
13552 | } | |
7c673cae FG |
13553 | |
13554 | Dentry *de; | |
13555 | res = get_or_create(dir, name, &de); | |
13556 | if (res < 0) | |
13557 | goto fail; | |
13558 | req->set_dentry(de); | |
13559 | ||
13560 | ldout(cct, 10) << "_mkdir: making request" << dendl; | |
13561 | res = make_request(req, perm, inp); | |
13562 | ldout(cct, 10) << "_mkdir result is " << res << dendl; | |
13563 | ||
13564 | trim_cache(); | |
13565 | ||
1adf2230 | 13566 | ldout(cct, 8) << "_mkdir(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13567 | return res; |
13568 | ||
13569 | fail: | |
13570 | put_request(req); | |
13571 | return res; | |
13572 | } | |
13573 | ||
13574 | int Client::ll_mkdir(Inode *parent, const char *name, mode_t mode, | |
13575 | struct stat *attr, Inode **out, const UserPerm& perm) | |
13576 | { | |
f67539c2 TL |
13577 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13578 | if (!mref_reader.is_state_satisfied()) | |
13579 | return -CEPHFS_ENOTCONN; | |
181888fb | 13580 | |
7c673cae FG |
13581 | vinodeno_t vparent = _get_vino(parent); |
13582 | ||
13583 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name << dendl; | |
13584 | tout(cct) << "ll_mkdir" << std::endl; | |
13585 | tout(cct) << vparent.ino.val << std::endl; | |
13586 | tout(cct) << name << std::endl; | |
13587 | tout(cct) << mode << std::endl; | |
13588 | ||
f67539c2 TL |
13589 | std::scoped_lock lock(client_lock); |
13590 | ||
11fdf7f2 | 13591 | if (!fuse_default_permissions) { |
7c673cae FG |
13592 | int r = may_create(parent, perm); |
13593 | if (r < 0) | |
13594 | return r; | |
13595 | } | |
13596 | ||
13597 | InodeRef in; | |
13598 | int r = _mkdir(parent, name, mode, perm, &in); | |
13599 | if (r == 0) { | |
13600 | fill_stat(in, attr); | |
13601 | _ll_get(in.get()); | |
13602 | } | |
13603 | tout(cct) << attr->st_ino << std::endl; | |
13604 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name | |
13605 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13606 | *out = in.get(); | |
13607 | return r; | |
13608 | } | |
13609 | ||
13610 | int Client::ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out, | |
13611 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13612 | const UserPerm& perms) | |
13613 | { | |
f67539c2 TL |
13614 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13615 | if (!mref_reader.is_state_satisfied()) | |
13616 | return -CEPHFS_ENOTCONN; | |
181888fb | 13617 | |
7c673cae FG |
13618 | vinodeno_t vparent = _get_vino(parent); |
13619 | ||
13620 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name << dendl; | |
13621 | tout(cct) << "ll_mkdirx" << std::endl; | |
13622 | tout(cct) << vparent.ino.val << std::endl; | |
13623 | tout(cct) << name << std::endl; | |
13624 | tout(cct) << mode << std::endl; | |
13625 | ||
f67539c2 TL |
13626 | std::scoped_lock lock(client_lock); |
13627 | ||
11fdf7f2 | 13628 | if (!fuse_default_permissions) { |
7c673cae FG |
13629 | int r = may_create(parent, perms); |
13630 | if (r < 0) | |
13631 | return r; | |
13632 | } | |
13633 | ||
13634 | InodeRef in; | |
13635 | int r = _mkdir(parent, name, mode, perms, &in); | |
13636 | if (r == 0) { | |
13637 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13638 | _ll_get(in.get()); | |
13639 | } else { | |
13640 | stx->stx_ino = 0; | |
13641 | stx->stx_mask = 0; | |
13642 | } | |
13643 | tout(cct) << stx->stx_ino << std::endl; | |
13644 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name | |
13645 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13646 | *out = in.get(); | |
13647 | return r; | |
13648 | } | |
13649 | ||
13650 | int Client::_symlink(Inode *dir, const char *name, const char *target, | |
f67539c2 | 13651 | const UserPerm& perms, std::string alternate_name, InodeRef *inp) |
7c673cae | 13652 | { |
1adf2230 | 13653 | ldout(cct, 8) << "_symlink(" << dir->ino << " " << name << ", " << target |
7c673cae FG |
13654 | << ", uid " << perms.uid() << ", gid " << perms.gid() << ")" |
13655 | << dendl; | |
13656 | ||
13657 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13658 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13659 | |
13660 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13661 | return -CEPHFS_EROFS; |
7c673cae FG |
13662 | } |
13663 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13664 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13665 | } |
13666 | ||
13667 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SYMLINK); | |
13668 | ||
13669 | filepath path; | |
13670 | dir->make_nosnap_relative_path(path); | |
13671 | path.push_dentry(name); | |
13672 | req->set_filepath(path); | |
f67539c2 | 13673 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13674 | req->set_inode(dir); |
13675 | req->set_string2(target); | |
13676 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13677 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13678 | ||
13679 | Dentry *de; | |
13680 | int res = get_or_create(dir, name, &de); | |
13681 | if (res < 0) | |
13682 | goto fail; | |
13683 | req->set_dentry(de); | |
13684 | ||
13685 | res = make_request(req, perms, inp); | |
13686 | ||
13687 | trim_cache(); | |
1adf2230 | 13688 | ldout(cct, 8) << "_symlink(\"" << path << "\", \"" << target << "\") = " << |
7c673cae FG |
13689 | res << dendl; |
13690 | return res; | |
13691 | ||
13692 | fail: | |
13693 | put_request(req); | |
13694 | return res; | |
13695 | } | |
13696 | ||
13697 | int Client::ll_symlink(Inode *parent, const char *name, const char *value, | |
13698 | struct stat *attr, Inode **out, const UserPerm& perms) | |
13699 | { | |
f67539c2 TL |
13700 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13701 | if (!mref_reader.is_state_satisfied()) | |
13702 | return -CEPHFS_ENOTCONN; | |
181888fb | 13703 | |
7c673cae FG |
13704 | vinodeno_t vparent = _get_vino(parent); |
13705 | ||
13706 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name << " -> " << value | |
13707 | << dendl; | |
13708 | tout(cct) << "ll_symlink" << std::endl; | |
13709 | tout(cct) << vparent.ino.val << std::endl; | |
13710 | tout(cct) << name << std::endl; | |
13711 | tout(cct) << value << std::endl; | |
13712 | ||
f67539c2 TL |
13713 | std::scoped_lock lock(client_lock); |
13714 | ||
11fdf7f2 | 13715 | if (!fuse_default_permissions) { |
7c673cae FG |
13716 | int r = may_create(parent, perms); |
13717 | if (r < 0) | |
13718 | return r; | |
13719 | } | |
13720 | ||
13721 | InodeRef in; | |
f67539c2 | 13722 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13723 | if (r == 0) { |
13724 | fill_stat(in, attr); | |
13725 | _ll_get(in.get()); | |
13726 | } | |
13727 | tout(cct) << attr->st_ino << std::endl; | |
13728 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name | |
13729 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13730 | *out = in.get(); | |
13731 | return r; | |
13732 | } | |
13733 | ||
13734 | int Client::ll_symlinkx(Inode *parent, const char *name, const char *value, | |
13735 | Inode **out, struct ceph_statx *stx, unsigned want, | |
13736 | unsigned flags, const UserPerm& perms) | |
13737 | { | |
f67539c2 TL |
13738 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13739 | if (!mref_reader.is_state_satisfied()) | |
13740 | return -CEPHFS_ENOTCONN; | |
181888fb | 13741 | |
7c673cae FG |
13742 | vinodeno_t vparent = _get_vino(parent); |
13743 | ||
13744 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name << " -> " << value | |
13745 | << dendl; | |
13746 | tout(cct) << "ll_symlinkx" << std::endl; | |
13747 | tout(cct) << vparent.ino.val << std::endl; | |
13748 | tout(cct) << name << std::endl; | |
13749 | tout(cct) << value << std::endl; | |
13750 | ||
f67539c2 TL |
13751 | std::scoped_lock lock(client_lock); |
13752 | ||
11fdf7f2 | 13753 | if (!fuse_default_permissions) { |
7c673cae FG |
13754 | int r = may_create(parent, perms); |
13755 | if (r < 0) | |
13756 | return r; | |
13757 | } | |
13758 | ||
13759 | InodeRef in; | |
f67539c2 | 13760 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13761 | if (r == 0) { |
13762 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13763 | _ll_get(in.get()); | |
13764 | } | |
13765 | tout(cct) << stx->stx_ino << std::endl; | |
13766 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name | |
13767 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13768 | *out = in.get(); | |
13769 | return r; | |
13770 | } | |
13771 | ||
13772 | int Client::_unlink(Inode *dir, const char *name, const UserPerm& perm) | |
13773 | { | |
1adf2230 | 13774 | ldout(cct, 8) << "_unlink(" << dir->ino << " " << name |
7c673cae FG |
13775 | << " uid " << perm.uid() << " gid " << perm.gid() |
13776 | << ")" << dendl; | |
13777 | ||
13778 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13779 | return -CEPHFS_EROFS; |
7c673cae FG |
13780 | } |
13781 | ||
13782 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_UNLINK); | |
13783 | ||
13784 | filepath path; | |
13785 | dir->make_nosnap_relative_path(path); | |
13786 | path.push_dentry(name); | |
13787 | req->set_filepath(path); | |
13788 | ||
13789 | InodeRef otherin; | |
b32b8144 | 13790 | Inode *in; |
7c673cae | 13791 | Dentry *de; |
b32b8144 | 13792 | |
7c673cae FG |
13793 | int res = get_or_create(dir, name, &de); |
13794 | if (res < 0) | |
13795 | goto fail; | |
13796 | req->set_dentry(de); | |
13797 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13798 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13799 | ||
13800 | res = _lookup(dir, name, 0, &otherin, perm); | |
13801 | if (res < 0) | |
13802 | goto fail; | |
b32b8144 FG |
13803 | |
13804 | in = otherin.get(); | |
13805 | req->set_other_inode(in); | |
13806 | in->break_all_delegs(); | |
7c673cae FG |
13807 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
13808 | ||
13809 | req->set_inode(dir); | |
13810 | ||
13811 | res = make_request(req, perm); | |
13812 | ||
13813 | trim_cache(); | |
1adf2230 | 13814 | ldout(cct, 8) << "unlink(" << path << ") = " << res << dendl; |
7c673cae FG |
13815 | return res; |
13816 | ||
13817 | fail: | |
13818 | put_request(req); | |
13819 | return res; | |
13820 | } | |
13821 | ||
13822 | int Client::ll_unlink(Inode *in, const char *name, const UserPerm& perm) | |
13823 | { | |
f67539c2 TL |
13824 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13825 | if (!mref_reader.is_state_satisfied()) | |
13826 | return -CEPHFS_ENOTCONN; | |
181888fb | 13827 | |
7c673cae FG |
13828 | vinodeno_t vino = _get_vino(in); |
13829 | ||
13830 | ldout(cct, 3) << "ll_unlink " << vino << " " << name << dendl; | |
13831 | tout(cct) << "ll_unlink" << std::endl; | |
13832 | tout(cct) << vino.ino.val << std::endl; | |
13833 | tout(cct) << name << std::endl; | |
13834 | ||
f67539c2 TL |
13835 | std::scoped_lock lock(client_lock); |
13836 | ||
11fdf7f2 | 13837 | if (!fuse_default_permissions) { |
7c673cae FG |
13838 | int r = may_delete(in, name, perm); |
13839 | if (r < 0) | |
13840 | return r; | |
13841 | } | |
13842 | return _unlink(in, name, perm); | |
13843 | } | |
13844 | ||
13845 | int Client::_rmdir(Inode *dir, const char *name, const UserPerm& perms) | |
13846 | { | |
1adf2230 | 13847 | ldout(cct, 8) << "_rmdir(" << dir->ino << " " << name << " uid " |
7c673cae FG |
13848 | << perms.uid() << " gid " << perms.gid() << ")" << dendl; |
13849 | ||
13850 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13851 | return -CEPHFS_EROFS; |
7c673cae | 13852 | } |
b32b8144 FG |
13853 | |
13854 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_RMSNAP : CEPH_MDS_OP_RMDIR; | |
13855 | MetaRequest *req = new MetaRequest(op); | |
7c673cae FG |
13856 | filepath path; |
13857 | dir->make_nosnap_relative_path(path); | |
13858 | path.push_dentry(name); | |
13859 | req->set_filepath(path); | |
11fdf7f2 | 13860 | req->set_inode(dir); |
