]>
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 | ||
adb31ebb TL |
152 | using namespace TOPNSPC::common; |
153 | ||
f67539c2 TL |
154 | namespace bs = boost::system; |
155 | namespace ca = ceph::async; | |
156 | ||
7c673cae FG |
157 | void client_flush_set_callback(void *p, ObjectCacher::ObjectSet *oset) |
158 | { | |
159 | Client *client = static_cast<Client*>(p); | |
160 | client->flush_set_callback(oset); | |
161 | } | |
162 | ||
b3b6e05e TL |
163 | bool Client::is_reserved_vino(vinodeno_t &vino) { |
164 | if (MDS_IS_PRIVATE_INO(vino.ino)) { | |
165 | ldout(cct, -1) << __func__ << " attempt to access reserved inode number " << vino << dendl; | |
166 | return true; | |
167 | } | |
168 | return false; | |
169 | } | |
170 | ||
7c673cae FG |
171 | |
172 | // ------------- | |
173 | ||
174 | Client::CommandHook::CommandHook(Client *client) : | |
175 | m_client(client) | |
176 | { | |
177 | } | |
178 | ||
9f95a23c TL |
179 | int Client::CommandHook::call( |
180 | std::string_view command, | |
181 | const cmdmap_t& cmdmap, | |
182 | Formatter *f, | |
183 | std::ostream& errss, | |
184 | bufferlist& out) | |
7c673cae | 185 | { |
7c673cae | 186 | f->open_object_section("result"); |
9f95a23c | 187 | { |
f67539c2 | 188 | std::scoped_lock l{m_client->client_lock}; |
9f95a23c TL |
189 | if (command == "mds_requests") |
190 | m_client->dump_mds_requests(f); | |
adb31ebb TL |
191 | else if (command == "mds_sessions") { |
192 | bool cap_dump = false; | |
193 | cmd_getval(cmdmap, "cap_dump", cap_dump); | |
194 | m_client->dump_mds_sessions(f, cap_dump); | |
195 | } else if (command == "dump_cache") | |
9f95a23c TL |
196 | m_client->dump_cache(f); |
197 | else if (command == "kick_stale_sessions") | |
198 | m_client->_kick_stale_sessions(); | |
199 | else if (command == "status") | |
200 | m_client->dump_status(f); | |
201 | else | |
202 | ceph_abort_msg("bad command registered"); | |
203 | } | |
7c673cae | 204 | f->close_section(); |
9f95a23c | 205 | return 0; |
7c673cae FG |
206 | } |
207 | ||
208 | ||
209 | // ------------- | |
210 | ||
b3b6e05e TL |
211 | int Client::get_fd_inode(int fd, InodeRef *in) { |
212 | int r = 0; | |
213 | if (fd == CEPHFS_AT_FDCWD) { | |
214 | *in = cwd; | |
215 | } else { | |
216 | Fh *f = get_filehandle(fd); | |
217 | if (!f) { | |
218 | r = -CEPHFS_EBADF; | |
219 | } else { | |
220 | *in = f->inode; | |
221 | } | |
222 | } | |
223 | return r; | |
224 | } | |
225 | ||
7c673cae FG |
226 | dir_result_t::dir_result_t(Inode *in, const UserPerm& perms) |
227 | : inode(in), offset(0), next_offset(2), | |
228 | release_count(0), ordered_count(0), cache_index(0), start_shared_gen(0), | |
229 | perms(perms) | |
230 | { } | |
231 | ||
232 | void Client::_reset_faked_inos() | |
233 | { | |
234 | ino_t start = 1024; | |
235 | free_faked_inos.clear(); | |
236 | free_faked_inos.insert(start, (uint32_t)-1 - start + 1); | |
237 | last_used_faked_ino = 0; | |
11fdf7f2 | 238 | last_used_faked_root = 0; |
f67539c2 TL |
239 | #ifdef _WIN32 |
240 | // On Windows, sizeof(ino_t) is just 2. Despite that, most "native" | |
241 | // Windows structures, including Dokan ones, are using 64B identifiers. | |
242 | _use_faked_inos = false; | |
243 | #else | |
7c673cae | 244 | _use_faked_inos = sizeof(ino_t) < 8 || cct->_conf->client_use_faked_inos; |
f67539c2 | 245 | #endif |
7c673cae FG |
246 | } |
247 | ||
248 | void Client::_assign_faked_ino(Inode *in) | |
249 | { | |
11fdf7f2 TL |
250 | if (0 == last_used_faked_ino) |
251 | last_used_faked_ino = last_used_faked_ino + 2048; // start(1024)~2048 reserved for _assign_faked_root | |
7c673cae FG |
252 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
253 | if (it == free_faked_inos.end() && last_used_faked_ino > 0) { | |
11fdf7f2 | 254 | last_used_faked_ino = 2048; |
7c673cae FG |
255 | it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
256 | } | |
11fdf7f2 | 257 | ceph_assert(it != free_faked_inos.end()); |
7c673cae | 258 | if (last_used_faked_ino < it.get_start()) { |
11fdf7f2 | 259 | ceph_assert(it.get_len() > 0); |
7c673cae FG |
260 | last_used_faked_ino = it.get_start(); |
261 | } else { | |
262 | ++last_used_faked_ino; | |
11fdf7f2 | 263 | ceph_assert(it.get_start() + it.get_len() > last_used_faked_ino); |
7c673cae FG |
264 | } |
265 | in->faked_ino = last_used_faked_ino; | |
266 | free_faked_inos.erase(in->faked_ino); | |
267 | faked_ino_map[in->faked_ino] = in->vino(); | |
268 | } | |
269 | ||
11fdf7f2 TL |
270 | /* |
271 | * In the faked mode, if you export multiple subdirectories, | |
272 | * you will see that the inode numbers of the exported subdirectories | |
273 | * are the same. so we distinguish the mount point by reserving | |
274 | * the "fake ids" between "1024~2048" and combining the last | |
275 | * 10bits(0x3ff) of the "root inodes". | |
276 | */ | |
277 | void Client::_assign_faked_root(Inode *in) | |
278 | { | |
279 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
280 | if (it == free_faked_inos.end() && last_used_faked_root > 0) { | |
281 | last_used_faked_root = 0; | |
282 | it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
283 | } | |
284 | assert(it != free_faked_inos.end()); | |
285 | vinodeno_t inode_info = in->vino(); | |
286 | uint64_t inode_num = (uint64_t)inode_info.ino; | |
287 | ldout(cct, 10) << "inode_num " << inode_num << "inode_num & 0x3ff=" << (inode_num & 0x3ff)<< dendl; | |
288 | last_used_faked_root = it.get_start() + (inode_num & 0x3ff); // 0x3ff mask and get_start will not exceed 2048 | |
289 | assert(it.get_start() + it.get_len() > last_used_faked_root); | |
290 | ||
291 | in->faked_ino = last_used_faked_root; | |
292 | free_faked_inos.erase(in->faked_ino); | |
293 | faked_ino_map[in->faked_ino] = in->vino(); | |
294 | } | |
295 | ||
7c673cae FG |
296 | void Client::_release_faked_ino(Inode *in) |
297 | { | |
298 | free_faked_inos.insert(in->faked_ino); | |
299 | faked_ino_map.erase(in->faked_ino); | |
300 | } | |
301 | ||
302 | vinodeno_t Client::_map_faked_ino(ino_t ino) | |
303 | { | |
304 | vinodeno_t vino; | |
305 | if (ino == 1) | |
306 | vino = root->vino(); | |
307 | else if (faked_ino_map.count(ino)) | |
308 | vino = faked_ino_map[ino]; | |
309 | else | |
310 | vino = vinodeno_t(0, CEPH_NOSNAP); | |
11fdf7f2 | 311 | ldout(cct, 10) << __func__ << " " << ino << " -> " << vino << dendl; |
7c673cae FG |
312 | return vino; |
313 | } | |
314 | ||
315 | vinodeno_t Client::map_faked_ino(ino_t ino) | |
316 | { | |
f67539c2 | 317 | std::scoped_lock lock(client_lock); |
7c673cae FG |
318 | return _map_faked_ino(ino); |
319 | } | |
320 | ||
321 | // cons/des | |
322 | ||
323 | Client::Client(Messenger *m, MonClient *mc, Objecter *objecter_) | |
f67539c2 TL |
324 | : Dispatcher(m->cct->get()), |
325 | timer(m->cct, timer_lock, false), | |
11fdf7f2 TL |
326 | messenger(m), |
327 | monclient(mc), | |
328 | objecter(objecter_), | |
329 | whoami(mc->get_global_id()), | |
f67539c2 TL |
330 | mount_state(CLIENT_UNMOUNTED, "Client::mountstate_lock"), |
331 | initialize_state(CLIENT_NEW, "Client::initstate_lock"), | |
332 | cct_deleter{m->cct, [](CephContext *p) {p->put();}}, | |
7c673cae FG |
333 | async_ino_invalidator(m->cct), |
334 | async_dentry_invalidator(m->cct), | |
335 | interrupt_finisher(m->cct), | |
336 | remount_finisher(m->cct), | |
e306af50 | 337 | async_ino_releasor(m->cct), |
7c673cae | 338 | objecter_finisher(m->cct), |
11fdf7f2 TL |
339 | m_command_hook(this), |
340 | fscid(0) | |
7c673cae FG |
341 | { |
342 | _reset_faked_inos(); | |
7c673cae | 343 | |
7c673cae FG |
344 | user_id = cct->_conf->client_mount_uid; |
345 | group_id = cct->_conf->client_mount_gid; | |
92f5a8d4 TL |
346 | fuse_default_permissions = cct->_conf.get_val<bool>( |
347 | "fuse_default_permissions"); | |
7c673cae | 348 | |
7c673cae FG |
349 | if (cct->_conf->client_acl_type == "posix_acl") |
350 | acl_type = POSIX_ACL; | |
351 | ||
7c673cae FG |
352 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
353 | ||
354 | // file handles | |
355 | free_fd_set.insert(10, 1<<30); | |
356 | ||
357 | mdsmap.reset(new MDSMap); | |
358 | ||
359 | // osd interfaces | |
360 | writeback_handler.reset(new ObjecterWriteback(objecter, &objecter_finisher, | |
361 | &client_lock)); | |
362 | objectcacher.reset(new ObjectCacher(cct, "libcephfs", *writeback_handler, client_lock, | |
363 | client_flush_set_callback, // all commit callback | |
364 | (void*)this, | |
365 | cct->_conf->client_oc_size, | |
366 | cct->_conf->client_oc_max_objects, | |
367 | cct->_conf->client_oc_max_dirty, | |
368 | cct->_conf->client_oc_target_dirty, | |
369 | cct->_conf->client_oc_max_dirty_age, | |
370 | true)); | |
7c673cae FG |
371 | } |
372 | ||
373 | ||
374 | Client::~Client() | |
375 | { | |
9f95a23c | 376 | ceph_assert(ceph_mutex_is_not_locked(client_lock)); |
7c673cae | 377 | |
f67539c2 TL |
378 | // If the task is crashed or aborted and doesn't |
379 | // get any chance to run the umount and shutdow. | |
380 | { | |
381 | std::scoped_lock l{client_lock}; | |
382 | tick_thread_stopped = true; | |
383 | upkeep_cond.notify_one(); | |
384 | } | |
385 | ||
386 | if (upkeeper.joinable()) | |
387 | upkeeper.join(); | |
388 | ||
31f18b77 FG |
389 | // It is necessary to hold client_lock, because any inode destruction |
390 | // may call into ObjectCacher, which asserts that it's lock (which is | |
391 | // client_lock) is held. | |
f67539c2 | 392 | std::scoped_lock l{client_lock}; |
7c673cae FG |
393 | tear_down_cache(); |
394 | } | |
395 | ||
396 | void Client::tear_down_cache() | |
397 | { | |
398 | // fd's | |
f67539c2 TL |
399 | for (auto &[fd, fh] : fd_map) { |
400 | ldout(cct, 1) << __func__ << " forcing close of fh " << fd << " ino " << fh->inode->ino << dendl; | |
7c673cae FG |
401 | _release_fh(fh); |
402 | } | |
403 | fd_map.clear(); | |
404 | ||
405 | while (!opened_dirs.empty()) { | |
406 | dir_result_t *dirp = *opened_dirs.begin(); | |
11fdf7f2 | 407 | ldout(cct, 1) << __func__ << " forcing close of dir " << dirp << " ino " << dirp->inode->ino << dendl; |
7c673cae FG |
408 | _closedir(dirp); |
409 | } | |
410 | ||
411 | // caps! | |
412 | // *** FIXME *** | |
413 | ||
414 | // empty lru | |
7c673cae | 415 | trim_cache(); |
11fdf7f2 | 416 | ceph_assert(lru.lru_get_size() == 0); |
7c673cae FG |
417 | |
418 | // close root ino | |
11fdf7f2 | 419 | ceph_assert(inode_map.size() <= 1 + root_parents.size()); |
7c673cae | 420 | if (root && inode_map.size() == 1 + root_parents.size()) { |
b3b6e05e | 421 | root.reset(); |
7c673cae FG |
422 | } |
423 | ||
11fdf7f2 | 424 | ceph_assert(inode_map.empty()); |
7c673cae FG |
425 | } |
426 | ||
427 | inodeno_t Client::get_root_ino() | |
428 | { | |
f67539c2 | 429 | std::scoped_lock l(client_lock); |
7c673cae FG |
430 | if (use_faked_inos()) |
431 | return root->faked_ino; | |
432 | else | |
433 | return root->ino; | |
434 | } | |
435 | ||
436 | Inode *Client::get_root() | |
437 | { | |
f67539c2 | 438 | std::scoped_lock l(client_lock); |
7c673cae | 439 | root->ll_get(); |
b3b6e05e | 440 | return root.get(); |
7c673cae FG |
441 | } |
442 | ||
443 | ||
444 | // debug crapola | |
445 | ||
446 | void Client::dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected) | |
447 | { | |
448 | filepath path; | |
449 | in->make_long_path(path); | |
450 | ldout(cct, 1) << "dump_inode: " | |
451 | << (disconnected ? "DISCONNECTED ":"") | |
452 | << "inode " << in->ino | |
453 | << " " << path | |
b3b6e05e | 454 | << " ref " << in->get_nref() |
f67539c2 | 455 | << " " << *in << dendl; |
7c673cae FG |
456 | |
457 | if (f) { | |
458 | f->open_object_section("inode"); | |
459 | f->dump_stream("path") << path; | |
460 | if (disconnected) | |
461 | f->dump_int("disconnected", 1); | |
462 | in->dump(f); | |
463 | f->close_section(); | |
464 | } | |
465 | ||
466 | did.insert(in); | |
467 | if (in->dir) { | |
468 | ldout(cct, 1) << " dir " << in->dir << " size " << in->dir->dentries.size() << dendl; | |
469 | for (ceph::unordered_map<string, Dentry*>::iterator it = in->dir->dentries.begin(); | |
470 | it != in->dir->dentries.end(); | |
471 | ++it) { | |
472 | ldout(cct, 1) << " " << in->ino << " dn " << it->first << " " << it->second << " ref " << it->second->ref << dendl; | |
473 | if (f) { | |
474 | f->open_object_section("dentry"); | |
475 | it->second->dump(f); | |
476 | f->close_section(); | |
477 | } | |
478 | if (it->second->inode) | |
479 | dump_inode(f, it->second->inode.get(), did, false); | |
480 | } | |
481 | } | |
482 | } | |
483 | ||
484 | void Client::dump_cache(Formatter *f) | |
485 | { | |
486 | set<Inode*> did; | |
487 | ||
11fdf7f2 | 488 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
489 | |
490 | if (f) | |
491 | f->open_array_section("cache"); | |
492 | ||
493 | if (root) | |
b3b6e05e | 494 | dump_inode(f, root.get(), did, true); |
7c673cae FG |
495 | |
496 | // make a second pass to catch anything disconnected | |
497 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
498 | it != inode_map.end(); | |
499 | ++it) { | |
500 | if (did.count(it->second)) | |
501 | continue; | |
502 | dump_inode(f, it->second, did, true); | |
503 | } | |
504 | ||
505 | if (f) | |
506 | f->close_section(); | |
507 | } | |
508 | ||
509 | void Client::dump_status(Formatter *f) | |
510 | { | |
9f95a23c | 511 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
512 | |
513 | ldout(cct, 1) << __func__ << dendl; | |
514 | ||
515 | const epoch_t osd_epoch | |
516 | = objecter->with_osdmap(std::mem_fn(&OSDMap::get_epoch)); | |
517 | ||
518 | if (f) { | |
519 | f->open_object_section("metadata"); | |
520 | for (const auto& kv : metadata) | |
521 | f->dump_string(kv.first.c_str(), kv.second); | |
522 | f->close_section(); | |
523 | ||
524 | f->dump_int("dentry_count", lru.lru_get_size()); | |
525 | f->dump_int("dentry_pinned_count", lru.lru_get_num_pinned()); | |
526 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 527 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 | 528 | f->dump_object("inst", inst); |
11fdf7f2 TL |
529 | f->dump_object("addr", inst.addr); |
530 | f->dump_stream("inst_str") << inst.name << " " << inst.addr.get_legacy_str(); | |
531 | f->dump_string("addr_str", inst.addr.get_legacy_str()); | |
7c673cae FG |
532 | f->dump_int("inode_count", inode_map.size()); |
533 | f->dump_int("mds_epoch", mdsmap->get_epoch()); | |
534 | f->dump_int("osd_epoch", osd_epoch); | |
535 | f->dump_int("osd_epoch_barrier", cap_epoch_barrier); | |
f67539c2 | 536 | f->dump_bool("blocklisted", blocklisted); |
adb31ebb | 537 | f->dump_string("fs_name", mdsmap->get_fs_name()); |
7c673cae FG |
538 | } |
539 | } | |
540 | ||
e306af50 | 541 | void Client::_pre_init() |
7c673cae FG |
542 | { |
543 | timer.init(); | |
e306af50 TL |
544 | |
545 | objecter_finisher.start(); | |
546 | filer.reset(new Filer(objecter, &objecter_finisher)); | |
f67539c2 | 547 | objecter->enable_blocklist_events(); |
e306af50 | 548 | |
7c673cae | 549 | objectcacher->start(); |
e306af50 TL |
550 | } |
551 | ||
552 | int Client::init() | |
553 | { | |
f67539c2 TL |
554 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
555 | ceph_assert(iref_writer.is_first_writer()); | |
556 | ||
e306af50 | 557 | _pre_init(); |
9f95a23c | 558 | { |
f67539c2 | 559 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
560 | messenger->add_dispatcher_tail(this); |
561 | } | |
7c673cae | 562 | _finish_init(); |
f67539c2 | 563 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
564 | return 0; |
565 | } | |
566 | ||
567 | void Client::_finish_init() | |
568 | { | |
9f95a23c | 569 | { |
f67539c2 | 570 | std::scoped_lock l{client_lock}; |
9f95a23c TL |
571 | // logger |
572 | PerfCountersBuilder plb(cct, "client", l_c_first, l_c_last); | |
573 | plb.add_time_avg(l_c_reply, "reply", "Latency of receiving a reply on metadata request"); | |
574 | plb.add_time_avg(l_c_lat, "lat", "Latency of processing a metadata request"); | |
575 | plb.add_time_avg(l_c_wrlat, "wrlat", "Latency of a file data write operation"); | |
576 | plb.add_time_avg(l_c_read, "rdlat", "Latency of a file data read operation"); | |
577 | plb.add_time_avg(l_c_fsync, "fsync", "Latency of a file sync operation"); | |
578 | logger.reset(plb.create_perf_counters()); | |
579 | cct->get_perfcounters_collection()->add(logger.get()); | |
580 | } | |
7c673cae | 581 | |
11fdf7f2 | 582 | cct->_conf.add_observer(this); |
7c673cae FG |
583 | |
584 | AdminSocket* admin_socket = cct->get_admin_socket(); | |
585 | int ret = admin_socket->register_command("mds_requests", | |
7c673cae FG |
586 | &m_command_hook, |
587 | "show in-progress mds requests"); | |
588 | if (ret < 0) { | |
589 | lderr(cct) << "error registering admin socket command: " | |
590 | << cpp_strerror(-ret) << dendl; | |
591 | } | |
adb31ebb TL |
592 | ret = admin_socket->register_command("mds_sessions " |
593 | "name=cap_dump,type=CephBool,req=false", | |
7c673cae FG |
594 | &m_command_hook, |
595 | "show mds session state"); | |
596 | if (ret < 0) { | |
597 | lderr(cct) << "error registering admin socket command: " | |
598 | << cpp_strerror(-ret) << dendl; | |
599 | } | |
600 | ret = admin_socket->register_command("dump_cache", | |
7c673cae FG |
601 | &m_command_hook, |
602 | "show in-memory metadata cache contents"); | |
603 | if (ret < 0) { | |
604 | lderr(cct) << "error registering admin socket command: " | |
605 | << cpp_strerror(-ret) << dendl; | |
606 | } | |
607 | ret = admin_socket->register_command("kick_stale_sessions", | |
7c673cae FG |
608 | &m_command_hook, |
609 | "kick sessions that were remote reset"); | |
610 | if (ret < 0) { | |
611 | lderr(cct) << "error registering admin socket command: " | |
612 | << cpp_strerror(-ret) << dendl; | |
613 | } | |
614 | ret = admin_socket->register_command("status", | |
7c673cae FG |
615 | &m_command_hook, |
616 | "show overall client status"); | |
617 | if (ret < 0) { | |
618 | lderr(cct) << "error registering admin socket command: " | |
619 | << cpp_strerror(-ret) << dendl; | |
620 | } | |
7c673cae FG |
621 | } |
622 | ||
623 | void Client::shutdown() | |
624 | { | |
11fdf7f2 | 625 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
626 | |
627 | // If we were not mounted, but were being used for sending | |
628 | // MDS commands, we may have sessions that need closing. | |
9f95a23c | 629 | { |
f67539c2 TL |
630 | std::scoped_lock l{client_lock}; |
631 | ||
632 | // To make sure the tick thread will be stoppped before | |
633 | // destructing the Client, just in case like the _mount() | |
634 | // failed but didn't not get a chance to stop the tick | |
635 | // thread | |
636 | tick_thread_stopped = true; | |
637 | upkeep_cond.notify_one(); | |
638 | ||
9f95a23c TL |
639 | _close_sessions(); |
640 | } | |
11fdf7f2 | 641 | cct->_conf.remove_observer(this); |
7c673cae | 642 | |
11fdf7f2 | 643 | cct->get_admin_socket()->unregister_commands(&m_command_hook); |
7c673cae FG |
644 | |
645 | if (ino_invalidate_cb) { | |
646 | ldout(cct, 10) << "shutdown stopping cache invalidator finisher" << dendl; | |
647 | async_ino_invalidator.wait_for_empty(); | |
648 | async_ino_invalidator.stop(); | |
649 | } | |
650 | ||
651 | if (dentry_invalidate_cb) { | |
652 | ldout(cct, 10) << "shutdown stopping dentry invalidator finisher" << dendl; | |
653 | async_dentry_invalidator.wait_for_empty(); | |
654 | async_dentry_invalidator.stop(); | |
655 | } | |
656 | ||
657 | if (switch_interrupt_cb) { | |
658 | ldout(cct, 10) << "shutdown stopping interrupt finisher" << dendl; | |
659 | interrupt_finisher.wait_for_empty(); | |
660 | interrupt_finisher.stop(); | |
661 | } | |
662 | ||
663 | if (remount_cb) { | |
664 | ldout(cct, 10) << "shutdown stopping remount finisher" << dendl; | |
665 | remount_finisher.wait_for_empty(); | |
666 | remount_finisher.stop(); | |
667 | } | |
668 | ||
e306af50 TL |
669 | if (ino_release_cb) { |
670 | ldout(cct, 10) << "shutdown stopping inode release finisher" << dendl; | |
671 | async_ino_releasor.wait_for_empty(); | |
672 | async_ino_releasor.stop(); | |
673 | } | |
674 | ||
7c673cae | 675 | objectcacher->stop(); // outside of client_lock! this does a join. |
f67539c2 TL |
676 | |
677 | /* | |
678 | * We are shuting down the client. | |
679 | * | |
680 | * Just declare the state to CLIENT_NEW to block and fail any | |
681 | * new comming "reader" and then try to wait all the in-flight | |
682 | * "readers" to finish. | |
683 | */ | |
684 | RWRef_t iref_writer(initialize_state, CLIENT_NEW, false); | |
685 | if (!iref_writer.is_first_writer()) | |
686 | return; | |
687 | iref_writer.wait_readers_done(); | |
688 | ||
9f95a23c | 689 | { |
f67539c2 | 690 | std::scoped_lock l(timer_lock); |
9f95a23c TL |
691 | timer.shutdown(); |
692 | } | |
f67539c2 | 693 | |
7c673cae FG |
694 | objecter_finisher.wait_for_empty(); |
695 | objecter_finisher.stop(); | |
696 | ||
697 | if (logger) { | |
698 | cct->get_perfcounters_collection()->remove(logger.get()); | |
699 | logger.reset(); | |
700 | } | |
701 | } | |
702 | ||
703 | ||
704 | // =================== | |
705 | // metadata cache stuff | |
706 | ||
707 | void Client::trim_cache(bool trim_kernel_dcache) | |
708 | { | |
181888fb FG |
709 | uint64_t max = cct->_conf->client_cache_size; |
710 | ldout(cct, 20) << "trim_cache size " << lru.lru_get_size() << " max " << max << dendl; | |
7c673cae FG |
711 | unsigned last = 0; |
712 | while (lru.lru_get_size() != last) { | |
713 | last = lru.lru_get_size(); | |
714 | ||
f67539c2 | 715 | if (!is_unmounting() && lru.lru_get_size() <= max) break; |
7c673cae FG |
716 | |
717 | // trim! | |
31f18b77 | 718 | Dentry *dn = static_cast<Dentry*>(lru.lru_get_next_expire()); |
7c673cae FG |
719 | if (!dn) |
720 | break; // done | |
f67539c2 | 721 | |
7c673cae FG |
722 | trim_dentry(dn); |
723 | } | |
724 | ||
181888fb | 725 | if (trim_kernel_dcache && lru.lru_get_size() > max) |
7c673cae FG |
726 | _invalidate_kernel_dcache(); |
727 | ||
728 | // hose root? | |
b3b6e05e | 729 | if (lru.lru_get_size() == 0 && root && root->get_nref() == 1 && inode_map.size() == 1 + root_parents.size()) { |
7c673cae | 730 | ldout(cct, 15) << "trim_cache trimmed root " << root << dendl; |
b3b6e05e | 731 | root.reset(); |
7c673cae FG |
732 | } |
733 | } | |
734 | ||
735 | void Client::trim_cache_for_reconnect(MetaSession *s) | |
736 | { | |
737 | mds_rank_t mds = s->mds_num; | |
11fdf7f2 | 738 | ldout(cct, 20) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
739 | |
740 | int trimmed = 0; | |
741 | list<Dentry*> skipped; | |
742 | while (lru.lru_get_size() > 0) { | |
743 | Dentry *dn = static_cast<Dentry*>(lru.lru_expire()); | |
744 | if (!dn) | |
745 | break; | |
746 | ||
747 | if ((dn->inode && dn->inode->caps.count(mds)) || | |
748 | dn->dir->parent_inode->caps.count(mds)) { | |
749 | trim_dentry(dn); | |
750 | trimmed++; | |
751 | } else | |
752 | skipped.push_back(dn); | |
753 | } | |
754 | ||
755 | for(list<Dentry*>::iterator p = skipped.begin(); p != skipped.end(); ++p) | |
756 | lru.lru_insert_mid(*p); | |
757 | ||
11fdf7f2 | 758 | ldout(cct, 20) << __func__ << " mds." << mds |
7c673cae FG |
759 | << " trimmed " << trimmed << " dentries" << dendl; |
760 | ||
761 | if (s->caps.size() > 0) | |
762 | _invalidate_kernel_dcache(); | |
763 | } | |
764 | ||
765 | void Client::trim_dentry(Dentry *dn) | |
766 | { | |
767 | ldout(cct, 15) << "trim_dentry unlinking dn " << dn->name | |
11fdf7f2 TL |
768 | << " in dir " |
769 | << std::hex << dn->dir->parent_inode->ino << std::dec | |
7c673cae FG |
770 | << dendl; |
771 | if (dn->inode) { | |
772 | Inode *diri = dn->dir->parent_inode; | |
7c673cae FG |
773 | clear_dir_complete_and_ordered(diri, true); |
774 | } | |
775 | unlink(dn, false, false); // drop dir, drop dentry | |
776 | } | |
777 | ||
778 | ||
1adf2230 AA |
779 | void Client::update_inode_file_size(Inode *in, int issued, uint64_t size, |
780 | uint64_t truncate_seq, uint64_t truncate_size) | |
7c673cae | 781 | { |
7c673cae FG |
782 | uint64_t prior_size = in->size; |
783 | ||
7c673cae FG |
784 | if (truncate_seq > in->truncate_seq || |
785 | (truncate_seq == in->truncate_seq && size > in->size)) { | |
786 | ldout(cct, 10) << "size " << in->size << " -> " << size << dendl; | |
787 | in->size = size; | |
788 | in->reported_size = size; | |
789 | if (truncate_seq != in->truncate_seq) { | |
790 | ldout(cct, 10) << "truncate_seq " << in->truncate_seq << " -> " | |
791 | << truncate_seq << dendl; | |
792 | in->truncate_seq = truncate_seq; | |
793 | in->oset.truncate_seq = truncate_seq; | |
794 | ||
795 | // truncate cached file data | |
796 | if (prior_size > size) { | |
797 | _invalidate_inode_cache(in, truncate_size, prior_size - truncate_size); | |
798 | } | |
799 | } | |
800 | ||
801 | // truncate inline data | |
802 | if (in->inline_version < CEPH_INLINE_NONE) { | |
803 | uint32_t len = in->inline_data.length(); | |
804 | if (size < len) | |
805 | in->inline_data.splice(size, len - size); | |
806 | } | |
807 | } | |
808 | if (truncate_seq >= in->truncate_seq && | |
809 | in->truncate_size != truncate_size) { | |
810 | if (in->is_file()) { | |
811 | ldout(cct, 10) << "truncate_size " << in->truncate_size << " -> " | |
812 | << truncate_size << dendl; | |
813 | in->truncate_size = truncate_size; | |
814 | in->oset.truncate_size = truncate_size; | |
815 | } else { | |
816 | ldout(cct, 0) << "Hmmm, truncate_seq && truncate_size changed on non-file inode!" << dendl; | |
817 | } | |
818 | } | |
1adf2230 AA |
819 | } |
820 | ||
821 | void Client::update_inode_file_time(Inode *in, int issued, uint64_t time_warp_seq, | |
822 | utime_t ctime, utime_t mtime, utime_t atime) | |
823 | { | |
824 | ldout(cct, 10) << __func__ << " " << *in << " " << ccap_string(issued) | |
825 | << " ctime " << ctime << " mtime " << mtime << dendl; | |
826 | ||
827 | if (time_warp_seq > in->time_warp_seq) | |
828 | ldout(cct, 10) << " mds time_warp_seq " << time_warp_seq | |
829 | << " is higher than local time_warp_seq " | |
830 | << in->time_warp_seq << dendl; | |
831 | ||
832 | int warn = false; | |
7c673cae FG |
833 | // be careful with size, mtime, atime |
834 | if (issued & (CEPH_CAP_FILE_EXCL| | |
835 | CEPH_CAP_FILE_WR| | |
836 | CEPH_CAP_FILE_BUFFER| | |
837 | CEPH_CAP_AUTH_EXCL| | |
838 | CEPH_CAP_XATTR_EXCL)) { | |
839 | ldout(cct, 30) << "Yay have enough caps to look at our times" << dendl; | |
840 | if (ctime > in->ctime) | |
841 | in->ctime = ctime; | |
842 | if (time_warp_seq > in->time_warp_seq) { | |
7c673cae FG |
843 | //the mds updated times, so take those! |
844 | in->mtime = mtime; | |
845 | in->atime = atime; | |
846 | in->time_warp_seq = time_warp_seq; | |
847 | } else if (time_warp_seq == in->time_warp_seq) { | |
848 | //take max times | |
849 | if (mtime > in->mtime) | |
850 | in->mtime = mtime; | |
851 | if (atime > in->atime) | |
852 | in->atime = atime; | |
853 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
854 | //ignore mds values as we have a higher seq | |
855 | } else warn = true; | |
856 | } else { | |
857 | ldout(cct, 30) << "Don't have enough caps, just taking mds' time values" << dendl; | |
858 | if (time_warp_seq >= in->time_warp_seq) { | |
859 | in->ctime = ctime; | |
860 | in->mtime = mtime; | |
861 | in->atime = atime; | |
862 | in->time_warp_seq = time_warp_seq; | |
863 | } else warn = true; | |
864 | } | |
865 | if (warn) { | |
866 | ldout(cct, 0) << "WARNING: " << *in << " mds time_warp_seq " | |
867 | << time_warp_seq << " is lower than local time_warp_seq " | |
868 | << in->time_warp_seq | |
869 | << dendl; | |
870 | } | |
871 | } | |
872 | ||
873 | void Client::_fragmap_remove_non_leaves(Inode *in) | |
874 | { | |
875 | for (map<frag_t,int>::iterator p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
876 | if (!in->dirfragtree.is_leaf(p->first)) | |
877 | in->fragmap.erase(p++); | |
878 | else | |
879 | ++p; | |
880 | } | |
881 | ||
882 | void Client::_fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds) | |
883 | { | |
884 | for (auto p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
885 | if (p->second == mds) | |
886 | in->fragmap.erase(p++); | |
887 | else | |
888 | ++p; | |
889 | } | |
890 | ||
891 | Inode * Client::add_update_inode(InodeStat *st, utime_t from, | |
892 | MetaSession *session, | |
893 | const UserPerm& request_perms) | |
894 | { | |
895 | Inode *in; | |
896 | bool was_new = false; | |
897 | if (inode_map.count(st->vino)) { | |
898 | in = inode_map[st->vino]; | |
11fdf7f2 | 899 | ldout(cct, 12) << __func__ << " had " << *in << " caps " << ccap_string(st->cap.caps) << dendl; |
7c673cae FG |
900 | } else { |
901 | in = new Inode(this, st->vino, &st->layout); | |
902 | inode_map[st->vino] = in; | |
903 | ||
904 | if (use_faked_inos()) | |
905 | _assign_faked_ino(in); | |
906 | ||
907 | if (!root) { | |
908 | root = in; | |
11fdf7f2 | 909 | if (use_faked_inos()) |
b3b6e05e | 910 | _assign_faked_root(root.get()); |
7c673cae FG |
911 | root_ancestor = in; |
912 | cwd = root; | |
f67539c2 | 913 | } else if (is_mounting()) { |
7c673cae FG |
914 | root_parents[root_ancestor] = in; |
915 | root_ancestor = in; | |
916 | } | |
917 | ||
918 | // immutable bits | |
919 | in->ino = st->vino.ino; | |
920 | in->snapid = st->vino.snapid; | |
921 | in->mode = st->mode & S_IFMT; | |
922 | was_new = true; | |
923 | } | |
924 | ||
925 | in->rdev = st->rdev; | |
926 | if (in->is_symlink()) | |
927 | in->symlink = st->symlink; | |
928 | ||
7c673cae | 929 | // only update inode if mds info is strictly newer, or it is the same and projected (odd). |
1adf2230 AA |
930 | bool new_version = false; |
931 | if (in->version == 0 || | |
932 | ((st->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
933 | (in->version & ~1) < st->version)) | |
934 | new_version = true; | |
7c673cae | 935 | |
1adf2230 AA |
936 | int issued; |
937 | in->caps_issued(&issued); | |
938 | issued |= in->caps_dirty(); | |
939 | int new_issued = ~issued & (int)st->cap.caps; | |
7c673cae | 940 | |
1adf2230 AA |
941 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
942 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
943 | in->mode = st->mode; | |
944 | in->uid = st->uid; | |
945 | in->gid = st->gid; | |
946 | in->btime = st->btime; | |
81eedcae | 947 | in->snap_btime = st->snap_btime; |
f67539c2 | 948 | in->snap_metadata = st->snap_metadata; |
1adf2230 | 949 | } |
7c673cae | 950 | |
1adf2230 AA |
951 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
952 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
953 | in->nlink = st->nlink; | |
954 | } | |
7c673cae | 955 | |
1adf2230 AA |
956 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
957 | update_inode_file_time(in, issued, st->time_warp_seq, | |
958 | st->ctime, st->mtime, st->atime); | |
959 | } | |
7c673cae | 960 | |
1adf2230 AA |
961 | if (new_version || |
962 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7c673cae | 963 | in->layout = st->layout; |
1adf2230 AA |
964 | update_inode_file_size(in, issued, st->size, st->truncate_seq, st->truncate_size); |
965 | } | |
7c673cae | 966 | |
1adf2230 AA |
967 | if (in->is_dir()) { |
968 | if (new_version || (new_issued & CEPH_CAP_FILE_SHARED)) { | |
969 | in->dirstat = st->dirstat; | |
970 | } | |
971 | // dir_layout/rstat/quota are not tracked by capability, update them only if | |
972 | // the inode stat is from auth mds | |
973 | if (new_version || (st->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
7c673cae FG |
974 | in->dir_layout = st->dir_layout; |
975 | ldout(cct, 20) << " dir hash is " << (int)in->dir_layout.dl_dir_hash << dendl; | |
1adf2230 AA |
976 | in->rstat = st->rstat; |
977 | in->quota = st->quota; | |
11fdf7f2 | 978 | in->dir_pin = st->dir_pin; |
1adf2230 AA |
979 | } |
980 | // move me if/when version reflects fragtree changes. | |
981 | if (in->dirfragtree != st->dirfragtree) { | |
982 | in->dirfragtree = st->dirfragtree; | |
983 | _fragmap_remove_non_leaves(in); | |
7c673cae | 984 | } |
7c673cae FG |
985 | } |
986 | ||
987 | if ((in->xattr_version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && | |
988 | st->xattrbl.length() && | |
989 | st->xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
990 | auto p = st->xattrbl.cbegin(); |
991 | decode(in->xattrs, p); | |
7c673cae FG |
992 | in->xattr_version = st->xattr_version; |
993 | } | |
994 | ||
1adf2230 AA |
995 | if (st->inline_version > in->inline_version) { |
996 | in->inline_data = st->inline_data; | |
997 | in->inline_version = st->inline_version; | |
7c673cae FG |
998 | } |
999 | ||
1adf2230 AA |
1000 | /* always take a newer change attr */ |
1001 | if (st->change_attr > in->change_attr) | |
1002 | in->change_attr = st->change_attr; | |
1003 | ||
1004 | if (st->version > in->version) | |
1005 | in->version = st->version; | |
1006 | ||
1007 | if (was_new) | |
1008 | ldout(cct, 12) << __func__ << " adding " << *in << " caps " << ccap_string(st->cap.caps) << dendl; | |
1009 | ||
1010 | if (!st->cap.caps) | |
1011 | return in; // as with readdir returning indoes in different snaprealms (no caps!) | |
1012 | ||
7c673cae | 1013 | if (in->snapid == CEPH_NOSNAP) { |
a8e16298 TL |
1014 | add_update_cap(in, session, st->cap.cap_id, st->cap.caps, st->cap.wanted, |
1015 | st->cap.seq, st->cap.mseq, inodeno_t(st->cap.realm), | |
1016 | st->cap.flags, request_perms); | |
28e407b8 | 1017 | if (in->auth_cap && in->auth_cap->session == session) { |
7c673cae | 1018 | in->max_size = st->max_size; |
28e407b8 AA |
1019 | in->rstat = st->rstat; |
1020 | } | |
7c673cae | 1021 | |
1adf2230 AA |
1022 | // setting I_COMPLETE needs to happen after adding the cap |
1023 | if (in->is_dir() && | |
1024 | (st->cap.caps & CEPH_CAP_FILE_SHARED) && | |
1025 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
1026 | in->dirstat.nfiles == 0 && | |
1027 | in->dirstat.nsubdirs == 0) { | |
1028 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on empty dir " << *in << dendl; | |
1029 | in->flags |= I_COMPLETE | I_DIR_ORDERED; | |
1030 | if (in->dir) { | |
1031 | ldout(cct, 10) << " dir is open on empty dir " << in->ino << " with " | |
1032 | << in->dir->dentries.size() << " entries, marking all dentries null" << dendl; | |
1033 | in->dir->readdir_cache.clear(); | |
1034 | for (const auto& p : in->dir->dentries) { | |
1035 | unlink(p.second, true, true); // keep dir, keep dentry | |
1036 | } | |
1037 | if (in->dir->dentries.empty()) | |
1038 | close_dir(in->dir); | |
7c673cae | 1039 | } |
7c673cae | 1040 | } |
1adf2230 AA |
1041 | } else { |
1042 | in->snap_caps |= st->cap.caps; | |
7c673cae FG |
1043 | } |
1044 | ||
f67539c2 | 1045 | in->fscrypt = st->fscrypt; |
7c673cae FG |
1046 | return in; |
1047 | } | |
1048 | ||
1049 | ||
1050 | /* | |
1051 | * insert_dentry_inode - insert + link a single dentry + inode into the metadata cache. | |
1052 | */ | |
1053 | Dentry *Client::insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease, | |
1054 | Inode *in, utime_t from, MetaSession *session, | |
1055 | Dentry *old_dentry) | |
1056 | { | |
1057 | Dentry *dn = NULL; | |
1058 | if (dir->dentries.count(dname)) | |
1059 | dn = dir->dentries[dname]; | |
1060 | ||
11fdf7f2 | 1061 | ldout(cct, 12) << __func__ << " '" << dname << "' vino " << in->vino() |
7c673cae FG |
1062 | << " in dir " << dir->parent_inode->vino() << " dn " << dn |
1063 | << dendl; | |
1064 | ||
1065 | if (dn && dn->inode) { | |
1066 | if (dn->inode->vino() == in->vino()) { | |
1067 | touch_dn(dn); | |
1068 | ldout(cct, 12) << " had dentry " << dname | |
1069 | << " with correct vino " << dn->inode->vino() | |
1070 | << dendl; | |
1071 | } else { | |
1072 | ldout(cct, 12) << " had dentry " << dname | |
1073 | << " with WRONG vino " << dn->inode->vino() | |
1074 | << dendl; | |
1075 | unlink(dn, true, true); // keep dir, keep dentry | |
1076 | } | |
1077 | } | |
1078 | ||
1079 | if (!dn || !dn->inode) { | |
1080 | InodeRef tmp_ref(in); | |
1081 | if (old_dentry) { | |
1082 | if (old_dentry->dir != dir) { | |
1083 | Inode *old_diri = old_dentry->dir->parent_inode; | |
7c673cae FG |
1084 | clear_dir_complete_and_ordered(old_diri, false); |
1085 | } | |
1086 | unlink(old_dentry, dir == old_dentry->dir, false); // drop dentry, keep dir open if its the same dir | |
1087 | } | |
1088 | Inode *diri = dir->parent_inode; | |
7c673cae FG |
1089 | clear_dir_complete_and_ordered(diri, false); |
1090 | dn = link(dir, dname, in, dn); | |
1091 | } | |
1092 | ||
1093 | update_dentry_lease(dn, dlease, from, session); | |
1094 | return dn; | |
1095 | } | |
1096 | ||
1097 | void Client::update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session) | |
1098 | { | |
1099 | utime_t dttl = from; | |
1100 | dttl += (float)dlease->duration_ms / 1000.0; | |
f67539c2 TL |
1101 | |
1102 | ldout(cct, 15) << __func__ << " " << *dn << " " << *dlease << " from " << from << dendl; | |
7c673cae | 1103 | |
11fdf7f2 | 1104 | ceph_assert(dn); |
7c673cae | 1105 | |
9f95a23c | 1106 | if (dlease->mask & CEPH_LEASE_VALID) { |
7c673cae FG |
1107 | if (dttl > dn->lease_ttl) { |
1108 | ldout(cct, 10) << "got dentry lease on " << dn->name | |
1109 | << " dur " << dlease->duration_ms << "ms ttl " << dttl << dendl; | |
1110 | dn->lease_ttl = dttl; | |
1111 | dn->lease_mds = session->mds_num; | |
1112 | dn->lease_seq = dlease->seq; | |
1113 | dn->lease_gen = session->cap_gen; | |
1114 | } | |
1115 | } | |
1116 | dn->cap_shared_gen = dn->dir->parent_inode->shared_gen; | |
f91f0fd5 TL |
1117 | if (dlease->mask & CEPH_LEASE_PRIMARY_LINK) |
1118 | dn->mark_primary(); | |
f67539c2 | 1119 | dn->alternate_name = std::move(dlease->alternate_name); |
7c673cae FG |
1120 | } |
1121 | ||
1122 | ||
1123 | /* | |
1124 | * update MDS location cache for a single inode | |
1125 | */ | |
1126 | void Client::update_dir_dist(Inode *in, DirStat *dst) | |
1127 | { | |
1128 | // auth | |
1129 | ldout(cct, 20) << "got dirfrag map for " << in->ino << " frag " << dst->frag << " to mds " << dst->auth << dendl; | |
1130 | if (dst->auth >= 0) { | |
1131 | in->fragmap[dst->frag] = dst->auth; | |
1132 | } else { | |
1133 | in->fragmap.erase(dst->frag); | |
1134 | } | |
1135 | if (!in->dirfragtree.is_leaf(dst->frag)) { | |
1136 | in->dirfragtree.force_to_leaf(cct, dst->frag); | |
1137 | _fragmap_remove_non_leaves(in); | |
1138 | } | |
1139 | ||
1140 | // replicated | |
f67539c2 TL |
1141 | in->dir_replicated = !dst->dist.empty(); |
1142 | if (!dst->dist.empty()) | |
1143 | in->frag_repmap[dst->frag].assign(dst->dist.begin(), dst->dist.end()) ; | |
1144 | else | |
1145 | in->frag_repmap.erase(dst->frag); | |
7c673cae FG |
1146 | } |
1147 | ||
1148 | void Client::clear_dir_complete_and_ordered(Inode *diri, bool complete) | |
1149 | { | |
f91f0fd5 TL |
1150 | if (complete) |
1151 | diri->dir_release_count++; | |
1152 | else | |
1153 | diri->dir_ordered_count++; | |
7c673cae FG |
1154 | if (diri->flags & I_COMPLETE) { |
1155 | if (complete) { | |
1156 | ldout(cct, 10) << " clearing (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
1157 | diri->flags &= ~(I_COMPLETE | I_DIR_ORDERED); | |
1158 | } else { | |
1159 | if (diri->flags & I_DIR_ORDERED) { | |
1160 | ldout(cct, 10) << " clearing I_DIR_ORDERED on " << *diri << dendl; | |
1161 | diri->flags &= ~I_DIR_ORDERED; | |
1162 | } | |
1163 | } | |
1164 | if (diri->dir) | |
1165 | diri->dir->readdir_cache.clear(); | |
1166 | } | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * insert results from readdir or lssnap into the metadata cache. | |
1171 | */ | |
1172 | void Client::insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri) { | |
1173 | ||
11fdf7f2 | 1174 | auto& reply = request->reply; |
7c673cae | 1175 | ConnectionRef con = request->reply->get_connection(); |
11fdf7f2 TL |
1176 | uint64_t features; |
1177 | if(session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1178 | features = (uint64_t)-1; | |
1179 | } | |
1180 | else { | |
1181 | features = con->get_features(); | |
1182 | } | |
7c673cae FG |
1183 | |
1184 | dir_result_t *dirp = request->dirp; | |
11fdf7f2 | 1185 | ceph_assert(dirp); |
7c673cae FG |
1186 | |
1187 | // the extra buffer list is only set for readdir and lssnap replies | |
11fdf7f2 | 1188 | auto p = reply->get_extra_bl().cbegin(); |
7c673cae FG |
1189 | if (!p.end()) { |
1190 | // snapdir? | |
1191 | if (request->head.op == CEPH_MDS_OP_LSSNAP) { | |
11fdf7f2 | 1192 | ceph_assert(diri); |
7c673cae FG |
1193 | diri = open_snapdir(diri); |
1194 | } | |
1195 | ||
1196 | // only open dir if we're actually adding stuff to it! | |
1197 | Dir *dir = diri->open_dir(); | |
11fdf7f2 | 1198 | ceph_assert(dir); |
7c673cae FG |
1199 | |
1200 | // dirstat | |
11fdf7f2 | 1201 | DirStat dst(p, features); |
7c673cae FG |
1202 | __u32 numdn; |
1203 | __u16 flags; | |
11fdf7f2 TL |
1204 | decode(numdn, p); |
1205 | decode(flags, p); | |
7c673cae FG |
1206 | |
1207 | bool end = ((unsigned)flags & CEPH_READDIR_FRAG_END); | |
1208 | bool hash_order = ((unsigned)flags & CEPH_READDIR_HASH_ORDER); | |
1209 | ||
1210 | frag_t fg = (unsigned)request->head.args.readdir.frag; | |
1211 | unsigned readdir_offset = dirp->next_offset; | |
1212 | string readdir_start = dirp->last_name; | |
11fdf7f2 | 1213 | ceph_assert(!readdir_start.empty() || readdir_offset == 2); |
7c673cae FG |
1214 | |
1215 | unsigned last_hash = 0; | |
1216 | if (hash_order) { | |
1217 | if (!readdir_start.empty()) { | |
1218 | last_hash = ceph_frag_value(diri->hash_dentry_name(readdir_start)); | |
1219 | } else if (flags & CEPH_READDIR_OFFSET_HASH) { | |
1220 | /* mds understands offset_hash */ | |
1221 | last_hash = (unsigned)request->head.args.readdir.offset_hash; | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | if (fg != dst.frag) { | |
1226 | ldout(cct, 10) << "insert_trace got new frag " << fg << " -> " << dst.frag << dendl; | |
1227 | fg = dst.frag; | |
1228 | if (!hash_order) { | |
1229 | readdir_offset = 2; | |
1230 | readdir_start.clear(); | |
1231 | dirp->offset = dir_result_t::make_fpos(fg, readdir_offset, false); | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | ldout(cct, 10) << __func__ << " " << numdn << " readdir items, end=" << end | |
1236 | << ", hash_order=" << hash_order | |
1237 | << ", readdir_start " << readdir_start | |
1238 | << ", last_hash " << last_hash | |
1239 | << ", next_offset " << readdir_offset << dendl; | |
1240 | ||
1241 | if (diri->snapid != CEPH_SNAPDIR && | |
1242 | fg.is_leftmost() && readdir_offset == 2 && | |
1243 | !(hash_order && last_hash)) { | |
1244 | dirp->release_count = diri->dir_release_count; | |
1245 | dirp->ordered_count = diri->dir_ordered_count; | |
1246 | dirp->start_shared_gen = diri->shared_gen; | |
1247 | dirp->cache_index = 0; | |
1248 | } | |
1249 | ||
1250 | dirp->buffer_frag = fg; | |
1251 | ||
1252 | _readdir_drop_dirp_buffer(dirp); | |
1253 | dirp->buffer.reserve(numdn); | |
1254 | ||
1255 | string dname; | |
1256 | LeaseStat dlease; | |
1257 | for (unsigned i=0; i<numdn; i++) { | |
11fdf7f2 TL |
1258 | decode(dname, p); |
1259 | dlease.decode(p, features); | |
7c673cae FG |
1260 | InodeStat ist(p, features); |
1261 | ||
1262 | ldout(cct, 15) << "" << i << ": '" << dname << "'" << dendl; | |
1263 | ||
1264 | Inode *in = add_update_inode(&ist, request->sent_stamp, session, | |
1265 | request->perms); | |
1266 | Dentry *dn; | |
1267 | if (diri->dir->dentries.count(dname)) { | |
1268 | Dentry *olddn = diri->dir->dentries[dname]; | |
1269 | if (olddn->inode != in) { | |
1270 | // replace incorrect dentry | |
1271 | unlink(olddn, true, true); // keep dir, dentry | |
1272 | dn = link(dir, dname, in, olddn); | |
11fdf7f2 | 1273 | ceph_assert(dn == olddn); |
7c673cae FG |
1274 | } else { |
1275 | // keep existing dn | |
1276 | dn = olddn; | |
1277 | touch_dn(dn); | |
1278 | } | |
1279 | } else { | |
1280 | // new dn | |
1281 | dn = link(dir, dname, in, NULL); | |
1282 | } | |
f67539c2 | 1283 | dn->alternate_name = std::move(dlease.alternate_name); |
7c673cae FG |
1284 | |
1285 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1286 | if (hash_order) { | |
1287 | unsigned hash = ceph_frag_value(diri->hash_dentry_name(dname)); | |
1288 | if (hash != last_hash) | |
1289 | readdir_offset = 2; | |
1290 | last_hash = hash; | |
1291 | dn->offset = dir_result_t::make_fpos(hash, readdir_offset++, true); | |
1292 | } else { | |
1293 | dn->offset = dir_result_t::make_fpos(fg, readdir_offset++, false); | |
1294 | } | |
1295 | // add to readdir cache | |
1296 | if (dirp->release_count == diri->dir_release_count && | |
1297 | dirp->ordered_count == diri->dir_ordered_count && | |
1298 | dirp->start_shared_gen == diri->shared_gen) { | |
1299 | if (dirp->cache_index == dir->readdir_cache.size()) { | |
1300 | if (i == 0) { | |
11fdf7f2 | 1301 | ceph_assert(!dirp->inode->is_complete_and_ordered()); |
7c673cae FG |
1302 | dir->readdir_cache.reserve(dirp->cache_index + numdn); |
1303 | } | |
1304 | dir->readdir_cache.push_back(dn); | |
1305 | } else if (dirp->cache_index < dir->readdir_cache.size()) { | |
1306 | if (dirp->inode->is_complete_and_ordered()) | |
11fdf7f2 | 1307 | ceph_assert(dir->readdir_cache[dirp->cache_index] == dn); |
7c673cae FG |
1308 | else |
1309 | dir->readdir_cache[dirp->cache_index] = dn; | |
1310 | } else { | |
11fdf7f2 | 1311 | ceph_abort_msg("unexpected readdir buffer idx"); |
7c673cae FG |
1312 | } |
1313 | dirp->cache_index++; | |
1314 | } | |
1315 | // add to cached result list | |
f67539c2 | 1316 | dirp->buffer.push_back(dir_result_t::dentry(dn->offset, dname, dn->alternate_name, in)); |
7c673cae FG |
1317 | ldout(cct, 15) << __func__ << " " << hex << dn->offset << dec << ": '" << dname << "' -> " << in->ino << dendl; |
1318 | } | |
1319 | ||
1320 | if (numdn > 0) | |
1321 | dirp->last_name = dname; | |
1322 | if (end) | |
1323 | dirp->next_offset = 2; | |
1324 | else | |
1325 | dirp->next_offset = readdir_offset; | |
1326 | ||
1327 | if (dir->is_empty()) | |
1328 | close_dir(dir); | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | /** insert_trace | |
1333 | * | |
1334 | * insert a trace from a MDS reply into the cache. | |
1335 | */ | |
1336 | Inode* Client::insert_trace(MetaRequest *request, MetaSession *session) | |
1337 | { | |
11fdf7f2 | 1338 | auto& reply = request->reply; |
7c673cae FG |
1339 | int op = request->get_op(); |
1340 | ||
1341 | ldout(cct, 10) << "insert_trace from " << request->sent_stamp << " mds." << session->mds_num | |
1342 | << " is_target=" << (int)reply->head.is_target | |
1343 | << " is_dentry=" << (int)reply->head.is_dentry | |
1344 | << dendl; | |
1345 | ||
11fdf7f2 | 1346 | auto p = reply->get_trace_bl().cbegin(); |
7c673cae FG |
1347 | if (request->got_unsafe) { |
1348 | ldout(cct, 10) << "insert_trace -- already got unsafe; ignoring" << dendl; | |
11fdf7f2 | 1349 | ceph_assert(p.end()); |
7c673cae FG |
1350 | return NULL; |
1351 | } | |
1352 | ||
1353 | if (p.end()) { | |
1354 | ldout(cct, 10) << "insert_trace -- no trace" << dendl; | |
1355 | ||
1356 | Dentry *d = request->dentry(); | |
1357 | if (d) { | |
1358 | Inode *diri = d->dir->parent_inode; | |
7c673cae FG |
1359 | clear_dir_complete_and_ordered(diri, true); |
1360 | } | |
1361 | ||
1362 | if (d && reply->get_result() == 0) { | |
1363 | if (op == CEPH_MDS_OP_RENAME) { | |
1364 | // rename | |
1365 | Dentry *od = request->old_dentry(); | |
1366 | ldout(cct, 10) << " unlinking rename src dn " << od << " for traceless reply" << dendl; | |
11fdf7f2 | 1367 | ceph_assert(od); |
7c673cae FG |
1368 | unlink(od, true, true); // keep dir, dentry |
1369 | } else if (op == CEPH_MDS_OP_RMDIR || | |
1370 | op == CEPH_MDS_OP_UNLINK) { | |
1371 | // unlink, rmdir | |
1372 | ldout(cct, 10) << " unlinking unlink/rmdir dn " << d << " for traceless reply" << dendl; | |
1373 | unlink(d, true, true); // keep dir, dentry | |
1374 | } | |
1375 | } | |
1376 | return NULL; | |
1377 | } | |
1378 | ||
1379 | ConnectionRef con = request->reply->get_connection(); | |
11fdf7f2 TL |
1380 | uint64_t features; |
1381 | if (session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1382 | features = (uint64_t)-1; | |
1383 | } | |
1384 | else { | |
1385 | features = con->get_features(); | |
1386 | } | |
7c673cae FG |
1387 | ldout(cct, 10) << " features 0x" << hex << features << dec << dendl; |
1388 | ||
1389 | // snap trace | |
1390 | SnapRealm *realm = NULL; | |
1391 | if (reply->snapbl.length()) | |
1392 | update_snap_trace(reply->snapbl, &realm); | |
1393 | ||
1394 | ldout(cct, 10) << " hrm " | |
1395 | << " is_target=" << (int)reply->head.is_target | |
1396 | << " is_dentry=" << (int)reply->head.is_dentry | |
1397 | << dendl; | |
1398 | ||
1399 | InodeStat dirst; | |
1400 | DirStat dst; | |
1401 | string dname; | |
1402 | LeaseStat dlease; | |
1403 | InodeStat ist; | |
1404 | ||
1405 | if (reply->head.is_dentry) { | |
1406 | dirst.decode(p, features); | |
11fdf7f2 TL |
1407 | dst.decode(p, features); |
1408 | decode(dname, p); | |
1409 | dlease.decode(p, features); | |
7c673cae FG |
1410 | } |
1411 | ||
1412 | Inode *in = 0; | |
1413 | if (reply->head.is_target) { | |
1414 | ist.decode(p, features); | |
1415 | if (cct->_conf->client_debug_getattr_caps) { | |
1416 | unsigned wanted = 0; | |
1417 | if (op == CEPH_MDS_OP_GETATTR || op == CEPH_MDS_OP_LOOKUP) | |
1418 | wanted = request->head.args.getattr.mask; | |
1419 | else if (op == CEPH_MDS_OP_OPEN || op == CEPH_MDS_OP_CREATE) | |
1420 | wanted = request->head.args.open.mask; | |
1421 | ||
1422 | if ((wanted & CEPH_CAP_XATTR_SHARED) && | |
1423 | !(ist.xattr_version > 0 && ist.xattrbl.length() > 0)) | |
11fdf7f2 | 1424 | ceph_abort_msg("MDS reply does not contain xattrs"); |
7c673cae FG |
1425 | } |
1426 | ||
1427 | in = add_update_inode(&ist, request->sent_stamp, session, | |
1428 | request->perms); | |
1429 | } | |
1430 | ||
1431 | Inode *diri = NULL; | |
1432 | if (reply->head.is_dentry) { | |
1433 | diri = add_update_inode(&dirst, request->sent_stamp, session, | |
1434 | request->perms); | |
1435 | update_dir_dist(diri, &dst); // dir stat info is attached to .. | |
1436 | ||
1437 | if (in) { | |
1438 | Dir *dir = diri->open_dir(); | |
1439 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session, | |
1440 | (op == CEPH_MDS_OP_RENAME) ? request->old_dentry() : NULL); | |
1441 | } else { | |
1442 | Dentry *dn = NULL; | |
1443 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1444 | dn = diri->dir->dentries[dname]; | |
1445 | if (dn->inode) { | |
7c673cae FG |
1446 | clear_dir_complete_and_ordered(diri, false); |
1447 | unlink(dn, true, true); // keep dir, dentry | |
1448 | } | |
1449 | } | |
1450 | if (dlease.duration_ms > 0) { | |
1451 | if (!dn) { | |
1452 | Dir *dir = diri->open_dir(); | |
1453 | dn = link(dir, dname, NULL, NULL); | |
1454 | } | |
1455 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1456 | } | |
1457 | } | |
1458 | } else if (op == CEPH_MDS_OP_LOOKUPSNAP || | |
1459 | op == CEPH_MDS_OP_MKSNAP) { | |
1460 | ldout(cct, 10) << " faking snap lookup weirdness" << dendl; | |
1461 | // fake it for snap lookup | |
1462 | vinodeno_t vino = ist.vino; | |
1463 | vino.snapid = CEPH_SNAPDIR; | |
11fdf7f2 | 1464 | ceph_assert(inode_map.count(vino)); |
7c673cae FG |
1465 | diri = inode_map[vino]; |
1466 | ||
1467 | string dname = request->path.last_dentry(); | |
1468 | ||
1469 | LeaseStat dlease; | |
1470 | dlease.duration_ms = 0; | |
1471 | ||
1472 | if (in) { | |
1473 | Dir *dir = diri->open_dir(); | |
1474 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session); | |
1475 | } else { | |
1476 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1477 | Dentry *dn = diri->dir->dentries[dname]; | |
1478 | if (dn->inode) | |
1479 | unlink(dn, true, true); // keep dir, dentry | |
1480 | } | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | if (in) { | |
1485 | if (op == CEPH_MDS_OP_READDIR || | |
1486 | op == CEPH_MDS_OP_LSSNAP) { | |
1487 | insert_readdir_results(request, session, in); | |
1488 | } else if (op == CEPH_MDS_OP_LOOKUPNAME) { | |
1489 | // hack: return parent inode instead | |
1490 | in = diri; | |
1491 | } | |
1492 | ||
1493 | if (request->dentry() == NULL && in != request->inode()) { | |
1494 | // pin the target inode if its parent dentry is not pinned | |
1495 | request->set_other_inode(in); | |
1496 | } | |
1497 | } | |
1498 | ||
1499 | if (realm) | |
1500 | put_snap_realm(realm); | |
1501 | ||
1502 | request->target = in; | |
1503 | return in; | |
1504 | } | |
1505 | ||
1506 | // ------- | |
1507 | ||
1508 | mds_rank_t Client::choose_target_mds(MetaRequest *req, Inode** phash_diri) | |
1509 | { | |
1510 | mds_rank_t mds = MDS_RANK_NONE; | |
1511 | __u32 hash = 0; | |
1512 | bool is_hash = false; | |
1513 | ||
1514 | Inode *in = NULL; | |
1515 | Dentry *de = NULL; | |
7c673cae FG |
1516 | |
1517 | if (req->resend_mds >= 0) { | |
1518 | mds = req->resend_mds; | |
1519 | req->resend_mds = -1; | |
11fdf7f2 | 1520 | ldout(cct, 10) << __func__ << " resend_mds specified as mds." << mds << dendl; |
7c673cae FG |
1521 | goto out; |
1522 | } | |
1523 | ||
1524 | if (cct->_conf->client_use_random_mds) | |
1525 | goto random_mds; | |
1526 | ||
1527 | in = req->inode(); | |
1528 | de = req->dentry(); | |
1529 | if (in) { | |
11fdf7f2 | 1530 | ldout(cct, 20) << __func__ << " starting with req->inode " << *in << dendl; |
7c673cae FG |
1531 | if (req->path.depth()) { |
1532 | hash = in->hash_dentry_name(req->path[0]); | |
11fdf7f2 | 1533 | ldout(cct, 20) << __func__ << " inode dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1534 | << " on " << req->path[0] |
1535 | << " => " << hash << dendl; | |
1536 | is_hash = true; | |
1537 | } | |
1538 | } else if (de) { | |
1539 | if (de->inode) { | |
1540 | in = de->inode.get(); | |
11fdf7f2 | 1541 | ldout(cct, 20) << __func__ << " starting with req->dentry inode " << *in << dendl; |
7c673cae FG |
1542 | } else { |
1543 | in = de->dir->parent_inode; | |
1544 | hash = in->hash_dentry_name(de->name); | |
11fdf7f2 | 1545 | ldout(cct, 20) << __func__ << " dentry dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1546 | << " on " << de->name |
1547 | << " => " << hash << dendl; | |
1548 | is_hash = true; | |
1549 | } | |
1550 | } | |
1551 | if (in) { | |
1552 | if (in->snapid != CEPH_NOSNAP) { | |
11fdf7f2 | 1553 | ldout(cct, 10) << __func__ << " " << *in << " is snapped, using nonsnap parent" << dendl; |
7c673cae FG |
1554 | while (in->snapid != CEPH_NOSNAP) { |
1555 | if (in->snapid == CEPH_SNAPDIR) | |
1556 | in = in->snapdir_parent.get(); | |
11fdf7f2 | 1557 | else if (!in->dentries.empty()) |
7c673cae FG |
1558 | /* In most cases there will only be one dentry, so getting it |
1559 | * will be the correct action. If there are multiple hard links, | |
1560 | * I think the MDS should be able to redirect as needed*/ | |
1561 | in = in->get_first_parent()->dir->parent_inode; | |
1562 | else { | |
1563 | ldout(cct, 10) << "got unlinked inode, can't look at parent" << dendl; | |
1564 | break; | |
1565 | } | |
1566 | } | |
1567 | is_hash = false; | |
1568 | } | |
1569 | ||
11fdf7f2 | 1570 | ldout(cct, 20) << __func__ << " " << *in << " is_hash=" << is_hash |
7c673cae FG |
1571 | << " hash=" << hash << dendl; |
1572 | ||
f67539c2 | 1573 | if (is_hash && S_ISDIR(in->mode) && (!in->fragmap.empty() || !in->frag_repmap.empty())) { |
7c673cae | 1574 | frag_t fg = in->dirfragtree[hash]; |
f67539c2 TL |
1575 | if (!req->auth_is_best()) { |
1576 | auto repmapit = in->frag_repmap.find(fg); | |
1577 | if (repmapit != in->frag_repmap.end()) { | |
1578 | auto& repmap = repmapit->second; | |
1579 | auto r = ceph::util::generate_random_number<uint64_t>(0, repmap.size()-1); | |
1580 | mds = repmap.at(r); | |
1581 | } | |
1582 | } else if (in->fragmap.count(fg)) { | |
7c673cae FG |
1583 | mds = in->fragmap[fg]; |
1584 | if (phash_diri) | |
1585 | *phash_diri = in; | |
91327a77 | 1586 | } else if (in->auth_cap) { |
f67539c2 | 1587 | req->send_to_auth = true; |
91327a77 AA |
1588 | mds = in->auth_cap->session->mds_num; |
1589 | } | |
1590 | if (mds >= 0) { | |
11fdf7f2 | 1591 | ldout(cct, 10) << __func__ << " from dirfragtree hash" << dendl; |
7c673cae FG |
1592 | goto out; |
1593 | } | |
1594 | } | |
1595 | ||
11fdf7f2 TL |
1596 | if (in->auth_cap && req->auth_is_best()) { |
1597 | mds = in->auth_cap->session->mds_num; | |
1598 | } else if (!in->caps.empty()) { | |
1599 | mds = in->caps.begin()->second.session->mds_num; | |
1600 | } else { | |
7c673cae | 1601 | goto random_mds; |
11fdf7f2 TL |
1602 | } |
1603 | ldout(cct, 10) << __func__ << " from caps on inode " << *in << dendl; | |
7c673cae FG |
1604 | |
1605 | goto out; | |
1606 | } | |
1607 | ||
1608 | random_mds: | |
1609 | if (mds < 0) { | |
1610 | mds = _get_random_up_mds(); | |
1611 | ldout(cct, 10) << "did not get mds through better means, so chose random mds " << mds << dendl; | |
1612 | } | |
1613 | ||
1614 | out: | |
1615 | ldout(cct, 20) << "mds is " << mds << dendl; | |
1616 | return mds; | |
1617 | } | |
1618 | ||
7c673cae FG |
1619 | void Client::connect_mds_targets(mds_rank_t mds) |
1620 | { | |
11fdf7f2 TL |
1621 | ldout(cct, 10) << __func__ << " for mds." << mds << dendl; |
1622 | ceph_assert(mds_sessions.count(mds)); | |
7c673cae | 1623 | const MDSMap::mds_info_t& info = mdsmap->get_mds_info(mds); |
f67539c2 TL |
1624 | for (const auto &rank : info.export_targets) { |
1625 | if (mds_sessions.count(rank) == 0 && | |
1626 | mdsmap->is_clientreplay_or_active_or_stopping(rank)) { | |
7c673cae | 1627 | ldout(cct, 10) << "check_mds_sessions opening mds." << mds |
f67539c2 TL |
1628 | << " export target mds." << rank << dendl; |
1629 | _open_mds_session(rank); | |
7c673cae FG |
1630 | } |
1631 | } | |
1632 | } | |
1633 | ||
adb31ebb | 1634 | void Client::dump_mds_sessions(Formatter *f, bool cap_dump) |
7c673cae FG |
1635 | { |
1636 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 1637 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 AA |
1638 | f->dump_object("inst", inst); |
1639 | f->dump_stream("inst_str") << inst; | |
1640 | f->dump_stream("addr_str") << inst.addr; | |
7c673cae | 1641 | f->open_array_section("sessions"); |
11fdf7f2 | 1642 | for (const auto &p : mds_sessions) { |
7c673cae | 1643 | f->open_object_section("session"); |
adb31ebb | 1644 | p.second.dump(f, cap_dump); |
7c673cae FG |
1645 | f->close_section(); |
1646 | } | |
1647 | f->close_section(); | |
1648 | f->dump_int("mdsmap_epoch", mdsmap->get_epoch()); | |
1649 | } | |
f67539c2 | 1650 | |
7c673cae FG |
1651 | void Client::dump_mds_requests(Formatter *f) |
1652 | { | |
1653 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
1654 | p != mds_requests.end(); | |
1655 | ++p) { | |
1656 | f->open_object_section("request"); | |
1657 | p->second->dump(f); | |
1658 | f->close_section(); | |
1659 | } | |
1660 | } | |
1661 | ||
9f95a23c | 1662 | int Client::verify_reply_trace(int r, MetaSession *session, |
11fdf7f2 | 1663 | MetaRequest *request, const MConstRef<MClientReply>& reply, |
7c673cae FG |
1664 | InodeRef *ptarget, bool *pcreated, |
1665 | const UserPerm& perms) | |
1666 | { | |
1667 | // check whether this request actually did the create, and set created flag | |
1668 | bufferlist extra_bl; | |
1669 | inodeno_t created_ino; | |
1670 | bool got_created_ino = false; | |
1671 | ceph::unordered_map<vinodeno_t, Inode*>::iterator p; | |
1672 | ||
11fdf7f2 | 1673 | extra_bl = reply->get_extra_bl(); |
7c673cae | 1674 | if (extra_bl.length() >= 8) { |
9f95a23c TL |
1675 | if (session->mds_features.test(CEPHFS_FEATURE_DELEG_INO)) { |
1676 | struct openc_response_t ocres; | |
1677 | ||
1678 | decode(ocres, extra_bl); | |
1679 | created_ino = ocres.created_ino; | |
1680 | /* | |
1681 | * The userland cephfs client doesn't have a way to do an async create | |
1682 | * (yet), so just discard delegated_inos for now. Eventually we should | |
1683 | * store them and use them in create calls, even if they are synchronous, | |
1684 | * if only for testing purposes. | |
1685 | */ | |
1686 | ldout(cct, 10) << "delegated_inos: " << ocres.delegated_inos << dendl; | |
1687 | } else { | |
1688 | // u64 containing number of created ino | |
1689 | decode(created_ino, extra_bl); | |
1690 | } | |
7c673cae | 1691 | ldout(cct, 10) << "make_request created ino " << created_ino << dendl; |
9f95a23c | 1692 | got_created_ino = true; |
7c673cae FG |
1693 | } |
1694 | ||
1695 | if (pcreated) | |
1696 | *pcreated = got_created_ino; | |
1697 | ||
1698 | if (request->target) { | |
1699 | *ptarget = request->target; | |
1700 | ldout(cct, 20) << "make_request target is " << *ptarget->get() << dendl; | |
1701 | } else { | |
1702 | if (got_created_ino && (p = inode_map.find(vinodeno_t(created_ino, CEPH_NOSNAP))) != inode_map.end()) { | |
1703 | (*ptarget) = p->second; | |
1704 | ldout(cct, 20) << "make_request created, target is " << *ptarget->get() << dendl; | |
1705 | } else { | |
1706 | // we got a traceless reply, and need to look up what we just | |
1707 | // created. for now, do this by name. someday, do this by the | |
1708 | // ino... which we know! FIXME. | |
1709 | InodeRef target; | |
1710 | Dentry *d = request->dentry(); | |
1711 | if (d) { | |
1712 | if (d->dir) { | |
1713 | ldout(cct, 10) << "make_request got traceless reply, looking up #" | |
1714 | << d->dir->parent_inode->ino << "/" << d->name | |
1715 | << " got_ino " << got_created_ino | |
1716 | << " ino " << created_ino | |
1717 | << dendl; | |
1718 | r = _do_lookup(d->dir->parent_inode, d->name, request->regetattr_mask, | |
1719 | &target, perms); | |
1720 | } else { | |
1721 | // if the dentry is not linked, just do our best. see #5021. | |
11fdf7f2 | 1722 | ceph_abort_msg("how did this happen? i want logs!"); |
7c673cae FG |
1723 | } |
1724 | } else { | |
1725 | Inode *in = request->inode(); | |
1726 | ldout(cct, 10) << "make_request got traceless reply, forcing getattr on #" | |
1727 | << in->ino << dendl; | |
1728 | r = _getattr(in, request->regetattr_mask, perms, true); | |
1729 | target = in; | |
1730 | } | |
1731 | if (r >= 0) { | |
1732 | // verify ino returned in reply and trace_dist are the same | |
1733 | if (got_created_ino && | |
1734 | created_ino.val != target->ino.val) { | |
1735 | ldout(cct, 5) << "create got ino " << created_ino << " but then failed on lookup; EINTR?" << dendl; | |
f67539c2 | 1736 | r = -CEPHFS_EINTR; |
7c673cae FG |
1737 | } |
1738 | if (ptarget) | |
1739 | ptarget->swap(target); | |
1740 | } | |
1741 | } | |
1742 | } | |
1743 | ||
1744 | return r; | |
1745 | } | |
1746 | ||
1747 | ||
1748 | /** | |
1749 | * make a request | |
1750 | * | |
1751 | * Blocking helper to make an MDS request. | |
1752 | * | |
1753 | * If the ptarget flag is set, behavior changes slightly: the caller | |
1754 | * expects to get a pointer to the inode we are creating or operating | |
1755 | * on. As a result, we will follow up any traceless mutation reply | |
1756 | * with a getattr or lookup to transparently handle a traceless reply | |
1757 | * from the MDS (as when the MDS restarts and the client has to replay | |
1758 | * a request). | |
1759 | * | |
1760 | * @param request the MetaRequest to execute | |
1761 | * @param perms The user uid/gid to execute as (eventually, full group lists?) | |
1762 | * @param ptarget [optional] address to store a pointer to the target inode we want to create or operate on | |
1763 | * @param pcreated [optional; required if ptarget] where to store a bool of whether our create atomically created a file | |
1764 | * @param use_mds [optional] prefer a specific mds (-1 for default) | |
1765 | * @param pdirbl [optional; disallowed if ptarget] where to pass extra reply payload to the caller | |
1766 | */ | |
1767 | int Client::make_request(MetaRequest *request, | |
1768 | const UserPerm& perms, | |
1769 | InodeRef *ptarget, bool *pcreated, | |
1770 | mds_rank_t use_mds, | |
1771 | bufferlist *pdirbl) | |
1772 | { | |
1773 | int r = 0; | |
1774 | ||
1775 | // assign a unique tid | |
1776 | ceph_tid_t tid = ++last_tid; | |
1777 | request->set_tid(tid); | |
1778 | ||
1779 | // and timestamp | |
1780 | request->op_stamp = ceph_clock_now(); | |
1781 | ||
1782 | // make note | |
1783 | mds_requests[tid] = request->get(); | |
1784 | if (oldest_tid == 0 && request->get_op() != CEPH_MDS_OP_SETFILELOCK) | |
1785 | oldest_tid = tid; | |
1786 | ||
1787 | request->set_caller_perms(perms); | |
1788 | ||
1789 | if (cct->_conf->client_inject_fixed_oldest_tid) { | |
1790 | ldout(cct, 20) << __func__ << " injecting fixed oldest_client_tid(1)" << dendl; | |
1791 | request->set_oldest_client_tid(1); | |
1792 | } else { | |
1793 | request->set_oldest_client_tid(oldest_tid); | |
1794 | } | |
1795 | ||
1796 | // hack target mds? | |
1797 | if (use_mds >= 0) | |
1798 | request->resend_mds = use_mds; | |
1799 | ||
9f95a23c | 1800 | MetaSession *session = NULL; |
7c673cae FG |
1801 | while (1) { |
1802 | if (request->aborted()) | |
1803 | break; | |
1804 | ||
f67539c2 TL |
1805 | if (blocklisted) { |
1806 | request->abort(-CEPHFS_EBLOCKLISTED); | |
31f18b77 FG |
1807 | break; |
1808 | } | |
1809 | ||
7c673cae | 1810 | // set up wait cond |
9f95a23c | 1811 | ceph::condition_variable caller_cond; |
7c673cae FG |
1812 | request->caller_cond = &caller_cond; |
1813 | ||
1814 | // choose mds | |
1815 | Inode *hash_diri = NULL; | |
1816 | mds_rank_t mds = choose_target_mds(request, &hash_diri); | |
1817 | int mds_state = (mds == MDS_RANK_NONE) ? MDSMap::STATE_NULL : mdsmap->get_state(mds); | |
1818 | if (mds_state != MDSMap::STATE_ACTIVE && mds_state != MDSMap::STATE_STOPPING) { | |
1819 | if (mds_state == MDSMap::STATE_NULL && mds >= mdsmap->get_max_mds()) { | |
1820 | if (hash_diri) { | |
1821 | ldout(cct, 10) << " target mds." << mds << " has stopped, remove it from fragmap" << dendl; | |
1822 | _fragmap_remove_stopped_mds(hash_diri, mds); | |
1823 | } else { | |
1824 | ldout(cct, 10) << " target mds." << mds << " has stopped, trying a random mds" << dendl; | |
1825 | request->resend_mds = _get_random_up_mds(); | |
1826 | } | |
1827 | } else { | |
1828 | ldout(cct, 10) << " target mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
1829 | wait_on_list(waiting_for_mdsmap); | |
1830 | } | |
1831 | continue; | |
1832 | } | |
1833 | ||
1834 | // open a session? | |
7c673cae FG |
1835 | if (!have_open_session(mds)) { |
1836 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 1837 | if (session->state == MetaSession::STATE_REJECTED) { |
f67539c2 | 1838 | request->abort(-CEPHFS_EPERM); |
f6b5b4d7 TL |
1839 | break; |
1840 | } | |
7c673cae FG |
1841 | // wait |
1842 | if (session->state == MetaSession::STATE_OPENING) { | |