7c673cae FG |
13861 | |
13862 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13863 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13864 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
13865 | ||
13866 | InodeRef in; | |
13867 | ||
13868 | Dentry *de; | |
13869 | int res = get_or_create(dir, name, &de); | |
13870 | if (res < 0) | |
13871 | goto fail; | |
b32b8144 FG |
13872 | if (op == CEPH_MDS_OP_RMDIR) |
13873 | req->set_dentry(de); | |
13874 | else | |
13875 | de->get(); | |
13876 | ||
7c673cae FG |
13877 | res = _lookup(dir, name, 0, &in, perms); |
13878 | if (res < 0) | |
13879 | goto fail; | |
11fdf7f2 TL |
13880 | |
13881 | if (op == CEPH_MDS_OP_RMSNAP) { | |
7c673cae | 13882 | unlink(de, true, true); |
b32b8144 | 13883 | de->put(); |
7c673cae | 13884 | } |
11fdf7f2 | 13885 | req->set_other_inode(in.get()); |
7c673cae FG |
13886 | |
13887 | res = make_request(req, perms); | |
13888 | ||
13889 | trim_cache(); | |
1adf2230 | 13890 | ldout(cct, 8) << "rmdir(" << path << ") = " << res << dendl; |
7c673cae FG |
13891 | return res; |
13892 | ||
13893 | fail: | |
13894 | put_request(req); | |
13895 | return res; | |
13896 | } | |
13897 | ||
13898 | int Client::ll_rmdir(Inode *in, const char *name, const UserPerm& perms) | |
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 | ||
13906 | ldout(cct, 3) << "ll_rmdir " << vino << " " << name << dendl; | |
13907 | tout(cct) << "ll_rmdir" << std::endl; | |
13908 | tout(cct) << vino.ino.val << std::endl; | |
13909 | tout(cct) << name << std::endl; | |
13910 | ||
f67539c2 TL |
13911 | std::scoped_lock lock(client_lock); |
13912 | ||
11fdf7f2 | 13913 | if (!fuse_default_permissions) { |
7c673cae FG |
13914 | int r = may_delete(in, name, perms); |
13915 | if (r < 0) | |
13916 | return r; | |
13917 | } | |
13918 | ||
13919 | return _rmdir(in, name, perms); | |
13920 | } | |
13921 | ||
f67539c2 | 13922 | int Client::_rename(Inode *fromdir, const char *fromname, Inode *todir, const char *toname, const UserPerm& perm, std::string alternate_name) |
7c673cae | 13923 | { |
1adf2230 | 13924 | ldout(cct, 8) << "_rename(" << fromdir->ino << " " << fromname << " to " |
7c673cae FG |
13925 | << todir->ino << " " << toname |
13926 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" | |
13927 | << dendl; | |
13928 | ||
13929 | if (fromdir->snapid != todir->snapid) | |
f67539c2 | 13930 | return -CEPHFS_EXDEV; |
7c673cae FG |
13931 | |
13932 | int op = CEPH_MDS_OP_RENAME; | |
13933 | if (fromdir->snapid != CEPH_NOSNAP) { | |
13934 | if (fromdir == todir && fromdir->snapid == CEPH_SNAPDIR) | |
13935 | op = CEPH_MDS_OP_RENAMESNAP; | |
13936 | else | |
f67539c2 TL |
13937 | return -CEPHFS_EROFS; |
13938 | } | |
2a845540 | 13939 | if (cct->_conf.get_val<bool>("client_quota") && fromdir != todir) { |
f67539c2 TL |
13940 | Inode *fromdir_root = |
13941 | fromdir->quota.is_enable() ? fromdir : get_quota_root(fromdir, perm); | |
13942 | Inode *todir_root = | |
13943 | todir->quota.is_enable() ? todir : get_quota_root(todir, perm); | |
13944 | if (fromdir_root != todir_root) { | |
13945 | return -CEPHFS_EXDEV; | |
13946 | } | |
7c673cae | 13947 | } |
7c673cae FG |
13948 | |
13949 | InodeRef target; | |
13950 | MetaRequest *req = new MetaRequest(op); | |
13951 | ||
13952 | filepath from; | |
13953 | fromdir->make_nosnap_relative_path(from); | |
13954 | from.push_dentry(fromname); | |
13955 | filepath to; | |
13956 | todir->make_nosnap_relative_path(to); | |
13957 | to.push_dentry(toname); | |
13958 | req->set_filepath(to); | |
13959 | req->set_filepath2(from); | |
f67539c2 | 13960 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13961 | |
13962 | Dentry *oldde; | |
13963 | int res = get_or_create(fromdir, fromname, &oldde); | |
13964 | if (res < 0) | |
13965 | goto fail; | |
13966 | Dentry *de; | |
13967 | res = get_or_create(todir, toname, &de); | |
13968 | if (res < 0) | |
13969 | goto fail; | |
13970 | ||
13971 | if (op == CEPH_MDS_OP_RENAME) { | |
13972 | req->set_old_dentry(oldde); | |
13973 | req->old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
13974 | req->old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
13975 | ||
13976 | req->set_dentry(de); | |
13977 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13978 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13979 | ||
13980 | InodeRef oldin, otherin; | |
f67539c2 | 13981 | res = _lookup(fromdir, fromname, 0, &oldin, perm); |
7c673cae FG |
13982 | if (res < 0) |
13983 | goto fail; | |
b32b8144 FG |
13984 | |
13985 | Inode *oldinode = oldin.get(); | |
13986 | oldinode->break_all_delegs(); | |
13987 | req->set_old_inode(oldinode); | |
7c673cae FG |
13988 | req->old_inode_drop = CEPH_CAP_LINK_SHARED; |
13989 | ||
13990 | res = _lookup(todir, toname, 0, &otherin, perm); | |
b32b8144 FG |
13991 | switch (res) { |
13992 | case 0: | |
13993 | { | |
13994 | Inode *in = otherin.get(); | |
13995 | req->set_other_inode(in); | |
13996 | in->break_all_delegs(); | |
13997 | } | |
7c673cae | 13998 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
b32b8144 | 13999 | break; |
f67539c2 | 14000 | case -CEPHFS_ENOENT: |
b32b8144 FG |
14001 | break; |
14002 | default: | |
14003 | goto fail; | |
7c673cae FG |
14004 | } |
14005 | ||
14006 | req->set_inode(todir); | |
14007 | } else { | |
14008 | // renamesnap reply contains no tracedn, so we need to invalidate | |
14009 | // dentry manually | |
14010 | unlink(oldde, true, true); | |
14011 | unlink(de, true, true); | |
11fdf7f2 TL |
14012 | |
14013 | req->set_inode(todir); | |
7c673cae FG |
14014 | } |
14015 | ||
14016 | res = make_request(req, perm, &target); | |
14017 | ldout(cct, 10) << "rename result is " << res << dendl; | |
14018 | ||
14019 | // renamed item from our cache | |
14020 | ||
14021 | trim_cache(); | |
1adf2230 | 14022 | ldout(cct, 8) << "_rename(" << from << ", " << to << ") = " << res << dendl; |
7c673cae FG |
14023 | return res; |
14024 | ||
14025 | fail: | |
14026 | put_request(req); | |
14027 | return res; | |
14028 | } | |
14029 | ||
14030 | int Client::ll_rename(Inode *parent, const char *name, Inode *newparent, | |
14031 | const char *newname, const UserPerm& perm) | |
14032 | { | |
f67539c2 TL |
14033 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14034 | if (!mref_reader.is_state_satisfied()) | |
14035 | return -CEPHFS_ENOTCONN; | |
181888fb | 14036 | |
7c673cae FG |
14037 | vinodeno_t vparent = _get_vino(parent); |
14038 | vinodeno_t vnewparent = _get_vino(newparent); | |
14039 | ||
14040 | ldout(cct, 3) << "ll_rename " << vparent << " " << name << " to " | |
14041 | << vnewparent << " " << newname << dendl; | |
14042 | tout(cct) << "ll_rename" << std::endl; | |
14043 | tout(cct) << vparent.ino.val << std::endl; | |
14044 | tout(cct) << name << std::endl; | |
14045 | tout(cct) << vnewparent.ino.val << std::endl; | |
14046 | tout(cct) << newname << std::endl; | |
14047 | ||
f67539c2 TL |
14048 | std::scoped_lock lock(client_lock); |
14049 | ||
11fdf7f2 | 14050 | if (!fuse_default_permissions) { |
7c673cae FG |
14051 | int r = may_delete(parent, name, perm); |
14052 | if (r < 0) | |
14053 | return r; | |
14054 | r = may_delete(newparent, newname, perm); | |
f67539c2 | 14055 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
14056 | return r; |
14057 | } | |
14058 | ||
f67539c2 | 14059 | return _rename(parent, name, newparent, newname, perm, ""); |
7c673cae FG |
14060 | } |
14061 | ||
f67539c2 | 14062 | int Client::_link(Inode *in, Inode *dir, const char *newname, const UserPerm& perm, std::string alternate_name, InodeRef *inp) |
7c673cae | 14063 | { |
1adf2230 | 14064 | ldout(cct, 8) << "_link(" << in->ino << " to " << dir->ino << " " << newname |
7c673cae FG |
14065 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" << dendl; |
14066 | ||
14067 | if (strlen(newname) > NAME_MAX) | |
f67539c2 | 14068 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
14069 | |
14070 | if (in->snapid != CEPH_NOSNAP || dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 14071 | return -CEPHFS_EROFS; |
7c673cae FG |
14072 | } |
14073 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 14074 | return -CEPHFS_EDQUOT; |
7c673cae FG |
14075 | } |
14076 | ||
b32b8144 | 14077 | in->break_all_delegs(); |
7c673cae FG |
14078 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LINK); |
14079 | ||
14080 | filepath path(newname, dir->ino); | |
14081 | req->set_filepath(path); | |
f67539c2 | 14082 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
14083 | filepath existing(in->ino); |
14084 | req->set_filepath2(existing); | |
14085 | ||
14086 | req->set_inode(dir); | |
14087 | req->inode_drop = CEPH_CAP_FILE_SHARED; | |
14088 | req->inode_unless = CEPH_CAP_FILE_EXCL; | |
14089 | ||
14090 | Dentry *de; | |
14091 | int res = get_or_create(dir, newname, &de); | |
14092 | if (res < 0) | |
14093 | goto fail; | |
14094 | req->set_dentry(de); | |
14095 | ||
14096 | res = make_request(req, perm, inp); | |
14097 | ldout(cct, 10) << "link result is " << res << dendl; | |
14098 | ||
14099 | trim_cache(); | |
1adf2230 | 14100 | ldout(cct, 8) << "link(" << existing << ", " << path << ") = " << res << dendl; |
7c673cae FG |
14101 | return res; |
14102 | ||
14103 | fail: | |
14104 | put_request(req); | |
14105 | return res; | |
14106 | } | |
14107 | ||
14108 | int Client::ll_link(Inode *in, Inode *newparent, const char *newname, | |
14109 | const UserPerm& perm) | |
14110 | { | |
f67539c2 TL |
14111 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14112 | if (!mref_reader.is_state_satisfied()) | |
14113 | return -CEPHFS_ENOTCONN; | |
181888fb | 14114 | |
7c673cae FG |
14115 | vinodeno_t vino = _get_vino(in); |
14116 | vinodeno_t vnewparent = _get_vino(newparent); | |
14117 | ||
31f18b77 | 14118 | ldout(cct, 3) << "ll_link " << vino << " to " << vnewparent << " " << |
7c673cae FG |
14119 | newname << dendl; |
14120 | tout(cct) << "ll_link" << std::endl; | |
14121 | tout(cct) << vino.ino.val << std::endl; | |
14122 | tout(cct) << vnewparent << std::endl; | |
14123 | tout(cct) << newname << std::endl; | |
14124 | ||
7c673cae FG |
14125 | InodeRef target; |
14126 | ||
f67539c2 TL |
14127 | std::scoped_lock lock(client_lock); |
14128 | ||
11fdf7f2 | 14129 | if (!fuse_default_permissions) { |
7c673cae | 14130 | if (S_ISDIR(in->mode)) |
f67539c2 | 14131 | return -CEPHFS_EPERM; |
7c673cae | 14132 | |
11fdf7f2 | 14133 | int r = may_hardlink(in, perm); |
7c673cae FG |
14134 | if (r < 0) |
14135 | return r; | |
14136 | ||
14137 | r = may_create(newparent, perm); | |
14138 | if (r < 0) | |
14139 | return r; | |
14140 | } | |
14141 | ||
f67539c2 | 14142 | return _link(in, newparent, newname, perm, "", &target); |
7c673cae FG |
14143 | } |
14144 | ||
14145 | int Client::ll_num_osds(void) | |
14146 | { | |
f67539c2 | 14147 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14148 | return objecter->with_osdmap(std::mem_fn(&OSDMap::get_num_osds)); |
14149 | } | |
14150 | ||
14151 | int Client::ll_osdaddr(int osd, uint32_t *addr) | |
14152 | { | |
f67539c2 | 14153 | std::scoped_lock lock(client_lock); |
181888fb | 14154 | |
7c673cae FG |
14155 | entity_addr_t g; |
14156 | bool exists = objecter->with_osdmap([&](const OSDMap& o) { | |
14157 | if (!o.exists(osd)) | |
14158 | return false; | |
11fdf7f2 | 14159 | g = o.get_addrs(osd).front(); |
7c673cae FG |
14160 | return true; |
14161 | }); | |
14162 | if (!exists) | |
14163 | return -1; | |
14164 | uint32_t nb_addr = (g.in4_addr()).sin_addr.s_addr; | |
14165 | *addr = ntohl(nb_addr); | |
14166 | return 0; | |
14167 | } | |
181888fb | 14168 | |
7c673cae FG |
14169 | uint32_t Client::ll_stripe_unit(Inode *in) |
14170 | { | |
f67539c2 | 14171 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14172 | return in->layout.stripe_unit; |
14173 | } | |
14174 | ||
14175 | uint64_t Client::ll_snap_seq(Inode *in) | |
14176 | { | |
f67539c2 | 14177 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14178 | return in->snaprealm->seq; |
14179 | } | |
14180 | ||
14181 | int Client::ll_file_layout(Inode *in, file_layout_t *layout) | |
14182 | { | |
f67539c2 | 14183 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14184 | *layout = in->layout; |
14185 | return 0; | |
14186 | } | |
14187 | ||
14188 | int Client::ll_file_layout(Fh *fh, file_layout_t *layout) | |
14189 | { | |
14190 | return ll_file_layout(fh->inode.get(), layout); | |
14191 | } | |
14192 | ||
14193 | /* Currently we cannot take advantage of redundancy in reads, since we | |
14194 | would have to go through all possible placement groups (a | |
14195 | potentially quite large number determined by a hash), and use CRUSH | |
14196 | to calculate the appropriate set of OSDs for each placement group, | |
14197 | then index into that. An array with one entry per OSD is much more | |
14198 | tractable and works for demonstration purposes. */ | |
14199 | ||
14200 | int Client::ll_get_stripe_osd(Inode *in, uint64_t blockno, | |
14201 | file_layout_t* layout) | |
14202 | { | |
f67539c2 | 14203 | std::scoped_lock lock(client_lock); |
181888fb | 14204 | |
28e407b8 | 14205 | inodeno_t ino = in->ino; |
7c673cae FG |
14206 | uint32_t object_size = layout->object_size; |
14207 | uint32_t su = layout->stripe_unit; | |
14208 | uint32_t stripe_count = layout->stripe_count; | |
14209 | uint64_t stripes_per_object = object_size / su; | |
11fdf7f2 | 14210 | uint64_t stripeno = 0, stripepos = 0; |
7c673cae | 14211 | |
11fdf7f2 TL |
14212 | if(stripe_count) { |
14213 | stripeno = blockno / stripe_count; // which horizontal stripe (Y) | |
14214 | stripepos = blockno % stripe_count; // which object in the object set (X) | |
14215 | } | |
7c673cae FG |
14216 | uint64_t objectsetno = stripeno / stripes_per_object; // which object set |
14217 | uint64_t objectno = objectsetno * stripe_count + stripepos; // object id | |
14218 | ||
14219 | object_t oid = file_object_t(ino, objectno); | |
14220 | return objecter->with_osdmap([&](const OSDMap& o) { | |
14221 | ceph_object_layout olayout = | |
14222 | o.file_to_object_layout(oid, *layout); | |
14223 | pg_t pg = (pg_t)olayout.ol_pgid; | |
14224 | vector<int> osds; | |
14225 | int primary; | |
14226 | o.pg_to_acting_osds(pg, &osds, &primary); | |
14227 | return primary; | |
14228 | }); | |
14229 | } | |
14230 | ||
14231 | /* Return the offset of the block, internal to the object */ | |
14232 | ||
14233 | uint64_t Client::ll_get_internal_offset(Inode *in, uint64_t blockno) | |
14234 | { | |
f67539c2 | 14235 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14236 | file_layout_t *layout=&(in->layout); |
14237 | uint32_t object_size = layout->object_size; | |
14238 | uint32_t su = layout->stripe_unit; | |
14239 | uint64_t stripes_per_object = object_size / su; | |
14240 | ||
14241 | return (blockno % stripes_per_object) * su; | |
14242 | } | |
14243 | ||
14244 | int Client::ll_opendir(Inode *in, int flags, dir_result_t** dirpp, | |
14245 | const UserPerm& perms) | |
14246 | { | |
f67539c2 TL |
14247 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14248 | if (!mref_reader.is_state_satisfied()) | |
14249 | return -CEPHFS_ENOTCONN; | |
181888fb | 14250 | |
7c673cae FG |
14251 | vinodeno_t vino = _get_vino(in); |
14252 | ||
14253 | ldout(cct, 3) << "ll_opendir " << vino << dendl; | |
14254 | tout(cct) << "ll_opendir" << std::endl; | |
14255 | tout(cct) << vino.ino.val << std::endl; | |
14256 | ||
f67539c2 TL |
14257 | std::scoped_lock lock(client_lock); |
14258 | ||
11fdf7f2 | 14259 | if (!fuse_default_permissions) { |
7c673cae FG |
14260 | int r = may_open(in, flags, perms); |
14261 | if (r < 0) | |
14262 | return r; | |
14263 | } | |
14264 | ||
14265 | int r = _opendir(in, dirpp, perms); | |
f67539c2 | 14266 | tout(cct) << (uintptr_t)*dirpp << std::endl; |
7c673cae FG |
14267 | |
14268 | ldout(cct, 3) << "ll_opendir " << vino << " = " << r << " (" << *dirpp << ")" | |
14269 | << dendl; | |
14270 | return r; | |
14271 | } | |
14272 | ||
14273 | int Client::ll_releasedir(dir_result_t *dirp) | |
14274 | { | |
f67539c2 TL |
14275 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14276 | if (!mref_reader.is_state_satisfied()) | |
14277 | return -CEPHFS_ENOTCONN; | |
14278 | ||
7c673cae FG |
14279 | ldout(cct, 3) << "ll_releasedir " << dirp << dendl; |
14280 | tout(cct) << "ll_releasedir" << std::endl; | |
f67539c2 | 14281 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 14282 | |
f67539c2 | 14283 | std::scoped_lock lock(client_lock); |
181888fb | 14284 | |
7c673cae FG |
14285 | _closedir(dirp); |
14286 | return 0; | |
14287 | } | |
14288 | ||
14289 | int Client::ll_fsyncdir(dir_result_t *dirp) | |
14290 | { | |
f67539c2 TL |
14291 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14292 | if (!mref_reader.is_state_satisfied()) | |
14293 | return -CEPHFS_ENOTCONN; | |
14294 | ||
7c673cae FG |
14295 | ldout(cct, 3) << "ll_fsyncdir " << dirp << dendl; |
14296 | tout(cct) << "ll_fsyncdir" << std::endl; | |
f67539c2 | 14297 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 14298 | |
f67539c2 | 14299 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14300 | return _fsync(dirp->inode.get(), false); |
14301 | } | |
14302 | ||
14303 | int Client::ll_open(Inode *in, int flags, Fh **fhp, const UserPerm& perms) | |
14304 | { | |
11fdf7f2 | 14305 | ceph_assert(!(flags & O_CREAT)); |
7c673cae | 14306 | |
f67539c2 TL |
14307 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14308 | if (!mref_reader.is_state_satisfied()) | |
14309 | return -CEPHFS_ENOTCONN; | |
181888fb | 14310 | |
7c673cae FG |
14311 | vinodeno_t vino = _get_vino(in); |
14312 | ||
14313 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << dendl; | |
14314 | tout(cct) << "ll_open" << std::endl; | |
14315 | tout(cct) << vino.ino.val << std::endl; | |
14316 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
14317 | ||
f67539c2 TL |
14318 | std::scoped_lock lock(client_lock); |
14319 | ||
7c673cae | 14320 | int r; |
11fdf7f2 | 14321 | if (!fuse_default_permissions) { |
7c673cae FG |
14322 | r = may_open(in, flags, perms); |
14323 | if (r < 0) | |
14324 | goto out; | |
14325 | } | |
14326 | ||
14327 | r = _open(in, flags, 0, fhp /* may be NULL */, perms); | |
14328 | ||
14329 | out: | |
14330 | Fh *fhptr = fhp ? *fhp : NULL; | |
14331 | if (fhptr) { | |
14332 | ll_unclosed_fh_set.insert(fhptr); | |
14333 | } | |
f67539c2 | 14334 | tout(cct) << (uintptr_t)fhptr << std::endl; |
7c673cae FG |
14335 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << |
14336 | " = " << r << " (" << fhptr << ")" << dendl; | |
14337 | return r; | |
14338 | } | |
14339 | ||
14340 | int Client::_ll_create(Inode *parent, const char *name, mode_t mode, | |
14341 | int flags, InodeRef *in, int caps, Fh **fhp, | |
14342 | const UserPerm& perms) | |
14343 | { | |
14344 | *fhp = NULL; | |
14345 | ||
14346 | vinodeno_t vparent = _get_vino(parent); | |
14347 | ||
1adf2230 | 14348 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14349 | mode << dec << " " << ceph_flags_sys2wire(flags) << ", uid " << perms.uid() |
14350 | << ", gid " << perms.gid() << dendl; | |
14351 | tout(cct) << "ll_create" << std::endl; | |
14352 | tout(cct) << vparent.ino.val << std::endl; | |
14353 | tout(cct) << name << std::endl; | |
14354 | tout(cct) << mode << std::endl; | |
14355 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
14356 | ||
14357 | bool created = false; | |
14358 | int r = _lookup(parent, name, caps, in, perms); | |
14359 | ||
14360 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
f67539c2 | 14361 | return -CEPHFS_EEXIST; |
7c673cae | 14362 | |
f67539c2 | 14363 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
11fdf7f2 | 14364 | if (!fuse_default_permissions) { |
7c673cae FG |
14365 | r = may_create(parent, perms); |
14366 | if (r < 0) | |
14367 | goto out; | |
14368 | } | |
14369 | r = _create(parent, name, flags, mode, in, fhp, 0, 0, 0, NULL, &created, | |
f67539c2 | 14370 | perms, ""); |
7c673cae FG |
14371 | if (r < 0) |
14372 | goto out; | |
14373 | } | |
14374 | ||
14375 | if (r < 0) | |
14376 | goto out; | |
14377 | ||
11fdf7f2 | 14378 | ceph_assert(*in); |
7c673cae FG |
14379 | |
14380 | ldout(cct, 20) << "_ll_create created = " << created << dendl; | |
14381 | if (!created) { | |
11fdf7f2 | 14382 | if (!fuse_default_permissions) { |
7c673cae FG |
14383 | r = may_open(in->get(), flags, perms); |
14384 | if (r < 0) { | |
14385 | if (*fhp) { | |
14386 | int release_r = _release_fh(*fhp); | |
11fdf7f2 | 14387 | ceph_assert(release_r == 0); // during create, no async data ops should have happened |
7c673cae FG |
14388 | } |
14389 | goto out; | |
14390 | } | |
14391 | } | |
14392 | if (*fhp == NULL) { | |
14393 | r = _open(in->get(), flags, mode, fhp, perms); | |
14394 | if (r < 0) | |
14395 | goto out; | |
14396 | } | |
14397 | } | |
14398 | ||
14399 | out: | |
14400 | if (*fhp) { | |
14401 | ll_unclosed_fh_set.insert(*fhp); | |
14402 | } | |
14403 | ||
14404 | ino_t ino = 0; | |
14405 | if (r >= 0) { | |
14406 | Inode *inode = in->get(); | |
14407 | if (use_faked_inos()) | |
14408 | ino = inode->faked_ino; | |
14409 | else | |
14410 | ino = inode->ino; | |
14411 | } | |
14412 | ||
f67539c2 | 14413 | tout(cct) << (uintptr_t)*fhp << std::endl; |
7c673cae | 14414 | tout(cct) << ino << std::endl; |
1adf2230 | 14415 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14416 | mode << dec << " " << ceph_flags_sys2wire(flags) << " = " << r << " (" << |
14417 | *fhp << " " << hex << ino << dec << ")" << dendl; | |
14418 | ||
14419 | return r; | |
14420 | } | |
14421 | ||
14422 | int Client::ll_create(Inode *parent, const char *name, mode_t mode, | |
14423 | int flags, struct stat *attr, Inode **outp, Fh **fhp, | |
14424 | const UserPerm& perms) | |
14425 | { | |
f67539c2 TL |
14426 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14427 | if (!mref_reader.is_state_satisfied()) | |
14428 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14429 | |
f67539c2 TL |
14430 | std::scoped_lock lock(client_lock); |
14431 | InodeRef in; | |
181888fb | 14432 | |
7c673cae FG |
14433 | int r = _ll_create(parent, name, mode, flags, &in, CEPH_STAT_CAP_INODE_ALL, |
14434 | fhp, perms); | |
14435 | if (r >= 0) { | |
11fdf7f2 | 14436 | ceph_assert(in); |
7c673cae FG |
14437 | |
14438 | // passing an Inode in outp requires an additional ref | |
14439 | if (outp) { | |
14440 | _ll_get(in.get()); | |
14441 | *outp = in.get(); | |
14442 | } | |
14443 | fill_stat(in, attr); | |
14444 | } else { | |
14445 | attr->st_ino = 0; | |
14446 | } | |
14447 | ||
14448 | return r; | |
14449 | } | |
14450 | ||
14451 | int Client::ll_createx(Inode *parent, const char *name, mode_t mode, | |
14452 | int oflags, Inode **outp, Fh **fhp, | |
14453 | struct ceph_statx *stx, unsigned want, unsigned lflags, | |
14454 | const UserPerm& perms) | |
14455 | { | |
14456 | unsigned caps = statx_to_mask(lflags, want); | |
f67539c2 TL |
14457 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14458 | if (!mref_reader.is_state_satisfied()) | |
14459 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14460 | |
f67539c2 TL |
14461 | std::scoped_lock lock(client_lock); |
14462 | InodeRef in; | |
7c673cae FG |
14463 | |
14464 | int r = _ll_create(parent, name, mode, oflags, &in, caps, fhp, perms); | |
14465 | if (r >= 0) { | |
11fdf7f2 | 14466 | ceph_assert(in); |
7c673cae FG |
14467 | |
14468 | // passing an Inode in outp requires an additional ref | |
14469 | if (outp) { | |
14470 | _ll_get(in.get()); | |
14471 | *outp = in.get(); | |
14472 | } | |
14473 | fill_statx(in, caps, stx); | |