1843 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
1844 | wait_on_context_list(session->waiting_for_open); | |
7c673cae FG |
1845 | continue; |
1846 | } | |
1847 | ||
1848 | if (!have_open_session(mds)) | |
1849 | continue; | |
1850 | } else { | |
11fdf7f2 | 1851 | session = &mds_sessions.at(mds); |
7c673cae FG |
1852 | } |
1853 | ||
1854 | // send request. | |
1855 | send_request(request, session); | |
1856 | ||
1857 | // wait for signal | |
1858 | ldout(cct, 20) << "awaiting reply|forward|kick on " << &caller_cond << dendl; | |
1859 | request->kick = false; | |
9f95a23c TL |
1860 | std::unique_lock l{client_lock, std::adopt_lock}; |
1861 | caller_cond.wait(l, [request] { | |
1862 | return (request->reply || // reply | |
1863 | request->resend_mds >= 0 || // forward | |
1864 | request->kick); | |
1865 | }); | |
1866 | l.release(); | |
1867 | request->caller_cond = nullptr; | |
7c673cae FG |
1868 | |
1869 | // did we get a reply? | |
1870 | if (request->reply) | |
1871 | break; | |
1872 | } | |
1873 | ||
1874 | if (!request->reply) { | |
11fdf7f2 TL |
1875 | ceph_assert(request->aborted()); |
1876 | ceph_assert(!request->got_unsafe); | |
7c673cae FG |
1877 | r = request->get_abort_code(); |
1878 | request->item.remove_myself(); | |
1879 | unregister_request(request); | |
11fdf7f2 | 1880 | put_request(request); |
7c673cae FG |
1881 | return r; |
1882 | } | |
1883 | ||
1884 | // got it! | |
11fdf7f2 | 1885 | auto reply = std::move(request->reply); |
7c673cae FG |
1886 | r = reply->get_result(); |
1887 | if (r >= 0) | |
1888 | request->success = true; | |
1889 | ||
1890 | // kick dispatcher (we've got it!) | |
11fdf7f2 | 1891 | ceph_assert(request->dispatch_cond); |
9f95a23c | 1892 | request->dispatch_cond->notify_all(); |
7c673cae FG |
1893 | ldout(cct, 20) << "sendrecv kickback on tid " << tid << " " << request->dispatch_cond << dendl; |
1894 | request->dispatch_cond = 0; | |
1895 | ||
1896 | if (r >= 0 && ptarget) | |
9f95a23c | 1897 | r = verify_reply_trace(r, session, request, reply, ptarget, pcreated, perms); |
7c673cae FG |
1898 | |
1899 | if (pdirbl) | |
11fdf7f2 | 1900 | *pdirbl = reply->get_extra_bl(); |
7c673cae FG |
1901 | |
1902 | // -- log times -- | |
1903 | utime_t lat = ceph_clock_now(); | |
1904 | lat -= request->sent_stamp; | |
1905 | ldout(cct, 20) << "lat " << lat << dendl; | |
1906 | logger->tinc(l_c_lat, lat); | |
1907 | logger->tinc(l_c_reply, lat); | |
1908 | ||
1909 | put_request(request); | |
7c673cae FG |
1910 | return r; |
1911 | } | |
1912 | ||
1913 | void Client::unregister_request(MetaRequest *req) | |
1914 | { | |
1915 | mds_requests.erase(req->tid); | |
1916 | if (req->tid == oldest_tid) { | |
1917 | map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.upper_bound(oldest_tid); | |
1918 | while (true) { | |
1919 | if (p == mds_requests.end()) { | |
1920 | oldest_tid = 0; | |
1921 | break; | |
1922 | } | |
1923 | if (p->second->get_op() != CEPH_MDS_OP_SETFILELOCK) { | |
1924 | oldest_tid = p->first; | |
1925 | break; | |
1926 | } | |
1927 | ++p; | |
1928 | } | |
1929 | } | |
1930 | put_request(req); | |
1931 | } | |
1932 | ||
1933 | void Client::put_request(MetaRequest *request) | |
1934 | { | |
1935 | if (request->_put()) { | |
1936 | int op = -1; | |
1937 | if (request->success) | |
1938 | op = request->get_op(); | |
1939 | InodeRef other_in; | |
1940 | request->take_other_inode(&other_in); | |
1941 | delete request; | |
1942 | ||
1943 | if (other_in && | |
1944 | (op == CEPH_MDS_OP_RMDIR || | |
1945 | op == CEPH_MDS_OP_RENAME || | |
1946 | op == CEPH_MDS_OP_RMSNAP)) { | |
1947 | _try_to_trim_inode(other_in.get(), false); | |
1948 | } | |
1949 | } | |
1950 | } | |
1951 | ||
1952 | int Client::encode_inode_release(Inode *in, MetaRequest *req, | |
1953 | mds_rank_t mds, int drop, | |
1954 | int unless, int force) | |
1955 | { | |
11fdf7f2 | 1956 | ldout(cct, 20) << __func__ << " enter(in:" << *in << ", req:" << req |
f67539c2 | 1957 | << " mds:" << mds << ", drop:" << ccap_string(drop) << ", unless:" << ccap_string(unless) |
1911f103 | 1958 | << ", force:" << force << ")" << dendl; |
7c673cae | 1959 | int released = 0; |
11fdf7f2 TL |
1960 | auto it = in->caps.find(mds); |
1961 | if (it != in->caps.end()) { | |
1962 | Cap &cap = it->second; | |
7c673cae | 1963 | drop &= ~(in->dirty_caps | get_caps_used(in)); |
11fdf7f2 TL |
1964 | if ((drop & cap.issued) && |
1965 | !(unless & cap.issued)) { | |
1911f103 | 1966 | ldout(cct, 25) << "dropping caps " << ccap_string(drop) << dendl; |
11fdf7f2 TL |
1967 | cap.issued &= ~drop; |
1968 | cap.implemented &= ~drop; | |
7c673cae | 1969 | released = 1; |
7c673cae FG |
1970 | } else { |
1971 | released = force; | |
1972 | } | |
1973 | if (released) { | |
1911f103 TL |
1974 | cap.wanted = in->caps_wanted(); |
1975 | if (&cap == in->auth_cap && | |
1976 | !(cap.wanted & CEPH_CAP_ANY_FILE_WR)) { | |
1977 | in->requested_max_size = 0; | |
1978 | ldout(cct, 25) << "reset requested_max_size due to not wanting any file write cap" << dendl; | |
1979 | } | |
7c673cae FG |
1980 | ceph_mds_request_release rel; |
1981 | rel.ino = in->ino; | |
11fdf7f2 TL |
1982 | rel.cap_id = cap.cap_id; |
1983 | rel.seq = cap.seq; | |
1984 | rel.issue_seq = cap.issue_seq; | |
1985 | rel.mseq = cap.mseq; | |
1986 | rel.caps = cap.implemented; | |
1987 | rel.wanted = cap.wanted; | |
7c673cae FG |
1988 | rel.dname_len = 0; |
1989 | rel.dname_seq = 0; | |
1990 | req->cap_releases.push_back(MClientRequest::Release(rel,"")); | |
1991 | } | |
1992 | } | |
11fdf7f2 | 1993 | ldout(cct, 25) << __func__ << " exit(in:" << *in << ") released:" |
7c673cae FG |
1994 | << released << dendl; |
1995 | return released; | |
1996 | } | |
1997 | ||
1998 | void Client::encode_dentry_release(Dentry *dn, MetaRequest *req, | |
1999 | mds_rank_t mds, int drop, int unless) | |
2000 | { | |
11fdf7f2 | 2001 | ldout(cct, 20) << __func__ << " enter(dn:" |
7c673cae FG |
2002 | << dn << ")" << dendl; |
2003 | int released = 0; | |
2004 | if (dn->dir) | |
2005 | released = encode_inode_release(dn->dir->parent_inode, req, | |
2006 | mds, drop, unless, 1); | |
2007 | if (released && dn->lease_mds == mds) { | |
2008 | ldout(cct, 25) << "preemptively releasing dn to mds" << dendl; | |
11fdf7f2 | 2009 | auto& rel = req->cap_releases.back(); |
7c673cae FG |
2010 | rel.item.dname_len = dn->name.length(); |
2011 | rel.item.dname_seq = dn->lease_seq; | |
2012 | rel.dname = dn->name; | |
adb31ebb | 2013 | dn->lease_mds = -1; |
7c673cae | 2014 | } |
11fdf7f2 | 2015 | ldout(cct, 25) << __func__ << " exit(dn:" |
7c673cae FG |
2016 | << dn << ")" << dendl; |
2017 | } | |
2018 | ||
2019 | ||
2020 | /* | |
2021 | * This requires the MClientRequest *request member to be set. | |
2022 | * It will error out horribly without one. | |
2023 | * Additionally, if you set any *drop member, you'd better have | |
2024 | * set the corresponding dentry! | |
2025 | */ | |
2026 | void Client::encode_cap_releases(MetaRequest *req, mds_rank_t mds) | |
2027 | { | |
11fdf7f2 | 2028 | ldout(cct, 20) << __func__ << " enter (req: " |
7c673cae FG |
2029 | << req << ", mds: " << mds << ")" << dendl; |
2030 | if (req->inode_drop && req->inode()) | |
2031 | encode_inode_release(req->inode(), req, | |
2032 | mds, req->inode_drop, | |
2033 | req->inode_unless); | |
2034 | ||
2035 | if (req->old_inode_drop && req->old_inode()) | |
2036 | encode_inode_release(req->old_inode(), req, | |
2037 | mds, req->old_inode_drop, | |
2038 | req->old_inode_unless); | |
2039 | if (req->other_inode_drop && req->other_inode()) | |
2040 | encode_inode_release(req->other_inode(), req, | |
2041 | mds, req->other_inode_drop, | |
2042 | req->other_inode_unless); | |
2043 | ||
2044 | if (req->dentry_drop && req->dentry()) | |
2045 | encode_dentry_release(req->dentry(), req, | |
2046 | mds, req->dentry_drop, | |
2047 | req->dentry_unless); | |
2048 | ||
2049 | if (req->old_dentry_drop && req->old_dentry()) | |
2050 | encode_dentry_release(req->old_dentry(), req, | |
2051 | mds, req->old_dentry_drop, | |
2052 | req->old_dentry_unless); | |
11fdf7f2 | 2053 | ldout(cct, 25) << __func__ << " exit (req: " |
7c673cae FG |
2054 | << req << ", mds " << mds <<dendl; |
2055 | } | |
2056 | ||
2057 | bool Client::have_open_session(mds_rank_t mds) | |
2058 | { | |
11fdf7f2 TL |
2059 | const auto &it = mds_sessions.find(mds); |
2060 | return it != mds_sessions.end() && | |
2061 | (it->second.state == MetaSession::STATE_OPEN || | |
2062 | it->second.state == MetaSession::STATE_STALE); | |
7c673cae FG |
2063 | } |
2064 | ||
2065 | MetaSession *Client::_get_mds_session(mds_rank_t mds, Connection *con) | |
2066 | { | |
11fdf7f2 TL |
2067 | const auto &it = mds_sessions.find(mds); |
2068 | if (it == mds_sessions.end() || it->second.con != con) { | |
7c673cae | 2069 | return NULL; |
11fdf7f2 TL |
2070 | } else { |
2071 | return &it->second; | |
2072 | } | |
7c673cae FG |
2073 | } |
2074 | ||
2075 | MetaSession *Client::_get_or_open_mds_session(mds_rank_t mds) | |
2076 | { | |
11fdf7f2 TL |
2077 | auto it = mds_sessions.find(mds); |
2078 | return it == mds_sessions.end() ? _open_mds_session(mds) : &it->second; | |
7c673cae FG |
2079 | } |
2080 | ||
2081 | /** | |
2082 | * Populate a map of strings with client-identifying metadata, | |
2083 | * such as the hostname. Call this once at initialization. | |
2084 | */ | |
2085 | void Client::populate_metadata(const std::string &mount_root) | |
2086 | { | |
2087 | // Hostname | |
f67539c2 TL |
2088 | #ifdef _WIN32 |
2089 | // TODO: move this to compat.h | |
2090 | char hostname[64]; | |
2091 | DWORD hostname_sz = 64; | |
2092 | GetComputerNameA(hostname, &hostname_sz); | |
2093 | metadata["hostname"] = hostname; | |
2094 | #else | |
7c673cae FG |
2095 | struct utsname u; |
2096 | int r = uname(&u); | |
2097 | if (r >= 0) { | |
2098 | metadata["hostname"] = u.nodename; | |
2099 | ldout(cct, 20) << __func__ << " read hostname '" << u.nodename << "'" << dendl; | |
2100 | } else { | |
2101 | ldout(cct, 1) << __func__ << " failed to read hostname (" << cpp_strerror(r) << ")" << dendl; | |
2102 | } | |
f67539c2 | 2103 | #endif |
7c673cae FG |
2104 | |
2105 | metadata["pid"] = stringify(getpid()); | |
2106 | ||
2107 | // Ceph entity id (the '0' in "client.0") | |
2108 | metadata["entity_id"] = cct->_conf->name.get_id(); | |
2109 | ||
2110 | // Our mount position | |
2111 | if (!mount_root.empty()) { | |
2112 | metadata["root"] = mount_root; | |
2113 | } | |
2114 | ||
2115 | // Ceph version | |
2116 | metadata["ceph_version"] = pretty_version_to_str(); | |
2117 | metadata["ceph_sha1"] = git_version_to_str(); | |
2118 | ||
2119 | // Apply any metadata from the user's configured overrides | |
2120 | std::vector<std::string> tokens; | |
2121 | get_str_vec(cct->_conf->client_metadata, ",", tokens); | |
2122 | for (const auto &i : tokens) { | |
2123 | auto eqpos = i.find("="); | |
2124 | // Throw out anything that isn't of the form "<str>=<str>" | |
2125 | if (eqpos == 0 || eqpos == std::string::npos || eqpos == i.size()) { | |
2126 | lderr(cct) << "Invalid metadata keyval pair: '" << i << "'" << dendl; | |
2127 | continue; | |
2128 | } | |
2129 | metadata[i.substr(0, eqpos)] = i.substr(eqpos + 1); | |
2130 | } | |
2131 | } | |
2132 | ||
2133 | /** | |
2134 | * Optionally add or override client metadata fields. | |
2135 | */ | |
2136 | void Client::update_metadata(std::string const &k, std::string const &v) | |
2137 | { | |
f67539c2 TL |
2138 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2139 | ceph_assert(iref_reader.is_state_satisfied()); | |
2140 | ||
2141 | std::scoped_lock l(client_lock); | |
7c673cae | 2142 | |
11fdf7f2 TL |
2143 | auto it = metadata.find(k); |
2144 | if (it != metadata.end()) { | |
7c673cae | 2145 | ldout(cct, 1) << __func__ << " warning, overriding metadata field '" << k |
11fdf7f2 | 2146 | << "' from '" << it->second << "' to '" << v << "'" << dendl; |
7c673cae FG |
2147 | } |
2148 | ||
2149 | metadata[k] = v; | |
2150 | } | |
2151 | ||
2152 | MetaSession *Client::_open_mds_session(mds_rank_t mds) | |
2153 | { | |
11fdf7f2 TL |
2154 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
2155 | auto addrs = mdsmap->get_addrs(mds); | |
2156 | auto em = mds_sessions.emplace(std::piecewise_construct, | |
2157 | std::forward_as_tuple(mds), | |
2158 | std::forward_as_tuple(mds, messenger->connect_to_mds(addrs), addrs)); | |
2159 | ceph_assert(em.second); /* not already present */ | |
2160 | MetaSession *session = &em.first->second; | |
7c673cae | 2161 | |
9f95a23c | 2162 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_OPEN); |
11fdf7f2 TL |
2163 | m->metadata = metadata; |
2164 | m->supported_features = feature_bitset_t(CEPHFS_FEATURES_CLIENT_SUPPORTED); | |
f67539c2 | 2165 | m->metric_spec = feature_bitset_t(CEPHFS_METRIC_FEATURES_ALL); |
11fdf7f2 | 2166 | session->con->send_message2(std::move(m)); |
7c673cae FG |
2167 | return session; |
2168 | } | |
2169 | ||
2170 | void Client::_close_mds_session(MetaSession *s) | |
2171 | { | |
11fdf7f2 | 2172 | ldout(cct, 2) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
7c673cae | 2173 | s->state = MetaSession::STATE_CLOSING; |
9f95a23c | 2174 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
2175 | } |
2176 | ||
f6b5b4d7 | 2177 | void Client::_closed_mds_session(MetaSession *s, int err, bool rejected) |
7c673cae | 2178 | { |
11fdf7f2 | 2179 | ldout(cct, 5) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
f6b5b4d7 TL |
2180 | if (rejected && s->state != MetaSession::STATE_CLOSING) |
2181 | s->state = MetaSession::STATE_REJECTED; | |
2182 | else | |
2183 | s->state = MetaSession::STATE_CLOSED; | |
7c673cae FG |
2184 | s->con->mark_down(); |
2185 | signal_context_list(s->waiting_for_open); | |
9f95a23c | 2186 | mount_cond.notify_all(); |
f6b5b4d7 | 2187 | remove_session_caps(s, err); |
7c673cae | 2188 | kick_requests_closed(s); |
f6b5b4d7 TL |
2189 | mds_ranks_closing.erase(s->mds_num); |
2190 | if (s->state == MetaSession::STATE_CLOSED) | |
2191 | mds_sessions.erase(s->mds_num); | |
7c673cae FG |
2192 | } |
2193 | ||
11fdf7f2 | 2194 | void Client::handle_client_session(const MConstRef<MClientSession>& m) |
7c673cae FG |
2195 | { |
2196 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
11fdf7f2 | 2197 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << from << dendl; |
7c673cae | 2198 | |
f67539c2 | 2199 | std::scoped_lock cl(client_lock); |
7c673cae FG |
2200 | MetaSession *session = _get_mds_session(from, m->get_connection().get()); |
2201 | if (!session) { | |
2202 | ldout(cct, 10) << " discarding session message from sessionless mds " << m->get_source_inst() << dendl; | |
7c673cae FG |
2203 | return; |
2204 | } | |
2205 | ||
2206 | switch (m->get_op()) { | |
2207 | case CEPH_SESSION_OPEN: | |
11fdf7f2 TL |
2208 | { |
2209 | feature_bitset_t missing_features(CEPHFS_FEATURES_CLIENT_REQUIRED); | |
2210 | missing_features -= m->supported_features; | |
2211 | if (!missing_features.empty()) { | |
2212 | lderr(cct) << "mds." << from << " lacks required features '" | |
2213 | << missing_features << "', closing session " << dendl; | |
11fdf7f2 | 2214 | _close_mds_session(session); |
f67539c2 | 2215 | _closed_mds_session(session, -CEPHFS_EPERM, true); |
11fdf7f2 TL |
2216 | break; |
2217 | } | |
2218 | session->mds_features = std::move(m->supported_features); | |
2219 | ||
2220 | renew_caps(session); | |
2221 | session->state = MetaSession::STATE_OPEN; | |
f67539c2 | 2222 | if (is_unmounting()) |
9f95a23c | 2223 | mount_cond.notify_all(); |
11fdf7f2 TL |
2224 | else |
2225 | connect_mds_targets(from); | |
2226 | signal_context_list(session->waiting_for_open); | |
2227 | break; | |
2228 | } | |
7c673cae FG |
2229 | |
2230 | case CEPH_SESSION_CLOSE: | |
2231 | _closed_mds_session(session); | |
2232 | break; | |
2233 | ||
2234 | case CEPH_SESSION_RENEWCAPS: | |
2235 | if (session->cap_renew_seq == m->get_seq()) { | |
a8e16298 | 2236 | bool was_stale = ceph_clock_now() >= session->cap_ttl; |
7c673cae FG |
2237 | session->cap_ttl = |
2238 | session->last_cap_renew_request + mdsmap->get_session_timeout(); | |
a8e16298 TL |
2239 | if (was_stale) |
2240 | wake_up_session_caps(session, false); | |
7c673cae FG |
2241 | } |
2242 | break; | |
2243 | ||
2244 | case CEPH_SESSION_STALE: | |
28e407b8 AA |
2245 | // invalidate session caps/leases |
2246 | session->cap_gen++; | |
2247 | session->cap_ttl = ceph_clock_now(); | |
2248 | session->cap_ttl -= 1; | |
7c673cae FG |
2249 | renew_caps(session); |
2250 | break; | |
2251 | ||
2252 | case CEPH_SESSION_RECALL_STATE: | |
f67539c2 TL |
2253 | /* |
2254 | * Call the renew caps and flush cap releases just before | |
2255 | * triming the caps in case the tick() won't get a chance | |
2256 | * to run them, which could cause the client to be blocklisted | |
2257 | * and MDS daemons trying to recall the caps again and | |
2258 | * again. | |
2259 | * | |
2260 | * In most cases it will do nothing, and the new cap releases | |
2261 | * added by trim_caps() followed will be deferred flushing | |
2262 | * by tick(). | |
2263 | */ | |
2264 | renew_and_flush_cap_releases(); | |
7c673cae FG |
2265 | trim_caps(session, m->get_max_caps()); |
2266 | break; | |
2267 | ||
2268 | case CEPH_SESSION_FLUSHMSG: | |
a8e16298 | 2269 | /* flush cap release */ |
11fdf7f2 TL |
2270 | if (auto& m = session->release; m) { |
2271 | session->con->send_message2(std::move(m)); | |
a8e16298 | 2272 | } |
9f95a23c | 2273 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_FLUSHMSG_ACK, m->get_seq())); |
7c673cae FG |
2274 | break; |
2275 | ||
2276 | case CEPH_SESSION_FORCE_RO: | |
2277 | force_session_readonly(session); | |
2278 | break; | |
2279 | ||
2280 | case CEPH_SESSION_REJECT: | |
11fdf7f2 TL |
2281 | { |
2282 | std::string_view error_str; | |
2283 | auto it = m->metadata.find("error_string"); | |
2284 | if (it != m->metadata.end()) | |
2285 | error_str = it->second; | |
2286 | else | |
2287 | error_str = "unknown error"; | |
2288 | lderr(cct) << "mds." << from << " rejected us (" << error_str << ")" << dendl; | |
7c673cae | 2289 | |
f67539c2 | 2290 | _closed_mds_session(session, -CEPHFS_EPERM, true); |
11fdf7f2 | 2291 | } |
7c673cae FG |
2292 | break; |
2293 | ||
2294 | default: | |
2295 | ceph_abort(); | |
2296 | } | |
7c673cae FG |
2297 | } |
2298 | ||
2299 | bool Client::_any_stale_sessions() const | |
2300 | { | |
9f95a23c | 2301 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae | 2302 | |
11fdf7f2 TL |
2303 | for (const auto &p : mds_sessions) { |
2304 | if (p.second.state == MetaSession::STATE_STALE) { | |
7c673cae FG |
2305 | return true; |
2306 | } | |
2307 | } | |
2308 | ||
2309 | return false; | |
2310 | } | |
2311 | ||
2312 | void Client::_kick_stale_sessions() | |
2313 | { | |
11fdf7f2 | 2314 | ldout(cct, 1) << __func__ << dendl; |
7c673cae | 2315 | |
11fdf7f2 TL |
2316 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
2317 | MetaSession &s = it->second; | |
f6b5b4d7 TL |
2318 | if (s.state == MetaSession::STATE_REJECTED) { |
2319 | mds_sessions.erase(it++); | |
2320 | continue; | |
2321 | } | |
11fdf7f2 TL |
2322 | ++it; |
2323 | if (s.state == MetaSession::STATE_STALE) | |
2324 | _closed_mds_session(&s); | |
7c673cae FG |
2325 | } |
2326 | } | |
2327 | ||
2328 | void Client::send_request(MetaRequest *request, MetaSession *session, | |
2329 | bool drop_cap_releases) | |
2330 | { | |
2331 | // make the request | |
2332 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2333 | ldout(cct, 10) << __func__ << " rebuilding request " << request->get_tid() |
7c673cae | 2334 | << " for mds." << mds << dendl; |
11fdf7f2 | 2335 | auto r = build_client_request(request); |
7c673cae FG |
2336 | if (request->dentry()) { |
2337 | r->set_dentry_wanted(); | |
2338 | } | |
2339 | if (request->got_unsafe) { | |
2340 | r->set_replayed_op(); | |
2341 | if (request->target) | |
2342 | r->head.ino = request->target->ino; | |
2343 | } else { | |
2344 | encode_cap_releases(request, mds); | |
2345 | if (drop_cap_releases) // we haven't send cap reconnect yet, drop cap releases | |
2346 | request->cap_releases.clear(); | |
2347 | else | |
2348 | r->releases.swap(request->cap_releases); | |
2349 | } | |
2350 | r->set_mdsmap_epoch(mdsmap->get_epoch()); | |
2351 | if (r->head.op == CEPH_MDS_OP_SETXATTR) { | |
2352 | objecter->with_osdmap([r](const OSDMap& o) { | |
2353 | r->set_osdmap_epoch(o.get_epoch()); | |
2354 | }); | |
2355 | } | |
2356 | ||
2357 | if (request->mds == -1) { | |
2358 | request->sent_stamp = ceph_clock_now(); | |
11fdf7f2 | 2359 | ldout(cct, 20) << __func__ << " set sent_stamp to " << request->sent_stamp << dendl; |
7c673cae FG |
2360 | } |
2361 | request->mds = mds; | |
2362 | ||
2363 | Inode *in = request->inode(); | |
11fdf7f2 TL |
2364 | if (in) { |
2365 | auto it = in->caps.find(mds); | |
2366 | if (it != in->caps.end()) { | |
2367 | request->sent_on_mseq = it->second.mseq; | |
2368 | } | |
2369 | } | |
7c673cae FG |
2370 | |
2371 | session->requests.push_back(&request->item); | |
2372 | ||
11fdf7f2 TL |
2373 | ldout(cct, 10) << __func__ << " " << *r << " to mds." << mds << dendl; |
2374 | session->con->send_message2(std::move(r)); | |
7c673cae FG |
2375 | } |
2376 | ||
9f95a23c | 2377 | ref_t<MClientRequest> Client::build_client_request(MetaRequest *request) |
7c673cae | 2378 | { |
9f95a23c | 2379 | auto req = make_message<MClientRequest>(request->get_op()); |
7c673cae FG |
2380 | req->set_tid(request->tid); |
2381 | req->set_stamp(request->op_stamp); | |
2382 | memcpy(&req->head, &request->head, sizeof(ceph_mds_request_head)); | |
2383 | ||
2384 | // if the filepath's haven't been set, set them! | |
2385 | if (request->path.empty()) { | |
2386 | Inode *in = request->inode(); | |
2387 | Dentry *de = request->dentry(); | |
2388 | if (in) | |
2389 | in->make_nosnap_relative_path(request->path); | |
2390 | else if (de) { | |
2391 | if (de->inode) | |
2392 | de->inode->make_nosnap_relative_path(request->path); | |
2393 | else if (de->dir) { | |
2394 | de->dir->parent_inode->make_nosnap_relative_path(request->path); | |
2395 | request->path.push_dentry(de->name); | |
2396 | } | |
2397 | else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2398 | << " No path, inode, or appropriately-endowed dentry given!" | |
2399 | << dendl; | |
2400 | } else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2401 | << " No path, inode, or dentry given!" | |
2402 | << dendl; | |
2403 | } | |
2404 | req->set_filepath(request->get_filepath()); | |
2405 | req->set_filepath2(request->get_filepath2()); | |
f67539c2 | 2406 | req->set_alternate_name(request->alternate_name); |
7c673cae FG |
2407 | req->set_data(request->data); |
2408 | req->set_retry_attempt(request->retry_attempt++); | |
2409 | req->head.num_fwd = request->num_fwd; | |
2410 | const gid_t *_gids; | |
2411 | int gid_count = request->perms.get_gids(&_gids); | |
2412 | req->set_gid_list(gid_count, _gids); | |
2413 | return req; | |
2414 | } | |
2415 | ||
2416 | ||
2417 | ||
11fdf7f2 | 2418 | void Client::handle_client_request_forward(const MConstRef<MClientRequestForward>& fwd) |
7c673cae FG |
2419 | { |
2420 | mds_rank_t mds = mds_rank_t(fwd->get_source().num()); | |
f67539c2 TL |
2421 | |
2422 | std::scoped_lock cl(client_lock); | |
7c673cae FG |
2423 | MetaSession *session = _get_mds_session(mds, fwd->get_connection().get()); |
2424 | if (!session) { | |
7c673cae FG |
2425 | return; |
2426 | } | |
2427 | ceph_tid_t tid = fwd->get_tid(); | |
2428 | ||
2429 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2430 | ldout(cct, 10) << __func__ << " no pending request on tid " << tid << dendl; |
7c673cae FG |
2431 | return; |
2432 | } | |
2433 | ||
2434 | MetaRequest *request = mds_requests[tid]; | |
11fdf7f2 | 2435 | ceph_assert(request); |
7c673cae FG |
2436 | |
2437 | // reset retry counter | |
2438 | request->retry_attempt = 0; | |
2439 | ||
2440 | // request not forwarded, or dest mds has no session. | |
2441 | // resend. | |
11fdf7f2 | 2442 | ldout(cct, 10) << __func__ << " tid " << tid |
7c673cae FG |
2443 | << " fwd " << fwd->get_num_fwd() |
2444 | << " to mds." << fwd->get_dest_mds() | |
2445 | << ", resending to " << fwd->get_dest_mds() | |
2446 | << dendl; | |
2447 | ||
2448 | request->mds = -1; | |
2449 | request->item.remove_myself(); | |
2450 | request->num_fwd = fwd->get_num_fwd(); | |
2451 | request->resend_mds = fwd->get_dest_mds(); | |
9f95a23c | 2452 | request->caller_cond->notify_all(); |
7c673cae FG |
2453 | } |
2454 | ||
2455 | bool Client::is_dir_operation(MetaRequest *req) | |
2456 | { | |
2457 | int op = req->get_op(); | |
2458 | if (op == CEPH_MDS_OP_MKNOD || op == CEPH_MDS_OP_LINK || | |
2459 | op == CEPH_MDS_OP_UNLINK || op == CEPH_MDS_OP_RENAME || | |
2460 | op == CEPH_MDS_OP_MKDIR || op == CEPH_MDS_OP_RMDIR || | |
2461 | op == CEPH_MDS_OP_SYMLINK || op == CEPH_MDS_OP_CREATE) | |
2462 | return true; | |
2463 | return false; | |
2464 | } | |
2465 | ||
11fdf7f2 | 2466 | void Client::handle_client_reply(const MConstRef<MClientReply>& reply) |
7c673cae FG |
2467 | { |
2468 | mds_rank_t mds_num = mds_rank_t(reply->get_source().num()); | |
f67539c2 TL |
2469 | |
2470 | std::scoped_lock cl(client_lock); | |
7c673cae FG |
2471 | MetaSession *session = _get_mds_session(mds_num, reply->get_connection().get()); |
2472 | if (!session) { | |
7c673cae FG |
2473 | return; |
2474 | } | |
2475 | ||
2476 | ceph_tid_t tid = reply->get_tid(); | |
2477 | bool is_safe = reply->is_safe(); | |
2478 | ||
2479 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2480 | lderr(cct) << __func__ << " no pending request on tid " << tid |
7c673cae | 2481 | << " safe is:" << is_safe << dendl; |
7c673cae FG |
2482 | return; |
2483 | } | |
2484 | MetaRequest *request = mds_requests.at(tid); | |
2485 | ||
11fdf7f2 | 2486 | ldout(cct, 20) << __func__ << " got a reply. Safe:" << is_safe |
7c673cae FG |
2487 | << " tid " << tid << dendl; |
2488 | ||
2489 | if (request->got_unsafe && !is_safe) { | |
2490 | //duplicate response | |
2491 | ldout(cct, 0) << "got a duplicate reply on tid " << tid << " from mds " | |
2492 | << mds_num << " safe:" << is_safe << dendl; | |
7c673cae FG |
2493 | return; |
2494 | } | |
2495 | ||
f67539c2 | 2496 | if (-CEPHFS_ESTALE == reply->get_result()) { // see if we can get to proper MDS |
7c673cae FG |
2497 | ldout(cct, 20) << "got ESTALE on tid " << request->tid |
2498 | << " from mds." << request->mds << dendl; | |
2499 | request->send_to_auth = true; | |
2500 | request->resend_mds = choose_target_mds(request); | |
2501 | Inode *in = request->inode(); | |
11fdf7f2 | 2502 | std::map<mds_rank_t, Cap>::const_iterator it; |
7c673cae FG |
2503 | if (request->resend_mds >= 0 && |
2504 | request->resend_mds == request->mds && | |
2505 | (in == NULL || | |
11fdf7f2 TL |
2506 | (it = in->caps.find(request->resend_mds)) != in->caps.end() || |
2507 | request->sent_on_mseq == it->second.mseq)) { | |
2508 | ldout(cct, 20) << "have to return ESTALE" << dendl; | |
7c673cae | 2509 | } else { |
9f95a23c | 2510 | request->caller_cond->notify_all(); |
7c673cae FG |
2511 | return; |
2512 | } | |
7c673cae FG |
2513 | } |
2514 | ||
11fdf7f2 | 2515 | ceph_assert(!request->reply); |
7c673cae FG |
2516 | request->reply = reply; |
2517 | insert_trace(request, session); | |
2518 | ||
2519 | // Handle unsafe reply | |
2520 | if (!is_safe) { | |
2521 | request->got_unsafe = true; | |
2522 | session->unsafe_requests.push_back(&request->unsafe_item); | |
2523 | if (is_dir_operation(request)) { | |
2524 | Inode *dir = request->inode(); | |
11fdf7f2 | 2525 | ceph_assert(dir); |
7c673cae FG |
2526 | dir->unsafe_ops.push_back(&request->unsafe_dir_item); |
2527 | } | |
2528 | if (request->target) { | |
2529 | InodeRef &in = request->target; | |
2530 | in->unsafe_ops.push_back(&request->unsafe_target_item); | |
2531 | } | |
2532 | } | |
2533 | ||
2534 | // Only signal the caller once (on the first reply): | |
2535 | // Either its an unsafe reply, or its a safe reply and no unsafe reply was sent. | |
2536 | if (!is_safe || !request->got_unsafe) { | |
9f95a23c | 2537 | ceph::condition_variable cond; |
7c673cae FG |
2538 | request->dispatch_cond = &cond; |
2539 | ||
2540 | // wake up waiter | |
11fdf7f2 | 2541 | ldout(cct, 20) << __func__ << " signalling caller " << (void*)request->caller_cond << dendl; |
9f95a23c | 2542 | request->caller_cond->notify_all(); |
7c673cae FG |
2543 | |
2544 | // wake for kick back | |
9f95a23c TL |
2545 | std::unique_lock l{client_lock, std::adopt_lock}; |
2546 | cond.wait(l, [tid, request, &cond, this] { | |
2547 | if (request->dispatch_cond) { | |
2548 | ldout(cct, 20) << "handle_client_reply awaiting kickback on tid " | |
2549 | << tid << " " << &cond << dendl; | |
2550 | } | |
2551 | return !request->dispatch_cond; | |
2552 | }); | |
2553 | l.release(); | |
7c673cae FG |
2554 | } |
2555 | ||
2556 | if (is_safe) { | |
2557 | // the filesystem change is committed to disk | |
2558 | // we're done, clean up | |
2559 | if (request->got_unsafe) { | |
2560 | request->unsafe_item.remove_myself(); | |
2561 | request->unsafe_dir_item.remove_myself(); | |
2562 | request->unsafe_target_item.remove_myself(); | |
2563 | signal_cond_list(request->waitfor_safe); | |
2564 | } | |
2565 | request->item.remove_myself(); | |
2566 | unregister_request(request); | |
2567 | } | |
f67539c2 | 2568 | if (is_unmounting()) |
9f95a23c | 2569 | mount_cond.notify_all(); |
7c673cae FG |
2570 | } |
2571 | ||
2572 | void Client::_handle_full_flag(int64_t pool) | |
2573 | { | |
2574 | ldout(cct, 1) << __func__ << ": FULL: cancelling outstanding operations " | |
2575 | << "on " << pool << dendl; | |
f67539c2 | 2576 | // Cancel all outstanding ops in this pool with -CEPHFS_ENOSPC: it is necessary |
7c673cae FG |
2577 | // to do this rather than blocking, because otherwise when we fill up we |
2578 | // potentially lock caps forever on files with dirty pages, and we need | |
2579 | // to be able to release those caps to the MDS so that it can delete files | |
2580 | // and free up space. | |
f67539c2 | 2581 | epoch_t cancelled_epoch = objecter->op_cancel_writes(-CEPHFS_ENOSPC, pool); |
7c673cae FG |
2582 | |
2583 | // For all inodes with layouts in this pool and a pending flush write op | |
2584 | // (i.e. one of the ones we will cancel), we've got to purge_set their data | |
2585 | // from ObjectCacher so that it doesn't re-issue the write in response to | |
2586 | // the ENOSPC error. | |
2587 | // Fortunately since we're cancelling everything in a given pool, we don't | |
2588 | // need to know which ops belong to which ObjectSet, we can just blow all | |
2589 | // the un-flushed cached data away and mark any dirty inodes' async_err | |
f67539c2 | 2590 | // field with -CEPHFS_ENOSPC as long as we're sure all the ops we cancelled were |
7c673cae FG |
2591 | // affecting this pool, and all the objectsets we're purging were also |
2592 | // in this pool. | |
2593 | for (unordered_map<vinodeno_t,Inode*>::iterator i = inode_map.begin(); | |
2594 | i != inode_map.end(); ++i) | |
2595 | { | |
2596 | Inode *inode = i->second; | |
2597 | if (inode->oset.dirty_or_tx | |
2598 | && (pool == -1 || inode->layout.pool_id == pool)) { | |
2599 | ldout(cct, 4) << __func__ << ": FULL: inode 0x" << std::hex << i->first << std::dec | |
2600 | << " has dirty objects, purging and setting ENOSPC" << dendl; | |
2601 | objectcacher->purge_set(&inode->oset); | |
f67539c2 | 2602 | inode->set_async_err(-CEPHFS_ENOSPC); |
7c673cae FG |
2603 | } |
2604 | } | |
2605 | ||
2606 | if (cancelled_epoch != (epoch_t)-1) { | |
2607 | set_cap_epoch_barrier(cancelled_epoch); | |
2608 | } | |
2609 | } | |
2610 | ||
11fdf7f2 | 2611 | void Client::handle_osd_map(const MConstRef<MOSDMap>& m) |
7c673cae | 2612 | { |
f67539c2 TL |
2613 | std::set<entity_addr_t> new_blocklists; |
2614 | ||
2615 | std::scoped_lock cl(client_lock); | |
2616 | objecter->consume_blocklist_events(&new_blocklists); | |
31f18b77 | 2617 | |
11fdf7f2 | 2618 | const auto myaddrs = messenger->get_myaddrs(); |
f67539c2 | 2619 | bool new_blocklist = false; |
11fdf7f2 TL |
2620 | bool prenautilus = objecter->with_osdmap( |
2621 | [&](const OSDMap& o) { | |
9f95a23c | 2622 | return o.require_osd_release < ceph_release_t::nautilus; |
11fdf7f2 | 2623 | }); |
f67539c2 | 2624 | if (!blocklisted) { |
11fdf7f2 | 2625 | for (auto a : myaddrs.v) { |
f67539c2 | 2626 | // blocklist entries are always TYPE_ANY for nautilus+ |
11fdf7f2 | 2627 | a.set_type(entity_addr_t::TYPE_ANY); |
f67539c2 TL |
2628 | if (new_blocklists.count(a)) { |
2629 | new_blocklist = true; | |
11fdf7f2 TL |
2630 | break; |
2631 | } | |
2632 | if (prenautilus) { | |
2633 | // ...except pre-nautilus, they were TYPE_LEGACY | |
2634 | a.set_type(entity_addr_t::TYPE_LEGACY); | |
f67539c2 TL |
2635 | if (new_blocklists.count(a)) { |
2636 | new_blocklist = true; | |
11fdf7f2 TL |
2637 | break; |
2638 | } | |
2639 | } | |
2640 | } | |
2641 | } | |
f67539c2 | 2642 | if (new_blocklist) { |
31f18b77 FG |
2643 | auto epoch = objecter->with_osdmap([](const OSDMap &o){ |
2644 | return o.get_epoch(); | |
2645 | }); | |
f67539c2 TL |
2646 | lderr(cct) << "I was blocklisted at osd epoch " << epoch << dendl; |
2647 | blocklisted = true; | |
31f18b77 | 2648 | |
f67539c2 | 2649 | _abort_mds_sessions(-CEPHFS_EBLOCKLISTED); |
31f18b77 FG |
2650 | |
2651 | // Since we know all our OSD ops will fail, cancel them all preemtively, | |
2652 | // so that on an unhealthy cluster we can umount promptly even if e.g. | |
2653 | // some PGs were inaccessible. | |
f67539c2 TL |
2654 | objecter->op_cancel_writes(-CEPHFS_EBLOCKLISTED); |
2655 | ||
2656 | } | |
31f18b77 | 2657 | |
f67539c2 TL |
2658 | if (blocklisted) { |
2659 | // Handle case where we were blocklisted but no longer are | |
2660 | blocklisted = objecter->with_osdmap([myaddrs](const OSDMap &o){ | |
2661 | return o.is_blocklisted(myaddrs);}); | |
31f18b77 FG |
2662 | } |
2663 | ||
f67539c2 TL |
2664 | // Always subscribe to next osdmap for blocklisted client |
2665 | // until this client is not blocklisted. | |
2666 | if (blocklisted) { | |
f64942e4 AA |
2667 | objecter->maybe_request_map(); |
2668 | } | |
2669 | ||
7c673cae FG |
2670 | if (objecter->osdmap_full_flag()) { |
2671 | _handle_full_flag(-1); | |
2672 | } else { | |
2673 | // Accumulate local list of full pools so that I can drop | |
2674 | // the objecter lock before re-entering objecter in | |
2675 | // cancel_writes | |
2676 | std::vector<int64_t> full_pools; | |
2677 | ||
2678 | objecter->with_osdmap([&full_pools](const OSDMap &o) { | |
2679 | for (const auto& kv : o.get_pools()) { | |
2680 | if (kv.second.has_flag(pg_pool_t::FLAG_FULL)) { | |
2681 | full_pools.push_back(kv.first); | |
2682 | } | |
2683 | } | |
2684 | }); | |
2685 | ||
2686 | for (auto p : full_pools) | |
2687 | _handle_full_flag(p); | |
2688 | ||
2689 | // Subscribe to subsequent maps to watch for the full flag going | |
2690 | // away. For the global full flag objecter does this for us, but | |
2691 | // it pays no attention to the per-pool full flag so in this branch | |
2692 | // we do it ourselves. | |
2693 | if (!full_pools.empty()) { | |
2694 | objecter->maybe_request_map(); | |
2695 | } | |
2696 | } | |
7c673cae FG |
2697 | } |
2698 | ||
2699 | ||
2700 | // ------------------------ | |
2701 | // incoming messages | |
2702 | ||
2703 | ||
11fdf7f2 | 2704 | bool Client::ms_dispatch2(const MessageRef &m) |
7c673cae | 2705 | { |
f67539c2 TL |
2706 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
2707 | if (!iref_reader.is_state_satisfied()) { | |
7c673cae | 2708 | ldout(cct, 10) << "inactive, discarding " << *m << dendl; |
7c673cae FG |
2709 | return true; |
2710 | } | |
2711 | ||
2712 | switch (m->get_type()) { | |
2713 | // mounting and mds sessions | |
2714 | case CEPH_MSG_MDS_MAP: | |
9f95a23c | 2715 | handle_mds_map(ref_cast<MMDSMap>(m)); |
7c673cae FG |
2716 | break; |
2717 | case CEPH_MSG_FS_MAP: | |
9f95a23c | 2718 | handle_fs_map(ref_cast<MFSMap>(m)); |
7c673cae FG |
2719 | break; |
2720 | case CEPH_MSG_FS_MAP_USER: | |
9f95a23c | 2721 | handle_fs_map_user(ref_cast<MFSMapUser>(m)); |
7c673cae FG |
2722 | break; |
2723 | case CEPH_MSG_CLIENT_SESSION: | |
9f95a23c | 2724 | handle_client_session(ref_cast<MClientSession>(m)); |
7c673cae FG |
2725 | break; |
2726 | ||
2727 | case CEPH_MSG_OSD_MAP: | |
9f95a23c | 2728 | handle_osd_map(ref_cast<MOSDMap>(m)); |
7c673cae FG |
2729 | break; |
2730 | ||
2731 | // requests | |
2732 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: | |
9f95a23c | 2733 | handle_client_request_forward(ref_cast<MClientRequestForward>(m)); |
7c673cae FG |
2734 | break; |
2735 | case CEPH_MSG_CLIENT_REPLY: | |
9f95a23c | 2736 | handle_client_reply(ref_cast<MClientReply>(m)); |
11fdf7f2 TL |
2737 | break; |
2738 | ||
2739 | // reclaim reply | |
2740 | case CEPH_MSG_CLIENT_RECLAIM_REPLY: | |
9f95a23c | 2741 | handle_client_reclaim_reply(ref_cast<MClientReclaimReply>(m)); |
7c673cae FG |
2742 | break; |
2743 | ||
2744 | case CEPH_MSG_CLIENT_SNAP: | |
9f95a23c | 2745 | handle_snap(ref_cast<MClientSnap>(m)); |
7c673cae FG |
2746 | break; |
2747 | case CEPH_MSG_CLIENT_CAPS: | |
9f95a23c | 2748 | handle_caps(ref_cast<MClientCaps>(m)); |
7c673cae FG |
2749 | break; |
2750 | case CEPH_MSG_CLIENT_LEASE: | |
9f95a23c | 2751 | handle_lease(ref_cast<MClientLease>(m)); |
7c673cae FG |
2752 | break; |
2753 | case MSG_COMMAND_REPLY: | |
2754 | if (m->get_source().type() == CEPH_ENTITY_TYPE_MDS) { | |
9f95a23c | 2755 | handle_command_reply(ref_cast<MCommandReply>(m)); |
7c673cae FG |
2756 | } else { |
2757 | return false; | |
2758 | } | |
2759 | break; | |
2760 | case CEPH_MSG_CLIENT_QUOTA: | |
9f95a23c | 2761 | handle_quota(ref_cast<MClientQuota>(m)); |
7c673cae FG |
2762 | break; |
2763 | ||
2764 | default: | |
2765 | return false; | |
2766 | } | |
2767 | ||
2768 | // unmounting? | |
f67539c2 TL |
2769 | std::scoped_lock cl(client_lock); |
2770 | if (is_unmounting()) { | |
7c673cae FG |
2771 | ldout(cct, 10) << "unmounting: trim pass, size was " << lru.lru_get_size() |
2772 | << "+" << inode_map.size() << dendl; | |
f67539c2 | 2773 | uint64_t size = lru.lru_get_size() + inode_map.size(); |
7c673cae | 2774 | trim_cache(); |
f67539c2 | 2775 | if (size > lru.lru_get_size() + inode_map.size()) { |
7c673cae | 2776 | ldout(cct, 10) << "unmounting: trim pass, cache shrank, poking unmount()" << dendl; |
9f95a23c | 2777 | mount_cond.notify_all(); |
7c673cae FG |
2778 | } else { |
2779 | ldout(cct, 10) << "unmounting: trim pass, size still " << lru.lru_get_size() | |
2780 | << "+" << inode_map.size() << dendl; | |
2781 | } | |
2782 | } | |
2783 | ||
2784 | return true; | |
2785 | } | |
2786 | ||
11fdf7f2 | 2787 | void Client::handle_fs_map(const MConstRef<MFSMap>& m) |
7c673cae | 2788 | { |
f67539c2 | 2789 | std::scoped_lock cl(client_lock); |
7c673cae | 2790 | fsmap.reset(new FSMap(m->get_fsmap())); |
7c673cae FG |
2791 | |
2792 | signal_cond_list(waiting_for_fsmap); | |
2793 | ||
2794 | monclient->sub_got("fsmap", fsmap->get_epoch()); | |
2795 | } | |
2796 | ||
11fdf7f2 | 2797 | void Client::handle_fs_map_user(const MConstRef<MFSMapUser>& m) |
7c673cae | 2798 | { |
f67539c2 | 2799 | std::scoped_lock cl(client_lock); |
7c673cae FG |
2800 | fsmap_user.reset(new FSMapUser); |
2801 | *fsmap_user = m->get_fsmap(); | |
7c673cae FG |
2802 | |
2803 | monclient->sub_got("fsmap.user", fsmap_user->get_epoch()); | |
2804 | signal_cond_list(waiting_for_fsmap); | |
2805 | } | |
2806 | ||
f67539c2 TL |
2807 | // Cancel all the commands for missing or laggy GIDs |
2808 | void Client::cancel_commands(const MDSMap& newmap) | |
7c673cae | 2809 | { |
f67539c2 | 2810 | std::vector<ceph_tid_t> cancel_ops; |
7c673cae | 2811 | |
f67539c2 | 2812 | std::scoped_lock cmd_lock(command_lock); |
7c673cae | 2813 | auto &commands = command_table.get_commands(); |
f67539c2 | 2814 | for (const auto &[tid, op] : commands) { |
7c673cae | 2815 | const mds_gid_t op_mds_gid = op.mds_gid; |
f67539c2 TL |
2816 | if (newmap.is_dne_gid(op_mds_gid) || newmap.is_laggy_gid(op_mds_gid)) { |
2817 | ldout(cct, 1) << __func__ << ": cancelling command op " << tid << dendl; | |
2818 | cancel_ops.push_back(tid); | |
7c673cae FG |
2819 | if (op.outs) { |
2820 | std::ostringstream ss; | |
2821 | ss << "MDS " << op_mds_gid << " went away"; | |
2822 | *(op.outs) = ss.str(); | |
2823 | } | |
f67539c2 TL |
2824 | /* |
2825 | * No need to make the con->mark_down under | |
2826 | * client_lock here, because the con will | |
2827 | * has its own lock. | |
2828 | */ | |
7c673cae | 2829 | op.con->mark_down(); |
f67539c2 TL |
2830 | if (op.on_finish) |
2831 | op.on_finish->complete(-CEPHFS_ETIMEDOUT); | |
7c673cae FG |
2832 | } |
2833 | } | |
2834 | ||
f67539c2 TL |
2835 | for (const auto &tid : cancel_ops) |
2836 | command_table.erase(tid); | |
2837 | } | |
2838 | ||
2839 | void Client::handle_mds_map(const MConstRef<MMDSMap>& m) | |
2840 | { | |
2841 | std::unique_lock cl(client_lock); | |
2842 | if (m->get_epoch() <= mdsmap->get_epoch()) { | |
2843 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() | |
2844 | << " is identical to or older than our " | |
2845 | << mdsmap->get_epoch() << dendl; | |
2846 | return; | |
7c673cae FG |
2847 | } |
2848 | ||
f67539c2 TL |
2849 | cl.unlock(); |
2850 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() << dendl; | |
2851 | std::unique_ptr<MDSMap> _mdsmap(new MDSMap); | |
2852 | _mdsmap->decode(m->get_encoded()); | |
2853 | cancel_commands(*_mdsmap.get()); | |
2854 | cl.lock(); | |
2855 | ||
2856 | _mdsmap.swap(mdsmap); | |
2857 | ||
7c673cae | 2858 | // reset session |
11fdf7f2 | 2859 | for (auto p = mds_sessions.begin(); p != mds_sessions.end(); ) { |
7c673cae | 2860 | mds_rank_t mds = p->first; |
11fdf7f2 | 2861 | MetaSession *session = &p->second; |
7c673cae FG |
2862 | ++p; |
2863 | ||
f67539c2 | 2864 | int oldstate = _mdsmap->get_state(mds); |
7c673cae FG |
2865 | int newstate = mdsmap->get_state(mds); |
2866 | if (!mdsmap->is_up(mds)) { | |
2867 | session->con->mark_down(); | |
11fdf7f2 | 2868 | } else if (mdsmap->get_addrs(mds) != session->addrs) { |
f67539c2 TL |
2869 | auto old_inc = _mdsmap->get_incarnation(mds); |
2870 | auto new_inc = mdsmap->get_incarnation(mds); | |
f64942e4 AA |
2871 | if (old_inc != new_inc) { |
2872 | ldout(cct, 1) << "mds incarnation changed from " | |
2873 | << old_inc << " to " << new_inc << dendl; | |
2874 | oldstate = MDSMap::STATE_NULL; | |
2875 | } | |
7c673cae | 2876 | session->con->mark_down(); |
11fdf7f2 | 2877 | session->addrs = mdsmap->get_addrs(mds); |
7c673cae FG |
2878 | // When new MDS starts to take over, notify kernel to trim unused entries |
2879 | // in its dcache/icache. Hopefully, the kernel will release some unused | |
2880 | // inodes before the new MDS enters reconnect state. | |
2881 | trim_cache_for_reconnect(session); | |
2882 | } else if (oldstate == newstate) | |
2883 | continue; // no change | |
f67539c2 | 2884 | |
7c673cae FG |
2885 | session->mds_state = newstate; |
2886 | if (newstate == MDSMap::STATE_RECONNECT) { | |
11fdf7f2 | 2887 | session->con = messenger->connect_to_mds(session->addrs); |
7c673cae | 2888 | send_reconnect(session); |
81eedcae TL |
2889 | } else if (newstate > MDSMap::STATE_RECONNECT) { |
2890 | if (oldstate < MDSMap::STATE_RECONNECT) { | |
2891 | ldout(cct, 1) << "we may miss the MDSMap::RECONNECT, close mds session ... " << dendl; | |
2892 | _closed_mds_session(session); | |
2893 | continue; | |
2894 | } | |
2895 | if (newstate >= MDSMap::STATE_ACTIVE) { | |
2896 | if (oldstate < MDSMap::STATE_ACTIVE) { | |
2897 | // kick new requests | |
2898 | kick_requests(session); | |
2899 | kick_flushing_caps(session); | |
2900 | signal_context_list(session->waiting_for_open); | |
2901 | wake_up_session_caps(session, true); | |
2902 | } | |
2903 | connect_mds_targets(mds); | |
7c673cae | 2904 | } |
7c673cae FG |
2905 | } else if (newstate == MDSMap::STATE_NULL && |
2906 | mds >= mdsmap->get_max_mds()) { | |
2907 | _closed_mds_session(session); | |
2908 | } | |
2909 | } | |
2910 | ||
2911 | // kick any waiting threads | |
2912 | signal_cond_list(waiting_for_mdsmap); | |
2913 | ||
7c673cae FG |
2914 | monclient->sub_got("mdsmap", mdsmap->get_epoch()); |
2915 | } | |
2916 | ||
2917 | void Client::send_reconnect(MetaSession *session) | |
2918 | { | |
2919 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2920 | ldout(cct, 10) << __func__ << " to mds." << mds << dendl; |
7c673cae FG |
2921 | |
2922 | // trim unused caps to reduce MDS's cache rejoin time | |
2923 | trim_cache_for_reconnect(session); | |
2924 | ||
2925 | session->readonly = false; | |
2926 | ||
11fdf7f2 | 2927 | session->release.reset(); |
7c673cae FG |
2928 | |
2929 | // reset my cap seq number | |
2930 | session->seq = 0; | |
2931 | //connect to the mds' offload targets | |
2932 | connect_mds_targets(mds); | |
2933 | //make sure unsafe requests get saved | |
2934 | resend_unsafe_requests(session); | |
2935 | ||
11fdf7f2 TL |
2936 | early_kick_flushing_caps(session); |
2937 | ||
9f95a23c | 2938 | auto m = make_message<MClientReconnect>(); |
11fdf7f2 | 2939 | bool allow_multi = session->mds_features.test(CEPHFS_FEATURE_MULTI_RECONNECT); |
7c673cae FG |
2940 | |
2941 | // i have an open session. | |
2942 | ceph::unordered_set<inodeno_t> did_snaprealm; | |
2943 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator p = inode_map.begin(); | |
2944 | p != inode_map.end(); | |
2945 | ++p) { | |
2946 | Inode *in = p->second; | |
11fdf7f2 TL |
2947 | auto it = in->caps.find(mds); |
2948 | if (it != in->caps.end()) { | |
2949 | if (allow_multi && | |
9f95a23c TL |
2950 | m->get_approx_size() >= |
2951 | static_cast<size_t>((std::numeric_limits<int>::max() >> 1))) { | |
11fdf7f2 TL |
2952 | m->mark_more(); |
2953 | session->con->send_message2(std::move(m)); | |
2954 | ||
9f95a23c | 2955 | m = make_message<MClientReconnect>(); |
11fdf7f2 TL |
2956 | } |
2957 | ||
2958 | Cap &cap = it->second; | |
7c673cae | 2959 | ldout(cct, 10) << " caps on " << p->first |
11fdf7f2 | 2960 | << " " << ccap_string(cap.issued) |
7c673cae FG |
2961 | << " wants " << ccap_string(in->caps_wanted()) |
2962 | << dendl; | |
2963 | filepath path; | |
f91f0fd5 | 2964 | in->make_short_path(path); |
7c673cae FG |
2965 | ldout(cct, 10) << " path " << path << dendl; |
2966 | ||
2967 | bufferlist flockbl; | |
2968 | _encode_filelocks(in, flockbl); | |
2969 | ||
11fdf7f2 TL |
2970 | cap.seq = 0; // reset seq. |
2971 | cap.issue_seq = 0; // reset seq. | |
2972 | cap.mseq = 0; // reset seq. | |
2973 | // cap gen should catch up with session cap_gen | |
2974 | if (cap.gen < session->cap_gen) { | |
2975 | cap.gen = session->cap_gen; | |
2976 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
2977 | } else { | |
2978 | cap.issued = cap.implemented; | |
2979 | } | |
7c673cae FG |
2980 | snapid_t snap_follows = 0; |
2981 | if (!in->cap_snaps.empty()) | |
2982 | snap_follows = in->cap_snaps.begin()->first; | |
2983 | ||
2984 | m->add_cap(p->first.ino, | |
11fdf7f2 | 2985 | cap.cap_id, |
7c673cae FG |
2986 | path.get_ino(), path.get_path(), // ino |
2987 | in->caps_wanted(), // wanted | |
11fdf7f2 | 2988 | cap.issued, // issued |
7c673cae FG |
2989 | in->snaprealm->ino, |
2990 | snap_follows, | |
2991 | flockbl); | |
2992 | ||
2993 | if (did_snaprealm.count(in->snaprealm->ino) == 0) { | |
2994 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
2995 | m->add_snaprealm(in->snaprealm->ino, in->snaprealm->seq, in->snaprealm->parent); | |
2996 | did_snaprealm.insert(in->snaprealm->ino); | |
2997 | } | |
2998 | } | |
2999 | } | |
3000 | ||
11fdf7f2 TL |
3001 | if (!allow_multi) |
3002 | m->set_encoding_version(0); // use connection features to choose encoding | |
3003 | session->con->send_message2(std::move(m)); | |
7c673cae | 3004 | |
9f95a23c | 3005 | mount_cond.notify_all(); |
11fdf7f2 TL |
3006 | |
3007 | if (session->reclaim_state == MetaSession::RECLAIMING) | |
3008 | signal_cond_list(waiting_for_reclaim); | |
7c673cae FG |
3009 | } |
3010 | ||
3011 | ||
3012 | void Client::kick_requests(MetaSession *session) | |
3013 | { | |
11fdf7f2 | 3014 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3015 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3016 | p != mds_requests.end(); | |
3017 | ++p) { | |
31f18b77 FG |
3018 | MetaRequest *req = p->second; |
3019 | if (req->got_unsafe) | |
3020 | continue; | |
3021 | if (req->aborted()) { | |
3022 | if (req->caller_cond) { | |
3023 | req->kick = true; | |
9f95a23c | 3024 | req->caller_cond->notify_all(); |
31f18b77 | 3025 | } |
7c673cae | 3026 | continue; |
31f18b77 FG |
3027 | } |
3028 | if (req->retry_attempt > 0) | |
7c673cae | 3029 | continue; // new requests only |
31f18b77 | 3030 | if (req->mds == session->mds_num) { |
7c673cae FG |
3031 | send_request(p->second, session); |
3032 | } | |
3033 | } | |
3034 | } | |
3035 | ||
3036 | void Client::resend_unsafe_requests(MetaSession *session) | |
3037 | { | |
3038 | for (xlist<MetaRequest*>::iterator iter = session->unsafe_requests.begin(); | |
3039 | !iter.end(); | |
3040 | ++iter) | |
3041 | send_request(*iter, session); | |
3042 | ||
3043 | // also re-send old requests when MDS enters reconnect stage. So that MDS can | |
3044 | // process completed requests in clientreplay stage. | |
3045 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
3046 | p != mds_requests.end(); | |
3047 | ++p) { | |
3048 | MetaRequest *req = p->second; | |
3049 | if (req->got_unsafe) | |
3050 | continue; | |
31f18b77 FG |
3051 | if (req->aborted()) |
3052 | continue; | |
7c673cae FG |
3053 | if (req->retry_attempt == 0) |
3054 | continue; // old requests only | |
3055 | if (req->mds == session->mds_num) | |
3056 | send_request(req, session, true); | |
3057 | } | |
3058 | } | |
3059 | ||
3060 | void Client::wait_unsafe_requests() | |
3061 | { | |
3062 | list<MetaRequest*> last_unsafe_reqs; | |
11fdf7f2 TL |
3063 | for (const auto &p : mds_sessions) { |
3064 | const MetaSession &s = p.second; | |
3065 | if (!s.unsafe_requests.empty()) { | |
3066 | MetaRequest *req = s.unsafe_requests.back(); | |
7c673cae FG |
3067 | req->get(); |
3068 | last_unsafe_reqs.push_back(req); | |
3069 | } | |
3070 | } | |
3071 | ||
3072 | for (list<MetaRequest*>::iterator p = last_unsafe_reqs.begin(); | |
3073 | p != last_unsafe_reqs.end(); | |
3074 | ++p) { | |
3075 | MetaRequest *req = *p; | |
3076 | if (req->unsafe_item.is_on_list()) | |
3077 | wait_on_list(req->waitfor_safe); | |
3078 | put_request(req); | |
3079 | } | |
3080 | } | |
3081 | ||
3082 | void Client::kick_requests_closed(MetaSession *session) | |
3083 | { | |
11fdf7f2 | 3084 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
3085 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
3086 | p != mds_requests.end(); ) { | |
3087 | MetaRequest *req = p->second; | |
3088 | ++p; | |
3089 | if (req->mds == session->mds_num) { | |
3090 | if (req->caller_cond) { | |
3091 | req->kick = true; | |
9f95a23c | 3092 | req->caller_cond->notify_all(); |
7c673cae FG |
3093 | } |
3094 | req->item.remove_myself(); | |
3095 | if (req->got_unsafe) { | |
11fdf7f2 | 3096 | lderr(cct) << __func__ << " removing unsafe request " << req->get_tid() << dendl; |
7c673cae | 3097 | req->unsafe_item.remove_myself(); |
eafe8130 TL |
3098 | if (is_dir_operation(req)) { |
3099 | Inode *dir = req->inode(); | |
3100 | assert(dir); | |
f67539c2 | 3101 | dir->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3102 | lderr(cct) << "kick_requests_closed drop req of inode(dir) : " |
3103 | << dir->ino << " " << req->get_tid() << dendl; | |
3104 | req->unsafe_dir_item.remove_myself(); | |
3105 | } | |
3106 | if (req->target) { | |
3107 | InodeRef &in = req->target; | |
f67539c2 | 3108 | in->set_async_err(-CEPHFS_EIO); |
eafe8130 TL |
3109 | lderr(cct) << "kick_requests_closed drop req of inode : " |
3110 | << in->ino << " " << req->get_tid() << dendl; | |
3111 | req->unsafe_target_item.remove_myself(); | |
3112 | } | |
7c673cae FG |
3113 | signal_cond_list(req->waitfor_safe); |
3114 | unregister_request(req); | |
3115 | } | |
3116 | } | |
3117 | } | |
11fdf7f2 TL |
3118 | ceph_assert(session->requests.empty()); |
3119 | ceph_assert(session->unsafe_requests.empty()); | |
7c673cae FG |
3120 | } |
3121 | ||
3122 | ||
3123 | ||
3124 | ||
3125 | /************ | |
3126 | * leases | |
3127 | */ | |
3128 | ||
3129 | void Client::got_mds_push(MetaSession *s) | |
3130 | { | |
3131 | s->seq++; | |
3132 | ldout(cct, 10) << " mds." << s->mds_num << " seq now " << s->seq << dendl; | |
3133 | if (s->state == MetaSession::STATE_CLOSING) { | |
9f95a23c | 3134 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
3135 | } |
3136 | } | |
3137 | ||
11fdf7f2 | 3138 | void Client::handle_lease(const MConstRef<MClientLease>& m) |
7c673cae | 3139 | { |
11fdf7f2 | 3140 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 3141 | |
11fdf7f2 | 3142 | ceph_assert(m->get_action() == CEPH_MDS_LEASE_REVOKE); |
7c673cae | 3143 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
3144 | |
3145 | std::scoped_lock cl(client_lock); | |
7c673cae FG |
3146 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); |
3147 | if (!session) { | |
7c673cae FG |
3148 | return; |
3149 | } | |
3150 | ||
3151 | got_mds_push(session); | |
3152 | ||
3153 | ceph_seq_t seq = m->get_seq(); | |
3154 | ||
3155 | Inode *in; | |
3156 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); | |
3157 | if (inode_map.count(vino) == 0) { | |
3158 | ldout(cct, 10) << " don't have vino " << vino << dendl; | |
3159 | goto revoke; | |
3160 | } | |
3161 | in = inode_map[vino]; | |
3162 | ||
9f95a23c | 3163 | if (m->get_mask() & CEPH_LEASE_VALID) { |
7c673cae FG |
3164 | if (!in->dir || in->dir->dentries.count(m->dname) == 0) { |
3165 | ldout(cct, 10) << " don't have dir|dentry " << m->get_ino() << "/" << m->dname <<dendl; | |
3166 | goto revoke; | |
3167 | } | |
3168 | Dentry *dn = in->dir->dentries[m->dname]; | |
3169 | ldout(cct, 10) << " revoked DN lease on " << dn << dendl; | |
3170 | dn->lease_mds = -1; | |
3171 | } | |
3172 | ||
3173 | revoke: | |
11fdf7f2 | 3174 | { |
9f95a23c TL |
3175 | auto reply = make_message<MClientLease>(CEPH_MDS_LEASE_RELEASE, seq, |
3176 | m->get_mask(), m->get_ino(), | |
3177 | m->get_first(), m->get_last(), m->dname); | |
11fdf7f2 TL |
3178 | m->get_connection()->send_message2(std::move(reply)); |
3179 | } | |
7c673cae FG |
3180 | } |
3181 | ||
f67539c2 | 3182 | void Client::_put_inode(Inode *in, int n) |
7c673cae | 3183 | { |
f67539c2 TL |
3184 | ldout(cct, 10) << __func__ << " on " << *in << " n = " << n << dendl; |
3185 | ||
b3b6e05e TL |
3186 | int left = in->get_nref(); |
3187 | ceph_assert(left >= n + 1); | |
3188 | in->iput(n); | |
3189 | left -= n; | |
3190 | if (left == 1) { // the last one will be held by the inode_map | |
7c673cae FG |
3191 | // release any caps |
3192 | remove_all_caps(in); | |
3193 | ||
11fdf7f2 | 3194 | ldout(cct, 10) << __func__ << " deleting " << *in << dendl; |
7c673cae | 3195 | bool unclean = objectcacher->release_set(&in->oset); |
11fdf7f2 | 3196 | ceph_assert(!unclean); |
7c673cae FG |
3197 | inode_map.erase(in->vino()); |
3198 | if (use_faked_inos()) | |
3199 | _release_faked_ino(in); | |
3200 | ||
b3b6e05e | 3201 | if (root == nullptr) { |
7c673cae FG |
3202 | root_ancestor = 0; |
3203 | while (!root_parents.empty()) | |
3204 | root_parents.erase(root_parents.begin()); | |
3205 | } | |
3206 | ||
b3b6e05e | 3207 | in->iput(); |
7c673cae FG |
3208 | } |
3209 | } | |
3210 | ||
f67539c2 TL |
3211 | void Client::delay_put_inodes(bool wakeup) |
3212 | { | |
3213 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
3214 | ||
3215 | std::map<Inode*,int> release; | |
3216 | { | |
3217 | std::scoped_lock dl(delay_i_lock); | |
3218 | release.swap(delay_i_release); | |
3219 | } | |
3220 | ||
3221 | if (release.empty()) | |
3222 | return; | |
3223 | ||
3224 | for (auto &[in, cnt] : release) | |
3225 | _put_inode(in, cnt); | |
3226 | ||
3227 | if (wakeup) | |
3228 | mount_cond.notify_all(); | |
3229 | } | |
3230 | ||
3231 | void Client::put_inode(Inode *in, int n) | |
3232 | { | |
3233 | ldout(cct, 20) << __func__ << " on " << *in << " n = " << n << dendl; | |
3234 | ||
3235 | std::scoped_lock dl(delay_i_lock); | |
3236 | delay_i_release[in] += n; | |
3237 | } | |
3238 | ||
7c673cae FG |
3239 | void Client::close_dir(Dir *dir) |
3240 | { | |
3241 | Inode *in = dir->parent_inode; | |
11fdf7f2 TL |
3242 | ldout(cct, 15) << __func__ << " dir " << dir << " on " << in << dendl; |
3243 | ceph_assert(dir->is_empty()); | |
3244 | ceph_assert(in->dir == dir); | |
3245 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked | |
3246 | if (!in->dentries.empty()) | |
7c673cae FG |
3247 | in->get_first_parent()->put(); // unpin dentry |
3248 | ||
3249 | delete in->dir; | |
3250 | in->dir = 0; | |
3251 | put_inode(in); // unpin inode | |
3252 | } | |
3253 | ||
3254 | /** | |
3255 | * Don't call this with in==NULL, use get_or_create for that | |
3256 | * leave dn set to default NULL unless you're trying to add | |
3257 | * a new inode to a pre-created Dentry | |
3258 | */ | |
3259 | Dentry* Client::link(Dir *dir, const string& name, Inode *in, Dentry *dn) | |
3260 | { | |
3261 | if (!dn) { | |
3262 | // create a new Dentry | |
11fdf7f2 TL |
3263 | dn = new Dentry(dir, name); |
3264 | ||
7c673cae FG |
3265 | lru.lru_insert_mid(dn); // mid or top? |
3266 | ||
3267 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in | |
3268 | << " dn " << dn << " (new dn)" << dendl; | |
3269 | } else { | |
11fdf7f2 | 3270 | ceph_assert(!dn->inode); |
7c673cae FG |
3271 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in |
3272 | << " dn " << dn << " (old dn)" << dendl; | |
3273 | } | |
3274 | ||
3275 | if (in) { // link to inode | |
11fdf7f2 | 3276 | InodeRef tmp_ref; |
7c673cae | 3277 | // only one parent for directories! |
11fdf7f2 TL |
3278 | if (in->is_dir() && !in->dentries.empty()) { |
3279 | tmp_ref = in; // prevent unlink below from freeing the inode. | |
7c673cae | 3280 | Dentry *olddn = in->get_first_parent(); |
11fdf7f2 | 3281 | ceph_assert(olddn->dir != dir || olddn->name != name); |
7c673cae | 3282 | Inode *old_diri = olddn->dir->parent_inode; |
7c673cae FG |
3283 | clear_dir_complete_and_ordered(old_diri, true); |
3284 | unlink(olddn, true, true); // keep dir, dentry | |
3285 | } | |
3286 | ||
11fdf7f2 | 3287 | dn->link(in); |
f67539c2 | 3288 | inc_dentry_nr(); |
11fdf7f2 | 3289 | ldout(cct, 20) << "link inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3290 | } |
3291 | ||
3292 | return dn; | |
3293 | } | |
3294 | ||
3295 | void Client::unlink(Dentry *dn, bool keepdir, bool keepdentry) | |
3296 | { | |
11fdf7f2 | 3297 | InodeRef in(dn->inode); |
7c673cae FG |
3298 | ldout(cct, 15) << "unlink dir " << dn->dir->parent_inode << " '" << dn->name << "' dn " << dn |
3299 | << " inode " << dn->inode << dendl; | |
3300 | ||
3301 | // unlink from inode | |
11fdf7f2 TL |
3302 | if (dn->inode) { |
3303 | dn->unlink(); | |
f67539c2 | 3304 | dec_dentry_nr(); |
11fdf7f2 | 3305 | ldout(cct, 20) << "unlink inode " << in << " parents now " << in->dentries << dendl; |
7c673cae FG |
3306 | } |
3307 | ||
3308 | if (keepdentry) { | |
3309 | dn->lease_mds = -1; | |
3310 | } else { | |
3311 | ldout(cct, 15) << "unlink removing '" << dn->name << "' dn " << dn << dendl; | |
3312 | ||
3313 | // unlink from dir | |
11fdf7f2 TL |
3314 | Dir *dir = dn->dir; |
3315 | dn->detach(); | |
7c673cae FG |
3316 | |
3317 | // delete den | |
3318 | lru.lru_remove(dn); | |
3319 | dn->put(); | |
11fdf7f2 TL |
3320 | |
3321 | if (dir->is_empty() && !keepdir) | |
3322 | close_dir(dir); | |
7c673cae FG |
3323 | } |
3324 | } | |
3325 | ||
3326 | /** | |
3327 | * For asynchronous flushes, check for errors from the IO and | |
3328 | * update the inode if necessary | |
3329 | */ | |
3330 | class C_Client_FlushComplete : public Context { | |
3331 | private: | |
3332 | Client *client; | |
3333 | InodeRef inode; | |
3334 | public: | |
3335 | C_Client_FlushComplete(Client *c, Inode *in) : client(c), inode(in) { } | |
3336 | void finish(int r) override { | |
9f95a23c | 3337 | ceph_assert(ceph_mutex_is_locked_by_me(client->client_lock)); |
7c673cae FG |
3338 | if (r != 0) { |
3339 | client_t const whoami = client->whoami; // For the benefit of ldout prefix | |
3340 | ldout(client->cct, 1) << "I/O error from flush on inode " << inode | |
3341 | << " 0x" << std::hex << inode->ino << std::dec | |
3342 | << ": " << r << "(" << cpp_strerror(r) << ")" << dendl; | |
3343 | inode->set_async_err(r); | |
3344 | } | |
3345 | } | |
3346 | }; | |
3347 | ||
3348 | ||
3349 | /**** | |
3350 | * caps | |
3351 | */ | |
3352 | ||
3353 | void Client::get_cap_ref(Inode *in, int cap) | |
3354 | { | |
3355 | if ((cap & CEPH_CAP_FILE_BUFFER) && | |
3356 | in->cap_refs[CEPH_CAP_FILE_BUFFER] == 0) { | |
11fdf7f2 | 3357 | ldout(cct, 5) << __func__ << " got first FILE_BUFFER ref on " << *in << dendl; |
b3b6e05e | 3358 | in->iget(); |
7c673cae FG |
3359 | } |
3360 | if ((cap & CEPH_CAP_FILE_CACHE) && | |
3361 | in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
11fdf7f2 | 3362 | ldout(cct, 5) << __func__ << " got first FILE_CACHE ref on " << *in << dendl; |
b3b6e05e | 3363 | in->iget(); |
7c673cae FG |
3364 | } |
3365 | in->get_cap_ref(cap); | |
3366 | } | |
3367 | ||
3368 | void Client::put_cap_ref(Inode *in, int cap) | |
3369 | { | |
3370 | int last = in->put_cap_ref(cap); | |
3371 | if (last) { | |
3372 | int put_nref = 0; | |
3373 | int drop = last & ~in->caps_issued(); | |
3374 | if (in->snapid == CEPH_NOSNAP) { | |
f67539c2 | 3375 | if ((last & (CEPH_CAP_FILE_WR | CEPH_CAP_FILE_BUFFER)) && |
7c673cae FG |
3376 | !in->cap_snaps.empty() && |
3377 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3378 | ldout(cct, 10) << __func__ << " finishing pending cap_snap on " << *in << dendl; |
7c673cae FG |
3379 | in->cap_snaps.rbegin()->second.writing = 0; |
3380 | finish_cap_snap(in, in->cap_snaps.rbegin()->second, get_caps_used(in)); | |
3381 | signal_cond_list(in->waitfor_caps); // wake up blocked sync writers | |
3382 | } | |
3383 | if (last & CEPH_CAP_FILE_BUFFER) { | |
3384 | for (auto &p : in->cap_snaps) | |
3385 | p.second.dirty_data = 0; | |
3386 | signal_cond_list(in->waitfor_commit); | |
11fdf7f2 | 3387 | ldout(cct, 5) << __func__ << " dropped last FILE_BUFFER ref on " << *in << dendl; |
7c673cae FG |
3388 | ++put_nref; |
3389 | } | |
3390 | } | |
3391 | if (last & CEPH_CAP_FILE_CACHE) { | |
11fdf7f2 | 3392 | ldout(cct, 5) << __func__ << " dropped last FILE_CACHE ref on " << *in << dendl; |
7c673cae FG |
3393 | ++put_nref; |
3394 | } | |
3395 | if (drop) | |
3396 | check_caps(in, 0); | |
3397 | if (put_nref) | |
3398 | put_inode(in, put_nref); | |
3399 | } | |
3400 | } | |
3401 | ||
f67539c2 TL |
3402 | // get caps for a given file handle -- the inode should have @need caps |
3403 | // issued by the mds and @want caps not revoked (or not under revocation). | |
3404 | // this routine blocks till the cap requirement is satisfied. also account | |
3405 | // (track) for capability hit when required (when cap requirement succeedes). | |
f6b5b4d7 | 3406 | int Client::get_caps(Fh *fh, int need, int want, int *phave, loff_t endoff) |
7c673cae | 3407 | { |
f6b5b4d7 TL |
3408 | Inode *in = fh->inode.get(); |
3409 | ||
7c673cae FG |
3410 | int r = check_pool_perm(in, need); |
3411 | if (r < 0) | |
3412 | return r; | |
3413 | ||
3414 | while (1) { | |
3415 | int file_wanted = in->caps_file_wanted(); | |
3416 | if ((file_wanted & need) != need) { | |
3417 | ldout(cct, 10) << "get_caps " << *in << " need " << ccap_string(need) | |
3418 | << " file_wanted " << ccap_string(file_wanted) << ", EBADF " | |
3419 | << dendl; | |
f67539c2 | 3420 | return -CEPHFS_EBADF; |
7c673cae FG |
3421 | } |
3422 | ||
f6b5b4d7 | 3423 | if ((fh->mode & CEPH_FILE_MODE_WR) && fh->gen != fd_gen) |
f67539c2 | 3424 | return -CEPHFS_EBADF; |
f6b5b4d7 TL |
3425 | |
3426 | if ((in->flags & I_ERROR_FILELOCK) && fh->has_any_filelocks()) | |
f67539c2 | 3427 | return -CEPHFS_EIO; |
f6b5b4d7 | 3428 | |
7c673cae FG |
3429 | int implemented; |
3430 | int have = in->caps_issued(&implemented); | |
3431 | ||
3432 | bool waitfor_caps = false; | |
3433 | bool waitfor_commit = false; | |
3434 | ||
3435 | if (have & need & CEPH_CAP_FILE_WR) { | |
1911f103 TL |
3436 | if (endoff > 0) { |
3437 | if ((endoff >= (loff_t)in->max_size || | |
3438 | endoff > (loff_t)(in->size << 1)) && | |
3439 | endoff > (loff_t)in->wanted_max_size) { | |
3440 | ldout(cct, 10) << "wanted_max_size " << in->wanted_max_size << " -> " << endoff << dendl; | |
3441 | in->wanted_max_size = endoff; | |
3442 | } | |
3443 | if (in->wanted_max_size > in->max_size && | |
3444 | in->wanted_max_size > in->requested_max_size) | |
3445 | check_caps(in, 0); | |
7c673cae FG |
3446 | } |
3447 | ||
3448 | if (endoff >= 0 && endoff > (loff_t)in->max_size) { | |
3449 | ldout(cct, 10) << "waiting on max_size, endoff " << endoff << " max_size " << in->max_size << " on " << *in << dendl; | |
3450 | waitfor_caps = true; | |
3451 | } | |
3452 | if (!in->cap_snaps.empty()) { | |
3453 | if (in->cap_snaps.rbegin()->second.writing) { | |
3454 | ldout(cct, 10) << "waiting on cap_snap write to complete" << dendl; | |
3455 | waitfor_caps = true; | |
3456 | } | |
3457 | for (auto &p : in->cap_snaps) { | |
3458 | if (p.second.dirty_data) { | |
3459 | waitfor_commit = true; | |
3460 | break; | |
3461 | } | |
3462 | } | |
3463 | if (waitfor_commit) { | |
3464 | _flush(in, new C_Client_FlushComplete(this, in)); | |
3465 | ldout(cct, 10) << "waiting for WRBUFFER to get dropped" << dendl; | |
3466 | } | |
3467 | } | |
3468 | } | |
3469 | ||
3470 | if (!waitfor_caps && !waitfor_commit) { | |
3471 | if ((have & need) == need) { | |
7c673cae FG |
3472 | int revoking = implemented & ~have; |
3473 | ldout(cct, 10) << "get_caps " << *in << " have " << ccap_string(have) | |
3474 | << " need " << ccap_string(need) << " want " << ccap_string(want) | |
c07f9fc5 | 3475 | << " revoking " << ccap_string(revoking) |
7c673cae | 3476 | << dendl; |
c07f9fc5 | 3477 | if ((revoking & want) == 0) { |
7c673cae FG |
3478 | *phave = need | (have & want); |
3479 | in->get_cap_ref(need); | |
f67539c2 | 3480 | cap_hit(); |
7c673cae FG |
3481 | return 0; |
3482 | } | |
3483 | } | |
3484 | ldout(cct, 10) << "waiting for caps " << *in << " need " << ccap_string(need) << " want " << ccap_string(want) << dendl; | |
3485 | waitfor_caps = true; | |