14474 | } else { | |
14475 | stx->stx_ino = 0; | |
14476 | stx->stx_mask = 0; | |
14477 | } | |
14478 | ||
14479 | return r; | |
14480 | } | |
14481 | ||
14482 | loff_t Client::ll_lseek(Fh *fh, loff_t offset, int whence) | |
14483 | { | |
f67539c2 TL |
14484 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14485 | if (!mref_reader.is_state_satisfied()) | |
14486 | return -CEPHFS_ENOTCONN; | |
14487 | ||
7c673cae FG |
14488 | tout(cct) << "ll_lseek" << std::endl; |
14489 | tout(cct) << offset << std::endl; | |
14490 | tout(cct) << whence << std::endl; | |
14491 | ||
f67539c2 | 14492 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14493 | return _lseek(fh, offset, whence); |
14494 | } | |
14495 | ||
14496 | int Client::ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl) | |
14497 | { | |
f67539c2 TL |
14498 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14499 | if (!mref_reader.is_state_satisfied()) | |
14500 | return -CEPHFS_ENOTCONN; | |
14501 | ||
7c673cae FG |
14502 | ldout(cct, 3) << "ll_read " << fh << " " << fh->inode->ino << " " << " " << off << "~" << len << dendl; |
14503 | tout(cct) << "ll_read" << std::endl; | |
f67539c2 | 14504 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14505 | tout(cct) << off << std::endl; |
14506 | tout(cct) << len << std::endl; | |
14507 | ||
11fdf7f2 TL |
14508 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14509 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14510 | std::scoped_lock lock(client_lock); |
14511 | ||
f6b5b4d7 TL |
14512 | int r = _read(fh, off, len, bl); |
14513 | ldout(cct, 3) << "ll_read " << fh << " " << off << "~" << len << " = " << r | |
14514 | << dendl; | |
14515 | return r; | |
7c673cae FG |
14516 | } |
14517 | ||
14518 | int Client::ll_read_block(Inode *in, uint64_t blockid, | |
14519 | char *buf, | |
14520 | uint64_t offset, | |
14521 | uint64_t length, | |
14522 | file_layout_t* layout) | |
14523 | { | |
f67539c2 TL |
14524 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14525 | if (!mref_reader.is_state_satisfied()) | |
14526 | return -CEPHFS_ENOTCONN; | |
181888fb | 14527 | |
b32b8144 | 14528 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14529 | object_t oid = file_object_t(vino.ino, blockid); |
14530 | C_SaferCond onfinish; | |
14531 | bufferlist bl; | |
14532 | ||
14533 | objecter->read(oid, | |
14534 | object_locator_t(layout->pool_id), | |
14535 | offset, | |
14536 | length, | |
14537 | vino.snapid, | |
14538 | &bl, | |
14539 | CEPH_OSD_FLAG_READ, | |
14540 | &onfinish); | |
14541 | ||
7c673cae | 14542 | int r = onfinish.wait(); |
7c673cae | 14543 | if (r >= 0) { |
9f95a23c | 14544 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
14545 | r = bl.length(); |
14546 | } | |
14547 | ||
14548 | return r; | |
14549 | } | |
14550 | ||
14551 | /* It appears that the OSD doesn't return success unless the entire | |
14552 | buffer was written, return the write length on success. */ | |
14553 | ||
14554 | int Client::ll_write_block(Inode *in, uint64_t blockid, | |
14555 | char* buf, uint64_t offset, | |
14556 | uint64_t length, file_layout_t* layout, | |
14557 | uint64_t snapseq, uint32_t sync) | |
14558 | { | |
7c673cae | 14559 | vinodeno_t vino = ll_get_vino(in); |
7c673cae | 14560 | int r = 0; |
11fdf7f2 | 14561 | std::unique_ptr<C_SaferCond> onsafe = nullptr; |
f67539c2 TL |
14562 | |
14563 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
14564 | if (!mref_reader.is_state_satisfied()) | |
14565 | return -CEPHFS_ENOTCONN; | |
14566 | ||
7c673cae | 14567 | if (length == 0) { |
f67539c2 | 14568 | return -CEPHFS_EINVAL; |
7c673cae FG |
14569 | } |
14570 | if (true || sync) { | |
14571 | /* if write is stable, the epilogue is waiting on | |
14572 | * flock */ | |
11fdf7f2 | 14573 | onsafe.reset(new C_SaferCond("Client::ll_write_block flock")); |
7c673cae FG |
14574 | } |
14575 | object_t oid = file_object_t(vino.ino, blockid); | |
14576 | SnapContext fakesnap; | |
11fdf7f2 TL |
14577 | ceph::bufferlist bl; |
14578 | if (length > 0) { | |
14579 | bl.push_back(buffer::copy(buf, length)); | |
14580 | } | |
7c673cae FG |
14581 | |
14582 | ldout(cct, 1) << "ll_block_write for " << vino.ino << "." << blockid | |
14583 | << dendl; | |
14584 | ||
14585 | fakesnap.seq = snapseq; | |
14586 | ||
14587 | /* lock just in time */ | |
7c673cae FG |
14588 | objecter->write(oid, |
14589 | object_locator_t(layout->pool_id), | |
14590 | offset, | |
14591 | length, | |
14592 | fakesnap, | |
14593 | bl, | |
14594 | ceph::real_clock::now(), | |
14595 | 0, | |
11fdf7f2 | 14596 | onsafe.get()); |
7c673cae | 14597 | |
11fdf7f2 TL |
14598 | if (nullptr != onsafe) { |
14599 | r = onsafe->wait(); | |
7c673cae FG |
14600 | } |
14601 | ||
14602 | if (r < 0) { | |
14603 | return r; | |
14604 | } else { | |
14605 | return length; | |
14606 | } | |
14607 | } | |
14608 | ||
14609 | int Client::ll_commit_blocks(Inode *in, | |
14610 | uint64_t offset, | |
14611 | uint64_t length) | |
14612 | { | |
7c673cae FG |
14613 | /* |
14614 | BarrierContext *bctx; | |
b32b8144 | 14615 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14616 | uint64_t ino = vino.ino; |
14617 | ||
14618 | ldout(cct, 1) << "ll_commit_blocks for " << vino.ino << " from " | |
14619 | << offset << " to " << length << dendl; | |
14620 | ||
14621 | if (length == 0) { | |
f67539c2 | 14622 | return -CEPHFS_EINVAL; |
7c673cae FG |
14623 | } |
14624 | ||
f67539c2 | 14625 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14626 | map<uint64_t, BarrierContext*>::iterator p = barriers.find(ino); |
14627 | if (p != barriers.end()) { | |
14628 | barrier_interval civ(offset, offset + length); | |
14629 | p->second->commit_barrier(civ); | |
14630 | } | |
14631 | */ | |
14632 | return 0; | |
14633 | } | |
14634 | ||
14635 | int Client::ll_write(Fh *fh, loff_t off, loff_t len, const char *data) | |
14636 | { | |
7c673cae FG |
14637 | ldout(cct, 3) << "ll_write " << fh << " " << fh->inode->ino << " " << off << |
14638 | "~" << len << dendl; | |
14639 | tout(cct) << "ll_write" << std::endl; | |
f67539c2 | 14640 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14641 | tout(cct) << off << std::endl; |
14642 | tout(cct) << len << std::endl; | |
14643 | ||
f67539c2 TL |
14644 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14645 | if (!mref_reader.is_state_satisfied()) | |
14646 | return -CEPHFS_ENOTCONN; | |
181888fb | 14647 | |
11fdf7f2 TL |
14648 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14649 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14650 | std::scoped_lock lock(client_lock); |
14651 | ||
7c673cae FG |
14652 | int r = _write(fh, off, len, data, NULL, 0); |
14653 | ldout(cct, 3) << "ll_write " << fh << " " << off << "~" << len << " = " << r | |
14654 | << dendl; | |
14655 | return r; | |
14656 | } | |
14657 | ||
11fdf7f2 TL |
14658 | int64_t Client::ll_writev(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) |
14659 | { | |
f67539c2 TL |
14660 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14661 | if (!mref_reader.is_state_satisfied()) | |
14662 | return -CEPHFS_ENOTCONN; | |
14663 | ||
20effc67 TL |
14664 | std::scoped_lock cl(client_lock); |
14665 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, true, false); | |
11fdf7f2 TL |
14666 | } |
14667 | ||
14668 | int64_t Client::ll_readv(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) | |
14669 | { | |
f67539c2 TL |
14670 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14671 | if (!mref_reader.is_state_satisfied()) | |
14672 | return -CEPHFS_ENOTCONN; | |
14673 | ||
20effc67 TL |
14674 | std::scoped_lock cl(client_lock); |
14675 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, false, false); | |
11fdf7f2 TL |
14676 | } |
14677 | ||
7c673cae FG |
14678 | int Client::ll_flush(Fh *fh) |
14679 | { | |
f67539c2 TL |
14680 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14681 | if (!mref_reader.is_state_satisfied()) | |
14682 | return -CEPHFS_ENOTCONN; | |
14683 | ||
7c673cae FG |
14684 | ldout(cct, 3) << "ll_flush " << fh << " " << fh->inode->ino << " " << dendl; |
14685 | tout(cct) << "ll_flush" << std::endl; | |
f67539c2 | 14686 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14687 | |
f67539c2 | 14688 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14689 | return _flush(fh); |
14690 | } | |
14691 | ||
14692 | int Client::ll_fsync(Fh *fh, bool syncdataonly) | |
14693 | { | |
f67539c2 TL |
14694 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14695 | if (!mref_reader.is_state_satisfied()) | |
14696 | return -CEPHFS_ENOTCONN; | |
14697 | ||
7c673cae FG |
14698 | ldout(cct, 3) << "ll_fsync " << fh << " " << fh->inode->ino << " " << dendl; |
14699 | tout(cct) << "ll_fsync" << std::endl; | |
f67539c2 | 14700 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14701 | |
f67539c2 | 14702 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14703 | int r = _fsync(fh, syncdataonly); |
14704 | if (r) { | |
14705 | // If we're returning an error, clear it from the FH | |
14706 | fh->take_async_err(); | |
14707 | } | |
14708 | return r; | |
14709 | } | |
14710 | ||
28e407b8 AA |
14711 | int Client::ll_sync_inode(Inode *in, bool syncdataonly) |
14712 | { | |
f67539c2 TL |
14713 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14714 | if (!mref_reader.is_state_satisfied()) | |
14715 | return -CEPHFS_ENOTCONN; | |
14716 | ||
28e407b8 AA |
14717 | ldout(cct, 3) << "ll_sync_inode " << *in << " " << dendl; |
14718 | tout(cct) << "ll_sync_inode" << std::endl; | |
f67539c2 | 14719 | tout(cct) << (uintptr_t)in << std::endl; |
28e407b8 | 14720 | |
f67539c2 | 14721 | std::scoped_lock lock(client_lock); |
28e407b8 AA |
14722 | return _fsync(in, syncdataonly); |
14723 | } | |
14724 | ||
7c673cae FG |
14725 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
14726 | { | |
f67539c2 TL |
14727 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
14728 | ||
7c673cae | 14729 | if (offset < 0 || length <= 0) |
f67539c2 | 14730 | return -CEPHFS_EINVAL; |
7c673cae FG |
14731 | |
14732 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
f67539c2 | 14733 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14734 | |
14735 | if ((mode & FALLOC_FL_PUNCH_HOLE) && !(mode & FALLOC_FL_KEEP_SIZE)) | |
f67539c2 | 14736 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14737 | |
14738 | Inode *in = fh->inode.get(); | |
14739 | ||
14740 | if (objecter->osdmap_pool_full(in->layout.pool_id) && | |
14741 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
f67539c2 | 14742 | return -CEPHFS_ENOSPC; |
7c673cae FG |
14743 | } |
14744 | ||
14745 | if (in->snapid != CEPH_NOSNAP) | |
f67539c2 | 14746 | return -CEPHFS_EROFS; |
7c673cae FG |
14747 | |
14748 | if ((fh->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 14749 | return -CEPHFS_EBADF; |
7c673cae FG |
14750 | |
14751 | uint64_t size = offset + length; | |
14752 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && | |
14753 | size > in->size && | |
11fdf7f2 | 14754 | is_quota_bytes_exceeded(in, size - in->size, fh->actor_perms)) { |
f67539c2 | 14755 | return -CEPHFS_EDQUOT; |
7c673cae FG |
14756 | } |
14757 | ||
14758 | int have; | |
f6b5b4d7 | 14759 | int r = get_caps(fh, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER, &have, -1); |
7c673cae FG |
14760 | if (r < 0) |
14761 | return r; | |
14762 | ||
11fdf7f2 | 14763 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
7c673cae FG |
14764 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
14765 | if (in->inline_version < CEPH_INLINE_NONE && | |
14766 | (have & CEPH_CAP_FILE_BUFFER)) { | |
14767 | bufferlist bl; | |
9f95a23c | 14768 | auto inline_iter = in->inline_data.cbegin(); |
7c673cae FG |
14769 | int len = in->inline_data.length(); |
14770 | if (offset < len) { | |
14771 | if (offset > 0) | |
9f95a23c | 14772 | inline_iter.copy(offset, bl); |
7c673cae FG |
14773 | int size = length; |
14774 | if (offset + size > len) | |
14775 | size = len - offset; | |
14776 | if (size > 0) | |
14777 | bl.append_zero(size); | |
9f95a23c TL |
14778 | if (offset + size < len) { |
14779 | inline_iter += size; | |
14780 | inline_iter.copy(len - offset - size, bl); | |
14781 | } | |
7c673cae FG |
14782 | in->inline_data = bl; |
14783 | in->inline_version++; | |
14784 | } | |
91327a77 | 14785 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14786 | in->change_attr++; |
28e407b8 | 14787 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14788 | } else { |
14789 | if (in->inline_version < CEPH_INLINE_NONE) { | |
11fdf7f2 TL |
14790 | onuninline.reset(new C_SaferCond("Client::_fallocate_uninline_data flock")); |
14791 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
14792 | } |
14793 | ||
11fdf7f2 | 14794 | C_SaferCond onfinish("Client::_punch_hole flock"); |
7c673cae | 14795 | |
7c673cae FG |
14796 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
14797 | ||
14798 | _invalidate_inode_cache(in, offset, length); | |
14799 | filer->zero(in->ino, &in->layout, | |
14800 | in->snaprealm->get_snap_context(), | |
14801 | offset, length, | |
14802 | ceph::real_clock::now(), | |
11fdf7f2 | 14803 | 0, true, &onfinish); |
91327a77 | 14804 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14805 | in->change_attr++; |
28e407b8 | 14806 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14807 | |
9f95a23c | 14808 | client_lock.unlock(); |
11fdf7f2 | 14809 | onfinish.wait(); |
9f95a23c | 14810 | client_lock.lock(); |
f67539c2 | 14811 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
14812 | } |
14813 | } else if (!(mode & FALLOC_FL_KEEP_SIZE)) { | |
14814 | uint64_t size = offset + length; | |
14815 | if (size > in->size) { | |
14816 | in->size = size; | |
91327a77 | 14817 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14818 | in->change_attr++; |
28e407b8 | 14819 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14820 | |
11fdf7f2 | 14821 | if (is_quota_bytes_approaching(in, fh->actor_perms)) { |
7c673cae | 14822 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
14823 | } else if (is_max_size_approaching(in)) { |
14824 | check_caps(in, 0); | |
7c673cae FG |
14825 | } |
14826 | } | |
14827 | } | |
14828 | ||
11fdf7f2 | 14829 | if (nullptr != onuninline) { |
9f95a23c | 14830 | client_lock.unlock(); |
11fdf7f2 | 14831 | int ret = onuninline->wait(); |
9f95a23c | 14832 | client_lock.lock(); |
7c673cae | 14833 | |
f67539c2 | 14834 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
14835 | in->inline_data.clear(); |
14836 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 14837 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14838 | check_caps(in, 0); |
14839 | } else | |
11fdf7f2 | 14840 | r = ret; |
7c673cae FG |
14841 | } |
14842 | ||
14843 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
14844 | return r; | |
14845 | } | |
7c673cae | 14846 | |
11fdf7f2 | 14847 | int Client::ll_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
7c673cae | 14848 | { |
f67539c2 TL |
14849 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14850 | if (!mref_reader.is_state_satisfied()) | |
14851 | return -CEPHFS_ENOTCONN; | |
14852 | ||
11fdf7f2 TL |
14853 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << dendl; |
14854 | tout(cct) << __func__ << " " << mode << " " << offset << " " << length << std::endl; | |
f67539c2 | 14855 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14856 | |
f67539c2 | 14857 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14858 | return _fallocate(fh, mode, offset, length); |
14859 | } | |
14860 | ||
14861 | int Client::fallocate(int fd, int mode, loff_t offset, loff_t length) | |
14862 | { | |
f67539c2 TL |
14863 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14864 | if (!mref_reader.is_state_satisfied()) | |
14865 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14866 | |
f67539c2 | 14867 | tout(cct) << __func__ << " " << " " << fd << mode << " " << offset << " " << length << std::endl; |
181888fb | 14868 | |
f67539c2 | 14869 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14870 | Fh *fh = get_filehandle(fd); |
14871 | if (!fh) | |
f67539c2 | 14872 | return -CEPHFS_EBADF; |
7c673cae FG |
14873 | #if defined(__linux__) && defined(O_PATH) |
14874 | if (fh->flags & O_PATH) | |
f67539c2 | 14875 | return -CEPHFS_EBADF; |
7c673cae FG |
14876 | #endif |
14877 | return _fallocate(fh, mode, offset, length); | |
14878 | } | |
14879 | ||
14880 | int Client::ll_release(Fh *fh) | |
14881 | { | |
f67539c2 TL |
14882 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14883 | if (!mref_reader.is_state_satisfied()) | |
14884 | return -CEPHFS_ENOTCONN; | |
91327a77 | 14885 | |
11fdf7f2 | 14886 | ldout(cct, 3) << __func__ << " (fh)" << fh << " " << fh->inode->ino << " " << |
7c673cae | 14887 | dendl; |
11fdf7f2 | 14888 | tout(cct) << __func__ << " (fh)" << std::endl; |
f67539c2 TL |
14889 | tout(cct) << (uintptr_t)fh << std::endl; |
14890 | ||
14891 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
14892 | |
14893 | if (ll_unclosed_fh_set.count(fh)) | |
14894 | ll_unclosed_fh_set.erase(fh); | |
14895 | return _release_fh(fh); | |
14896 | } | |
14897 | ||
14898 | int Client::ll_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
14899 | { | |
f67539c2 TL |
14900 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14901 | if (!mref_reader.is_state_satisfied()) | |
14902 | return -CEPHFS_ENOTCONN; | |
7c673cae FG |
14903 | |
14904 | ldout(cct, 3) << "ll_getlk (fh)" << fh << " " << fh->inode->ino << dendl; | |
f67539c2 | 14905 | tout(cct) << "ll_getk (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14906 | |
f67539c2 | 14907 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14908 | return _getlk(fh, fl, owner); |
14909 | } | |
14910 | ||
14911 | int Client::ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
14912 | { | |
f67539c2 TL |
14913 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14914 | if (!mref_reader.is_state_satisfied()) | |
14915 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14916 | |
11fdf7f2 | 14917 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14918 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14919 | |
f67539c2 | 14920 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14921 | return _setlk(fh, fl, owner, sleep); |
14922 | } | |
14923 | ||
14924 | int Client::ll_flock(Fh *fh, int cmd, uint64_t owner) | |
14925 | { | |
f67539c2 TL |
14926 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14927 | if (!mref_reader.is_state_satisfied()) | |
14928 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14929 | |
11fdf7f2 | 14930 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14931 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14932 | |
f67539c2 | 14933 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14934 | return _flock(fh, cmd, owner); |
14935 | } | |
14936 | ||
b32b8144 FG |
14937 | int Client::set_deleg_timeout(uint32_t timeout) |
14938 | { | |
f67539c2 | 14939 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14940 | |
14941 | /* | |
f67539c2 | 14942 | * The whole point is to prevent blocklisting so we must time out the |
b32b8144 FG |
14943 | * delegation before the session autoclose timeout kicks in. |
14944 | */ | |
14945 | if (timeout >= mdsmap->get_session_autoclose()) | |
f67539c2 | 14946 | return -CEPHFS_EINVAL; |
b32b8144 FG |
14947 | |
14948 | deleg_timeout = timeout; | |
14949 | return 0; | |
14950 | } | |
14951 | ||
14952 | int Client::ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv) | |
14953 | { | |
f67539c2 | 14954 | int ret = -CEPHFS_EINVAL; |
b32b8144 | 14955 | |
f67539c2 TL |
14956 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14957 | if (!mref_reader.is_state_satisfied()) | |
14958 | return -CEPHFS_ENOTCONN; | |
b32b8144 | 14959 | |
f67539c2 | 14960 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14961 | |
14962 | Inode *inode = fh->inode.get(); | |
14963 | ||
14964 | switch(cmd) { | |
14965 | case CEPH_DELEGATION_NONE: | |
14966 | inode->unset_deleg(fh); | |
14967 | ret = 0; | |
14968 | break; | |
14969 | default: | |
14970 | try { | |
14971 | ret = inode->set_deleg(fh, cmd, cb, priv); | |
11fdf7f2 | 14972 | } catch (std::bad_alloc&) { |
f67539c2 | 14973 | ret = -CEPHFS_ENOMEM; |
b32b8144 FG |
14974 | } |
14975 | break; | |
14976 | } | |
14977 | return ret; | |
14978 | } | |
14979 | ||
7c673cae FG |
14980 | class C_Client_RequestInterrupt : public Context { |
14981 | private: | |
14982 | Client *client; | |
14983 | MetaRequest *req; | |
14984 | public: | |
14985 | C_Client_RequestInterrupt(Client *c, MetaRequest *r) : client(c), req(r) { | |
14986 | req->get(); | |
14987 | } | |
14988 | void finish(int r) override { | |
f67539c2 | 14989 | std::scoped_lock l(client->client_lock); |
11fdf7f2 | 14990 | ceph_assert(req->head.op == CEPH_MDS_OP_SETFILELOCK); |
7c673cae FG |
14991 | client->_interrupt_filelock(req); |
14992 | client->put_request(req); | |
14993 | } | |
14994 | }; | |
14995 | ||
14996 | void Client::ll_interrupt(void *d) | |
14997 | { | |
14998 | MetaRequest *req = static_cast<MetaRequest*>(d); | |
11fdf7f2 TL |
14999 | ldout(cct, 3) << __func__ << " tid " << req->get_tid() << dendl; |
15000 | tout(cct) << __func__ << " tid " << req->get_tid() << std::endl; | |
7c673cae FG |
15001 | interrupt_finisher.queue(new C_Client_RequestInterrupt(this, req)); |
15002 | } | |
15003 | ||
15004 | // ========================================= | |
15005 | // layout | |
15006 | ||
15007 | // expose file layouts | |
15008 | ||
15009 | int Client::describe_layout(const char *relpath, file_layout_t *lp, | |
15010 | const UserPerm& perms) | |
15011 | { | |
f67539c2 TL |
15012 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15013 | if (!mref_reader.is_state_satisfied()) | |
15014 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15015 | |
f67539c2 | 15016 | std::scoped_lock lock(client_lock); |
181888fb | 15017 | |
7c673cae FG |
15018 | filepath path(relpath); |
15019 | InodeRef in; | |
15020 | int r = path_walk(path, &in, perms); | |
15021 | if (r < 0) | |
15022 | return r; | |
15023 | ||
15024 | *lp = in->layout; | |
15025 | ||
11fdf7f2 | 15026 | ldout(cct, 3) << __func__ << "(" << relpath << ") = 0" << dendl; |
7c673cae FG |
15027 | return 0; |
15028 | } | |
15029 | ||
15030 | int Client::fdescribe_layout(int fd, file_layout_t *lp) | |
15031 | { | |
f67539c2 TL |
15032 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15033 | if (!mref_reader.is_state_satisfied()) | |
15034 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15035 | |
f67539c2 | 15036 | std::scoped_lock lock(client_lock); |
181888fb | 15037 | |
7c673cae FG |
15038 | Fh *f = get_filehandle(fd); |
15039 | if (!f) | |
f67539c2 | 15040 | return -CEPHFS_EBADF; |
7c673cae FG |
15041 | Inode *in = f->inode.get(); |
15042 | ||
15043 | *lp = in->layout; | |
15044 | ||
11fdf7f2 | 15045 | ldout(cct, 3) << __func__ << "(" << fd << ") = 0" << dendl; |
7c673cae FG |
15046 | return 0; |
15047 | } | |
15048 | ||
d2e6a577 FG |
15049 | int64_t Client::get_default_pool_id() |
15050 | { | |
f67539c2 TL |
15051 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15052 | if (!mref_reader.is_state_satisfied()) | |
15053 | return -CEPHFS_ENOTCONN; | |
181888fb | 15054 | |
f67539c2 | 15055 | std::scoped_lock lock(client_lock); |
181888fb | 15056 | |
d2e6a577 FG |
15057 | /* first data pool is the default */ |
15058 | return mdsmap->get_first_data_pool(); | |
15059 | } | |
7c673cae FG |
15060 | |
15061 | // expose osdmap | |
15062 | ||
15063 | int64_t Client::get_pool_id(const char *pool_name) | |