3486 | } | |
3487 | ||
3488 | if ((need & CEPH_CAP_FILE_WR) && in->auth_cap && | |
3489 | in->auth_cap->session->readonly) | |
f67539c2 | 3490 | return -CEPHFS_EROFS; |
7c673cae FG |
3491 | |
3492 | if (in->flags & I_CAP_DROPPED) { | |
3493 | int mds_wanted = in->caps_mds_wanted(); | |
3494 | if ((mds_wanted & need) != need) { | |
3495 | int ret = _renew_caps(in); | |
3496 | if (ret < 0) | |
3497 | return ret; | |
3498 | continue; | |
3499 | } | |
a8e16298 | 3500 | if (!(file_wanted & ~mds_wanted)) |
7c673cae | 3501 | in->flags &= ~I_CAP_DROPPED; |
7c673cae FG |
3502 | } |
3503 | ||
3504 | if (waitfor_caps) | |
3505 | wait_on_list(in->waitfor_caps); | |
3506 | else if (waitfor_commit) | |
3507 | wait_on_list(in->waitfor_commit); | |
3508 | } | |
3509 | } | |
3510 | ||
3511 | int Client::get_caps_used(Inode *in) | |
3512 | { | |
3513 | unsigned used = in->caps_used(); | |
3514 | if (!(used & CEPH_CAP_FILE_CACHE) && | |
3515 | !objectcacher->set_is_empty(&in->oset)) | |
3516 | used |= CEPH_CAP_FILE_CACHE; | |
3517 | return used; | |
3518 | } | |
3519 | ||
3520 | void Client::cap_delay_requeue(Inode *in) | |
3521 | { | |
11fdf7f2 | 3522 | ldout(cct, 10) << __func__ << " on " << *in << dendl; |
7c673cae FG |
3523 | in->hold_caps_until = ceph_clock_now(); |
3524 | in->hold_caps_until += cct->_conf->client_caps_release_delay; | |
28e407b8 | 3525 | delayed_list.push_back(&in->delay_cap_item); |
7c673cae FG |
3526 | } |
3527 | ||
3528 | void Client::send_cap(Inode *in, MetaSession *session, Cap *cap, | |
eafe8130 | 3529 | int flags, int used, int want, int retain, |
7c673cae FG |
3530 | int flush, ceph_tid_t flush_tid) |
3531 | { | |
3532 | int held = cap->issued | cap->implemented; | |
3533 | int revoking = cap->implemented & ~cap->issued; | |
3534 | retain &= ~revoking; | |
3535 | int dropping = cap->issued & ~retain; | |
3536 | int op = CEPH_CAP_OP_UPDATE; | |
3537 | ||
11fdf7f2 | 3538 | ldout(cct, 10) << __func__ << " " << *in |
7c673cae | 3539 | << " mds." << session->mds_num << " seq " << cap->seq |
7c673cae FG |
3540 | << " used " << ccap_string(used) |
3541 | << " want " << ccap_string(want) | |
3542 | << " flush " << ccap_string(flush) | |
3543 | << " retain " << ccap_string(retain) | |
3544 | << " held "<< ccap_string(held) | |
3545 | << " revoking " << ccap_string(revoking) | |
3546 | << " dropping " << ccap_string(dropping) | |
3547 | << dendl; | |
3548 | ||
3549 | if (cct->_conf->client_inject_release_failure && revoking) { | |
3550 | const int would_have_issued = cap->issued & retain; | |
3551 | const int would_have_implemented = cap->implemented & (cap->issued | used); | |
3552 | // Simulated bug: | |
3553 | // - tell the server we think issued is whatever they issued plus whatever we implemented | |
3554 | // - leave what we have implemented in place | |
3555 | ldout(cct, 20) << __func__ << " injecting failure to release caps" << dendl; | |
3556 | cap->issued = cap->issued | cap->implemented; | |
3557 | ||
3558 | // Make an exception for revoking xattr caps: we are injecting | |
3559 | // failure to release other caps, but allow xattr because client | |
3560 | // will block on xattr ops if it can't release these to MDS (#9800) | |
3561 | const int xattr_mask = CEPH_CAP_XATTR_SHARED | CEPH_CAP_XATTR_EXCL; | |
3562 | cap->issued ^= xattr_mask & revoking; | |
3563 | cap->implemented ^= xattr_mask & revoking; | |
3564 | ||
3565 | ldout(cct, 20) << __func__ << " issued " << ccap_string(cap->issued) << " vs " << ccap_string(would_have_issued) << dendl; | |
3566 | ldout(cct, 20) << __func__ << " implemented " << ccap_string(cap->implemented) << " vs " << ccap_string(would_have_implemented) << dendl; | |
3567 | } else { | |
3568 | // Normal behaviour | |
3569 | cap->issued &= retain; | |
3570 | cap->implemented &= cap->issued | used; | |
3571 | } | |
3572 | ||
3573 | snapid_t follows = 0; | |
3574 | ||
3575 | if (flush) | |
3576 | follows = in->snaprealm->get_snap_context().seq; | |
3577 | ||
9f95a23c | 3578 | auto m = make_message<MClientCaps>(op, |
7c673cae FG |
3579 | in->ino, |
3580 | 0, | |
3581 | cap->cap_id, cap->seq, | |
3582 | cap->implemented, | |
3583 | want, | |
3584 | flush, | |
3585 | cap->mseq, | |
3586 | cap_epoch_barrier); | |
3587 | m->caller_uid = in->cap_dirtier_uid; | |
3588 | m->caller_gid = in->cap_dirtier_gid; | |
3589 | ||
3590 | m->head.issue_seq = cap->issue_seq; | |
3591 | m->set_tid(flush_tid); | |
3592 | ||
3593 | m->head.uid = in->uid; | |
3594 | m->head.gid = in->gid; | |
3595 | m->head.mode = in->mode; | |
3596 | ||
3597 | m->head.nlink = in->nlink; | |
3598 | ||
3599 | if (flush & CEPH_CAP_XATTR_EXCL) { | |
11fdf7f2 | 3600 | encode(in->xattrs, m->xattrbl); |
7c673cae FG |
3601 | m->head.xattr_version = in->xattr_version; |
3602 | } | |
3603 | ||
3604 | m->size = in->size; | |
3605 | m->max_size = in->max_size; | |
3606 | m->truncate_seq = in->truncate_seq; | |
3607 | m->truncate_size = in->truncate_size; | |
3608 | m->mtime = in->mtime; | |
3609 | m->atime = in->atime; | |
3610 | m->ctime = in->ctime; | |
3611 | m->btime = in->btime; | |
3612 | m->time_warp_seq = in->time_warp_seq; | |
3613 | m->change_attr = in->change_attr; | |
eafe8130 TL |
3614 | |
3615 | if (!(flags & MClientCaps::FLAG_PENDING_CAPSNAP) && | |
3616 | !in->cap_snaps.empty() && | |
3617 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3618 | flags |= MClientCaps::FLAG_PENDING_CAPSNAP; | |
3619 | m->flags = flags; | |
3620 | ||
7c673cae FG |
3621 | if (flush & CEPH_CAP_FILE_WR) { |
3622 | m->inline_version = in->inline_version; | |
3623 | m->inline_data = in->inline_data; | |
3624 | } | |
3625 | ||
3626 | in->reported_size = in->size; | |
3627 | m->set_snap_follows(follows); | |
3628 | cap->wanted = want; | |
3629 | if (cap == in->auth_cap) { | |
1911f103 TL |
3630 | if (want & CEPH_CAP_ANY_FILE_WR) { |
3631 | m->set_max_size(in->wanted_max_size); | |
3632 | in->requested_max_size = in->wanted_max_size; | |
3633 | ldout(cct, 15) << "auth cap, requesting max_size " << in->requested_max_size << dendl; | |
3634 | } else { | |
3635 | in->requested_max_size = 0; | |
3636 | ldout(cct, 15) << "auth cap, reset requested_max_size due to not wanting any file write cap" << dendl; | |
3637 | } | |
7c673cae FG |
3638 | } |
3639 | ||
3640 | if (!session->flushing_caps_tids.empty()) | |
3641 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3642 | ||
11fdf7f2 | 3643 | session->con->send_message2(std::move(m)); |
7c673cae FG |
3644 | } |
3645 | ||
31f18b77 FG |
3646 | static bool is_max_size_approaching(Inode *in) |
3647 | { | |
3648 | /* mds will adjust max size according to the reported size */ | |
3649 | if (in->flushing_caps & CEPH_CAP_FILE_WR) | |
3650 | return false; | |
3651 | if (in->size >= in->max_size) | |
3652 | return true; | |
3653 | /* half of previous max_size increment has been used */ | |
3654 | if (in->max_size > in->reported_size && | |
3655 | (in->size << 1) >= in->max_size + in->reported_size) | |
3656 | return true; | |
3657 | return false; | |
3658 | } | |
7c673cae | 3659 | |
11fdf7f2 TL |
3660 | static int adjust_caps_used_for_lazyio(int used, int issued, int implemented) |
3661 | { | |
3662 | if (!(used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER))) | |
3663 | return used; | |
3664 | if (!(implemented & CEPH_CAP_FILE_LAZYIO)) | |
3665 | return used; | |
3666 | ||
3667 | if (issued & CEPH_CAP_FILE_LAZYIO) { | |
3668 | if (!(issued & CEPH_CAP_FILE_CACHE)) { | |
3669 | used &= ~CEPH_CAP_FILE_CACHE; | |
3670 | used |= CEPH_CAP_FILE_LAZYIO; | |
3671 | } | |
3672 | if (!(issued & CEPH_CAP_FILE_BUFFER)) { | |
3673 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3674 | used |= CEPH_CAP_FILE_LAZYIO; | |
3675 | } | |
3676 | } else { | |
3677 | if (!(implemented & CEPH_CAP_FILE_CACHE)) { | |
3678 | used &= ~CEPH_CAP_FILE_CACHE; | |
3679 | used |= CEPH_CAP_FILE_LAZYIO; | |
3680 | } | |
3681 | if (!(implemented & CEPH_CAP_FILE_BUFFER)) { | |
3682 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3683 | used |= CEPH_CAP_FILE_LAZYIO; | |
3684 | } | |
3685 | } | |
3686 | return used; | |
3687 | } | |
3688 | ||
7c673cae FG |
3689 | /** |
3690 | * check_caps | |
3691 | * | |
3692 | * Examine currently used and wanted versus held caps. Release, flush or ack | |
3693 | * revoked caps to the MDS as appropriate. | |
3694 | * | |
3695 | * @param in the inode to check | |
3696 | * @param flags flags to apply to cap check | |
3697 | */ | |
3698 | void Client::check_caps(Inode *in, unsigned flags) | |
3699 | { | |
3700 | unsigned wanted = in->caps_wanted(); | |
3701 | unsigned used = get_caps_used(in); | |
3702 | unsigned cap_used; | |
3703 | ||
7c673cae FG |
3704 | int implemented; |
3705 | int issued = in->caps_issued(&implemented); | |
3706 | int revoking = implemented & ~issued; | |
3707 | ||
11fdf7f2 TL |
3708 | int orig_used = used; |
3709 | used = adjust_caps_used_for_lazyio(used, issued, implemented); | |
3710 | ||
7c673cae | 3711 | int retain = wanted | used | CEPH_CAP_PIN; |
f67539c2 | 3712 | if (!is_unmounting() && in->nlink > 0) { |
a8e16298 | 3713 | if (wanted) { |
7c673cae | 3714 | retain |= CEPH_CAP_ANY; |
a8e16298 TL |
3715 | } else if (in->is_dir() && |
3716 | (issued & CEPH_CAP_FILE_SHARED) && | |
3717 | (in->flags & I_COMPLETE)) { | |
3718 | // we do this here because we don't want to drop to Fs (and then | |
3719 | // drop the Fs if we do a create!) if that alone makes us send lookups | |
3720 | // to the MDS. Doing it in in->caps_wanted() has knock-on effects elsewhere | |
3721 | wanted = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; | |
3722 | retain |= wanted; | |
3723 | } else { | |
7c673cae | 3724 | retain |= CEPH_CAP_ANY_SHARED; |
a8e16298 TL |
3725 | // keep RD only if we didn't have the file open RW, |
3726 | // because then the mds would revoke it anyway to | |
3727 | // journal max_size=0. | |
3728 | if (in->max_size == 0) | |
3729 | retain |= CEPH_CAP_ANY_RD; | |
3730 | } | |
7c673cae FG |
3731 | } |
3732 | ||
11fdf7f2 | 3733 | ldout(cct, 10) << __func__ << " on " << *in |
7c673cae FG |
3734 | << " wanted " << ccap_string(wanted) |
3735 | << " used " << ccap_string(used) | |
3736 | << " issued " << ccap_string(issued) | |
3737 | << " revoking " << ccap_string(revoking) | |
3738 | << " flags=" << flags | |
3739 | << dendl; | |
3740 | ||
3741 | if (in->snapid != CEPH_NOSNAP) | |
3742 | return; //snap caps last forever, can't write | |
3743 | ||
3744 | if (in->caps.empty()) | |
3745 | return; // guard if at end of func | |
3746 | ||
11fdf7f2 TL |
3747 | if (!(orig_used & CEPH_CAP_FILE_BUFFER) && |
3748 | (revoking & used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
94b18763 | 3749 | if (_release(in)) |
11fdf7f2 | 3750 | used &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
94b18763 | 3751 | } |
7c673cae | 3752 | |
7c673cae | 3753 | |
11fdf7f2 TL |
3754 | for (auto &p : in->caps) { |
3755 | mds_rank_t mds = p.first; | |
3756 | Cap &cap = p.second; | |
7c673cae | 3757 | |
11fdf7f2 | 3758 | MetaSession *session = &mds_sessions.at(mds); |
7c673cae FG |
3759 | |
3760 | cap_used = used; | |
11fdf7f2 | 3761 | if (in->auth_cap && &cap != in->auth_cap) |
7c673cae FG |
3762 | cap_used &= ~in->auth_cap->issued; |
3763 | ||
11fdf7f2 | 3764 | revoking = cap.implemented & ~cap.issued; |
7c673cae FG |
3765 | |
3766 | ldout(cct, 10) << " cap mds." << mds | |
11fdf7f2 TL |
3767 | << " issued " << ccap_string(cap.issued) |
3768 | << " implemented " << ccap_string(cap.implemented) | |
7c673cae FG |
3769 | << " revoking " << ccap_string(revoking) << dendl; |
3770 | ||
3771 | if (in->wanted_max_size > in->max_size && | |
3772 | in->wanted_max_size > in->requested_max_size && | |
11fdf7f2 | 3773 | &cap == in->auth_cap) |
7c673cae FG |
3774 | goto ack; |
3775 | ||
3776 | /* approaching file_max? */ | |
11fdf7f2 TL |
3777 | if ((cap.issued & CEPH_CAP_FILE_WR) && |
3778 | &cap == in->auth_cap && | |
31f18b77 | 3779 | is_max_size_approaching(in)) { |
7c673cae | 3780 | ldout(cct, 10) << "size " << in->size << " approaching max_size " << in->max_size |
31f18b77 | 3781 | << ", reported " << in->reported_size << dendl; |
7c673cae FG |
3782 | goto ack; |
3783 | } | |
3784 | ||
3785 | /* completed revocation? */ | |
3786 | if (revoking && (revoking & cap_used) == 0) { | |
11fdf7f2 | 3787 | ldout(cct, 10) << "completed revocation of " << ccap_string(cap.implemented & ~cap.issued) << dendl; |
7c673cae FG |
3788 | goto ack; |
3789 | } | |
3790 | ||
3791 | /* want more caps from mds? */ | |
11fdf7f2 | 3792 | if (wanted & ~(cap.wanted | cap.issued)) |
7c673cae FG |
3793 | goto ack; |
3794 | ||
f67539c2 | 3795 | if (!revoking && is_unmounting() && (cap_used == 0)) |
7c673cae FG |
3796 | goto ack; |
3797 | ||
11fdf7f2 | 3798 | if ((cap.issued & ~retain) == 0 && // and we don't have anything we wouldn't like |
a8e16298 | 3799 | !in->dirty_caps) // and we have no dirty caps |
7c673cae FG |
3800 | continue; |
3801 | ||
11fdf7f2 | 3802 | if (!(flags & CHECK_CAPS_NODELAY)) { |
7c673cae | 3803 | ldout(cct, 10) << "delaying cap release" << dendl; |
11fdf7f2 | 3804 | cap_delay_requeue(in); |
7c673cae FG |
3805 | continue; |
3806 | } | |
3807 | ||
3808 | ack: | |
eafe8130 TL |
3809 | if (&cap == in->auth_cap) { |
3810 | if (in->flags & I_KICK_FLUSH) { | |
3811 | ldout(cct, 20) << " reflushing caps (check_caps) on " << *in | |
3812 | << " to mds." << mds << dendl; | |
3813 | kick_flushing_caps(in, session); | |
3814 | } | |
3815 | if (!in->cap_snaps.empty() && | |
3816 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3817 | flush_snaps(in); | |
7c673cae FG |
3818 | } |
3819 | ||
3820 | int flushing; | |
e306af50 | 3821 | int msg_flags = 0; |
7c673cae | 3822 | ceph_tid_t flush_tid; |
11fdf7f2 | 3823 | if (in->auth_cap == &cap && in->dirty_caps) { |
7c673cae | 3824 | flushing = mark_caps_flushing(in, &flush_tid); |
e306af50 TL |
3825 | if (flags & CHECK_CAPS_SYNCHRONOUS) |
3826 | msg_flags |= MClientCaps::FLAG_SYNC; | |
7c673cae FG |
3827 | } else { |
3828 | flushing = 0; | |
3829 | flush_tid = 0; | |
3830 | } | |
3831 | ||
eafe8130 TL |
3832 | send_cap(in, session, &cap, msg_flags, cap_used, wanted, retain, |
3833 | flushing, flush_tid); | |
7c673cae FG |
3834 | } |
3835 | } | |
3836 | ||
3837 | ||
3838 | void Client::queue_cap_snap(Inode *in, SnapContext& old_snapc) | |
3839 | { | |
3840 | int used = get_caps_used(in); | |
3841 | int dirty = in->caps_dirty(); | |
11fdf7f2 | 3842 | ldout(cct, 10) << __func__ << " " << *in << " snapc " << old_snapc << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3843 | |
3844 | if (in->cap_snaps.size() && | |
3845 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3846 | ldout(cct, 10) << __func__ << " already have pending cap_snap on " << *in << dendl; |
7c673cae FG |
3847 | return; |
3848 | } else if (in->caps_dirty() || | |
3849 | (used & CEPH_CAP_FILE_WR) || | |
3850 | (dirty & CEPH_CAP_ANY_WR)) { | |
3851 | const auto &capsnapem = in->cap_snaps.emplace(std::piecewise_construct, std::make_tuple(old_snapc.seq), std::make_tuple(in)); | |
11fdf7f2 | 3852 | ceph_assert(capsnapem.second); /* element inserted */ |
7c673cae FG |
3853 | CapSnap &capsnap = capsnapem.first->second; |
3854 | capsnap.context = old_snapc; | |
3855 | capsnap.issued = in->caps_issued(); | |
3856 | capsnap.dirty = in->caps_dirty(); | |
f67539c2 | 3857 | |
7c673cae | 3858 | capsnap.dirty_data = (used & CEPH_CAP_FILE_BUFFER); |
f67539c2 | 3859 | |
7c673cae FG |
3860 | capsnap.uid = in->uid; |
3861 | capsnap.gid = in->gid; | |
3862 | capsnap.mode = in->mode; | |
3863 | capsnap.btime = in->btime; | |
3864 | capsnap.xattrs = in->xattrs; | |
3865 | capsnap.xattr_version = in->xattr_version; | |
11fdf7f2 TL |
3866 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; |
3867 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
f67539c2 | 3868 | |
7c673cae | 3869 | if (used & CEPH_CAP_FILE_WR) { |
11fdf7f2 | 3870 | ldout(cct, 10) << __func__ << " WR used on " << *in << dendl; |
7c673cae FG |
3871 | capsnap.writing = 1; |
3872 | } else { | |
3873 | finish_cap_snap(in, capsnap, used); | |
3874 | } | |
3875 | } else { | |
11fdf7f2 | 3876 | ldout(cct, 10) << __func__ << " not dirty|writing on " << *in << dendl; |
7c673cae FG |
3877 | } |
3878 | } | |
3879 | ||
3880 | void Client::finish_cap_snap(Inode *in, CapSnap &capsnap, int used) | |
3881 | { | |
11fdf7f2 | 3882 | ldout(cct, 10) << __func__ << " " << *in << " capsnap " << (void *)&capsnap << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3883 | capsnap.size = in->size; |
3884 | capsnap.mtime = in->mtime; | |
3885 | capsnap.atime = in->atime; | |
3886 | capsnap.ctime = in->ctime; | |
3887 | capsnap.time_warp_seq = in->time_warp_seq; | |
3888 | capsnap.change_attr = in->change_attr; | |
7c673cae FG |
3889 | capsnap.dirty |= in->caps_dirty(); |
3890 | ||
11fdf7f2 TL |
3891 | /* Only reset it if it wasn't set before */ |
3892 | if (capsnap.cap_dirtier_uid == -1) { | |
3893 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; | |
3894 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
3895 | } | |
3896 | ||
7c673cae FG |
3897 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { |
3898 | capsnap.inline_data = in->inline_data; | |
3899 | capsnap.inline_version = in->inline_version; | |
3900 | } | |
3901 | ||
3902 | if (used & CEPH_CAP_FILE_BUFFER) { | |
f67539c2 | 3903 | capsnap.writing = 1; |
11fdf7f2 | 3904 | ldout(cct, 10) << __func__ << " " << *in << " cap_snap " << &capsnap << " used " << used |
7c673cae FG |
3905 | << " WRBUFFER, delaying" << dendl; |
3906 | } else { | |
3907 | capsnap.dirty_data = 0; | |
3908 | flush_snaps(in); | |
3909 | } | |
3910 | } | |
3911 | ||
eafe8130 TL |
3912 | void Client::send_flush_snap(Inode *in, MetaSession *session, |
3913 | snapid_t follows, CapSnap& capsnap) | |
3914 | { | |
9f95a23c TL |
3915 | auto m = make_message<MClientCaps>(CEPH_CAP_OP_FLUSHSNAP, |
3916 | in->ino, in->snaprealm->ino, 0, | |
3917 | in->auth_cap->mseq, cap_epoch_barrier); | |
eafe8130 TL |
3918 | m->caller_uid = capsnap.cap_dirtier_uid; |
3919 | m->caller_gid = capsnap.cap_dirtier_gid; | |
3920 | ||
3921 | m->set_client_tid(capsnap.flush_tid); | |
3922 | m->head.snap_follows = follows; | |
3923 | ||
3924 | m->head.caps = capsnap.issued; | |
3925 | m->head.dirty = capsnap.dirty; | |
3926 | ||
3927 | m->head.uid = capsnap.uid; | |
3928 | m->head.gid = capsnap.gid; | |
3929 | m->head.mode = capsnap.mode; | |
3930 | m->btime = capsnap.btime; | |
3931 | ||
3932 | m->size = capsnap.size; | |
3933 | ||
3934 | m->head.xattr_version = capsnap.xattr_version; | |
3935 | encode(capsnap.xattrs, m->xattrbl); | |
3936 | ||
3937 | m->ctime = capsnap.ctime; | |
3938 | m->btime = capsnap.btime; | |
3939 | m->mtime = capsnap.mtime; | |
3940 | m->atime = capsnap.atime; | |
3941 | m->time_warp_seq = capsnap.time_warp_seq; | |
3942 | m->change_attr = capsnap.change_attr; | |
3943 | ||
3944 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { | |
3945 | m->inline_version = in->inline_version; | |
3946 | m->inline_data = in->inline_data; | |
3947 | } | |
3948 | ||
3949 | ceph_assert(!session->flushing_caps_tids.empty()); | |
3950 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3951 | ||
3952 | session->con->send_message2(std::move(m)); | |
3953 | } | |
3954 | ||
3955 | void Client::flush_snaps(Inode *in) | |
7c673cae | 3956 | { |
eafe8130 | 3957 | ldout(cct, 10) << "flush_snaps on " << *in << dendl; |
11fdf7f2 | 3958 | ceph_assert(in->cap_snaps.size()); |
7c673cae FG |
3959 | |
3960 | // pick auth mds | |
11fdf7f2 | 3961 | ceph_assert(in->auth_cap); |
7c673cae | 3962 | MetaSession *session = in->auth_cap->session; |
7c673cae FG |
3963 | |
3964 | for (auto &p : in->cap_snaps) { | |
3965 | CapSnap &capsnap = p.second; | |
eafe8130 TL |
3966 | // only do new flush |
3967 | if (capsnap.flush_tid > 0) | |
3968 | continue; | |
7c673cae FG |
3969 | |
3970 | ldout(cct, 10) << "flush_snaps mds." << session->mds_num | |
3971 | << " follows " << p.first | |
3972 | << " size " << capsnap.size | |
3973 | << " mtime " << capsnap.mtime | |
3974 | << " dirty_data=" << capsnap.dirty_data | |
3975 | << " writing=" << capsnap.writing | |
3976 | << " on " << *in << dendl; | |
3977 | if (capsnap.dirty_data || capsnap.writing) | |
eafe8130 | 3978 | break; |
f67539c2 | 3979 | |
eafe8130 TL |
3980 | capsnap.flush_tid = ++last_flush_tid; |
3981 | session->flushing_caps_tids.insert(capsnap.flush_tid); | |
3982 | in->flushing_cap_tids[capsnap.flush_tid] = 0; | |
3983 | if (!in->flushing_cap_item.is_on_list()) | |
3984 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
7c673cae | 3985 | |
eafe8130 | 3986 | send_flush_snap(in, session, p.first, capsnap); |
7c673cae FG |
3987 | } |
3988 | } | |
3989 | ||
9f95a23c | 3990 | void Client::wait_on_list(list<ceph::condition_variable*>& ls) |
7c673cae | 3991 | { |
9f95a23c | 3992 | ceph::condition_variable cond; |
7c673cae | 3993 | ls.push_back(&cond); |
9f95a23c TL |
3994 | std::unique_lock l{client_lock, std::adopt_lock}; |
3995 | cond.wait(l); | |
3996 | l.release(); | |
7c673cae FG |
3997 | ls.remove(&cond); |
3998 | } | |
3999 | ||
9f95a23c | 4000 | void Client::signal_cond_list(list<ceph::condition_variable*>& ls) |
7c673cae | 4001 | { |
9f95a23c TL |
4002 | for (auto cond : ls) { |
4003 | cond->notify_all(); | |
4004 | } | |
7c673cae FG |
4005 | } |
4006 | ||
4007 | void Client::wait_on_context_list(list<Context*>& ls) | |
4008 | { | |
9f95a23c | 4009 | ceph::condition_variable cond; |
7c673cae FG |
4010 | bool done = false; |
4011 | int r; | |
9f95a23c TL |
4012 | ls.push_back(new C_Cond(cond, &done, &r)); |
4013 | std::unique_lock l{client_lock, std::adopt_lock}; | |
4014 | cond.wait(l, [&done] { return done;}); | |
4015 | l.release(); | |
7c673cae FG |
4016 | } |
4017 | ||
4018 | void Client::signal_context_list(list<Context*>& ls) | |
4019 | { | |
4020 | while (!ls.empty()) { | |
4021 | ls.front()->complete(0); | |
4022 | ls.pop_front(); | |
4023 | } | |
4024 | } | |
4025 | ||
a8e16298 | 4026 | void Client::wake_up_session_caps(MetaSession *s, bool reconnect) |
7c673cae | 4027 | { |
11fdf7f2 TL |
4028 | for (const auto &cap : s->caps) { |
4029 | auto &in = cap->inode; | |
a8e16298 | 4030 | if (reconnect) { |
11fdf7f2 TL |
4031 | in.requested_max_size = 0; |
4032 | in.wanted_max_size = 0; | |
a8e16298 TL |
4033 | } else { |
4034 | if (cap->gen < s->cap_gen) { | |
4035 | // mds did not re-issue stale cap. | |
4036 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
4037 | // make sure mds knows what we want. | |
11fdf7f2 TL |
4038 | if (in.caps_file_wanted() & ~cap->wanted) |
4039 | in.flags |= I_CAP_DROPPED; | |
a8e16298 TL |
4040 | } |
4041 | } | |
11fdf7f2 | 4042 | signal_cond_list(in.waitfor_caps); |
7c673cae FG |
4043 | } |
4044 | } | |
4045 | ||
4046 | ||
4047 | // flush dirty data (from objectcache) | |
4048 | ||
4049 | class C_Client_CacheInvalidate : public Context { | |
4050 | private: | |
4051 | Client *client; | |
4052 | vinodeno_t ino; | |
4053 | int64_t offset, length; | |
4054 | public: | |
4055 | C_Client_CacheInvalidate(Client *c, Inode *in, int64_t off, int64_t len) : | |
4056 | client(c), offset(off), length(len) { | |
4057 | if (client->use_faked_inos()) | |
4058 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4059 | else | |
4060 | ino = in->vino(); | |
4061 | } | |
4062 | void finish(int r) override { | |
4063 | // _async_invalidate takes the lock when it needs to, call this back from outside of lock. | |
9f95a23c | 4064 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
4065 | client->_async_invalidate(ino, offset, length); |
4066 | } | |
4067 | }; | |
4068 | ||
4069 | void Client::_async_invalidate(vinodeno_t ino, int64_t off, int64_t len) | |
4070 | { | |
f67539c2 TL |
4071 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4072 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4073 | return; |
f67539c2 | 4074 | |
11fdf7f2 | 4075 | ldout(cct, 10) << __func__ << " " << ino << " " << off << "~" << len << dendl; |
7c673cae FG |
4076 | ino_invalidate_cb(callback_handle, ino, off, len); |
4077 | } | |
4078 | ||
4079 | void Client::_schedule_invalidate_callback(Inode *in, int64_t off, int64_t len) { | |
4080 | ||
4081 | if (ino_invalidate_cb) | |
4082 | // we queue the invalidate, which calls the callback and decrements the ref | |
4083 | async_ino_invalidator.queue(new C_Client_CacheInvalidate(this, in, off, len)); | |
4084 | } | |
4085 | ||
4086 | void Client::_invalidate_inode_cache(Inode *in) | |
4087 | { | |
11fdf7f2 | 4088 | ldout(cct, 10) << __func__ << " " << *in << dendl; |
7c673cae FG |
4089 | |
4090 | // invalidate our userspace inode cache | |
94b18763 | 4091 | if (cct->_conf->client_oc) { |
7c673cae | 4092 | objectcacher->release_set(&in->oset); |
94b18763 FG |
4093 | if (!objectcacher->set_is_empty(&in->oset)) |
4094 | lderr(cct) << "failed to invalidate cache for " << *in << dendl; | |
4095 | } | |
7c673cae FG |
4096 | |
4097 | _schedule_invalidate_callback(in, 0, 0); | |
4098 | } | |
4099 | ||
4100 | void Client::_invalidate_inode_cache(Inode *in, int64_t off, int64_t len) | |
4101 | { | |
11fdf7f2 | 4102 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
4103 | |
4104 | // invalidate our userspace inode cache | |
4105 | if (cct->_conf->client_oc) { | |
4106 | vector<ObjectExtent> ls; | |
4107 | Striper::file_to_extents(cct, in->ino, &in->layout, off, len, in->truncate_size, ls); | |
28e407b8 | 4108 | objectcacher->discard_writeback(&in->oset, ls, nullptr); |
7c673cae FG |
4109 | } |
4110 | ||
4111 | _schedule_invalidate_callback(in, off, len); | |
4112 | } | |
4113 | ||
4114 | bool Client::_release(Inode *in) | |
4115 | { | |
4116 | ldout(cct, 20) << "_release " << *in << dendl; | |
4117 | if (in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
4118 | _invalidate_inode_cache(in); | |
4119 | return true; | |
4120 | } | |
4121 | return false; | |
4122 | } | |
4123 | ||
4124 | bool Client::_flush(Inode *in, Context *onfinish) | |
4125 | { | |
4126 | ldout(cct, 10) << "_flush " << *in << dendl; | |
4127 | ||
4128 | if (!in->oset.dirty_or_tx) { | |
4129 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4130 | onfinish->complete(0); | |
4131 | return true; | |
4132 | } | |
4133 | ||
4134 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
1adf2230 | 4135 | ldout(cct, 8) << __func__ << ": FULL, purging for ENOSPC" << dendl; |
7c673cae FG |
4136 | objectcacher->purge_set(&in->oset); |
4137 | if (onfinish) { | |
f67539c2 | 4138 | onfinish->complete(-CEPHFS_ENOSPC); |
7c673cae FG |
4139 | } |
4140 | return true; | |
4141 | } | |
4142 | ||
4143 | return objectcacher->flush_set(&in->oset, onfinish); | |
4144 | } | |
4145 | ||
4146 | void Client::_flush_range(Inode *in, int64_t offset, uint64_t size) | |
4147 | { | |
f67539c2 | 4148 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
4149 | if (!in->oset.dirty_or_tx) { |
4150 | ldout(cct, 10) << " nothing to flush" << dendl; | |
4151 | return; | |
4152 | } | |
4153 | ||
11fdf7f2 | 4154 | C_SaferCond onflush("Client::_flush_range flock"); |
7c673cae | 4155 | bool ret = objectcacher->file_flush(&in->oset, &in->layout, in->snaprealm->get_snap_context(), |
11fdf7f2 | 4156 | offset, size, &onflush); |
7c673cae FG |
4157 | if (!ret) { |
4158 | // wait for flush | |
9f95a23c | 4159 | client_lock.unlock(); |
11fdf7f2 | 4160 | onflush.wait(); |
9f95a23c | 4161 | client_lock.lock(); |
7c673cae FG |
4162 | } |
4163 | } | |
4164 | ||
4165 | void Client::flush_set_callback(ObjectCacher::ObjectSet *oset) | |
4166 | { | |
f67539c2 TL |
4167 | // std::scoped_lock l(client_lock); |
4168 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); // will be called via dispatch() -> objecter -> ... | |
7c673cae | 4169 | Inode *in = static_cast<Inode *>(oset->parent); |
11fdf7f2 | 4170 | ceph_assert(in); |
7c673cae FG |
4171 | _flushed(in); |
4172 | } | |
4173 | ||
4174 | void Client::_flushed(Inode *in) | |
4175 | { | |
4176 | ldout(cct, 10) << "_flushed " << *in << dendl; | |
4177 | ||
4178 | put_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
4179 | } | |
4180 | ||
4181 | ||
4182 | ||
4183 | // checks common to add_update_cap, handle_cap_grant | |
11fdf7f2 | 4184 | void Client::check_cap_issue(Inode *in, unsigned issued) |
7c673cae FG |
4185 | { |
4186 | unsigned had = in->caps_issued(); | |
4187 | ||
4188 | if ((issued & CEPH_CAP_FILE_CACHE) && | |
4189 | !(had & CEPH_CAP_FILE_CACHE)) | |
4190 | in->cache_gen++; | |
4191 | ||
f91f0fd5 TL |
4192 | if ((issued & CEPH_CAP_FILE_SHARED) != |
4193 | (had & CEPH_CAP_FILE_SHARED)) { | |
4194 | if (issued & CEPH_CAP_FILE_SHARED) | |
4195 | in->shared_gen++; | |
7c673cae FG |
4196 | if (in->is_dir()) |
4197 | clear_dir_complete_and_ordered(in, true); | |
4198 | } | |
4199 | } | |
4200 | ||
4201 | void Client::add_update_cap(Inode *in, MetaSession *mds_session, uint64_t cap_id, | |
a8e16298 TL |
4202 | unsigned issued, unsigned wanted, unsigned seq, unsigned mseq, |
4203 | inodeno_t realm, int flags, const UserPerm& cap_perms) | |
7c673cae | 4204 | { |
11fdf7f2 TL |
4205 | if (!in->is_any_caps()) { |
4206 | ceph_assert(in->snaprealm == 0); | |
4207 | in->snaprealm = get_snap_realm(realm); | |
4208 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4209 | ldout(cct, 15) << __func__ << " first one, opened snaprealm " << in->snaprealm << dendl; | |
4210 | } else { | |
4211 | ceph_assert(in->snaprealm); | |
4212 | if ((flags & CEPH_CAP_FLAG_AUTH) && | |
4213 | realm != inodeno_t(-1) && in->snaprealm->ino != realm) { | |
4214 | in->snaprealm_item.remove_myself(); | |
4215 | auto oldrealm = in->snaprealm; | |
4216 | in->snaprealm = get_snap_realm(realm); | |
4217 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4218 | put_snap_realm(oldrealm); | |
4219 | } | |
4220 | } | |
4221 | ||
7c673cae | 4222 | mds_rank_t mds = mds_session->mds_num; |
11fdf7f2 TL |
4223 | const auto &capem = in->caps.emplace(std::piecewise_construct, std::forward_as_tuple(mds), std::forward_as_tuple(*in, mds_session)); |
4224 | Cap &cap = capem.first->second; | |
4225 | if (!capem.second) { | |
4226 | if (cap.gen < mds_session->cap_gen) | |
4227 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
7c673cae FG |
4228 | |
4229 | /* | |
4230 | * auth mds of the inode changed. we received the cap export | |
4231 | * message, but still haven't received the cap import message. | |
4232 | * handle_cap_export() updated the new auth MDS' cap. | |
4233 | * | |
4234 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
4235 | * a message that was send before the cap import message. So | |
4236 | * don't remove caps. | |
4237 | */ | |
11fdf7f2 | 4238 | if (ceph_seq_cmp(seq, cap.seq) <= 0) { |
92f5a8d4 TL |
4239 | if (&cap != in->auth_cap) |
4240 | ldout(cct, 0) << "WARNING: " << "inode " << *in << " caps on mds." << mds << " != auth_cap." << dendl; | |
4241 | ||
11fdf7f2 TL |
4242 | ceph_assert(cap.cap_id == cap_id); |
4243 | seq = cap.seq; | |
4244 | mseq = cap.mseq; | |
4245 | issued |= cap.issued; | |
7c673cae FG |
4246 | flags |= CEPH_CAP_FLAG_AUTH; |
4247 | } | |
f67539c2 TL |
4248 | } else { |
4249 | inc_pinned_icaps(); | |
7c673cae FG |
4250 | } |
4251 | ||
11fdf7f2 | 4252 | check_cap_issue(in, issued); |
7c673cae FG |
4253 | |
4254 | if (flags & CEPH_CAP_FLAG_AUTH) { | |
11fdf7f2 | 4255 | if (in->auth_cap != &cap && |
7c673cae FG |
4256 | (!in->auth_cap || ceph_seq_cmp(in->auth_cap->mseq, mseq) < 0)) { |
4257 | if (in->auth_cap && in->flushing_cap_item.is_on_list()) { | |
11fdf7f2 | 4258 | ldout(cct, 10) << __func__ << " changing auth cap: " |
7c673cae FG |
4259 | << "add myself to new auth MDS' flushing caps list" << dendl; |
4260 | adjust_session_flushing_caps(in, in->auth_cap->session, mds_session); | |
4261 | } | |
11fdf7f2 | 4262 | in->auth_cap = ∩ |
7c673cae FG |
4263 | } |
4264 | } | |
4265 | ||
11fdf7f2 TL |
4266 | unsigned old_caps = cap.issued; |
4267 | cap.cap_id = cap_id; | |
4268 | cap.issued = issued; | |
4269 | cap.implemented |= issued; | |
4270 | if (ceph_seq_cmp(mseq, cap.mseq) > 0) | |
4271 | cap.wanted = wanted; | |
a8e16298 | 4272 | else |
11fdf7f2 TL |
4273 | cap.wanted |= wanted; |
4274 | cap.seq = seq; | |
4275 | cap.issue_seq = seq; | |
4276 | cap.mseq = mseq; | |
4277 | cap.gen = mds_session->cap_gen; | |
4278 | cap.latest_perms = cap_perms; | |
4279 | ldout(cct, 10) << __func__ << " issued " << ccap_string(old_caps) << " -> " << ccap_string(cap.issued) | |
4280 | << " from mds." << mds | |
4281 | << " on " << *in | |
4282 | << dendl; | |
4283 | ||
4284 | if ((issued & ~old_caps) && in->auth_cap == &cap) { | |
7c673cae | 4285 | // non-auth MDS is revoking the newly grant caps ? |
11fdf7f2 TL |
4286 | for (auto &p : in->caps) { |
4287 | if (&p.second == &cap) | |
7c673cae | 4288 | continue; |
11fdf7f2 | 4289 | if (p.second.implemented & ~p.second.issued & issued) { |
7c673cae FG |
4290 | check_caps(in, CHECK_CAPS_NODELAY); |
4291 | break; | |
4292 | } | |
4293 | } | |
4294 | } | |
4295 | ||
4296 | if (issued & ~old_caps) | |
4297 | signal_cond_list(in->waitfor_caps); | |
4298 | } | |
4299 | ||
4300 | void Client::remove_cap(Cap *cap, bool queue_release) | |
4301 | { | |
11fdf7f2 | 4302 | auto &in = cap->inode; |
7c673cae FG |
4303 | MetaSession *session = cap->session; |
4304 | mds_rank_t mds = cap->session->mds_num; | |
4305 | ||
11fdf7f2 | 4306 | ldout(cct, 10) << __func__ << " mds." << mds << " on " << in << dendl; |
7c673cae FG |
4307 | |
4308 | if (queue_release) { | |
4309 | session->enqueue_cap_release( | |
11fdf7f2 | 4310 | in.ino, |
7c673cae FG |
4311 | cap->cap_id, |
4312 | cap->issue_seq, | |
4313 | cap->mseq, | |
4314 | cap_epoch_barrier); | |
f67539c2 TL |
4315 | } else { |
4316 | dec_pinned_icaps(); | |
7c673cae FG |
4317 | } |
4318 | ||
f67539c2 | 4319 | |
11fdf7f2 TL |
4320 | if (in.auth_cap == cap) { |
4321 | if (in.flushing_cap_item.is_on_list()) { | |
7c673cae | 4322 | ldout(cct, 10) << " removing myself from flushing_cap list" << dendl; |
11fdf7f2 | 4323 | in.flushing_cap_item.remove_myself(); |
7c673cae | 4324 | } |
11fdf7f2 | 4325 | in.auth_cap = NULL; |
7c673cae | 4326 | } |
11fdf7f2 TL |
4327 | size_t n = in.caps.erase(mds); |
4328 | ceph_assert(n == 1); | |
7c673cae FG |
4329 | cap = nullptr; |
4330 | ||
11fdf7f2 TL |
4331 | if (!in.is_any_caps()) { |
4332 | ldout(cct, 15) << __func__ << " last one, closing snaprealm " << in.snaprealm << dendl; | |
4333 | in.snaprealm_item.remove_myself(); | |
4334 | put_snap_realm(in.snaprealm); | |
4335 | in.snaprealm = 0; | |
7c673cae FG |
4336 | } |
4337 | } | |
4338 | ||
4339 | void Client::remove_all_caps(Inode *in) | |
4340 | { | |
4341 | while (!in->caps.empty()) | |
11fdf7f2 | 4342 | remove_cap(&in->caps.begin()->second, true); |
7c673cae FG |
4343 | } |
4344 | ||
f6b5b4d7 | 4345 | void Client::remove_session_caps(MetaSession *s, int err) |
7c673cae | 4346 | { |
11fdf7f2 | 4347 | ldout(cct, 10) << __func__ << " mds." << s->mds_num << dendl; |
7c673cae FG |
4348 | |
4349 | while (s->caps.size()) { | |
4350 | Cap *cap = *s->caps.begin(); | |
11fdf7f2 | 4351 | InodeRef in(&cap->inode); |
eafe8130 | 4352 | bool dirty_caps = false; |
7c673cae | 4353 | if (in->auth_cap == cap) { |
7c673cae FG |
4354 | dirty_caps = in->dirty_caps | in->flushing_caps; |
4355 | in->wanted_max_size = 0; | |
4356 | in->requested_max_size = 0; | |
f6b5b4d7 TL |
4357 | if (in->has_any_filelocks()) |
4358 | in->flags |= I_ERROR_FILELOCK; | |
7c673cae | 4359 | } |
f6b5b4d7 | 4360 | auto caps = cap->implemented; |
a8e16298 TL |
4361 | if (cap->wanted | cap->issued) |
4362 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 4363 | remove_cap(cap, false); |
eafe8130 | 4364 | in->cap_snaps.clear(); |
7c673cae | 4365 | if (dirty_caps) { |
11fdf7f2 | 4366 | lderr(cct) << __func__ << " still has dirty|flushing caps on " << *in << dendl; |
7c673cae FG |
4367 | if (in->flushing_caps) { |
4368 | num_flushing_caps--; | |
4369 | in->flushing_cap_tids.clear(); | |
4370 | } | |
4371 | in->flushing_caps = 0; | |
28e407b8 | 4372 | in->mark_caps_clean(); |
11fdf7f2 | 4373 | put_inode(in.get()); |
7c673cae | 4374 | } |
f6b5b4d7 TL |
4375 | caps &= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER; |
4376 | if (caps && !in->caps_issued_mask(caps, true)) { | |
f67539c2 | 4377 | if (err == -CEPHFS_EBLOCKLISTED) { |
f6b5b4d7 TL |
4378 | if (in->oset.dirty_or_tx) { |
4379 | lderr(cct) << __func__ << " still has dirty data on " << *in << dendl; | |
4380 | in->set_async_err(err); | |
4381 | } | |
4382 | objectcacher->purge_set(&in->oset); | |
4383 | } else { | |
4384 | objectcacher->release_set(&in->oset); | |
4385 | } | |
4386 | _schedule_invalidate_callback(in.get(), 0, 0); | |
4387 | } | |
4388 | ||
a8e16298 | 4389 | signal_cond_list(in->waitfor_caps); |
7c673cae FG |
4390 | } |
4391 | s->flushing_caps_tids.clear(); | |
9f95a23c | 4392 | sync_cond.notify_all(); |
7c673cae FG |
4393 | } |
4394 | ||
91327a77 | 4395 | int Client::_do_remount(bool retry_on_error) |
b32b8144 | 4396 | { |
adb31ebb | 4397 | uint64_t max_retries = cct->_conf.get_val<uint64_t>("mds_max_retries_on_remount_failure"); |
91327a77 | 4398 | |
b32b8144 FG |
4399 | errno = 0; |
4400 | int r = remount_cb(callback_handle); | |
91327a77 AA |
4401 | if (r == 0) { |
4402 | retries_on_invalidate = 0; | |
4403 | } else { | |
b32b8144 FG |
4404 | int e = errno; |
4405 | client_t whoami = get_nodeid(); | |
4406 | if (r == -1) { | |
4407 | lderr(cct) << | |
4408 | "failed to remount (to trim kernel dentries): " | |
4409 | "errno = " << e << " (" << strerror(e) << ")" << dendl; | |
4410 | } else { | |
4411 | lderr(cct) << | |
4412 | "failed to remount (to trim kernel dentries): " | |
4413 | "return code = " << r << dendl; | |
4414 | } | |
91327a77 | 4415 | bool should_abort = |
11fdf7f2 TL |
4416 | (cct->_conf.get_val<bool>("client_die_on_failed_remount") || |
4417 | cct->_conf.get_val<bool>("client_die_on_failed_dentry_invalidate")) && | |
91327a77 | 4418 | !(retry_on_error && (++retries_on_invalidate < max_retries)); |
f67539c2 | 4419 | if (should_abort && !is_unmounting()) { |
b32b8144 FG |
4420 | lderr(cct) << "failed to remount for kernel dentry trimming; quitting!" << dendl; |
4421 | ceph_abort(); | |
4422 | } | |
4423 | } | |
4424 | return r; | |
4425 | } | |
4426 | ||
7c673cae FG |
4427 | class C_Client_Remount : public Context { |
4428 | private: | |
4429 | Client *client; | |
4430 | public: | |
4431 | explicit C_Client_Remount(Client *c) : client(c) {} | |
4432 | void finish(int r) override { | |
11fdf7f2 | 4433 | ceph_assert(r == 0); |
91327a77 | 4434 | client->_do_remount(true); |
7c673cae FG |
4435 | } |
4436 | }; | |
4437 | ||
4438 | void Client::_invalidate_kernel_dcache() | |
4439 | { | |
f67539c2 TL |
4440 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4441 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 4442 | return; |
f67539c2 | 4443 | |
94b18763 FG |
4444 | if (can_invalidate_dentries) { |
4445 | if (dentry_invalidate_cb && root->dir) { | |
4446 | for (ceph::unordered_map<string, Dentry*>::iterator p = root->dir->dentries.begin(); | |
4447 | p != root->dir->dentries.end(); | |
4448 | ++p) { | |
4449 | if (p->second->inode) | |
4450 | _schedule_invalidate_dentry_callback(p->second, false); | |
4451 | } | |
7c673cae FG |
4452 | } |
4453 | } else if (remount_cb) { | |
4454 | // Hacky: | |
4455 | // when remounting a file system, linux kernel trims all unused dentries in the fs | |
4456 | remount_finisher.queue(new C_Client_Remount(this)); | |
4457 | } | |
4458 | } | |
4459 | ||
91327a77 AA |
4460 | void Client::_trim_negative_child_dentries(InodeRef& in) |
4461 | { | |
4462 | if (!in->is_dir()) | |
4463 | return; | |
4464 | ||
4465 | Dir* dir = in->dir; | |
4466 | if (dir && dir->dentries.size() == dir->num_null_dentries) { | |
4467 | for (auto p = dir->dentries.begin(); p != dir->dentries.end(); ) { | |
4468 | Dentry *dn = p->second; | |
4469 | ++p; | |
11fdf7f2 | 4470 | ceph_assert(!dn->inode); |
91327a77 AA |
4471 | if (dn->lru_is_expireable()) |
4472 | unlink(dn, true, false); // keep dir, drop dentry | |
4473 | } | |
4474 | if (dir->dentries.empty()) { | |
4475 | close_dir(dir); | |
4476 | } | |
4477 | } | |
4478 | ||
4479 | if (in->flags & I_SNAPDIR_OPEN) { | |
4480 | InodeRef snapdir = open_snapdir(in.get()); | |
4481 | _trim_negative_child_dentries(snapdir); | |
4482 | } | |
4483 | } | |
4484 | ||
e306af50 TL |
4485 | class C_Client_CacheRelease : public Context { |
4486 | private: | |
4487 | Client *client; | |
4488 | vinodeno_t ino; | |
4489 | public: | |
4490 | C_Client_CacheRelease(Client *c, Inode *in) : | |
4491 | client(c) { | |
4492 | if (client->use_faked_inos()) | |
4493 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
4494 | else | |
4495 | ino = in->vino(); | |
4496 | } | |
4497 | void finish(int r) override { | |
4498 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); | |
4499 | client->_async_inode_release(ino); | |
4500 | } | |
4501 | }; | |
4502 | ||
4503 | void Client::_async_inode_release(vinodeno_t ino) | |
4504 | { | |
f67539c2 TL |
4505 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
4506 | if (!mref_reader.is_state_satisfied()) | |
e306af50 | 4507 | return; |
f67539c2 | 4508 | |
e306af50 TL |
4509 | ldout(cct, 10) << __func__ << " " << ino << dendl; |
4510 | ino_release_cb(callback_handle, ino); | |
4511 | } | |
4512 | ||
4513 | void Client::_schedule_ino_release_callback(Inode *in) { | |
4514 | ||
4515 | if (ino_release_cb) | |
4516 | // we queue the invalidate, which calls the callback and decrements the ref | |
4517 | async_ino_releasor.queue(new C_Client_CacheRelease(this, in)); | |
4518 | } | |
4519 | ||
28e407b8 | 4520 | void Client::trim_caps(MetaSession *s, uint64_t max) |
7c673cae FG |
4521 | { |
4522 | mds_rank_t mds = s->mds_num; | |
28e407b8 | 4523 | size_t caps_size = s->caps.size(); |
11fdf7f2 | 4524 | ldout(cct, 10) << __func__ << " mds." << mds << " max " << max |
7c673cae FG |
4525 | << " caps " << caps_size << dendl; |
4526 | ||
28e407b8 AA |
4527 | uint64_t trimmed = 0; |
4528 | auto p = s->caps.begin(); | |
4529 | std::set<Dentry *> to_trim; /* this avoids caps other than the one we're | |
4530 | * looking at from getting deleted during traversal. */ | |
7c673cae FG |
4531 | while ((caps_size - trimmed) > max && !p.end()) { |
4532 | Cap *cap = *p; | |
11fdf7f2 | 4533 | InodeRef in(&cap->inode); |
7c673cae FG |
4534 | |
4535 | // Increment p early because it will be invalidated if cap | |
4536 | // is deleted inside remove_cap | |
4537 | ++p; | |
4538 | ||
4539 | if (in->caps.size() > 1 && cap != in->auth_cap) { | |
4540 | int mine = cap->issued | cap->implemented; | |
4541 | int oissued = in->auth_cap ? in->auth_cap->issued : 0; | |
4542 | // disposable non-auth cap | |
b32b8144 | 4543 | if (!(get_caps_used(in.get()) & ~oissued & mine)) { |
7c673cae | 4544 | ldout(cct, 20) << " removing unused, unneeded non-auth cap on " << *in << dendl; |
28e407b8 | 4545 | cap = (remove_cap(cap, true), nullptr); |
7c673cae FG |
4546 | trimmed++; |
4547 | } | |
4548 | } else { | |
4549 | ldout(cct, 20) << " trying to trim dentries for " << *in << dendl; | |
91327a77 | 4550 | _trim_negative_child_dentries(in); |
7c673cae | 4551 | bool all = true; |
11fdf7f2 TL |
4552 | auto q = in->dentries.begin(); |
4553 | while (q != in->dentries.end()) { | |
4554 | Dentry *dn = *q; | |
4555 | ++q; | |
7c673cae FG |
4556 | if (dn->lru_is_expireable()) { |
4557 | if (can_invalidate_dentries && | |
b3b6e05e | 4558 | dn->dir->parent_inode->ino == CEPH_INO_ROOT) { |
7c673cae FG |
4559 | // Only issue one of these per DN for inodes in root: handle |
4560 | // others more efficiently by calling for root-child DNs at | |
4561 | // the end of this function. | |
4562 | _schedule_invalidate_dentry_callback(dn, true); | |
4563 | } | |
28e407b8 AA |
4564 | ldout(cct, 20) << " queueing dentry for trimming: " << dn->name << dendl; |
4565 | to_trim.insert(dn); | |
7c673cae FG |
4566 | } else { |
4567 | ldout(cct, 20) << " not expirable: " << dn->name << dendl; | |
4568 | all = false; | |
4569 | } | |
4570 | } | |
b3b6e05e | 4571 | if (in->ll_ref == 1 && in->ino != CEPH_INO_ROOT) { |
f91f0fd5 TL |
4572 | _schedule_ino_release_callback(in.get()); |
4573 | } | |
b3b6e05e | 4574 | if (all && in->ino != CEPH_INO_ROOT) { |
7c673cae FG |
4575 | ldout(cct, 20) << __func__ << " counting as trimmed: " << *in << dendl; |
4576 | trimmed++; | |
4577 | } | |
4578 | } | |
4579 | } | |
28e407b8 AA |
4580 | ldout(cct, 20) << " trimming queued dentries: " << dendl; |
4581 | for (const auto &dn : to_trim) { | |
4582 | trim_dentry(dn); | |
4583 | } | |
4584 | to_trim.clear(); | |
7c673cae | 4585 | |
b32b8144 | 4586 | caps_size = s->caps.size(); |
11fdf7f2 | 4587 | if (caps_size > (size_t)max) |
7c673cae FG |
4588 | _invalidate_kernel_dcache(); |
4589 | } | |
4590 | ||
4591 | void Client::force_session_readonly(MetaSession *s) | |
4592 | { | |
4593 | s->readonly = true; | |
4594 | for (xlist<Cap*>::iterator p = s->caps.begin(); !p.end(); ++p) { | |
11fdf7f2 TL |
4595 | auto &in = (*p)->inode; |
4596 | if (in.caps_wanted() & CEPH_CAP_FILE_WR) | |
4597 | signal_cond_list(in.waitfor_caps); | |
7c673cae FG |
4598 | } |
4599 | } | |
4600 | ||
7c673cae FG |
4601 | int Client::mark_caps_flushing(Inode *in, ceph_tid_t* ptid) |
4602 | { | |
4603 | MetaSession *session = in->auth_cap->session; | |
4604 | ||
4605 | int flushing = in->dirty_caps; | |
11fdf7f2 | 4606 | ceph_assert(flushing); |
7c673cae FG |
4607 | |
4608 | ceph_tid_t flush_tid = ++last_flush_tid; | |
4609 | in->flushing_cap_tids[flush_tid] = flushing; | |
4610 | ||
4611 | if (!in->flushing_caps) { | |
11fdf7f2 | 4612 | ldout(cct, 10) << __func__ << " " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4613 | num_flushing_caps++; |
4614 | } else { | |
11fdf7f2 | 4615 | ldout(cct, 10) << __func__ << " (more) " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4616 | } |
4617 | ||
4618 | in->flushing_caps |= flushing; | |
28e407b8 | 4619 | in->mark_caps_clean(); |
7c673cae FG |
4620 | |
4621 | if (!in->flushing_cap_item.is_on_list()) | |
4622 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
4623 | session->flushing_caps_tids.insert(flush_tid); | |
4624 | ||
4625 | *ptid = flush_tid; | |
4626 | return flushing; | |
4627 | } | |
4628 | ||
4629 | void Client::adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s) | |
4630 | { | |
4631 | for (auto &p : in->cap_snaps) { | |
4632 | CapSnap &capsnap = p.second; | |
4633 | if (capsnap.flush_tid > 0) { | |
4634 | old_s->flushing_caps_tids.erase(capsnap.flush_tid); | |
4635 | new_s->flushing_caps_tids.insert(capsnap.flush_tid); | |
4636 | } | |
4637 | } | |
4638 | for (map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4639 | it != in->flushing_cap_tids.end(); | |
4640 | ++it) { | |
4641 | old_s->flushing_caps_tids.erase(it->first); | |
4642 | new_s->flushing_caps_tids.insert(it->first); | |
4643 | } | |
4644 | new_s->flushing_caps.push_back(&in->flushing_cap_item); | |
4645 | } | |
4646 | ||
4647 | /* | |
4648 | * Flush all caps back to the MDS. Because the callers generally wait on the | |
4649 | * result of this function (syncfs and umount cases), we set | |
4650 | * CHECK_CAPS_SYNCHRONOUS on the last check_caps call. | |
4651 | */ | |
4652 | void Client::flush_caps_sync() | |
4653 | { | |
4654 | ldout(cct, 10) << __func__ << dendl; | |
28e407b8 | 4655 | xlist<Inode*>::iterator p = delayed_list.begin(); |
7c673cae FG |
4656 | while (!p.end()) { |
4657 | unsigned flags = CHECK_CAPS_NODELAY; | |
4658 | Inode *in = *p; | |
4659 | ||
4660 | ++p; | |
28e407b8 AA |
4661 | delayed_list.pop_front(); |
4662 | if (p.end() && dirty_list.empty()) | |
7c673cae FG |
4663 | flags |= CHECK_CAPS_SYNCHRONOUS; |
4664 | check_caps(in, flags); | |
4665 | } | |
4666 | ||
4667 | // other caps, too | |
28e407b8 | 4668 | p = dirty_list.begin(); |
7c673cae FG |
4669 | while (!p.end()) { |
4670 | unsigned flags = CHECK_CAPS_NODELAY; | |
4671 | Inode *in = *p; | |
4672 | ||
4673 | ++p; | |
4674 | if (p.end()) | |
4675 | flags |= CHECK_CAPS_SYNCHRONOUS; | |
4676 | check_caps(in, flags); | |
4677 | } | |
4678 | } | |
4679 | ||
7c673cae FG |
4680 | void Client::wait_sync_caps(Inode *in, ceph_tid_t want) |
4681 | { | |
4682 | while (in->flushing_caps) { | |
4683 | map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
11fdf7f2 | 4684 | ceph_assert(it != in->flushing_cap_tids.end()); |
7c673cae FG |
4685 | if (it->first > want) |
4686 | break; | |
11fdf7f2 | 4687 | ldout(cct, 10) << __func__ << " on " << *in << " flushing " |
7c673cae FG |
4688 | << ccap_string(it->second) << " want " << want |
4689 | << " last " << it->first << dendl; | |
4690 | wait_on_list(in->waitfor_caps); | |
4691 | } | |
4692 | } | |
4693 | ||
4694 | void Client::wait_sync_caps(ceph_tid_t want) | |
4695 | { | |
4696 | retry: | |
11fdf7f2 | 4697 | ldout(cct, 10) << __func__ << " want " << want << " (last is " << last_flush_tid << ", " |
7c673cae | 4698 | << num_flushing_caps << " total flushing)" << dendl; |
11fdf7f2 TL |
4699 | for (auto &p : mds_sessions) { |
4700 | MetaSession *s = &p.second; | |
7c673cae FG |
4701 | if (s->flushing_caps_tids.empty()) |
4702 | continue; | |
4703 | ceph_tid_t oldest_tid = *s->flushing_caps_tids.begin(); | |
4704 | if (oldest_tid <= want) { | |
11fdf7f2 | 4705 | ldout(cct, 10) << " waiting on mds." << p.first << " tid " << oldest_tid |
7c673cae | 4706 | << " (want " << want << ")" << dendl; |
9f95a23c TL |
4707 | std::unique_lock l{client_lock, std::adopt_lock}; |
4708 | sync_cond.wait(l); | |
4709 | l.release(); | |
7c673cae FG |
4710 | goto retry; |
4711 | } | |
4712 | } | |
4713 | } | |
4714 | ||
eafe8130 TL |
4715 | void Client::kick_flushing_caps(Inode *in, MetaSession *session) |
4716 | { | |
4717 | in->flags &= ~I_KICK_FLUSH; | |
4718 | ||
4719 | Cap *cap = in->auth_cap; | |
4720 | ceph_assert(cap->session == session); | |
4721 | ||
4722 | ceph_tid_t last_snap_flush = 0; | |
4723 | for (auto p = in->flushing_cap_tids.rbegin(); | |
4724 | p != in->flushing_cap_tids.rend(); | |
4725 | ++p) { | |
4726 | if (!p->second) { | |
4727 | last_snap_flush = p->first; | |
4728 | break; | |
4729 | } | |
4730 | } | |
4731 | ||
4732 | int wanted = in->caps_wanted(); | |
4733 | int used = get_caps_used(in) | in->caps_dirty(); | |
4734 | auto it = in->cap_snaps.begin(); | |
4735 | for (auto& p : in->flushing_cap_tids) { | |
4736 | if (p.second) { | |
4737 | int msg_flags = p.first < last_snap_flush ? MClientCaps::FLAG_PENDING_CAPSNAP : 0; | |
4738 | send_cap(in, session, cap, msg_flags, used, wanted, (cap->issued | cap->implemented), | |
4739 | p.second, p.first); | |
4740 | } else { | |
4741 | ceph_assert(it != in->cap_snaps.end()); | |
4742 | ceph_assert(it->second.flush_tid == p.first); | |
4743 | send_flush_snap(in, session, it->first, it->second); | |
4744 | ++it; | |
4745 | } | |
4746 | } | |
4747 | } | |
4748 | ||
7c673cae FG |
4749 | void Client::kick_flushing_caps(MetaSession *session) |
4750 | { | |
4751 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 4752 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
4753 | |
4754 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { | |
4755 | Inode *in = *p; | |
eafe8130 TL |
4756 | if (in->flags & I_KICK_FLUSH) { |
4757 | ldout(cct, 20) << " reflushing caps on " << *in << " to mds." << mds << dendl; | |
4758 | kick_flushing_caps(in, session); | |
4759 | } | |
7c673cae | 4760 | } |
7c673cae FG |
4761 | } |
4762 | ||
4763 | void Client::early_kick_flushing_caps(MetaSession *session) | |
4764 | { | |
7c673cae FG |
4765 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { |
4766 | Inode *in = *p; | |
11fdf7f2 TL |
4767 | Cap *cap = in->auth_cap; |
4768 | ceph_assert(cap); | |
7c673cae FG |
4769 | |
4770 | // if flushing caps were revoked, we re-send the cap flush in client reconnect | |
4771 | // stage. This guarantees that MDS processes the cap flush message before issuing | |
4772 | // the flushing caps to other client. | |
eafe8130 TL |
4773 | if ((in->flushing_caps & in->auth_cap->issued) == in->flushing_caps) { |
4774 | in->flags |= I_KICK_FLUSH; | |
7c673cae | 4775 | continue; |
eafe8130 | 4776 | } |
7c673cae FG |
4777 | |
4778 | ldout(cct, 20) << " reflushing caps (early_kick) on " << *in | |
4779 | << " to mds." << session->mds_num << dendl; | |
11fdf7f2 TL |
4780 | // send_reconnect() also will reset these sequence numbers. make sure |
4781 | // sequence numbers in cap flush message match later reconnect message. | |
4782 | cap->seq = 0; | |
4783 | cap->issue_seq = 0; | |
4784 | cap->mseq = 0; | |
4785 | cap->issued = cap->implemented; | |
4786 | ||
eafe8130 | 4787 | kick_flushing_caps(in, session); |
7c673cae FG |
4788 | } |
4789 | } | |
4790 | ||
7c673cae FG |
4791 | void SnapRealm::build_snap_context() |
4792 | { | |
4793 | set<snapid_t> snaps; | |
4794 | snapid_t max_seq = seq; | |
4795 | ||
4796 | // start with prior_parents? | |
4797 | for (unsigned i=0; i<prior_parent_snaps.size(); i++) | |
4798 | snaps.insert(prior_parent_snaps[i]); | |
4799 | ||
4800 | // current parent's snaps | |
4801 | if (pparent) { | |
4802 | const SnapContext& psnapc = pparent->get_snap_context(); | |
4803 | for (unsigned i=0; i<psnapc.snaps.size(); i++) | |
4804 | if (psnapc.snaps[i] >= parent_since) | |
4805 | snaps.insert(psnapc.snaps[i]); | |
4806 | if (psnapc.seq > max_seq) | |
4807 | max_seq = psnapc.seq; | |
4808 | } | |
4809 | ||
4810 | // my snaps | |
4811 | for (unsigned i=0; i<my_snaps.size(); i++) | |
4812 | snaps.insert(my_snaps[i]); | |
4813 | ||
4814 | // ok! | |
4815 | cached_snap_context.seq = max_seq; | |
4816 | cached_snap_context.snaps.resize(0); | |
4817 | cached_snap_context.snaps.reserve(snaps.size()); | |
4818 | for (set<snapid_t>::reverse_iterator p = snaps.rbegin(); p != snaps.rend(); ++p) | |
4819 | cached_snap_context.snaps.push_back(*p); | |
4820 | } | |
4821 | ||
4822 | void Client::invalidate_snaprealm_and_children(SnapRealm *realm) | |
4823 | { | |
4824 | list<SnapRealm*> q; | |
4825 | q.push_back(realm); | |
4826 | ||
4827 | while (!q.empty()) { | |
4828 | realm = q.front(); | |
4829 | q.pop_front(); | |
4830 | ||
11fdf7f2 | 4831 | ldout(cct, 10) << __func__ << " " << *realm << dendl; |
7c673cae FG |
4832 | realm->invalidate_cache(); |
4833 | ||
4834 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4835 | p != realm->pchildren.end(); | |
4836 | ++p) | |
4837 | q.push_back(*p); | |
4838 | } | |
4839 | } | |
4840 | ||
4841 | SnapRealm *Client::get_snap_realm(inodeno_t r) | |
4842 | { | |
4843 | SnapRealm *realm = snap_realms[r]; | |
4844 | if (!realm) | |
4845 | snap_realms[r] = realm = new SnapRealm(r); | |
11fdf7f2 | 4846 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4847 | realm->nref++; |
4848 | return realm; | |
4849 | } | |
4850 | ||
4851 | SnapRealm *Client::get_snap_realm_maybe(inodeno_t r) | |
4852 | { | |
4853 | if (snap_realms.count(r) == 0) { | |
11fdf7f2 | 4854 | ldout(cct, 20) << __func__ << " " << r << " fail" << dendl; |
7c673cae FG |
4855 | return NULL; |
4856 | } | |
4857 | SnapRealm *realm = snap_realms[r]; | |
11fdf7f2 | 4858 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4859 | realm->nref++; |
4860 | return realm; | |
4861 | } | |
4862 | ||
4863 | void Client::put_snap_realm(SnapRealm *realm) | |
4864 | { | |
11fdf7f2 | 4865 | ldout(cct, 20) << __func__ << " " << realm->ino << " " << realm |
7c673cae FG |
4866 | << " " << realm->nref << " -> " << (realm->nref - 1) << dendl; |
4867 | if (--realm->nref == 0) { | |
4868 | snap_realms.erase(realm->ino); | |
4869 | if (realm->pparent) { | |
4870 | realm->pparent->pchildren.erase(realm); | |
4871 | put_snap_realm(realm->pparent); | |
4872 | } | |
4873 | delete realm; | |
4874 | } | |
4875 | } | |
4876 | ||
4877 | bool Client::adjust_realm_parent(SnapRealm *realm, inodeno_t parent) | |
4878 | { | |
4879 | if (realm->parent != parent) { | |
11fdf7f2 | 4880 | ldout(cct, 10) << __func__ << " " << *realm |
7c673cae FG |
4881 | << " " << realm->parent << " -> " << parent << dendl; |
4882 | realm->parent = parent; | |
4883 | if (realm->pparent) { | |
4884 | realm->pparent->pchildren.erase(realm); | |
4885 | put_snap_realm(realm->pparent); | |
4886 | } | |
4887 | realm->pparent = get_snap_realm(parent); | |
4888 | realm->pparent->pchildren.insert(realm); | |
4889 | return true; | |
4890 | } | |
4891 | return false; | |
4892 | } | |
4893 | ||
4894 | static bool has_new_snaps(const SnapContext& old_snapc, | |
4895 | const SnapContext& new_snapc) | |
4896 | { | |
4897 | return !new_snapc.snaps.empty() && new_snapc.snaps[0] > old_snapc.seq; | |
4898 | } | |
4899 | ||
4900 | ||
11fdf7f2 | 4901 | void Client::update_snap_trace(const bufferlist& bl, SnapRealm **realm_ret, bool flush) |
7c673cae FG |
4902 | { |
4903 | SnapRealm *first_realm = NULL; | |
11fdf7f2 | 4904 | ldout(cct, 10) << __func__ << " len " << bl.length() << dendl; |
7c673cae FG |
4905 | |
4906 | map<SnapRealm*, SnapContext> dirty_realms; | |
4907 | ||
11fdf7f2 | 4908 | auto p = bl.cbegin(); |
7c673cae FG |
4909 | while (!p.end()) { |
4910 | SnapRealmInfo info; | |
11fdf7f2 | 4911 | decode(info, p); |
7c673cae FG |
4912 | SnapRealm *realm = get_snap_realm(info.ino()); |
4913 | ||
4914 | bool invalidate = false; | |
4915 | ||
4916 | if (info.seq() > realm->seq) { | |
11fdf7f2 | 4917 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() << " > " << realm->seq |
7c673cae FG |
4918 | << dendl; |
4919 | ||
4920 | if (flush) { | |
4921 | // writeback any dirty caps _before_ updating snap list (i.e. with old snap info) | |
4922 | // flush me + children | |
4923 | list<SnapRealm*> q; | |
4924 | q.push_back(realm); | |
4925 | while (!q.empty()) { | |
4926 | SnapRealm *realm = q.front(); | |
4927 | q.pop_front(); | |
4928 | ||
4929 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4930 | p != realm->pchildren.end(); | |
4931 | ++p) | |
4932 | q.push_back(*p); | |
4933 | ||
4934 | if (dirty_realms.count(realm) == 0) { | |
4935 | realm->nref++; | |
4936 | dirty_realms[realm] = realm->get_snap_context(); | |
4937 | } | |
4938 | } | |
4939 | } | |
4940 | ||
4941 | // update | |
4942 | realm->seq = info.seq(); | |
4943 | realm->created = info.created(); | |
4944 | realm->parent_since = info.parent_since(); | |
4945 | realm->prior_parent_snaps = info.prior_parent_snaps; | |
4946 | realm->my_snaps = info.my_snaps; | |
4947 | invalidate = true; | |
4948 | } | |
4949 | ||
4950 | // _always_ verify parent | |
4951 | if (adjust_realm_parent(realm, info.parent())) | |
4952 | invalidate = true; | |
4953 | ||
4954 | if (invalidate) { | |
4955 | invalidate_snaprealm_and_children(realm); | |
11fdf7f2 | 4956 | ldout(cct, 15) << __func__ << " " << *realm << " self|parent updated" << dendl; |
7c673cae FG |
4957 | ldout(cct, 15) << " snapc " << realm->get_snap_context() << dendl; |
4958 | } else { | |
11fdf7f2 | 4959 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() |
7c673cae FG |
4960 | << " <= " << realm->seq << " and same parent, SKIPPING" << dendl; |
4961 | } | |
f67539c2 | 4962 | |
7c673cae FG |
4963 | if (!first_realm) |
4964 | first_realm = realm; | |
4965 | else | |
4966 | put_snap_realm(realm); | |
4967 | } | |
4968 | ||
f67539c2 | 4969 | for (auto &[realm, snapc] : dirty_realms) { |
7c673cae | 4970 | // if there are new snaps ? |
f67539c2 | 4971 | if (has_new_snaps(snapc, realm->get_snap_context())) { |
7c673cae | 4972 | ldout(cct, 10) << " flushing caps on " << *realm << dendl; |
f67539c2 TL |
4973 | for (auto&& in : realm->inodes_with_caps) { |
4974 | queue_cap_snap(in, snapc); | |
7c673cae FG |
4975 | } |
4976 | } else { | |
4977 | ldout(cct, 10) << " no new snap on " << *realm << dendl; | |
4978 | } | |
4979 | put_snap_realm(realm); | |
4980 | } | |
4981 | ||
4982 | if (realm_ret) | |
4983 | *realm_ret = first_realm; | |
4984 | else | |
4985 | put_snap_realm(first_realm); | |
4986 | } | |
4987 | ||
11fdf7f2 | 4988 | void Client::handle_snap(const MConstRef<MClientSnap>& m) |
7c673cae | 4989 | { |
11fdf7f2 | 4990 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 4991 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
f67539c2 TL |
4992 | |
4993 | std::scoped_lock cl(client_lock); | |
7c673cae FG |
4994 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); |
4995 | if (!session) { | |
7c673cae FG |
4996 | return; |
4997 | } | |
4998 | ||
4999 | got_mds_push(session); | |
5000 | ||
5001 | map<Inode*, SnapContext> to_move; | |
5002 | SnapRealm *realm = 0; | |
5003 | ||
5004 | if (m->head.op == CEPH_SNAP_OP_SPLIT) { | |
11fdf7f2 | 5005 | ceph_assert(m->head.split); |
7c673cae | 5006 | SnapRealmInfo info; |
11fdf7f2 TL |
5007 | auto p = m->bl.cbegin(); |
5008 | decode(info, p); | |
5009 | ceph_assert(info.ino() == m->head.split); | |
7c673cae FG |
5010 | |
5011 | // flush, then move, ino's. | |
5012 | realm = get_snap_realm(info.ino()); | |
5013 | ldout(cct, 10) << " splitting off " << *realm << dendl; | |
11fdf7f2 TL |
5014 | for (auto& ino : m->split_inos) { |
5015 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
7c673cae FG |
5016 | if (inode_map.count(vino)) { |
5017 | Inode *in = inode_map[vino]; | |
5018 | if (!in->snaprealm || in->snaprealm == realm) | |
5019 | continue; | |
5020 | if (in->snaprealm->created > info.created()) { | |
5021 | ldout(cct, 10) << " NOT moving " << *in << " from _newer_ realm " | |
5022 | << *in->snaprealm << dendl; | |
5023 | continue; | |
5024 | } | |
5025 | ldout(cct, 10) << " moving " << *in << " from " << *in->snaprealm << dendl; | |
5026 | ||
5027 | ||
5028 | in->snaprealm_item.remove_myself(); | |
5029 | to_move[in] = in->snaprealm->get_snap_context(); | |
5030 | put_snap_realm(in->snaprealm); | |
5031 | } | |
5032 | } | |
5033 | ||
5034 | // move child snaprealms, too | |
11fdf7f2 TL |
5035 | for (auto& child_realm : m->split_realms) { |
5036 | ldout(cct, 10) << "adjusting snaprealm " << child_realm << " parent" << dendl; | |
5037 | SnapRealm *child = get_snap_realm_maybe(child_realm); | |
7c673cae FG |
5038 | if (!child) |
5039 | continue; | |
5040 | adjust_realm_parent(child, realm->ino); | |
5041 | put_snap_realm(child); | |
5042 | } | |
5043 | } | |
5044 | ||
5045 | update_snap_trace(m->bl, NULL, m->head.op != CEPH_SNAP_OP_DESTROY); | |
5046 | ||
5047 | if (realm) { | |
5048 | for (auto p = to_move.begin(); p != to_move.end(); ++p) { | |
5049 | Inode *in = p->first; | |
5050 | in->snaprealm = realm; | |
5051 | realm->inodes_with_caps.push_back(&in->snaprealm_item); | |
5052 | realm->nref++; | |
5053 | // queue for snap writeback | |
5054 | if (has_new_snaps(p->second, realm->get_snap_context())) | |
5055 | queue_cap_snap(in, p->second); | |
5056 | } | |
5057 | put_snap_realm(realm); | |
5058 | } | |
7c673cae FG |
5059 | } |
5060 | ||
11fdf7f2 | 5061 | void Client::handle_quota(const MConstRef<MClientQuota>& m) |
7c673cae FG |
5062 | { |
5063 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5064 | |
5065 | std::scoped_lock cl(client_lock); | |
7c673cae FG |
5066 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); |
5067 | if (!session) { | |
7c673cae FG |
5068 | return; |
5069 | } | |
5070 | ||
5071 | got_mds_push(session); | |
5072 | ||
11fdf7f2 | 5073 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << mds << dendl; |
7c673cae FG |
5074 | |
5075 | vinodeno_t vino(m->ino, CEPH_NOSNAP); | |
5076 | if (inode_map.count(vino)) { | |
5077 | Inode *in = NULL; | |
5078 | in = inode_map[vino]; | |
5079 | ||
5080 | if (in) { | |
5081 | in->quota = m->quota; | |
5082 | in->rstat = m->rstat; | |
5083 | } | |
5084 | } | |
7c673cae FG |
5085 | } |
5086 | ||
11fdf7f2 | 5087 | void Client::handle_caps(const MConstRef<MClientCaps>& m) |
7c673cae FG |
5088 | { |
5089 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
f67539c2 TL |
5090 | |
5091 | std::scoped_lock cl(client_lock); | |
7c673cae FG |
5092 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); |
5093 | if (!session) { | |
7c673cae FG |
5094 | return; |
5095 | } | |
5096 | ||
5097 | if (m->osd_epoch_barrier && !objecter->have_map(m->osd_epoch_barrier)) { | |
5098 | // Pause RADOS operations until we see the required epoch | |
5099 | objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
5100 | } | |
5101 | ||
5102 | if (m->osd_epoch_barrier > cap_epoch_barrier) { | |
5103 | // Record the barrier so that we will transmit it to MDS when releasing | |
5104 | set_cap_epoch_barrier(m->osd_epoch_barrier); | |
5105 | } | |
5106 | ||
5107 | got_mds_push(session); | |
5108 | ||
11fdf7f2 | 5109 | Inode *in; |
7c673cae | 5110 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); |
11fdf7f2 TL |
5111 | if (auto it = inode_map.find(vino); it != inode_map.end()) { |
5112 | in = it->second; | |
5113 | } else { | |
7c673cae | 5114 | if (m->get_op() == CEPH_CAP_OP_IMPORT) { |
11fdf7f2 | 5115 | ldout(cct, 5) << __func__ << " don't have vino " << vino << " on IMPORT, immediately releasing" << dendl; |
7c673cae FG |
5116 | session->enqueue_cap_release( |
5117 | m->get_ino(), | |
5118 | m->get_cap_id(), | |
5119 | m->get_seq(), | |
5120 | m->get_mseq(), | |
5121 | cap_epoch_barrier); | |
5122 | } else { | |
11fdf7f2 | 5123 | ldout(cct, 5) << __func__ << " don't have vino " << vino << ", dropping" << dendl; |
7c673cae | 5124 | } |
7c673cae FG |
5125 | |
5126 | // in case the mds is waiting on e.g. a revocation | |
5127 | flush_cap_releases(); | |
5128 | return; | |
5129 | } | |
5130 | ||
5131 | switch (m->get_op()) { | |
11fdf7f2 TL |
5132 | case CEPH_CAP_OP_EXPORT: return handle_cap_export(session, in, m); |
5133 | case CEPH_CAP_OP_FLUSHSNAP_ACK: return handle_cap_flushsnap_ack(session, in, m); | |
5134 | case CEPH_CAP_OP_IMPORT: /* no return */ handle_cap_import(session, in, m); | |
7c673cae FG |
5135 | } |
5136 | ||
11fdf7f2 TL |
5137 | if (auto it = in->caps.find(mds); it != in->caps.end()) { |
5138 | Cap &cap = in->caps.at(mds); | |
7c673cae | 5139 | |
11fdf7f2 TL |
5140 | switch (m->get_op()) { |
5141 | case CEPH_CAP_OP_TRUNC: return handle_cap_trunc(session, in, m); | |
5142 | case CEPH_CAP_OP_IMPORT: | |
5143 | case CEPH_CAP_OP_REVOKE: | |
5144 | case CEPH_CAP_OP_GRANT: return handle_cap_grant(session, in, &cap, m); | |
5145 | case CEPH_CAP_OP_FLUSH_ACK: return handle_cap_flush_ack(session, in, &cap, m); | |
5146 | } | |
5147 | } else { | |
5148 | ldout(cct, 5) << __func__ << " don't have " << *in << " cap on mds." << mds << dendl; | |
5149 | return; | |
7c673cae FG |
5150 | } |
5151 | } | |
5152 | ||
11fdf7f2 | 5153 | void Client::handle_cap_import(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5154 | { |
5155 | mds_rank_t mds = session->mds_num; | |
5156 | ||
11fdf7f2 | 5157 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5158 | << " IMPORT from mds." << mds << dendl; |
5159 | ||
5160 | const mds_rank_t peer_mds = mds_rank_t(m->peer.mds); | |
5161 | Cap *cap = NULL; | |
5162 | UserPerm cap_perms; | |
11fdf7f2 TL |
5163 | if (auto it = in->caps.find(peer_mds); m->peer.cap_id && it != in->caps.end()) { |
5164 | cap = &it->second; | |
5165 | cap_perms = cap->latest_perms; | |
7c673cae FG |
5166 | } |
5167 | ||
5168 | // add/update it | |
5169 | SnapRealm *realm = NULL; | |
5170 | update_snap_trace(m->snapbl, &realm); | |
5171 | ||
1911f103 TL |
5172 | int issued = m->get_caps(); |
5173 | int wanted = m->get_wanted(); | |
7c673cae | 5174 | add_update_cap(in, session, m->get_cap_id(), |
1911f103 | 5175 | issued, wanted, m->get_seq(), m->get_mseq(), |
a8e16298 | 5176 | m->get_realm(), CEPH_CAP_FLAG_AUTH, cap_perms); |
7c673cae FG |
5177 | |
5178 | if (cap && cap->cap_id == m->peer.cap_id) { | |
5179 | remove_cap(cap, (m->peer.flags & CEPH_CAP_FLAG_RELEASE)); | |
5180 | } | |
5181 | ||
5182 | if (realm) | |
5183 | put_snap_realm(realm); | |
5184 | ||
eafe8130 | 5185 | if (in->auth_cap && in->auth_cap->session == session) { |
1911f103 TL |
5186 | if (!(wanted & CEPH_CAP_ANY_FILE_WR) || |
5187 | in->requested_max_size > m->get_max_size()) { | |
5188 | in->requested_max_size = 0; | |
5189 | ldout(cct, 15) << "reset requested_max_size after cap import" << dendl; | |
5190 | } | |
7c673cae | 5191 | // reflush any/all caps (if we are now the auth_cap) |
eafe8130 | 5192 | kick_flushing_caps(in, session); |
7c673cae FG |
5193 | } |
5194 | } | |
5195 | ||
11fdf7f2 | 5196 | void Client::handle_cap_export(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5197 | { |
5198 | mds_rank_t mds = session->mds_num; | |
5199 | ||
11fdf7f2 | 5200 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
5201 | << " EXPORT from mds." << mds << dendl; |
5202 | ||
11fdf7f2 TL |
5203 | auto it = in->caps.find(mds); |
5204 | if (it != in->caps.end()) { | |
5205 | Cap &cap = it->second; | |
5206 | if (cap.cap_id == m->get_cap_id()) { | |
5207 | if (m->peer.cap_id) { | |
5208 | const auto peer_mds = mds_rank_t(m->peer.mds); | |
5209 | MetaSession *tsession = _get_or_open_mds_session(peer_mds); | |
5210 | auto it = in->caps.find(peer_mds); | |
5211 | if (it != in->caps.end()) { | |
5212 | Cap &tcap = it->second; | |
5213 | if (tcap.cap_id == m->peer.cap_id && | |
5214 | ceph_seq_cmp(tcap.seq, m->peer.seq) < 0) { | |
5215 | tcap.cap_id = m->peer.cap_id; | |
5216 | tcap.seq = m->peer.seq - 1; | |
5217 | tcap.issue_seq = tcap.seq; | |
5218 | tcap.issued |= cap.issued; | |
5219 | tcap.implemented |= cap.issued; | |
5220 | if (&cap == in->auth_cap) | |
5221 | in->auth_cap = &tcap; | |
5222 | if (in->auth_cap == &tcap && in->flushing_cap_item.is_on_list()) | |
5223 | adjust_session_flushing_caps(in, session, tsession); | |
5224 | } | |
5225 | } else { | |
5226 | add_update_cap(in, tsession, m->peer.cap_id, cap.issued, 0, | |
5227 | m->peer.seq - 1, m->peer.mseq, (uint64_t)-1, | |
5228 | &cap == in->auth_cap ? CEPH_CAP_FLAG_AUTH : 0, | |
5229 | cap.latest_perms); | |
5230 | } | |
7c673cae | 5231 | } else { |
11fdf7f2 TL |
5232 | if (cap.wanted | cap.issued) |
5233 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 5234 | } |
7c673cae | 5235 | |
11fdf7f2 TL |
5236 | remove_cap(&cap, false); |
5237 | } | |
7c673cae | 5238 | } |
7c673cae FG |
5239 | } |
5240 | ||
11fdf7f2 | 5241 | void Client::handle_cap_trunc(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5242 | { |
5243 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 5244 | ceph_assert(in->caps.count(mds)); |
7c673cae | 5245 | |
11fdf7f2 | 5246 | ldout(cct, 10) << __func__ << " on ino " << *in |
7c673cae FG |
5247 | << " size " << in->size << " -> " << m->get_size() |
5248 | << dendl; | |
5249 | ||
1adf2230 AA |
5250 | int issued; |
5251 | in->caps_issued(&issued); | |
5252 | issued |= in->caps_dirty(); | |
5253 | update_inode_file_size(in, issued, m->get_size(), | |
5254 | m->get_truncate_seq(), m->get_truncate_size()); | |
7c673cae FG |
5255 | } |
5256 | ||
11fdf7f2 | 5257 | void Client::handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5258 | { |
5259 | ceph_tid_t flush_ack_tid = m->get_client_tid(); | |
5260 | int dirty = m->get_dirty(); | |
5261 | int cleaned = 0; | |
5262 | int flushed = 0; | |
5263 | ||
11fdf7f2 TL |
5264 | auto it = in->flushing_cap_tids.begin(); |
5265 | if (it->first < flush_ack_tid) { | |
5266 | ldout(cct, 0) << __func__ << " mds." << session->mds_num | |
5267 | << " got unexpected flush ack tid " << flush_ack_tid | |
5268 | << " expected is " << it->first << dendl; | |
5269 | } | |
5270 | for (; it != in->flushing_cap_tids.end(); ) { | |
eafe8130 TL |
5271 | if (!it->second) { |
5272 | // cap snap | |
5273 | ++it; | |
5274 | continue; | |
5275 | } | |
7c673cae FG |
5276 | if (it->first == flush_ack_tid) |
5277 | cleaned = it->second; | |
5278 | if (it->first <= flush_ack_tid) { | |
5279 | session->flushing_caps_tids.erase(it->first); | |
5280 | in->flushing_cap_tids.erase(it++); | |
5281 | ++flushed; | |
5282 | continue; | |
5283 | } | |
5284 | cleaned &= ~it->second; | |
5285 | if (!cleaned) | |
5286 | break; | |
5287 | ++it; | |
5288 | } | |
5289 | ||
11fdf7f2 | 5290 | ldout(cct, 5) << __func__ << " mds." << session->mds_num |
7c673cae FG |
5291 | << " cleaned " << ccap_string(cleaned) << " on " << *in |
5292 | << " with " << ccap_string(dirty) << dendl; | |
5293 | ||
5294 | if (flushed) { | |
5295 | signal_cond_list(in->waitfor_caps); | |
5296 | if (session->flushing_caps_tids.empty() || | |
5297 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5298 | sync_cond.notify_all(); |
7c673cae FG |
5299 | } |
5300 | ||
5301 | if (!dirty) { | |
5302 | in->cap_dirtier_uid = -1; | |
5303 | in->cap_dirtier_gid = -1; | |
5304 | } | |
5305 | ||
5306 | if (!cleaned) { | |
5307 | ldout(cct, 10) << " tid " << m->get_client_tid() << " != any cap bit tids" << dendl; | |
5308 | } else { | |
5309 | if (in->flushing_caps) { | |
5310 | ldout(cct, 5) << " flushing_caps " << ccap_string(in->flushing_caps) | |
5311 | << " -> " << ccap_string(in->flushing_caps & ~cleaned) << dendl; | |
5312 | in->flushing_caps &= ~cleaned; | |
5313 | if (in->flushing_caps == 0) { | |
5314 | ldout(cct, 10) << " " << *in << " !flushing" << dendl; | |
5315 | num_flushing_caps--; | |
eafe8130 | 5316 | if (in->flushing_cap_tids.empty()) |
7c673cae FG |
5317 | in->flushing_cap_item.remove_myself(); |
5318 | } | |
5319 | if (!in->caps_dirty()) | |
5320 | put_inode(in); | |
5321 | } | |
5322 | } | |
7c673cae FG |
5323 | } |
5324 | ||
5325 | ||
11fdf7f2 | 5326 | void Client::handle_cap_flushsnap_ack(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae | 5327 | { |
eafe8130 | 5328 | ceph_tid_t flush_ack_tid = m->get_client_tid(); |
7c673cae | 5329 | mds_rank_t mds = session->mds_num; |
11fdf7f2 | 5330 | ceph_assert(in->caps.count(mds)); |
7c673cae FG |
5331 | snapid_t follows = m->get_snap_follows(); |
5332 | ||
11fdf7f2 TL |
5333 | if (auto it = in->cap_snaps.find(follows); it != in->cap_snaps.end()) { |
5334 | auto& capsnap = it->second; | |
eafe8130 TL |
5335 | if (flush_ack_tid != capsnap.flush_tid) { |
5336 | ldout(cct, 10) << " tid " << flush_ack_tid << " != " << capsnap.flush_tid << dendl; | |
7c673cae | 5337 | } else { |
eafe8130 | 5338 | InodeRef tmp_ref(in); |
11fdf7f2 | 5339 | ldout(cct, 5) << __func__ << " mds." << mds << " flushed snap follows " << follows |
7c673cae | 5340 | << " on " << *in << dendl; |
7c673cae | 5341 | session->flushing_caps_tids.erase(capsnap.flush_tid); |
eafe8130 TL |
5342 | in->flushing_cap_tids.erase(capsnap.flush_tid); |
5343 | if (in->flushing_caps == 0 && in->flushing_cap_tids.empty()) | |
5344 | in->flushing_cap_item.remove_myself(); | |
11fdf7f2 | 5345 | in->cap_snaps.erase(it); |
eafe8130 TL |
5346 | |
5347 | signal_cond_list(in->waitfor_caps); | |
5348 | if (session->flushing_caps_tids.empty() || | |
5349 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5350 | sync_cond.notify_all(); |
7c673cae FG |
5351 | } |
5352 | } else { | |
11fdf7f2 | 5353 | ldout(cct, 5) << __func__ << " DUP(?) mds." << mds << " flushed snap follows " << follows |
7c673cae FG |
5354 | << " on " << *in << dendl; |
5355 | // we may not have it if we send multiple FLUSHSNAP requests and (got multiple FLUSHEDSNAPs back) | |
5356 | } | |
7c673cae FG |
5357 | } |
5358 | ||
5359 | class C_Client_DentryInvalidate : public Context { | |
5360 | private: | |
5361 | Client *client; | |
5362 | vinodeno_t dirino; | |
5363 | vinodeno_t ino; | |
5364 | string name; | |
5365 | public: | |
5366 | C_Client_DentryInvalidate(Client *c, Dentry *dn, bool del) : | |
5367 | client(c), name(dn->name) { | |
5368 | if (client->use_faked_inos()) { | |
5369 | dirino.ino = dn->dir->parent_inode->faked_ino; | |
5370 | if (del) | |
5371 | ino.ino = dn->inode->faked_ino; | |
5372 | } else { | |
5373 | dirino = dn->dir->parent_inode->vino(); | |
5374 | if (del) | |
5375 | ino = dn->inode->vino(); | |
5376 | } | |
5377 | if (!del) | |
5378 | ino.ino = inodeno_t(); | |
5379 | } | |
5380 | void finish(int r) override { | |
5381 | // _async_dentry_invalidate is responsible for its own locking | |
9f95a23c | 5382 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
5383 | client->_async_dentry_invalidate(dirino, ino, name); |
5384 | } | |
5385 | }; | |
5386 | ||
5387 | void Client::_async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name) | |
5388 | { | |
f67539c2 TL |
5389 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
5390 | if (!mref_reader.is_state_satisfied()) | |
7c673cae | 5391 | return; |
f67539c2 | 5392 | |
11fdf7f2 | 5393 | ldout(cct, 10) << __func__ << " '" << name << "' ino " << ino |
7c673cae | 5394 | << " in dir " << dirino << dendl; |
e306af50 | 5395 | dentry_invalidate_cb(callback_handle, dirino, ino, name.c_str(), name.length()); |
7c673cae FG |
5396 | } |
5397 | ||
5398 | void Client::_schedule_invalidate_dentry_callback(Dentry *dn, bool del) | |
5399 | { | |
5400 | if (dentry_invalidate_cb && dn->inode->ll_ref > 0) | |
5401 | async_dentry_invalidator.queue(new C_Client_DentryInvalidate(this, dn, del)); | |
5402 | } | |
5403 | ||
5404 | void Client::_try_to_trim_inode(Inode *in, bool sched_inval) | |
5405 | { | |
b3b6e05e | 5406 | int ref = in->get_nref(); |
494da23a | 5407 | ldout(cct, 5) << __func__ << " in " << *in <<dendl; |
7c673cae FG |
5408 | |
5409 | if (in->dir && !in->dir->dentries.empty()) { | |
5410 | for (auto p = in->dir->dentries.begin(); | |
5411 | p != in->dir->dentries.end(); ) { | |
5412 | Dentry *dn = p->second; | |
5413 | ++p; | |
5414 | /* rmsnap removes whole subtree, need trim inodes recursively. | |
5415 | * we don't need to invalidate dentries recursively. because | |
5416 | * invalidating a directory dentry effectively invalidate | |
5417 | * whole subtree */ | |
5418 | if (in->snapid != CEPH_NOSNAP && dn->inode && dn->inode->is_dir()) | |
5419 | _try_to_trim_inode(dn->inode.get(), false); | |
5420 | ||
5421 | if (dn->lru_is_expireable()) | |
5422 | unlink(dn, true, false); // keep dir, drop dentry | |
5423 | } | |
5424 | if (in->dir->dentries.empty()) { | |
5425 | close_dir(in->dir); | |
5426 | --ref; | |
5427 | } | |
5428 | } | |
5429 | ||
b3b6e05e | 5430 | if (ref > 1 && (in->flags & I_SNAPDIR_OPEN)) { |
7c673cae FG |
5431 | InodeRef snapdir = open_snapdir(in); |
5432 | _try_to_trim_inode(snapdir.get(), false); | |
5433 | --ref; | |
5434 | } | |
5435 | ||
b3b6e05e | 5436 | if (ref > 1) { |
11fdf7f2 TL |
5437 | auto q = in->dentries.begin(); |
5438 | while (q != in->dentries.end()) { | |
5439 | Dentry *dn = *q; | |
5440 | ++q; | |
494da23a TL |
5441 | if( in->ll_ref > 0 && sched_inval) { |
5442 | // FIXME: we play lots of unlink/link tricks when handling MDS replies, | |
5443 | // so in->dentries doesn't always reflect the state of kernel's dcache. | |
5444 | _schedule_invalidate_dentry_callback(dn, true); | |
5445 | } | |
7c673cae FG |
5446 | unlink(dn, true, true); |
5447 | } | |
5448 | } | |
5449 | } | |
5450 | ||
11fdf7f2 | 5451 | void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5452 | { |
5453 | mds_rank_t mds = session->mds_num; | |
5454 | int used = get_caps_used(in); | |
5455 | int wanted = in->caps_wanted(); | |
5456 | ||
a8e16298 TL |
5457 | const unsigned new_caps = m->get_caps(); |
5458 | const bool was_stale = session->cap_gen > cap->gen; | |
11fdf7f2 | 5459 | ldout(cct, 5) << __func__ << " on in " << m->get_ino() |
7c673cae FG |
5460 | << " mds." << mds << " seq " << m->get_seq() |
5461 | << " caps now " << ccap_string(new_caps) | |
a8e16298 | 5462 | << " was " << ccap_string(cap->issued) |
92f5a8d4 | 5463 | << (was_stale ? " (stale)" : "") << dendl; |
a8e16298 TL |
5464 | |
5465 | if (was_stale) | |
5466 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
7c673cae | 5467 | cap->seq = m->get_seq(); |
28e407b8 | 5468 | cap->gen = session->cap_gen; |
7c673cae | 5469 | |
11fdf7f2 | 5470 | check_cap_issue(in, new_caps); |
a8e16298 | 5471 | |
7c673cae | 5472 | // update inode |
1adf2230 AA |
5473 | int issued; |
5474 | in->caps_issued(&issued); | |
5475 | issued |= in->caps_dirty(); | |
7c673cae | 5476 | |
1adf2230 AA |
5477 | if ((new_caps & CEPH_CAP_AUTH_SHARED) && |
5478 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
5479 | in->mode = m->head.mode; |
5480 | in->uid = m->head.uid; | |
5481 | in->gid = m->head.gid; | |
5482 | in->btime = m->btime; | |
5483 | } | |
5484 | bool deleted_inode = false; | |
1adf2230 AA |
5485 | if ((new_caps & CEPH_CAP_LINK_SHARED) && |
5486 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
7c673cae FG |
5487 | in->nlink = m->head.nlink; |
5488 | if (in->nlink == 0 && | |
5489 | (new_caps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL))) | |
5490 | deleted_inode = true; | |
5491 | } | |
1adf2230 | 5492 | if (!(issued & CEPH_CAP_XATTR_EXCL) && |
7c673cae FG |
5493 | m->xattrbl.length() && |
5494 | m->head.xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
5495 | auto p = m->xattrbl.cbegin(); |
5496 | decode(in->xattrs, p); | |
7c673cae FG |
5497 | in->xattr_version = m->head.xattr_version; |
5498 | } | |
28e407b8 AA |
5499 | |
5500 | if ((new_caps & CEPH_CAP_FILE_SHARED) && m->dirstat_is_valid()) { | |
5501 | in->dirstat.nfiles = m->get_nfiles(); | |
5502 | in->dirstat.nsubdirs = m->get_nsubdirs(); | |
5503 | } | |
5504 | ||
1adf2230 AA |
5505 | if (new_caps & CEPH_CAP_ANY_RD) { |
5506 | update_inode_file_time(in, issued, m->get_time_warp_seq(), | |
5507 | m->get_ctime(), m->get_mtime(), m->get_atime()); | |
5508 | } | |
5509 | ||
5510 | if (new_caps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { | |
5511 | in->layout = m->get_layout(); | |
5512 | update_inode_file_size(in, issued, m->get_size(), | |
5513 | m->get_truncate_seq(), m->get_truncate_size()); | |
5514 | } | |
5515 | ||
5516 | if (m->inline_version > in->inline_version) { | |
5517 | in->inline_data = m->inline_data; | |
5518 | in->inline_version = m->inline_version; | |
5519 | } | |
5520 | ||
5521 | /* always take a newer change attr */ | |
5522 | if (m->get_change_attr() > in->change_attr) | |
5523 | in->change_attr = m->get_change_attr(); | |
7c673cae FG |
5524 | |
5525 | // max_size | |
5526 | if (cap == in->auth_cap && | |
1adf2230 AA |
5527 | (new_caps & CEPH_CAP_ANY_FILE_WR) && |
5528 | (m->get_max_size() != in->max_size)) { | |
7c673cae FG |
5529 | ldout(cct, 10) << "max_size " << in->max_size << " -> " << m->get_max_size() << dendl; |
5530 | in->max_size = m->get_max_size(); | |
5531 | if (in->max_size > in->wanted_max_size) { | |
5532 | in->wanted_max_size = 0; | |
5533 | in->requested_max_size = 0; | |
5534 | } | |
5535 | } | |
5536 | ||
5537 | bool check = false; | |
a8e16298 TL |
5538 | if ((was_stale || m->get_op() == CEPH_CAP_OP_IMPORT) && |
5539 | (wanted & ~(cap->wanted | new_caps))) { | |
5540 | // If mds is importing cap, prior cap messages that update 'wanted' | |
5541 | // may get dropped by mds (migrate seq mismatch). | |
5542 | // | |
5543 | // We don't send cap message to update 'wanted' if what we want are | |
5544 | // already issued. If mds revokes caps, cap message that releases caps | |
5545 | // also tells mds what we want. But if caps got revoked by mds forcedly | |
5546 | // (session stale). We may haven't told mds what we want. | |
7c673cae | 5547 | check = true; |
a8e16298 | 5548 | } |
7c673cae | 5549 | |
7c673cae FG |
5550 | |
5551 | // update caps | |
a8e16298 | 5552 | auto revoked = cap->issued & ~new_caps; |
b32b8144 FG |
5553 | if (revoked) { |
5554 | ldout(cct, 10) << " revocation of " << ccap_string(revoked) << dendl; | |
7c673cae FG |
5555 | cap->issued = new_caps; |
5556 | cap->implemented |= new_caps; | |
5557 | ||
b32b8144 FG |
5558 | // recall delegations if we're losing caps necessary for them |
5559 | if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_RD)) | |
5560 | in->recall_deleg(false); | |
5561 | else if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_WR)) | |
5562 | in->recall_deleg(true); | |
5563 | ||
11fdf7f2 TL |
5564 | used = adjust_caps_used_for_lazyio(used, cap->issued, cap->implemented); |
5565 | if ((used & revoked & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO)) && | |
28e407b8 | 5566 | !_flush(in, new C_Client_FlushComplete(this, in))) { |
7c673cae | 5567 | // waitin' for flush |
11fdf7f2 | 5568 | } else if (used & revoked & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) { |
7c673cae FG |
5569 | if (_release(in)) |
5570 | check = true; | |
5571 | } else { | |
5572 | cap->wanted = 0; // don't let check_caps skip sending a response to MDS | |
5573 | check = true; | |
5574 | } | |
a8e16298 TL |
5575 | } else if (cap->issued == new_caps) { |
5576 | ldout(cct, 10) << " caps unchanged at " << ccap_string(cap->issued) << dendl; | |
7c673cae | 5577 | } else { |
a8e16298 | 5578 | ldout(cct, 10) << " grant, new caps are " << ccap_string(new_caps & ~cap->issued) << dendl; |
7c673cae FG |
5579 | cap->issued = new_caps; |
5580 | cap->implemented |= new_caps; | |
5581 | ||
5582 | if (cap == in->auth_cap) { | |
5583 | // non-auth MDS is revoking the newly grant caps ? | |
11fdf7f2 TL |
5584 | for (const auto &p : in->caps) { |
5585 | if (&p.second == cap) | |
7c673cae | 5586 | continue; |
11fdf7f2 | 5587 | if (p.second.implemented & ~p.second.issued & new_caps) { |
7c673cae FG |
5588 | check = true; |
5589 | break; | |
5590 | } | |
5591 | } | |
5592 | } | |
5593 | } | |
5594 | ||
5595 | if (check) | |
5596 | check_caps(in, 0); | |
5597 | ||
5598 | // wake up waiters | |
5599 | if (new_caps) | |
5600 | signal_cond_list(in->waitfor_caps); | |
5601 | ||
5602 | // may drop inode's last ref | |
5603 | if (deleted_inode) | |
5604 | _try_to_trim_inode(in, true); | |
7c673cae FG |
5605 | } |
5606 | ||
7c673cae FG |
5607 | int Client::inode_permission(Inode *in, const UserPerm& perms, unsigned want) |
5608 | { | |
b3b6e05e TL |
5609 | if (perms.uid() == 0) { |
5610 | // Executable are overridable when there is at least one exec bit set | |
5611 | if((want & MAY_EXEC) && !(in->mode & S_IXUGO)) | |
5612 | return -CEPHFS_EACCES; | |
7c673cae | 5613 | return 0; |
b3b6e05e | 5614 | } |
7c673cae FG |
5615 | |
5616 | if (perms.uid() != in->uid && (in->mode & S_IRWXG)) { | |
5617 | int ret = _posix_acl_permission(in, perms, want); | |
f67539c2 | 5618 | if (ret != -CEPHFS_EAGAIN) |
7c673cae FG |
5619 | return ret; |
5620 | } | |
5621 | ||
5622 | // check permissions before doing anything else | |
5623 | if (!in->check_mode(perms, want)) | |
f67539c2 | 5624 | return -CEPHFS_EACCES; |
7c673cae FG |
5625 | return 0; |
5626 | } | |
5627 | ||
5628 | int Client::xattr_permission(Inode *in, const char *name, unsigned want, | |
5629 | const UserPerm& perms) | |
5630 | { | |
5631 | int r = _getattr_for_perm(in, perms); | |
5632 | if (r < 0) | |
5633 | goto out; | |
5634 | ||
5635 | r = 0; | |
5636 | if (strncmp(name, "system.", 7) == 0) { | |
5637 | if ((want & MAY_WRITE) && (perms.uid() != 0 && perms.uid() != in->uid)) | |
f67539c2 | 5638 | r = -CEPHFS_EPERM; |
7c673cae FG |
5639 | } else { |
5640 | r = inode_permission(in, perms, want); | |
5641 | } | |
5642 | out: | |
1adf2230 | 5643 | ldout(cct, 5) << __func__ << " " << in << " = " << r << dendl; |
7c673cae FG |
5644 | return r; |
5645 | } | |
5646 | ||
5647 | ostream& operator<<(ostream &out, const UserPerm& perm) { | |
5648 | out << "UserPerm(uid: " << perm.uid() << ", gid: " << perm.gid() << ")"; | |
5649 | return out; | |
5650 | } | |
5651 | ||
5652 | int Client::may_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
5653 | const UserPerm& perms) | |
5654 | { | |
181888fb | 5655 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5656 | int r = _getattr_for_perm(in, perms); |
5657 | if (r < 0) | |
5658 | goto out; | |
5659 | ||
5660 | if (mask & CEPH_SETATTR_SIZE) { | |
5661 | r = inode_permission(in, perms, MAY_WRITE); | |
5662 | if (r < 0) | |
5663 | goto out; | |
5664 | } | |
5665 | ||
f67539c2 | 5666 | r = -CEPHFS_EPERM; |
7c673cae FG |
5667 | if (mask & CEPH_SETATTR_UID) { |
5668 | if (perms.uid() != 0 && (perms.uid() != in->uid || stx->stx_uid != in->uid)) | |
5669 | goto out; | |
5670 | } | |
5671 | if (mask & CEPH_SETATTR_GID) { | |
5672 | if (perms.uid() != 0 && (perms.uid() != in->uid || | |
5673 | (!perms.gid_in_groups(stx->stx_gid) && stx->stx_gid != in->gid))) | |
5674 | goto out; | |
5675 | } | |
5676 | ||
5677 | if (mask & CEPH_SETATTR_MODE) { | |
5678 | if (perms.uid() != 0 && perms.uid() != in->uid) | |
5679 | goto out; | |
5680 | ||
5681 | gid_t i_gid = (mask & CEPH_SETATTR_GID) ? stx->stx_gid : in->gid; | |
5682 | if (perms.uid() != 0 && !perms.gid_in_groups(i_gid)) | |
5683 | stx->stx_mode &= ~S_ISGID; | |
5684 | } | |
5685 | ||
5686 | if (mask & (CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME | | |
5687 | CEPH_SETATTR_MTIME | CEPH_SETATTR_ATIME)) { | |
5688 | if (perms.uid() != 0 && perms.uid() != in->uid) { | |
5689 | int check_mask = CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME; | |
5690 | if (!(mask & CEPH_SETATTR_MTIME_NOW)) | |
5691 | check_mask |= CEPH_SETATTR_MTIME; | |
5692 | if (!(mask & CEPH_SETATTR_ATIME_NOW)) | |
5693 | check_mask |= CEPH_SETATTR_ATIME; | |
5694 | if (check_mask & mask) { | |
5695 | goto out; | |
5696 | } else { | |
5697 | r = inode_permission(in, perms, MAY_WRITE); | |
5698 | if (r < 0) | |
5699 | goto out; | |
5700 | } | |
5701 | } | |
5702 | } | |
5703 | r = 0; | |
5704 | out: | |
5705 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5706 | return r; | |
5707 | } | |
5708 | ||
5709 | int Client::may_open(Inode *in, int flags, const UserPerm& perms) | |
5710 | { | |
181888fb | 5711 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5712 | unsigned want = 0; |
5713 | ||
5714 | if ((flags & O_ACCMODE) == O_WRONLY) | |
5715 | want = MAY_WRITE; | |
5716 | else if ((flags & O_ACCMODE) == O_RDWR) | |
5717 | want = MAY_READ | MAY_WRITE; | |
5718 | else if ((flags & O_ACCMODE) == O_RDONLY) | |
5719 | want = MAY_READ; | |
5720 | if (flags & O_TRUNC) | |
5721 | want |= MAY_WRITE; | |
5722 | ||
5723 | int r = 0; | |
5724 | switch (in->mode & S_IFMT) { | |
5725 | case S_IFLNK: | |
f67539c2 | 5726 | r = -CEPHFS_ELOOP; |
7c673cae FG |
5727 | goto out; |
5728 | case S_IFDIR: | |
5729 | if (want & MAY_WRITE) { | |
f67539c2 | 5730 | r = -CEPHFS_EISDIR; |
7c673cae FG |
5731 | goto out; |
5732 | } | |
5733 | break; | |
5734 | } | |
5735 | ||
5736 | r = _getattr_for_perm(in, perms); | |
5737 | if (r < 0) | |
5738 | goto out; | |
5739 | ||
5740 | r = inode_permission(in, perms, want); | |
5741 | out: | |
5742 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5743 | return r; | |
5744 | } | |
5745 | ||
5746 | int Client::may_lookup(Inode *dir, const UserPerm& perms) | |
5747 | { | |
181888fb | 5748 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5749 | int r = _getattr_for_perm(dir, perms); |
5750 | if (r < 0) | |
5751 | goto out; | |
5752 | ||
5753 | r = inode_permission(dir, perms, MAY_EXEC); | |
5754 | out: | |
5755 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5756 | return r; | |
5757 | } | |
5758 | ||
5759 | int Client::may_create(Inode *dir, const UserPerm& perms) | |
5760 | { | |
181888fb | 5761 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5762 | int r = _getattr_for_perm(dir, perms); |
5763 | if (r < 0) | |
5764 | goto out; | |
5765 | ||
5766 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5767 | out: | |
5768 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5769 | return r; | |
5770 | } | |
5771 | ||
5772 | int Client::may_delete(Inode *dir, const char *name, const UserPerm& perms) | |
5773 | { | |
181888fb | 5774 | ldout(cct, 20) << __func__ << " " << *dir << "; " << "; name " << name << "; " << perms << dendl; |
7c673cae FG |
5775 | int r = _getattr_for_perm(dir, perms); |
5776 | if (r < 0) | |
5777 | goto out; | |
5778 | ||
5779 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5780 | if (r < 0) | |
5781 | goto out; | |
5782 | ||
f67539c2 | 5783 | /* 'name == NULL' means rmsnap w/o permission checks */ |
7c673cae FG |
5784 | if (perms.uid() != 0 && name && (dir->mode & S_ISVTX)) { |
5785 | InodeRef otherin; | |
5786 | r = _lookup(dir, name, CEPH_CAP_AUTH_SHARED, &otherin, perms); | |
5787 | if (r < 0) | |
5788 | goto out; | |
5789 | if (dir->uid != perms.uid() && otherin->uid != perms.uid()) | |
f67539c2 | 5790 | r = -CEPHFS_EPERM; |
7c673cae FG |
5791 | } |
5792 | out: | |
5793 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5794 | return r; | |
5795 | } | |
5796 | ||
f67539c2 TL |
5797 | int Client::may_delete(const char *relpath, const UserPerm& perms) { |
5798 | ldout(cct, 20) << __func__ << " " << relpath << "; " << perms << dendl; | |
5799 | ||
5800 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
5801 | if (!mref_reader.is_state_satisfied()) | |
5802 | return -ENOTCONN; | |
5803 | ||
5804 | filepath path(relpath); | |
5805 | string name = path.last_dentry(); | |
5806 | path.pop_dentry(); | |
5807 | InodeRef dir; | |
5808 | ||
5809 | std::scoped_lock lock(client_lock); | |
5810 | int r = path_walk(path, &dir, perms); | |
5811 | if (r < 0) | |
5812 | return r; | |
5813 | if (cct->_conf->client_permissions) { | |
5814 | int r = may_delete(dir.get(), name.c_str(), perms); | |
5815 | if (r < 0) | |
5816 | return r; | |
5817 | } | |
5818 | ||
5819 | return 0; | |
5820 | } | |
5821 | ||
7c673cae FG |
5822 | int Client::may_hardlink(Inode *in, const UserPerm& perms) |
5823 | { | |
181888fb | 5824 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5825 | int r = _getattr_for_perm(in, perms); |
5826 | if (r < 0) | |
5827 | goto out; | |
5828 | ||
5829 | if (perms.uid() == 0 || perms.uid() == in->uid) { | |
5830 | r = 0; | |
5831 | goto out; | |
5832 | } | |
5833 | ||
f67539c2 | 5834 | r = -CEPHFS_EPERM; |
7c673cae FG |
5835 | if (!S_ISREG(in->mode)) |
5836 | goto out; | |
5837 | ||
5838 | if (in->mode & S_ISUID) | |
5839 | goto out; | |
5840 | ||
5841 | if ((in->mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
5842 | goto out; | |
5843 | ||
5844 | r = inode_permission(in, perms, MAY_READ | MAY_WRITE); | |
5845 | out: | |
5846 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5847 | return r; | |
5848 | } | |
5849 | ||
5850 | int Client::_getattr_for_perm(Inode *in, const UserPerm& perms) | |
5851 | { | |
5852 | int mask = CEPH_STAT_CAP_MODE; | |
5853 | bool force = false; | |
5854 | if (acl_type != NO_ACL) { | |
5855 | mask |= CEPH_STAT_CAP_XATTR; | |
5856 | force = in->xattr_version == 0; | |
5857 | } | |
5858 | return _getattr(in, mask, perms, force); | |
5859 | } | |
5860 | ||
5861 | vinodeno_t Client::_get_vino(Inode *in) | |
5862 | { | |
5863 | /* The caller must hold the client lock */ | |
5864 | return vinodeno_t(in->ino, in->snapid); | |
5865 | } | |
5866 | ||
7c673cae FG |
5867 | /** |
5868 | * Resolve an MDS spec to a list of MDS daemon GIDs. | |
5869 | * | |
5870 | * The spec is a string representing a GID, rank, filesystem:rank, or name/id. | |
5871 | * It may be '*' in which case it matches all GIDs. | |
5872 | * | |
5873 | * If no error is returned, the `targets` vector will be populated with at least | |
5874 | * one MDS. | |
5875 | */ | |
5876 | int Client::resolve_mds( | |
5877 | const std::string &mds_spec, | |
5878 | std::vector<mds_gid_t> *targets) | |
5879 | { | |
11fdf7f2 TL |
5880 | ceph_assert(fsmap); |
5881 | ceph_assert(targets != nullptr); | |
7c673cae FG |
5882 | |
5883 | mds_role_t role; | |
f67539c2 TL |
5884 | CachedStackStringStream css; |
5885 | int role_r = fsmap->parse_role(mds_spec, &role, *css); | |
7c673cae FG |
5886 | if (role_r == 0) { |
5887 | // We got a role, resolve it to a GID | |
f67539c2 TL |
5888 | auto& info = fsmap->get_filesystem(role.fscid)->mds_map.get_info(role.rank); |
5889 | ldout(cct, 10) << __func__ << ": resolved " << mds_spec << " to role '" | |
5890 | << role << "' aka " << info.human_name() << dendl; | |
5891 | targets->push_back(info.global_id); | |
7c673cae FG |
5892 | return 0; |
5893 | } | |
5894 | ||
5895 | std::string strtol_err; | |
5896 | long long rank_or_gid = strict_strtoll(mds_spec.c_str(), 10, &strtol_err); | |
5897 | if (strtol_err.empty()) { | |
5898 | // It is a possible GID | |
5899 | const mds_gid_t mds_gid = mds_gid_t(rank_or_gid); | |
5900 | if (fsmap->gid_exists(mds_gid)) { | |
f67539c2 TL |
5901 | auto& info = fsmap->get_info_gid(mds_gid); |
5902 | ldout(cct, 10) << __func__ << ": validated gid " << mds_gid << " aka " | |
5903 | << info.human_name() << dendl; | |
7c673cae | 5904 | targets->push_back(mds_gid); |
f67539c2 | 5905 | return 0; |
7c673cae | 5906 | } else { |
f67539c2 | 5907 | lderr(cct) << __func__ << ": gid " << mds_gid << " not in MDS map" |
7c673cae | 5908 | << dendl; |
f67539c2 TL |
5909 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
5910 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5911 | } |
5912 | } else if (mds_spec == "*") { | |
5913 | // It is a wildcard: use all MDSs | |
f67539c2 | 5914 | const auto& mds_info = fsmap->get_mds_info(); |
7c673cae | 5915 | |
f67539c2 | 5916 | ldout(cct, 10) << __func__ << ": resolving `*' to all MDS daemons" << dendl; |
7c673cae | 5917 | if (mds_info.empty()) { |
f67539c2 TL |
5918 | lderr(cct) << __func__ << ": no MDS daemons found" << dendl; |
5919 | lderr(cct) << "FSMap: " << *fsmap << dendl; | |
5920 | return -CEPHFS_ENOENT; | |
7c673cae FG |
5921 | } |
5922 | ||
f67539c2 TL |
5923 | for (const auto& [gid, info] : mds_info) { |
5924 | ldout(cct, 10) << __func__ << ": appending " << info.human_name() << " to targets" << dendl; | |
5925 | targets->push_back(gid); | |
7c673cae | 5926 | } |
f67539c2 | 5927 | return 0; |
7c673cae FG |
5928 | } else { |
5929 | // It did not parse as an integer, it is not a wildcard, it must be a name | |
5930 | const mds_gid_t mds_gid = fsmap->find_mds_gid_by_name(mds_spec); | |
5931 | if (mds_gid == 0) { | |
f67539c2 | 5932 | lderr(cct) << __func__ << ": no MDS daemons found by name `" << mds_spec << "'" << dendl; |
7c673cae | 5933 | lderr(cct) << "FSMap: " << *fsmap << dendl; |
f67539c2 | 5934 | return -CEPHFS_ENOENT; |
7c673cae | 5935 | } else { |
f67539c2 TL |
5936 | auto& info = fsmap->get_info_gid(mds_gid); |
5937 | ldout(cct, 10) << __func__ << ": resolved name '" << mds_spec | |
5938 | << "' to " << info.human_name() << dendl; | |
7c673cae FG |
5939 | targets->push_back(mds_gid); |
5940 | } | |
f67539c2 | 5941 | return 0; |
7c673cae | 5942 | } |
7c673cae FG |
5943 | } |
5944 | ||
5945 | ||
5946 | /** | |
5947 | * Authenticate with mon and establish global ID | |
5948 | */ | |
5949 | int Client::authenticate() | |
5950 | { | |
9f95a23c | 5951 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
5952 | |
5953 | if (monclient->is_authenticated()) { | |
5954 | return 0; | |
5955 | } | |
5956 | ||
9f95a23c | 5957 | client_lock.unlock(); |
7c673cae | 5958 | int r = monclient->authenticate(cct->_conf->client_mount_timeout); |
9f95a23c | 5959 | client_lock.lock(); |
7c673cae FG |
5960 | if (r < 0) { |
5961 | return r; | |
5962 | } | |
5963 | ||
5964 | whoami = monclient->get_global_id(); | |
5965 | messenger->set_myname(entity_name_t::CLIENT(whoami.v)); | |
5966 | ||
5967 | return 0; | |
5968 | } | |
5969 | ||
5970 | int Client::fetch_fsmap(bool user) | |
5971 | { | |
f67539c2 TL |
5972 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
5973 | ||
7c673cae FG |
5974 | // Retrieve FSMap to enable looking up daemon addresses. We need FSMap |
5975 | // rather than MDSMap because no one MDSMap contains all the daemons, and | |
5976 | // a `tell` can address any daemon. | |
5977 | version_t fsmap_latest; | |
f67539c2 | 5978 | bs::error_code ec; |
7c673cae | 5979 | do { |
9f95a23c | 5980 | client_lock.unlock(); |
f67539c2 TL |
5981 | std::tie(fsmap_latest, std::ignore) = |
5982 | monclient->get_version("fsmap", ca::use_blocked[ec]); | |
9f95a23c | 5983 | client_lock.lock(); |
f67539c2 | 5984 | } while (ec == bs::errc::resource_unavailable_try_again); |
7c673cae | 5985 | |
f67539c2 TL |
5986 | if (ec) { |
5987 | lderr(cct) << "Failed to learn FSMap version: " << ec << dendl; | |
5988 | return ceph::from_error_code(ec); | |
7c673cae FG |
5989 | } |
5990 | ||
5991 | ldout(cct, 10) << __func__ << " learned FSMap version " << fsmap_latest << dendl; | |
5992 | ||
5993 | if (user) { | |
5994 | if (!fsmap_user || fsmap_user->get_epoch() < fsmap_latest) { | |
5995 | monclient->sub_want("fsmap.user", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5996 | monclient->renew_subs(); | |
5997 | wait_on_list(waiting_for_fsmap); | |
5998 | } | |
11fdf7f2 TL |
5999 | ceph_assert(fsmap_user); |
6000 | ceph_assert(fsmap_user->get_epoch() >= fsmap_latest); | |
7c673cae FG |
6001 | } else { |
6002 | if (!fsmap || fsmap->get_epoch() < fsmap_latest) { | |
6003 | monclient->sub_want("fsmap", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
6004 | monclient->renew_subs(); | |
6005 | wait_on_list(waiting_for_fsmap); | |
6006 | } | |
11fdf7f2 TL |
6007 | ceph_assert(fsmap); |
6008 | ceph_assert(fsmap->get_epoch() >= fsmap_latest); | |
7c673cae FG |
6009 | } |
6010 | ldout(cct, 10) << __func__ << " finished waiting for FSMap version " | |
6011 | << fsmap_latest << dendl; | |
6012 | return 0; | |
6013 | } | |
6014 | ||
6015 | /** | |
6016 | * | |
6017 | * @mds_spec one of ID, rank, GID, "*" | |
6018 | * | |
6019 | */ | |
6020 | int Client::mds_command( | |
6021 | const std::string &mds_spec, | |
6022 | const vector<string>& cmd, | |
6023 | const bufferlist& inbl, | |
6024 | bufferlist *outbl, | |
6025 | string *outs, | |
6026 | Context *onfinish) | |
6027 | { | |
f67539c2 TL |
6028 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
6029 | if (!iref_reader.is_state_satisfied()) | |
6030 | return -CEPHFS_ENOTCONN; | |
7c673cae | 6031 | |
f67539c2 | 6032 | std::unique_lock cl(client_lock); |
7c673cae FG |
6033 | |
6034 | int r; | |
6035 | r = authenticate(); | |
6036 | if (r < 0) { | |
6037 | return r; | |
6038 | } | |
6039 | ||
6040 | r = fetch_fsmap(false); | |
6041 | if (r < 0) { | |
6042 | return r; | |
6043 | } | |
6044 | ||
6045 | // Look up MDS target(s) of the command | |
6046 | std::vector<mds_gid_t> targets; | |
6047 | r = resolve_mds(mds_spec, &targets); | |
6048 | if (r < 0) { | |
6049 | return r; | |
6050 | } | |
6051 | ||
6052 | // If daemons are laggy, we won't send them commands. If all | |
6053 | // are laggy then we fail. | |
6054 | std::vector<mds_gid_t> non_laggy; | |
f67539c2 | 6055 | for (const auto& gid : targets) { |
7c673cae FG |
6056 | const auto info = fsmap->get_info_gid(gid); |
6057 | if (!info.laggy()) { | |
6058 | non_laggy.push_back(gid); | |
6059 | } | |
6060 | } | |
6061 | if (non_laggy.size() == 0) { | |
6062 | *outs = "All targeted MDS daemons are laggy"; | |
f67539c2 | 6063 | return -CEPHFS_ENOENT; |
7c673cae FG |
6064 | } |
6065 | ||
6066 | if (metadata.empty()) { | |
6067 | // We are called on an unmounted client, so metadata | |
6068 | // won't be initialized yet. | |
6069 | populate_metadata(""); | |
6070 | } | |
6071 | ||
6072 | // Send commands to targets | |
6073 | C_GatherBuilder gather(cct, onfinish); | |
f67539c2 | 6074 | for (const auto& target_gid : non_laggy) { |
7c673cae FG |
6075 | const auto info = fsmap->get_info_gid(target_gid); |
6076 | ||
6077 | // Open a connection to the target MDS | |
11fdf7f2 | 6078 | ConnectionRef conn = messenger->connect_to_mds(info.get_addrs()); |
7c673cae | 6079 | |
f67539c2 TL |
6080 | cl.unlock(); |
6081 | { | |
6082 | std::scoped_lock cmd_lock(command_lock); | |
6083 | // Generate MDSCommandOp state | |
6084 | auto &op = command_table.start_command(); | |
7c673cae | 6085 | |
f67539c2 TL |
6086 | op.on_finish = gather.new_sub(); |
6087 | op.cmd = cmd; | |
6088 | op.outbl = outbl; | |
6089 | op.outs = outs; | |
6090 | op.inbl = inbl; | |
6091 | op.mds_gid = target_gid; | |
6092 | op.con = conn; | |
7c673cae | 6093 | |
f67539c2 TL |
6094 | ldout(cct, 4) << __func__ << ": new command op to " << target_gid |
6095 | << " tid=" << op.tid << cmd << dendl; | |
7c673cae | 6096 | |
f67539c2 TL |
6097 | // Construct and send MCommand |
6098 | MessageRef m = op.get_message(monclient->get_fsid()); | |
6099 | conn->send_message2(std::move(m)); | |
6100 | } | |
6101 | cl.lock(); | |
7c673cae FG |
6102 | } |
6103 | gather.activate(); | |
6104 | ||
6105 | return 0; | |
6106 | } | |
6107 | ||
11fdf7f2 | 6108 | void Client::handle_command_reply(const MConstRef<MCommandReply>& m) |
7c673cae FG |
6109 | { |
6110 | ceph_tid_t const tid = m->get_tid(); | |
6111 | ||
6112 | ldout(cct, 10) << __func__ << ": tid=" << m->get_tid() << dendl; | |
6113 | ||
f67539c2 | 6114 | std::scoped_lock cmd_lock(command_lock); |
7c673cae FG |
6115 | if (!command_table.exists(tid)) { |
6116 | ldout(cct, 1) << __func__ << ": unknown tid " << tid << ", dropping" << dendl; | |
7c673cae FG |
6117 | return; |
6118 | } | |
6119 | ||
6120 | auto &op = command_table.get_command(tid); | |
6121 | if (op.outbl) { | |
11fdf7f2 | 6122 | *op.outbl = m->get_data(); |
7c673cae FG |
6123 | } |
6124 | if (op.outs) { | |
6125 | *op.outs = m->rs; | |
6126 | } | |
6127 | ||
6128 | if (op.on_finish) { | |
6129 | op.on_finish->complete(m->r); | |
6130 | } | |
6131 | ||
6132 | command_table.erase(tid); | |
7c673cae FG |
6133 | } |
6134 | ||
6135 | // ------------------- | |
6136 | // MOUNT | |
6137 | ||
11fdf7f2 | 6138 | int Client::subscribe_mdsmap(const std::string &fs_name) |
7c673cae | 6139 | { |
7c673cae FG |
6140 | int r = authenticate(); |
6141 | if (r < 0) { | |
6142 | lderr(cct) << "authentication failed: " << cpp_strerror(r) << dendl; | |
6143 | return r; | |
6144 | } | |
6145 | ||
11fdf7f2 TL |
6146 | std::string resolved_fs_name; |
6147 | if (fs_name.empty()) { | |
9f95a23c TL |
6148 | resolved_fs_name = cct->_conf.get_val<std::string>("client_fs"); |
6149 | if (resolved_fs_name.empty()) | |
6150 | // Try the backwards compatibility fs name option | |
6151 | resolved_fs_name = cct->_conf.get_val<std::string>("client_mds_namespace"); | |
11fdf7f2 TL |
6152 | } else { |
6153 | resolved_fs_name = fs_name; | |
6154 | } | |
6155 | ||
7c673cae | 6156 | std::string want = "mdsmap"; |
11fdf7f2 | 6157 | if (!resolved_fs_name.empty()) { |
7c673cae FG |
6158 | r = fetch_fsmap(true); |
6159 | if (r < 0) | |
6160 | return r; | |
11fdf7f2 TL |
6161 | fscid = fsmap_user->get_fs_cid(resolved_fs_name); |
6162 | if (fscid == FS_CLUSTER_ID_NONE) { | |
f67539c2 | 6163 | return -CEPHFS_ENOENT; |
11fdf7f2 | 6164 | } |
7c673cae FG |
6165 | |
6166 | std::ostringstream oss; | |
11fdf7f2 | 6167 | oss << want << "." << fscid; |
7c673cae FG |
6168 | want = oss.str(); |
6169 | } | |
6170 | ldout(cct, 10) << "Subscribing to map '" << want << "'" << dendl; | |
6171 | ||
6172 | monclient->sub_want(want, 0, 0); | |
6173 | monclient->renew_subs(); | |
6174 | ||
11fdf7f2 TL |
6175 | return 0; |
6176 | } | |
6177 | ||
6178 | int Client::mount(const std::string &mount_root, const UserPerm& perms, | |
6179 | bool require_mds, const std::string &fs_name) | |
6180 | { | |
f67539c2 | 6181 | ceph_assert(is_initialized()); |
11fdf7f2 | 6182 | |
f67539c2 TL |
6183 | /* |
6184 | * To make sure that the _unmount() must wait until the mount() | |
6185 | * is done. | |
6186 | */ | |
6187 | RWRef_t mref_writer(mount_state, CLIENT_MOUNTING, false); | |
6188 | if (!mref_writer.is_first_writer()) // already mounting or mounted | |
11fdf7f2 | 6189 | return 0; |
11fdf7f2 | 6190 | |
f67539c2 | 6191 | std::unique_lock cl(client_lock); |
11fdf7f2 TL |
6192 | |
6193 | int r = subscribe_mdsmap(fs_name); | |
6194 | if (r < 0) { | |
6195 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
6196 | return r; | |
6197 | } | |
6198 | ||
f67539c2 TL |
6199 | start_tick_thread(); // start tick thread |
6200 | ||
7c673cae FG |
6201 | if (require_mds) { |
6202 | while (1) { | |
6203 | auto availability = mdsmap->is_cluster_available(); | |
6204 | if (availability == MDSMap::STUCK_UNAVAILABLE) { | |
6205 | // Error out | |
6206 | ldout(cct, 10) << "mds cluster unavailable: epoch=" << mdsmap->get_epoch() << dendl; | |
6207 | return CEPH_FUSE_NO_MDS_UP; | |
6208 | } else if (availability == MDSMap::AVAILABLE) { | |
6209 | // Continue to mount | |
6210 | break; | |
6211 | } else if (availability == MDSMap::TRANSIENT_UNAVAILABLE) { | |
6212 | // Else, wait. MDSMonitor will update the map to bring | |
6213 | // us to a conclusion eventually. | |
6214 | wait_on_list(waiting_for_mdsmap); | |
6215 | } else { | |
6216 | // Unexpected value! | |
6217 | ceph_abort(); | |
6218 | } | |
6219 | } | |
6220 | } | |
6221 | ||
6222 | populate_metadata(mount_root.empty() ? "/" : mount_root); | |
6223 | ||
6224 | filepath fp(CEPH_INO_ROOT); | |
6225 | if (!mount_root.empty()) { | |
6226 | fp = filepath(mount_root.c_str()); | |
6227 | } | |
6228 | while (true) { | |
6229 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
6230 | req->set_filepath(fp); | |
6231 | req->head.args.getattr.mask = CEPH_STAT_CAP_INODE_ALL; | |
6232 | int res = make_request(req, perms); | |
6233 | if (res < 0) { | |
f67539c2 | 6234 | if (res == -CEPHFS_EACCES && root) { |
7c673cae FG |
6235 | ldout(cct, 1) << __func__ << " EACCES on parent of mount point; quotas may not work" << dendl; |
6236 | break; | |
6237 | } | |
6238 | return res; | |
6239 | } | |
6240 | ||
6241 | if (fp.depth()) | |
6242 | fp.pop_dentry(); | |
6243 | else | |
6244 | break; | |
6245 | } | |
6246 | ||
11fdf7f2 | 6247 | ceph_assert(root); |
b3b6e05e | 6248 | _ll_get(root.get()); |
7c673cae | 6249 | |
7c673cae FG |
6250 | // trace? |
6251 | if (!cct->_conf->client_trace.empty()) { | |
6252 | traceout.open(cct->_conf->client_trace.c_str()); | |
6253 | if (traceout.is_open()) { | |
6254 | ldout(cct, 1) << "opened trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6255 | } else { | |
6256 | ldout(cct, 1) << "FAILED to open trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6257 | } | |
6258 | } | |
6259 | ||
6260 | /* | |
6261 | ldout(cct, 3) << "op: // client trace data structs" << dendl; | |
6262 | ldout(cct, 3) << "op: struct stat st;" << dendl; | |
6263 | ldout(cct, 3) << "op: struct utimbuf utim;" << dendl; | |
6264 | ldout(cct, 3) << "op: int readlinkbuf_len = 1000;" << dendl; | |
6265 | ldout(cct, 3) << "op: char readlinkbuf[readlinkbuf_len];" << dendl; | |
6266 | ldout(cct, 3) << "op: map<string, inode_t*> dir_contents;" << dendl; | |
6267 | ldout(cct, 3) << "op: map<int, int> open_files;" << dendl; | |
6268 | ldout(cct, 3) << "op: int fd;" << dendl; | |
6269 | */ | |
f67539c2 TL |
6270 | |
6271 | mref_writer.update_state(CLIENT_MOUNTED); | |
7c673cae FG |
6272 | return 0; |
6273 | } | |
6274 | ||
6275 | // UNMOUNT | |
6276 | ||
6277 | void Client::_close_sessions() | |
6278 | { | |
f6b5b4d7 TL |
6279 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
6280 | if (it->second.state == MetaSession::STATE_REJECTED) | |
6281 | mds_sessions.erase(it++); | |
6282 | else | |
6283 | ++it; | |
6284 | } | |
6285 | ||
7c673cae FG |
6286 | while (!mds_sessions.empty()) { |
6287 | // send session closes! | |
11fdf7f2 TL |
6288 | for (auto &p : mds_sessions) { |
6289 | if (p.second.state != MetaSession::STATE_CLOSING) { | |
6290 | _close_mds_session(&p.second); | |
f6b5b4d7 | 6291 | mds_ranks_closing.insert(p.first); |
7c673cae FG |
6292 | } |
6293 | } | |
6294 | ||
6295 | // wait for sessions to close | |
f6b5b4d7 TL |
6296 | double timo = cct->_conf.get_val<std::chrono::seconds>("client_shutdown_timeout").count(); |
6297 | ldout(cct, 2) << "waiting for " << mds_ranks_closing.size() << " mds session(s) to close (timeout: " | |
6298 | << timo << "s)" << dendl; | |
9f95a23c | 6299 | std::unique_lock l{client_lock, std::adopt_lock}; |
f6b5b4d7 TL |
6300 | if (!timo) { |
6301 | mount_cond.wait(l); | |
6302 | } else if (!mount_cond.wait_for(l, ceph::make_timespan(timo), [this] { return mds_ranks_closing.empty(); })) { | |
6303 | ldout(cct, 1) << mds_ranks_closing.size() << " mds(s) did not respond to session close -- timing out." << dendl; | |
6304 | while (!mds_ranks_closing.empty()) { | |
6305 | auto session = mds_sessions.at(*mds_ranks_closing.begin()); | |
6306 | // this prunes entry from mds_sessions and mds_ranks_closing | |
f67539c2 | 6307 | _closed_mds_session(&session, -CEPHFS_ETIMEDOUT); |
f6b5b4d7 TL |
6308 | } |
6309 | } | |
6310 | ||
6311 | mds_ranks_closing.clear(); | |
9f95a23c | 6312 | l.release(); |
7c673cae FG |
6313 | } |
6314 | } | |
6315 | ||
31f18b77 FG |
6316 | void Client::flush_mdlog_sync() |
6317 | { | |
6318 | if (mds_requests.empty()) | |
6319 | return; | |
11fdf7f2 TL |
6320 | for (auto &p : mds_sessions) { |
6321 | flush_mdlog(&p.second); | |
31f18b77 FG |
6322 | } |
6323 | } | |
6324 | ||
6325 | void Client::flush_mdlog(MetaSession *session) | |
6326 | { | |
6327 | // Only send this to Luminous or newer MDS daemons, older daemons | |
6328 | // will crash if they see an unknown CEPH_SESSION_* value in this msg. | |
6329 | const uint64_t features = session->con->get_features(); | |
6330 | if (HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
9f95a23c | 6331 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_FLUSH_MDLOG); |
11fdf7f2 | 6332 | session->con->send_message2(std::move(m)); |
31f18b77 FG |
6333 | } |
6334 | } | |
6335 | ||
6336 | ||
11fdf7f2 TL |
6337 | void Client::_abort_mds_sessions(int err) |
6338 | { | |
6339 | for (auto p = mds_requests.begin(); p != mds_requests.end(); ) { | |
6340 | auto req = p->second; | |
6341 | ++p; | |
6342 | // unsafe requests will be removed during close session below. | |
6343 | if (req->got_unsafe) | |
6344 | continue; | |
6345 | ||
6346 | req->abort(err); | |
6347 | if (req->caller_cond) { | |
6348 | req->kick = true; | |
9f95a23c | 6349 | req->caller_cond->notify_all(); |
11fdf7f2 TL |
6350 | } |
6351 | } | |
6352 | ||
6353 | // Process aborts on any requests that were on this waitlist. | |
6354 | // Any requests that were on a waiting_for_open session waitlist | |
6355 | // will get kicked during close session below. | |
6356 | signal_cond_list(waiting_for_mdsmap); | |
6357 | ||
6358 | // Force-close all sessions | |
6359 | while(!mds_sessions.empty()) { | |
6360 | auto& session = mds_sessions.begin()->second; | |
f6b5b4d7 | 6361 | _closed_mds_session(&session, err); |
11fdf7f2 TL |
6362 | } |
6363 | } | |
6364 | ||
6365 | void Client::_unmount(bool abort) | |
7c673cae | 6366 | { |
f67539c2 TL |
6367 | /* |
6368 | * We are unmounting the client. | |
6369 | * | |
6370 | * Just declare the state to STATE_UNMOUNTING to block and fail | |
6371 | * any new comming "reader" and then try to wait all the in-flight | |
6372 | * "readers" to finish. | |
6373 | */ | |
6374 | RWRef_t mref_writer(mount_state, CLIENT_UNMOUNTING, false); | |
6375 | if (!mref_writer.is_first_writer()) | |
181888fb | 6376 | return; |
f67539c2 | 6377 | mref_writer.wait_readers_done(); |
7c673cae | 6378 | |
f67539c2 TL |
6379 | std::unique_lock lock{client_lock}; |
6380 | ||
6381 | if (abort || blocklisted) { | |
6382 | ldout(cct, 2) << "unmounting (" << (abort ? "abort)" : "blocklisted)") << dendl; | |
11fdf7f2 TL |
6383 | } else { |
6384 | ldout(cct, 2) << "unmounting" << dendl; | |
6385 | } | |
7c673cae | 6386 | |
b32b8144 FG |
6387 | deleg_timeout = 0; |
6388 | ||
11fdf7f2 | 6389 | if (abort) { |
f67539c2 | 6390 | mount_aborted = true; |
11fdf7f2 | 6391 | // Abort all mds sessions |
f67539c2 | 6392 | _abort_mds_sessions(-CEPHFS_ENOTCONN); |
11fdf7f2 | 6393 | |
f67539c2 | 6394 | objecter->op_cancel_writes(-CEPHFS_ENOTCONN); |
11fdf7f2 TL |
6395 | } else { |
6396 | // flush the mdlog for pending requests, if any | |
6397 | flush_mdlog_sync(); | |
6398 | } | |
6399 | ||
9f95a23c TL |
6400 | mount_cond.wait(lock, [this] { |
6401 | if (!mds_requests.empty()) { | |
6402 | ldout(cct, 10) << "waiting on " << mds_requests.size() << " requests" | |
6403 | << dendl; | |
6404 | } | |
6405 | return mds_requests.empty(); | |
6406 | }); | |
7c673cae FG |
6407 | |
6408 | cwd.reset(); | |
b3b6e05e | 6409 | root.reset(); |
7c673cae FG |
6410 | |
6411 | // clean up any unclosed files | |
6412 | while (!fd_map.empty()) { | |
6413 | Fh *fh = fd_map.begin()->second; | |
6414 | fd_map.erase(fd_map.begin()); | |
6415 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *fh->inode << dendl; | |
6416 | _release_fh(fh); | |
6417 | } | |
6418 | ||
6419 | while (!ll_unclosed_fh_set.empty()) { | |
6420 | set<Fh*>::iterator it = ll_unclosed_fh_set.begin(); | |
6421 | Fh *fh = *it; | |
6422 | ll_unclosed_fh_set.erase(fh); | |
6423 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *(fh->inode) << dendl; | |
6424 | _release_fh(fh); | |
6425 | } | |
6426 | ||
6427 | while (!opened_dirs.empty()) { | |
6428 | dir_result_t *dirp = *opened_dirs.begin(); | |
6429 | ldout(cct, 0) << " destroyed lost open dir " << dirp << " on " << *dirp->inode << dendl; | |
6430 | _closedir(dirp); | |
6431 | } | |
6432 | ||
6433 | _ll_drop_pins(); | |
6434 | ||
7c673cae FG |
6435 | if (cct->_conf->client_oc) { |
6436 | // flush/release all buffered data | |
11fdf7f2 TL |
6437 | std::list<InodeRef> anchor; |
6438 | for (auto& p : inode_map) { | |
6439 | Inode *in = p.second; | |
7c673cae | 6440 | if (!in) { |
11fdf7f2 TL |
6441 | ldout(cct, 0) << "null inode_map entry ino " << p.first << dendl; |
6442 | ceph_assert(in); | |
7c673cae | 6443 | } |
11fdf7f2 TL |
6444 | |
6445 | // prevent inode from getting freed | |
6446 | anchor.emplace_back(in); | |
6447 | ||
f67539c2 | 6448 | if (abort || blocklisted) { |
11fdf7f2 TL |
6449 | objectcacher->purge_set(&in->oset); |
6450 | } else if (!in->caps.empty()) { | |
7c673cae FG |
6451 | _release(in); |
6452 | _flush(in, new C_Client_FlushComplete(this, in)); | |
6453 | } | |
6454 | } | |
6455 | } | |
6456 | ||
f67539c2 | 6457 | if (abort || blocklisted) { |
11fdf7f2 TL |
6458 | for (auto p = dirty_list.begin(); !p.end(); ) { |
6459 | Inode *in = *p; | |
6460 | ++p; | |
6461 | if (in->dirty_caps) { | |
6462 | ldout(cct, 0) << " drop dirty caps on " << *in << dendl; | |
6463 | in->mark_caps_clean(); | |
6464 | put_inode(in); | |
6465 | } | |
6466 | } | |
6467 | } else { | |
6468 | flush_caps_sync(); | |
6469 | wait_sync_caps(last_flush_tid); | |
6470 | } | |
7c673cae FG |
6471 | |
6472 | // empty lru cache | |
7c673cae FG |
6473 | trim_cache(); |
6474 | ||
f67539c2 TL |
6475 | delay_put_inodes(); |
6476 | ||
7c673cae FG |
6477 | while (lru.lru_get_size() > 0 || |
6478 | !inode_map.empty()) { | |
6479 | ldout(cct, 2) << "cache still has " << lru.lru_get_size() | |
6480 | << "+" << inode_map.size() << " items" | |
6481 | << ", waiting (for caps to release?)" | |
6482 | << dendl; | |
f67539c2 | 6483 | |
9f95a23c TL |
6484 | if (auto r = mount_cond.wait_for(lock, ceph::make_timespan(5)); |
6485 | r == std::cv_status::timeout) { | |
7c673cae FG |
6486 | dump_cache(NULL); |
6487 | } | |
6488 | } | |
11fdf7f2 TL |
6489 | ceph_assert(lru.lru_get_size() == 0); |
6490 | ceph_assert(inode_map.empty()); | |
7c673cae FG |
6491 | |
6492 | // stop tracing | |
6493 | if (!cct->_conf->client_trace.empty()) { | |
6494 | ldout(cct, 1) << "closing trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6495 | traceout.close(); | |
6496 | } | |
6497 | ||
f67539c2 TL |
6498 | // stop the tick thread |
6499 | tick_thread_stopped = true; | |
6500 | upkeep_cond.notify_one(); | |
6501 | ||
7c673cae FG |
6502 | _close_sessions(); |
6503 | ||
f67539c2 | 6504 | mref_writer.update_state(CLIENT_UNMOUNTED); |
7c673cae FG |
6505 | |
6506 | ldout(cct, 2) << "unmounted." << dendl; | |
6507 | } | |
6508 | ||
b32b8144 FG |
6509 | void Client::unmount() |
6510 | { | |
11fdf7f2 TL |
6511 | _unmount(false); |
6512 | } | |
6513 | ||
6514 | void Client::abort_conn() | |
6515 | { | |
11fdf7f2 | 6516 | _unmount(true); |
b32b8144 FG |
6517 | } |
6518 | ||
7c673cae FG |
6519 | void Client::flush_cap_releases() |
6520 | { | |
f67539c2 TL |
6521 | uint64_t nr_caps = 0; |
6522 | ||
7c673cae | 6523 | // send any cap releases |
11fdf7f2 TL |
6524 | for (auto &p : mds_sessions) { |
6525 | auto &session = p.second; | |
6526 | if (session.release && mdsmap->is_clientreplay_or_active_or_stopping( | |
6527 | p.first)) { | |
f67539c2 | 6528 | nr_caps += session.release->caps.size(); |
7c673cae FG |
6529 | if (cct->_conf->client_inject_release_failure) { |
6530 | ldout(cct, 20) << __func__ << " injecting failure to send cap release message" << dendl; | |
7c673cae | 6531 | } else { |
11fdf7f2 | 6532 | session.con->send_message2(std::move(session.release)); |
7c673cae | 6533 | } |
11fdf7f2 | 6534 | session.release.reset(); |
7c673cae FG |
6535 | } |
6536 | } | |
f67539c2 TL |
6537 | |
6538 | if (nr_caps > 0) { | |
6539 | dec_pinned_icaps(nr_caps); | |
6540 | } | |
7c673cae FG |
6541 | } |
6542 | ||
f67539c2 | 6543 | void Client::renew_and_flush_cap_releases() |
7c673cae | 6544 | { |
f67539c2 TL |
6545 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
6546 | ||
6547 | if (!mount_aborted && mdsmap->get_epoch()) { | |
6548 | // renew caps? | |
6549 | utime_t el = ceph_clock_now() - last_cap_renew; | |
6550 | if (unlikely(el > mdsmap->get_session_timeout() / 3.0)) | |
6551 | renew_caps(); | |
6552 | ||
6553 | flush_cap_releases(); | |
7c673cae | 6554 | } |
f67539c2 TL |
6555 | } |
6556 | ||
6557 | void Client::tick() | |
6558 | { | |
6559 | ldout(cct, 20) << "tick" << dendl; | |
7c673cae | 6560 | |
7c673cae FG |
6561 | utime_t now = ceph_clock_now(); |
6562 | ||
f67539c2 TL |
6563 | /* |
6564 | * If the mount() is not finished | |
6565 | */ | |
6566 | if (is_mounting() && !mds_requests.empty()) { | |
7c673cae | 6567 | MetaRequest *req = mds_requests.begin()->second; |
f67539c2 | 6568 | |
7c673cae | 6569 | if (req->op_stamp + cct->_conf->client_mount_timeout < now) { |
f67539c2 | 6570 | req->abort(-CEPHFS_ETIMEDOUT); |
7c673cae | 6571 | if (req->caller_cond) { |
f67539c2 TL |
6572 | req->kick = true; |
6573 | req->caller_cond->notify_all(); | |
7c673cae FG |
6574 | } |
6575 | signal_cond_list(waiting_for_mdsmap); | |
11fdf7f2 | 6576 | for (auto &p : mds_sessions) { |
f67539c2 | 6577 | signal_context_list(p.second.waiting_for_open); |
11fdf7f2 | 6578 | } |
7c673cae FG |
6579 | } |
6580 | } | |
6581 | ||
f67539c2 | 6582 | renew_and_flush_cap_releases(); |
7c673cae FG |
6583 | |
6584 | // delayed caps | |
28e407b8 | 6585 | xlist<Inode*>::iterator p = delayed_list.begin(); |
7c673cae FG |
6586 | while (!p.end()) { |
6587 | Inode *in = *p; | |
6588 | ++p; | |
f67539c2 | 6589 | if (!mount_aborted && in->hold_caps_until > now) |
7c673cae | 6590 | break; |
28e407b8 | 6591 | delayed_list.pop_front(); |
f67539c2 TL |
6592 | if (!mount_aborted) |
6593 | check_caps(in, CHECK_CAPS_NODELAY); | |
7c673cae FG |
6594 | } |
6595 | ||
f67539c2 TL |
6596 | if (!mount_aborted) |
6597 | collect_and_send_metrics(); | |
6598 | ||
6599 | delay_put_inodes(is_unmounting()); | |
7c673cae | 6600 | trim_cache(true); |
f6b5b4d7 | 6601 | |
f67539c2 | 6602 | if (blocklisted && (is_mounted() || is_unmounting()) && |
f6b5b4d7 TL |
6603 | last_auto_reconnect + 30 * 60 < now && |
6604 | cct->_conf.get_val<bool>("client_reconnect_stale")) { | |
6605 | messenger->client_reset(); | |
6606 | fd_gen++; // invalidate open files | |
f67539c2 | 6607 | blocklisted = false; |
f6b5b4d7 TL |
6608 | _kick_stale_sessions(); |
6609 | last_auto_reconnect = now; | |
6610 | } | |
7c673cae FG |
6611 | } |
6612 | ||
f67539c2 TL |
6613 | void Client::start_tick_thread() |
6614 | { | |
6615 | upkeeper = std::thread([this]() { | |
6616 | using time = ceph::coarse_mono_time; | |
6617 | using sec = std::chrono::seconds; | |
6618 | ||
6619 | auto last_tick = time::min(); | |
6620 | ||
6621 | std::unique_lock cl(client_lock); | |
6622 | while (!tick_thread_stopped) { | |
6623 | auto now = clock::now(); | |
6624 | auto since = now - last_tick; | |
6625 | ||
6626 | auto t_interval = clock::duration(cct->_conf.get_val<sec>("client_tick_interval")); | |
6627 | auto d_interval = clock::duration(cct->_conf.get_val<sec>("client_debug_inject_tick_delay")); | |
6628 | ||
6629 | auto interval = std::max(t_interval, d_interval); | |
6630 | if (likely(since >= interval*.90)) { | |
6631 | tick(); | |
6632 | last_tick = clock::now(); | |
6633 | } else { | |
6634 | interval -= since; | |
6635 | } | |
6636 | ||
6637 | ldout(cct, 20) << "upkeep thread waiting interval " << interval << dendl; | |
6638 | if (!tick_thread_stopped) | |
6639 | upkeep_cond.wait_for(cl, interval); | |
6640 | } | |
6641 | }); | |
6642 | } | |
6643 | ||
6644 | void Client::collect_and_send_metrics() { | |
6645 | ldout(cct, 20) << __func__ << dendl; | |
6646 | ||
6647 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6648 | ||
6649 | // right now, we only track and send global metrics. its sufficient | |
6650 | // to send these metrics to MDS rank0. | |
6651 | collect_and_send_global_metrics(); | |
6652 | } | |
6653 | ||
6654 | void Client::collect_and_send_global_metrics() { | |
6655 | ldout(cct, 20) << __func__ << dendl; | |
6656 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6657 | ||
6658 | if (!have_open_session((mds_rank_t)0)) { | |
6659 | ldout(cct, 5) << __func__ << ": no session with rank=0 -- not sending metric" | |
6660 | << dendl; | |
6661 | return; | |
6662 | } | |
6663 | auto session = _get_or_open_mds_session((mds_rank_t)0); | |
6664 | if (!session->mds_features.test(CEPHFS_FEATURE_METRIC_COLLECT)) { | |
6665 | ldout(cct, 5) << __func__ << ": rank=0 does not support metrics" << dendl; | |
6666 | return; | |
6667 | } | |
6668 | ||
6669 | ClientMetricMessage metric; | |
6670 | std::vector<ClientMetricMessage> message; | |
6671 | ||
6672 | // read latency | |
6673 | metric = ClientMetricMessage(ReadLatencyPayload(logger->tget(l_c_read))); | |
6674 | message.push_back(metric); | |
6675 | ||
6676 | // write latency | |
6677 | metric = ClientMetricMessage(WriteLatencyPayload(logger->tget(l_c_wrlat))); | |
6678 | message.push_back(metric); | |
6679 | ||
6680 | // metadata latency | |
6681 | metric = ClientMetricMessage(MetadataLatencyPayload(logger->tget(l_c_lat))); | |
6682 | message.push_back(metric); | |
6683 | ||
6684 | // cap hit ratio -- nr_caps is unused right now | |
6685 | auto [cap_hits, cap_misses] = get_cap_hit_rates(); | |
6686 | metric = ClientMetricMessage(CapInfoPayload(cap_hits, cap_misses, 0)); | |
6687 | message.push_back(metric); | |
6688 | ||
6689 | // dentry lease hit ratio | |
6690 | auto [dlease_hits, dlease_misses, nr] = get_dlease_hit_rates(); | |
6691 | metric = ClientMetricMessage(DentryLeasePayload(dlease_hits, dlease_misses, nr)); | |
6692 | message.push_back(metric); | |
6693 | ||
6694 | // opened files | |
6695 | { | |
6696 | auto [opened_files, total_inodes] = get_opened_files_rates(); | |
6697 | metric = ClientMetricMessage(OpenedFilesPayload(opened_files, total_inodes)); | |
6698 | } | |
6699 | message.push_back(metric); | |
6700 | ||
6701 | // pinned i_caps | |
6702 | { | |
6703 | auto [pinned_icaps, total_inodes] = get_pinned_icaps_rates(); | |
6704 | metric = ClientMetricMessage(PinnedIcapsPayload(pinned_icaps, total_inodes)); | |
6705 | } | |
6706 | message.push_back(metric); | |
6707 | ||
6708 | // opened inodes | |
6709 | { | |
6710 | auto [opened_inodes, total_inodes] = get_opened_inodes_rates(); | |
6711 | metric = ClientMetricMessage(OpenedInodesPayload(opened_inodes, total_inodes)); | |
6712 | } | |
6713 | message.push_back(metric); | |
6714 | ||
6715 | session->con->send_message2(make_message<MClientMetrics>(std::move(message))); | |
6716 | } | |
6717 | ||
7c673cae FG |
6718 | void Client::renew_caps() |
6719 | { | |
6720 | ldout(cct, 10) << "renew_caps()" << dendl; | |
6721 | last_cap_renew = ceph_clock_now(); | |
6722 | ||
11fdf7f2 TL |
6723 | for (auto &p : mds_sessions) { |
6724 | ldout(cct, 15) << "renew_caps requesting from mds." << p.first << dendl; | |
6725 | if (mdsmap->get_state(p.first) >= MDSMap::STATE_REJOIN) | |
6726 | renew_caps(&p.second); | |
7c673cae FG |
6727 | } |
6728 | } | |
6729 | ||
6730 | void Client::renew_caps(MetaSession *session) | |
6731 | { | |
6732 | ldout(cct, 10) << "renew_caps mds." << session->mds_num << dendl; | |
6733 | session->last_cap_renew_request = ceph_clock_now(); | |
6734 | uint64_t seq = ++session->cap_renew_seq; | |
9f95a23c | 6735 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_RENEWCAPS, seq)); |
7c673cae FG |
6736 | } |
6737 | ||
6738 | ||
6739 | // =============================================================== | |
6740 | // high level (POSIXy) interface | |
6741 | ||
6742 | int Client::_do_lookup(Inode *dir, const string& name, int mask, | |
6743 | InodeRef *target, const UserPerm& perms) | |
6744 | { | |
6745 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
6746 | MetaRequest *req = new MetaRequest(op); | |
6747 | filepath path; | |
6748 | dir->make_nosnap_relative_path(path); | |
6749 | path.push_dentry(name); | |
6750 | req->set_filepath(path); | |
6751 | req->set_inode(dir); | |
6752 | if (cct->_conf->client_debug_getattr_caps && op == CEPH_MDS_OP_LOOKUP) | |
6753 | mask |= DEBUG_GETATTR_CAPS; | |
6754 | req->head.args.getattr.mask = mask; | |
6755 | ||
11fdf7f2 | 6756 | ldout(cct, 10) << __func__ << " on " << path << dendl; |
7c673cae FG |
6757 | |
6758 | int r = make_request(req, perms, target); | |
11fdf7f2 | 6759 | ldout(cct, 10) << __func__ << " res is " << r << dendl; |
7c673cae FG |
6760 | return r; |
6761 | } | |
6762 | ||
f67539c2 TL |
6763 | bool Client::_dentry_valid(const Dentry *dn) |
6764 | { | |
6765 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); | |
6766 | ||
6767 | // is dn lease valid? | |
6768 | utime_t now = ceph_clock_now(); | |
6769 | if (dn->lease_mds >= 0 && dn->lease_ttl > now && | |
6770 | mds_sessions.count(dn->lease_mds)) { | |
6771 | MetaSession &s = mds_sessions.at(dn->lease_mds); | |
6772 | if (s.cap_ttl > now && s.cap_gen == dn->lease_gen) { | |
6773 | dlease_hit(); | |
6774 | return true; | |
6775 | } | |
6776 | ||
6777 | ldout(cct, 20) << " bad lease, cap_ttl " << s.cap_ttl << ", cap_gen " << s.cap_gen | |
6778 | << " vs lease_gen " << dn->lease_gen << dendl; | |
6779 | } | |
6780 | ||
6781 | dlease_miss(); | |
6782 | return false; | |
6783 | } | |
6784 | ||
7c673cae | 6785 | int Client::_lookup(Inode *dir, const string& dname, int mask, InodeRef *target, |
f67539c2 | 6786 | const UserPerm& perms, std::string* alternate_name) |
7c673cae FG |
6787 | { |
6788 | int r = 0; | |
6789 | Dentry *dn = NULL; | |
f67539c2 | 6790 | bool did_lookup_request = false; |
f91f0fd5 TL |
6791 | // can only request shared caps |
6792 | mask &= CEPH_CAP_ANY_SHARED | CEPH_STAT_RSTAT; | |
7c673cae | 6793 | |
7c673cae | 6794 | if (dname == "..") { |
11fdf7f2 TL |
6795 | if (dir->dentries.empty()) { |
6796 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); | |
6797 | filepath path(dir->ino); | |
6798 | req->set_filepath(path); | |
6799 | ||
6800 | InodeRef tmptarget; | |
6801 | int r = make_request(req, perms, &tmptarget, NULL, rand() % mdsmap->get_num_in_mds()); | |
6802 | ||
6803 | if (r == 0) { | |
f91f0fd5 | 6804 | *target = std::move(tmptarget); |
11fdf7f2 TL |
6805 | ldout(cct, 8) << __func__ << " found target " << (*target)->ino << dendl; |
6806 | } else { | |
6807 | *target = dir; | |
6808 | } | |
6809 | } | |
7c673cae FG |
6810 | else |
6811 | *target = dir->get_first_parent()->dir->parent_inode; //dirs can't be hard-linked | |
6812 | goto done; | |
6813 | } | |
6814 | ||
6815 | if (dname == ".") { | |
6816 | *target = dir; | |
6817 | goto done; | |
6818 | } | |
6819 | ||
11fdf7f2 | 6820 | if (!dir->is_dir()) { |
f67539c2 | 6821 | r = -CEPHFS_ENOTDIR; |
11fdf7f2 TL |
6822 | goto done; |
6823 | } | |
6824 | ||
7c673cae | 6825 | if (dname.length() > NAME_MAX) { |
f67539c2 | 6826 | r = -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
6827 | goto done; |
6828 | } | |
6829 | ||
6830 | if (dname == cct->_conf->client_snapdir && | |
6831 | dir->snapid == CEPH_NOSNAP) { | |
6832 | *target = open_snapdir(dir); | |
6833 | goto done; | |
6834 | } | |
6835 | ||
f67539c2 | 6836 | relookup: |
7c673cae FG |
6837 | if (dir->dir && |
6838 | dir->dir->dentries.count(dname)) { | |
6839 | dn = dir->dir->dentries[dname]; | |
6840 | ||
f67539c2 TL |
6841 | ldout(cct, 20) << __func__ << " have " << *dn << " from mds." << dn->lease_mds |
6842 | << " ttl " << dn->lease_ttl << " seq " << dn->lease_seq << dendl; | |
7c673cae | 6843 | |
94b18763 | 6844 | if (!dn->inode || dn->inode->caps_issued_mask(mask, true)) { |
f67539c2 TL |
6845 | if (_dentry_valid(dn)) { |
6846 | // touch this mds's dir cap too, even though we don't _explicitly_ use it here, to | |
6847 | // make trim_caps() behave. | |
6848 | dir->try_touch_cap(dn->lease_mds); | |
6849 | goto hit_dn; | |
7c673cae | 6850 | } |
92f5a8d4 | 6851 | // dir shared caps? |
94b18763 | 6852 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 6853 | if (dn->cap_shared_gen == dir->shared_gen && |
94b18763 | 6854 | (!dn->inode || dn->inode->caps_issued_mask(mask, true))) |
7c673cae FG |
6855 | goto hit_dn; |
6856 | if (!dn->inode && (dir->flags & I_COMPLETE)) { | |
11fdf7f2 | 6857 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " |
7c673cae | 6858 | << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 6859 | return -CEPHFS_ENOENT; |
7c673cae FG |
6860 | } |
6861 | } | |
6862 | } else { | |
6863 | ldout(cct, 20) << " no cap on " << dn->inode->vino() << dendl; | |
6864 | } | |