15064 | { | |
f67539c2 TL |
15065 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15066 | if (!mref_reader.is_state_satisfied()) | |
15067 | return -CEPHFS_ENOTCONN; | |
181888fb | 15068 | |
f67539c2 | 15069 | std::scoped_lock lock(client_lock); |
181888fb | 15070 | |
7c673cae FG |
15071 | return objecter->with_osdmap(std::mem_fn(&OSDMap::lookup_pg_pool_name), |
15072 | pool_name); | |
15073 | } | |
15074 | ||
15075 | string Client::get_pool_name(int64_t pool) | |
15076 | { | |
f67539c2 TL |
15077 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15078 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
15079 | return string(); |
15080 | ||
f67539c2 TL |
15081 | std::scoped_lock lock(client_lock); |
15082 | ||
7c673cae FG |
15083 | return objecter->with_osdmap([pool](const OSDMap& o) { |
15084 | return o.have_pg_pool(pool) ? o.get_pool_name(pool) : string(); | |
15085 | }); | |
15086 | } | |
15087 | ||
15088 | int Client::get_pool_replication(int64_t pool) | |
15089 | { | |
f67539c2 TL |
15090 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15091 | if (!mref_reader.is_state_satisfied()) | |
15092 | return -CEPHFS_ENOTCONN; | |
181888fb | 15093 | |
f67539c2 | 15094 | std::scoped_lock lock(client_lock); |
181888fb | 15095 | |
7c673cae | 15096 | return objecter->with_osdmap([pool](const OSDMap& o) { |
f67539c2 | 15097 | return o.have_pg_pool(pool) ? o.get_pg_pool(pool)->get_size() : -CEPHFS_ENOENT; |
7c673cae FG |
15098 | }); |
15099 | } | |
15100 | ||
15101 | int Client::get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds) | |
15102 | { | |
f67539c2 TL |
15103 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15104 | if (!mref_reader.is_state_satisfied()) | |
15105 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15106 | |
f67539c2 | 15107 | std::scoped_lock lock(client_lock); |
181888fb | 15108 | |
7c673cae FG |
15109 | Fh *f = get_filehandle(fd); |
15110 | if (!f) | |
f67539c2 | 15111 | return -CEPHFS_EBADF; |
7c673cae FG |
15112 | Inode *in = f->inode.get(); |
15113 | ||
15114 | vector<ObjectExtent> extents; | |
15115 | Striper::file_to_extents(cct, in->ino, &in->layout, off, 1, in->truncate_size, extents); | |
11fdf7f2 | 15116 | ceph_assert(extents.size() == 1); |
7c673cae FG |
15117 | |
15118 | objecter->with_osdmap([&](const OSDMap& o) { | |
15119 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
15120 | o.pg_to_acting_osds(pg, osds); | |
15121 | }); | |
15122 | ||
15123 | if (osds.empty()) | |
f67539c2 | 15124 | return -CEPHFS_EINVAL; |
7c673cae FG |
15125 | |
15126 | /* | |
15127 | * Return the remainder of the extent (stripe unit) | |
15128 | * | |
15129 | * If length = 1 is passed to Striper::file_to_extents we get a single | |
15130 | * extent back, but its length is one so we still need to compute the length | |
15131 | * to the end of the stripe unit. | |
15132 | * | |
15133 | * If length = su then we may get 1 or 2 objects back in the extents vector | |
15134 | * which would have to be examined. Even then, the offsets are local to the | |
15135 | * object, so matching up to the file offset is extra work. | |
15136 | * | |
15137 | * It seems simpler to stick with length = 1 and manually compute the | |
15138 | * remainder. | |
15139 | */ | |
15140 | if (len) { | |
15141 | uint64_t su = in->layout.stripe_unit; | |
15142 | *len = su - (off % su); | |
15143 | } | |
15144 | ||
15145 | return 0; | |
15146 | } | |
15147 | ||
15148 | int Client::get_osd_crush_location(int id, vector<pair<string, string> >& path) | |
15149 | { | |
f67539c2 TL |
15150 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15151 | if (!mref_reader.is_state_satisfied()) | |
15152 | return -CEPHFS_ENOTCONN; | |
181888fb | 15153 | |
f67539c2 | 15154 | std::scoped_lock lock(client_lock); |
181888fb | 15155 | |
7c673cae | 15156 | if (id < 0) |
f67539c2 | 15157 | return -CEPHFS_EINVAL; |
7c673cae FG |
15158 | return objecter->with_osdmap([&](const OSDMap& o) { |
15159 | return o.crush->get_full_location_ordered(id, path); | |
15160 | }); | |
15161 | } | |
15162 | ||
15163 | int Client::get_file_stripe_address(int fd, loff_t offset, | |
15164 | vector<entity_addr_t>& address) | |
15165 | { | |
f67539c2 TL |
15166 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15167 | if (!mref_reader.is_state_satisfied()) | |
15168 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15169 | |
f67539c2 | 15170 | std::scoped_lock lock(client_lock); |
181888fb | 15171 | |
7c673cae FG |
15172 | Fh *f = get_filehandle(fd); |
15173 | if (!f) | |
f67539c2 | 15174 | return -CEPHFS_EBADF; |
7c673cae FG |
15175 | Inode *in = f->inode.get(); |
15176 | ||
15177 | // which object? | |
15178 | vector<ObjectExtent> extents; | |
15179 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, 1, | |
15180 | in->truncate_size, extents); | |
11fdf7f2 | 15181 | ceph_assert(extents.size() == 1); |
7c673cae FG |
15182 | |
15183 | // now we have the object and its 'layout' | |
15184 | return objecter->with_osdmap([&](const OSDMap& o) { | |
15185 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
15186 | vector<int> osds; | |
15187 | o.pg_to_acting_osds(pg, osds); | |
15188 | if (osds.empty()) | |
f67539c2 | 15189 | return -CEPHFS_EINVAL; |
7c673cae | 15190 | for (unsigned i = 0; i < osds.size(); i++) { |
11fdf7f2 | 15191 | entity_addr_t addr = o.get_addrs(osds[i]).front(); |
7c673cae FG |
15192 | address.push_back(addr); |
15193 | } | |
15194 | return 0; | |
15195 | }); | |
15196 | } | |
15197 | ||
15198 | int Client::get_osd_addr(int osd, entity_addr_t& addr) | |
15199 | { | |
f67539c2 TL |
15200 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15201 | if (!mref_reader.is_state_satisfied()) | |
15202 | return -CEPHFS_ENOTCONN; | |
181888fb | 15203 | |
f67539c2 | 15204 | std::scoped_lock lock(client_lock); |
181888fb | 15205 | |
7c673cae FG |
15206 | return objecter->with_osdmap([&](const OSDMap& o) { |
15207 | if (!o.exists(osd)) | |
f67539c2 | 15208 | return -CEPHFS_ENOENT; |
7c673cae | 15209 | |
11fdf7f2 | 15210 | addr = o.get_addrs(osd).front(); |
7c673cae FG |
15211 | return 0; |
15212 | }); | |
15213 | } | |
15214 | ||
15215 | int Client::enumerate_layout(int fd, vector<ObjectExtent>& result, | |
15216 | loff_t length, loff_t offset) | |
15217 | { | |
f67539c2 TL |
15218 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15219 | if (!mref_reader.is_state_satisfied()) | |
15220 | return -CEPHFS_ENOTCONN; | |
7c673cae | 15221 | |
f67539c2 | 15222 | std::scoped_lock lock(client_lock); |
181888fb | 15223 | |
7c673cae FG |
15224 | Fh *f = get_filehandle(fd); |
15225 | if (!f) | |
f67539c2 | 15226 | return -CEPHFS_EBADF; |
7c673cae FG |
15227 | Inode *in = f->inode.get(); |
15228 | ||
15229 | // map to a list of extents | |
15230 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, length, in->truncate_size, result); | |
15231 | ||
11fdf7f2 | 15232 | ldout(cct, 3) << __func__ << "(" << fd << ", " << length << ", " << offset << ") = 0" << dendl; |
7c673cae FG |
15233 | return 0; |
15234 | } | |
15235 | ||
15236 | ||
f67539c2 | 15237 | /* find an osd with the same ip. -CEPHFS_ENXIO if none. */ |
7c673cae FG |
15238 | int Client::get_local_osd() |
15239 | { | |
f67539c2 TL |
15240 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
15241 | if (!mref_reader.is_state_satisfied()) | |
15242 | return -CEPHFS_ENOTCONN; | |
181888fb | 15243 | |
f67539c2 | 15244 | std::scoped_lock lock(client_lock); |
181888fb | 15245 | |
7c673cae FG |
15246 | objecter->with_osdmap([this](const OSDMap& o) { |
15247 | if (o.get_epoch() != local_osd_epoch) { | |
11fdf7f2 | 15248 | local_osd = o.find_osd_on_ip(messenger->get_myaddrs().front()); |
7c673cae FG |
15249 | local_osd_epoch = o.get_epoch(); |
15250 | } | |
15251 | }); | |
15252 | return local_osd; | |
15253 | } | |
15254 | ||
15255 | ||
15256 | ||
15257 | ||
15258 | ||
15259 | ||
15260 | // =============================== | |
15261 | ||
15262 | void Client::ms_handle_connect(Connection *con) | |
15263 | { | |
11fdf7f2 | 15264 | ldout(cct, 10) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15265 | } |
15266 | ||
15267 | bool Client::ms_handle_reset(Connection *con) | |
15268 | { | |
11fdf7f2 | 15269 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15270 | return false; |
15271 | } | |
15272 | ||
15273 | void Client::ms_handle_remote_reset(Connection *con) | |
15274 | { | |
f67539c2 | 15275 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 15276 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15277 | switch (con->get_peer_type()) { |
15278 | case CEPH_ENTITY_TYPE_MDS: | |
15279 | { | |
15280 | // kludge to figure out which mds this is; fixme with a Connection* state | |
15281 | mds_rank_t mds = MDS_RANK_NONE; | |
20effc67 | 15282 | MetaSessionRef s = NULL; |
11fdf7f2 | 15283 | for (auto &p : mds_sessions) { |
b3b6e05e | 15284 | if (mdsmap->have_inst(p.first) && mdsmap->get_addrs(p.first) == con->get_peer_addrs()) { |
11fdf7f2 | 15285 | mds = p.first; |
20effc67 | 15286 | s = p.second; |
7c673cae FG |
15287 | } |
15288 | } | |
15289 | if (mds >= 0) { | |
20effc67 | 15290 | ceph_assert(s != NULL); |
7c673cae FG |
15291 | switch (s->state) { |
15292 | case MetaSession::STATE_CLOSING: | |
15293 | ldout(cct, 1) << "reset from mds we were closing; we'll call that closed" << dendl; | |
20effc67 | 15294 | _closed_mds_session(s.get()); |
7c673cae FG |
15295 | break; |
15296 | ||
15297 | case MetaSession::STATE_OPENING: | |
15298 | { | |
15299 | ldout(cct, 1) << "reset from mds we were opening; retrying" << dendl; | |
15300 | list<Context*> waiters; | |
15301 | waiters.swap(s->waiting_for_open); | |
20effc67 TL |
15302 | _closed_mds_session(s.get()); |
15303 | auto news = _get_or_open_mds_session(mds); | |
7c673cae FG |
15304 | news->waiting_for_open.swap(waiters); |
15305 | } | |
15306 | break; | |
15307 | ||
15308 | case MetaSession::STATE_OPEN: | |
15309 | { | |
f67539c2 | 15310 | objecter->maybe_request_map(); /* to check if we are blocklisted */ |
f6b5b4d7 | 15311 | if (cct->_conf.get_val<bool>("client_reconnect_stale")) { |
7c673cae | 15312 | ldout(cct, 1) << "reset from mds we were open; close mds session for reconnect" << dendl; |
20effc67 | 15313 | _closed_mds_session(s.get()); |
7c673cae FG |
15314 | } else { |
15315 | ldout(cct, 1) << "reset from mds we were open; mark session as stale" << dendl; | |
15316 | s->state = MetaSession::STATE_STALE; | |
15317 | } | |
15318 | } | |
15319 | break; | |
15320 | ||
15321 | case MetaSession::STATE_NEW: | |
15322 | case MetaSession::STATE_CLOSED: | |
15323 | default: | |
15324 | break; | |
15325 | } | |
15326 | } | |
15327 | } | |
15328 | break; | |
15329 | } | |
15330 | } | |
15331 | ||
15332 | bool Client::ms_handle_refused(Connection *con) | |
15333 | { | |
11fdf7f2 | 15334 | ldout(cct, 1) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15335 | return false; |
15336 | } | |
15337 | ||
7c673cae FG |
15338 | Inode *Client::get_quota_root(Inode *in, const UserPerm& perms) |
15339 | { | |
11fdf7f2 TL |
15340 | Inode *quota_in = root_ancestor; |
15341 | SnapRealm *realm = in->snaprealm; | |
2a845540 TL |
15342 | |
15343 | if (!cct->_conf.get_val<bool>("client_quota")) | |
15344 | return NULL; | |
15345 | ||
11fdf7f2 TL |
15346 | while (realm) { |
15347 | ldout(cct, 10) << __func__ << " realm " << realm->ino << dendl; | |
15348 | if (realm->ino != in->ino) { | |
15349 | auto p = inode_map.find(vinodeno_t(realm->ino, CEPH_NOSNAP)); | |
15350 | if (p == inode_map.end()) | |
15351 | break; | |
7c673cae | 15352 | |
11fdf7f2 TL |
15353 | if (p->second->quota.is_enable()) { |
15354 | quota_in = p->second; | |
15355 | break; | |
7c673cae | 15356 | } |
7c673cae | 15357 | } |
11fdf7f2 | 15358 | realm = realm->pparent; |
7c673cae | 15359 | } |
11fdf7f2 TL |