6865 | } else { | |
6866 | // can we conclude ENOENT locally? | |
94b18763 | 6867 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true) && |
7c673cae | 6868 | (dir->flags & I_COMPLETE)) { |
11fdf7f2 | 6869 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " << *dir << " dn '" << dname << "'" << dendl; |
f67539c2 | 6870 | return -CEPHFS_ENOENT; |
7c673cae FG |
6871 | } |
6872 | } | |
6873 | ||
f67539c2 TL |
6874 | if (did_lookup_request) { |
6875 | r = 0; | |
6876 | goto done; | |
6877 | } | |
7c673cae | 6878 | r = _do_lookup(dir, dname, mask, target, perms); |
f67539c2 TL |
6879 | did_lookup_request = true; |
6880 | if (r == 0) { | |
6881 | /* complete lookup to get dentry for alternate_name */ | |
6882 | goto relookup; | |
6883 | } else { | |
6884 | goto done; | |
6885 | } | |
6886 | ||
6887 | hit_dn: | |
6888 | if (dn->inode) { | |
7c673cae | 6889 | *target = dn->inode; |
f67539c2 TL |
6890 | if (alternate_name) |
6891 | *alternate_name = dn->alternate_name; | |
7c673cae | 6892 | } else { |
f67539c2 | 6893 | r = -CEPHFS_ENOENT; |
7c673cae FG |
6894 | } |
6895 | touch_dn(dn); | |
f67539c2 | 6896 | goto done; |
7c673cae FG |
6897 | |
6898 | done: | |
6899 | if (r < 0) | |
11fdf7f2 | 6900 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << r << dendl; |
7c673cae | 6901 | else |
11fdf7f2 | 6902 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << **target << dendl; |
7c673cae FG |
6903 | return r; |
6904 | } | |
6905 | ||
6906 | int Client::get_or_create(Inode *dir, const char* name, | |
6907 | Dentry **pdn, bool expect_null) | |
6908 | { | |
6909 | // lookup | |
11fdf7f2 | 6910 | ldout(cct, 20) << __func__ << " " << *dir << " name " << name << dendl; |
7c673cae FG |
6911 | dir->open_dir(); |
6912 | if (dir->dir->dentries.count(name)) { | |
6913 | Dentry *dn = dir->dir->dentries[name]; | |
f67539c2 TL |
6914 | if (_dentry_valid(dn)) { |
6915 | if (expect_null) | |
6916 | return -CEPHFS_EEXIST; | |
7c673cae FG |
6917 | } |
6918 | *pdn = dn; | |
6919 | } else { | |
6920 | // otherwise link up a new one | |
6921 | *pdn = link(dir->dir, name, NULL, NULL); | |
6922 | } | |
6923 | ||
6924 | // success | |
6925 | return 0; | |
6926 | } | |
6927 | ||
f67539c2 TL |
6928 | int Client::walk(std::string_view path, walk_dentry_result* wdr, const UserPerm& perms, bool followsym) |
6929 | { | |
6930 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
6931 | if (!mref_reader.is_state_satisfied()) | |
6932 | return -CEPHFS_ENOTCONN; | |
6933 | ||
6934 | ldout(cct, 10) << __func__ << ": " << path << dendl; | |
6935 | ||
6936 | std::scoped_lock lock(client_lock); | |
6937 | ||
6938 | return path_walk(path, wdr, perms, followsym); | |
6939 | } | |
6940 | ||
7c673cae | 6941 | int Client::path_walk(const filepath& origpath, InodeRef *end, |
b3b6e05e | 6942 | const UserPerm& perms, bool followsym, int mask, InodeRef dirinode) |
f67539c2 TL |
6943 | { |
6944 | walk_dentry_result wdr; | |
b3b6e05e | 6945 | int rc = path_walk(origpath, &wdr, perms, followsym, mask, dirinode); |
f67539c2 TL |
6946 | *end = std::move(wdr.in); |
6947 | return rc; | |
6948 | } | |
6949 | ||
b3b6e05e TL |
6950 | int Client::path_walk(const filepath& origpath, walk_dentry_result* result, const UserPerm& perms, |
6951 | bool followsym, int mask, InodeRef dirinode) | |
7c673cae FG |
6952 | { |
6953 | filepath path = origpath; | |
6954 | InodeRef cur; | |
f67539c2 | 6955 | std::string alternate_name; |
7c673cae FG |
6956 | if (origpath.absolute()) |
6957 | cur = root; | |
b3b6e05e | 6958 | else if (!dirinode) |
7c673cae | 6959 | cur = cwd; |
b3b6e05e TL |
6960 | else { |
6961 | cur = dirinode; | |
6962 | } | |
11fdf7f2 | 6963 | ceph_assert(cur); |
7c673cae | 6964 | |
b3b6e05e | 6965 | ldout(cct, 20) << __func__ << " cur=" << *cur << dendl; |
11fdf7f2 | 6966 | ldout(cct, 10) << __func__ << " " << path << dendl; |
7c673cae FG |
6967 | |
6968 | int symlinks = 0; | |
6969 | ||
6970 | unsigned i=0; | |
6971 | while (i < path.depth() && cur) { | |
6972 | int caps = 0; | |
6973 | const string &dname = path[i]; | |
6974 | ldout(cct, 10) << " " << i << " " << *cur << " " << dname << dendl; | |
6975 | ldout(cct, 20) << " (path is " << path << ")" << dendl; | |
6976 | InodeRef next; | |
6977 | if (cct->_conf->client_permissions) { | |
6978 | int r = may_lookup(cur.get(), perms); | |
6979 | if (r < 0) | |
6980 | return r; | |
6981 | caps = CEPH_CAP_AUTH_SHARED; | |
6982 | } | |
6983 | ||
6984 | /* Get extra requested caps on the last component */ | |
6985 | if (i == (path.depth() - 1)) | |
6986 | caps |= mask; | |
f67539c2 | 6987 | int r = _lookup(cur.get(), dname, caps, &next, perms, &alternate_name); |
7c673cae FG |
6988 | if (r < 0) |
6989 | return r; | |
6990 | // only follow trailing symlink if followsym. always follow | |
6991 | // 'directory' symlinks. | |
6992 | if (next && next->is_symlink()) { | |
6993 | symlinks++; | |
6994 | ldout(cct, 20) << " symlink count " << symlinks << ", value is '" << next->symlink << "'" << dendl; | |
6995 | if (symlinks > MAXSYMLINKS) { | |
f67539c2 | 6996 | return -CEPHFS_ELOOP; |
7c673cae FG |
6997 | } |
6998 | ||
6999 | if (i < path.depth() - 1) { | |
7000 | // dir symlink | |
7001 | // replace consumed components of path with symlink dir target | |
7002 | filepath resolved(next->symlink.c_str()); | |
7003 | resolved.append(path.postfixpath(i + 1)); | |
7004 | path = resolved; | |
7005 | i = 0; | |
7006 | if (next->symlink[0] == '/') { | |
7007 | cur = root; | |
7008 | } | |
7009 | continue; | |
7010 | } else if (followsym) { | |
7011 | if (next->symlink[0] == '/') { | |
7012 | path = next->symlink.c_str(); | |
7013 | i = 0; | |
7014 | // reset position | |
7015 | cur = root; | |
7016 | } else { | |
7017 | filepath more(next->symlink.c_str()); | |
7018 | // we need to remove the symlink component from off of the path | |
7019 | // before adding the target that the symlink points to. remain | |
7020 | // at the same position in the path. | |
7021 | path.pop_dentry(); | |
7022 | path.append(more); | |
7023 | } | |
7024 | continue; | |
7025 | } | |
7026 | } | |
7027 | cur.swap(next); | |
7028 | i++; | |
7029 | } | |
7030 | if (!cur) | |
f67539c2 TL |
7031 | return -CEPHFS_ENOENT; |
7032 | if (result) { | |
7033 | result->in = std::move(cur); | |
7034 | result->alternate_name = std::move(alternate_name); | |
7035 | } | |
7c673cae FG |
7036 | return 0; |
7037 | } | |
7038 | ||
7039 | ||
7040 | // namespace ops | |
7041 | ||
f67539c2 | 7042 | int Client::link(const char *relexisting, const char *relpath, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7043 | { |
f67539c2 TL |
7044 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7045 | if (!mref_reader.is_state_satisfied()) | |
7046 | return -CEPHFS_ENOTCONN; | |
7047 | ||
7c673cae FG |
7048 | tout(cct) << "link" << std::endl; |
7049 | tout(cct) << relexisting << std::endl; | |
7050 | tout(cct) << relpath << std::endl; | |
7051 | ||
7052 | filepath existing(relexisting); | |
7053 | ||
7054 | InodeRef in, dir; | |
f67539c2 TL |
7055 | |
7056 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7057 | int r = path_walk(existing, &in, perm, true); |
7058 | if (r < 0) | |
7059 | return r; | |
7060 | if (std::string(relpath) == "/") { | |
f67539c2 | 7061 | r = -CEPHFS_EEXIST; |
7c673cae FG |
7062 | return r; |
7063 | } | |
7064 | filepath path(relpath); | |
7065 | string name = path.last_dentry(); | |
7066 | path.pop_dentry(); | |
7067 | ||
7068 | r = path_walk(path, &dir, perm, true); | |
7069 | if (r < 0) | |
7070 | return r; | |
7071 | if (cct->_conf->client_permissions) { | |
7072 | if (S_ISDIR(in->mode)) { | |
f67539c2 | 7073 | r = -CEPHFS_EPERM; |
7c673cae FG |
7074 | return r; |
7075 | } | |
7076 | r = may_hardlink(in.get(), perm); | |
7077 | if (r < 0) | |
7078 | return r; | |
7079 | r = may_create(dir.get(), perm); | |
7080 | if (r < 0) | |
7081 | return r; | |
7082 | } | |
f67539c2 | 7083 | r = _link(in.get(), dir.get(), name.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7084 | return r; |
7085 | } | |
7086 | ||
7087 | int Client::unlink(const char *relpath, const UserPerm& perm) | |
b3b6e05e TL |
7088 | { |
7089 | return unlinkat(CEPHFS_AT_FDCWD, relpath, 0, perm); | |
7090 | } | |
7091 | ||
7092 | int Client::unlinkat(int dirfd, const char *relpath, int flags, const UserPerm& perm) | |
7c673cae | 7093 | { |
f67539c2 | 7094 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7095 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7096 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7097 | } |
f67539c2 | 7098 | |
11fdf7f2 | 7099 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7100 | tout(cct) << dirfd << std::endl; |
7c673cae | 7101 | tout(cct) << relpath << std::endl; |
b3b6e05e | 7102 | tout(cct) << flags << std::endl; |
7c673cae | 7103 | |
b3b6e05e TL |
7104 | if (std::string(relpath) == "/") { |
7105 | return flags & AT_REMOVEDIR ? -CEPHFS_EBUSY : -CEPHFS_EISDIR; | |
7106 | } | |
7c673cae FG |
7107 | |
7108 | filepath path(relpath); | |
7109 | string name = path.last_dentry(); | |
7110 | path.pop_dentry(); | |
7111 | InodeRef dir; | |
f67539c2 TL |
7112 | |
7113 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7114 | |
7115 | InodeRef dirinode; | |
7116 | int r = get_fd_inode(dirfd, &dirinode); | |
7117 | if (r < 0) { | |
7118 | return r; | |
7119 | } | |
7120 | ||
7121 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7122 | if (r < 0) { | |
7c673cae | 7123 | return r; |
b3b6e05e | 7124 | } |
7c673cae FG |
7125 | if (cct->_conf->client_permissions) { |
7126 | r = may_delete(dir.get(), name.c_str(), perm); | |
b3b6e05e | 7127 | if (r < 0) { |
7c673cae | 7128 | return r; |
b3b6e05e | 7129 | } |
7c673cae | 7130 | } |
b3b6e05e TL |
7131 | if (flags & AT_REMOVEDIR) { |
7132 | r = _rmdir(dir.get(), name.c_str(), perm); | |
7133 | } else { | |
7134 | r = _unlink(dir.get(), name.c_str(), perm); | |
7135 | } | |
7136 | return r; | |
7c673cae FG |
7137 | } |
7138 | ||
f67539c2 | 7139 | int Client::rename(const char *relfrom, const char *relto, const UserPerm& perm, std::string alternate_name) |
7c673cae | 7140 | { |
f67539c2 TL |
7141 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7142 | if (!mref_reader.is_state_satisfied()) | |
7143 | return -CEPHFS_ENOTCONN; | |
7144 | ||
11fdf7f2 | 7145 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7146 | tout(cct) << relfrom << std::endl; |
7147 | tout(cct) << relto << std::endl; | |
7148 | ||
7149 | if (std::string(relfrom) == "/" || std::string(relto) == "/") | |
f67539c2 | 7150 | return -CEPHFS_EBUSY; |
7c673cae FG |
7151 | |
7152 | filepath from(relfrom); | |
7153 | filepath to(relto); | |
7154 | string fromname = from.last_dentry(); | |
7155 | from.pop_dentry(); | |
7156 | string toname = to.last_dentry(); | |
7157 | to.pop_dentry(); | |
7158 | ||
7159 | InodeRef fromdir, todir; | |
f67539c2 TL |
7160 | |
7161 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7162 | int r = path_walk(from, &fromdir, perm); |
7163 | if (r < 0) | |
7164 | goto out; | |
7165 | r = path_walk(to, &todir, perm); | |
7166 | if (r < 0) | |
7167 | goto out; | |
7168 | ||
7169 | if (cct->_conf->client_permissions) { | |
7170 | int r = may_delete(fromdir.get(), fromname.c_str(), perm); | |
7171 | if (r < 0) | |
7172 | return r; | |
7173 | r = may_delete(todir.get(), toname.c_str(), perm); | |
f67539c2 | 7174 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
7175 | return r; |
7176 | } | |
f67539c2 | 7177 | r = _rename(fromdir.get(), fromname.c_str(), todir.get(), toname.c_str(), perm, std::move(alternate_name)); |
7c673cae FG |
7178 | out: |
7179 | return r; | |
7180 | } | |
7181 | ||
7182 | // dirs | |
7183 | ||
f67539c2 | 7184 | int Client::mkdir(const char *relpath, mode_t mode, const UserPerm& perm, std::string alternate_name) |
b3b6e05e TL |
7185 | { |
7186 | return mkdirat(CEPHFS_AT_FDCWD, relpath, mode, perm, alternate_name); | |
7187 | } | |
7188 | ||
7189 | int Client::mkdirat(int dirfd, const char *relpath, mode_t mode, const UserPerm& perm, | |
7190 | std::string alternate_name) | |
7c673cae | 7191 | { |
f67539c2 TL |
7192 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7193 | if (!mref_reader.is_state_satisfied()) | |
7194 | return -CEPHFS_ENOTCONN; | |
7195 | ||
11fdf7f2 | 7196 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7197 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7198 | tout(cct) << relpath << std::endl; |
7199 | tout(cct) << mode << std::endl; | |
11fdf7f2 | 7200 | ldout(cct, 10) << __func__ << ": " << relpath << dendl; |
7c673cae | 7201 | |
b3b6e05e | 7202 | if (std::string(relpath) == "/") { |
f67539c2 | 7203 | return -CEPHFS_EEXIST; |
b3b6e05e | 7204 | } |
7c673cae FG |
7205 | |
7206 | filepath path(relpath); | |
7207 | string name = path.last_dentry(); | |
7208 | path.pop_dentry(); | |
7209 | InodeRef dir; | |
f67539c2 TL |
7210 | |
7211 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7212 | |
7213 | InodeRef dirinode; | |
7214 | int r = get_fd_inode(dirfd, &dirinode); | |
7215 | if (r < 0) { | |
7c673cae | 7216 | return r; |
b3b6e05e TL |
7217 | } |
7218 | ||
7219 | r = path_walk(path, &dir, perm, true, 0, dirinode); | |
7220 | if (r < 0) { | |
7221 | return r; | |
7222 | } | |
7c673cae FG |
7223 | if (cct->_conf->client_permissions) { |
7224 | r = may_create(dir.get(), perm); | |
b3b6e05e | 7225 | if (r < 0) { |
7c673cae | 7226 | return r; |
b3b6e05e | 7227 | } |
7c673cae | 7228 | } |
f67539c2 | 7229 | return _mkdir(dir.get(), name.c_str(), mode, perm, 0, {}, std::move(alternate_name)); |
7c673cae FG |
7230 | } |
7231 | ||
7232 | int Client::mkdirs(const char *relpath, mode_t mode, const UserPerm& perms) | |
7233 | { | |
f67539c2 TL |
7234 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7235 | if (!mref_reader.is_state_satisfied()) | |
7236 | return -CEPHFS_ENOTCONN; | |
7237 | ||
7c673cae | 7238 | ldout(cct, 10) << "Client::mkdirs " << relpath << dendl; |
11fdf7f2 | 7239 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7240 | tout(cct) << relpath << std::endl; |
7241 | tout(cct) << mode << std::endl; | |
7242 | ||
7243 | //get through existing parts of path | |
7244 | filepath path(relpath); | |
7245 | unsigned int i; | |
7246 | int r = 0, caps = 0; | |
7247 | InodeRef cur, next; | |
f67539c2 TL |
7248 | |
7249 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7250 | cur = cwd; |
7251 | for (i=0; i<path.depth(); ++i) { | |
7252 | if (cct->_conf->client_permissions) { | |
7253 | r = may_lookup(cur.get(), perms); | |
7254 | if (r < 0) | |
7255 | break; | |
7256 | caps = CEPH_CAP_AUTH_SHARED; | |
7257 | } | |
7258 | r = _lookup(cur.get(), path[i].c_str(), caps, &next, perms); | |
7259 | if (r < 0) | |
7260 | break; | |
7261 | cur.swap(next); | |
7262 | } | |
f67539c2 | 7263 | if (r!=-CEPHFS_ENOENT) return r; |
11fdf7f2 | 7264 | ldout(cct, 20) << __func__ << " got through " << i << " directories on path " << relpath << dendl; |
7c673cae FG |
7265 | //make new directory at each level |
7266 | for (; i<path.depth(); ++i) { | |
7267 | if (cct->_conf->client_permissions) { | |
7268 | r = may_create(cur.get(), perms); | |
7269 | if (r < 0) | |
7270 | return r; | |
7271 | } | |
7272 | //make new dir | |
7273 | r = _mkdir(cur.get(), path[i].c_str(), mode, perms, &next); | |
c07f9fc5 | 7274 | |
7c673cae | 7275 | //check proper creation/existence |
f67539c2 | 7276 | if(-CEPHFS_EEXIST == r && i < path.depth() - 1) { |
c07f9fc5 FG |
7277 | r = _lookup(cur.get(), path[i].c_str(), CEPH_CAP_AUTH_SHARED, &next, perms); |
7278 | } | |
7279 | if (r < 0) | |
7280 | return r; | |
7c673cae FG |
7281 | //move to new dir and continue |
7282 | cur.swap(next); | |
11fdf7f2 | 7283 | ldout(cct, 20) << __func__ << ": successfully created directory " |
7c673cae FG |
7284 | << filepath(cur->ino).get_path() << dendl; |
7285 | } | |
7286 | return 0; | |
7287 | } | |
7288 | ||
7289 | int Client::rmdir(const char *relpath, const UserPerm& perms) | |
7290 | { | |
b3b6e05e | 7291 | return unlinkat(CEPHFS_AT_FDCWD, relpath, AT_REMOVEDIR, perms); |
7c673cae FG |
7292 | } |
7293 | ||
7294 | int Client::mknod(const char *relpath, mode_t mode, const UserPerm& perms, dev_t rdev) | |
f67539c2 TL |
7295 | { |
7296 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
7297 | if (!mref_reader.is_state_satisfied()) | |
7298 | return -CEPHFS_ENOTCONN; | |
7299 | ||
11fdf7f2 | 7300 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7301 | tout(cct) << relpath << std::endl; |
7302 | tout(cct) << mode << std::endl; | |
7303 | tout(cct) << rdev << std::endl; | |
7304 | ||
7305 | if (std::string(relpath) == "/") | |
f67539c2 | 7306 | return -CEPHFS_EEXIST; |
7c673cae FG |
7307 | |
7308 | filepath path(relpath); | |
7309 | string name = path.last_dentry(); | |
7310 | path.pop_dentry(); | |
7311 | InodeRef dir; | |
f67539c2 TL |
7312 | |
7313 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7314 | int r = path_walk(path, &dir, perms); |
7315 | if (r < 0) | |
7316 | return r; | |
7317 | if (cct->_conf->client_permissions) { | |
7318 | int r = may_create(dir.get(), perms); | |
7319 | if (r < 0) | |
7320 | return r; | |
7321 | } | |
7322 | return _mknod(dir.get(), name.c_str(), mode, rdev, perms); | |
7323 | } | |
7324 | ||
7325 | // symlinks | |
7326 | ||
f67539c2 | 7327 | int Client::symlink(const char *target, const char *relpath, const UserPerm& perms, std::string alternate_name) |
b3b6e05e TL |
7328 | { |
7329 | return symlinkat(target, CEPHFS_AT_FDCWD, relpath, perms, alternate_name); | |
7330 | } | |
7331 | ||
7332 | int Client::symlinkat(const char *target, int dirfd, const char *relpath, const UserPerm& perms, | |
7333 | std::string alternate_name) | |
7c673cae | 7334 | { |
f67539c2 | 7335 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7336 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7337 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7338 | } |
f67539c2 | 7339 | |
11fdf7f2 | 7340 | tout(cct) << __func__ << std::endl; |
7c673cae | 7341 | tout(cct) << target << std::endl; |
b3b6e05e | 7342 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7343 | tout(cct) << relpath << std::endl; |
7344 | ||
b3b6e05e | 7345 | if (std::string(relpath) == "/") { |
f67539c2 | 7346 | return -CEPHFS_EEXIST; |
b3b6e05e | 7347 | } |
7c673cae FG |
7348 | |
7349 | filepath path(relpath); | |
7350 | string name = path.last_dentry(); | |
7351 | path.pop_dentry(); | |
7352 | InodeRef dir; | |
f67539c2 TL |
7353 | |
7354 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
7355 | |
7356 | InodeRef dirinode; | |
7357 | int r = get_fd_inode(dirfd, &dirinode); | |
7358 | if (r < 0) { | |
7c673cae | 7359 | return r; |
b3b6e05e TL |
7360 | } |
7361 | r = path_walk(path, &dir, perms, true, 0, dirinode); | |
7362 | if (r < 0) { | |
7363 | return r; | |
7364 | } | |
7c673cae FG |
7365 | if (cct->_conf->client_permissions) { |
7366 | int r = may_create(dir.get(), perms); | |
b3b6e05e | 7367 | if (r < 0) { |
7c673cae | 7368 | return r; |
b3b6e05e | 7369 | } |
7c673cae | 7370 | } |
f67539c2 | 7371 | return _symlink(dir.get(), name.c_str(), target, perms, std::move(alternate_name)); |
7c673cae FG |
7372 | } |
7373 | ||
7374 | int Client::readlink(const char *relpath, char *buf, loff_t size, const UserPerm& perms) | |
7375 | { | |
b3b6e05e TL |
7376 | return readlinkat(CEPHFS_AT_FDCWD, relpath, buf, size, perms); |
7377 | } | |
7378 | ||
7379 | int Client::readlinkat(int dirfd, const char *relpath, char *buf, loff_t size, const UserPerm& perms) { | |
f67539c2 | 7380 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 7381 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 7382 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 7383 | } |
f67539c2 | 7384 | |
11fdf7f2 | 7385 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 7386 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
7387 | tout(cct) << relpath << std::endl; |
7388 | ||
b3b6e05e | 7389 | InodeRef dirinode; |
f67539c2 | 7390 | std::scoped_lock lock(client_lock); |
b3b6e05e TL |
7391 | int r = get_fd_inode(dirfd, &dirinode); |
7392 | if (r < 0) { | |
7c673cae | 7393 | return r; |
b3b6e05e TL |
7394 | } |
7395 | ||
7396 | InodeRef in; | |
7397 | filepath path(relpath); | |
7398 | r = path_walk(path, &in, perms, false, 0, dirinode); | |
7399 | if (r < 0) { | |
7400 | return r; | |
7401 | } | |
7c673cae FG |
7402 | |
7403 | return _readlink(in.get(), buf, size); | |
7404 | } | |
7405 | ||
7406 | int Client::_readlink(Inode *in, char *buf, size_t size) | |
7407 | { | |
7408 | if (!in->is_symlink()) | |
f67539c2 | 7409 | return -CEPHFS_EINVAL; |
7c673cae FG |
7410 | |
7411 | // copy into buf (at most size bytes) | |
7412 | int r = in->symlink.length(); | |
7413 | if (r > (int)size) | |
7414 | r = size; | |
7415 | memcpy(buf, in->symlink.c_str(), r); | |
7416 | return r; | |
7417 | } | |
7418 | ||
7419 | ||
7420 | // inode stuff | |
7421 | ||
7422 | int Client::_getattr(Inode *in, int mask, const UserPerm& perms, bool force) | |
7423 | { | |
94b18763 | 7424 | bool yes = in->caps_issued_mask(mask, true); |
7c673cae | 7425 | |
11fdf7f2 | 7426 | ldout(cct, 10) << __func__ << " mask " << ccap_string(mask) << " issued=" << yes << dendl; |
7c673cae FG |
7427 | if (yes && !force) |
7428 | return 0; | |
7429 | ||
7430 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
7431 | filepath path; | |
7432 | in->make_nosnap_relative_path(path); | |
7433 | req->set_filepath(path); | |
7434 | req->set_inode(in); | |
7435 | req->head.args.getattr.mask = mask; | |
7436 | ||
7437 | int res = make_request(req, perms); | |
11fdf7f2 | 7438 | ldout(cct, 10) << __func__ << " result=" << res << dendl; |
7c673cae FG |
7439 | return res; |
7440 | } | |
7441 | ||
7442 | int Client::_do_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
7443 | const UserPerm& perms, InodeRef *inp) | |
7444 | { | |
7445 | int issued = in->caps_issued(); | |
7446 | ||
11fdf7f2 | 7447 | ldout(cct, 10) << __func__ << " mask " << mask << " issued " << |
7c673cae FG |
7448 | ccap_string(issued) << dendl; |
7449 | ||
7450 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 7451 | return -CEPHFS_EROFS; |
7c673cae FG |
7452 | } |
7453 | if ((mask & CEPH_SETATTR_SIZE) && | |
f67539c2 TL |
7454 | (uint64_t)stx->stx_size > in->size && |
7455 | is_quota_bytes_exceeded(in, (uint64_t)stx->stx_size - in->size, | |
7c673cae | 7456 | perms)) { |
f67539c2 | 7457 | return -CEPHFS_EDQUOT; |
7c673cae FG |
7458 | } |
7459 | ||
7460 | // make the change locally? | |
7461 | if ((in->cap_dirtier_uid >= 0 && perms.uid() != in->cap_dirtier_uid) || | |
7462 | (in->cap_dirtier_gid >= 0 && perms.gid() != in->cap_dirtier_gid)) { | |
7463 | ldout(cct, 10) << __func__ << " caller " << perms.uid() << ":" << perms.gid() | |
7464 | << " != cap dirtier " << in->cap_dirtier_uid << ":" | |
7465 | << in->cap_dirtier_gid << ", forcing sync setattr" | |
7466 | << dendl; | |
7467 | /* | |
7468 | * This works because we implicitly flush the caps as part of the | |
7469 | * request, so the cap update check will happen with the writeback | |
7470 | * cap context, and then the setattr check will happen with the | |
7471 | * caller's context. | |
7472 | * | |
7473 | * In reality this pattern is likely pretty rare (different users | |
7474 | * setattr'ing the same file). If that turns out not to be the | |
7475 | * case later, we can build a more complex pipelined cap writeback | |
7476 | * infrastructure... | |
7477 | */ | |
7478 | if (!mask) | |
7479 | mask |= CEPH_SETATTR_CTIME; | |
7480 | goto force_request; | |
7481 | } | |
7482 | ||
7483 | if (!mask) { | |
7484 | // caller just needs us to bump the ctime | |
7485 | in->ctime = ceph_clock_now(); | |
7486 | in->cap_dirtier_uid = perms.uid(); | |
7487 | in->cap_dirtier_gid = perms.gid(); | |
7488 | if (issued & CEPH_CAP_AUTH_EXCL) | |
28e407b8 | 7489 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 7490 | else if (issued & CEPH_CAP_FILE_EXCL) |
28e407b8 | 7491 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
7c673cae | 7492 | else if (issued & CEPH_CAP_XATTR_EXCL) |
28e407b8 | 7493 | in->mark_caps_dirty(CEPH_CAP_XATTR_EXCL); |
7c673cae FG |
7494 | else |
7495 | mask |= CEPH_SETATTR_CTIME; | |
7496 | } | |
7497 | ||
7498 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
7499 | bool kill_sguid = mask & (CEPH_SETATTR_SIZE|CEPH_SETATTR_KILL_SGUID); | |
7500 | ||
7501 | mask &= ~CEPH_SETATTR_KILL_SGUID; | |
7502 | ||
7503 | if (mask & CEPH_SETATTR_UID) { | |
7504 | in->ctime = ceph_clock_now(); | |
7505 | in->cap_dirtier_uid = perms.uid(); | |
7506 | in->cap_dirtier_gid = perms.gid(); | |
7507 | in->uid = stx->stx_uid; | |
28e407b8 | 7508 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7509 | mask &= ~CEPH_SETATTR_UID; |
7510 | kill_sguid = true; | |
7511 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
7512 | } | |
7513 | if (mask & CEPH_SETATTR_GID) { | |
7514 | in->ctime = ceph_clock_now(); | |
7515 | in->cap_dirtier_uid = perms.uid(); | |
7516 | in->cap_dirtier_gid = perms.gid(); | |
7517 | in->gid = stx->stx_gid; | |
28e407b8 | 7518 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7519 | mask &= ~CEPH_SETATTR_GID; |
7520 | kill_sguid = true; | |
7521 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
7522 | } | |
7523 | ||
7524 | if (mask & CEPH_SETATTR_MODE) { | |
7525 | in->ctime = ceph_clock_now(); | |
7526 | in->cap_dirtier_uid = perms.uid(); | |
7527 | in->cap_dirtier_gid = perms.gid(); | |
7528 | in->mode = (in->mode & ~07777) | (stx->stx_mode & 07777); | |
28e407b8 | 7529 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7530 | mask &= ~CEPH_SETATTR_MODE; |
7531 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
181888fb | 7532 | } else if (kill_sguid && S_ISREG(in->mode) && (in->mode & (S_IXUSR|S_IXGRP|S_IXOTH))) { |
7c673cae | 7533 | /* Must squash the any setuid/setgid bits with an ownership change */ |
181888fb | 7534 | in->mode &= ~(S_ISUID|S_ISGID); |
28e407b8 | 7535 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7536 | } |
7537 | ||
7538 | if (mask & CEPH_SETATTR_BTIME) { | |
7539 | in->ctime = ceph_clock_now(); | |
7540 | in->cap_dirtier_uid = perms.uid(); | |
7541 | in->cap_dirtier_gid = perms.gid(); | |
7542 | in->btime = utime_t(stx->stx_btime); | |
28e407b8 | 7543 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
7544 | mask &= ~CEPH_SETATTR_BTIME; |
7545 | ldout(cct,10) << "changing btime to " << in->btime << dendl; | |
7546 | } | |
7547 | } else if (mask & CEPH_SETATTR_SIZE) { | |
7548 | /* If we don't have Ax, then we must ask the server to clear them on truncate */ | |
7549 | mask |= CEPH_SETATTR_KILL_SGUID; | |
7550 | } | |
7551 | ||
7552 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
7553 | if (mask & (CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME)) { | |
7554 | if (mask & CEPH_SETATTR_MTIME) | |
7555 | in->mtime = utime_t(stx->stx_mtime); | |
7556 | if (mask & CEPH_SETATTR_ATIME) | |
7557 | in->atime = utime_t(stx->stx_atime); | |
7558 | in->ctime = ceph_clock_now(); | |
7559 | in->cap_dirtier_uid = perms.uid(); | |
7560 | in->cap_dirtier_gid = perms.gid(); | |
7561 | in->time_warp_seq++; | |
28e407b8 | 7562 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
7c673cae FG |
7563 | mask &= ~(CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME); |
7564 | } | |
7565 | } | |
7566 | if (!mask) { | |
7567 | in->change_attr++; | |
7568 | return 0; | |
7569 | } | |
7570 | ||
7571 | force_request: | |
7572 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETATTR); | |
7573 | ||
7574 | filepath path; | |
7575 | ||
7576 | in->make_nosnap_relative_path(path); | |
7577 | req->set_filepath(path); | |
7578 | req->set_inode(in); | |
7579 | ||
7580 | if (mask & CEPH_SETATTR_KILL_SGUID) { | |
7581 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7582 | } | |
7583 | if (mask & CEPH_SETATTR_MODE) { | |
7584 | req->head.args.setattr.mode = stx->stx_mode; | |
7585 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7586 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
7587 | } | |
7588 | if (mask & CEPH_SETATTR_UID) { | |
7589 | req->head.args.setattr.uid = stx->stx_uid; | |
7590 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7591 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
7592 | } | |
7593 | if (mask & CEPH_SETATTR_GID) { | |
7594 | req->head.args.setattr.gid = stx->stx_gid; | |
7595 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7596 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
7597 | } | |
7598 | if (mask & CEPH_SETATTR_BTIME) { | |
7599 | req->head.args.setattr.btime = utime_t(stx->stx_btime); | |
7600 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7601 | } | |
7602 | if (mask & CEPH_SETATTR_MTIME) { | |
7603 | req->head.args.setattr.mtime = utime_t(stx->stx_mtime); | |
94b18763 | 7604 | req->inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | |
7c673cae FG |
7605 | CEPH_CAP_FILE_WR; |
7606 | } | |
7607 | if (mask & CEPH_SETATTR_ATIME) { | |
7608 | req->head.args.setattr.atime = utime_t(stx->stx_atime); | |
7609 | req->inode_drop |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
7610 | CEPH_CAP_FILE_WR; | |
7611 | } | |
7612 | if (mask & CEPH_SETATTR_SIZE) { | |
f67539c2 | 7613 | if ((uint64_t)stx->stx_size < mdsmap->get_max_filesize()) { |
7c673cae FG |
7614 | req->head.args.setattr.size = stx->stx_size; |
7615 | ldout(cct,10) << "changing size to " << stx->stx_size << dendl; | |
7616 | } else { //too big! | |
7617 | put_request(req); | |
7618 | ldout(cct,10) << "unable to set size to " << stx->stx_size << ". Too large!" << dendl; | |
f67539c2 | 7619 | return -CEPHFS_EFBIG; |
7c673cae | 7620 | } |
94b18763 | 7621 | req->inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | |
7c673cae FG |
7622 | CEPH_CAP_FILE_WR; |
7623 | } | |
7624 | req->head.args.setattr.mask = mask; | |
7625 | ||
7626 | req->regetattr_mask = mask; | |
7627 | ||
7628 | int res = make_request(req, perms, inp); | |
7629 | ldout(cct, 10) << "_setattr result=" << res << dendl; | |
7630 | return res; | |
7631 | } | |
7632 | ||
7633 | /* Note that we only care about attrs that setattr cares about */ | |
7634 | void Client::stat_to_statx(struct stat *st, struct ceph_statx *stx) | |
7635 | { | |
7636 | stx->stx_size = st->st_size; | |
7637 | stx->stx_mode = st->st_mode; | |
7638 | stx->stx_uid = st->st_uid; | |
7639 | stx->stx_gid = st->st_gid; | |
11fdf7f2 TL |
7640 | #ifdef __APPLE__ |
7641 | stx->stx_mtime = st->st_mtimespec; | |
7642 | stx->stx_atime = st->st_atimespec; | |
f67539c2 TL |
7643 | #elif __WIN32 |
7644 | stx->stx_mtime.tv_sec = st->st_mtime; | |
7645 | stx->stx_atime.tv_sec = st->st_atime; | |
11fdf7f2 | 7646 | #else |
7c673cae FG |
7647 | stx->stx_mtime = st->st_mtim; |
7648 | stx->stx_atime = st->st_atim; | |
11fdf7f2 | 7649 | #endif |
7c673cae FG |
7650 | } |
7651 | ||
7652 | int Client::__setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
7653 | const UserPerm& perms, InodeRef *inp) | |
7654 | { | |
7655 | int ret = _do_setattr(in, stx, mask, perms, inp); | |
7656 | if (ret < 0) | |
7657 | return ret; | |
7658 | if (mask & CEPH_SETATTR_MODE) | |
7659 | ret = _posix_acl_chmod(in, stx->stx_mode, perms); | |
7660 | return ret; | |
7661 | } | |
7662 | ||
7663 | int Client::_setattrx(InodeRef &in, struct ceph_statx *stx, int mask, | |
7664 | const UserPerm& perms) | |
7665 | { | |
7666 | mask &= (CEPH_SETATTR_MODE | CEPH_SETATTR_UID | | |
7667 | CEPH_SETATTR_GID | CEPH_SETATTR_MTIME | | |
7668 | CEPH_SETATTR_ATIME | CEPH_SETATTR_SIZE | | |
7669 | CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME); | |
7670 | if (cct->_conf->client_permissions) { | |
7671 | int r = may_setattr(in.get(), stx, mask, perms); | |
7672 | if (r < 0) | |
7673 | return r; | |
7674 | } | |
7675 | return __setattrx(in.get(), stx, mask, perms); | |
7676 | } | |
7677 | ||
7678 | int Client::_setattr(InodeRef &in, struct stat *attr, int mask, | |
7679 | const UserPerm& perms) | |
7680 | { | |
7681 | struct ceph_statx stx; | |
7682 | ||
7683 | stat_to_statx(attr, &stx); | |
7684 | mask &= ~CEPH_SETATTR_BTIME; | |
181888fb FG |
7685 | |
7686 | if ((mask & CEPH_SETATTR_UID) && attr->st_uid == static_cast<uid_t>(-1)) { | |
7687 | mask &= ~CEPH_SETATTR_UID; | |
7688 | } | |
7689 | if ((mask & CEPH_SETATTR_GID) && attr->st_gid == static_cast<uid_t>(-1)) { | |
7690 | mask &= ~CEPH_SETATTR_GID; | |
7691 | } | |
7692 | ||
7c673cae FG |
7693 | return _setattrx(in, &stx, mask, perms); |
7694 | } | |
7695 | ||
7696 | int Client::setattr(const char *relpath, struct stat *attr, int mask, | |
7697 | const UserPerm& perms) | |
7698 | { | |
f67539c2 TL |
7699 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7700 | if (!mref_reader.is_state_satisfied()) | |
7701 | return -CEPHFS_ENOTCONN; | |
7702 | ||
11fdf7f2 | 7703 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7704 | tout(cct) << relpath << std::endl; |
7705 | tout(cct) << mask << std::endl; | |
7706 | ||
7707 | filepath path(relpath); | |
7708 | InodeRef in; | |
f67539c2 TL |
7709 | |
7710 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7711 | int r = path_walk(path, &in, perms); |
7712 | if (r < 0) | |
7713 | return r; | |
7714 | return _setattr(in, attr, mask, perms); | |
7715 | } | |
7716 | ||
7717 | int Client::setattrx(const char *relpath, struct ceph_statx *stx, int mask, | |
7718 | const UserPerm& perms, int flags) | |
7719 | { | |
f67539c2 TL |
7720 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7721 | if (!mref_reader.is_state_satisfied()) | |
7722 | return -CEPHFS_ENOTCONN; | |
7723 | ||
11fdf7f2 | 7724 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7725 | tout(cct) << relpath << std::endl; |
7726 | tout(cct) << mask << std::endl; | |
7727 | ||
7728 | filepath path(relpath); | |
7729 | InodeRef in; | |
f67539c2 TL |
7730 | |
7731 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7732 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW)); |
7733 | if (r < 0) | |
7734 | return r; | |
7735 | return _setattrx(in, stx, mask, perms); | |
7736 | } | |
7737 | ||
7738 | int Client::fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms) | |
7739 | { | |
f67539c2 TL |
7740 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7741 | if (!mref_reader.is_state_satisfied()) | |
7742 | return -CEPHFS_ENOTCONN; | |
7743 | ||
11fdf7f2 | 7744 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7745 | tout(cct) << fd << std::endl; |
7746 | tout(cct) << mask << std::endl; | |
7747 | ||
f67539c2 | 7748 | std::scoped_lock lock(client_lock); |
7c673cae FG |
7749 | Fh *f = get_filehandle(fd); |
7750 | if (!f) | |
f67539c2 | 7751 | return -CEPHFS_EBADF; |
7c673cae FG |
7752 | #if defined(__linux__) && defined(O_PATH) |
7753 | if (f->flags & O_PATH) | |
f67539c2 | 7754 | return -CEPHFS_EBADF; |
7c673cae FG |
7755 | #endif |
7756 | return _setattr(f->inode, attr, mask, perms); | |
7757 | } | |
7758 | ||
7759 | int Client::fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms) | |
7760 | { | |
f67539c2 TL |
7761 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7762 | if (!mref_reader.is_state_satisfied()) | |
7763 | return -CEPHFS_ENOTCONN; | |
7764 | ||
11fdf7f2 | 7765 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
7766 | tout(cct) << fd << std::endl; |
7767 | tout(cct) << mask << std::endl; | |
7768 | ||
f67539c2 | 7769 | std::scoped_lock lock(client_lock); |
7c673cae FG |
7770 | Fh *f = get_filehandle(fd); |
7771 | if (!f) | |
f67539c2 | 7772 | return -CEPHFS_EBADF; |
7c673cae FG |
7773 | #if defined(__linux__) && defined(O_PATH) |
7774 | if (f->flags & O_PATH) | |
f67539c2 | 7775 | return -CEPHFS_EBADF; |
7c673cae FG |
7776 | #endif |
7777 | return _setattrx(f->inode, stx, mask, perms); | |
7778 | } | |
7779 | ||
7780 | int Client::stat(const char *relpath, struct stat *stbuf, const UserPerm& perms, | |
7781 | frag_info_t *dirstat, int mask) | |
7782 | { | |
f67539c2 TL |
7783 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7784 | if (!mref_reader.is_state_satisfied()) | |
7785 | return -CEPHFS_ENOTCONN; | |
7786 | ||
11fdf7f2 | 7787 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7788 | tout(cct) << "stat" << std::endl; |
7789 | tout(cct) << relpath << std::endl; | |
181888fb | 7790 | |
7c673cae FG |
7791 | filepath path(relpath); |
7792 | InodeRef in; | |
f67539c2 TL |
7793 | |
7794 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7795 | int r = path_walk(path, &in, perms, true, mask); |
7796 | if (r < 0) | |
7797 | return r; | |
7798 | r = _getattr(in, mask, perms); | |
7799 | if (r < 0) { | |
11fdf7f2 | 7800 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7801 | return r; |
7802 | } | |
7803 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 7804 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7805 | return r; |
7806 | } | |
7807 | ||
7808 | unsigned Client::statx_to_mask(unsigned int flags, unsigned int want) | |
7809 | { | |
7810 | unsigned mask = 0; | |
7811 | ||
7812 | /* if NO_ATTR_SYNC is set, then we don't need any -- just use what's in cache */ | |
7813 | if (flags & AT_NO_ATTR_SYNC) | |
7814 | goto out; | |
7815 | ||
7816 | /* Always set PIN to distinguish from AT_NO_ATTR_SYNC case */ | |
7817 | mask |= CEPH_CAP_PIN; | |
7818 | if (want & (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7819 | mask |= CEPH_CAP_AUTH_SHARED; | |
7820 | if (want & (CEPH_STATX_NLINK|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7821 | mask |= CEPH_CAP_LINK_SHARED; | |
adb31ebb | 7822 | 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 |
7823 | mask |= CEPH_CAP_FILE_SHARED; |
7824 | if (want & (CEPH_STATX_VERSION|CEPH_STATX_CTIME)) | |
7825 | mask |= CEPH_CAP_XATTR_SHARED; | |
7826 | out: | |
7827 | return mask; | |
7828 | } | |
7829 | ||
7830 | int Client::statx(const char *relpath, struct ceph_statx *stx, | |
7831 | const UserPerm& perms, | |
7832 | unsigned int want, unsigned int flags) | |
7833 | { | |
b3b6e05e | 7834 | return statxat(CEPHFS_AT_FDCWD, relpath, stx, perms, want, flags); |
7c673cae FG |
7835 | } |
7836 | ||
7837 | int Client::lstat(const char *relpath, struct stat *stbuf, | |
7838 | const UserPerm& perms, frag_info_t *dirstat, int mask) | |
7839 | { | |
f67539c2 TL |
7840 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
7841 | if (!mref_reader.is_state_satisfied()) | |
7842 | return -CEPHFS_ENOTCONN; | |
7843 | ||
11fdf7f2 | 7844 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
11fdf7f2 | 7845 | tout(cct) << __func__ << std::endl; |
7c673cae | 7846 | tout(cct) << relpath << std::endl; |
181888fb | 7847 | |
7c673cae FG |
7848 | filepath path(relpath); |
7849 | InodeRef in; | |
f67539c2 TL |
7850 | |
7851 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
7852 | // don't follow symlinks |
7853 | int r = path_walk(path, &in, perms, false, mask); | |
7854 | if (r < 0) | |
7855 | return r; | |
7856 | r = _getattr(in, mask, perms); | |
7857 | if (r < 0) { | |
11fdf7f2 | 7858 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7859 | return r; |
7860 | } | |
7861 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 7862 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7863 | return r; |
7864 | } | |
7865 | ||
7866 | int Client::fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat, nest_info_t *rstat) | |
7867 | { | |
11fdf7f2 | 7868 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
7869 | << " mode 0" << oct << in->mode << dec |
7870 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
7871 | memset(st, 0, sizeof(struct stat)); | |
7872 | if (use_faked_inos()) | |
7873 | st->st_ino = in->faked_ino; | |
7874 | else | |
7875 | st->st_ino = in->ino; | |
7876 | st->st_dev = in->snapid; | |
7877 | st->st_mode = in->mode; | |
7878 | st->st_rdev = in->rdev; | |
28e407b8 AA |
7879 | if (in->is_dir()) { |
7880 | switch (in->nlink) { | |
7881 | case 0: | |
7882 | st->st_nlink = 0; /* dir is unlinked */ | |
7883 | break; | |
7884 | case 1: | |
7885 | st->st_nlink = 1 /* parent dentry */ | |
7886 | + 1 /* <dir>/. */ | |
7887 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
7888 | break; | |
7889 | default: | |
7890 | ceph_abort(); | |
7891 | } | |
7892 | } else { | |
7893 | st->st_nlink = in->nlink; | |
7894 | } | |
7c673cae FG |
7895 | st->st_uid = in->uid; |
7896 | st->st_gid = in->gid; | |
7897 | if (in->ctime > in->mtime) { | |
7898 | stat_set_ctime_sec(st, in->ctime.sec()); | |
7899 | stat_set_ctime_nsec(st, in->ctime.nsec()); | |
7900 | } else { | |
7901 | stat_set_ctime_sec(st, in->mtime.sec()); | |
7902 | stat_set_ctime_nsec(st, in->mtime.nsec()); | |
7903 | } | |
7904 | stat_set_atime_sec(st, in->atime.sec()); | |
7905 | stat_set_atime_nsec(st, in->atime.nsec()); | |
7906 | stat_set_mtime_sec(st, in->mtime.sec()); | |
7907 | stat_set_mtime_nsec(st, in->mtime.nsec()); | |
7908 | if (in->is_dir()) { | |
7909 | if (cct->_conf->client_dirsize_rbytes) | |
7910 | st->st_size = in->rstat.rbytes; | |
7911 | else | |
7912 | st->st_size = in->dirstat.size(); | |
f67539c2 TL |
7913 | // The Windows "stat" structure provides just a subset of the fields that are |
7914 | // available on Linux. | |
7915 | #ifndef _WIN32 | |
7c673cae | 7916 | st->st_blocks = 1; |
f67539c2 | 7917 | #endif |
7c673cae FG |
7918 | } else { |
7919 | st->st_size = in->size; | |
f67539c2 | 7920 | #ifndef _WIN32 |
7c673cae | 7921 | st->st_blocks = (in->size + 511) >> 9; |
f67539c2 | 7922 | #endif |
7c673cae | 7923 | } |
f67539c2 | 7924 | #ifndef _WIN32 |
11fdf7f2 | 7925 | st->st_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
f67539c2 | 7926 | #endif |
7c673cae FG |
7927 | |
7928 | if (dirstat) | |
7929 | *dirstat = in->dirstat; | |
7930 | if (rstat) | |
7931 | *rstat = in->rstat; | |
7932 | ||
7933 | return in->caps_issued(); | |
7934 | } | |
7935 | ||
7936 | void Client::fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx) | |
7937 | { | |
11fdf7f2 | 7938 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
7939 | << " mode 0" << oct << in->mode << dec |
7940 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
7941 | memset(stx, 0, sizeof(struct ceph_statx)); | |
7942 | ||
7943 | /* | |
7944 | * If mask is 0, then the caller set AT_NO_ATTR_SYNC. Reset the mask | |
7945 | * so that all bits are set. | |
7946 | */ | |
7947 | if (!mask) | |
7948 | mask = ~0; | |
7949 | ||
7950 | /* These are always considered to be available */ | |
7951 | stx->stx_dev = in->snapid; | |
11fdf7f2 | 7952 | stx->stx_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
7c673cae FG |
7953 | |
7954 | /* Type bits are always set, even when CEPH_STATX_MODE is not */ | |
7955 | stx->stx_mode = S_IFMT & in->mode; | |
7956 | stx->stx_ino = use_faked_inos() ? in->faked_ino : (ino_t)in->ino; | |
7957 | stx->stx_rdev = in->rdev; | |
7958 | stx->stx_mask |= (CEPH_STATX_INO|CEPH_STATX_RDEV); | |
7959 | ||
7960 | if (mask & CEPH_CAP_AUTH_SHARED) { | |
7961 | stx->stx_uid = in->uid; | |
7962 | stx->stx_gid = in->gid; | |
7963 | stx->stx_mode = in->mode; | |
7964 | in->btime.to_timespec(&stx->stx_btime); | |
7965 | stx->stx_mask |= (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME); | |
7966 | } | |
7967 | ||
7968 | if (mask & CEPH_CAP_LINK_SHARED) { | |
28e407b8 AA |
7969 | if (in->is_dir()) { |
7970 | switch (in->nlink) { | |
7971 | case 0: | |
7972 | stx->stx_nlink = 0; /* dir is unlinked */ | |
7973 | break; | |
7974 | case 1: | |
7975 | stx->stx_nlink = 1 /* parent dentry */ | |
7976 | + 1 /* <dir>/. */ | |
7977 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
7978 | break; | |
7979 | default: | |
7980 | ceph_abort(); | |
7981 | } | |
7982 | } else { | |
7983 | stx->stx_nlink = in->nlink; | |
7984 | } | |
7c673cae FG |
7985 | stx->stx_mask |= CEPH_STATX_NLINK; |
7986 | } | |
7987 | ||
7988 | if (mask & CEPH_CAP_FILE_SHARED) { | |
7989 | ||
7990 | in->atime.to_timespec(&stx->stx_atime); | |
7991 | in->mtime.to_timespec(&stx->stx_mtime); | |
7992 | ||
7993 | if (in->is_dir()) { | |
7994 | if (cct->_conf->client_dirsize_rbytes) | |
7995 | stx->stx_size = in->rstat.rbytes; | |
7996 | else | |
7997 | stx->stx_size = in->dirstat.size(); | |
7998 | stx->stx_blocks = 1; | |
7999 | } else { | |
8000 | stx->stx_size = in->size; | |
8001 | stx->stx_blocks = (in->size + 511) >> 9; | |
8002 | } | |
8003 | stx->stx_mask |= (CEPH_STATX_ATIME|CEPH_STATX_MTIME| | |
8004 | CEPH_STATX_SIZE|CEPH_STATX_BLOCKS); | |
8005 | } | |
8006 | ||
8007 | /* Change time and change_attr both require all shared caps to view */ | |
8008 | if ((mask & CEPH_STAT_CAP_INODE_ALL) == CEPH_STAT_CAP_INODE_ALL) { | |
8009 | stx->stx_version = in->change_attr; | |
8010 | if (in->ctime > in->mtime) | |
8011 | in->ctime.to_timespec(&stx->stx_ctime); | |
8012 | else | |
8013 | in->mtime.to_timespec(&stx->stx_ctime); | |
8014 | stx->stx_mask |= (CEPH_STATX_CTIME|CEPH_STATX_VERSION); | |
8015 | } | |
8016 | ||
8017 | } | |
8018 | ||
8019 | void Client::touch_dn(Dentry *dn) | |
8020 | { | |
8021 | lru.lru_touch(dn); | |
8022 | } | |
8023 | ||
8024 | int Client::chmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
8025 | { | |
b3b6e05e | 8026 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, 0, perms); |
7c673cae FG |
8027 | } |
8028 | ||
8029 | int Client::fchmod(int fd, mode_t mode, const UserPerm& perms) | |
8030 | { | |
f67539c2 TL |
8031 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8032 | if (!mref_reader.is_state_satisfied()) | |
8033 | return -CEPHFS_ENOTCONN; | |
8034 | ||
11fdf7f2 | 8035 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8036 | tout(cct) << fd << std::endl; |
8037 | tout(cct) << mode << std::endl; | |
181888fb | 8038 | |
f67539c2 | 8039 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8040 | Fh *f = get_filehandle(fd); |
8041 | if (!f) | |
f67539c2 | 8042 | return -CEPHFS_EBADF; |
7c673cae FG |
8043 | #if defined(__linux__) && defined(O_PATH) |
8044 | if (f->flags & O_PATH) | |
f67539c2 | 8045 | return -CEPHFS_EBADF; |
7c673cae FG |
8046 | #endif |
8047 | struct stat attr; | |
8048 | attr.st_mode = mode; | |
8049 | return _setattr(f->inode, &attr, CEPH_SETATTR_MODE, perms); | |
8050 | } | |
8051 | ||
b3b6e05e TL |
8052 | int Client::chmodat(int dirfd, const char *relpath, mode_t mode, int flags, |
8053 | const UserPerm& perms) { | |
f67539c2 | 8054 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8055 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8056 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8057 | } |
f67539c2 | 8058 | |
11fdf7f2 | 8059 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8060 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8061 | tout(cct) << relpath << std::endl; |
8062 | tout(cct) << mode << std::endl; | |
b3b6e05e | 8063 | tout(cct) << flags << std::endl; |
181888fb | 8064 | |
7c673cae FG |
8065 | filepath path(relpath); |
8066 | InodeRef in; | |
b3b6e05e | 8067 | InodeRef dirinode; |
f67539c2 TL |
8068 | |
8069 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8070 | int r = get_fd_inode(dirfd, &dirinode); |
8071 | if (r < 0) { | |
8072 | return r; | |
8073 | } | |
8074 | ||
8075 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8076 | if (r < 0) { | |
7c673cae | 8077 | return r; |
b3b6e05e | 8078 | } |
7c673cae FG |
8079 | struct stat attr; |
8080 | attr.st_mode = mode; | |
8081 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
8082 | } | |
8083 | ||
b3b6e05e TL |
8084 | int Client::lchmod(const char *relpath, mode_t mode, const UserPerm& perms) |
8085 | { | |
8086 | return chmodat(CEPHFS_AT_FDCWD, relpath, mode, AT_SYMLINK_NOFOLLOW, perms); | |
8087 | } | |
8088 | ||
7c673cae FG |
8089 | int Client::chown(const char *relpath, uid_t new_uid, gid_t new_gid, |
8090 | const UserPerm& perms) | |
8091 | { | |
b3b6e05e | 8092 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, 0, perms); |
7c673cae FG |
8093 | } |
8094 | ||
8095 | int Client::fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms) | |
8096 | { | |
f67539c2 TL |
8097 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8098 | if (!mref_reader.is_state_satisfied()) | |
8099 | return -CEPHFS_ENOTCONN; | |
8100 | ||
11fdf7f2 | 8101 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8102 | tout(cct) << fd << std::endl; |
8103 | tout(cct) << new_uid << std::endl; | |
8104 | tout(cct) << new_gid << std::endl; | |
181888fb | 8105 | |
f67539c2 | 8106 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8107 | Fh *f = get_filehandle(fd); |
8108 | if (!f) | |
f67539c2 | 8109 | return -CEPHFS_EBADF; |
7c673cae FG |
8110 | #if defined(__linux__) && defined(O_PATH) |
8111 | if (f->flags & O_PATH) | |
f67539c2 | 8112 | return -CEPHFS_EBADF; |
7c673cae FG |
8113 | #endif |
8114 | struct stat attr; | |
8115 | attr.st_uid = new_uid; | |
8116 | attr.st_gid = new_gid; | |
8117 | int mask = 0; | |
8118 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
8119 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
8120 | return _setattr(f->inode, &attr, mask, perms); | |
8121 | } | |
8122 | ||
8123 | int Client::lchown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
8124 | const UserPerm& perms) | |
8125 | { | |
b3b6e05e TL |
8126 | return chownat(CEPHFS_AT_FDCWD, relpath, new_uid, new_gid, AT_SYMLINK_NOFOLLOW, perms); |
8127 | } | |
8128 | ||
8129 | int Client::chownat(int dirfd, const char *relpath, uid_t new_uid, gid_t new_gid, | |
8130 | int flags, const UserPerm& perms) { | |
f67539c2 | 8131 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
b3b6e05e | 8132 | if (!mref_reader.is_state_satisfied()) { |
f67539c2 | 8133 | return -CEPHFS_ENOTCONN; |
b3b6e05e | 8134 | } |
f67539c2 | 8135 | |
11fdf7f2 | 8136 | tout(cct) << __func__ << std::endl; |
b3b6e05e | 8137 | tout(cct) << dirfd << std::endl; |
7c673cae FG |
8138 | tout(cct) << relpath << std::endl; |
8139 | tout(cct) << new_uid << std::endl; | |
8140 | tout(cct) << new_gid << std::endl; | |
b3b6e05e | 8141 | tout(cct) << flags << std::endl; |
181888fb | 8142 | |
7c673cae FG |
8143 | filepath path(relpath); |
8144 | InodeRef in; | |
b3b6e05e | 8145 | InodeRef dirinode; |
f67539c2 TL |
8146 | |
8147 | std::scoped_lock lock(client_lock); | |
b3b6e05e TL |
8148 | int r = get_fd_inode(dirfd, &dirinode); |
8149 | if (r < 0) { | |
7c673cae | 8150 | return r; |
b3b6e05e TL |
8151 | } |
8152 | ||
8153 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8154 | if (r < 0) { | |
8155 | return r; | |
8156 | } | |
7c673cae FG |
8157 | struct stat attr; |
8158 | attr.st_uid = new_uid; | |
8159 | attr.st_gid = new_gid; | |
b3b6e05e | 8160 | return _setattr(in, &attr, CEPH_SETATTR_UID|CEPH_SETATTR_GID, perms); |
7c673cae FG |
8161 | } |
8162 | ||
11fdf7f2 TL |
8163 | static void attr_set_atime_and_mtime(struct stat *attr, |
8164 | const utime_t &atime, | |
8165 | const utime_t &mtime) | |
8166 | { | |
8167 | stat_set_atime_sec(attr, atime.tv.tv_sec); | |
8168 | stat_set_atime_nsec(attr, atime.tv.tv_nsec); | |
8169 | stat_set_mtime_sec(attr, mtime.tv.tv_sec); | |
8170 | stat_set_mtime_nsec(attr, mtime.tv.tv_nsec); | |
8171 | } | |
8172 | ||
8173 | // for [l]utime() invoke the timeval variant as the timespec | |
8174 | // variant are not yet implemented. for futime[s](), invoke | |
8175 | // the timespec variant. | |
7c673cae FG |
8176 | int Client::utime(const char *relpath, struct utimbuf *buf, |
8177 | const UserPerm& perms) | |
8178 | { | |
11fdf7f2 TL |
8179 | struct timeval tv[2]; |
8180 | tv[0].tv_sec = buf->actime; | |
8181 | tv[0].tv_usec = 0; | |
8182 | tv[1].tv_sec = buf->modtime; | |
8183 | tv[1].tv_usec = 0; | |
8184 | ||
8185 | return utimes(relpath, tv, perms); | |
8186 | } | |
8187 | ||
8188 | int Client::lutime(const char *relpath, struct utimbuf *buf, | |
8189 | const UserPerm& perms) | |
8190 | { | |
8191 | struct timeval tv[2]; | |
8192 | tv[0].tv_sec = buf->actime; | |
8193 | tv[0].tv_usec = 0; | |
8194 | tv[1].tv_sec = buf->modtime; | |
8195 | tv[1].tv_usec = 0; | |
8196 | ||
8197 | return lutimes(relpath, tv, perms); | |
8198 | } | |
8199 | ||
8200 | int Client::futime(int fd, struct utimbuf *buf, const UserPerm& perms) | |
8201 | { | |
8202 | struct timespec ts[2]; | |
8203 | ts[0].tv_sec = buf->actime; | |
8204 | ts[0].tv_nsec = 0; | |
8205 | ts[1].tv_sec = buf->modtime; | |
8206 | ts[1].tv_nsec = 0; | |
8207 | ||
8208 | return futimens(fd, ts, perms); | |
8209 | } | |
8210 | ||
8211 | int Client::utimes(const char *relpath, struct timeval times[2], | |
8212 | const UserPerm& perms) | |
8213 | { | |
f67539c2 TL |
8214 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8215 | if (!mref_reader.is_state_satisfied()) | |
8216 | return -CEPHFS_ENOTCONN; | |
8217 | ||
11fdf7f2 | 8218 | tout(cct) << __func__ << std::endl; |
7c673cae | 8219 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8220 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8221 | << std::endl; | |
8222 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8223 | << std::endl; | |
181888fb | 8224 | |
7c673cae FG |
8225 | filepath path(relpath); |
8226 | InodeRef in; | |
f67539c2 TL |
8227 | |
8228 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8229 | int r = path_walk(path, &in, perms); |
8230 | if (r < 0) | |
8231 | return r; | |
8232 | struct stat attr; | |
11fdf7f2 TL |
8233 | utime_t atime(times[0]); |
8234 | utime_t mtime(times[1]); | |
8235 | ||
8236 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8237 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8238 | } | |
8239 | ||
11fdf7f2 TL |
8240 | int Client::lutimes(const char *relpath, struct timeval times[2], |
8241 | const UserPerm& perms) | |
7c673cae | 8242 | { |
f67539c2 TL |
8243 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8244 | if (!mref_reader.is_state_satisfied()) | |
8245 | return -CEPHFS_ENOTCONN; | |
8246 | ||
11fdf7f2 | 8247 | tout(cct) << __func__ << std::endl; |
7c673cae | 8248 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
8249 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
8250 | << std::endl; | |
8251 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
8252 | << std::endl; | |
181888fb | 8253 | |
7c673cae FG |
8254 | filepath path(relpath); |
8255 | InodeRef in; | |
f67539c2 TL |
8256 | |
8257 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8258 | int r = path_walk(path, &in, perms, false); |
8259 | if (r < 0) | |
8260 | return r; | |
8261 | struct stat attr; | |
11fdf7f2 TL |
8262 | utime_t atime(times[0]); |
8263 | utime_t mtime(times[1]); | |
8264 | ||
8265 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
8266 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
8267 | } | |
8268 | ||
11fdf7f2 TL |
8269 | int Client::futimes(int fd, struct timeval times[2], const UserPerm& perms) |
8270 | { | |
8271 | struct timespec ts[2]; | |
8272 | ts[0].tv_sec = times[0].tv_sec; | |
8273 | ts[0].tv_nsec = times[0].tv_usec * 1000; | |
8274 | ts[1].tv_sec = times[1].tv_sec; | |
8275 | ts[1].tv_nsec = times[1].tv_usec * 1000; | |
8276 | ||
8277 | return futimens(fd, ts, perms); | |
8278 | } | |
8279 | ||
8280 | int Client::futimens(int fd, struct timespec times[2], const UserPerm& perms) | |
8281 | { | |
f67539c2 TL |
8282 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8283 | if (!mref_reader.is_state_satisfied()) | |
8284 | return -CEPHFS_ENOTCONN; | |
8285 | ||
11fdf7f2 TL |
8286 | tout(cct) << __func__ << std::endl; |
8287 | tout(cct) << fd << std::endl; | |
8288 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8289 | << std::endl; | |
8290 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8291 | << std::endl; | |
8292 | ||
f67539c2 | 8293 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
8294 | Fh *f = get_filehandle(fd); |
8295 | if (!f) | |
f67539c2 | 8296 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8297 | #if defined(__linux__) && defined(O_PATH) |
8298 | if (f->flags & O_PATH) | |
f67539c2 | 8299 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
8300 | #endif |
8301 | struct stat attr; | |
8302 | utime_t atime(times[0]); | |
8303 | utime_t mtime(times[1]); | |
8304 | ||
8305 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8306 | return _setattr(f->inode, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8307 | } | |
8308 | ||
b3b6e05e TL |
8309 | int Client::utimensat(int dirfd, const char *relpath, struct timespec times[2], int flags, |
8310 | const UserPerm& perms) { | |
8311 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8312 | if (!mref_reader.is_state_satisfied()) { | |
8313 | return -CEPHFS_ENOTCONN; | |
8314 | } | |
8315 | ||
8316 | tout(cct) << __func__ << std::endl; | |
8317 | tout(cct) << dirfd << std::endl; | |
8318 | tout(cct) << relpath << std::endl; | |
8319 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
8320 | << std::endl; | |
8321 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
8322 | << std::endl; | |
8323 | tout(cct) << flags << std::endl; | |
8324 | ||
8325 | filepath path(relpath); | |
8326 | InodeRef in; | |
8327 | InodeRef dirinode; | |
8328 | ||
8329 | std::scoped_lock lock(client_lock); | |
8330 | int r = get_fd_inode(dirfd, &dirinode); | |
8331 | if (r < 0) { | |
8332 | return r; | |
8333 | } | |
8334 | ||
8335 | #if defined(__linux__) && defined(O_PATH) | |
8336 | if (flags & O_PATH) { | |
8337 | return -CEPHFS_EBADF; | |
8338 | } | |
8339 | #endif | |
8340 | ||
8341 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), 0, dirinode); | |
8342 | if (r < 0) { | |
8343 | return r; | |
8344 | } | |
8345 | struct stat attr; | |
8346 | utime_t atime(times[0]); | |
8347 | utime_t mtime(times[1]); | |
8348 | ||
8349 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
8350 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
8351 | } | |
8352 | ||
7c673cae FG |
8353 | int Client::flock(int fd, int operation, uint64_t owner) |
8354 | { | |
f67539c2 TL |
8355 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8356 | if (!mref_reader.is_state_satisfied()) | |
8357 | return -CEPHFS_ENOTCONN; | |
8358 | ||
11fdf7f2 | 8359 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
8360 | tout(cct) << fd << std::endl; |
8361 | tout(cct) << operation << std::endl; | |
8362 | tout(cct) << owner << std::endl; | |
181888fb | 8363 | |
f67539c2 | 8364 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8365 | Fh *f = get_filehandle(fd); |
8366 | if (!f) | |
f67539c2 | 8367 | return -CEPHFS_EBADF; |
7c673cae FG |
8368 | |
8369 | return _flock(f, operation, owner); | |
8370 | } | |
8371 | ||
8372 | int Client::opendir(const char *relpath, dir_result_t **dirpp, const UserPerm& perms) | |
8373 | { | |
f67539c2 TL |
8374 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8375 | if (!mref_reader.is_state_satisfied()) | |
8376 | return -CEPHFS_ENOTCONN; | |
8377 | ||
11fdf7f2 | 8378 | tout(cct) << __func__ << std::endl; |
7c673cae | 8379 | tout(cct) << relpath << std::endl; |
181888fb | 8380 | |
7c673cae FG |
8381 | filepath path(relpath); |
8382 | InodeRef in; | |
f67539c2 TL |
8383 | |
8384 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
8385 | int r = path_walk(path, &in, perms, true); |
8386 | if (r < 0) | |
8387 | return r; | |
8388 | if (cct->_conf->client_permissions) { | |
8389 | int r = may_open(in.get(), O_RDONLY, perms); | |
8390 | if (r < 0) | |
8391 | return r; | |
8392 | } | |
8393 | r = _opendir(in.get(), dirpp, perms); | |
8394 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
f67539c2 TL |
8395 | if (r != -CEPHFS_ENOTDIR) |
8396 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
7c673cae FG |
8397 | return r; |
8398 | } | |
8399 | ||
b3b6e05e TL |
8400 | int Client::fdopendir(int dirfd, dir_result_t **dirpp, const UserPerm &perms) { |
8401 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
8402 | if (!mref_reader.is_state_satisfied()) { | |
8403 | return -CEPHFS_ENOTCONN; | |
8404 | } | |
8405 | ||
8406 | tout(cct) << __func__ << std::endl; | |
8407 | tout(cct) << dirfd << std::endl; | |
8408 | ||
8409 | InodeRef dirinode; | |
8410 | std::scoped_lock locker(client_lock); | |
8411 | int r = get_fd_inode(dirfd, &dirinode); | |
8412 | if (r < 0) { | |
8413 | return r; | |
8414 | } | |
8415 | ||
8416 | if (cct->_conf->client_permissions) { | |
8417 | r = may_open(dirinode.get(), O_RDONLY, perms); | |
8418 | if (r < 0) { | |
8419 | return r; | |
8420 | } | |
8421 | } | |
8422 | r = _opendir(dirinode.get(), dirpp, perms); | |
8423 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
8424 | if (r != -CEPHFS_ENOTDIR) { | |
8425 | tout(cct) << (uintptr_t)*dirpp << std::endl; | |
8426 | } | |
8427 | return r; | |
8428 | } | |
8429 | ||
7c673cae FG |
8430 | int Client::_opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms) |
8431 | { | |
8432 | if (!in->is_dir()) | |
f67539c2 | 8433 | return -CEPHFS_ENOTDIR; |
7c673cae FG |
8434 | *dirpp = new dir_result_t(in, perms); |
8435 | opened_dirs.insert(*dirpp); | |
11fdf7f2 | 8436 | ldout(cct, 8) << __func__ << "(" << in->ino << ") = " << 0 << " (" << *dirpp << ")" << dendl; |
7c673cae FG |
8437 | return 0; |
8438 | } | |
8439 | ||
8440 | ||
8441 | int Client::closedir(dir_result_t *dir) | |
8442 | { | |
11fdf7f2 | 8443 | tout(cct) << __func__ << std::endl; |
f67539c2 | 8444 | tout(cct) << (uintptr_t)dir << std::endl; |
7c673cae | 8445 | |
11fdf7f2 | 8446 | ldout(cct, 3) << __func__ << "(" << dir << ") = 0" << dendl; |
f67539c2 | 8447 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8448 | _closedir(dir); |
8449 | return 0; | |
8450 | } | |
8451 | ||
8452 | void Client::_closedir(dir_result_t *dirp) | |
8453 | { | |
11fdf7f2 | 8454 | ldout(cct, 10) << __func__ << "(" << dirp << ")" << dendl; |
f67539c2 | 8455 | |
7c673cae | 8456 | if (dirp->inode) { |
11fdf7f2 | 8457 | ldout(cct, 10) << __func__ << " detaching inode " << dirp->inode << dendl; |
7c673cae FG |
8458 | dirp->inode.reset(); |
8459 | } | |
8460 | _readdir_drop_dirp_buffer(dirp); | |
8461 | opened_dirs.erase(dirp); | |
8462 | delete dirp; | |
8463 | } | |
8464 | ||
8465 | void Client::rewinddir(dir_result_t *dirp) | |
8466 | { | |
11fdf7f2 | 8467 | ldout(cct, 3) << __func__ << "(" << dirp << ")" << dendl; |
181888fb | 8468 | |
f67539c2 TL |
8469 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8470 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8471 | return; |
8472 | ||
f67539c2 | 8473 | std::scoped_lock lock(client_lock); |
7c673cae FG |
8474 | dir_result_t *d = static_cast<dir_result_t*>(dirp); |
8475 | _readdir_drop_dirp_buffer(d); | |
8476 | d->reset(); | |
8477 | } | |
8478 | ||
8479 | loff_t Client::telldir(dir_result_t *dirp) | |
8480 | { | |
8481 | dir_result_t *d = static_cast<dir_result_t*>(dirp); | |
11fdf7f2 | 8482 | ldout(cct, 3) << __func__ << "(" << dirp << ") = " << d->offset << dendl; |
7c673cae FG |
8483 | return d->offset; |
8484 | } | |
8485 | ||
8486 | void Client::seekdir(dir_result_t *dirp, loff_t offset) | |
8487 | { | |
11fdf7f2 | 8488 | ldout(cct, 3) << __func__ << "(" << dirp << ", " << offset << ")" << dendl; |
7c673cae | 8489 | |
f67539c2 TL |
8490 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8491 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
8492 | return; |
8493 | ||
f67539c2 TL |
8494 | std::scoped_lock lock(client_lock); |
8495 | ||
7c673cae FG |
8496 | if (offset == dirp->offset) |
8497 | return; | |
8498 | ||
8499 | if (offset > dirp->offset) | |
8500 | dirp->release_count = 0; // bump if we do a forward seek | |
8501 | else | |
8502 | dirp->ordered_count = 0; // disable filling readdir cache | |
8503 | ||
8504 | if (dirp->hash_order()) { | |
8505 | if (dirp->offset > offset) { | |
8506 | _readdir_drop_dirp_buffer(dirp); | |
8507 | dirp->reset(); | |
8508 | } | |
8509 | } else { | |
8510 | if (offset == 0 || | |
8511 | dirp->buffer_frag != frag_t(dir_result_t::fpos_high(offset)) || | |
8512 | dirp->offset_low() > dir_result_t::fpos_low(offset)) { | |
8513 | _readdir_drop_dirp_buffer(dirp); | |
8514 | dirp->reset(); | |
8515 | } | |
8516 | } | |
8517 | ||
8518 | dirp->offset = offset; | |
8519 | } | |
8520 | ||
8521 | ||
8522 | //struct dirent { | |
8523 | // ino_t d_ino; /* inode number */ | |
8524 | // off_t d_off; /* offset to the next dirent */ | |
8525 | // unsigned short d_reclen; /* length of this record */ | |
8526 | // unsigned char d_type; /* type of file */ | |
8527 | // char d_name[256]; /* filename */ | |
8528 | //}; | |
8529 | void Client::fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off) | |
8530 | { | |
8531 | strncpy(de->d_name, name, 255); | |
8532 | de->d_name[255] = '\0'; | |
f67539c2 | 8533 | #if !defined(__CYGWIN__) && !(defined(_WIN32)) |
7c673cae | 8534 | de->d_ino = ino; |
11fdf7f2 | 8535 | #if !defined(__APPLE__) && !defined(__FreeBSD__) |
7c673cae FG |
8536 | de->d_off = next_off; |
8537 | #endif | |
8538 | de->d_reclen = 1; | |
8539 | de->d_type = IFTODT(type); | |
11fdf7f2 | 8540 | ldout(cct, 10) << __func__ << " '" << de->d_name << "' -> " << inodeno_t(de->d_ino) |
7c673cae FG |
8541 | << " type " << (int)de->d_type << " w/ next_off " << hex << next_off << dec << dendl; |
8542 | #endif | |
8543 | } | |
8544 | ||
8545 | void Client::_readdir_next_frag(dir_result_t *dirp) | |
8546 | { | |
8547 | frag_t fg = dirp->buffer_frag; | |
8548 | ||
8549 | if (fg.is_rightmost()) { | |
11fdf7f2 | 8550 | ldout(cct, 10) << __func__ << " advance from " << fg << " to END" << dendl; |
7c673cae FG |
8551 | dirp->set_end(); |
8552 | return; | |
8553 | } | |
8554 | ||
8555 | // advance | |
8556 | fg = fg.next(); | |
11fdf7f2 | 8557 | ldout(cct, 10) << __func__ << " advance from " << dirp->buffer_frag << " to " << fg << dendl; |
7c673cae FG |
8558 | |
8559 | if (dirp->hash_order()) { | |
8560 | // keep last_name | |
8561 | int64_t new_offset = dir_result_t::make_fpos(fg.value(), 2, true); | |
8562 | if (dirp->offset < new_offset) // don't decrease offset | |
8563 | dirp->offset = new_offset; | |
8564 | } else { | |
8565 | dirp->last_name.clear(); | |
8566 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); | |
8567 | _readdir_rechoose_frag(dirp); | |
8568 | } | |
8569 | } | |
8570 | ||
8571 | void Client::_readdir_rechoose_frag(dir_result_t *dirp) | |
8572 | { | |
11fdf7f2 | 8573 | ceph_assert(dirp->inode); |
7c673cae FG |
8574 | |
8575 | if (dirp->hash_order()) | |
8576 | return; | |
8577 | ||
8578 | frag_t cur = frag_t(dirp->offset_high()); | |
8579 | frag_t fg = dirp->inode->dirfragtree[cur.value()]; | |
8580 | if (fg != cur) { | |
11fdf7f2 | 8581 | ldout(cct, 10) << __func__ << " frag " << cur << " maps to " << fg << dendl; |
7c673cae FG |
8582 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); |
8583 | dirp->last_name.clear(); | |
8584 | dirp->next_offset = 2; | |
8585 | } | |
8586 | } | |
8587 | ||
8588 | void Client::_readdir_drop_dirp_buffer(dir_result_t *dirp) | |
8589 | { | |
11fdf7f2 | 8590 | ldout(cct, 10) << __func__ << " " << dirp << dendl; |
7c673cae FG |
8591 | dirp->buffer.clear(); |
8592 | } | |
8593 | ||
8594 | int Client::_readdir_get_frag(dir_result_t *dirp) | |
8595 | { | |
11fdf7f2 TL |
8596 | ceph_assert(dirp); |
8597 | ceph_assert(dirp->inode); | |
7c673cae FG |
8598 | |
8599 | // get the current frag. | |
8600 | frag_t fg; | |
8601 | if (dirp->hash_order()) | |
8602 | fg = dirp->inode->dirfragtree[dirp->offset_high()]; | |
8603 | else | |
8604 | fg = frag_t(dirp->offset_high()); | |
8605 | ||
11fdf7f2 | 8606 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " fg " << fg |
7c673cae FG |
8607 | << " offset " << hex << dirp->offset << dec << dendl; |
8608 | ||
8609 | int op = CEPH_MDS_OP_READDIR; | |
8610 | if (dirp->inode && dirp->inode->snapid == CEPH_SNAPDIR) | |
8611 | op = CEPH_MDS_OP_LSSNAP; | |
8612 | ||
8613 | InodeRef& diri = dirp->inode; | |
8614 | ||
8615 | MetaRequest *req = new MetaRequest(op); | |
8616 | filepath path; | |
8617 | diri->make_nosnap_relative_path(path); | |
8618 | req->set_filepath(path); | |
8619 | req->set_inode(diri.get()); | |
8620 | req->head.args.readdir.frag = fg; | |
8621 | req->head.args.readdir.flags = CEPH_READDIR_REPLY_BITFLAGS; | |
8622 | if (dirp->last_name.length()) { | |
94b18763 | 8623 | req->path2.set_path(dirp->last_name); |
7c673cae FG |
8624 | } else if (dirp->hash_order()) { |
8625 | req->head.args.readdir.offset_hash = dirp->offset_high(); | |
8626 | } | |
8627 | req->dirp = dirp; | |
8628 | ||
8629 | bufferlist dirbl; | |
8630 | int res = make_request(req, dirp->perms, NULL, NULL, -1, &dirbl); | |
8631 | ||
f67539c2 | 8632 | if (res == -CEPHFS_EAGAIN) { |
11fdf7f2 | 8633 | ldout(cct, 10) << __func__ << " got EAGAIN, retrying" << dendl; |
7c673cae FG |
8634 | _readdir_rechoose_frag(dirp); |
8635 | return _readdir_get_frag(dirp); | |
8636 | } | |
8637 | ||
8638 | if (res == 0) { | |
11fdf7f2 | 8639 | ldout(cct, 10) << __func__ << " " << dirp << " got frag " << dirp->buffer_frag |
7c673cae FG |
8640 | << " size " << dirp->buffer.size() << dendl; |
8641 | } else { | |
11fdf7f2 | 8642 | ldout(cct, 10) << __func__ << " got error " << res << ", setting end flag" << dendl; |
7c673cae FG |
8643 | dirp->set_end(); |
8644 | } | |
8645 | ||
8646 | return res; | |
8647 | } | |
8648 | ||
8649 | struct dentry_off_lt { | |
8650 | bool operator()(const Dentry* dn, int64_t off) const { | |
8651 | return dir_result_t::fpos_cmp(dn->offset, off) < 0; | |
8652 | } | |
8653 | }; | |
8654 | ||
8655 | int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, | |
8656 | int caps, bool getref) | |
8657 | { | |
f67539c2 | 8658 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11fdf7f2 | 8659 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino |
7c673cae FG |
8660 | << " last_name " << dirp->last_name << " offset " << hex << dirp->offset << dec |
8661 | << dendl; | |
8662 | Dir *dir = dirp->inode->dir; | |
8663 | ||
8664 | if (!dir) { | |
8665 | ldout(cct, 10) << " dir is empty" << dendl; | |
8666 | dirp->set_end(); | |
8667 | return 0; | |
8668 | } | |
8669 | ||
8670 | vector<Dentry*>::iterator pd = std::lower_bound(dir->readdir_cache.begin(), | |
8671 | dir->readdir_cache.end(), | |
8672 | dirp->offset, dentry_off_lt()); | |
8673 | ||
8674 | string dn_name; | |
8675 | while (true) { | |
adb31ebb | 8676 | int mask = caps; |
7c673cae | 8677 | if (!dirp->inode->is_complete_and_ordered()) |
f67539c2 | 8678 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8679 | if (pd == dir->readdir_cache.end()) |
8680 | break; | |
8681 | Dentry *dn = *pd; | |
8682 | if (dn->inode == NULL) { | |
8683 | ldout(cct, 15) << " skipping null '" << dn->name << "'" << dendl; | |
8684 | ++pd; | |
8685 | continue; | |
8686 | } | |
8687 | if (dn->cap_shared_gen != dir->parent_inode->shared_gen) { | |
8688 | ldout(cct, 15) << " skipping mismatch shared gen '" << dn->name << "'" << dendl; | |
8689 | ++pd; | |
8690 | continue; | |
8691 | } | |
8692 | ||
92f5a8d4 | 8693 | int idx = pd - dir->readdir_cache.begin(); |
adb31ebb TL |
8694 | if (dn->inode->is_dir()) { |
8695 | mask |= CEPH_STAT_RSTAT; | |
8696 | } | |
8697 | int r = _getattr(dn->inode, mask, dirp->perms); | |
7c673cae FG |
8698 | if (r < 0) |
8699 | return r; | |
92f5a8d4 TL |
8700 | |
8701 | // the content of readdir_cache may change after _getattr(), so pd may be invalid iterator | |
8702 | pd = dir->readdir_cache.begin() + idx; | |
8703 | if (pd >= dir->readdir_cache.end() || *pd != dn) | |
f67539c2 | 8704 | return -CEPHFS_EAGAIN; |
7c673cae FG |
8705 | |
8706 | struct ceph_statx stx; | |
8707 | struct dirent de; | |
8708 | fill_statx(dn->inode, caps, &stx); | |
8709 | ||
8710 | uint64_t next_off = dn->offset + 1; | |
eafe8130 | 8711 | fill_dirent(&de, dn->name.c_str(), stx.stx_mode, stx.stx_ino, next_off); |
7c673cae FG |
8712 | ++pd; |
8713 | if (pd == dir->readdir_cache.end()) | |
8714 | next_off = dir_result_t::END; | |
8715 | ||
8716 | Inode *in = NULL; | |
7c673cae FG |
8717 | if (getref) { |
8718 | in = dn->inode.get(); | |
8719 | _ll_get(in); | |
8720 | } | |
8721 | ||
8722 | dn_name = dn->name; // fill in name while we have lock | |
8723 | ||
9f95a23c | 8724 | client_lock.unlock(); |
7c673cae | 8725 | r = cb(p, &de, &stx, next_off, in); // _next_ offset |
9f95a23c | 8726 | client_lock.lock(); |
7c673cae FG |
8727 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << dn->offset << dec |
8728 | << " = " << r << dendl; | |
8729 | if (r < 0) { | |
8730 | return r; | |
8731 | } | |
8732 | ||
8733 | dirp->offset = next_off; | |
8734 | if (dirp->at_end()) | |
8735 | dirp->next_offset = 2; | |
8736 | else | |
8737 | dirp->next_offset = dirp->offset_low(); | |
8738 | dirp->last_name = dn_name; // we successfully returned this one; update! | |
28e407b8 | 8739 | dirp->release_count = 0; // last_name no longer match cache index |
7c673cae FG |
8740 | if (r > 0) |
8741 | return r; | |
8742 | } | |
8743 | ||
11fdf7f2 | 8744 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " at end" << dendl; |
7c673cae FG |
8745 | dirp->set_end(); |
8746 | return 0; | |
8747 | } | |
8748 | ||
8749 | int Client::readdir_r_cb(dir_result_t *d, add_dirent_cb_t cb, void *p, | |
8750 | unsigned want, unsigned flags, bool getref) | |
8751 | { | |
8752 | int caps = statx_to_mask(flags, want); | |
8753 | ||
f67539c2 TL |
8754 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
8755 | if (!mref_reader.is_state_satisfied()) | |
8756 | return -CEPHFS_ENOTCONN; | |
7c673cae | 8757 | |
f67539c2 | 8758 | std::unique_lock cl(client_lock); |
181888fb | 8759 | |
7c673cae FG |
8760 | dir_result_t *dirp = static_cast<dir_result_t*>(d); |
8761 | ||
11fdf7f2 | 8762 | ldout(cct, 10) << __func__ << " " << *dirp->inode << " offset " << hex << dirp->offset |
7c673cae FG |
8763 | << dec << " at_end=" << dirp->at_end() |
8764 | << " hash_order=" << dirp->hash_order() << dendl; | |
8765 | ||
8766 | struct dirent de; | |
8767 | struct ceph_statx stx; | |
8768 | memset(&de, 0, sizeof(de)); | |
8769 | memset(&stx, 0, sizeof(stx)); | |
8770 | ||
8771 | InodeRef& diri = dirp->inode; | |
8772 | ||
8773 | if (dirp->at_end()) | |
8774 | return 0; | |
8775 | ||
8776 | if (dirp->offset == 0) { | |
8777 | ldout(cct, 15) << " including ." << dendl; | |
11fdf7f2 | 8778 | ceph_assert(diri->dentries.size() < 2); // can't have multiple hard-links to a dir |
7c673cae FG |
8779 | uint64_t next_off = 1; |
8780 | ||
8781 | int r; | |
adb31ebb | 8782 | r = _getattr(diri, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
8783 | if (r < 0) |
8784 | return r; | |
8785 | ||
8786 | fill_statx(diri, caps, &stx); | |
8787 | fill_dirent(&de, ".", S_IFDIR, stx.stx_ino, next_off); | |
8788 | ||
8789 | Inode *inode = NULL; | |
8790 | if (getref) { | |
8791 | inode = diri.get(); | |
8792 | _ll_get(inode); | |
8793 | } | |
8794 | ||
f67539c2 | 8795 | cl.unlock(); |
7c673cae | 8796 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 8797 | cl.lock(); |
7c673cae FG |
8798 | if (r < 0) |
8799 | return r; | |
8800 | ||
8801 | dirp->offset = next_off; | |
8802 | if (r > 0) | |
8803 | return r; | |
8804 | } | |
8805 | if (dirp->offset == 1) { | |
8806 | ldout(cct, 15) << " including .." << dendl; | |
8807 | uint64_t next_off = 2; | |
8808 | InodeRef in; | |
11fdf7f2 | 8809 | if (diri->dentries.empty()) |
7c673cae FG |
8810 | in = diri; |
8811 | else | |
94b18763 | 8812 | in = diri->get_first_parent()->dir->parent_inode; |
7c673cae FG |
8813 | |
8814 | int r; | |
adb31ebb | 8815 | r = _getattr(in, caps | CEPH_STAT_RSTAT, dirp->perms); |
7c673cae FG |
8816 | if (r < 0) |
8817 | return r; | |
8818 | ||
8819 | fill_statx(in, caps, &stx); | |
8820 | fill_dirent(&de, "..", S_IFDIR, stx.stx_ino, next_off); | |
8821 | ||
8822 | Inode *inode = NULL; | |
8823 | if (getref) { | |
8824 | inode = in.get(); | |
8825 | _ll_get(inode); | |
8826 | } | |
8827 | ||
f67539c2 | 8828 | cl.unlock(); |
7c673cae | 8829 | r = cb(p, &de, &stx, next_off, inode); |
f67539c2 | 8830 | cl.lock(); |
7c673cae FG |
8831 | if (r < 0) |
8832 | return r; | |
8833 | ||
8834 | dirp->offset = next_off; | |
8835 | if (r > 0) | |
8836 | return r; | |
8837 | } | |
8838 | ||
8839 | // can we read from our cache? | |
8840 | ldout(cct, 10) << "offset " << hex << dirp->offset << dec | |
8841 | << " snapid " << dirp->inode->snapid << " (complete && ordered) " | |
8842 | << dirp->inode->is_complete_and_ordered() | |
8843 | << " issued " << ccap_string(dirp->inode->caps_issued()) | |
8844 | << dendl; | |
8845 | if (dirp->inode->snapid != CEPH_SNAPDIR && | |
8846 | dirp->inode->is_complete_and_ordered() && | |
94b18763 | 8847 | dirp->inode->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 8848 | int err = _readdir_cache_cb(dirp, cb, p, caps, getref); |
f67539c2 | 8849 | if (err != -CEPHFS_EAGAIN) |
7c673cae FG |
8850 | return err; |
8851 | } | |
8852 | ||
8853 | while (1) { | |
8854 | if (dirp->at_end()) | |
8855 | return 0; | |
8856 | ||
8857 | bool check_caps = true; | |
8858 | if (!dirp->is_cached()) { | |
8859 | int r = _readdir_get_frag(dirp); | |
8860 | if (r) | |
8861 | return r; | |
8862 | // _readdir_get_frag () may updates dirp->offset if the replied dirfrag is | |
8863 | // different than the requested one. (our dirfragtree was outdated) | |
8864 | check_caps = false; | |
8865 | } | |
8866 | frag_t fg = dirp->buffer_frag; | |
8867 | ||
8868 | ldout(cct, 10) << "frag " << fg << " buffer size " << dirp->buffer.size() | |
8869 | << " offset " << hex << dirp->offset << dendl; | |
8870 | ||
8871 | for (auto it = std::lower_bound(dirp->buffer.begin(), dirp->buffer.end(), | |
8872 | dirp->offset, dir_result_t::dentry_off_lt()); | |
8873 | it != dirp->buffer.end(); | |
8874 | ++it) { | |
8875 | dir_result_t::dentry &entry = *it; | |
8876 | ||
8877 | uint64_t next_off = entry.offset + 1; | |
8878 | ||
8879 | int r; | |
8880 | if (check_caps) { | |
adb31ebb TL |
8881 | int mask = caps; |
8882 | if(entry.inode->is_dir()){ | |
8883 | mask |= CEPH_STAT_RSTAT; | |
8884 | } | |
8885 | r = _getattr(entry.inode, mask, dirp->perms); | |
7c673cae FG |
8886 | if (r < 0) |
8887 | return r; | |
8888 | } | |
8889 | ||
8890 | fill_statx(entry.inode, caps, &stx); | |
8891 | fill_dirent(&de, entry.name.c_str(), stx.stx_mode, stx.stx_ino, next_off); | |
8892 | ||
8893 | Inode *inode = NULL; | |
8894 | if (getref) { | |
8895 | inode = entry.inode.get(); | |
8896 | _ll_get(inode); | |
8897 | } | |
8898 | ||
f67539c2 | 8899 | cl.unlock(); |
7c673cae | 8900 | r = cb(p, &de, &stx, next_off, inode); // _next_ offset |
f67539c2 | 8901 | cl.lock(); |
7c673cae FG |
8902 | |
8903 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << next_off - 1 << dec | |
8904 | << " = " << r << dendl; | |
8905 | if (r < 0) | |
8906 | return r; | |
8907 | ||
8908 | dirp->offset = next_off; | |
8909 | if (r > 0) | |
8910 | return r; | |
8911 | } | |
8912 | ||
8913 | if (dirp->next_offset > 2) { | |
8914 | ldout(cct, 10) << " fetching next chunk of this frag" << dendl; | |
8915 | _readdir_drop_dirp_buffer(dirp); | |
8916 | continue; // more! | |
8917 | } | |
8918 | ||
8919 | if (!fg.is_rightmost()) { | |
8920 | // next frag! | |
8921 | _readdir_next_frag(dirp); | |
8922 | continue; | |
8923 | } | |
8924 | ||
8925 | if (diri->shared_gen == dirp->start_shared_gen && | |
8926 | diri->dir_release_count == dirp->release_count) { | |
8927 | if (diri->dir_ordered_count == dirp->ordered_count) { | |
8928 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
8929 | if (diri->dir) { | |
11fdf7f2 | 8930 | ceph_assert(diri->dir->readdir_cache.size() >= dirp->cache_index); |
7c673cae FG |
8931 | diri->dir->readdir_cache.resize(dirp->cache_index); |
8932 | } | |
8933 | diri->flags |= I_COMPLETE | I_DIR_ORDERED; | |
8934 | } else { | |
8935 | ldout(cct, 10) << " marking I_COMPLETE on " << *diri << dendl; | |
8936 | diri->flags |= I_COMPLETE; | |
8937 | } | |
8938 | } | |
8939 | ||
8940 | dirp->set_end(); | |
8941 | return 0; | |
8942 | } | |
8943 | ceph_abort(); | |
8944 | return 0; | |
8945 | } | |
8946 | ||
8947 | ||
8948 | int Client::readdir_r(dir_result_t *d, struct dirent *de) | |
8949 | { | |
8950 | return readdirplus_r(d, de, 0, 0, 0, NULL); | |
8951 | } | |
8952 | ||
8953 | /* | |
8954 | * readdirplus_r | |
8955 | * | |
8956 | * returns | |
8957 | * 1 if we got a dirent | |
8958 | * 0 for end of directory | |
8959 | * <0 on error | |
8960 | */ | |
8961 | ||
8962 | struct single_readdir { | |
8963 | struct dirent *de; | |
8964 | struct ceph_statx *stx; | |
8965 | Inode *inode; | |
8966 | bool full; | |
8967 | }; | |
8968 | ||
8969 | static int _readdir_single_dirent_cb(void *p, struct dirent *de, | |
8970 | struct ceph_statx *stx, off_t off, | |
8971 | Inode *in) | |
8972 | { | |
8973 | single_readdir *c = static_cast<single_readdir *>(p); | |
8974 | ||
8975 | if (c->full) | |
8976 | return -1; // already filled this dirent | |
8977 | ||
8978 | *c->de = *de; | |
8979 | if (c->stx) | |
8980 | *c->stx = *stx; | |
8981 | c->inode = in; | |
8982 | c->full = true; | |
8983 | return 1; | |
8984 | } | |
8985 | ||
8986 | struct dirent *Client::readdir(dir_result_t *d) | |
8987 | { | |
8988 | int ret; | |
f91f0fd5 | 8989 | auto& de = d->de; |
7c673cae FG |
8990 | single_readdir sr; |
8991 | sr.de = &de; | |
8992 | sr.stx = NULL; | |
8993 | sr.inode = NULL; | |
8994 | sr.full = false; | |
8995 | ||
8996 | // our callback fills the dirent and sets sr.full=true on first | |
8997 | // call, and returns -1 the second time around. | |
8998 | ret = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr); | |
8999 | if (ret < -1) { | |
9000 | errno = -ret; // this sucks. | |
9001 | return (dirent *) NULL; | |
9002 | } | |
9003 | if (sr.full) { | |
9004 | return &de; | |
9005 | } | |
9006 | return (dirent *) NULL; | |
9007 | } | |
9008 | ||
9009 | int Client::readdirplus_r(dir_result_t *d, struct dirent *de, | |
9010 | struct ceph_statx *stx, unsigned want, | |
9011 | unsigned flags, Inode **out) | |
9012 | { | |
9013 | single_readdir sr; | |
9014 | sr.de = de; | |
9015 | sr.stx = stx; | |
9016 | sr.inode = NULL; | |
9017 | sr.full = false; | |
9018 | ||
9019 | // our callback fills the dirent and sets sr.full=true on first | |
9020 | // call, and returns -1 the second time around. | |
9021 | int r = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr, want, flags, out); | |
9022 | if (r < -1) | |
9023 | return r; | |
9024 | if (out) | |
9025 | *out = sr.inode; | |
9026 | if (sr.full) | |
9027 | return 1; | |
9028 | return 0; | |
9029 | } | |
9030 | ||
9031 | ||
9032 | /* getdents */ | |
9033 | struct getdents_result { | |
9034 | char *buf; | |
9035 | int buflen; | |
9036 | int pos; | |
9037 | bool fullent; | |
9038 | }; | |
9039 | ||
9040 | static int _readdir_getdent_cb(void *p, struct dirent *de, | |
9041 | struct ceph_statx *stx, off_t off, Inode *in) | |
9042 | { | |
9043 | struct getdents_result *c = static_cast<getdents_result *>(p); | |
9044 | ||
9045 | int dlen; | |
9046 | if (c->fullent) | |
9047 | dlen = sizeof(*de); | |
9048 | else | |
9049 | dlen = strlen(de->d_name) + 1; | |
9050 | ||
9051 | if (c->pos + dlen > c->buflen) | |
9052 | return -1; // doesn't fit | |
9053 | ||
9054 | if (c->fullent) { | |
9055 | memcpy(c->buf + c->pos, de, sizeof(*de)); | |
9056 | } else { | |
9057 | memcpy(c->buf + c->pos, de->d_name, dlen); | |
9058 | } | |
9059 | c->pos += dlen; | |
9060 | return 0; | |
9061 | } | |
9062 | ||
9063 | int Client::_getdents(dir_result_t *dir, char *buf, int buflen, bool fullent) | |
9064 | { | |
9065 | getdents_result gr; | |
9066 | gr.buf = buf; | |
9067 | gr.buflen = buflen; | |
9068 | gr.fullent = fullent; | |
9069 | gr.pos = 0; | |
9070 | ||
9071 | int r = readdir_r_cb(dir, _readdir_getdent_cb, (void *)&gr); | |
9072 | ||
9073 | if (r < 0) { // some error | |
9074 | if (r == -1) { // buffer ran out of space | |
9075 | if (gr.pos) { // but we got some entries already! | |
9076 | return gr.pos; | |
9077 | } // or we need a larger buffer | |
f67539c2 | 9078 | return -CEPHFS_ERANGE; |
7c673cae FG |
9079 | } else { // actual error, return it |
9080 | return r; | |
9081 | } | |
9082 | } | |
9083 | return gr.pos; | |
9084 | } | |
9085 | ||
9086 | ||
9087 | /* getdir */ | |
9088 | struct getdir_result { | |
9089 | list<string> *contents; | |
9090 | int num; | |
9091 | }; | |
9092 | ||
9093 | static int _getdir_cb(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in) | |
9094 | { | |
9095 | getdir_result *r = static_cast<getdir_result *>(p); | |
9096 | ||
9097 | r->contents->push_back(de->d_name); | |
9098 | r->num++; | |
9099 | return 0; | |
9100 | } | |
9101 | ||
9102 | int Client::getdir(const char *relpath, list<string>& contents, | |
9103 | const UserPerm& perms) | |
9104 | { | |
9105 | ldout(cct, 3) << "getdir(" << relpath << ")" << dendl; | |
f67539c2 TL |
9106 | tout(cct) << "getdir" << std::endl; |
9107 | tout(cct) << relpath << std::endl; | |
7c673cae FG |
9108 | |
9109 | dir_result_t *d; | |
9110 | int r = opendir(relpath, &d, perms); | |
9111 | if (r < 0) | |
9112 | return r; | |
9113 | ||
9114 | getdir_result gr; | |
9115 | gr.contents = &contents; | |
9116 | gr.num = 0; | |
9117 | r = readdir_r_cb(d, _getdir_cb, (void *)&gr); | |
9118 | ||
9119 | closedir(d); | |
9120 | ||
9121 | if (r < 0) | |
9122 | return r; | |
9123 | return gr.num; | |
9124 | } | |
9125 | ||
9126 | ||
9127 | /****** file i/o **********/ | |
f67539c2 | 9128 | |
b3b6e05e TL |
9129 | // common parts for open and openat. call with client_lock locked. |
9130 | int Client::create_and_open(std::optional<int> dirfd, const char *relpath, int flags, | |
9131 | const UserPerm& perms, mode_t mode, int stripe_unit, | |
9132 | int stripe_count, int object_size, const char *data_pool, | |
9133 | std::string alternate_name) { | |
9134 | ceph_assert(ceph_mutex_is_locked(client_lock)); | |
f91f0fd5 | 9135 | int cflags = ceph_flags_sys2wire(flags); |
f91f0fd5 | 9136 | tout(cct) << cflags << std::endl; |
7c673cae FG |
9137 | |
9138 | Fh *fh = NULL; | |
9139 | ||
9140 | #if defined(__linux__) && defined(O_PATH) | |
9141 | /* When the O_PATH is being specified, others flags than O_DIRECTORY | |
9142 | * and O_NOFOLLOW are ignored. Please refer do_entry_open() function | |
9143 | * in kernel (fs/open.c). */ | |
9144 | if (flags & O_PATH) | |
9145 | flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; | |
9146 | #endif | |
9147 | ||
9148 | filepath path(relpath); | |
9149 | InodeRef in; | |
9150 | bool created = false; | |
9151 | /* O_CREATE with O_EXCL enforces O_NOFOLLOW. */ | |
9152 | bool followsym = !((flags & O_NOFOLLOW) || ((flags & O_CREAT) && (flags & O_EXCL))); | |
f91f0fd5 TL |
9153 | int mask = ceph_caps_for_mode(ceph_flags_to_mode(cflags)); |
9154 | ||
b3b6e05e TL |
9155 | InodeRef dirinode = nullptr; |
9156 | if (dirfd) { | |
9157 | int r = get_fd_inode(*dirfd, &dirinode); | |
9158 | if (r < 0) { | |
9159 | return r; | |
9160 | } | |
9161 | } | |
7c673cae | 9162 | |
b3b6e05e | 9163 | int r = path_walk(path, &in, perms, followsym, mask, dirinode); |
7c673cae | 9164 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) |
f67539c2 | 9165 | return -CEPHFS_EEXIST; |
7c673cae FG |
9166 | |
9167 | #if defined(__linux__) && defined(O_PATH) | |
9168 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW) && !(flags & O_PATH)) | |
9169 | #else | |
b3b6e05e | 9170 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW)) |
7c673cae | 9171 | #endif |
f67539c2 | 9172 | return -CEPHFS_ELOOP; |
7c673cae | 9173 | |
f67539c2 | 9174 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
7c673cae FG |
9175 | filepath dirpath = path; |
9176 | string dname = dirpath.last_dentry(); | |
9177 | dirpath.pop_dentry(); | |
9178 | InodeRef dir; | |
9179 | r = path_walk(dirpath, &dir, perms, true, | |
b3b6e05e TL |
9180 | cct->_conf->client_permissions ? CEPH_CAP_AUTH_SHARED : 0, dirinode); |
9181 | if (r < 0) { | |
7c673cae | 9182 | goto out; |
b3b6e05e | 9183 | } |
7c673cae FG |
9184 | if (cct->_conf->client_permissions) { |
9185 | r = may_create(dir.get(), perms); | |
9186 | if (r < 0) | |
b3b6e05e | 9187 | goto out; |
7c673cae FG |
9188 | } |
9189 | r = _create(dir.get(), dname.c_str(), flags, mode, &in, &fh, stripe_unit, | |
f67539c2 TL |
9190 | stripe_count, object_size, data_pool, &created, perms, |
9191 | std::move(alternate_name)); | |
7c673cae FG |
9192 | } |
9193 | if (r < 0) | |
9194 | goto out; | |
9195 | ||
9196 | if (!created) { | |
9197 | // posix says we can only check permissions of existing files | |
9198 | if (cct->_conf->client_permissions) { | |
9199 | r = may_open(in.get(), flags, perms); | |
9200 | if (r < 0) | |
b3b6e05e | 9201 | goto out; |
7c673cae FG |
9202 | } |
9203 | } | |
9204 | ||
9205 | if (!fh) | |
9206 | r = _open(in.get(), flags, mode, &fh, perms); | |
9207 | if (r >= 0) { | |
9208 | // allocate a integer file descriptor | |
11fdf7f2 | 9209 | ceph_assert(fh); |
7c673cae | 9210 | r = get_fd(); |
11fdf7f2 | 9211 | ceph_assert(fd_map.count(r) == 0); |
7c673cae FG |
9212 | fd_map[r] = fh; |
9213 | } | |
9214 | ||
9215 | out: | |
b3b6e05e TL |
9216 | return r; |
9217 | } | |
9218 | ||
9219 | int Client::open(const char *relpath, int flags, const UserPerm& perms, | |
9220 | mode_t mode, int stripe_unit, int stripe_count, | |
9221 | int object_size, const char *data_pool, std::string alternate_name) | |
9222 | { | |
9223 | return openat(CEPHFS_AT_FDCWD, relpath, flags, perms, mode, stripe_unit, | |
9224 | stripe_count, object_size, data_pool, alternate_name); | |
9225 | } | |
9226 | ||
9227 | int Client::_openat(int dirfd, const char *relpath, int flags, const UserPerm& perms, | |
9228 | mode_t mode, std::string alternate_name) { | |
9229 | return create_and_open(dirfd, relpath, flags, perms, mode, 0, 0, 0, NULL, alternate_name); | |
9230 | } | |
9231 | ||
9232 | int Client::openat(int dirfd, const char *relpath, int flags, const UserPerm& perms, | |
9233 | mode_t mode, int stripe_unit, int stripe_count, int object_size, | |
9234 | const char *data_pool, std::string alternate_name) { | |
9235 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9236 | if (!mref_reader.is_state_satisfied()) { | |
9237 | return -CEPHFS_ENOTCONN; | |
9238 | } | |
9239 | ||
9240 | ldout(cct, 3) << "openat enter(" << relpath << ")" << dendl; | |
9241 | tout(cct) << dirfd << std::endl; | |
9242 | tout(cct) << relpath << std::endl; | |
9243 | tout(cct) << flags << std::endl; | |
9244 | tout(cct) << mode << std::endl; | |
9245 | ||
9246 | std::scoped_lock locker(client_lock); | |
9247 | int r = create_and_open(dirfd, relpath, flags, perms, mode, stripe_unit, stripe_count, | |
9248 | object_size, data_pool, alternate_name); | |
9249 | ||
7c673cae | 9250 | tout(cct) << r << std::endl; |
b3b6e05e | 9251 | ldout(cct, 3) << "openat exit(" << relpath << ")" << dendl; |
7c673cae FG |
9252 | return r; |
9253 | } | |
9254 | ||
7c673cae FG |
9255 | int Client::lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name, |
9256 | const UserPerm& perms) | |
9257 | { | |
11fdf7f2 | 9258 | ldout(cct, 3) << __func__ << " enter(" << ino << ", #" << dirino << "/" << name << ")" << dendl; |
7c673cae | 9259 | |
f67539c2 TL |
9260 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9261 | if (!mref_reader.is_state_satisfied()) | |
9262 | return -CEPHFS_ENOTCONN; | |
181888fb | 9263 | |
f67539c2 | 9264 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9265 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPHASH); |
9266 | filepath path(ino); | |
9267 | req->set_filepath(path); | |
9268 | ||
9269 | uint32_t h = ceph_str_hash(CEPH_STR_HASH_RJENKINS, name, strlen(name)); | |
9270 | char f[30]; | |
9271 | sprintf(f, "%u", h); | |
9272 | filepath path2(dirino); | |
9273 | path2.push_dentry(string(f)); | |
9274 | req->set_filepath2(path2); | |
9275 | ||
9276 | int r = make_request(req, perms, NULL, NULL, | |
9277 | rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9278 | ldout(cct, 3) << __func__ << " exit(" << ino << ", #" << dirino << "/" << name << ") = " << r << dendl; |
7c673cae FG |
9279 | return r; |
9280 | } | |
9281 | ||
9282 | ||
9283 | /** | |
9284 | * Load inode into local cache. | |
9285 | * | |
9286 | * If inode pointer is non-NULL, and take a reference on | |
9287 | * the resulting Inode object in one operation, so that caller | |
9288 | * can safely assume inode will still be there after return. | |
9289 | */ | |
f67539c2 | 9290 | int Client::_lookup_vino(vinodeno_t vino, const UserPerm& perms, Inode **inode) |
7c673cae | 9291 | { |
f67539c2 | 9292 | ldout(cct, 8) << __func__ << " enter(" << vino << ")" << dendl; |
7c673cae | 9293 | |
f67539c2 TL |
9294 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9295 | if (!mref_reader.is_state_satisfied()) | |
9296 | return -CEPHFS_ENOTCONN; | |
181888fb | 9297 | |
b3b6e05e TL |
9298 | if (is_reserved_vino(vino)) |
9299 | return -CEPHFS_ESTALE; | |
9300 | ||
7c673cae | 9301 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPINO); |
f67539c2 | 9302 | filepath path(vino.ino); |
7c673cae FG |
9303 | req->set_filepath(path); |
9304 | ||
f67539c2 TL |
9305 | /* |
9306 | * The MDS expects either a "real" snapid here or 0. The special value | |
9307 | * carveouts for the snapid are all at the end of the range so we can | |
9308 | * just look for any snapid below this value. | |
9309 | */ | |
9310 | if (vino.snapid < CEPH_NOSNAP) | |
9311 | req->head.args.lookupino.snapid = vino.snapid; | |
9312 | ||
7c673cae FG |
9313 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); |
9314 | if (r == 0 && inode != NULL) { | |
7c673cae | 9315 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
11fdf7f2 | 9316 | ceph_assert(p != inode_map.end()); |
7c673cae FG |
9317 | *inode = p->second; |
9318 | _ll_get(*inode); | |
9319 | } | |
f67539c2 | 9320 | ldout(cct, 8) << __func__ << " exit(" << vino << ") = " << r << dendl; |
7c673cae FG |
9321 | return r; |
9322 | } | |
9323 | ||
1adf2230 AA |
9324 | int Client::lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode) |
9325 | { | |
f67539c2 TL |
9326 | vinodeno_t vino(ino, CEPH_NOSNAP); |
9327 | std::scoped_lock lock(client_lock); | |
9328 | return _lookup_vino(vino, perms, inode); | |
1adf2230 | 9329 | } |
7c673cae FG |
9330 | |
9331 | /** | |
9332 | * Find the parent inode of `ino` and insert it into | |
9333 | * our cache. Conditionally also set `parent` to a referenced | |
9334 | * Inode* if caller provides non-NULL value. | |
9335 | */ | |
1adf2230 | 9336 | int Client::_lookup_parent(Inode *ino, const UserPerm& perms, Inode **parent) |
7c673cae | 9337 | { |
11fdf7f2 | 9338 | ldout(cct, 8) << __func__ << " enter(" << ino->ino << ")" << dendl; |
7c673cae | 9339 | |
7c673cae FG |
9340 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); |
9341 | filepath path(ino->ino); | |
9342 | req->set_filepath(path); | |
9343 | ||
9344 | InodeRef target; | |
9345 | int r = make_request(req, perms, &target, NULL, rand() % mdsmap->get_num_in_mds()); | |
9346 | // Give caller a reference to the parent ino if they provided a pointer. | |
9347 | if (parent != NULL) { | |
9348 | if (r == 0) { | |
9349 | *parent = target.get(); | |
9350 | _ll_get(*parent); | |
11fdf7f2 | 9351 | ldout(cct, 8) << __func__ << " found parent " << (*parent)->ino << dendl; |
7c673cae FG |
9352 | } else { |
9353 | *parent = NULL; | |
9354 | } | |
9355 | } | |
11fdf7f2 | 9356 | ldout(cct, 8) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9357 | return r; |
9358 | } | |
9359 | ||
7c673cae FG |
9360 | /** |
9361 | * Populate the parent dentry for `ino`, provided it is | |
9362 | * a child of `parent`. | |
9363 | */ | |
1adf2230 | 9364 | int Client::_lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
7c673cae | 9365 | { |
11fdf7f2 TL |
9366 | ceph_assert(parent->is_dir()); |
9367 | ldout(cct, 3) << __func__ << " enter(" << ino->ino << ")" << dendl; | |
7c673cae | 9368 | |
f67539c2 TL |
9369 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9370 | if (!mref_reader.is_state_satisfied()) | |
9371 | return -CEPHFS_ENOTCONN; | |
181888fb | 9372 | |
7c673cae FG |
9373 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
9374 | req->set_filepath2(filepath(parent->ino)); | |
9375 | req->set_filepath(filepath(ino->ino)); | |
9376 | req->set_inode(ino); | |
9377 | ||
9378 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 9379 | ldout(cct, 3) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
9380 | return r; |
9381 | } | |
9382 | ||
1adf2230 AA |
9383 | int Client::lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
9384 | { | |
f67539c2 | 9385 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
9386 | return _lookup_name(ino, parent, perms); |
9387 | } | |
7c673cae | 9388 | |
11fdf7f2 | 9389 | Fh *Client::_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms) |
7c673cae | 9390 | { |
11fdf7f2 | 9391 | ceph_assert(in); |
f6b5b4d7 | 9392 | Fh *f = new Fh(in, flags, cmode, fd_gen, perms); |
7c673cae | 9393 | |
11fdf7f2 | 9394 | ldout(cct, 10) << __func__ << " " << in->ino << " mode " << cmode << dendl; |
7c673cae FG |
9395 | |
9396 | if (in->snapid != CEPH_NOSNAP) { | |
9397 | in->snap_cap_refs++; | |
9398 | ldout(cct, 5) << "open success, fh is " << f << " combined IMMUTABLE SNAP caps " | |
9399 | << ccap_string(in->caps_issued()) << dendl; | |
9400 | } | |
9401 | ||
11fdf7f2 | 9402 | const auto& conf = cct->_conf; |
7c673cae FG |
9403 | f->readahead.set_trigger_requests(1); |
9404 | f->readahead.set_min_readahead_size(conf->client_readahead_min); | |
9405 | uint64_t max_readahead = Readahead::NO_LIMIT; | |
9406 | if (conf->client_readahead_max_bytes) { | |
11fdf7f2 | 9407 | max_readahead = std::min(max_readahead, (uint64_t)conf->client_readahead_max_bytes); |
7c673cae FG |
9408 | } |
9409 | if (conf->client_readahead_max_periods) { | |
11fdf7f2 | 9410 | max_readahead = std::min(max_readahead, in->layout.get_period()*(uint64_t)conf->client_readahead_max_periods); |
7c673cae FG |
9411 | } |
9412 | f->readahead.set_max_readahead_size(max_readahead); | |
9413 | vector<uint64_t> alignments; | |
9414 | alignments.push_back(in->layout.get_period()); | |
9415 | alignments.push_back(in->layout.stripe_unit); | |
9416 | f->readahead.set_alignments(alignments); | |
9417 | ||
9418 | return f; | |
9419 | } | |
9420 | ||
9421 | int Client::_release_fh(Fh *f) | |
9422 | { | |
9423 | //ldout(cct, 3) << "op: client->close(open_files[ " << fh << " ]);" << dendl; | |
9424 | //ldout(cct, 3) << "op: open_files.erase( " << fh << " );" << dendl; | |
9425 | Inode *in = f->inode.get(); | |
11fdf7f2 | 9426 | ldout(cct, 8) << __func__ << " " << f << " mode " << f->mode << " on " << *in << dendl; |
7c673cae | 9427 | |
b32b8144 FG |
9428 | in->unset_deleg(f); |
9429 | ||
7c673cae FG |
9430 | if (in->snapid == CEPH_NOSNAP) { |
9431 | if (in->put_open_ref(f->mode)) { | |
9432 | _flush(in, new C_Client_FlushComplete(this, in)); | |
9433 | check_caps(in, 0); | |
9434 | } | |
9435 | } else { | |
11fdf7f2 | 9436 | ceph_assert(in->snap_cap_refs > 0); |
7c673cae FG |
9437 | in->snap_cap_refs--; |
9438 | } | |
9439 | ||
9440 | _release_filelocks(f); | |
9441 | ||
9442 | // Finally, read any async err (i.e. from flushes) | |
9443 | int err = f->take_async_err(); | |
9444 | if (err != 0) { | |
11fdf7f2 | 9445 | ldout(cct, 1) << __func__ << " " << f << " on inode " << *in << " caught async_err = " |
7c673cae FG |
9446 | << cpp_strerror(err) << dendl; |
9447 | } else { | |
11fdf7f2 | 9448 | ldout(cct, 10) << __func__ << " " << f << " on inode " << *in << " no async_err state" << dendl; |
7c673cae FG |
9449 | } |
9450 | ||
9451 | _put_fh(f); | |
9452 | ||
9453 | return err; | |
9454 | } | |
9455 | ||
9456 | void Client::_put_fh(Fh *f) | |
9457 | { | |
9458 | int left = f->put(); | |
9459 | if (!left) { | |
9460 | delete f; | |
9461 | } | |
9462 | } | |
9463 | ||
9464 | int Client::_open(Inode *in, int flags, mode_t mode, Fh **fhp, | |
9465 | const UserPerm& perms) | |
9466 | { | |
9467 | if (in->snapid != CEPH_NOSNAP && | |
9468 | (flags & (O_WRONLY | O_RDWR | O_CREAT | O_TRUNC | O_APPEND))) { | |
f67539c2 | 9469 | return -CEPHFS_EROFS; |
7c673cae FG |
9470 | } |
9471 | ||
9472 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
9473 | int cflags = ceph_flags_sys2wire(flags); |
9474 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
9475 | cflags |= CEPH_O_LAZY; | |
9476 | ||
9477 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
9478 | int want = ceph_caps_for_mode(cmode); |
9479 | int result = 0; | |
9480 | ||
9481 | in->get_open_ref(cmode); // make note of pending open, since it effects _wanted_ caps. | |
9482 | ||
b32b8144 | 9483 | if ((flags & O_TRUNC) == 0 && in->caps_issued_mask(want)) { |
7c673cae FG |
9484 | // update wanted? |
9485 | check_caps(in, CHECK_CAPS_NODELAY); | |
9486 | } else { | |
b32b8144 | 9487 | |
7c673cae FG |
9488 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); |
9489 | filepath path; | |
9490 | in->make_nosnap_relative_path(path); | |
9491 | req->set_filepath(path); | |
11fdf7f2 | 9492 | req->head.args.open.flags = cflags & ~CEPH_O_CREAT; |
7c673cae FG |
9493 | req->head.args.open.mode = mode; |
9494 | req->head.args.open.pool = -1; | |
9495 | if (cct->_conf->client_debug_getattr_caps) | |
9496 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9497 | else | |
9498 | req->head.args.open.mask = 0; | |
9499 | req->head.args.open.old_size = in->size; // for O_TRUNC | |
9500 | req->set_inode(in); | |
9501 | result = make_request(req, perms); | |
b32b8144 FG |
9502 | |
9503 | /* | |
9504 | * NFS expects that delegations will be broken on a conflicting open, | |
9505 | * not just when there is actual conflicting access to the file. SMB leases | |
9506 | * and oplocks also have similar semantics. | |
9507 | * | |
9508 | * Ensure that clients that have delegations enabled will wait on minimal | |
9509 | * caps during open, just to ensure that other clients holding delegations | |
9510 | * return theirs first. | |
9511 | */ | |
9512 | if (deleg_timeout && result == 0) { | |
9513 | int need = 0, have; | |
9514 | ||
9515 | if (cmode & CEPH_FILE_MODE_WR) | |
9516 | need |= CEPH_CAP_FILE_WR; | |
9517 | if (cmode & CEPH_FILE_MODE_RD) | |
9518 | need |= CEPH_CAP_FILE_RD; | |
9519 | ||
f6b5b4d7 TL |
9520 | Fh fh(in, flags, cmode, fd_gen, perms); |
9521 | result = get_caps(&fh, need, want, &have, -1); | |
b32b8144 | 9522 | if (result < 0) { |
1adf2230 | 9523 | ldout(cct, 8) << "Unable to get caps after open of inode " << *in << |
b32b8144 FG |
9524 | " . Denying open: " << |
9525 | cpp_strerror(result) << dendl; | |
b32b8144 FG |
9526 | } else { |
9527 | put_cap_ref(in, need); | |
9528 | } | |
9529 | } | |
7c673cae FG |
9530 | } |
9531 | ||
9532 | // success? | |
9533 | if (result >= 0) { | |
9534 | if (fhp) | |
9535 | *fhp = _create_fh(in, flags, cmode, perms); | |
9536 | } else { | |
9537 | in->put_open_ref(cmode); | |
9538 | } | |
9539 | ||
9540 | trim_cache(); | |
9541 | ||
9542 | return result; | |
9543 | } | |
9544 | ||
9545 | int Client::_renew_caps(Inode *in) | |
9546 | { | |
9547 | int wanted = in->caps_file_wanted(); | |
9548 | if (in->is_any_caps() && | |
9549 | ((wanted & CEPH_CAP_ANY_WR) == 0 || in->auth_cap)) { | |
9550 | check_caps(in, CHECK_CAPS_NODELAY); | |
9551 | return 0; | |
9552 | } | |
9553 | ||
9554 | int flags = 0; | |
9555 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
9556 | flags = O_RDWR; | |
9557 | else if (wanted & CEPH_CAP_FILE_RD) | |
9558 | flags = O_RDONLY; | |
9559 | else if (wanted & CEPH_CAP_FILE_WR) | |
9560 | flags = O_WRONLY; | |
9561 | ||
9562 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); | |
9563 | filepath path; | |
9564 | in->make_nosnap_relative_path(path); | |
9565 | req->set_filepath(path); | |
9566 | req->head.args.open.flags = flags; | |
9567 | req->head.args.open.pool = -1; | |
9568 | if (cct->_conf->client_debug_getattr_caps) | |
9569 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
9570 | else | |
9571 | req->head.args.open.mask = 0; | |
9572 | req->set_inode(in); | |
9573 | ||
9574 | // duplicate in case Cap goes away; not sure if that race is a concern? | |
9575 | const UserPerm *pperm = in->get_best_perms(); | |
9576 | UserPerm perms; | |
9577 | if (pperm != NULL) | |
9578 | perms = *pperm; | |
9579 | int ret = make_request(req, perms); | |
9580 | return ret; | |
9581 | } | |
9582 | ||
b3b6e05e | 9583 | int Client::_close(int fd) |
7c673cae FG |
9584 | { |
9585 | ldout(cct, 3) << "close enter(" << fd << ")" << dendl; | |
7c673cae FG |
9586 | tout(cct) << "close" << std::endl; |
9587 | tout(cct) << fd << std::endl; | |
9588 | ||
9589 | Fh *fh = get_filehandle(fd); | |
9590 | if (!fh) | |
f67539c2 | 9591 | return -CEPHFS_EBADF; |
7c673cae FG |
9592 | int err = _release_fh(fh); |
9593 | fd_map.erase(fd); | |
9594 | put_fd(fd); | |
9595 | ldout(cct, 3) << "close exit(" << fd << ")" << dendl; | |
9596 | return err; | |
9597 | } | |
9598 | ||
b3b6e05e TL |
9599 | int Client::close(int fd) { |
9600 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
9601 | if (!mref_reader.is_state_satisfied()) | |
9602 | return -CEPHFS_ENOTCONN; | |
9603 | ||
9604 | std::scoped_lock lock(client_lock); | |
9605 | return _close(fd); | |
9606 | } | |
7c673cae FG |
9607 | |
9608 | // ------------ | |
9609 | // read, write | |
9610 | ||
9611 | loff_t Client::lseek(int fd, loff_t offset, int whence) | |
9612 | { | |
f67539c2 TL |
9613 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9614 | if (!mref_reader.is_state_satisfied()) | |
9615 | return -CEPHFS_ENOTCONN; | |
9616 | ||
7c673cae FG |
9617 | tout(cct) << "lseek" << std::endl; |
9618 | tout(cct) << fd << std::endl; | |
9619 | tout(cct) << offset << std::endl; | |
9620 | tout(cct) << whence << std::endl; | |
9621 | ||
f67539c2 | 9622 | std::scoped_lock lock(client_lock); |
7c673cae FG |
9623 | Fh *f = get_filehandle(fd); |
9624 | if (!f) | |
f67539c2 | 9625 | return -CEPHFS_EBADF; |
7c673cae FG |
9626 | #if defined(__linux__) && defined(O_PATH) |
9627 | if (f->flags & O_PATH) | |
f67539c2 | 9628 | return -CEPHFS_EBADF; |
7c673cae FG |
9629 | #endif |
9630 | return _lseek(f, offset, whence); | |
9631 | } | |
9632 | ||
9633 | loff_t Client::_lseek(Fh *f, loff_t offset, int whence) | |
9634 | { | |
9635 | Inode *in = f->inode.get(); | |
9f95a23c | 9636 | bool whence_check = false; |
11fdf7f2 | 9637 | loff_t pos = -1; |
7c673cae | 9638 | |
9f95a23c TL |
9639 | switch (whence) { |
9640 | case SEEK_END: | |
9641 | whence_check = true; | |
9642 | break; | |
9643 | ||
9644 | #ifdef SEEK_DATA | |
9645 | case SEEK_DATA: | |
9646 | whence_check = true; | |
9647 | break; | |
9648 | #endif | |
9649 | ||
9650 | #ifdef SEEK_HOLE | |
9651 | case SEEK_HOLE: | |
9652 | whence_check = true; | |
9653 | break; | |
9654 | #endif | |
9655 | } | |
9656 | ||
9657 | if (whence_check) { | |
9658 | int r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
9659 | if (r < 0) | |
92f5a8d4 | 9660 | return r; |
92f5a8d4 TL |
9661 | } |
9662 | ||
7c673cae FG |
9663 | switch (whence) { |
9664 | case SEEK_SET: | |
11fdf7f2 | 9665 | pos = offset; |
7c673cae FG |
9666 | break; |
9667 | ||
9668 | case SEEK_CUR: | |
92f5a8d4 | 9669 | pos = f->pos + offset; |
7c673cae FG |
9670 | break; |
9671 | ||
9672 | case SEEK_END: | |
11fdf7f2 | 9673 | pos = in->size + offset; |
7c673cae FG |
9674 | break; |
9675 | ||
9f95a23c | 9676 | #ifdef SEEK_DATA |
92f5a8d4 | 9677 | case SEEK_DATA: |
9f95a23c | 9678 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9679 | return -CEPHFS_ENXIO; |
92f5a8d4 TL |
9680 | pos = offset; |
9681 | break; | |
9f95a23c | 9682 | #endif |
92f5a8d4 | 9683 | |
9f95a23c | 9684 | #ifdef SEEK_HOLE |
92f5a8d4 | 9685 | case SEEK_HOLE: |
9f95a23c | 9686 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
f67539c2 | 9687 | return -CEPHFS_ENXIO; |
9f95a23c | 9688 | pos = in->size; |
92f5a8d4 | 9689 | break; |
9f95a23c | 9690 | #endif |
92f5a8d4 | 9691 | |
7c673cae | 9692 | default: |
92f5a8d4 | 9693 | ldout(cct, 1) << __func__ << ": invalid whence value " << whence << dendl; |
f67539c2 | 9694 | return -CEPHFS_EINVAL; |
7c673cae FG |
9695 | } |
9696 | ||
11fdf7f2 | 9697 | if (pos < 0) { |
f67539c2 | 9698 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
9699 | } else { |
9700 | f->pos = pos; | |
9701 | } | |
9702 | ||
1adf2230 | 9703 | ldout(cct, 8) << "_lseek(" << f << ", " << offset << ", " << whence << ") = " << f->pos << dendl; |
7c673cae FG |
9704 | return f->pos; |
9705 | } | |
9706 | ||
9707 | ||
9708 | void Client::lock_fh_pos(Fh *f) | |
9709 | { | |
11fdf7f2 | 9710 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae FG |
9711 | |
9712 | if (f->pos_locked || !f->pos_waiters.empty()) { | |
9f95a23c | 9713 | ceph::condition_variable cond; |
7c673cae | 9714 | f->pos_waiters.push_back(&cond); |
11fdf7f2 | 9715 | ldout(cct, 10) << __func__ << " BLOCKING on " << f << dendl; |
9f95a23c TL |
9716 | std::unique_lock l{client_lock, std::adopt_lock}; |
9717 | cond.wait(l, [f, me=&cond] { | |
9718 | return !f->pos_locked && f->pos_waiters.front() == me; | |
9719 | }); | |
9720 | l.release(); | |
11fdf7f2 TL |
9721 | ldout(cct, 10) << __func__ << " UNBLOCKING on " << f << dendl; |
9722 | ceph_assert(f->pos_waiters.front() == &cond); | |
7c673cae FG |
9723 | f->pos_waiters.pop_front(); |
9724 | } | |
9725 | ||
9726 | f->pos_locked = true; | |
9727 | } | |
9728 | ||
9729 | void Client::unlock_fh_pos(Fh *f) | |
9730 | { | |
f67539c2 TL |
9731 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9732 | ||
11fdf7f2 | 9733 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae | 9734 | f->pos_locked = false; |
f67539c2 TL |
9735 | if (!f->pos_waiters.empty()) { |
9736 | // only wake up the oldest waiter | |
9737 | auto cond = f->pos_waiters.front(); | |
9738 | cond->notify_one(); | |
9739 | } | |
7c673cae FG |
9740 | } |
9741 | ||
9742 | int Client::uninline_data(Inode *in, Context *onfinish) | |
9743 | { | |
9744 | if (!in->inline_data.length()) { | |
9745 | onfinish->complete(0); | |
9746 | return 0; | |
9747 | } | |
9748 | ||
9749 | char oid_buf[32]; | |
9750 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (long long unsigned)in->ino); | |
9751 | object_t oid = oid_buf; | |
9752 | ||
9753 | ObjectOperation create_ops; | |
9754 | create_ops.create(false); | |
9755 | ||
9756 | objecter->mutate(oid, | |
9757 | OSDMap::file_to_object_locator(in->layout), | |
9758 | create_ops, | |
9759 | in->snaprealm->get_snap_context(), | |
9760 | ceph::real_clock::now(), | |
9761 | 0, | |
9762 | NULL); | |
9763 | ||
9764 | bufferlist inline_version_bl; | |
11fdf7f2 | 9765 | encode(in->inline_version, inline_version_bl); |
7c673cae FG |
9766 | |
9767 | ObjectOperation uninline_ops; | |
9768 | uninline_ops.cmpxattr("inline_version", | |
9769 | CEPH_OSD_CMPXATTR_OP_GT, | |
9770 | CEPH_OSD_CMPXATTR_MODE_U64, | |
9771 | inline_version_bl); | |
9772 | bufferlist inline_data = in->inline_data; | |
9773 | uninline_ops.write(0, inline_data, in->truncate_size, in->truncate_seq); | |
9774 | uninline_ops.setxattr("inline_version", stringify(in->inline_version)); | |
9775 | ||
9776 | objecter->mutate(oid, | |
9777 | OSDMap::file_to_object_locator(in->layout), | |
9778 | uninline_ops, | |
9779 | in->snaprealm->get_snap_context(), | |
9780 | ceph::real_clock::now(), | |
9781 | 0, | |
9782 | onfinish); | |
9783 | ||
9784 | return 0; | |
9785 | } | |
9786 | ||
9787 | // | |
9788 | ||
9789 | // blocking osd interface | |
9790 | ||
9791 | int Client::read(int fd, char *buf, loff_t size, loff_t offset) | |
9792 | { | |
f67539c2 TL |
9793 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
9794 | if (!mref_reader.is_state_satisfied()) | |
9795 | return -CEPHFS_ENOTCONN; | |
9796 | ||
7c673cae FG |
9797 | tout(cct) << "read" << std::endl; |
9798 | tout(cct) << fd << std::endl; | |
9799 | tout(cct) << size << std::endl; | |
9800 | tout(cct) << offset << std::endl; | |
9801 | ||
f67539c2 | 9802 | std::unique_lock lock(client_lock); |
7c673cae FG |
9803 | Fh *f = get_filehandle(fd); |
9804 | if (!f) | |
f67539c2 | 9805 | return -CEPHFS_EBADF; |
7c673cae FG |
9806 | #if defined(__linux__) && defined(O_PATH) |
9807 | if (f->flags & O_PATH) | |
f67539c2 | 9808 | return -CEPHFS_EBADF; |
7c673cae FG |
9809 | #endif |
9810 | bufferlist bl; | |
11fdf7f2 TL |
9811 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
9812 | size = std::min(size, (loff_t)INT_MAX); | |
7c673cae FG |
9813 | int r = _read(f, offset, size, &bl); |
9814 | ldout(cct, 3) << "read(" << fd << ", " << (void*)buf << ", " << size << ", " << offset << ") = " << r << dendl; | |
9815 | if (r >= 0) { | |
f6b5b4d7 | 9816 | lock.unlock(); |
9f95a23c | 9817 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
9818 | r = bl.length(); |
9819 | } | |
9820 | return r; | |
9821 | } | |
9822 | ||
9823 | int Client::preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset) | |
9824 | { | |
9825 | if (iovcnt < 0) | |
f67539c2 | 9826 | return -CEPHFS_EINVAL; |
7c673cae FG |
9827 | return _preadv_pwritev(fd, iov, iovcnt, offset, false); |
9828 | } | |
9829 | ||
11fdf7f2 | 9830 | int64_t Client::_read(Fh *f, int64_t offset, uint64_t size, bufferlist *bl) |
7c673cae | 9831 | { |
f67539c2 TL |
9832 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9833 | ||
11fdf7f2 TL |
9834 | int want, have = 0; |
9835 | bool movepos = false; | |
9836 | std::unique_ptr<C_SaferCond> onuninline; | |
adb31ebb | 9837 | int64_t rc = 0; |
11fdf7f2 | 9838 | const auto& conf = cct->_conf; |
7c673cae | 9839 | Inode *in = f->inode.get(); |
11fdf7f2 TL |
9840 | utime_t lat; |
9841 | utime_t start = ceph_clock_now(); | |
7c673cae FG |
9842 | |
9843 | if ((f->mode & CEPH_FILE_MODE_RD) == 0) | |
f67539c2 | 9844 | return -CEPHFS_EBADF; |
7c673cae FG |
9845 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
9846 | ||
7c673cae FG |
9847 | if (offset < 0) { |
9848 | lock_fh_pos(f); | |
9849 | offset = f->pos; | |
9850 | movepos = true; | |
9851 | } | |
9852 | loff_t start_pos = offset; | |
9853 | ||
9854 | if (in->inline_version == 0) { | |
adb31ebb | 9855 | auto r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); |
c07f9fc5 | 9856 | if (r < 0) { |
adb31ebb | 9857 | rc = r; |
11fdf7f2 | 9858 | goto done; |
c07f9fc5 | 9859 | } |
11fdf7f2 | 9860 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
9861 | } |
9862 | ||
9863 | retry: | |
11fdf7f2 TL |
9864 | if (f->mode & CEPH_FILE_MODE_LAZY) |
9865 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
9866 | else | |
9867 | want = CEPH_CAP_FILE_CACHE; | |
adb31ebb TL |
9868 | { |
9869 | auto r = get_caps(f, CEPH_CAP_FILE_RD, want, &have, -1); | |
9870 | if (r < 0) { | |
9871 | rc = r; | |
9872 | goto done; | |
9873 | } | |
c07f9fc5 | 9874 | } |
7c673cae | 9875 | if (f->flags & O_DIRECT) |
11fdf7f2 | 9876 | have &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
9877 | |
9878 | if (in->inline_version < CEPH_INLINE_NONE) { | |
9879 | if (!(have & CEPH_CAP_FILE_CACHE)) { | |
11fdf7f2 TL |
9880 | onuninline.reset(new C_SaferCond("Client::_read_uninline_data flock")); |
9881 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
9882 | } else { |
9883 | uint32_t len = in->inline_data.length(); | |
7c673cae FG |
9884 | uint64_t endoff = offset + size; |
9885 | if (endoff > in->size) | |
9886 | endoff = in->size; | |
9887 | ||
9888 | if (offset < len) { | |
9889 | if (endoff <= len) { | |
9890 | bl->substr_of(in->inline_data, offset, endoff - offset); | |
9891 | } else { | |
9892 | bl->substr_of(in->inline_data, offset, len - offset); | |
9893 | bl->append_zero(endoff - len); | |
9894 | } | |
adb31ebb | 9895 | rc = endoff - offset; |
7c673cae FG |
9896 | } else if ((uint64_t)offset < endoff) { |
9897 | bl->append_zero(endoff - offset); | |
adb31ebb | 9898 | rc = endoff - offset; |
11fdf7f2 | 9899 | } else { |
adb31ebb | 9900 | rc = 0; |
7c673cae | 9901 | } |
7c673cae FG |
9902 | goto success; |
9903 | } | |
9904 | } | |
9905 | ||
9906 | if (!conf->client_debug_force_sync_read && | |
11fdf7f2 TL |
9907 | conf->client_oc && |
9908 | (have & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
9909 | |
9910 | if (f->flags & O_RSYNC) { | |
9911 | _flush_range(in, offset, size); | |
9912 | } | |
adb31ebb TL |
9913 | rc = _read_async(f, offset, size, bl); |
9914 | if (rc < 0) | |
7c673cae FG |
9915 | goto done; |
9916 | } else { | |
9917 | if (f->flags & O_DIRECT) | |
9918 | _flush_range(in, offset, size); | |
9919 | ||
9920 | bool checkeof = false; | |
adb31ebb TL |
9921 | rc = _read_sync(f, offset, size, bl, &checkeof); |
9922 | if (rc < 0) | |
7c673cae FG |
9923 | goto done; |
9924 | if (checkeof) { | |
adb31ebb TL |
9925 | offset += rc; |
9926 | size -= rc; | |
7c673cae FG |
9927 | |
9928 | put_cap_ref(in, CEPH_CAP_FILE_RD); | |
9929 | have = 0; | |
9930 | // reverify size | |
adb31ebb TL |
9931 | { |
9932 | auto r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
9933 | if (r < 0) { | |
9934 | rc = r; | |
9935 | goto done; | |
9936 | } | |
9937 | } | |
7c673cae FG |
9938 | |
9939 | // eof? short read. | |
9940 | if ((uint64_t)offset < in->size) | |
9941 | goto retry; | |
9942 | } | |
9943 | } | |
9944 | ||
9945 | success: | |
adb31ebb | 9946 | ceph_assert(rc >= 0); |
7c673cae FG |
9947 | if (movepos) { |
9948 | // adjust fd pos | |
adb31ebb | 9949 | f->pos = start_pos + rc; |
7c673cae | 9950 | } |
11fdf7f2 TL |
9951 | |
9952 | lat = ceph_clock_now(); | |
9953 | lat -= start; | |
9954 | logger->tinc(l_c_read, lat); | |
7c673cae FG |
9955 | |
9956 | done: | |
9957 | // done! | |
11fdf7f2 | 9958 | |
7c673cae | 9959 | if (onuninline) { |
9f95a23c | 9960 | client_lock.unlock(); |
11fdf7f2 | 9961 | int ret = onuninline->wait(); |
9f95a23c | 9962 | client_lock.lock(); |
f67539c2 | 9963 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
9964 | in->inline_data.clear(); |
9965 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 9966 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
9967 | check_caps(in, 0); |
9968 | } else | |
adb31ebb | 9969 | rc = ret; |
7c673cae | 9970 | } |
11fdf7f2 | 9971 | if (have) { |
7c673cae | 9972 | put_cap_ref(in, CEPH_CAP_FILE_RD); |
11fdf7f2 TL |
9973 | } |
9974 | if (movepos) { | |
9975 | unlock_fh_pos(f); | |
9976 | } | |
adb31ebb | 9977 | return rc; |
7c673cae FG |
9978 | } |
9979 | ||
9980 | Client::C_Readahead::C_Readahead(Client *c, Fh *f) : | |
9981 | client(c), f(f) { | |
9982 | f->get(); | |
9983 | f->readahead.inc_pending(); | |
9984 | } | |
9985 | ||
9986 | Client::C_Readahead::~C_Readahead() { | |
9987 | f->readahead.dec_pending(); | |
9988 | client->_put_fh(f); | |
9989 | } | |
9990 | ||
9991 | void Client::C_Readahead::finish(int r) { | |
9992 | lgeneric_subdout(client->cct, client, 20) << "client." << client->get_nodeid() << " " << "C_Readahead on " << f->inode << dendl; | |
9993 | client->put_cap_ref(f->inode.get(), CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
9994 | } | |
9995 | ||
9996 | int Client::_read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl) | |
9997 | { | |
f67539c2 TL |
9998 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
9999 | ||
11fdf7f2 | 10000 | const auto& conf = cct->_conf; |
7c673cae FG |
10001 | Inode *in = f->inode.get(); |
10002 | ||
11fdf7f2 | 10003 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
10004 | |
10005 | // trim read based on file size? | |
10006 | if (off >= in->size) | |
10007 | return 0; | |
10008 | if (len == 0) | |
10009 | return 0; | |
10010 | if (off + len > in->size) { | |
10011 | len = in->size - off; | |
10012 | } | |
10013 | ||
10014 | ldout(cct, 10) << " min_bytes=" << f->readahead.get_min_readahead_size() | |
10015 | << " max_bytes=" << f->readahead.get_max_readahead_size() | |
10016 | << " max_periods=" << conf->client_readahead_max_periods << dendl; | |
10017 | ||
10018 | // read (and possibly block) | |
11fdf7f2 TL |
10019 | int r = 0; |
10020 | C_SaferCond onfinish("Client::_read_async flock"); | |
7c673cae | 10021 | r = objectcacher->file_read(&in->oset, &in->layout, in->snapid, |
11fdf7f2 | 10022 | off, len, bl, 0, &onfinish); |
7c673cae FG |
10023 | if (r == 0) { |
10024 | get_cap_ref(in, CEPH_CAP_FILE_CACHE); | |
9f95a23c | 10025 | client_lock.unlock(); |
11fdf7f2 | 10026 | r = onfinish.wait(); |
9f95a23c | 10027 | client_lock.lock(); |
7c673cae | 10028 | put_cap_ref(in, CEPH_CAP_FILE_CACHE); |
7c673cae FG |
10029 | } |
10030 | ||
10031 | if(f->readahead.get_min_readahead_size() > 0) { | |
10032 | pair<uint64_t, uint64_t> readahead_extent = f->readahead.update(off, len, in->size); | |
10033 | if (readahead_extent.second > 0) { | |
10034 | ldout(cct, 20) << "readahead " << readahead_extent.first << "~" << readahead_extent.second | |
10035 | << " (caller wants " << off << "~" << len << ")" << dendl; | |
10036 | Context *onfinish2 = new C_Readahead(this, f); | |
10037 | int r2 = objectcacher->file_read(&in->oset, &in->layout, in->snapid, | |
10038 | readahead_extent.first, readahead_extent.second, | |
10039 | NULL, 0, onfinish2); | |
10040 | if (r2 == 0) { | |
10041 | ldout(cct, 20) << "readahead initiated, c " << onfinish2 << dendl; | |
10042 | get_cap_ref(in, CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
10043 | } else { | |
10044 | ldout(cct, 20) << "readahead was no-op, already cached" << dendl; | |
10045 | delete onfinish2; | |
10046 | } | |
10047 | } | |
10048 | } | |
10049 | ||
10050 | return r; | |
10051 | } | |
10052 | ||
10053 | int Client::_read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, | |
10054 | bool *checkeof) | |
10055 | { | |
f67539c2 TL |
10056 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10057 | ||
7c673cae FG |
10058 | Inode *in = f->inode.get(); |
10059 | uint64_t pos = off; | |
10060 | int left = len; | |
10061 | int read = 0; | |
10062 | ||
11fdf7f2 | 10063 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae | 10064 | |
f67539c2 TL |
10065 | // 0 success, 1 continue and < 0 error happen. |
10066 | auto wait_and_copy = [&](C_SaferCond &onfinish, bufferlist &tbl, int wanted) { | |
11fdf7f2 | 10067 | int r = onfinish.wait(); |
7c673cae FG |
10068 | |
10069 | // if we get ENOENT from OSD, assume 0 bytes returned | |
f67539c2 | 10070 | if (r == -CEPHFS_ENOENT) |
7c673cae FG |
10071 | r = 0; |
10072 | if (r < 0) | |
10073 | return r; | |
f67539c2 | 10074 | |
7c673cae FG |
10075 | if (tbl.length()) { |
10076 | r = tbl.length(); | |
10077 | ||
10078 | read += r; | |
10079 | pos += r; | |
10080 | left -= r; | |
10081 | bl->claim_append(tbl); | |
10082 | } | |
10083 | // short read? | |
10084 | if (r >= 0 && r < wanted) { | |
10085 | if (pos < in->size) { | |
10086 | // zero up to known EOF | |
10087 | int64_t some = in->size - pos; | |
10088 | if (some > left) | |
10089 | some = left; | |
11fdf7f2 TL |
10090 | auto z = buffer::ptr_node::create(some); |
10091 | z->zero(); | |
10092 | bl->push_back(std::move(z)); | |
7c673cae FG |
10093 | read += some; |
10094 | pos += some; | |
10095 | left -= some; | |
10096 | if (left == 0) | |
f67539c2 | 10097 | return 0; |
7c673cae FG |
10098 | } |
10099 | ||
10100 | *checkeof = true; | |
f67539c2 | 10101 | return 0; |
7c673cae | 10102 | } |
f67539c2 TL |
10103 | return 1; |
10104 | }; | |
7c673cae | 10105 | |
f67539c2 TL |
10106 | while (left > 0) { |
10107 | C_SaferCond onfinish("Client::_read_sync flock"); | |
10108 | bufferlist tbl; | |
7c673cae | 10109 | |
f67539c2 TL |
10110 | int wanted = left; |
10111 | filer->read_trunc(in->ino, &in->layout, in->snapid, | |
10112 | pos, left, &tbl, 0, | |
10113 | in->truncate_size, in->truncate_seq, | |
10114 | &onfinish); | |
10115 | client_lock.unlock(); | |
10116 | int r = wait_and_copy(onfinish, tbl, wanted); | |
10117 | client_lock.lock(); | |
10118 | if (!r) | |
10119 | return read; | |
10120 | if (r < 0) | |
10121 | return r; | |
7c673cae | 10122 | } |
f67539c2 | 10123 | return read; |
7c673cae FG |
10124 | } |
10125 | ||
10126 | int Client::write(int fd, const char *buf, loff_t size, loff_t offset) | |
10127 | { | |
f67539c2 TL |
10128 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10129 | if (!mref_reader.is_state_satisfied()) | |
10130 | return -CEPHFS_ENOTCONN; | |
10131 | ||
7c673cae FG |
10132 | tout(cct) << "write" << std::endl; |
10133 | tout(cct) << fd << std::endl; | |
10134 | tout(cct) << size << std::endl; | |
10135 | tout(cct) << offset << std::endl; | |
10136 | ||
f67539c2 | 10137 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10138 | Fh *fh = get_filehandle(fd); |
10139 | if (!fh) | |
f67539c2 | 10140 | return -CEPHFS_EBADF; |
7c673cae FG |
10141 | #if defined(__linux__) && defined(O_PATH) |
10142 | if (fh->flags & O_PATH) | |
f67539c2 | 10143 | return -CEPHFS_EBADF; |
7c673cae | 10144 | #endif |
11fdf7f2 TL |
10145 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
10146 | size = std::min(size, (loff_t)INT_MAX); | |
10147 | int r = _write(fh, offset, size, buf, NULL, false); | |
7c673cae FG |
10148 | ldout(cct, 3) << "write(" << fd << ", \"...\", " << size << ", " << offset << ") = " << r << dendl; |
10149 | return r; | |
10150 | } | |
10151 | ||
10152 | int Client::pwritev(int fd, const struct iovec *iov, int iovcnt, int64_t offset) | |
10153 | { | |
10154 | if (iovcnt < 0) | |
f67539c2 | 10155 | return -CEPHFS_EINVAL; |
7c673cae FG |
10156 | return _preadv_pwritev(fd, iov, iovcnt, offset, true); |
10157 | } | |
10158 | ||
11fdf7f2 TL |
10159 | int64_t Client::_preadv_pwritev_locked(Fh *fh, const struct iovec *iov, |
10160 | unsigned iovcnt, int64_t offset, bool write, | |
f67539c2 | 10161 | bool clamp_to_int, std::unique_lock<ceph::mutex> &cl) |
7c673cae | 10162 | { |
7c673cae FG |
10163 | #if defined(__linux__) && defined(O_PATH) |
10164 | if (fh->flags & O_PATH) | |
f67539c2 | 10165 | return -CEPHFS_EBADF; |
7c673cae FG |
10166 | #endif |
10167 | loff_t totallen = 0; | |
10168 | for (unsigned i = 0; i < iovcnt; i++) { | |
10169 | totallen += iov[i].iov_len; | |
10170 | } | |
11fdf7f2 TL |
10171 | |
10172 | /* | |
10173 | * Some of the API functions take 64-bit size values, but only return | |
10174 | * 32-bit signed integers. Clamp the I/O sizes in those functions so that | |
10175 | * we don't do I/Os larger than the values we can return. | |
10176 | */ | |
10177 | if (clamp_to_int) { | |
10178 | totallen = std::min(totallen, (loff_t)INT_MAX); | |
10179 | } | |
7c673cae | 10180 | if (write) { |
11fdf7f2 TL |
10181 | int64_t w = _write(fh, offset, totallen, NULL, iov, iovcnt); |
10182 | ldout(cct, 3) << "pwritev(" << fh << ", \"...\", " << totallen << ", " << offset << ") = " << w << dendl; | |
7c673cae FG |
10183 | return w; |
10184 | } else { | |
10185 | bufferlist bl; | |
11fdf7f2 TL |
10186 | int64_t r = _read(fh, offset, totallen, &bl); |
10187 | ldout(cct, 3) << "preadv(" << fh << ", " << offset << ") = " << r << dendl; | |
7c673cae FG |
10188 | if (r <= 0) |
10189 | return r; | |
10190 | ||
f67539c2 | 10191 | cl.unlock(); |
9f95a23c | 10192 | auto iter = bl.cbegin(); |
7c673cae FG |
10193 | for (unsigned j = 0, resid = r; j < iovcnt && resid > 0; j++) { |
10194 | /* | |
f67539c2 TL |
10195 | * This piece of code aims to handle the case that bufferlist |
10196 | * does not have enough data to fill in the iov | |
7c673cae | 10197 | */ |
9f95a23c TL |
10198 | const auto round_size = std::min<unsigned>(resid, iov[j].iov_len); |
10199 | iter.copy(round_size, reinterpret_cast<char*>(iov[j].iov_base)); | |
10200 | resid -= round_size; | |
10201 | /* iter is self-updating */ | |
7c673cae | 10202 | } |
f67539c2 TL |
10203 | cl.lock(); |
10204 | return r; | |
7c673cae FG |
10205 | } |
10206 | } | |
10207 | ||
11fdf7f2 TL |
10208 | int Client::_preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write) |
10209 | { | |
f67539c2 TL |
10210 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10211 | if (!mref_reader.is_state_satisfied()) | |
10212 | return -CEPHFS_ENOTCONN; | |
10213 | ||
11fdf7f2 TL |
10214 | tout(cct) << fd << std::endl; |
10215 | tout(cct) << offset << std::endl; | |
10216 | ||
f67539c2 | 10217 | std::unique_lock cl(client_lock); |
11fdf7f2 TL |
10218 | Fh *fh = get_filehandle(fd); |
10219 | if (!fh) | |
f67539c2 TL |
10220 | return -CEPHFS_EBADF; |
10221 | return _preadv_pwritev_locked(fh, iov, iovcnt, offset, write, true, cl); | |
11fdf7f2 TL |
10222 | } |
10223 | ||
10224 | int64_t Client::_write(Fh *f, int64_t offset, uint64_t size, const char *buf, | |
10225 | const struct iovec *iov, int iovcnt) | |
7c673cae | 10226 | { |
f67539c2 TL |
10227 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10228 | ||
f64942e4 AA |
10229 | uint64_t fpos = 0; |
10230 | ||
7c673cae | 10231 | if ((uint64_t)(offset+size) > mdsmap->get_max_filesize()) //too large! |
f67539c2 | 10232 | return -CEPHFS_EFBIG; |
7c673cae FG |
10233 | |
10234 | //ldout(cct, 7) << "write fh " << fh << " size " << size << " offset " << offset << dendl; | |
10235 | Inode *in = f->inode.get(); | |
10236 | ||
10237 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
f67539c2 | 10238 | return -CEPHFS_ENOSPC; |
7c673cae FG |
10239 | } |
10240 | ||
11fdf7f2 | 10241 | ceph_assert(in->snapid == CEPH_NOSNAP); |
7c673cae FG |
10242 | |
10243 | // was Fh opened as writeable? | |
10244 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 10245 | return -CEPHFS_EBADF; |
7c673cae | 10246 | |
7c673cae FG |
10247 | // use/adjust fd pos? |
10248 | if (offset < 0) { | |
10249 | lock_fh_pos(f); | |
10250 | /* | |
10251 | * FIXME: this is racy in that we may block _after_ this point waiting for caps, and size may | |
10252 | * change out from under us. | |
10253 | */ | |
10254 | if (f->flags & O_APPEND) { | |
9f95a23c | 10255 | auto r = _lseek(f, 0, SEEK_END); |
7c673cae FG |
10256 | if (r < 0) { |
10257 | unlock_fh_pos(f); | |
10258 | return r; | |
10259 | } | |
10260 | } | |
10261 | offset = f->pos; | |
f64942e4 | 10262 | fpos = offset+size; |
7c673cae FG |
10263 | unlock_fh_pos(f); |
10264 | } | |
10265 | ||
11fdf7f2 TL |
10266 | // check quota |
10267 | uint64_t endoff = offset + size; | |
10268 | if (endoff > in->size && is_quota_bytes_exceeded(in, endoff - in->size, | |
10269 | f->actor_perms)) { | |
f67539c2 | 10270 | return -CEPHFS_EDQUOT; |
11fdf7f2 TL |
10271 | } |
10272 | ||
7c673cae FG |
10273 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
10274 | ||
10275 | ldout(cct, 10) << "cur file size is " << in->size << dendl; | |
10276 | ||
10277 | // time it. | |
10278 | utime_t start = ceph_clock_now(); | |
10279 | ||
10280 | if (in->inline_version == 0) { | |
10281 | int r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); | |
10282 | if (r < 0) | |
10283 | return r; | |
11fdf7f2 | 10284 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
10285 | } |
10286 | ||
10287 | // copy into fresh buffer (since our write may be resub, async) | |
10288 | bufferlist bl; | |
10289 | if (buf) { | |
10290 | if (size > 0) | |
10291 | bl.append(buf, size); | |
10292 | } else if (iov){ | |
10293 | for (int i = 0; i < iovcnt; i++) { | |
10294 | if (iov[i].iov_len > 0) { | |
10295 | bl.append((const char *)iov[i].iov_base, iov[i].iov_len); | |
10296 | } | |
10297 | } | |
10298 | } | |
10299 | ||
10300 | utime_t lat; | |
10301 | uint64_t totalwritten; | |
11fdf7f2 TL |
10302 | int want, have; |
10303 | if (f->mode & CEPH_FILE_MODE_LAZY) | |
10304 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
10305 | else | |
10306 | want = CEPH_CAP_FILE_BUFFER; | |
f6b5b4d7 | 10307 | int r = get_caps(f, CEPH_CAP_FILE_WR|CEPH_CAP_AUTH_SHARED, want, &have, endoff); |
7c673cae FG |
10308 | if (r < 0) |
10309 | return r; | |
10310 | ||
10311 | /* clear the setuid/setgid bits, if any */ | |
181888fb | 10312 | if (unlikely(in->mode & (S_ISUID|S_ISGID)) && size > 0) { |
7c673cae FG |
10313 | struct ceph_statx stx = { 0 }; |
10314 | ||
10315 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10316 | r = __setattrx(in, &stx, CEPH_SETATTR_KILL_SGUID, f->actor_perms); | |
10317 | if (r < 0) | |
10318 | return r; | |
10319 | } else { | |
10320 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
10321 | } | |
10322 | ||
10323 | if (f->flags & O_DIRECT) | |
11fdf7f2 | 10324 | have &= ~(CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
10325 | |
10326 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
10327 | ||
11fdf7f2 TL |
10328 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
10329 | ||
7c673cae FG |
10330 | if (in->inline_version < CEPH_INLINE_NONE) { |
10331 | if (endoff > cct->_conf->client_max_inline_size || | |
10332 | endoff > CEPH_INLINE_MAX_SIZE || | |
10333 | !(have & CEPH_CAP_FILE_BUFFER)) { | |
11fdf7f2 TL |
10334 | onuninline.reset(new C_SaferCond("Client::_write_uninline_data flock")); |
10335 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
10336 | } else { |
10337 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10338 | ||
10339 | uint32_t len = in->inline_data.length(); | |
10340 | ||
10341 | if (endoff < len) | |
9f95a23c | 10342 | in->inline_data.begin(endoff).copy(len - endoff, bl); // XXX |
7c673cae FG |
10343 | |
10344 | if (offset < len) | |
10345 | in->inline_data.splice(offset, len - offset); | |
10346 | else if (offset > len) | |
10347 | in->inline_data.append_zero(offset - len); | |
10348 | ||
10349 | in->inline_data.append(bl); | |
10350 | in->inline_version++; | |
10351 | ||
10352 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10353 | ||
10354 | goto success; | |
10355 | } | |
10356 | } | |
10357 | ||
11fdf7f2 TL |