15360 | ldout(cct, 10) << __func__ << " " << in->vino() << " -> " << quota_in->vino() << dendl; |
15361 | return quota_in; | |
7c673cae FG |
15362 | } |
15363 | ||
15364 | /** | |
15365 | * Traverse quota ancestors of the Inode, return true | |
15366 | * if any of them passes the passed function | |
15367 | */ | |
15368 | bool Client::check_quota_condition(Inode *in, const UserPerm& perms, | |
15369 | std::function<bool (const Inode &in)> test) | |
15370 | { | |
2a845540 TL |
15371 | if (!cct->_conf.get_val<bool>("client_quota")) |
15372 | return false; | |
15373 | ||
7c673cae | 15374 | while (true) { |
11fdf7f2 | 15375 | ceph_assert(in != NULL); |
7c673cae FG |
15376 | if (test(*in)) { |
15377 | return true; | |
15378 | } | |
15379 | ||
15380 | if (in == root_ancestor) { | |
15381 | // We're done traversing, drop out | |
15382 | return false; | |
15383 | } else { | |
15384 | // Continue up the tree | |
15385 | in = get_quota_root(in, perms); | |
15386 | } | |
15387 | } | |
15388 | ||
15389 | return false; | |
15390 | } | |
15391 | ||
15392 | bool Client::is_quota_files_exceeded(Inode *in, const UserPerm& perms) | |
15393 | { | |
15394 | return check_quota_condition(in, perms, | |
15395 | [](const Inode &in) { | |
15396 | return in.quota.max_files && in.rstat.rsize() >= in.quota.max_files; | |
15397 | }); | |
15398 | } | |
15399 | ||
15400 | bool Client::is_quota_bytes_exceeded(Inode *in, int64_t new_bytes, | |
11fdf7f2 | 15401 | const UserPerm& perms) |
7c673cae FG |
15402 | { |
15403 | return check_quota_condition(in, perms, | |
11fdf7f2 | 15404 | [&new_bytes](const Inode &in) { |
7c673cae FG |
15405 | return in.quota.max_bytes && (in.rstat.rbytes + new_bytes) |
15406 | > in.quota.max_bytes; | |
15407 | }); | |
15408 | } | |
15409 | ||
11fdf7f2 | 15410 | bool Client::is_quota_bytes_approaching(Inode *in, const UserPerm& perms) |
7c673cae | 15411 | { |
9f95a23c TL |
15412 | ceph_assert(in->size >= in->reported_size); |
15413 | const uint64_t size = in->size - in->reported_size; | |
11fdf7f2 | 15414 | return check_quota_condition(in, perms, |
9f95a23c | 15415 | [&size](const Inode &in) { |
11fdf7f2 TL |
15416 | if (in.quota.max_bytes) { |
15417 | if (in.rstat.rbytes >= in.quota.max_bytes) { | |
15418 | return true; | |
15419 | } | |
15420 | ||
11fdf7f2 | 15421 | const uint64_t space = in.quota.max_bytes - in.rstat.rbytes; |
11fdf7f2 TL |
15422 | return (space >> 4) < size; |
15423 | } else { | |
15424 | return false; | |
15425 | } | |
15426 | }); | |
7c673cae FG |
15427 | } |
15428 | ||
15429 | enum { | |
15430 | POOL_CHECKED = 1, | |
15431 | POOL_CHECKING = 2, | |
15432 | POOL_READ = 4, | |
15433 | POOL_WRITE = 8, | |
15434 | }; | |
15435 | ||
15436 | int Client::check_pool_perm(Inode *in, int need) | |
15437 | { | |
f67539c2 TL |
15438 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
15439 | ||
7c673cae FG |
15440 | if (!cct->_conf->client_check_pool_perm) |
15441 | return 0; | |
15442 | ||
f67539c2 TL |
15443 | /* Only need to do this for regular files */ |
15444 | if (!in->is_file()) | |
15445 | return 0; | |
15446 | ||
7c673cae FG |
15447 | int64_t pool_id = in->layout.pool_id; |
15448 | std::string pool_ns = in->layout.pool_ns; | |
15449 | std::pair<int64_t, std::string> perm_key(pool_id, pool_ns); | |
15450 | int have = 0; | |
15451 | while (true) { | |
15452 | auto it = pool_perms.find(perm_key); | |
15453 | if (it == pool_perms.end()) | |
15454 | break; | |
15455 | if (it->second == POOL_CHECKING) { | |
15456 | // avoid concurrent checkings | |
15457 | wait_on_list(waiting_for_pool_perm); | |
15458 | } else { | |
15459 | have = it->second; | |
11fdf7f2 | 15460 | ceph_assert(have & POOL_CHECKED); |
7c673cae FG |
15461 | break; |
15462 | } | |
15463 | } | |
15464 | ||
15465 | if (!have) { | |
15466 | if (in->snapid != CEPH_NOSNAP) { | |
15467 | // pool permission check needs to write to the first object. But for snapshot, | |
20effc67 | 15468 | // head of the first object may have already been deleted. To avoid creating |
7c673cae FG |
15469 | // orphan object, skip the check for now. |
15470 | return 0; | |
15471 | } | |
15472 | ||
15473 | pool_perms[perm_key] = POOL_CHECKING; | |
15474 | ||
15475 | char oid_buf[32]; | |
15476 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (unsigned long long)in->ino); | |
15477 | object_t oid = oid_buf; | |
15478 | ||
15479 | SnapContext nullsnapc; | |
15480 | ||
15481 | C_SaferCond rd_cond; | |
15482 | ObjectOperation rd_op; | |
f67539c2 | 15483 | rd_op.stat(nullptr, nullptr, nullptr); |
7c673cae FG |
15484 | |
15485 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), rd_op, | |
15486 | nullsnapc, ceph::real_clock::now(), 0, &rd_cond); | |
15487 | ||
15488 | C_SaferCond wr_cond; | |
15489 | ObjectOperation wr_op; | |
15490 | wr_op.create(true); | |
15491 | ||
15492 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), wr_op, | |
15493 | nullsnapc, ceph::real_clock::now(), 0, &wr_cond); | |
15494 | ||
9f95a23c | 15495 | client_lock.unlock(); |
7c673cae FG |
15496 | int rd_ret = rd_cond.wait(); |
15497 | int wr_ret = wr_cond.wait(); | |
9f95a23c | 15498 | client_lock.lock(); |
7c673cae FG |
15499 | |
15500 | bool errored = false; | |
15501 | ||
f67539c2 | 15502 | if (rd_ret == 0 || rd_ret == -CEPHFS_ENOENT) |
7c673cae | 15503 | have |= POOL_READ; |
f67539c2 | 15504 | else if (rd_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15505 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15506 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15507 | errored = true; | |
15508 | } | |
15509 | ||
f67539c2 | 15510 | if (wr_ret == 0 || wr_ret == -CEPHFS_EEXIST) |
7c673cae | 15511 | have |= POOL_WRITE; |
f67539c2 | 15512 | else if (wr_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15513 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15514 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15515 | errored = true; | |
15516 | } | |
15517 | ||
15518 | if (errored) { | |
15519 | // Indeterminate: erase CHECKING state so that subsequent calls re-check. | |
15520 | // Raise EIO because actual error code might be misleading for | |
15521 | // userspace filesystem user. | |
15522 | pool_perms.erase(perm_key); | |
15523 | signal_cond_list(waiting_for_pool_perm); | |
f67539c2 | 15524 | return -CEPHFS_EIO; |
7c673cae FG |
15525 | } |
15526 | ||
15527 | pool_perms[perm_key] = have | POOL_CHECKED; | |
15528 | signal_cond_list(waiting_for_pool_perm); | |
15529 | } | |
15530 | ||
15531 | if ((need & CEPH_CAP_FILE_RD) && !(have & POOL_READ)) { | |
11fdf7f2 | 15532 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15533 | << " need " << ccap_string(need) << ", but no read perm" << dendl; |
f67539c2 | 15534 | return -CEPHFS_EPERM; |
7c673cae FG |
15535 | } |
15536 | if ((need & CEPH_CAP_FILE_WR) && !(have & POOL_WRITE)) { | |
11fdf7f2 | 15537 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15538 | << " need " << ccap_string(need) << ", but no write perm" << dendl; |
f67539c2 | 15539 | return -CEPHFS_EPERM; |
7c673cae FG |
15540 | } |
15541 | ||
15542 | return 0; | |
15543 | } | |
15544 | ||
15545 | int Client::_posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want) | |
15546 | { | |
15547 | if (acl_type == POSIX_ACL) { | |
15548 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15549 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15550 | ||
15551 | return posix_acl_permits(access_acl, in->uid, in->gid, perms, want); | |
15552 | } | |
15553 | } | |
f67539c2 | 15554 | return -CEPHFS_EAGAIN; |
7c673cae FG |
15555 | } |
15556 | ||
15557 | int Client::_posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms) | |
15558 | { | |
15559 | if (acl_type == NO_ACL) | |
15560 | return 0; | |
15561 | ||
15562 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
15563 | if (r < 0) | |
15564 | goto out; | |
15565 | ||
15566 | if (acl_type == POSIX_ACL) { | |
15567 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15568 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15569 | bufferptr acl(access_acl.c_str(), access_acl.length()); | |
15570 | r = posix_acl_access_chmod(acl, mode); | |
15571 | if (r < 0) | |
15572 | goto out; | |
15573 | r = _do_setxattr(in, ACL_EA_ACCESS, acl.c_str(), acl.length(), 0, perms); | |
15574 | } else { | |
15575 | r = 0; | |
15576 | } | |
15577 | } | |
15578 | out: | |
15579 | ldout(cct, 10) << __func__ << " ino " << in->ino << " result=" << r << dendl; | |
15580 | return r; | |
15581 | } | |
15582 | ||
15583 | int Client::_posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl, | |
15584 | const UserPerm& perms) | |
15585 | { | |
15586 | if (acl_type == NO_ACL) | |
15587 | return 0; | |
15588 | ||
15589 | if (S_ISLNK(*mode)) | |
15590 | return 0; | |
15591 | ||
15592 | int r = _getattr(dir, CEPH_STAT_CAP_XATTR, perms, dir->xattr_version == 0); | |
15593 | if (r < 0) | |
15594 | goto out; | |
15595 | ||
15596 | if (acl_type == POSIX_ACL) { | |
15597 | if (dir->xattrs.count(ACL_EA_DEFAULT)) { | |
15598 | map<string, bufferptr> xattrs; | |
15599 | ||
15600 | const bufferptr& default_acl = dir->xattrs[ACL_EA_DEFAULT]; | |
15601 | bufferptr acl(default_acl.c_str(), default_acl.length()); | |
15602 | r = posix_acl_inherit_mode(acl, mode); | |
15603 | if (r < 0) | |
15604 | goto out; | |
15605 | ||
15606 | if (r > 0) { | |
15607 | r = posix_acl_equiv_mode(acl.c_str(), acl.length(), mode); | |
15608 | if (r < 0) | |
15609 | goto out; | |
15610 | if (r > 0) | |
15611 | xattrs[ACL_EA_ACCESS] = acl; | |
15612 | } | |
15613 | ||
15614 | if (S_ISDIR(*mode)) | |
15615 | xattrs[ACL_EA_DEFAULT] = dir->xattrs[ACL_EA_DEFAULT]; | |
15616 | ||
15617 | r = xattrs.size(); | |
15618 | if (r > 0) | |
11fdf7f2 | 15619 | encode(xattrs, xattrs_bl); |
7c673cae FG |
15620 | } else { |
15621 | if (umask_cb) | |
15622 | *mode &= ~umask_cb(callback_handle); | |
15623 | r = 0; | |
15624 | } | |
15625 | } | |
15626 | out: | |
15627 | ldout(cct, 10) << __func__ << " dir ino " << dir->ino << " result=" << r << dendl; | |
15628 | return r; | |
15629 | } | |
15630 | ||
15631 | void Client::set_filer_flags(int flags) | |
15632 | { | |
f67539c2 | 15633 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15634 | ceph_assert(flags == 0 || |
7c673cae FG |
15635 | flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
15636 | objecter->add_global_op_flags(flags); | |
15637 | } | |
15638 | ||
15639 | void Client::clear_filer_flags(int flags) | |
15640 | { | |
f67539c2 | 15641 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15642 | ceph_assert(flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
7c673cae FG |
15643 | objecter->clear_global_op_flag(flags); |
15644 | } | |
15645 | ||
11fdf7f2 TL |
15646 | // called before mount |
15647 | void Client::set_uuid(const std::string& uuid) | |
15648 | { | |
f67539c2 TL |
15649 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15650 | ceph_assert(iref_reader.is_state_satisfied()); | |
15651 | ||
15652 | std::scoped_lock l(client_lock); | |
20effc67 | 15653 | ceph_assert(!uuid.empty()); |
11fdf7f2 TL |
15654 | |
15655 | metadata["uuid"] = uuid; | |
15656 | _close_sessions(); | |
15657 | } | |
15658 | ||
15659 | // called before mount. 0 means infinite | |
15660 | void Client::set_session_timeout(unsigned timeout) | |
15661 | { | |
f67539c2 TL |
15662 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15663 | ceph_assert(iref_reader.is_state_satisfied()); | |
15664 | ||
15665 | std::scoped_lock l(client_lock); | |
11fdf7f2 TL |
15666 | |
15667 | metadata["timeout"] = stringify(timeout); | |
15668 | } | |
15669 | ||
15670 | // called before mount | |
15671 | int Client::start_reclaim(const std::string& uuid, unsigned flags, | |
15672 | const std::string& fs_name) | |
15673 | { | |
f67539c2 TL |
15674 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15675 | if (!iref_reader.is_state_satisfied()) | |
15676 | return -CEPHFS_ENOTCONN; | |
11fdf7f2 TL |
15677 | |
15678 | if (uuid.empty()) | |
f67539c2 | 15679 | return -CEPHFS_EINVAL; |
11fdf7f2 | 15680 | |
f67539c2 | 15681 | std::unique_lock l(client_lock); |
11fdf7f2 TL |
15682 | { |
15683 | auto it = metadata.find("uuid"); | |
15684 | if (it != metadata.end() && it->second == uuid) | |
f67539c2 | 15685 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15686 | } |
15687 | ||
15688 | int r = subscribe_mdsmap(fs_name); | |
15689 | if (r < 0) { | |
15690 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
15691 | return r; | |
15692 | } | |
15693 | ||
15694 | if (metadata.empty()) | |
15695 | populate_metadata(""); | |
15696 | ||
15697 | while (mdsmap->get_epoch() == 0) | |
15698 | wait_on_list(waiting_for_mdsmap); | |
15699 | ||
15700 | reclaim_errno = 0; | |
15701 | for (unsigned mds = 0; mds < mdsmap->get_num_in_mds(); ) { | |
15702 | if (!mdsmap->is_up(mds)) { | |
15703 | ldout(cct, 10) << "mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
15704 | wait_on_list(waiting_for_mdsmap); | |
15705 | continue; | |
15706 | } | |
15707 | ||
20effc67 | 15708 | MetaSessionRef session; |
11fdf7f2 TL |
15709 | if (!have_open_session(mds)) { |
15710 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 15711 | if (session->state == MetaSession::STATE_REJECTED) |
f67539c2 | 15712 | return -CEPHFS_EPERM; |
11fdf7f2 TL |
15713 | if (session->state != MetaSession::STATE_OPENING) { |
15714 | // umounting? | |
f67539c2 | 15715 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15716 | } |
15717 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
15718 | wait_on_context_list(session->waiting_for_open); | |
11fdf7f2 TL |
15719 | continue; |
15720 | } | |
15721 | ||
20effc67 | 15722 | session = mds_sessions.at(mds); |
11fdf7f2 | 15723 | if (!session->mds_features.test(CEPHFS_FEATURE_RECLAIM_CLIENT)) |
f67539c2 | 15724 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
15725 | |
15726 | if (session->reclaim_state == MetaSession::RECLAIM_NULL || | |
15727 | session->reclaim_state == MetaSession::RECLAIMING) { | |
15728 | session->reclaim_state = MetaSession::RECLAIMING; | |
9f95a23c | 15729 | auto m = make_message<MClientReclaim>(uuid, flags); |
11fdf7f2 TL |
15730 | session->con->send_message2(std::move(m)); |
15731 | wait_on_list(waiting_for_reclaim); | |
15732 | } else if (session->reclaim_state == MetaSession::RECLAIM_FAIL) { | |
f67539c2 | 15733 | return reclaim_errno ? : -CEPHFS_ENOTRECOVERABLE; |
11fdf7f2 TL |
15734 | } else { |
15735 | mds++; | |
15736 | } | |
15737 | } | |
15738 | ||
15739 | // didn't find target session in any mds | |
15740 | if (reclaim_target_addrs.empty()) { | |
15741 | if (flags & CEPH_RECLAIM_RESET) | |
f67539c2 TL |
15742 | return -CEPHFS_ENOENT; |
15743 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15744 | } |
15745 | ||
15746 | if (flags & CEPH_RECLAIM_RESET) | |
15747 | return 0; | |
15748 | ||
f67539c2 TL |
15749 | // use blocklist to check if target session was killed |
15750 | // (config option mds_session_blocklist_on_evict needs to be true) | |
15751 | ldout(cct, 10) << __func__ << ": waiting for OSD epoch " << reclaim_osd_epoch << dendl; | |
15752 | bs::error_code ec; | |
15753 | l.unlock(); | |
15754 | objecter->wait_for_map(reclaim_osd_epoch, ca::use_blocked[ec]); | |
15755 | l.lock(); | |
11fdf7f2 | 15756 | |
f67539c2 TL |
15757 | if (ec) |
15758 | return ceph::from_error_code(ec); | |
15759 | ||
15760 | bool blocklisted = objecter->with_osdmap( | |
11fdf7f2 | 15761 | [this](const OSDMap &osd_map) -> bool { |
f67539c2 | 15762 | return osd_map.is_blocklisted(reclaim_target_addrs); |
11fdf7f2 | 15763 | }); |
f67539c2 TL |
15764 | if (blocklisted) |
15765 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15766 | |
15767 | metadata["reclaiming_uuid"] = uuid; | |
15768 | return 0; | |
15769 | } | |
15770 | ||
15771 | void Client::finish_reclaim() | |
15772 | { | |
15773 | auto it = metadata.find("reclaiming_uuid"); | |
15774 | if (it == metadata.end()) { | |
15775 | for (auto &p : mds_sessions) | |
20effc67 | 15776 | p.second->reclaim_state = MetaSession::RECLAIM_NULL; |
11fdf7f2 TL |
15777 | return; |
15778 | } | |
15779 | ||
15780 | for (auto &p : mds_sessions) { | |
20effc67 | 15781 | p.second->reclaim_state = MetaSession::RECLAIM_NULL; |
9f95a23c | 15782 | auto m = make_message<MClientReclaim>("", MClientReclaim::FLAG_FINISH); |
20effc67 | 15783 | p.second->con->send_message2(std::move(m)); |
11fdf7f2 TL |
15784 | } |
15785 | ||
15786 | metadata["uuid"] = it->second; | |
15787 | metadata.erase(it); | |
15788 | } | |
15789 | ||
15790 | void Client::handle_client_reclaim_reply(const MConstRef<MClientReclaimReply>& reply) | |
15791 | { | |
15792 | mds_rank_t from = mds_rank_t(reply->get_source().num()); | |
15793 | ldout(cct, 10) << __func__ << " " << *reply << " from mds." << from << dendl; | |
15794 | ||
f67539c2 | 15795 | std::scoped_lock cl(client_lock); |
20effc67 | 15796 | auto session = _get_mds_session(from, reply->get_connection().get()); |
11fdf7f2 TL |
15797 | if (!session) { |
15798 | ldout(cct, 10) << " discarding reclaim reply from sessionless mds." << from << dendl; | |
15799 | return; | |
15800 | } | |
15801 | ||
15802 | if (reply->get_result() >= 0) { | |
15803 | session->reclaim_state = MetaSession::RECLAIM_OK; | |
15804 | if (reply->get_epoch() > reclaim_osd_epoch) | |
15805 | reclaim_osd_epoch = reply->get_epoch(); | |
15806 | if (!reply->get_addrs().empty()) | |
15807 | reclaim_target_addrs = reply->get_addrs(); | |
15808 | } else { | |
15809 | session->reclaim_state = MetaSession::RECLAIM_FAIL; | |
15810 | reclaim_errno = reply->get_result(); | |
15811 | } | |
15812 | ||
15813 | signal_cond_list(waiting_for_reclaim); | |
15814 | } | |
15815 | ||
7c673cae FG |
15816 | /** |
15817 | * This is included in cap release messages, to cause | |
15818 | * the MDS to wait until this OSD map epoch. It is necessary | |
15819 | * in corner cases where we cancel RADOS ops, so that | |
15820 | * nobody else tries to do IO to the same objects in | |
15821 | * the same epoch as the cancelled ops. | |
15822 | */ | |
15823 | void Client::set_cap_epoch_barrier(epoch_t e) | |
15824 | { | |
15825 | ldout(cct, 5) << __func__ << " epoch = " << e << dendl; | |
15826 | cap_epoch_barrier = e; | |
15827 | } | |
15828 | ||
15829 | const char** Client::get_tracked_conf_keys() const | |
15830 | { | |
15831 | static const char* keys[] = { | |
15832 | "client_cache_size", | |
15833 | "client_cache_mid", | |
15834 | "client_acl_type", | |
b32b8144 FG |
15835 | "client_deleg_timeout", |
15836 | "client_deleg_break_on_open", | |
f67539c2 TL |
15837 | "client_oc_size", |
15838 | "client_oc_max_objects", | |
15839 | "client_oc_max_dirty", | |
15840 | "client_oc_target_dirty", | |
15841 | "client_oc_max_dirty_age", | |
2a845540 TL |
15842 | "client_caps_release_delay", |
15843 | "client_mount_timeout", | |
7c673cae FG |
15844 | NULL |
15845 | }; | |
15846 | return keys; | |
15847 | } | |
15848 | ||
11fdf7f2 | 15849 | void Client::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
15850 | const std::set <std::string> &changed) |
15851 | { | |
f67539c2 | 15852 | std::scoped_lock lock(client_lock); |
7c673cae | 15853 | |
181888fb | 15854 | if (changed.count("client_cache_mid")) { |
7c673cae FG |
15855 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
15856 | } | |
15857 | if (changed.count("client_acl_type")) { | |
15858 | acl_type = NO_ACL; | |
15859 | if (cct->_conf->client_acl_type == "posix_acl") | |
15860 | acl_type = POSIX_ACL; | |
15861 | } | |
f67539c2 TL |
15862 | if (changed.count("client_oc_size")) { |
15863 | objectcacher->set_max_size(cct->_conf->client_oc_size); | |
15864 | } | |
15865 | if (changed.count("client_oc_max_objects")) { | |
15866 | objectcacher->set_max_objects(cct->_conf->client_oc_max_objects); | |
15867 | } | |
15868 | if (changed.count("client_oc_max_dirty")) { | |
15869 | objectcacher->set_max_dirty(cct->_conf->client_oc_max_dirty); | |
15870 | } | |
15871 | if (changed.count("client_oc_target_dirty")) { | |
15872 | objectcacher->set_target_dirty(cct->_conf->client_oc_target_dirty); | |
15873 | } | |
15874 | if (changed.count("client_oc_max_dirty_age")) { | |
15875 | objectcacher->set_max_dirty_age(cct->_conf->client_oc_max_dirty_age); | |
15876 | } | |
33c7a0ef TL |
15877 | if (changed.count("client_collect_and_send_global_metrics")) { |
15878 | _collect_and_send_global_metrics = cct->_conf.get_val<bool>( | |
15879 | "client_collect_and_send_global_metrics"); | |
15880 | } | |
2a845540 TL |
15881 | if (changed.count("client_caps_release_delay")) { |
15882 | caps_release_delay = cct->_conf.get_val<std::chrono::seconds>( | |
15883 | "client_caps_release_delay"); | |
15884 | } | |
15885 | if (changed.count("client_mount_timeout")) { | |
15886 | mount_timeout = cct->_conf.get_val<std::chrono::seconds>( | |
15887 | "client_mount_timeout"); | |
15888 | } | |
7c673cae FG |
15889 | } |
15890 | ||
7c673cae FG |
15891 | void intrusive_ptr_add_ref(Inode *in) |
15892 | { | |
b3b6e05e | 15893 | in->iget(); |
7c673cae | 15894 | } |
f67539c2 | 15895 | |
7c673cae FG |
15896 | void intrusive_ptr_release(Inode *in) |
15897 | { | |
15898 | in->client->put_inode(in); | |
15899 | } | |
15900 | ||
15901 | mds_rank_t Client::_get_random_up_mds() const | |
15902 | { | |
9f95a23c | 15903 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
15904 | |
15905 | std::set<mds_rank_t> up; | |
15906 | mdsmap->get_up_mds_set(up); | |
15907 | ||
15908 | if (up.empty()) | |
15909 | return MDS_RANK_NONE; | |
15910 | std::set<mds_rank_t>::const_iterator p = up.begin(); | |
15911 | for (int n = rand() % up.size(); n; n--) | |
15912 | ++p; | |
15913 | return *p; | |
15914 | } | |
15915 | ||
15916 | ||
f67539c2 TL |
15917 | StandaloneClient::StandaloneClient(Messenger *m, MonClient *mc, |
15918 | boost::asio::io_context& ictx) | |
15919 | : Client(m, mc, new Objecter(m->cct, m, mc, ictx)) | |
7c673cae FG |
15920 | { |
15921 | monclient->set_messenger(m); | |
15922 | objecter->set_client_incarnation(0); | |
15923 | } | |
15924 | ||
15925 | StandaloneClient::~StandaloneClient() | |
15926 | { | |
15927 | delete objecter; | |
15928 | objecter = nullptr; | |
15929 | } | |
15930 | ||
15931 | int StandaloneClient::init() | |
15932 | { | |
f67539c2 TL |
15933 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
15934 | ceph_assert(iref_writer.is_first_writer()); | |
15935 | ||
e306af50 | 15936 | _pre_init(); |
7c673cae FG |
15937 | objecter->init(); |
15938 | ||
9f95a23c | 15939 | client_lock.lock(); |
7c673cae FG |
15940 | |
15941 | messenger->add_dispatcher_tail(objecter); | |
15942 | messenger->add_dispatcher_tail(this); | |
15943 | ||
15944 | monclient->set_want_keys(CEPH_ENTITY_TYPE_MDS | CEPH_ENTITY_TYPE_OSD); | |
15945 | int r = monclient->init(); | |
15946 | if (r < 0) { | |
15947 | // need to do cleanup because we're in an intermediate init state | |
f67539c2 TL |
15948 | { |
15949 | std::scoped_lock l(timer_lock); | |
15950 | timer.shutdown(); | |
15951 | } | |
15952 | ||
9f95a23c | 15953 | client_lock.unlock(); |
7c673cae FG |
15954 | objecter->shutdown(); |
15955 | objectcacher->stop(); | |
15956 | monclient->shutdown(); | |
15957 | return r; | |
15958 | } | |
15959 | objecter->start(); | |
15960 | ||
9f95a23c | 15961 | client_lock.unlock(); |
7c673cae | 15962 | _finish_init(); |
f67539c2 | 15963 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
15964 | |
15965 | return 0; | |
15966 | } | |
15967 | ||
15968 | void StandaloneClient::shutdown() | |
15969 | { | |
15970 | Client::shutdown(); | |
15971 | objecter->shutdown(); | |
15972 | monclient->shutdown(); | |
15973 | } |