10358 | if (cct->_conf->client_oc && |
10359 | (have & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
10360 | // do buffered write |
10361 | if (!in->oset.dirty_or_tx) | |
10362 | get_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
10363 | ||
10364 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10365 | ||
10366 | // async, caching, non-blocking. | |
10367 | r = objectcacher->file_write(&in->oset, &in->layout, | |
10368 | in->snaprealm->get_snap_context(), | |
10369 | offset, size, bl, ceph::real_clock::now(), | |
10370 | 0); | |
10371 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
10372 | ||
10373 | if (r < 0) | |
10374 | goto done; | |
10375 | ||
10376 | // flush cached write if O_SYNC is set on file fh | |
10377 | // O_DSYNC == O_SYNC on linux < 2.6.33 | |
10378 | // O_SYNC = __O_SYNC | O_DSYNC on linux >= 2.6.33 | |
10379 | if ((f->flags & O_SYNC) || (f->flags & O_DSYNC)) { | |
10380 | _flush_range(in, offset, size); | |
10381 | } | |
10382 | } else { | |
10383 | if (f->flags & O_DIRECT) | |
10384 | _flush_range(in, offset, size); | |
10385 | ||
10386 | // simple, non-atomic sync write | |
11fdf7f2 | 10387 | C_SaferCond onfinish("Client::_write flock"); |
f67539c2 | 10388 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
10389 | |
10390 | filer->write_trunc(in->ino, &in->layout, in->snaprealm->get_snap_context(), | |
10391 | offset, size, bl, ceph::real_clock::now(), 0, | |
10392 | in->truncate_size, in->truncate_seq, | |
11fdf7f2 | 10393 | &onfinish); |
9f95a23c | 10394 | client_lock.unlock(); |
f6b5b4d7 | 10395 | r = onfinish.wait(); |
9f95a23c | 10396 | client_lock.lock(); |
f67539c2 | 10397 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
f6b5b4d7 TL |
10398 | if (r < 0) |
10399 | goto done; | |
7c673cae FG |
10400 | } |
10401 | ||
10402 | // if we get here, write was successful, update client metadata | |
10403 | success: | |
10404 | // time | |
10405 | lat = ceph_clock_now(); | |
10406 | lat -= start; | |
10407 | logger->tinc(l_c_wrlat, lat); | |
10408 | ||
f64942e4 AA |
10409 | if (fpos) { |
10410 | lock_fh_pos(f); | |
10411 | f->pos = fpos; | |
10412 | unlock_fh_pos(f); | |
10413 | } | |
7c673cae | 10414 | totalwritten = size; |
11fdf7f2 | 10415 | r = (int64_t)totalwritten; |
7c673cae FG |
10416 | |
10417 | // extend file? | |
10418 | if (totalwritten + offset > in->size) { | |
10419 | in->size = totalwritten + offset; | |
28e407b8 | 10420 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 10421 | |
11fdf7f2 | 10422 | if (is_quota_bytes_approaching(in, f->actor_perms)) { |
7c673cae | 10423 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
10424 | } else if (is_max_size_approaching(in)) { |
10425 | check_caps(in, 0); | |
7c673cae FG |
10426 | } |
10427 | ||
10428 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", extending file size" << dendl; | |
10429 | } else { | |
10430 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", leaving file size at " << in->size << dendl; | |
10431 | } | |
10432 | ||
10433 | // mtime | |
91327a77 | 10434 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 10435 | in->change_attr++; |
28e407b8 | 10436 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10437 | |
10438 | done: | |
10439 | ||
11fdf7f2 | 10440 | if (nullptr != onuninline) { |
9f95a23c | 10441 | client_lock.unlock(); |
11fdf7f2 | 10442 | int uninline_ret = onuninline->wait(); |
9f95a23c | 10443 | client_lock.lock(); |
7c673cae | 10444 | |
f67539c2 | 10445 | if (uninline_ret >= 0 || uninline_ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
10446 | in->inline_data.clear(); |
10447 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 10448 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
10449 | check_caps(in, 0); |
10450 | } else | |
10451 | r = uninline_ret; | |
10452 | } | |
10453 | ||
10454 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
10455 | return r; | |
10456 | } | |
10457 | ||
10458 | int Client::_flush(Fh *f) | |
10459 | { | |
10460 | Inode *in = f->inode.get(); | |
10461 | int err = f->take_async_err(); | |
10462 | if (err != 0) { | |
10463 | ldout(cct, 1) << __func__ << ": " << f << " on inode " << *in << " caught async_err = " | |
10464 | << cpp_strerror(err) << dendl; | |
10465 | } else { | |
10466 | ldout(cct, 10) << __func__ << ": " << f << " on inode " << *in << " no async_err state" << dendl; | |
10467 | } | |
10468 | ||
10469 | return err; | |
10470 | } | |
10471 | ||
10472 | int Client::truncate(const char *relpath, loff_t length, const UserPerm& perms) | |
10473 | { | |
10474 | struct ceph_statx stx; | |
10475 | stx.stx_size = length; | |
10476 | return setattrx(relpath, &stx, CEPH_SETATTR_SIZE, perms); | |
10477 | } | |
10478 | ||
10479 | int Client::ftruncate(int fd, loff_t length, const UserPerm& perms) | |
10480 | { | |
f67539c2 TL |
10481 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10482 | if (!mref_reader.is_state_satisfied()) | |
10483 | return -CEPHFS_ENOTCONN; | |
10484 | ||
11fdf7f2 | 10485 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
10486 | tout(cct) << fd << std::endl; |
10487 | tout(cct) << length << std::endl; | |
10488 | ||
f67539c2 | 10489 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10490 | Fh *f = get_filehandle(fd); |
10491 | if (!f) | |
f67539c2 | 10492 | return -CEPHFS_EBADF; |
7c673cae FG |
10493 | #if defined(__linux__) && defined(O_PATH) |
10494 | if (f->flags & O_PATH) | |
f67539c2 | 10495 | return -CEPHFS_EBADF; |
7c673cae | 10496 | #endif |
adb31ebb | 10497 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) |
f67539c2 | 10498 | return -CEPHFS_EBADF; |
7c673cae FG |
10499 | struct stat attr; |
10500 | attr.st_size = length; | |
10501 | return _setattr(f->inode, &attr, CEPH_SETATTR_SIZE, perms); | |
10502 | } | |
10503 | ||
10504 | int Client::fsync(int fd, bool syncdataonly) | |
10505 | { | |
f67539c2 TL |
10506 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10507 | if (!mref_reader.is_state_satisfied()) | |
10508 | return -CEPHFS_ENOTCONN; | |
10509 | ||
7c673cae FG |
10510 | tout(cct) << "fsync" << std::endl; |
10511 | tout(cct) << fd << std::endl; | |
10512 | tout(cct) << syncdataonly << std::endl; | |
10513 | ||
f67539c2 | 10514 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10515 | Fh *f = get_filehandle(fd); |
10516 | if (!f) | |
f67539c2 | 10517 | return -CEPHFS_EBADF; |
7c673cae FG |
10518 | #if defined(__linux__) && defined(O_PATH) |
10519 | if (f->flags & O_PATH) | |
f67539c2 | 10520 | return -CEPHFS_EBADF; |
7c673cae FG |
10521 | #endif |
10522 | int r = _fsync(f, syncdataonly); | |
10523 | if (r == 0) { | |
10524 | // The IOs in this fsync were okay, but maybe something happened | |
10525 | // in the background that we shoudl be reporting? | |
10526 | r = f->take_async_err(); | |
1adf2230 | 10527 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly |
7c673cae FG |
10528 | << ") = 0, async_err = " << r << dendl; |
10529 | } else { | |
10530 | // Assume that an error we encountered during fsync, even reported | |
10531 | // synchronously, would also have applied the error to the Fh, and we | |
10532 | // should clear it here to avoid returning the same error again on next | |
10533 | // call. | |
1adf2230 | 10534 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly << ") = " |
7c673cae FG |
10535 | << r << dendl; |
10536 | f->take_async_err(); | |
10537 | } | |
10538 | return r; | |
10539 | } | |
10540 | ||
10541 | int Client::_fsync(Inode *in, bool syncdataonly) | |
10542 | { | |
f67539c2 TL |
10543 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
10544 | ||
7c673cae | 10545 | int r = 0; |
11fdf7f2 | 10546 | std::unique_ptr<C_SaferCond> object_cacher_completion = nullptr; |
7c673cae FG |
10547 | ceph_tid_t flush_tid = 0; |
10548 | InodeRef tmp_ref; | |
11fdf7f2 TL |
10549 | utime_t lat; |
10550 | utime_t start = ceph_clock_now(); | |
7c673cae | 10551 | |
1adf2230 | 10552 | ldout(cct, 8) << "_fsync on " << *in << " " << (syncdataonly ? "(dataonly)":"(data+metadata)") << dendl; |
7c673cae FG |
10553 | |
10554 | if (cct->_conf->client_oc) { | |
11fdf7f2 TL |
10555 | object_cacher_completion.reset(new C_SaferCond("Client::_fsync::lock")); |
10556 | tmp_ref = in; // take a reference; C_SaferCond doesn't and _flush won't either | |
10557 | _flush(in, object_cacher_completion.get()); | |
7c673cae FG |
10558 | ldout(cct, 15) << "using return-valued form of _fsync" << dendl; |
10559 | } | |
10560 | ||
10561 | if (!syncdataonly && in->dirty_caps) { | |
10562 | check_caps(in, CHECK_CAPS_NODELAY|CHECK_CAPS_SYNCHRONOUS); | |
10563 | if (in->flushing_caps) | |
10564 | flush_tid = last_flush_tid; | |
10565 | } else ldout(cct, 10) << "no metadata needs to commit" << dendl; | |
10566 | ||
10567 | if (!syncdataonly && !in->unsafe_ops.empty()) { | |
28e407b8 AA |
10568 | flush_mdlog_sync(); |
10569 | ||
7c673cae FG |
10570 | MetaRequest *req = in->unsafe_ops.back(); |
10571 | ldout(cct, 15) << "waiting on unsafe requests, last tid " << req->get_tid() << dendl; | |
10572 | ||
10573 | req->get(); | |
10574 | wait_on_list(req->waitfor_safe); | |
10575 | put_request(req); | |
10576 | } | |
10577 | ||
11fdf7f2 | 10578 | if (nullptr != object_cacher_completion) { // wait on a real reply instead of guessing |
9f95a23c | 10579 | client_lock.unlock(); |
7c673cae | 10580 | ldout(cct, 15) << "waiting on data to flush" << dendl; |
11fdf7f2 | 10581 | r = object_cacher_completion->wait(); |
9f95a23c | 10582 | client_lock.lock(); |
7c673cae FG |
10583 | ldout(cct, 15) << "got " << r << " from flush writeback" << dendl; |
10584 | } else { | |
10585 | // FIXME: this can starve | |
10586 | while (in->cap_refs[CEPH_CAP_FILE_BUFFER] > 0) { | |
10587 | ldout(cct, 10) << "ino " << in->ino << " has " << in->cap_refs[CEPH_CAP_FILE_BUFFER] | |
10588 | << " uncommitted, waiting" << dendl; | |
10589 | wait_on_list(in->waitfor_commit); | |
10590 | } | |
10591 | } | |
10592 | ||
10593 | if (!r) { | |
10594 | if (flush_tid > 0) | |
10595 | wait_sync_caps(in, flush_tid); | |
10596 | ||
10597 | ldout(cct, 10) << "ino " << in->ino << " has no uncommitted writes" << dendl; | |
10598 | } else { | |
1adf2230 | 10599 | ldout(cct, 8) << "ino " << in->ino << " failed to commit to disk! " |
7c673cae FG |
10600 | << cpp_strerror(-r) << dendl; |
10601 | } | |
11fdf7f2 TL |
10602 | |
10603 | lat = ceph_clock_now(); | |
10604 | lat -= start; | |
10605 | logger->tinc(l_c_fsync, lat); | |
7c673cae FG |
10606 | |
10607 | return r; | |
10608 | } | |
10609 | ||
10610 | int Client::_fsync(Fh *f, bool syncdataonly) | |
10611 | { | |
1adf2230 | 10612 | ldout(cct, 8) << "_fsync(" << f << ", " << (syncdataonly ? "dataonly)":"data+metadata)") << dendl; |
7c673cae FG |
10613 | return _fsync(f->inode.get(), syncdataonly); |
10614 | } | |
10615 | ||
10616 | int Client::fstat(int fd, struct stat *stbuf, const UserPerm& perms, int mask) | |
10617 | { | |
f67539c2 TL |
10618 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10619 | if (!mref_reader.is_state_satisfied()) | |
10620 | return -CEPHFS_ENOTCONN; | |
10621 | ||
7c673cae FG |
10622 | tout(cct) << "fstat mask " << hex << mask << dec << std::endl; |
10623 | tout(cct) << fd << std::endl; | |
10624 | ||
f67539c2 | 10625 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10626 | Fh *f = get_filehandle(fd); |
10627 | if (!f) | |
f67539c2 | 10628 | return -CEPHFS_EBADF; |
7c673cae FG |
10629 | int r = _getattr(f->inode, mask, perms); |
10630 | if (r < 0) | |
10631 | return r; | |
10632 | fill_stat(f->inode, stbuf, NULL); | |
1adf2230 | 10633 | ldout(cct, 5) << "fstat(" << fd << ", " << stbuf << ") = " << r << dendl; |
7c673cae FG |
10634 | return r; |
10635 | } | |
10636 | ||
10637 | int Client::fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms, | |
10638 | unsigned int want, unsigned int flags) | |
10639 | { | |
f67539c2 TL |
10640 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10641 | if (!mref_reader.is_state_satisfied()) | |
10642 | return -CEPHFS_ENOTCONN; | |
10643 | ||
7c673cae FG |
10644 | tout(cct) << "fstatx flags " << hex << flags << " want " << want << dec << std::endl; |
10645 | tout(cct) << fd << std::endl; | |
10646 | ||
f67539c2 | 10647 | std::scoped_lock lock(client_lock); |
7c673cae FG |
10648 | Fh *f = get_filehandle(fd); |
10649 | if (!f) | |
f67539c2 | 10650 | return -CEPHFS_EBADF; |
7c673cae FG |
10651 | |
10652 | unsigned mask = statx_to_mask(flags, want); | |
10653 | ||
10654 | int r = 0; | |
b3b6e05e | 10655 | if (mask) { |
7c673cae FG |
10656 | r = _getattr(f->inode, mask, perms); |
10657 | if (r < 0) { | |
10658 | ldout(cct, 3) << "fstatx exit on error!" << dendl; | |
10659 | return r; | |
10660 | } | |
10661 | } | |
10662 | ||
10663 | fill_statx(f->inode, mask, stx); | |
10664 | ldout(cct, 3) << "fstatx(" << fd << ", " << stx << ") = " << r << dendl; | |
10665 | return r; | |
10666 | } | |
10667 | ||
b3b6e05e TL |
10668 | int Client::statxat(int dirfd, const char *relpath, |
10669 | struct ceph_statx *stx, const UserPerm& perms, | |
10670 | unsigned int want, unsigned int flags) { | |
10671 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
10672 | if (!mref_reader.is_state_satisfied()) { | |
10673 | return -CEPHFS_ENOTCONN; | |
10674 | } | |
10675 | ||
10676 | tout(cct) << __func__ << " flags " << hex << flags << " want " << want << dec << std::endl; | |
10677 | tout(cct) << dirfd << std::endl; | |
10678 | tout(cct) << relpath << std::endl; | |
10679 | ||
10680 | unsigned mask = statx_to_mask(flags, want); | |
10681 | ||
10682 | InodeRef dirinode; | |
10683 | std::scoped_lock lock(client_lock); | |
10684 | int r = get_fd_inode(dirfd, &dirinode); | |
10685 | if (r < 0) { | |
10686 | return r; | |
10687 | } | |
10688 | ||
10689 | InodeRef in; | |
10690 | filepath path(relpath); | |
10691 | r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask, dirinode); | |
10692 | if (r < 0) { | |
10693 | return r; | |
10694 | } | |
10695 | r = _getattr(in, mask, perms); | |
10696 | if (r < 0) { | |
10697 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; | |
10698 | return r; | |
10699 | } | |
10700 | ||
10701 | fill_statx(in, mask, stx); | |
10702 | ldout(cct, 3) << __func__ << " dirfd" << dirfd << ", r= " << r << dendl; | |
10703 | return r; | |
10704 | } | |
10705 | ||
7c673cae FG |
10706 | // not written yet, but i want to link! |
10707 | ||
10708 | int Client::chdir(const char *relpath, std::string &new_cwd, | |
10709 | const UserPerm& perms) | |
10710 | { | |
f67539c2 TL |
10711 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10712 | if (!mref_reader.is_state_satisfied()) | |
10713 | return -CEPHFS_ENOTCONN; | |
10714 | ||
7c673cae FG |
10715 | tout(cct) << "chdir" << std::endl; |
10716 | tout(cct) << relpath << std::endl; | |
181888fb | 10717 | |
7c673cae FG |
10718 | filepath path(relpath); |
10719 | InodeRef in; | |
f67539c2 TL |
10720 | |
10721 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
10722 | int r = path_walk(path, &in, perms); |
10723 | if (r < 0) | |
10724 | return r; | |
92f5a8d4 TL |
10725 | |
10726 | if (!(in.get()->is_dir())) | |
f67539c2 | 10727 | return -CEPHFS_ENOTDIR; |
92f5a8d4 | 10728 | |
7c673cae FG |
10729 | if (cwd != in) |
10730 | cwd.swap(in); | |
10731 | ldout(cct, 3) << "chdir(" << relpath << ") cwd now " << cwd->ino << dendl; | |
10732 | ||
b5b8bbf5 | 10733 | _getcwd(new_cwd, perms); |
7c673cae FG |
10734 | return 0; |
10735 | } | |
10736 | ||
b5b8bbf5 | 10737 | void Client::_getcwd(string& dir, const UserPerm& perms) |
7c673cae FG |
10738 | { |
10739 | filepath path; | |
11fdf7f2 | 10740 | ldout(cct, 10) << __func__ << " " << *cwd << dendl; |
7c673cae FG |
10741 | |
10742 | Inode *in = cwd.get(); | |
b3b6e05e | 10743 | while (in != root.get()) { |
11fdf7f2 | 10744 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked |
7c673cae FG |
10745 | |
10746 | // A cwd or ancester is unlinked | |
11fdf7f2 | 10747 | if (in->dentries.empty()) { |
7c673cae FG |
10748 | return; |
10749 | } | |
10750 | ||
10751 | Dentry *dn = in->get_first_parent(); | |
10752 | ||
10753 | ||
10754 | if (!dn) { | |
10755 | // look it up | |
11fdf7f2 | 10756 | ldout(cct, 10) << __func__ << " looking up parent for " << *in << dendl; |
7c673cae FG |
10757 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
10758 | filepath path(in->ino); | |
10759 | req->set_filepath(path); | |
10760 | req->set_inode(in); | |
10761 | int res = make_request(req, perms); | |
10762 | if (res < 0) | |
10763 | break; | |
10764 | ||
10765 | // start over | |
10766 | path = filepath(); | |
10767 | in = cwd.get(); | |
10768 | continue; | |
10769 | } | |
10770 | path.push_front_dentry(dn->name); | |
10771 | in = dn->dir->parent_inode; | |
10772 | } | |
10773 | dir = "/"; | |
10774 | dir += path.get_path(); | |
10775 | } | |
10776 | ||
b5b8bbf5 FG |
10777 | void Client::getcwd(string& dir, const UserPerm& perms) |
10778 | { | |
f67539c2 TL |
10779 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10780 | if (!mref_reader.is_state_satisfied()) | |
10781 | return; | |
10782 | ||
10783 | std::scoped_lock l(client_lock); | |
10784 | ||
10785 | _getcwd(dir, perms); | |
b5b8bbf5 FG |
10786 | } |
10787 | ||
7c673cae FG |
10788 | int Client::statfs(const char *path, struct statvfs *stbuf, |
10789 | const UserPerm& perms) | |
10790 | { | |
f67539c2 TL |
10791 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
10792 | if (!mref_reader.is_state_satisfied()) | |
10793 | return -CEPHFS_ENOTCONN; | |
10794 | ||
11fdf7f2 | 10795 | tout(cct) << __func__ << std::endl; |
91327a77 | 10796 | unsigned long int total_files_on_fs; |
7c673cae FG |
10797 | |
10798 | ceph_statfs stats; | |
10799 | C_SaferCond cond; | |
d2e6a577 | 10800 | |
f67539c2 | 10801 | std::unique_lock lock(client_lock); |
d2e6a577 FG |
10802 | const vector<int64_t> &data_pools = mdsmap->get_data_pools(); |
10803 | if (data_pools.size() == 1) { | |
10804 | objecter->get_fs_stats(stats, data_pools[0], &cond); | |
10805 | } else { | |
10806 | objecter->get_fs_stats(stats, boost::optional<int64_t>(), &cond); | |
10807 | } | |
7c673cae | 10808 | |
f67539c2 | 10809 | lock.unlock(); |
7c673cae | 10810 | int rval = cond.wait(); |
f67539c2 TL |
10811 | lock.lock(); |
10812 | ||
91327a77 AA |
10813 | assert(root); |
10814 | total_files_on_fs = root->rstat.rfiles + root->rstat.rsubdirs; | |
7c673cae FG |
10815 | |
10816 | if (rval < 0) { | |
10817 | ldout(cct, 1) << "underlying call to statfs returned error: " | |
10818 | << cpp_strerror(rval) | |
10819 | << dendl; | |
10820 | return rval; | |
10821 | } | |
10822 | ||
10823 | memset(stbuf, 0, sizeof(*stbuf)); | |
10824 | ||
10825 | /* | |
10826 | * we're going to set a block size of 4MB so we can represent larger | |
10827 | * FSes without overflowing. Additionally convert the space | |
10828 | * measurements from KB to bytes while making them in terms of | |
10829 | * blocks. We use 4MB only because it is big enough, and because it | |
10830 | * actually *is* the (ceph) default block size. | |
10831 | */ | |
10832 | const int CEPH_BLOCK_SHIFT = 22; | |
10833 | stbuf->f_frsize = 1 << CEPH_BLOCK_SHIFT; | |
10834 | stbuf->f_bsize = 1 << CEPH_BLOCK_SHIFT; | |
91327a77 | 10835 | stbuf->f_files = total_files_on_fs; |
f67539c2 | 10836 | stbuf->f_ffree = -1; |
7c673cae FG |
10837 | stbuf->f_favail = -1; |
10838 | stbuf->f_fsid = -1; // ?? | |
10839 | stbuf->f_flag = 0; // ?? | |
10840 | stbuf->f_namemax = NAME_MAX; | |
10841 | ||
10842 | // Usually quota_root will == root_ancestor, but if the mount root has no | |
10843 | // quota but we can see a parent of it that does have a quota, we'll | |
10844 | // respect that one instead. | |
11fdf7f2 | 10845 | ceph_assert(root != nullptr); |
b3b6e05e | 10846 | InodeRef quota_root = root->quota.is_enable() ? root : get_quota_root(root.get(), perms); |
7c673cae FG |
10847 | |
10848 | // get_quota_root should always give us something | |
10849 | // because client quotas are always enabled | |
11fdf7f2 | 10850 | ceph_assert(quota_root != nullptr); |
7c673cae FG |
10851 | |
10852 | if (quota_root && cct->_conf->client_quota_df && quota_root->quota.max_bytes) { | |
10853 | ||
10854 | // Skip the getattr if any sessions are stale, as we don't want to | |
10855 | // block `df` if this client has e.g. been evicted, or if the MDS cluster | |
10856 | // is unhealthy. | |
10857 | if (!_any_stale_sessions()) { | |
10858 | int r = _getattr(quota_root, 0, perms, true); | |
10859 | if (r != 0) { | |
10860 | // Ignore return value: error getting latest inode metadata is not a good | |
10861 | // reason to break "df". | |
10862 | lderr(cct) << "Error in getattr on quota root 0x" | |
10863 | << std::hex << quota_root->ino << std::dec | |
10864 | << " statfs result may be outdated" << dendl; | |
10865 | } | |
10866 | } | |
10867 | ||
10868 | // Special case: if there is a size quota set on the Inode acting | |
10869 | // as the root for this client mount, then report the quota status | |
10870 | // as the filesystem statistics. | |
10871 | const fsblkcnt_t total = quota_root->quota.max_bytes >> CEPH_BLOCK_SHIFT; | |
10872 | const fsblkcnt_t used = quota_root->rstat.rbytes >> CEPH_BLOCK_SHIFT; | |
31f18b77 FG |
10873 | // It is possible for a quota to be exceeded: arithmetic here must |
10874 | // handle case where used > total. | |
10875 | const fsblkcnt_t free = total > used ? total - used : 0; | |
7c673cae FG |
10876 | |
10877 | stbuf->f_blocks = total; | |
10878 | stbuf->f_bfree = free; | |
10879 | stbuf->f_bavail = free; | |
10880 | } else { | |
d2e6a577 | 10881 | // General case: report the cluster statistics returned from RADOS. Because |
7c673cae FG |
10882 | // multiple pools may be used without one filesystem namespace via |
10883 | // layouts, this is the most correct thing we can do. | |
10884 | stbuf->f_blocks = stats.kb >> (CEPH_BLOCK_SHIFT - 10); | |
10885 | stbuf->f_bfree = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
10886 | stbuf->f_bavail = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
10887 | } | |
10888 | ||
10889 | return rval; | |
10890 | } | |
10891 | ||
10892 | int Client::_do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep, | |
10893 | struct flock *fl, uint64_t owner, bool removing) | |
10894 | { | |
11fdf7f2 | 10895 | ldout(cct, 10) << __func__ << " ino " << in->ino |
7c673cae FG |
10896 | << (lock_type == CEPH_LOCK_FCNTL ? " fcntl" : " flock") |
10897 | << " type " << fl->l_type << " owner " << owner | |
10898 | << " " << fl->l_start << "~" << fl->l_len << dendl; | |
10899 | ||
f6b5b4d7 | 10900 | if (in->flags & I_ERROR_FILELOCK) |
f67539c2 | 10901 | return -CEPHFS_EIO; |
f6b5b4d7 | 10902 | |
7c673cae FG |
10903 | int lock_cmd; |
10904 | if (F_RDLCK == fl->l_type) | |
10905 | lock_cmd = CEPH_LOCK_SHARED; | |
10906 | else if (F_WRLCK == fl->l_type) | |
10907 | lock_cmd = CEPH_LOCK_EXCL; | |
10908 | else if (F_UNLCK == fl->l_type) | |
10909 | lock_cmd = CEPH_LOCK_UNLOCK; | |
10910 | else | |
f67539c2 | 10911 | return -CEPHFS_EIO; |
7c673cae FG |
10912 | |
10913 | if (op != CEPH_MDS_OP_SETFILELOCK || lock_cmd == CEPH_LOCK_UNLOCK) | |
10914 | sleep = 0; | |
10915 | ||
10916 | /* | |
10917 | * Set the most significant bit, so that MDS knows the 'owner' | |
10918 | * is sufficient to identify the owner of lock. (old code uses | |
10919 | * both 'owner' and 'pid') | |
10920 | */ | |
10921 | owner |= (1ULL << 63); | |
10922 | ||
10923 | MetaRequest *req = new MetaRequest(op); | |
10924 | filepath path; | |
10925 | in->make_nosnap_relative_path(path); | |
10926 | req->set_filepath(path); | |
10927 | req->set_inode(in); | |
10928 | ||
10929 | req->head.args.filelock_change.rule = lock_type; | |
10930 | req->head.args.filelock_change.type = lock_cmd; | |
10931 | req->head.args.filelock_change.owner = owner; | |
10932 | req->head.args.filelock_change.pid = fl->l_pid; | |
10933 | req->head.args.filelock_change.start = fl->l_start; | |
10934 | req->head.args.filelock_change.length = fl->l_len; | |
10935 | req->head.args.filelock_change.wait = sleep; | |
10936 | ||
10937 | int ret; | |
10938 | bufferlist bl; | |
10939 | ||
10940 | if (sleep && switch_interrupt_cb) { | |
10941 | // enable interrupt | |
10942 | switch_interrupt_cb(callback_handle, req->get()); | |
10943 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
7c673cae FG |
10944 | // disable interrupt |
10945 | switch_interrupt_cb(callback_handle, NULL); | |
31f18b77 FG |
10946 | if (ret == 0 && req->aborted()) { |
10947 | // effect of this lock request has been revoked by the 'lock intr' request | |
10948 | ret = req->get_abort_code(); | |
10949 | } | |
7c673cae FG |
10950 | put_request(req); |
10951 | } else { | |
10952 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
10953 | } | |
10954 | ||
10955 | if (ret == 0) { | |
10956 | if (op == CEPH_MDS_OP_GETFILELOCK) { | |
10957 | ceph_filelock filelock; | |
11fdf7f2 TL |
10958 | auto p = bl.cbegin(); |
10959 | decode(filelock, p); | |
7c673cae FG |
10960 | |
10961 | if (CEPH_LOCK_SHARED == filelock.type) | |
10962 | fl->l_type = F_RDLCK; | |
10963 | else if (CEPH_LOCK_EXCL == filelock.type) | |
10964 | fl->l_type = F_WRLCK; | |
10965 | else | |
10966 | fl->l_type = F_UNLCK; | |
10967 | ||
10968 | fl->l_whence = SEEK_SET; | |
10969 | fl->l_start = filelock.start; | |
10970 | fl->l_len = filelock.length; | |
10971 | fl->l_pid = filelock.pid; | |
10972 | } else if (op == CEPH_MDS_OP_SETFILELOCK) { | |
10973 | ceph_lock_state_t *lock_state; | |
10974 | if (lock_type == CEPH_LOCK_FCNTL) { | |
10975 | if (!in->fcntl_locks) | |
11fdf7f2 TL |
10976 | in->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
10977 | lock_state = in->fcntl_locks.get(); | |
7c673cae FG |
10978 | } else if (lock_type == CEPH_LOCK_FLOCK) { |
10979 | if (!in->flock_locks) | |
11fdf7f2 TL |
10980 | in->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
10981 | lock_state = in->flock_locks.get(); | |
7c673cae FG |
10982 | } else { |
10983 | ceph_abort(); | |
f67539c2 | 10984 | return -CEPHFS_EINVAL; |
7c673cae FG |
10985 | } |
10986 | _update_lock_state(fl, owner, lock_state); | |
10987 | ||
10988 | if (!removing) { | |
10989 | if (lock_type == CEPH_LOCK_FCNTL) { | |
10990 | if (!fh->fcntl_locks) | |
11fdf7f2 TL |
10991 | fh->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
10992 | lock_state = fh->fcntl_locks.get(); | |
7c673cae FG |
10993 | } else { |
10994 | if (!fh->flock_locks) | |
11fdf7f2 TL |
10995 | fh->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
10996 | lock_state = fh->flock_locks.get(); | |
7c673cae FG |
10997 | } |
10998 | _update_lock_state(fl, owner, lock_state); | |
10999 | } | |
11000 | } else | |
11001 | ceph_abort(); | |
11002 | } | |
11003 | return ret; | |
11004 | } | |
11005 | ||
11006 | int Client::_interrupt_filelock(MetaRequest *req) | |
11007 | { | |
31f18b77 FG |
11008 | // Set abort code, but do not kick. The abort code prevents the request |
11009 | // from being re-sent. | |
f67539c2 | 11010 | req->abort(-CEPHFS_EINTR); |
31f18b77 FG |
11011 | if (req->mds < 0) |
11012 | return 0; // haven't sent the request | |
11013 | ||
7c673cae FG |
11014 | Inode *in = req->inode(); |
11015 | ||
11016 | int lock_type; | |
11017 | if (req->head.args.filelock_change.rule == CEPH_LOCK_FLOCK) | |
11018 | lock_type = CEPH_LOCK_FLOCK_INTR; | |
11019 | else if (req->head.args.filelock_change.rule == CEPH_LOCK_FCNTL) | |
11020 | lock_type = CEPH_LOCK_FCNTL_INTR; | |
11021 | else { | |
11022 | ceph_abort(); | |
f67539c2 | 11023 | return -CEPHFS_EINVAL; |
7c673cae FG |
11024 | } |
11025 | ||
11026 | MetaRequest *intr_req = new MetaRequest(CEPH_MDS_OP_SETFILELOCK); | |
11027 | filepath path; | |
11028 | in->make_nosnap_relative_path(path); | |
11029 | intr_req->set_filepath(path); | |
11030 | intr_req->set_inode(in); | |
11031 | intr_req->head.args.filelock_change = req->head.args.filelock_change; | |
11032 | intr_req->head.args.filelock_change.rule = lock_type; | |
11033 | intr_req->head.args.filelock_change.type = CEPH_LOCK_UNLOCK; | |
11034 | ||
11035 | UserPerm perms(req->get_uid(), req->get_gid()); | |
11036 | return make_request(intr_req, perms, NULL, NULL, -1); | |
11037 | } | |
11038 | ||
11039 | void Client::_encode_filelocks(Inode *in, bufferlist& bl) | |
11040 | { | |
11041 | if (!in->fcntl_locks && !in->flock_locks) | |
11042 | return; | |
11043 | ||
11044 | unsigned nr_fcntl_locks = in->fcntl_locks ? in->fcntl_locks->held_locks.size() : 0; | |
11fdf7f2 | 11045 | encode(nr_fcntl_locks, bl); |
7c673cae | 11046 | if (nr_fcntl_locks) { |
11fdf7f2 | 11047 | auto &lock_state = in->fcntl_locks; |
7c673cae FG |
11048 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); |
11049 | p != lock_state->held_locks.end(); | |
11050 | ++p) | |
11fdf7f2 | 11051 | encode(p->second, bl); |
7c673cae FG |
11052 | } |
11053 | ||
11054 | unsigned nr_flock_locks = in->flock_locks ? in->flock_locks->held_locks.size() : 0; | |
11fdf7f2 | 11055 | encode(nr_flock_locks, bl); |
7c673cae | 11056 | if (nr_flock_locks) { |
11fdf7f2 | 11057 | auto &lock_state = in->flock_locks; |
7c673cae FG |
11058 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); |
11059 | p != lock_state->held_locks.end(); | |
11060 | ++p) | |
11fdf7f2 | 11061 | encode(p->second, bl); |
7c673cae FG |
11062 | } |
11063 | ||
11fdf7f2 | 11064 | ldout(cct, 10) << __func__ << " ino " << in->ino << ", " << nr_fcntl_locks |
7c673cae FG |
11065 | << " fcntl locks, " << nr_flock_locks << " flock locks" << dendl; |
11066 | } | |
11067 | ||
11068 | void Client::_release_filelocks(Fh *fh) | |
11069 | { | |
11070 | if (!fh->fcntl_locks && !fh->flock_locks) | |
11071 | return; | |
11072 | ||
11073 | Inode *in = fh->inode.get(); | |
11fdf7f2 | 11074 | ldout(cct, 10) << __func__ << " " << fh << " ino " << in->ino << dendl; |
7c673cae | 11075 | |
f6b5b4d7 TL |
11076 | list<ceph_filelock> activated_locks; |
11077 | ||
7c673cae FG |
11078 | list<pair<int, ceph_filelock> > to_release; |
11079 | ||
11080 | if (fh->fcntl_locks) { | |
11fdf7f2 | 11081 | auto &lock_state = fh->fcntl_locks; |
f6b5b4d7 TL |
11082 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11083 | auto q = p++; | |
11084 | if (in->flags & I_ERROR_FILELOCK) { | |
11085 | lock_state->remove_lock(q->second, activated_locks); | |
11086 | } else { | |
11087 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FCNTL, q->second)); | |
11088 | } | |
11089 | } | |
11fdf7f2 | 11090 | lock_state.reset(); |
7c673cae FG |
11091 | } |
11092 | if (fh->flock_locks) { | |
11fdf7f2 | 11093 | auto &lock_state = fh->flock_locks; |
f6b5b4d7 TL |
11094 | for(auto p = lock_state->held_locks.begin(); p != lock_state->held_locks.end(); ) { |
11095 | auto q = p++; | |
11096 | if (in->flags & I_ERROR_FILELOCK) { | |
11097 | lock_state->remove_lock(q->second, activated_locks); | |
11098 | } else { | |
11099 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FLOCK, q->second)); | |
11100 | } | |
11101 | } | |
11fdf7f2 | 11102 | lock_state.reset(); |
7c673cae FG |
11103 | } |
11104 | ||
f6b5b4d7 TL |
11105 | if ((in->flags & I_ERROR_FILELOCK) && !in->has_any_filelocks()) |
11106 | in->flags &= ~I_ERROR_FILELOCK; | |
7c673cae | 11107 | |
f6b5b4d7 | 11108 | if (to_release.empty()) |
11fdf7f2 TL |
11109 | return; |
11110 | ||
7c673cae FG |
11111 | struct flock fl; |
11112 | memset(&fl, 0, sizeof(fl)); | |
11113 | fl.l_whence = SEEK_SET; | |
11114 | fl.l_type = F_UNLCK; | |
11115 | ||
11116 | for (list<pair<int, ceph_filelock> >::iterator p = to_release.begin(); | |
11117 | p != to_release.end(); | |
11118 | ++p) { | |
11119 | fl.l_start = p->second.start; | |
11120 | fl.l_len = p->second.length; | |
11121 | fl.l_pid = p->second.pid; | |
11122 | _do_filelock(in, fh, p->first, CEPH_MDS_OP_SETFILELOCK, 0, &fl, | |
11123 | p->second.owner, true); | |
11124 | } | |
11125 | } | |
11126 | ||
11127 | void Client::_update_lock_state(struct flock *fl, uint64_t owner, | |
11128 | ceph_lock_state_t *lock_state) | |
11129 | { | |
11130 | int lock_cmd; | |
11131 | if (F_RDLCK == fl->l_type) | |
11132 | lock_cmd = CEPH_LOCK_SHARED; | |
11133 | else if (F_WRLCK == fl->l_type) | |
11134 | lock_cmd = CEPH_LOCK_EXCL; | |
11135 | else | |
11136 | lock_cmd = CEPH_LOCK_UNLOCK;; | |
11137 | ||
11138 | ceph_filelock filelock; | |
11139 | filelock.start = fl->l_start; | |
11140 | filelock.length = fl->l_len; | |
11141 | filelock.client = 0; | |
11142 | // see comment in _do_filelock() | |
11143 | filelock.owner = owner | (1ULL << 63); | |
11144 | filelock.pid = fl->l_pid; | |
11145 | filelock.type = lock_cmd; | |
11146 | ||
11147 | if (filelock.type == CEPH_LOCK_UNLOCK) { | |
11148 | list<ceph_filelock> activated_locks; | |
11149 | lock_state->remove_lock(filelock, activated_locks); | |
11150 | } else { | |
11151 | bool r = lock_state->add_lock(filelock, false, false, NULL); | |
11fdf7f2 | 11152 | ceph_assert(r); |
7c673cae FG |
11153 | } |
11154 | } | |
11155 | ||
11156 | int Client::_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
11157 | { | |
11158 | Inode *in = fh->inode.get(); | |
11159 | ldout(cct, 10) << "_getlk " << fh << " ino " << in->ino << dendl; | |
11160 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_GETFILELOCK, 0, fl, owner); | |
11161 | return ret; | |
11162 | } | |
11163 | ||
11164 | int Client::_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
11165 | { | |
11166 | Inode *in = fh->inode.get(); | |
11167 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << dendl; | |
11168 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_SETFILELOCK, sleep, fl, owner); | |
11169 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11170 | return ret; | |
11171 | } | |
11172 | ||
11173 | int Client::_flock(Fh *fh, int cmd, uint64_t owner) | |
11174 | { | |
11175 | Inode *in = fh->inode.get(); | |
11176 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << dendl; | |
11177 | ||
11178 | int sleep = !(cmd & LOCK_NB); | |
11179 | cmd &= ~LOCK_NB; | |
11180 | ||
11181 | int type; | |
11182 | switch (cmd) { | |
11183 | case LOCK_SH: | |
11184 | type = F_RDLCK; | |
11185 | break; | |
11186 | case LOCK_EX: | |
11187 | type = F_WRLCK; | |
11188 | break; | |
11189 | case LOCK_UN: | |
11190 | type = F_UNLCK; | |
11191 | break; | |
11192 | default: | |
f67539c2 | 11193 | return -CEPHFS_EINVAL; |
7c673cae FG |
11194 | } |
11195 | ||
11196 | struct flock fl; | |
11197 | memset(&fl, 0, sizeof(fl)); | |
11198 | fl.l_type = type; | |
11199 | fl.l_whence = SEEK_SET; | |
11200 | ||
11201 | int ret = _do_filelock(in, fh, CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK, sleep, &fl, owner); | |
11202 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
11203 | return ret; | |
11204 | } | |
11205 | ||
f67539c2 TL |
11206 | int Client::get_snap_info(const char *path, const UserPerm &perms, SnapInfo *snap_info) { |
11207 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11208 | if (!mref_reader.is_state_satisfied()) { | |
11209 | return -CEPHFS_ENOTCONN; | |
11210 | } | |
11211 | ||
11212 | std::unique_lock locker(client_lock); | |
11213 | InodeRef in; | |
11214 | int r = Client::path_walk(path, &in, perms, true); | |
11215 | if (r < 0) { | |
11216 | return r; | |
11217 | } | |
11218 | ||
11219 | if (in->snapid == CEPH_NOSNAP) { | |
11220 | return -CEPHFS_EINVAL; | |
11221 | } | |
11222 | ||
11223 | snap_info->id = in->snapid; | |
11224 | snap_info->metadata = in->snap_metadata; | |
11225 | return 0; | |
11226 | } | |
11227 | ||
11228 | int Client::ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms) | |
11229 | { | |
11230 | /* Since the only thing this does is wrap a call to statfs, and | |
11231 | statfs takes a lock, it doesn't seem we have a need to split it | |
11232 | out. */ | |
7c673cae FG |
11233 | return statfs(0, stbuf, perms); |
11234 | } | |
11235 | ||
e306af50 | 11236 | void Client::ll_register_callbacks(struct ceph_client_callback_args *args) |
7c673cae FG |
11237 | { |
11238 | if (!args) | |
11239 | return; | |
f67539c2 | 11240 | std::scoped_lock l(client_lock); |
11fdf7f2 | 11241 | ldout(cct, 10) << __func__ << " cb " << args->handle |
7c673cae FG |
11242 | << " invalidate_ino_cb " << args->ino_cb |
11243 | << " invalidate_dentry_cb " << args->dentry_cb | |
7c673cae FG |
11244 | << " switch_interrupt_cb " << args->switch_intr_cb |
11245 | << " remount_cb " << args->remount_cb | |
11246 | << dendl; | |
11247 | callback_handle = args->handle; | |
11248 | if (args->ino_cb) { | |
11249 | ino_invalidate_cb = args->ino_cb; | |
11250 | async_ino_invalidator.start(); | |
11251 | } | |
11252 | if (args->dentry_cb) { | |
11253 | dentry_invalidate_cb = args->dentry_cb; | |
11254 | async_dentry_invalidator.start(); | |
11255 | } | |
11256 | if (args->switch_intr_cb) { | |
11257 | switch_interrupt_cb = args->switch_intr_cb; | |
11258 | interrupt_finisher.start(); | |
11259 | } | |
11260 | if (args->remount_cb) { | |
11261 | remount_cb = args->remount_cb; | |
11262 | remount_finisher.start(); | |
11263 | } | |
e306af50 TL |
11264 | if (args->ino_release_cb) { |
11265 | ino_release_cb = args->ino_release_cb; | |
11266 | async_ino_releasor.start(); | |
11267 | } | |
11268 | if (args->umask_cb) | |
11269 | umask_cb = args->umask_cb; | |
7c673cae FG |
11270 | } |
11271 | ||
11272 | int Client::test_dentry_handling(bool can_invalidate) | |
11273 | { | |
11274 | int r = 0; | |
11275 | ||
f67539c2 TL |
11276 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
11277 | if (!iref_reader.is_state_satisfied()) | |
11278 | return -CEPHFS_ENOTCONN; | |
11279 | ||
7c673cae FG |
11280 | can_invalidate_dentries = can_invalidate; |
11281 | ||
11282 | if (can_invalidate_dentries) { | |
11fdf7f2 | 11283 | ceph_assert(dentry_invalidate_cb); |
7c673cae | 11284 | ldout(cct, 1) << "using dentry_invalidate_cb" << dendl; |
b32b8144 | 11285 | r = 0; |
11fdf7f2 TL |
11286 | } else { |
11287 | ceph_assert(remount_cb); | |
7c673cae | 11288 | ldout(cct, 1) << "using remount_cb" << dendl; |
91327a77 | 11289 | r = _do_remount(false); |
b32b8144 | 11290 | } |
11fdf7f2 | 11291 | |
7c673cae FG |
11292 | return r; |
11293 | } | |
11294 | ||
11295 | int Client::_sync_fs() | |
11296 | { | |
f67539c2 TL |
11297 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
11298 | ||
11fdf7f2 | 11299 | ldout(cct, 10) << __func__ << dendl; |
7c673cae FG |
11300 | |
11301 | // flush file data | |
11fdf7f2 TL |
11302 | std::unique_ptr<C_SaferCond> cond = nullptr; |
11303 | if (cct->_conf->client_oc) { | |
11304 | cond.reset(new C_SaferCond("Client::_sync_fs:lock")); | |
11305 | objectcacher->flush_all(cond.get()); | |
11306 | } | |
7c673cae FG |
11307 | |
11308 | // flush caps | |
11309 | flush_caps_sync(); | |
11310 | ceph_tid_t flush_tid = last_flush_tid; | |
11311 | ||
11312 | // wait for unsafe mds requests | |
11313 | wait_unsafe_requests(); | |
11314 | ||
11315 | wait_sync_caps(flush_tid); | |
11316 | ||
11fdf7f2 | 11317 | if (nullptr != cond) { |
9f95a23c | 11318 | client_lock.unlock(); |
11fdf7f2 TL |
11319 | ldout(cct, 15) << __func__ << " waiting on data to flush" << dendl; |
11320 | cond->wait(); | |
11321 | ldout(cct, 15) << __func__ << " flush finished" << dendl; | |
9f95a23c | 11322 | client_lock.lock(); |
7c673cae FG |
11323 | } |
11324 | ||
11325 | return 0; | |
11326 | } | |
11327 | ||
11328 | int Client::sync_fs() | |
11329 | { | |
f67539c2 TL |
11330 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11331 | if (!mref_reader.is_state_satisfied()) | |
11332 | return -CEPHFS_ENOTCONN; | |
181888fb | 11333 | |
f67539c2 | 11334 | std::scoped_lock l(client_lock); |
181888fb | 11335 | |
7c673cae FG |
11336 | return _sync_fs(); |
11337 | } | |
11338 | ||
11339 | int64_t Client::drop_caches() | |
11340 | { | |
f67539c2 | 11341 | std::scoped_lock l(client_lock); |
7c673cae FG |
11342 | return objectcacher->release_all(); |
11343 | } | |
11344 | ||
11fdf7f2 TL |
11345 | int Client::_lazyio(Fh *fh, int enable) |
11346 | { | |
11347 | Inode *in = fh->inode.get(); | |
11348 | ldout(cct, 20) << __func__ << " " << *in << " " << !!enable << dendl; | |
11349 | ||
11350 | if (!!(fh->mode & CEPH_FILE_MODE_LAZY) == !!enable) | |
11351 | return 0; | |
11352 | ||
11353 | int orig_mode = fh->mode; | |
11354 | if (enable) { | |
11355 | fh->mode |= CEPH_FILE_MODE_LAZY; | |
11356 | in->get_open_ref(fh->mode); | |
11357 | in->put_open_ref(orig_mode); | |
11358 | check_caps(in, CHECK_CAPS_NODELAY); | |
11359 | } else { | |
11360 | fh->mode &= ~CEPH_FILE_MODE_LAZY; | |
11361 | in->get_open_ref(fh->mode); | |
11362 | in->put_open_ref(orig_mode); | |
11363 | check_caps(in, 0); | |
11364 | } | |
11365 | ||
11366 | return 0; | |
11367 | } | |
11368 | ||
11369 | int Client::lazyio(int fd, int enable) | |
11370 | { | |
f67539c2 | 11371 | std::scoped_lock l(client_lock); |
11fdf7f2 TL |
11372 | Fh *f = get_filehandle(fd); |
11373 | if (!f) | |
f67539c2 | 11374 | return -CEPHFS_EBADF; |
11fdf7f2 TL |
11375 | |
11376 | return _lazyio(f, enable); | |
11377 | } | |
11378 | ||
11379 | int Client::ll_lazyio(Fh *fh, int enable) | |
11380 | { | |
11fdf7f2 TL |
11381 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << !!enable << dendl; |
11382 | tout(cct) << __func__ << std::endl; | |
11383 | ||
f67539c2 | 11384 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
11385 | return _lazyio(fh, enable); |
11386 | } | |
7c673cae | 11387 | |
92f5a8d4 | 11388 | int Client::lazyio_propagate(int fd, loff_t offset, size_t count) |
7c673cae | 11389 | { |
f67539c2 | 11390 | std::scoped_lock l(client_lock); |
92f5a8d4 | 11391 | ldout(cct, 3) << "op: client->lazyio_propagate(" << fd |
7c673cae FG |
11392 | << ", " << offset << ", " << count << ")" << dendl; |
11393 | ||
11394 | Fh *f = get_filehandle(fd); | |
11395 | if (!f) | |
f67539c2 | 11396 | return -CEPHFS_EBADF; |
7c673cae FG |
11397 | |
11398 | // for now | |
11399 | _fsync(f, true); | |
11400 | ||
11401 | return 0; | |
11402 | } | |
11403 | ||
11404 | int Client::lazyio_synchronize(int fd, loff_t offset, size_t count) | |
11405 | { | |
f67539c2 | 11406 | std::scoped_lock l(client_lock); |
7c673cae FG |
11407 | ldout(cct, 3) << "op: client->lazyio_synchronize(" << fd |
11408 | << ", " << offset << ", " << count << ")" << dendl; | |
11409 | ||
11410 | Fh *f = get_filehandle(fd); | |
11411 | if (!f) | |
f67539c2 | 11412 | return -CEPHFS_EBADF; |
7c673cae FG |
11413 | Inode *in = f->inode.get(); |
11414 | ||
11415 | _fsync(f, true); | |
92f5a8d4 TL |
11416 | if (_release(in)) { |
11417 | int r =_getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
11418 | if (r < 0) | |
11419 | return r; | |
11420 | } | |
7c673cae FG |
11421 | return 0; |
11422 | } | |
11423 | ||
11424 | ||
11425 | // ============================= | |
11426 | // snaps | |
11427 | ||
f67539c2 TL |
11428 | int Client::mksnap(const char *relpath, const char *name, const UserPerm& perm, |
11429 | mode_t mode, const std::map<std::string, std::string> &metadata) | |
7c673cae | 11430 | { |
f67539c2 TL |
11431 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11432 | if (!mref_reader.is_state_satisfied()) | |
11433 | return -CEPHFS_ENOTCONN; | |
181888fb | 11434 | |
f67539c2 | 11435 | std::scoped_lock l(client_lock); |
181888fb | 11436 | |
7c673cae FG |
11437 | filepath path(relpath); |
11438 | InodeRef in; | |
11439 | int r = path_walk(path, &in, perm); | |
11440 | if (r < 0) | |
11441 | return r; | |
11442 | if (cct->_conf->client_permissions) { | |
11443 | r = may_create(in.get(), perm); | |
11444 | if (r < 0) | |
11445 | return r; | |
11446 | } | |
11447 | Inode *snapdir = open_snapdir(in.get()); | |
f67539c2 | 11448 | return _mkdir(snapdir, name, mode, perm, nullptr, metadata); |
7c673cae | 11449 | } |
181888fb | 11450 | |
f67539c2 | 11451 | int Client::rmsnap(const char *relpath, const char *name, const UserPerm& perms, bool check_perms) |
7c673cae | 11452 | { |
f67539c2 TL |
11453 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11454 | if (!mref_reader.is_state_satisfied()) | |
11455 | return -CEPHFS_ENOTCONN; | |
181888fb | 11456 | |
f67539c2 | 11457 | std::scoped_lock l(client_lock); |
181888fb | 11458 | |
7c673cae FG |
11459 | filepath path(relpath); |
11460 | InodeRef in; | |
11461 | int r = path_walk(path, &in, perms); | |
11462 | if (r < 0) | |
11463 | return r; | |
f67539c2 | 11464 | Inode *snapdir = open_snapdir(in.get()); |
7c673cae | 11465 | if (cct->_conf->client_permissions) { |
f67539c2 | 11466 | r = may_delete(snapdir, check_perms ? name : NULL, perms); |
7c673cae FG |
11467 | if (r < 0) |
11468 | return r; | |
11469 | } | |
7c673cae FG |
11470 | return _rmdir(snapdir, name, perms); |
11471 | } | |
11472 | ||
11473 | // ============================= | |
11474 | // expose caps | |
11475 | ||
f67539c2 TL |
11476 | int Client::get_caps_issued(int fd) |
11477 | { | |
11478 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
11479 | if (!mref_reader.is_state_satisfied()) | |
11480 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11481 | |
f67539c2 | 11482 | std::scoped_lock lock(client_lock); |
181888fb | 11483 | |
7c673cae FG |
11484 | Fh *f = get_filehandle(fd); |
11485 | if (!f) | |
f67539c2 | 11486 | return -CEPHFS_EBADF; |
7c673cae FG |
11487 | |
11488 | return f->inode->caps_issued(); | |
11489 | } | |
11490 | ||
11491 | int Client::get_caps_issued(const char *path, const UserPerm& perms) | |
11492 | { | |
f67539c2 TL |
11493 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11494 | if (!mref_reader.is_state_satisfied()) | |
11495 | return -CEPHFS_ENOTCONN; | |
181888fb | 11496 | |
f67539c2 | 11497 | std::scoped_lock lock(client_lock); |
181888fb | 11498 | |
7c673cae FG |
11499 | filepath p(path); |
11500 | InodeRef in; | |
11501 | int r = path_walk(p, &in, perms, true); | |
11502 | if (r < 0) | |
11503 | return r; | |
11504 | return in->caps_issued(); | |
11505 | } | |
11506 | ||
11507 | // ========================================= | |
11508 | // low level | |
11509 | ||
11510 | Inode *Client::open_snapdir(Inode *diri) | |
11511 | { | |
11512 | Inode *in; | |
11513 | vinodeno_t vino(diri->ino, CEPH_SNAPDIR); | |
11514 | if (!inode_map.count(vino)) { | |
11515 | in = new Inode(this, vino, &diri->layout); | |
11516 | ||
11517 | in->ino = diri->ino; | |
11518 | in->snapid = CEPH_SNAPDIR; | |
11519 | in->mode = diri->mode; | |
11520 | in->uid = diri->uid; | |
11521 | in->gid = diri->gid; | |
494da23a | 11522 | in->nlink = 1; |
7c673cae FG |
11523 | in->mtime = diri->mtime; |
11524 | in->ctime = diri->ctime; | |
11525 | in->btime = diri->btime; | |
f6b5b4d7 | 11526 | in->atime = diri->atime; |
7c673cae FG |
11527 | in->size = diri->size; |
11528 | in->change_attr = diri->change_attr; | |
11529 | ||
11530 | in->dirfragtree.clear(); | |
11531 | in->snapdir_parent = diri; | |
11532 | diri->flags |= I_SNAPDIR_OPEN; | |
11533 | inode_map[vino] = in; | |
11534 | if (use_faked_inos()) | |
11535 | _assign_faked_ino(in); | |
11536 | ldout(cct, 10) << "open_snapdir created snapshot inode " << *in << dendl; | |
11537 | } else { | |
11538 | in = inode_map[vino]; | |
11539 | ldout(cct, 10) << "open_snapdir had snapshot inode " << *in << dendl; | |
11540 | } | |
11541 | return in; | |
11542 | } | |
11543 | ||
11544 | int Client::ll_lookup(Inode *parent, const char *name, struct stat *attr, | |
11545 | Inode **out, const UserPerm& perms) | |
11546 | { | |
f67539c2 TL |
11547 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11548 | if (!mref_reader.is_state_satisfied()) | |
11549 | return -CEPHFS_ENOTCONN; | |
11550 | ||
31f18b77 | 11551 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 TL |
11552 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
11553 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11554 | tout(cct) << name << std::endl; |
11555 | ||
f67539c2 | 11556 | std::scoped_lock lock(client_lock); |
181888fb | 11557 | |
7c673cae | 11558 | int r = 0; |
11fdf7f2 TL |
11559 | if (!fuse_default_permissions) { |
11560 | if (strcmp(name, ".") && strcmp(name, "..")) { | |
11561 | r = may_lookup(parent, perms); | |
11562 | if (r < 0) | |
11563 | return r; | |
11564 | } | |
7c673cae FG |
11565 | } |
11566 | ||
11567 | string dname(name); | |
11568 | InodeRef in; | |
11569 | ||
11570 | r = _lookup(parent, dname, CEPH_STAT_CAP_INODE_ALL, &in, perms); | |
11571 | if (r < 0) { | |
11572 | attr->st_ino = 0; | |
11573 | goto out; | |
11574 | } | |
11575 | ||
11fdf7f2 | 11576 | ceph_assert(in); |
7c673cae FG |
11577 | fill_stat(in, attr); |
11578 | _ll_get(in.get()); | |
11579 | ||
11580 | out: | |
11fdf7f2 | 11581 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11582 | << " -> " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; |
11583 | tout(cct) << attr->st_ino << std::endl; | |
11584 | *out = in.get(); | |
11585 | return r; | |
11586 | } | |
11587 | ||
f67539c2 TL |
11588 | int Client::ll_lookup_vino( |
11589 | vinodeno_t vino, | |
1adf2230 AA |
11590 | const UserPerm& perms, |
11591 | Inode **inode) | |
11592 | { | |
81eedcae | 11593 | ceph_assert(inode != NULL); |
f67539c2 TL |
11594 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11595 | if (!mref_reader.is_state_satisfied()) | |
11596 | return -CEPHFS_ENOTCONN; | |
81eedcae | 11597 | |
b3b6e05e TL |
11598 | if (is_reserved_vino(vino)) |
11599 | return -CEPHFS_ESTALE; | |
11600 | ||
f67539c2 TL |
11601 | std::scoped_lock lock(client_lock); |
11602 | ldout(cct, 3) << __func__ << " " << vino << dendl; | |
1adf2230 | 11603 | |
f67539c2 TL |
11604 | // Check the cache first |
11605 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
11606 | if (p != inode_map.end()) { | |
11607 | *inode = p->second; | |
11608 | _ll_get(*inode); | |
81eedcae TL |
11609 | return 0; |
11610 | } | |
11611 | ||
f67539c2 | 11612 | uint64_t snapid = vino.snapid; |
81eedcae | 11613 | |
f67539c2 TL |
11614 | // for snapdir, find the non-snapped dir inode |
11615 | if (snapid == CEPH_SNAPDIR) | |
11616 | vino.snapid = CEPH_NOSNAP; | |
11617 | ||
11618 | int r = _lookup_vino(vino, perms, inode); | |
11619 | if (r) | |
1adf2230 | 11620 | return r; |
f67539c2 | 11621 | ceph_assert(*inode != NULL); |
81eedcae | 11622 | |
f67539c2 TL |
11623 | if (snapid == CEPH_SNAPDIR) { |
11624 | Inode *tmp = *inode; | |
1adf2230 | 11625 | |
f67539c2 TL |
11626 | // open the snapdir and put the inode ref |
11627 | *inode = open_snapdir(tmp); | |
11628 | _ll_forget(tmp, 1); | |
11629 | _ll_get(*inode); | |
1adf2230 | 11630 | } |
1adf2230 AA |
11631 | return 0; |
11632 | } | |
11633 | ||
f67539c2 TL |
11634 | int Client::ll_lookup_inode( |
11635 | struct inodeno_t ino, | |
11636 | const UserPerm& perms, | |
11637 | Inode **inode) | |
11638 | { | |
11639 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
11640 | return ll_lookup_vino(vino, perms, inode); | |
11641 | } | |
11642 | ||
7c673cae FG |
11643 | int Client::ll_lookupx(Inode *parent, const char *name, Inode **out, |
11644 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
11645 | const UserPerm& perms) | |
11646 | { | |
f67539c2 TL |
11647 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11648 | if (!mref_reader.is_state_satisfied()) | |
11649 | return -CEPHFS_ENOTCONN; | |
11650 | ||
31f18b77 | 11651 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 | 11652 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
7c673cae FG |
11653 | tout(cct) << "ll_lookupx" << std::endl; |
11654 | tout(cct) << name << std::endl; | |
11655 | ||
f67539c2 | 11656 | std::scoped_lock lock(client_lock); |
181888fb | 11657 | |
7c673cae | 11658 | int r = 0; |
11fdf7f2 | 11659 | if (!fuse_default_permissions) { |
7c673cae FG |
11660 | r = may_lookup(parent, perms); |
11661 | if (r < 0) | |
11662 | return r; | |
11663 | } | |
11664 | ||
11665 | string dname(name); | |
11666 | InodeRef in; | |
11667 | ||
11668 | unsigned mask = statx_to_mask(flags, want); | |
11669 | r = _lookup(parent, dname, mask, &in, perms); | |
11670 | if (r < 0) { | |
11671 | stx->stx_ino = 0; | |
11672 | stx->stx_mask = 0; | |
11673 | } else { | |
11fdf7f2 | 11674 | ceph_assert(in); |
7c673cae FG |
11675 | fill_statx(in, mask, stx); |
11676 | _ll_get(in.get()); | |
11677 | } | |
11678 | ||
11fdf7f2 | 11679 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
11680 | << " -> " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; |
11681 | tout(cct) << stx->stx_ino << std::endl; | |
11682 | *out = in.get(); | |
11683 | return r; | |
11684 | } | |
11685 | ||
11686 | int Client::ll_walk(const char* name, Inode **out, struct ceph_statx *stx, | |
11687 | unsigned int want, unsigned int flags, const UserPerm& perms) | |
11688 | { | |
f67539c2 TL |
11689 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11690 | if (!mref_reader.is_state_satisfied()) | |
11691 | return -CEPHFS_ENOTCONN; | |
181888fb | 11692 | |
7c673cae FG |
11693 | filepath fp(name, 0); |
11694 | InodeRef in; | |
11695 | int rc; | |
11696 | unsigned mask = statx_to_mask(flags, want); | |
11697 | ||
11fdf7f2 TL |
11698 | ldout(cct, 3) << __func__ << " " << name << dendl; |
11699 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11700 | tout(cct) << name << std::endl; |
11701 | ||
f67539c2 | 11702 | std::scoped_lock lock(client_lock); |
7c673cae FG |
11703 | rc = path_walk(fp, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); |
11704 | if (rc < 0) { | |
11705 | /* zero out mask, just in case... */ | |
11706 | stx->stx_mask = 0; | |
11707 | stx->stx_ino = 0; | |
11708 | *out = NULL; | |
11709 | return rc; | |
11710 | } else { | |
11fdf7f2 | 11711 | ceph_assert(in); |
7c673cae FG |
11712 | fill_statx(in, mask, stx); |
11713 | _ll_get(in.get()); | |
11714 | *out = in.get(); | |
11715 | return 0; | |
11716 | } | |
11717 | } | |
11718 | ||
11719 | void Client::_ll_get(Inode *in) | |
11720 | { | |
11721 | if (in->ll_ref == 0) { | |
b3b6e05e | 11722 | in->iget(); |
11fdf7f2 TL |
11723 | if (in->is_dir() && !in->dentries.empty()) { |
11724 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
11725 | in->get_first_parent()->get(); // pin dentry |
11726 | } | |
11fdf7f2 TL |
11727 | if (in->snapid != CEPH_NOSNAP) |
11728 | ll_snap_ref[in->snapid]++; | |
7c673cae FG |
11729 | } |
11730 | in->ll_get(); | |
11fdf7f2 | 11731 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " -> " << in->ll_ref << dendl; |
7c673cae FG |
11732 | } |
11733 | ||
494da23a | 11734 | int Client::_ll_put(Inode *in, uint64_t num) |
7c673cae FG |
11735 | { |
11736 | in->ll_put(num); | |
11fdf7f2 | 11737 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " " << num << " -> " << in->ll_ref << dendl; |
7c673cae | 11738 | if (in->ll_ref == 0) { |
11fdf7f2 TL |
11739 | if (in->is_dir() && !in->dentries.empty()) { |
11740 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
11741 | in->get_first_parent()->put(); // unpin dentry |
11742 | } | |
11fdf7f2 TL |
11743 | if (in->snapid != CEPH_NOSNAP) { |
11744 | auto p = ll_snap_ref.find(in->snapid); | |
11745 | ceph_assert(p != ll_snap_ref.end()); | |
11746 | ceph_assert(p->second > 0); | |
11747 | if (--p->second == 0) | |
11748 | ll_snap_ref.erase(p); | |
11749 | } | |
7c673cae FG |
11750 | put_inode(in); |
11751 | return 0; | |
11752 | } else { | |
11753 | return in->ll_ref; | |
11754 | } | |
11755 | } | |
11756 | ||
11757 | void Client::_ll_drop_pins() | |
11758 | { | |
11fdf7f2 | 11759 | ldout(cct, 10) << __func__ << dendl; |
1adf2230 | 11760 | std::set<InodeRef> to_be_put; //this set will be deconstructed item by item when exit |
7c673cae FG |
11761 | ceph::unordered_map<vinodeno_t, Inode*>::iterator next; |
11762 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
11763 | it != inode_map.end(); | |
11764 | it = next) { | |
11765 | Inode *in = it->second; | |
11766 | next = it; | |
11767 | ++next; | |
1adf2230 AA |
11768 | if (in->ll_ref){ |
11769 | to_be_put.insert(in); | |
7c673cae | 11770 | _ll_put(in, in->ll_ref); |
1adf2230 | 11771 | } |
7c673cae FG |
11772 | } |
11773 | } | |
11774 | ||
494da23a | 11775 | bool Client::_ll_forget(Inode *in, uint64_t count) |
7c673cae | 11776 | { |
11fdf7f2 | 11777 | inodeno_t ino = in->ino; |
7c673cae | 11778 | |
11fdf7f2 TL |
11779 | ldout(cct, 8) << __func__ << " " << ino << " " << count << dendl; |
11780 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11781 | tout(cct) << ino.val << std::endl; |
11782 | tout(cct) << count << std::endl; | |
11783 | ||
181888fb | 11784 | // Ignore forget if we're no longer mounted |
f67539c2 TL |
11785 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11786 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
11787 | return true; |
11788 | ||
7c673cae FG |
11789 | if (ino == 1) return true; // ignore forget on root. |
11790 | ||
11791 | bool last = false; | |
11792 | if (in->ll_ref < count) { | |
11793 | ldout(cct, 1) << "WARNING: ll_forget on " << ino << " " << count | |
11794 | << ", which only has ll_ref=" << in->ll_ref << dendl; | |
11795 | _ll_put(in, in->ll_ref); | |
11796 | last = true; | |
11797 | } else { | |
11798 | if (_ll_put(in, count) == 0) | |
11799 | last = true; | |
11800 | } | |
11801 | ||
11802 | return last; | |
11803 | } | |
11804 | ||
494da23a | 11805 | bool Client::ll_forget(Inode *in, uint64_t count) |
1adf2230 | 11806 | { |
f67539c2 | 11807 | std::scoped_lock lock(client_lock); |
1adf2230 AA |
11808 | return _ll_forget(in, count); |
11809 | } | |
11810 | ||
7c673cae FG |
11811 | bool Client::ll_put(Inode *in) |
11812 | { | |
11813 | /* ll_forget already takes the lock */ | |
11814 | return ll_forget(in, 1); | |
11815 | } | |
11816 | ||
11fdf7f2 TL |
11817 | int Client::ll_get_snap_ref(snapid_t snap) |
11818 | { | |
f67539c2 | 11819 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
11820 | auto p = ll_snap_ref.find(snap); |
11821 | if (p != ll_snap_ref.end()) | |
11822 | return p->second; | |
11823 | return 0; | |
11824 | } | |
11825 | ||
7c673cae FG |
11826 | snapid_t Client::ll_get_snapid(Inode *in) |
11827 | { | |
f67539c2 | 11828 | std::scoped_lock lock(client_lock); |
7c673cae FG |
11829 | return in->snapid; |
11830 | } | |
11831 | ||
11832 | Inode *Client::ll_get_inode(ino_t ino) | |
11833 | { | |
f67539c2 TL |
11834 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11835 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
11836 | return NULL; |
11837 | ||
f67539c2 TL |
11838 | std::scoped_lock lock(client_lock); |
11839 | ||
7c673cae FG |
11840 | vinodeno_t vino = _map_faked_ino(ino); |
11841 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
11842 | if (p == inode_map.end()) | |
11843 | return NULL; | |
11844 | Inode *in = p->second; | |
11845 | _ll_get(in); | |
11846 | return in; | |
11847 | } | |
11848 | ||
11849 | Inode *Client::ll_get_inode(vinodeno_t vino) | |
11850 | { | |
f67539c2 TL |
11851 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11852 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
11853 | return NULL; |
11854 | ||
b3b6e05e TL |
11855 | if (is_reserved_vino(vino)) |
11856 | return NULL; | |
11857 | ||
f67539c2 TL |
11858 | std::scoped_lock lock(client_lock); |
11859 | ||
7c673cae FG |
11860 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
11861 | if (p == inode_map.end()) | |
11862 | return NULL; | |
11863 | Inode *in = p->second; | |
11864 | _ll_get(in); | |
11865 | return in; | |
11866 | } | |
11867 | ||
11868 | int Client::_ll_getattr(Inode *in, int caps, const UserPerm& perms) | |
11869 | { | |
11870 | vinodeno_t vino = _get_vino(in); | |
11871 | ||
11fdf7f2 TL |
11872 | ldout(cct, 8) << __func__ << " " << vino << dendl; |
11873 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11874 | tout(cct) << vino.ino.val << std::endl; |
11875 | ||
11876 | if (vino.snapid < CEPH_NOSNAP) | |
11877 | return 0; | |
11878 | else | |
11879 | return _getattr(in, caps, perms); | |
11880 | } | |
11881 | ||
11882 | int Client::ll_getattr(Inode *in, struct stat *attr, const UserPerm& perms) | |
11883 | { | |
f67539c2 TL |
11884 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11885 | if (!mref_reader.is_state_satisfied()) | |
11886 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11887 | |
f67539c2 | 11888 | std::scoped_lock lock(client_lock); |
181888fb | 11889 | |
7c673cae FG |
11890 | int res = _ll_getattr(in, CEPH_STAT_CAP_INODE_ALL, perms); |
11891 | ||
11892 | if (res == 0) | |
11893 | fill_stat(in, attr); | |
11fdf7f2 | 11894 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11895 | return res; |
11896 | } | |
11897 | ||
11898 | int Client::ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want, | |
11899 | unsigned int flags, const UserPerm& perms) | |
11900 | { | |
f67539c2 TL |
11901 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11902 | if (!mref_reader.is_state_satisfied()) | |
11903 | return -CEPHFS_ENOTCONN; | |
7c673cae | 11904 | |
f67539c2 | 11905 | std::scoped_lock lock(client_lock); |
181888fb | 11906 | |
7c673cae FG |
11907 | int res = 0; |
11908 | unsigned mask = statx_to_mask(flags, want); | |
11909 | ||
94b18763 | 11910 | if (mask && !in->caps_issued_mask(mask, true)) |
7c673cae FG |
11911 | res = _ll_getattr(in, mask, perms); |
11912 | ||
11913 | if (res == 0) | |
11914 | fill_statx(in, mask, stx); | |
11fdf7f2 | 11915 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11916 | return res; |
11917 | } | |
11918 | ||
11919 | int Client::_ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
11920 | const UserPerm& perms, InodeRef *inp) | |
11921 | { | |
11922 | vinodeno_t vino = _get_vino(in); | |
11923 | ||
11fdf7f2 | 11924 | ldout(cct, 8) << __func__ << " " << vino << " mask " << hex << mask << dec |
7c673cae | 11925 | << dendl; |
11fdf7f2 | 11926 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
11927 | tout(cct) << vino.ino.val << std::endl; |
11928 | tout(cct) << stx->stx_mode << std::endl; | |
11929 | tout(cct) << stx->stx_uid << std::endl; | |
11930 | tout(cct) << stx->stx_gid << std::endl; | |
11931 | tout(cct) << stx->stx_size << std::endl; | |
11932 | tout(cct) << stx->stx_mtime << std::endl; | |
11933 | tout(cct) << stx->stx_atime << std::endl; | |
11934 | tout(cct) << stx->stx_btime << std::endl; | |
11935 | tout(cct) << mask << std::endl; | |
11936 | ||
11fdf7f2 | 11937 | if (!fuse_default_permissions) { |
7c673cae FG |
11938 | int res = may_setattr(in, stx, mask, perms); |
11939 | if (res < 0) | |
11940 | return res; | |
11941 | } | |
11942 | ||
11943 | mask &= ~(CEPH_SETATTR_MTIME_NOW | CEPH_SETATTR_ATIME_NOW); | |
11944 | ||
11945 | return __setattrx(in, stx, mask, perms, inp); | |
11946 | } | |
11947 | ||
11948 | int Client::ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
11949 | const UserPerm& perms) | |
11950 | { | |
f67539c2 TL |
11951 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11952 | if (!mref_reader.is_state_satisfied()) | |
11953 | return -CEPHFS_ENOTCONN; | |
181888fb | 11954 | |
f67539c2 | 11955 | std::scoped_lock lock(client_lock); |
181888fb | 11956 | |
7c673cae FG |
11957 | InodeRef target(in); |
11958 | int res = _ll_setattrx(in, stx, mask, perms, &target); | |
11959 | if (res == 0) { | |
11fdf7f2 | 11960 | ceph_assert(in == target.get()); |
7c673cae FG |
11961 | fill_statx(in, in->caps_issued(), stx); |
11962 | } | |
11963 | ||
11fdf7f2 | 11964 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11965 | return res; |
11966 | } | |
11967 | ||
11968 | int Client::ll_setattr(Inode *in, struct stat *attr, int mask, | |
11969 | const UserPerm& perms) | |
11970 | { | |
11971 | struct ceph_statx stx; | |
11972 | stat_to_statx(attr, &stx); | |
11973 | ||
f67539c2 TL |
11974 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11975 | if (!mref_reader.is_state_satisfied()) | |
11976 | return -CEPHFS_ENOTCONN; | |
181888fb | 11977 | |
f67539c2 | 11978 | std::scoped_lock lock(client_lock); |
181888fb | 11979 | |
7c673cae FG |
11980 | InodeRef target(in); |
11981 | int res = _ll_setattrx(in, &stx, mask, perms, &target); | |
11982 | if (res == 0) { | |
11fdf7f2 | 11983 | ceph_assert(in == target.get()); |
7c673cae FG |
11984 | fill_stat(in, attr); |
11985 | } | |
11986 | ||
11fdf7f2 | 11987 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11988 | return res; |
11989 | } | |
11990 | ||
11991 | ||
11992 | // ---------- | |
11993 | // xattrs | |
11994 | ||
11995 | int Client::getxattr(const char *path, const char *name, void *value, size_t size, | |
11996 | const UserPerm& perms) | |
11997 | { | |
f67539c2 TL |
11998 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
11999 | if (!mref_reader.is_state_satisfied()) | |
12000 | return -CEPHFS_ENOTCONN; | |
181888fb | 12001 | |
f67539c2 | 12002 | std::scoped_lock lock(client_lock); |
181888fb | 12003 | |
7c673cae FG |
12004 | InodeRef in; |
12005 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12006 | if (r < 0) | |
12007 | return r; | |
12008 | return _getxattr(in, name, value, size, perms); | |
12009 | } | |
12010 | ||
12011 | int Client::lgetxattr(const char *path, const char *name, void *value, size_t size, | |
12012 | const UserPerm& perms) | |
12013 | { | |
f67539c2 TL |
12014 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12015 | if (!mref_reader.is_state_satisfied()) | |
12016 | return -CEPHFS_ENOTCONN; | |
181888fb | 12017 | |
f67539c2 | 12018 | std::scoped_lock lock(client_lock); |
181888fb | 12019 | |
7c673cae FG |
12020 | InodeRef in; |
12021 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12022 | if (r < 0) | |
12023 | return r; | |
12024 | return _getxattr(in, name, value, size, perms); | |
12025 | } | |
12026 | ||
12027 | int Client::fgetxattr(int fd, const char *name, void *value, size_t size, | |
12028 | const UserPerm& perms) | |
12029 | { | |
f67539c2 TL |
12030 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12031 | if (!mref_reader.is_state_satisfied()) | |
12032 | return -CEPHFS_ENOTCONN; | |
181888fb | 12033 | |
f67539c2 | 12034 | std::scoped_lock lock(client_lock); |
181888fb | 12035 | |
7c673cae FG |
12036 | Fh *f = get_filehandle(fd); |
12037 | if (!f) | |
f67539c2 | 12038 | return -CEPHFS_EBADF; |
7c673cae FG |
12039 | return _getxattr(f->inode, name, value, size, perms); |
12040 | } | |
12041 | ||
12042 | int Client::listxattr(const char *path, char *list, size_t size, | |
12043 | const UserPerm& perms) | |
12044 | { | |
f67539c2 TL |
12045 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12046 | if (!mref_reader.is_state_satisfied()) | |
12047 | return -CEPHFS_ENOTCONN; | |
181888fb | 12048 | |
f67539c2 | 12049 | std::scoped_lock lock(client_lock); |
181888fb | 12050 | |
7c673cae FG |
12051 | InodeRef in; |
12052 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
12053 | if (r < 0) | |
12054 | return r; | |
12055 | return Client::_listxattr(in.get(), list, size, perms); | |
12056 | } | |
12057 | ||
12058 | int Client::llistxattr(const char *path, char *list, size_t size, | |
12059 | const UserPerm& perms) | |
12060 | { | |
f67539c2 TL |
12061 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12062 | if (!mref_reader.is_state_satisfied()) | |
12063 | return -CEPHFS_ENOTCONN; | |
181888fb | 12064 | |
f67539c2 | 12065 | std::scoped_lock lock(client_lock); |
181888fb | 12066 | |
7c673cae FG |
12067 | InodeRef in; |
12068 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
12069 | if (r < 0) | |
12070 | return r; | |
12071 | return Client::_listxattr(in.get(), list, size, perms); | |
12072 | } | |
12073 | ||
12074 | int Client::flistxattr(int fd, char *list, size_t size, const UserPerm& perms) | |
12075 | { | |
f67539c2 TL |
12076 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12077 | if (!mref_reader.is_state_satisfied()) | |
12078 | return -CEPHFS_ENOTCONN; | |
181888fb | 12079 | |
f67539c2 | 12080 | std::scoped_lock lock(client_lock); |
181888fb | 12081 | |
7c673cae FG |
12082 | Fh *f = get_filehandle(fd); |
12083 | if (!f) | |
f67539c2 | 12084 | return -CEPHFS_EBADF; |
7c673cae FG |
12085 | return Client::_listxattr(f->inode.get(), list, size, perms); |
12086 | } | |
12087 | ||
12088 | int Client::removexattr(const char *path, const char *name, | |
12089 | const UserPerm& perms) | |
12090 | { | |
f67539c2 TL |
12091 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12092 | if (!mref_reader.is_state_satisfied()) | |
12093 | return -CEPHFS_ENOTCONN; | |
181888fb | 12094 | |
f67539c2 | 12095 | std::scoped_lock lock(client_lock); |
181888fb | 12096 | |
7c673cae FG |
12097 | InodeRef in; |
12098 | int r = Client::path_walk(path, &in, perms, true); | |
12099 | if (r < 0) | |
12100 | return r; | |
12101 | return _removexattr(in, name, perms); | |
12102 | } | |
12103 | ||
12104 | int Client::lremovexattr(const char *path, const char *name, | |
12105 | const UserPerm& perms) | |
12106 | { | |
f67539c2 TL |
12107 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12108 | if (!mref_reader.is_state_satisfied()) | |
12109 | return -CEPHFS_ENOTCONN; | |
181888fb | 12110 | |
f67539c2 | 12111 | std::scoped_lock lock(client_lock); |
181888fb | 12112 | |
7c673cae FG |
12113 | InodeRef in; |
12114 | int r = Client::path_walk(path, &in, perms, false); | |
12115 | if (r < 0) | |
12116 | return r; | |
12117 | return _removexattr(in, name, perms); | |
12118 | } | |
12119 | ||
12120 | int Client::fremovexattr(int fd, const char *name, const UserPerm& perms) | |
12121 | { | |
f67539c2 TL |
12122 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12123 | if (!mref_reader.is_state_satisfied()) | |
12124 | return -CEPHFS_ENOTCONN; | |
181888fb | 12125 | |
f67539c2 | 12126 | std::scoped_lock lock(client_lock); |
181888fb | 12127 | |
7c673cae FG |
12128 | Fh *f = get_filehandle(fd); |
12129 | if (!f) | |
f67539c2 | 12130 | return -CEPHFS_EBADF; |
7c673cae FG |
12131 | return _removexattr(f->inode, name, perms); |
12132 | } | |
12133 | ||
12134 | int Client::setxattr(const char *path, const char *name, const void *value, | |
12135 | size_t size, int flags, const UserPerm& perms) | |
12136 | { | |
f67539c2 TL |
12137 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12138 | if (!mref_reader.is_state_satisfied()) | |
12139 | return -CEPHFS_ENOTCONN; | |
12140 | ||
7c673cae FG |
12141 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
12142 | ||
f67539c2 | 12143 | std::scoped_lock lock(client_lock); |
181888fb | 12144 | |
7c673cae FG |
12145 | InodeRef in; |
12146 | int r = Client::path_walk(path, &in, perms, true); | |
12147 | if (r < 0) | |
12148 | return r; | |
12149 | return _setxattr(in, name, value, size, flags, perms); | |
12150 | } | |
12151 | ||
12152 | int Client::lsetxattr(const char *path, const char *name, const void *value, | |
12153 | size_t size, int flags, const UserPerm& perms) | |
12154 | { | |
f67539c2 TL |
12155 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12156 | if (!mref_reader.is_state_satisfied()) | |
12157 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12158 | |
f67539c2 | 12159 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12160 | |
f67539c2 | 12161 | std::scoped_lock lock(client_lock); |
181888fb | 12162 | |
7c673cae FG |
12163 | InodeRef in; |
12164 | int r = Client::path_walk(path, &in, perms, false); | |
12165 | if (r < 0) | |
12166 | return r; | |
12167 | return _setxattr(in, name, value, size, flags, perms); | |
12168 | } | |
12169 | ||
12170 | int Client::fsetxattr(int fd, const char *name, const void *value, size_t size, | |
12171 | int flags, 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 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12178 | |
f67539c2 | 12179 | std::scoped_lock lock(client_lock); |
181888fb | 12180 | |
7c673cae FG |
12181 | Fh *f = get_filehandle(fd); |
12182 | if (!f) | |
f67539c2 | 12183 | return -CEPHFS_EBADF; |
7c673cae FG |
12184 | return _setxattr(f->inode, name, value, size, flags, perms); |
12185 | } | |
12186 | ||
12187 | int Client::_getxattr(Inode *in, const char *name, void *value, size_t size, | |
12188 | const UserPerm& perms) | |
12189 | { | |
12190 | int r; | |
12191 | ||
12192 | const VXattr *vxattr = _match_vxattr(in, name); | |
12193 | if (vxattr) { | |
f67539c2 | 12194 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12195 | |
12196 | // Do a force getattr to get the latest quota before returning | |
12197 | // a value to userspace. | |
28e407b8 AA |
12198 | int flags = 0; |
12199 | if (vxattr->flags & VXATTR_RSTAT) { | |
12200 | flags |= CEPH_STAT_RSTAT; | |
12201 | } | |
adb31ebb TL |
12202 | if (vxattr->flags & VXATTR_DIRSTAT) { |
12203 | flags |= CEPH_CAP_FILE_SHARED; | |
12204 | } | |
f67539c2 | 12205 | r = _getattr(in, flags | CEPH_STAT_CAP_XATTR, perms, true); |
7c673cae FG |
12206 | if (r != 0) { |
12207 | // Error from getattr! | |
12208 | return r; | |
12209 | } | |
12210 | ||
12211 | // call pointer-to-member function | |
12212 | char buf[256]; | |
12213 | if (!(vxattr->exists_cb && !(this->*(vxattr->exists_cb))(in))) { | |
12214 | r = (this->*(vxattr->getxattr_cb))(in, buf, sizeof(buf)); | |
12215 | } else { | |
f67539c2 | 12216 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12217 | } |
12218 | ||
12219 | if (size != 0) { | |
12220 | if (r > (int)size) { | |
f67539c2 | 12221 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12222 | } else if (r > 0) { |
12223 | memcpy(value, buf, r); | |
12224 | } | |
12225 | } | |
12226 | goto out; | |
12227 | } | |
12228 | ||
12229 | if (acl_type == NO_ACL && !strncmp(name, "system.", 7)) { | |
f67539c2 | 12230 | r = -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12231 | goto out; |
12232 | } | |
12233 | ||
12234 | r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
12235 | if (r == 0) { | |
12236 | string n(name); | |
f67539c2 | 12237 | r = -CEPHFS_ENODATA; |
7c673cae FG |
12238 | if (in->xattrs.count(n)) { |
12239 | r = in->xattrs[n].length(); | |
12240 | if (r > 0 && size != 0) { | |
12241 | if (size >= (unsigned)r) | |
12242 | memcpy(value, in->xattrs[n].c_str(), r); | |
12243 | else | |
f67539c2 | 12244 | r = -CEPHFS_ERANGE; |
7c673cae FG |
12245 | } |
12246 | } | |
12247 | } | |
12248 | out: | |
1adf2230 | 12249 | ldout(cct, 8) << "_getxattr(" << in->ino << ", \"" << name << "\", " << size << ") = " << r << dendl; |
7c673cae FG |
12250 | return r; |
12251 | } | |
12252 | ||
12253 | int Client::_getxattr(InodeRef &in, const char *name, void *value, size_t size, | |
12254 | const UserPerm& perms) | |
12255 | { | |
12256 | if (cct->_conf->client_permissions) { | |
12257 | int r = xattr_permission(in.get(), name, MAY_READ, perms); | |
12258 | if (r < 0) | |
12259 | return r; | |
12260 | } | |
12261 | return _getxattr(in.get(), name, value, size, perms); | |
12262 | } | |
12263 | ||
12264 | int Client::ll_getxattr(Inode *in, const char *name, void *value, | |
12265 | size_t size, const UserPerm& perms) | |
12266 | { | |
f67539c2 TL |
12267 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12268 | if (!mref_reader.is_state_satisfied()) | |
12269 | return -CEPHFS_ENOTCONN; | |
181888fb | 12270 | |
7c673cae FG |
12271 | vinodeno_t vino = _get_vino(in); |
12272 | ||
11fdf7f2 TL |
12273 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12274 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12275 | tout(cct) << vino.ino.val << std::endl; |
12276 | tout(cct) << name << std::endl; | |
12277 | ||
f67539c2 | 12278 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12279 | if (!fuse_default_permissions) { |
7c673cae FG |
12280 | int r = xattr_permission(in, name, MAY_READ, perms); |
12281 | if (r < 0) | |
12282 | return r; | |
12283 | } | |
12284 | ||
12285 | return _getxattr(in, name, value, size, perms); | |
12286 | } | |
12287 | ||
12288 | int Client::_listxattr(Inode *in, char *name, size_t size, | |
12289 | const UserPerm& perms) | |
12290 | { | |
81eedcae | 12291 | bool len_only = (size == 0); |
7c673cae | 12292 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); |
81eedcae TL |
12293 | if (r != 0) { |
12294 | goto out; | |
12295 | } | |
7c673cae | 12296 | |
81eedcae | 12297 | r = 0; |
f67539c2 TL |
12298 | for ([[maybe_unused]] const auto &[xattr_name, xattr_value_bl] : in->xattrs) { |
12299 | if (xattr_name.rfind("ceph.", 0) == 0) { | |
12300 | continue; | |
12301 | } | |
12302 | ||
12303 | size_t this_len = xattr_name.length() + 1; | |
81eedcae TL |
12304 | r += this_len; |
12305 | if (len_only) | |
12306 | continue; | |
7c673cae | 12307 | |
81eedcae | 12308 | if (this_len > size) { |
f67539c2 | 12309 | r = -CEPHFS_ERANGE; |
81eedcae TL |
12310 | goto out; |
12311 | } | |
12312 | ||
f67539c2 | 12313 | memcpy(name, xattr_name.c_str(), this_len); |
81eedcae TL |
12314 | name += this_len; |
12315 | size -= this_len; | |
12316 | } | |
81eedcae | 12317 | out: |
11fdf7f2 | 12318 | ldout(cct, 8) << __func__ << "(" << in->ino << ", " << size << ") = " << r << dendl; |
7c673cae FG |
12319 | return r; |
12320 | } | |
12321 | ||
12322 | int Client::ll_listxattr(Inode *in, char *names, size_t size, | |
12323 | const UserPerm& perms) | |
12324 | { | |
f67539c2 TL |
12325 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12326 | if (!mref_reader.is_state_satisfied()) | |
12327 | return -CEPHFS_ENOTCONN; | |
181888fb | 12328 | |
7c673cae FG |
12329 | vinodeno_t vino = _get_vino(in); |
12330 | ||
11fdf7f2 TL |
12331 | ldout(cct, 3) << __func__ << " " << vino << " size " << size << dendl; |
12332 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12333 | tout(cct) << vino.ino.val << std::endl; |
12334 | tout(cct) << size << std::endl; | |
12335 | ||
f67539c2 | 12336 | std::scoped_lock lock(client_lock); |
7c673cae FG |
12337 | return _listxattr(in, names, size, perms); |
12338 | } | |
12339 | ||
12340 | int Client::_do_setxattr(Inode *in, const char *name, const void *value, | |
12341 | size_t size, int flags, const UserPerm& perms) | |
12342 | { | |
12343 | ||
12344 | int xattr_flags = 0; | |
12345 | if (!value) | |
12346 | xattr_flags |= CEPH_XATTR_REMOVE; | |
12347 | if (flags & XATTR_CREATE) | |
12348 | xattr_flags |= CEPH_XATTR_CREATE; | |
12349 | if (flags & XATTR_REPLACE) | |
12350 | xattr_flags |= CEPH_XATTR_REPLACE; | |
12351 | ||
12352 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETXATTR); | |
12353 | filepath path; | |
12354 | in->make_nosnap_relative_path(path); | |
12355 | req->set_filepath(path); | |
12356 | req->set_string2(name); | |
12357 | req->set_inode(in); | |
12358 | req->head.args.setxattr.flags = xattr_flags; | |
12359 | ||
12360 | bufferlist bl; | |
11fdf7f2 | 12361 | assert (value || size == 0); |
7c673cae FG |
12362 | bl.append((const char*)value, size); |
12363 | req->set_data(bl); | |
12364 | ||
12365 | int res = make_request(req, perms); | |
12366 | ||
12367 | trim_cache(); | |
11fdf7f2 | 12368 | ldout(cct, 3) << __func__ << "(" << in->ino << ", \"" << name << "\") = " << |
7c673cae FG |
12369 | res << dendl; |
12370 | return res; | |
12371 | } | |
12372 | ||
12373 | int Client::_setxattr(Inode *in, const char *name, const void *value, | |
12374 | size_t size, int flags, const UserPerm& perms) | |
12375 | { | |
12376 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12377 | return -CEPHFS_EROFS; |
7c673cae FG |
12378 | } |
12379 | ||
f6b5b4d7 TL |
12380 | if (size == 0) { |
12381 | value = ""; | |
12382 | } else if (value == NULL) { | |
f67539c2 | 12383 | return -CEPHFS_EINVAL; |
f6b5b4d7 TL |
12384 | } |
12385 | ||
7c673cae FG |
12386 | bool posix_acl_xattr = false; |
12387 | if (acl_type == POSIX_ACL) | |
12388 | posix_acl_xattr = !strncmp(name, "system.", 7); | |
12389 | ||
12390 | if (strncmp(name, "user.", 5) && | |
12391 | strncmp(name, "security.", 9) && | |
12392 | strncmp(name, "trusted.", 8) && | |
12393 | strncmp(name, "ceph.", 5) && | |
12394 | !posix_acl_xattr) | |
f67539c2 | 12395 | return -CEPHFS_EOPNOTSUPP; |
7c673cae | 12396 | |
11fdf7f2 TL |
12397 | bool check_realm = false; |
12398 | ||
7c673cae FG |
12399 | if (posix_acl_xattr) { |
12400 | if (!strcmp(name, ACL_EA_ACCESS)) { | |
12401 | mode_t new_mode = in->mode; | |
12402 | if (value) { | |
12403 | int ret = posix_acl_equiv_mode(value, size, &new_mode); | |
12404 | if (ret < 0) | |
12405 | return ret; | |
12406 | if (ret == 0) { | |
12407 | value = NULL; | |
12408 | size = 0; | |
12409 | } | |
12410 | if (new_mode != in->mode) { | |
12411 | struct ceph_statx stx; | |
12412 | stx.stx_mode = new_mode; | |
12413 | ret = _do_setattr(in, &stx, CEPH_SETATTR_MODE, perms, NULL); | |
12414 | if (ret < 0) | |
12415 | return ret; | |
12416 | } | |
12417 | } | |
12418 | } else if (!strcmp(name, ACL_EA_DEFAULT)) { | |
12419 | if (value) { | |
12420 | if (!S_ISDIR(in->mode)) | |
f67539c2 | 12421 | return -CEPHFS_EACCES; |
7c673cae FG |
12422 | int ret = posix_acl_check(value, size); |
12423 | if (ret < 0) | |
f67539c2 | 12424 | return -CEPHFS_EINVAL; |
7c673cae FG |
12425 | if (ret == 0) { |
12426 | value = NULL; | |
12427 | size = 0; | |
12428 | } | |
12429 | } | |
12430 | } else { | |
f67539c2 | 12431 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12432 | } |
12433 | } else { | |
12434 | const VXattr *vxattr = _match_vxattr(in, name); | |
11fdf7f2 TL |
12435 | if (vxattr) { |
12436 | if (vxattr->readonly) | |
f67539c2 | 12437 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12438 | if (vxattr->name.compare(0, 10, "ceph.quota") == 0 && value) |
12439 | check_realm = true; | |
12440 | } | |
7c673cae FG |
12441 | } |
12442 | ||
11fdf7f2 TL |
12443 | int ret = _do_setxattr(in, name, value, size, flags, perms); |
12444 | if (ret >= 0 && check_realm) { | |
12445 | // check if snaprealm was created for quota inode | |
12446 | if (in->quota.is_enable() && | |
12447 | !(in->snaprealm && in->snaprealm->ino == in->ino)) | |
f67539c2 | 12448 | ret = -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
12449 | } |
12450 | ||
12451 | return ret; | |
7c673cae FG |
12452 | } |
12453 | ||
12454 | int Client::_setxattr(InodeRef &in, const char *name, const void *value, | |
12455 | size_t size, int flags, const UserPerm& perms) | |
12456 | { | |
12457 | if (cct->_conf->client_permissions) { | |
12458 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12459 | if (r < 0) | |
12460 | return r; | |
12461 | } | |
12462 | return _setxattr(in.get(), name, value, size, flags, perms); | |
12463 | } | |
12464 | ||
12465 | int Client::_setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap) | |
12466 | { | |
12467 | string tmp; | |
12468 | if (name == "layout") { | |
12469 | string::iterator begin = value.begin(); | |
12470 | string::iterator end = value.end(); | |
12471 | keys_and_values<string::iterator> p; // create instance of parser | |
12472 | std::map<string, string> m; // map to receive results | |
12473 | if (!qi::parse(begin, end, p, m)) { // returns true if successful | |
f67539c2 | 12474 | return -CEPHFS_EINVAL; |
7c673cae FG |
12475 | } |
12476 | if (begin != end) | |
f67539c2 | 12477 | return -CEPHFS_EINVAL; |
7c673cae FG |
12478 | for (map<string,string>::iterator q = m.begin(); q != m.end(); ++q) { |
12479 | if (q->first == "pool") { | |
12480 | tmp = q->second; | |
12481 | break; | |
12482 | } | |
12483 | } | |
12484 | } else if (name == "layout.pool") { | |
12485 | tmp = value; | |
12486 | } | |
12487 | ||
12488 | if (tmp.length()) { | |
12489 | int64_t pool; | |
12490 | try { | |
12491 | pool = boost::lexical_cast<unsigned>(tmp); | |
12492 | if (!osdmap->have_pg_pool(pool)) | |
f67539c2 | 12493 | return -CEPHFS_ENOENT; |
7c673cae FG |
12494 | } catch (boost::bad_lexical_cast const&) { |
12495 | pool = osdmap->lookup_pg_pool_name(tmp); | |
12496 | if (pool < 0) { | |
f67539c2 | 12497 | return -CEPHFS_ENOENT; |
7c673cae FG |
12498 | } |
12499 | } | |
12500 | } | |
12501 | ||
12502 | return 0; | |
12503 | } | |
12504 | ||
12505 | void Client::_setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t size) | |
12506 | { | |
12507 | // For setting pool of layout, MetaRequest need osdmap epoch. | |
12508 | // There is a race which create a new data pool but client and mds both don't have. | |
12509 | // Make client got the latest osdmap which make mds quickly judge whether get newer osdmap. | |
f67539c2 | 12510 | ldout(cct, 15) << __func__ << ": name = " << name << dendl; |
7c673cae FG |
12511 | if (strcmp(name, "ceph.file.layout.pool") == 0 || strcmp(name, "ceph.dir.layout.pool") == 0 || |
12512 | strcmp(name, "ceph.file.layout") == 0 || strcmp(name, "ceph.dir.layout") == 0) { | |
12513 | string rest(strstr(name, "layout")); | |
12514 | string v((const char*)value, size); | |
12515 | int r = objecter->with_osdmap([&](const OSDMap& o) { | |
12516 | return _setxattr_check_data_pool(rest, v, &o); | |
12517 | }); | |
12518 | ||
f67539c2 TL |
12519 | if (r == -CEPHFS_ENOENT) { |
12520 | bs::error_code ec; | |
12521 | ldout(cct, 20) << __func__ << ": waiting for latest osdmap" << dendl; | |
12522 | objecter->wait_for_latest_osdmap(ca::use_blocked[ec]); | |
12523 | ldout(cct, 20) << __func__ << ": got latest osdmap: " << ec << dendl; | |
7c673cae FG |
12524 | } |
12525 | } | |
12526 | } | |
12527 | ||
12528 | int Client::ll_setxattr(Inode *in, const char *name, const void *value, | |
12529 | size_t size, int flags, const UserPerm& perms) | |
12530 | { | |
f67539c2 TL |
12531 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12532 | if (!mref_reader.is_state_satisfied()) | |
12533 | return -CEPHFS_ENOTCONN; | |
7c673cae | 12534 | |
f67539c2 | 12535 | _setxattr_maybe_wait_for_osdmap(name, value, size); |
181888fb | 12536 | |
7c673cae FG |
12537 | vinodeno_t vino = _get_vino(in); |
12538 | ||
11fdf7f2 TL |
12539 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
12540 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
12541 | tout(cct) << vino.ino.val << std::endl; |
12542 | tout(cct) << name << std::endl; | |
12543 | ||
f67539c2 | 12544 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12545 | if (!fuse_default_permissions) { |
7c673cae FG |
12546 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12547 | if (r < 0) | |
12548 | return r; | |
12549 | } | |
12550 | return _setxattr(in, name, value, size, flags, perms); | |
12551 | } | |
12552 | ||
12553 | int Client::_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12554 | { | |
12555 | if (in->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12556 | return -CEPHFS_EROFS; |
7c673cae FG |
12557 | } |
12558 | ||
12559 | // same xattrs supported by kernel client | |
12560 | if (strncmp(name, "user.", 5) && | |
12561 | strncmp(name, "system.", 7) && | |
12562 | strncmp(name, "security.", 9) && | |
12563 | strncmp(name, "trusted.", 8) && | |
12564 | strncmp(name, "ceph.", 5)) | |
f67539c2 | 12565 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12566 | |
12567 | const VXattr *vxattr = _match_vxattr(in, name); | |
12568 | if (vxattr && vxattr->readonly) | |
f67539c2 | 12569 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
12570 | |
12571 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_RMXATTR); | |
12572 | filepath path; | |
12573 | in->make_nosnap_relative_path(path); | |
12574 | req->set_filepath(path); | |
12575 | req->set_filepath2(name); | |
12576 | req->set_inode(in); | |
12577 | ||
12578 | int res = make_request(req, perms); | |
12579 | ||
12580 | trim_cache(); | |
1adf2230 | 12581 | ldout(cct, 8) << "_removexattr(" << in->ino << ", \"" << name << "\") = " << res << dendl; |
7c673cae FG |
12582 | return res; |
12583 | } | |
12584 | ||
12585 | int Client::_removexattr(InodeRef &in, const char *name, const UserPerm& perms) | |
12586 | { | |
12587 | if (cct->_conf->client_permissions) { | |
12588 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
12589 | if (r < 0) | |
12590 | return r; | |
12591 | } | |
12592 | return _removexattr(in.get(), name, perms); | |
12593 | } | |
12594 | ||
12595 | int Client::ll_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
12596 | { | |
f67539c2 TL |
12597 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12598 | if (!mref_reader.is_state_satisfied()) | |
12599 | return -CEPHFS_ENOTCONN; | |
181888fb | 12600 | |
7c673cae FG |
12601 | vinodeno_t vino = _get_vino(in); |
12602 | ||
12603 | ldout(cct, 3) << "ll_removexattr " << vino << " " << name << dendl; | |
12604 | tout(cct) << "ll_removexattr" << std::endl; | |
12605 | tout(cct) << vino.ino.val << std::endl; | |
12606 | tout(cct) << name << std::endl; | |
12607 | ||
f67539c2 | 12608 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 12609 | if (!fuse_default_permissions) { |
7c673cae FG |
12610 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
12611 | if (r < 0) | |
12612 | return r; | |
12613 | } | |
12614 | ||
12615 | return _removexattr(in, name, perms); | |
12616 | } | |
12617 | ||
12618 | bool Client::_vxattrcb_quota_exists(Inode *in) | |
12619 | { | |
11fdf7f2 | 12620 | return in->quota.is_enable() && |
f6b5b4d7 TL |
12621 | (in->snapid != CEPH_NOSNAP || |
12622 | (in->snaprealm && in->snaprealm->ino == in->ino)); | |
7c673cae FG |
12623 | } |
12624 | size_t Client::_vxattrcb_quota(Inode *in, char *val, size_t size) | |
12625 | { | |
12626 | return snprintf(val, size, | |
12627 | "max_bytes=%lld max_files=%lld", | |
12628 | (long long int)in->quota.max_bytes, | |
12629 | (long long int)in->quota.max_files); | |
12630 | } | |
12631 | size_t Client::_vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size) | |
12632 | { | |
12633 | return snprintf(val, size, "%lld", (long long int)in->quota.max_bytes); | |
12634 | } | |
12635 | size_t Client::_vxattrcb_quota_max_files(Inode *in, char *val, size_t size) | |
12636 | { | |
12637 | return snprintf(val, size, "%lld", (long long int)in->quota.max_files); | |
12638 | } | |
12639 | ||
12640 | bool Client::_vxattrcb_layout_exists(Inode *in) | |
12641 | { | |
12642 | return in->layout != file_layout_t(); | |
12643 | } | |
12644 | size_t Client::_vxattrcb_layout(Inode *in, char *val, size_t size) | |
12645 | { | |
12646 | int r = snprintf(val, size, | |
11fdf7f2 | 12647 | "stripe_unit=%llu stripe_count=%llu object_size=%llu pool=", |
7c673cae FG |
12648 | (unsigned long long)in->layout.stripe_unit, |
12649 | (unsigned long long)in->layout.stripe_count, | |
12650 | (unsigned long long)in->layout.object_size); | |
12651 | objecter->with_osdmap([&](const OSDMap& o) { | |
12652 | if (o.have_pg_pool(in->layout.pool_id)) | |
12653 | r += snprintf(val + r, size - r, "%s", | |
12654 | o.get_pool_name(in->layout.pool_id).c_str()); | |
12655 | else | |
12656 | r += snprintf(val + r, size - r, "%" PRIu64, | |
12657 | (uint64_t)in->layout.pool_id); | |
12658 | }); | |
12659 | if (in->layout.pool_ns.length()) | |
12660 | r += snprintf(val + r, size - r, " pool_namespace=%s", | |
12661 | in->layout.pool_ns.c_str()); | |
12662 | return r; | |
12663 | } | |
12664 | size_t Client::_vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size) | |
12665 | { | |
11fdf7f2 | 12666 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_unit); |
7c673cae FG |
12667 | } |
12668 | size_t Client::_vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size) | |
12669 | { | |
11fdf7f2 | 12670 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_count); |
7c673cae FG |
12671 | } |
12672 | size_t Client::_vxattrcb_layout_object_size(Inode *in, char *val, size_t size) | |
12673 | { | |
11fdf7f2 | 12674 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.object_size); |
7c673cae FG |
12675 | } |
12676 | size_t Client::_vxattrcb_layout_pool(Inode *in, char *val, size_t size) | |
12677 | { | |
12678 | size_t r; | |
12679 | objecter->with_osdmap([&](const OSDMap& o) { | |
12680 | if (o.have_pg_pool(in->layout.pool_id)) | |
12681 | r = snprintf(val, size, "%s", o.get_pool_name( | |
12682 | in->layout.pool_id).c_str()); | |
12683 | else | |
12684 | r = snprintf(val, size, "%" PRIu64, (uint64_t)in->layout.pool_id); | |
12685 | }); | |
12686 | return r; | |
12687 | } | |
12688 | size_t Client::_vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size) | |
12689 | { | |
12690 | return snprintf(val, size, "%s", in->layout.pool_ns.c_str()); | |
12691 | } | |
12692 | size_t Client::_vxattrcb_dir_entries(Inode *in, char *val, size_t size) | |
12693 | { | |
11fdf7f2 | 12694 | return snprintf(val, size, "%llu", (unsigned long long)(in->dirstat.nfiles + in->dirstat.nsubdirs)); |
7c673cae FG |
12695 | } |
12696 | size_t Client::_vxattrcb_dir_files(Inode *in, char *val, size_t size) | |
12697 | { | |
11fdf7f2 | 12698 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nfiles); |
7c673cae FG |
12699 | } |
12700 | size_t Client::_vxattrcb_dir_subdirs(Inode *in, char *val, size_t size) | |
12701 | { | |
11fdf7f2 | 12702 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nsubdirs); |
7c673cae FG |
12703 | } |
12704 | size_t Client::_vxattrcb_dir_rentries(Inode *in, char *val, size_t size) | |
12705 | { | |
11fdf7f2 | 12706 | return snprintf(val, size, "%llu", (unsigned long long)(in->rstat.rfiles + in->rstat.rsubdirs)); |
7c673cae FG |
12707 | } |
12708 | size_t Client::_vxattrcb_dir_rfiles(Inode *in, char *val, size_t size) | |
12709 | { | |
11fdf7f2 | 12710 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rfiles); |
7c673cae FG |
12711 | } |
12712 | size_t Client::_vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size) | |
12713 | { | |
11fdf7f2 | 12714 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsubdirs); |
7c673cae | 12715 | } |
f67539c2 TL |
12716 | size_t Client::_vxattrcb_dir_rsnaps(Inode *in, char *val, size_t size) |
12717 | { | |
12718 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsnaps); | |
12719 | } | |
7c673cae FG |
12720 | size_t Client::_vxattrcb_dir_rbytes(Inode *in, char *val, size_t size) |
12721 | { | |
11fdf7f2 | 12722 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rbytes); |
7c673cae FG |
12723 | } |
12724 | size_t Client::_vxattrcb_dir_rctime(Inode *in, char *val, size_t size) | |
12725 | { | |
81eedcae | 12726 | return snprintf(val, size, "%ld.%09ld", (long)in->rstat.rctime.sec(), |
7c673cae FG |
12727 | (long)in->rstat.rctime.nsec()); |
12728 | } | |
11fdf7f2 TL |
12729 | bool Client::_vxattrcb_dir_pin_exists(Inode *in) |
12730 | { | |
f67539c2 | 12731 | return in->dir_pin != -CEPHFS_ENODATA; |
11fdf7f2 TL |
12732 | } |
12733 | size_t Client::_vxattrcb_dir_pin(Inode *in, char *val, size_t size) | |
12734 | { | |
12735 | return snprintf(val, size, "%ld", (long)in->dir_pin); | |
12736 | } | |
7c673cae | 12737 | |
81eedcae TL |
12738 | bool Client::_vxattrcb_snap_btime_exists(Inode *in) |
12739 | { | |
12740 | return !in->snap_btime.is_zero(); | |
12741 | } | |
12742 | ||
12743 | size_t Client::_vxattrcb_snap_btime(Inode *in, char *val, size_t size) | |
12744 | { | |
12745 | return snprintf(val, size, "%llu.%09lu", | |
12746 | (long long unsigned)in->snap_btime.sec(), | |
12747 | (long unsigned)in->snap_btime.nsec()); | |
12748 | } | |
12749 | ||
f67539c2 TL |
12750 | bool Client::_vxattrcb_mirror_info_exists(Inode *in) |
12751 | { | |
12752 | // checking one of the xattrs would suffice | |
12753 | return in->xattrs.count("ceph.mirror.info.cluster_id") != 0; | |
12754 | } | |
12755 | ||
12756 | size_t Client::_vxattrcb_mirror_info(Inode *in, char *val, size_t size) | |
12757 | { | |
12758 | return snprintf(val, size, "cluster_id=%.*s fs_id=%.*s", | |
12759 | in->xattrs["ceph.mirror.info.cluster_id"].length(), | |
12760 | in->xattrs["ceph.mirror.info.cluster_id"].c_str(), | |
12761 | in->xattrs["ceph.mirror.info.fs_id"].length(), | |
12762 | in->xattrs["ceph.mirror.info.fs_id"].c_str()); | |
12763 | } | |
12764 | ||
adb31ebb TL |
12765 | size_t Client::_vxattrcb_cluster_fsid(Inode *in, char *val, size_t size) |
12766 | { | |
12767 | return snprintf(val, size, "%s", monclient->get_fsid().to_string().c_str()); | |
12768 | } | |
12769 | ||
12770 | size_t Client::_vxattrcb_client_id(Inode *in, char *val, size_t size) | |
12771 | { | |
12772 | auto name = messenger->get_myname(); | |
12773 | return snprintf(val, size, "%s%ld", name.type_str(), name.num()); | |
12774 | } | |
12775 | ||
7c673cae FG |
12776 | #define CEPH_XATTR_NAME(_type, _name) "ceph." #_type "." #_name |
12777 | #define CEPH_XATTR_NAME2(_type, _name, _name2) "ceph." #_type "." #_name "." #_name2 | |
12778 | ||
adb31ebb | 12779 | #define XATTR_NAME_CEPH(_type, _name, _flags) \ |
28e407b8 AA |
12780 | { \ |
12781 | name: CEPH_XATTR_NAME(_type, _name), \ | |
12782 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
12783 | readonly: true, \ | |
28e407b8 AA |
12784 | exists_cb: NULL, \ |
12785 | flags: _flags, \ | |
7c673cae FG |
12786 | } |
12787 | #define XATTR_LAYOUT_FIELD(_type, _name, _field) \ | |
12788 | { \ | |
12789 | name: CEPH_XATTR_NAME2(_type, _name, _field), \ | |
12790 | getxattr_cb: &Client::_vxattrcb_ ## _name ## _ ## _field, \ | |
12791 | readonly: false, \ | |
7c673cae | 12792 | exists_cb: &Client::_vxattrcb_layout_exists, \ |
28e407b8 | 12793 | flags: 0, \ |
7c673cae FG |
12794 | } |
12795 | #define XATTR_QUOTA_FIELD(_type, _name) \ | |
12796 | { \ | |
12797 | name: CEPH_XATTR_NAME(_type, _name), \ | |
12798 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
12799 | readonly: false, \ | |
7c673cae | 12800 | exists_cb: &Client::_vxattrcb_quota_exists, \ |
28e407b8 | 12801 | flags: 0, \ |
7c673cae FG |
12802 | } |
12803 | ||
12804 | const Client::VXattr Client::_dir_vxattrs[] = { | |
12805 | { | |
12806 | name: "ceph.dir.layout", | |
12807 | getxattr_cb: &Client::_vxattrcb_layout, | |
12808 | readonly: false, | |
7c673cae | 12809 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 12810 | flags: 0, |
7c673cae FG |
12811 | }, |
12812 | XATTR_LAYOUT_FIELD(dir, layout, stripe_unit), | |
12813 | XATTR_LAYOUT_FIELD(dir, layout, stripe_count), | |
12814 | XATTR_LAYOUT_FIELD(dir, layout, object_size), | |
12815 | XATTR_LAYOUT_FIELD(dir, layout, pool), | |
12816 | XATTR_LAYOUT_FIELD(dir, layout, pool_namespace), | |
adb31ebb TL |
12817 | XATTR_NAME_CEPH(dir, entries, VXATTR_DIRSTAT), |
12818 | XATTR_NAME_CEPH(dir, files, VXATTR_DIRSTAT), | |
12819 | XATTR_NAME_CEPH(dir, subdirs, VXATTR_DIRSTAT), | |
12820 | XATTR_NAME_CEPH(dir, rentries, VXATTR_RSTAT), | |
12821 | XATTR_NAME_CEPH(dir, rfiles, VXATTR_RSTAT), | |
12822 | XATTR_NAME_CEPH(dir, rsubdirs, VXATTR_RSTAT), | |
f67539c2 | 12823 | XATTR_NAME_CEPH(dir, rsnaps, VXATTR_RSTAT), |
adb31ebb TL |
12824 | XATTR_NAME_CEPH(dir, rbytes, VXATTR_RSTAT), |
12825 | XATTR_NAME_CEPH(dir, rctime, VXATTR_RSTAT), | |
7c673cae FG |
12826 | { |
12827 | name: "ceph.quota", | |
12828 | getxattr_cb: &Client::_vxattrcb_quota, | |
12829 | readonly: false, | |
7c673cae | 12830 | exists_cb: &Client::_vxattrcb_quota_exists, |
28e407b8 | 12831 | flags: 0, |
7c673cae FG |
12832 | }, |
12833 | XATTR_QUOTA_FIELD(quota, max_bytes), | |
12834 | XATTR_QUOTA_FIELD(quota, max_files), | |
11fdf7f2 TL |
12835 | { |
12836 | name: "ceph.dir.pin", | |
12837 | getxattr_cb: &Client::_vxattrcb_dir_pin, | |
12838 | readonly: false, | |
11fdf7f2 TL |
12839 | exists_cb: &Client::_vxattrcb_dir_pin_exists, |
12840 | flags: 0, | |
12841 | }, | |
81eedcae TL |
12842 | { |
12843 | name: "ceph.snap.btime", | |
12844 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
12845 | readonly: true, | |
81eedcae TL |
12846 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
12847 | flags: 0, | |
12848 | }, | |
f67539c2 TL |
12849 | { |
12850 | name: "ceph.mirror.info", | |
12851 | getxattr_cb: &Client::_vxattrcb_mirror_info, | |
12852 | readonly: false, | |
12853 | exists_cb: &Client::_vxattrcb_mirror_info_exists, | |
12854 | flags: 0, | |
12855 | }, | |
7c673cae FG |
12856 | { name: "" } /* Required table terminator */ |
12857 | }; | |
12858 | ||
12859 | const Client::VXattr Client::_file_vxattrs[] = { | |
12860 | { | |
12861 | name: "ceph.file.layout", | |
12862 | getxattr_cb: &Client::_vxattrcb_layout, | |
12863 | readonly: false, | |
7c673cae | 12864 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 12865 | flags: 0, |
7c673cae FG |
12866 | }, |
12867 | XATTR_LAYOUT_FIELD(file, layout, stripe_unit), | |
12868 | XATTR_LAYOUT_FIELD(file, layout, stripe_count), | |
12869 | XATTR_LAYOUT_FIELD(file, layout, object_size), | |
12870 | XATTR_LAYOUT_FIELD(file, layout, pool), | |
12871 | XATTR_LAYOUT_FIELD(file, layout, pool_namespace), | |
81eedcae TL |
12872 | { |
12873 | name: "ceph.snap.btime", | |
12874 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
12875 | readonly: true, | |
81eedcae TL |
12876 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
12877 | flags: 0, | |
12878 | }, | |
7c673cae FG |
12879 | { name: "" } /* Required table terminator */ |
12880 | }; | |
12881 | ||
adb31ebb TL |
12882 | const Client::VXattr Client::_common_vxattrs[] = { |
12883 | { | |
12884 | name: "ceph.cluster_fsid", | |
12885 | getxattr_cb: &Client::_vxattrcb_cluster_fsid, | |
12886 | readonly: true, | |
12887 | exists_cb: nullptr, | |
12888 | flags: 0, | |
12889 | }, | |
12890 | { | |
12891 | name: "ceph.client_id", | |
12892 | getxattr_cb: &Client::_vxattrcb_client_id, | |
12893 | readonly: true, | |
12894 | exists_cb: nullptr, | |
12895 | flags: 0, | |
12896 | }, | |
12897 | { name: "" } /* Required table terminator */ | |
12898 | }; | |
12899 | ||
7c673cae FG |
12900 | const Client::VXattr *Client::_get_vxattrs(Inode *in) |
12901 | { | |
12902 | if (in->is_dir()) | |
12903 | return _dir_vxattrs; | |
12904 | else if (in->is_file()) | |
12905 | return _file_vxattrs; | |
12906 | return NULL; | |
12907 | } | |
12908 | ||
12909 | const Client::VXattr *Client::_match_vxattr(Inode *in, const char *name) | |
12910 | { | |
12911 | if (strncmp(name, "ceph.", 5) == 0) { | |
12912 | const VXattr *vxattr = _get_vxattrs(in); | |
12913 | if (vxattr) { | |
12914 | while (!vxattr->name.empty()) { | |
12915 | if (vxattr->name == name) | |
12916 | return vxattr; | |
12917 | vxattr++; | |
12918 | } | |
12919 | } | |
adb31ebb TL |
12920 | |
12921 | // for common vxattrs | |
12922 | vxattr = _common_vxattrs; | |
12923 | while (!vxattr->name.empty()) { | |
12924 | if (vxattr->name == name) | |
12925 | return vxattr; | |
12926 | vxattr++; | |
12927 | } | |
7c673cae | 12928 | } |
adb31ebb | 12929 | |
7c673cae FG |
12930 | return NULL; |
12931 | } | |
12932 | ||
7c673cae FG |
12933 | int Client::ll_readlink(Inode *in, char *buf, size_t buflen, const UserPerm& perms) |
12934 | { | |
f67539c2 TL |
12935 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
12936 | if (!mref_reader.is_state_satisfied()) | |
12937 | return -CEPHFS_ENOTCONN; | |
181888fb | 12938 | |
7c673cae FG |
12939 | vinodeno_t vino = _get_vino(in); |
12940 | ||
12941 | ldout(cct, 3) << "ll_readlink " << vino << dendl; | |
12942 | tout(cct) << "ll_readlink" << std::endl; | |
12943 | tout(cct) << vino.ino.val << std::endl; | |
12944 | ||
f67539c2 | 12945 | std::scoped_lock lock(client_lock); |
11fdf7f2 TL |
12946 | for (auto dn : in->dentries) { |
12947 | touch_dn(dn); | |
7c673cae FG |
12948 | } |
12949 | ||
12950 | int r = _readlink(in, buf, buflen); // FIXME: no permission checking! | |
12951 | ldout(cct, 3) << "ll_readlink " << vino << " = " << r << dendl; | |
12952 | return r; | |
12953 | } | |
12954 | ||
12955 | int Client::_mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev, | |
12956 | const UserPerm& perms, InodeRef *inp) | |
12957 | { | |
1adf2230 | 12958 | ldout(cct, 8) << "_mknod(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
12959 | << mode << dec << ", " << rdev << ", uid " << perms.uid() |
12960 | << ", gid " << perms.gid() << ")" << dendl; | |
12961 | ||
12962 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 12963 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
12964 | |
12965 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 12966 | return -CEPHFS_EROFS; |
7c673cae FG |
12967 | } |
12968 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 12969 | return -CEPHFS_EDQUOT; |
7c673cae FG |
12970 | } |
12971 | ||
12972 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_MKNOD); | |
12973 | ||
12974 | filepath path; | |
12975 | dir->make_nosnap_relative_path(path); | |
12976 | path.push_dentry(name); | |
12977 | req->set_filepath(path); | |
12978 | req->set_inode(dir); | |
12979 | req->head.args.mknod.rdev = rdev; | |
12980 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12981 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12982 | ||
12983 | bufferlist xattrs_bl; | |
12984 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
12985 | if (res < 0) | |
12986 | goto fail; | |
12987 | req->head.args.mknod.mode = mode; | |
12988 | if (xattrs_bl.length() > 0) | |
12989 | req->set_data(xattrs_bl); | |
12990 | ||
12991 | Dentry *de; | |
12992 | res = get_or_create(dir, name, &de); | |
12993 | if (res < 0) | |
12994 | goto fail; | |
12995 | req->set_dentry(de); | |
12996 | ||
12997 | res = make_request(req, perms, inp); | |
12998 | ||
12999 | trim_cache(); | |
13000 | ||
1adf2230 | 13001 | ldout(cct, 8) << "mknod(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13002 | return res; |
13003 | ||
13004 | fail: | |
13005 | put_request(req); | |
13006 | return res; | |
13007 | } | |
13008 | ||
13009 | int Client::ll_mknod(Inode *parent, const char *name, mode_t mode, | |
13010 | dev_t rdev, struct stat *attr, Inode **out, | |
13011 | const UserPerm& perms) | |
13012 | { | |
f67539c2 TL |
13013 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13014 | if (!mref_reader.is_state_satisfied()) | |
13015 | return -CEPHFS_ENOTCONN; | |
181888fb | 13016 | |
7c673cae FG |
13017 | vinodeno_t vparent = _get_vino(parent); |
13018 | ||
13019 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name << dendl; | |
13020 | tout(cct) << "ll_mknod" << std::endl; | |
13021 | tout(cct) << vparent.ino.val << std::endl; | |
13022 | tout(cct) << name << std::endl; | |
13023 | tout(cct) << mode << std::endl; | |
13024 | tout(cct) << rdev << std::endl; | |
13025 | ||
f67539c2 | 13026 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 13027 | if (!fuse_default_permissions) { |
7c673cae FG |
13028 | int r = may_create(parent, perms); |
13029 | if (r < 0) | |
13030 | return r; | |
13031 | } | |
13032 | ||
13033 | InodeRef in; | |
13034 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13035 | if (r == 0) { | |
13036 | fill_stat(in, attr); | |
13037 | _ll_get(in.get()); | |
13038 | } | |
13039 | tout(cct) << attr->st_ino << std::endl; | |
13040 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name | |
13041 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13042 | *out = in.get(); | |
13043 | return r; | |
13044 | } | |
13045 | ||
13046 | int Client::ll_mknodx(Inode *parent, const char *name, mode_t mode, | |
13047 | dev_t rdev, Inode **out, | |
13048 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13049 | const UserPerm& perms) | |
13050 | { | |
f67539c2 TL |
13051 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13052 | if (!mref_reader.is_state_satisfied()) | |
13053 | return -CEPHFS_ENOTCONN; | |
7c673cae | 13054 | |
f67539c2 | 13055 | unsigned caps = statx_to_mask(flags, want); |
181888fb | 13056 | |
7c673cae FG |
13057 | vinodeno_t vparent = _get_vino(parent); |
13058 | ||
13059 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name << dendl; | |
13060 | tout(cct) << "ll_mknodx" << std::endl; | |
13061 | tout(cct) << vparent.ino.val << std::endl; | |
13062 | tout(cct) << name << std::endl; | |
13063 | tout(cct) << mode << std::endl; | |
13064 | tout(cct) << rdev << std::endl; | |
13065 | ||
f67539c2 TL |
13066 | std::scoped_lock lock(client_lock); |
13067 | ||
11fdf7f2 | 13068 | if (!fuse_default_permissions) { |
7c673cae FG |
13069 | int r = may_create(parent, perms); |
13070 | if (r < 0) | |
13071 | return r; | |
13072 | } | |
13073 | ||
13074 | InodeRef in; | |
13075 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
13076 | if (r == 0) { | |
13077 | fill_statx(in, caps, stx); | |
13078 | _ll_get(in.get()); | |
13079 | } | |
13080 | tout(cct) << stx->stx_ino << std::endl; | |
13081 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name | |
13082 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13083 | *out = in.get(); | |
13084 | return r; | |
13085 | } | |
13086 | ||
13087 | int Client::_create(Inode *dir, const char *name, int flags, mode_t mode, | |
13088 | InodeRef *inp, Fh **fhp, int stripe_unit, int stripe_count, | |
13089 | int object_size, const char *data_pool, bool *created, | |
f67539c2 | 13090 | const UserPerm& perms, std::string alternate_name) |
7c673cae | 13091 | { |
1adf2230 | 13092 | ldout(cct, 8) << "_create(" << dir->ino << " " << name << ", 0" << oct << |
7c673cae FG |
13093 | mode << dec << ")" << dendl; |
13094 | ||
13095 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13096 | return -CEPHFS_ENAMETOOLONG; |
7c673cae | 13097 | if (dir->snapid != CEPH_NOSNAP) { |
f67539c2 | 13098 | return -CEPHFS_EROFS; |
7c673cae FG |
13099 | } |
13100 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13101 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13102 | } |
13103 | ||
13104 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
13105 | int cflags = ceph_flags_sys2wire(flags); |
13106 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
13107 | cflags |= CEPH_O_LAZY; | |
13108 | ||
13109 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
13110 | |
13111 | int64_t pool_id = -1; | |
13112 | if (data_pool && *data_pool) { | |
13113 | pool_id = objecter->with_osdmap( | |
13114 | std::mem_fn(&OSDMap::lookup_pg_pool_name), data_pool); | |
13115 | if (pool_id < 0) | |
f67539c2 | 13116 | return -CEPHFS_EINVAL; |
7c673cae | 13117 | if (pool_id > 0xffffffffll) |
f67539c2 | 13118 | return -CEPHFS_ERANGE; // bummer! |
7c673cae FG |
13119 | } |
13120 | ||
13121 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_CREATE); | |
13122 | ||
13123 | filepath path; | |
13124 | dir->make_nosnap_relative_path(path); | |
13125 | path.push_dentry(name); | |
13126 | req->set_filepath(path); | |
f67539c2 | 13127 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae | 13128 | req->set_inode(dir); |
11fdf7f2 | 13129 | req->head.args.open.flags = cflags | CEPH_O_CREAT; |
7c673cae FG |
13130 | |
13131 | req->head.args.open.stripe_unit = stripe_unit; | |
13132 | req->head.args.open.stripe_count = stripe_count; | |
13133 | req->head.args.open.object_size = object_size; | |
13134 | if (cct->_conf->client_debug_getattr_caps) | |
13135 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
13136 | else | |
13137 | req->head.args.open.mask = 0; | |
13138 | req->head.args.open.pool = pool_id; | |
13139 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13140 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13141 | ||
13142 | mode |= S_IFREG; | |
13143 | bufferlist xattrs_bl; | |
13144 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
13145 | if (res < 0) | |
13146 | goto fail; | |
13147 | req->head.args.open.mode = mode; | |
13148 | if (xattrs_bl.length() > 0) | |
13149 | req->set_data(xattrs_bl); | |
13150 | ||
13151 | Dentry *de; | |
13152 | res = get_or_create(dir, name, &de); | |
13153 | if (res < 0) | |
13154 | goto fail; | |
13155 | req->set_dentry(de); | |
13156 | ||
13157 | res = make_request(req, perms, inp, created); | |
13158 | if (res < 0) { | |
13159 | goto reply_error; | |
13160 | } | |
13161 | ||
13162 | /* If the caller passed a value in fhp, do the open */ | |
13163 | if(fhp) { | |
13164 | (*inp)->get_open_ref(cmode); | |
13165 | *fhp = _create_fh(inp->get(), flags, cmode, perms); | |
13166 | } | |
13167 | ||
13168 | reply_error: | |
13169 | trim_cache(); | |
13170 | ||
1adf2230 | 13171 | ldout(cct, 8) << "create(" << path << ", 0" << oct << mode << dec |
7c673cae FG |
13172 | << " layout " << stripe_unit |
13173 | << ' ' << stripe_count | |
13174 | << ' ' << object_size | |
13175 | <<") = " << res << dendl; | |
13176 | return res; | |
13177 | ||
13178 | fail: | |
13179 | put_request(req); | |
13180 | return res; | |
13181 | } | |
13182 | ||
7c673cae | 13183 | int Client::_mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm, |
f67539c2 TL |
13184 | InodeRef *inp, const std::map<std::string, std::string> &metadata, |
13185 | std::string alternate_name) | |
7c673cae | 13186 | { |
1adf2230 | 13187 | ldout(cct, 8) << "_mkdir(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
13188 | << mode << dec << ", uid " << perm.uid() |
13189 | << ", gid " << perm.gid() << ")" << dendl; | |
13190 | ||
13191 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13192 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13193 | |
13194 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13195 | return -CEPHFS_EROFS; |
7c673cae FG |
13196 | } |
13197 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 13198 | return -CEPHFS_EDQUOT; |
7c673cae | 13199 | } |
f67539c2 TL |
13200 | |
13201 | bool is_snap_op = dir->snapid == CEPH_SNAPDIR; | |
13202 | MetaRequest *req = new MetaRequest(is_snap_op ? | |
7c673cae FG |
13203 | CEPH_MDS_OP_MKSNAP : CEPH_MDS_OP_MKDIR); |
13204 | ||
13205 | filepath path; | |
13206 | dir->make_nosnap_relative_path(path); | |
13207 | path.push_dentry(name); | |
13208 | req->set_filepath(path); | |
13209 | req->set_inode(dir); | |
13210 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13211 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
f67539c2 | 13212 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13213 | |
13214 | mode |= S_IFDIR; | |
f67539c2 TL |
13215 | bufferlist bl; |
13216 | int res = _posix_acl_create(dir, &mode, bl, perm); | |
7c673cae FG |
13217 | if (res < 0) |
13218 | goto fail; | |
13219 | req->head.args.mkdir.mode = mode; | |
f67539c2 TL |
13220 | if (is_snap_op) { |
13221 | SnapPayload payload; | |
13222 | // clear the bufferlist that may have been populated by the call | |
13223 | // to _posix_acl_create(). MDS mksnap does not make use of it. | |
13224 | // So, reuse it to pass metadata payload. | |
13225 | bl.clear(); | |
13226 | payload.metadata = metadata; | |
13227 | encode(payload, bl); | |
13228 | } | |
13229 | if (bl.length() > 0) { | |
13230 | req->set_data(bl); | |
13231 | } | |
7c673cae FG |
13232 | |
13233 | Dentry *de; | |
13234 | res = get_or_create(dir, name, &de); | |
13235 | if (res < 0) | |
13236 | goto fail; | |
13237 | req->set_dentry(de); | |
13238 | ||
13239 | ldout(cct, 10) << "_mkdir: making request" << dendl; | |
13240 | res = make_request(req, perm, inp); | |
13241 | ldout(cct, 10) << "_mkdir result is " << res << dendl; | |
13242 | ||
13243 | trim_cache(); | |
13244 | ||
1adf2230 | 13245 | ldout(cct, 8) << "_mkdir(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
13246 | return res; |
13247 | ||
13248 | fail: | |
13249 | put_request(req); | |
13250 | return res; | |
13251 | } | |
13252 | ||
13253 | int Client::ll_mkdir(Inode *parent, const char *name, mode_t mode, | |
13254 | struct stat *attr, Inode **out, const UserPerm& perm) | |
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 vparent = _get_vino(parent); |
13261 | ||
13262 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name << dendl; | |
13263 | tout(cct) << "ll_mkdir" << std::endl; | |
13264 | tout(cct) << vparent.ino.val << std::endl; | |
13265 | tout(cct) << name << std::endl; | |
13266 | tout(cct) << mode << std::endl; | |
13267 | ||
f67539c2 TL |
13268 | std::scoped_lock lock(client_lock); |
13269 | ||
11fdf7f2 | 13270 | if (!fuse_default_permissions) { |
7c673cae FG |
13271 | int r = may_create(parent, perm); |
13272 | if (r < 0) | |
13273 | return r; | |
13274 | } | |
13275 | ||
13276 | InodeRef in; | |
13277 | int r = _mkdir(parent, name, mode, perm, &in); | |
13278 | if (r == 0) { | |
13279 | fill_stat(in, attr); | |
13280 | _ll_get(in.get()); | |
13281 | } | |
13282 | tout(cct) << attr->st_ino << std::endl; | |
13283 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name | |
13284 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13285 | *out = in.get(); | |
13286 | return r; | |
13287 | } | |
13288 | ||
13289 | int Client::ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out, | |
13290 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
13291 | const UserPerm& perms) | |
13292 | { | |
f67539c2 TL |
13293 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13294 | if (!mref_reader.is_state_satisfied()) | |
13295 | return -CEPHFS_ENOTCONN; | |
181888fb | 13296 | |
7c673cae FG |
13297 | vinodeno_t vparent = _get_vino(parent); |
13298 | ||
13299 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name << dendl; | |
13300 | tout(cct) << "ll_mkdirx" << std::endl; | |
13301 | tout(cct) << vparent.ino.val << std::endl; | |
13302 | tout(cct) << name << std::endl; | |
13303 | tout(cct) << mode << std::endl; | |
13304 | ||
f67539c2 TL |
13305 | std::scoped_lock lock(client_lock); |
13306 | ||
11fdf7f2 | 13307 | if (!fuse_default_permissions) { |
7c673cae FG |
13308 | int r = may_create(parent, perms); |
13309 | if (r < 0) | |
13310 | return r; | |
13311 | } | |
13312 | ||
13313 | InodeRef in; | |
13314 | int r = _mkdir(parent, name, mode, perms, &in); | |
13315 | if (r == 0) { | |
13316 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13317 | _ll_get(in.get()); | |
13318 | } else { | |
13319 | stx->stx_ino = 0; | |
13320 | stx->stx_mask = 0; | |
13321 | } | |
13322 | tout(cct) << stx->stx_ino << std::endl; | |
13323 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name | |
13324 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13325 | *out = in.get(); | |
13326 | return r; | |
13327 | } | |
13328 | ||
13329 | int Client::_symlink(Inode *dir, const char *name, const char *target, | |
f67539c2 | 13330 | const UserPerm& perms, std::string alternate_name, InodeRef *inp) |
7c673cae | 13331 | { |
1adf2230 | 13332 | ldout(cct, 8) << "_symlink(" << dir->ino << " " << name << ", " << target |
7c673cae FG |
13333 | << ", uid " << perms.uid() << ", gid " << perms.gid() << ")" |
13334 | << dendl; | |
13335 | ||
13336 | if (strlen(name) > NAME_MAX) | |
f67539c2 | 13337 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13338 | |
13339 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13340 | return -CEPHFS_EROFS; |
7c673cae FG |
13341 | } |
13342 | if (is_quota_files_exceeded(dir, perms)) { | |
f67539c2 | 13343 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13344 | } |
13345 | ||
13346 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SYMLINK); | |
13347 | ||
13348 | filepath path; | |
13349 | dir->make_nosnap_relative_path(path); | |
13350 | path.push_dentry(name); | |
13351 | req->set_filepath(path); | |
f67539c2 | 13352 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13353 | req->set_inode(dir); |
13354 | req->set_string2(target); | |
13355 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13356 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13357 | ||
13358 | Dentry *de; | |
13359 | int res = get_or_create(dir, name, &de); | |
13360 | if (res < 0) | |
13361 | goto fail; | |
13362 | req->set_dentry(de); | |
13363 | ||
13364 | res = make_request(req, perms, inp); | |
13365 | ||
13366 | trim_cache(); | |
1adf2230 | 13367 | ldout(cct, 8) << "_symlink(\"" << path << "\", \"" << target << "\") = " << |
7c673cae FG |
13368 | res << dendl; |
13369 | return res; | |
13370 | ||
13371 | fail: | |
13372 | put_request(req); | |
13373 | return res; | |
13374 | } | |
13375 | ||
13376 | int Client::ll_symlink(Inode *parent, const char *name, const char *value, | |
13377 | struct stat *attr, Inode **out, const UserPerm& perms) | |
13378 | { | |
f67539c2 TL |
13379 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13380 | if (!mref_reader.is_state_satisfied()) | |
13381 | return -CEPHFS_ENOTCONN; | |
181888fb | 13382 | |
7c673cae FG |
13383 | vinodeno_t vparent = _get_vino(parent); |
13384 | ||
13385 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name << " -> " << value | |
13386 | << dendl; | |
13387 | tout(cct) << "ll_symlink" << std::endl; | |
13388 | tout(cct) << vparent.ino.val << std::endl; | |
13389 | tout(cct) << name << std::endl; | |
13390 | tout(cct) << value << std::endl; | |
13391 | ||
f67539c2 TL |
13392 | std::scoped_lock lock(client_lock); |
13393 | ||
11fdf7f2 | 13394 | if (!fuse_default_permissions) { |
7c673cae FG |
13395 | int r = may_create(parent, perms); |
13396 | if (r < 0) | |
13397 | return r; | |
13398 | } | |
13399 | ||
13400 | InodeRef in; | |
f67539c2 | 13401 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13402 | if (r == 0) { |
13403 | fill_stat(in, attr); | |
13404 | _ll_get(in.get()); | |
13405 | } | |
13406 | tout(cct) << attr->st_ino << std::endl; | |
13407 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name | |
13408 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
13409 | *out = in.get(); | |
13410 | return r; | |
13411 | } | |
13412 | ||
13413 | int Client::ll_symlinkx(Inode *parent, const char *name, const char *value, | |
13414 | Inode **out, struct ceph_statx *stx, unsigned want, | |
13415 | unsigned flags, const UserPerm& perms) | |
13416 | { | |
f67539c2 TL |
13417 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13418 | if (!mref_reader.is_state_satisfied()) | |
13419 | return -CEPHFS_ENOTCONN; | |
181888fb | 13420 | |
7c673cae FG |
13421 | vinodeno_t vparent = _get_vino(parent); |
13422 | ||
13423 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name << " -> " << value | |
13424 | << dendl; | |
13425 | tout(cct) << "ll_symlinkx" << std::endl; | |
13426 | tout(cct) << vparent.ino.val << std::endl; | |
13427 | tout(cct) << name << std::endl; | |
13428 | tout(cct) << value << std::endl; | |
13429 | ||
f67539c2 TL |
13430 | std::scoped_lock lock(client_lock); |
13431 | ||
11fdf7f2 | 13432 | if (!fuse_default_permissions) { |
7c673cae FG |
13433 | int r = may_create(parent, perms); |
13434 | if (r < 0) | |
13435 | return r; | |
13436 | } | |
13437 | ||
13438 | InodeRef in; | |
f67539c2 | 13439 | int r = _symlink(parent, name, value, perms, "", &in); |
7c673cae FG |
13440 | if (r == 0) { |
13441 | fill_statx(in, statx_to_mask(flags, want), stx); | |
13442 | _ll_get(in.get()); | |
13443 | } | |
13444 | tout(cct) << stx->stx_ino << std::endl; | |
13445 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name | |
13446 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
13447 | *out = in.get(); | |
13448 | return r; | |
13449 | } | |
13450 | ||
13451 | int Client::_unlink(Inode *dir, const char *name, const UserPerm& perm) | |
13452 | { | |
1adf2230 | 13453 | ldout(cct, 8) << "_unlink(" << dir->ino << " " << name |
7c673cae FG |
13454 | << " uid " << perm.uid() << " gid " << perm.gid() |
13455 | << ")" << dendl; | |
13456 | ||
13457 | if (dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13458 | return -CEPHFS_EROFS; |
7c673cae FG |
13459 | } |
13460 | ||
13461 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_UNLINK); | |
13462 | ||
13463 | filepath path; | |
13464 | dir->make_nosnap_relative_path(path); | |
13465 | path.push_dentry(name); | |
13466 | req->set_filepath(path); | |
13467 | ||
13468 | InodeRef otherin; | |
b32b8144 | 13469 | Inode *in; |
7c673cae | 13470 | Dentry *de; |
b32b8144 | 13471 | |
7c673cae FG |
13472 | int res = get_or_create(dir, name, &de); |
13473 | if (res < 0) | |
13474 | goto fail; | |
13475 | req->set_dentry(de); | |
13476 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13477 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13478 | ||
13479 | res = _lookup(dir, name, 0, &otherin, perm); | |
13480 | if (res < 0) | |
13481 | goto fail; | |
b32b8144 FG |
13482 | |
13483 | in = otherin.get(); | |
13484 | req->set_other_inode(in); | |
13485 | in->break_all_delegs(); | |
7c673cae FG |
13486 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
13487 | ||
13488 | req->set_inode(dir); | |
13489 | ||
13490 | res = make_request(req, perm); | |
13491 | ||
13492 | trim_cache(); | |
1adf2230 | 13493 | ldout(cct, 8) << "unlink(" << path << ") = " << res << dendl; |
7c673cae FG |
13494 | return res; |
13495 | ||
13496 | fail: | |
13497 | put_request(req); | |
13498 | return res; | |
13499 | } | |
13500 | ||
13501 | int Client::ll_unlink(Inode *in, const char *name, const UserPerm& perm) | |
13502 | { | |
f67539c2 TL |
13503 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13504 | if (!mref_reader.is_state_satisfied()) | |
13505 | return -CEPHFS_ENOTCONN; | |
181888fb | 13506 | |
7c673cae FG |
13507 | vinodeno_t vino = _get_vino(in); |
13508 | ||
13509 | ldout(cct, 3) << "ll_unlink " << vino << " " << name << dendl; | |
13510 | tout(cct) << "ll_unlink" << std::endl; | |
13511 | tout(cct) << vino.ino.val << std::endl; | |
13512 | tout(cct) << name << std::endl; | |
13513 | ||
f67539c2 TL |
13514 | std::scoped_lock lock(client_lock); |
13515 | ||
11fdf7f2 | 13516 | if (!fuse_default_permissions) { |
7c673cae FG |
13517 | int r = may_delete(in, name, perm); |
13518 | if (r < 0) | |
13519 | return r; | |
13520 | } | |
13521 | return _unlink(in, name, perm); | |
13522 | } | |
13523 | ||
13524 | int Client::_rmdir(Inode *dir, const char *name, const UserPerm& perms) | |
13525 | { | |
1adf2230 | 13526 | ldout(cct, 8) << "_rmdir(" << dir->ino << " " << name << " uid " |
7c673cae FG |
13527 | << perms.uid() << " gid " << perms.gid() << ")" << dendl; |
13528 | ||
13529 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
f67539c2 | 13530 | return -CEPHFS_EROFS; |
7c673cae | 13531 | } |
b32b8144 FG |
13532 | |
13533 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_RMSNAP : CEPH_MDS_OP_RMDIR; | |
13534 | MetaRequest *req = new MetaRequest(op); | |
7c673cae FG |
13535 | filepath path; |
13536 | dir->make_nosnap_relative_path(path); | |
13537 | path.push_dentry(name); | |
13538 | req->set_filepath(path); | |
11fdf7f2 | 13539 | req->set_inode(dir); |
7c673cae FG |
13540 | |
13541 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13542 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13543 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
13544 | ||
13545 | InodeRef in; | |
13546 | ||
13547 | Dentry *de; | |
13548 | int res = get_or_create(dir, name, &de); | |
13549 | if (res < 0) | |
13550 | goto fail; | |
b32b8144 FG |
13551 | if (op == CEPH_MDS_OP_RMDIR) |
13552 | req->set_dentry(de); | |
13553 | else | |
13554 | de->get(); | |
13555 | ||
7c673cae FG |
13556 | res = _lookup(dir, name, 0, &in, perms); |
13557 | if (res < 0) | |
13558 | goto fail; | |
11fdf7f2 TL |
13559 | |
13560 | if (op == CEPH_MDS_OP_RMSNAP) { | |
7c673cae | 13561 | unlink(de, true, true); |
b32b8144 | 13562 | de->put(); |
7c673cae | 13563 | } |
11fdf7f2 | 13564 | req->set_other_inode(in.get()); |
7c673cae FG |
13565 | |
13566 | res = make_request(req, perms); | |
13567 | ||
13568 | trim_cache(); | |
1adf2230 | 13569 | ldout(cct, 8) << "rmdir(" << path << ") = " << res << dendl; |
7c673cae FG |
13570 | return res; |
13571 | ||
13572 | fail: | |
13573 | put_request(req); | |
13574 | return res; | |
13575 | } | |
13576 | ||
13577 | int Client::ll_rmdir(Inode *in, const char *name, const UserPerm& perms) | |
13578 | { | |
f67539c2 TL |
13579 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13580 | if (!mref_reader.is_state_satisfied()) | |
13581 | return -CEPHFS_ENOTCONN; | |
181888fb | 13582 | |
7c673cae FG |
13583 | vinodeno_t vino = _get_vino(in); |
13584 | ||
13585 | ldout(cct, 3) << "ll_rmdir " << vino << " " << name << dendl; | |
13586 | tout(cct) << "ll_rmdir" << std::endl; | |
13587 | tout(cct) << vino.ino.val << std::endl; | |
13588 | tout(cct) << name << std::endl; | |
13589 | ||
f67539c2 TL |
13590 | std::scoped_lock lock(client_lock); |
13591 | ||
11fdf7f2 | 13592 | if (!fuse_default_permissions) { |
7c673cae FG |
13593 | int r = may_delete(in, name, perms); |
13594 | if (r < 0) | |
13595 | return r; | |
13596 | } | |
13597 | ||
13598 | return _rmdir(in, name, perms); | |
13599 | } | |
13600 | ||
f67539c2 | 13601 | int Client::_rename(Inode *fromdir, const char *fromname, Inode *todir, const char *toname, const UserPerm& perm, std::string alternate_name) |
7c673cae | 13602 | { |
1adf2230 | 13603 | ldout(cct, 8) << "_rename(" << fromdir->ino << " " << fromname << " to " |
7c673cae FG |
13604 | << todir->ino << " " << toname |
13605 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" | |
13606 | << dendl; | |
13607 | ||
13608 | if (fromdir->snapid != todir->snapid) | |
f67539c2 | 13609 | return -CEPHFS_EXDEV; |
7c673cae FG |
13610 | |
13611 | int op = CEPH_MDS_OP_RENAME; | |
13612 | if (fromdir->snapid != CEPH_NOSNAP) { | |
13613 | if (fromdir == todir && fromdir->snapid == CEPH_SNAPDIR) | |
13614 | op = CEPH_MDS_OP_RENAMESNAP; | |
13615 | else | |
f67539c2 TL |
13616 | return -CEPHFS_EROFS; |
13617 | } | |
13618 | if (fromdir != todir) { | |
13619 | Inode *fromdir_root = | |
13620 | fromdir->quota.is_enable() ? fromdir : get_quota_root(fromdir, perm); | |
13621 | Inode *todir_root = | |
13622 | todir->quota.is_enable() ? todir : get_quota_root(todir, perm); | |
13623 | if (fromdir_root != todir_root) { | |
13624 | return -CEPHFS_EXDEV; | |
13625 | } | |
7c673cae | 13626 | } |
7c673cae FG |
13627 | |
13628 | InodeRef target; | |
13629 | MetaRequest *req = new MetaRequest(op); | |
13630 | ||
13631 | filepath from; | |
13632 | fromdir->make_nosnap_relative_path(from); | |
13633 | from.push_dentry(fromname); | |
13634 | filepath to; | |
13635 | todir->make_nosnap_relative_path(to); | |
13636 | to.push_dentry(toname); | |
13637 | req->set_filepath(to); | |
13638 | req->set_filepath2(from); | |
f67539c2 | 13639 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13640 | |
13641 | Dentry *oldde; | |
13642 | int res = get_or_create(fromdir, fromname, &oldde); | |
13643 | if (res < 0) | |
13644 | goto fail; | |
13645 | Dentry *de; | |
13646 | res = get_or_create(todir, toname, &de); | |
13647 | if (res < 0) | |
13648 | goto fail; | |
13649 | ||
13650 | if (op == CEPH_MDS_OP_RENAME) { | |
13651 | req->set_old_dentry(oldde); | |
13652 | req->old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
13653 | req->old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
13654 | ||
13655 | req->set_dentry(de); | |
13656 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
13657 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
13658 | ||
13659 | InodeRef oldin, otherin; | |
f67539c2 | 13660 | res = _lookup(fromdir, fromname, 0, &oldin, perm); |
7c673cae FG |
13661 | if (res < 0) |
13662 | goto fail; | |
b32b8144 FG |
13663 | |
13664 | Inode *oldinode = oldin.get(); | |
13665 | oldinode->break_all_delegs(); | |
13666 | req->set_old_inode(oldinode); | |
7c673cae FG |
13667 | req->old_inode_drop = CEPH_CAP_LINK_SHARED; |
13668 | ||
13669 | res = _lookup(todir, toname, 0, &otherin, perm); | |
b32b8144 FG |
13670 | switch (res) { |
13671 | case 0: | |
13672 | { | |
13673 | Inode *in = otherin.get(); | |
13674 | req->set_other_inode(in); | |
13675 | in->break_all_delegs(); | |
13676 | } | |
7c673cae | 13677 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
b32b8144 | 13678 | break; |
f67539c2 | 13679 | case -CEPHFS_ENOENT: |
b32b8144 FG |
13680 | break; |
13681 | default: | |
13682 | goto fail; | |
7c673cae FG |
13683 | } |
13684 | ||
13685 | req->set_inode(todir); | |
13686 | } else { | |
13687 | // renamesnap reply contains no tracedn, so we need to invalidate | |
13688 | // dentry manually | |
13689 | unlink(oldde, true, true); | |
13690 | unlink(de, true, true); | |
11fdf7f2 TL |
13691 | |
13692 | req->set_inode(todir); | |
7c673cae FG |
13693 | } |
13694 | ||
13695 | res = make_request(req, perm, &target); | |
13696 | ldout(cct, 10) << "rename result is " << res << dendl; | |
13697 | ||
13698 | // renamed item from our cache | |
13699 | ||
13700 | trim_cache(); | |
1adf2230 | 13701 | ldout(cct, 8) << "_rename(" << from << ", " << to << ") = " << res << dendl; |
7c673cae FG |
13702 | return res; |
13703 | ||
13704 | fail: | |
13705 | put_request(req); | |
13706 | return res; | |
13707 | } | |
13708 | ||
13709 | int Client::ll_rename(Inode *parent, const char *name, Inode *newparent, | |
13710 | const char *newname, const UserPerm& perm) | |
13711 | { | |
f67539c2 TL |
13712 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13713 | if (!mref_reader.is_state_satisfied()) | |
13714 | return -CEPHFS_ENOTCONN; | |
181888fb | 13715 | |
7c673cae FG |
13716 | vinodeno_t vparent = _get_vino(parent); |
13717 | vinodeno_t vnewparent = _get_vino(newparent); | |
13718 | ||
13719 | ldout(cct, 3) << "ll_rename " << vparent << " " << name << " to " | |
13720 | << vnewparent << " " << newname << dendl; | |
13721 | tout(cct) << "ll_rename" << std::endl; | |
13722 | tout(cct) << vparent.ino.val << std::endl; | |
13723 | tout(cct) << name << std::endl; | |
13724 | tout(cct) << vnewparent.ino.val << std::endl; | |
13725 | tout(cct) << newname << std::endl; | |
13726 | ||
f67539c2 TL |
13727 | std::scoped_lock lock(client_lock); |
13728 | ||
11fdf7f2 | 13729 | if (!fuse_default_permissions) { |
7c673cae FG |
13730 | int r = may_delete(parent, name, perm); |
13731 | if (r < 0) | |
13732 | return r; | |
13733 | r = may_delete(newparent, newname, perm); | |
f67539c2 | 13734 | if (r < 0 && r != -CEPHFS_ENOENT) |
7c673cae FG |
13735 | return r; |
13736 | } | |
13737 | ||
f67539c2 | 13738 | return _rename(parent, name, newparent, newname, perm, ""); |
7c673cae FG |
13739 | } |
13740 | ||
f67539c2 | 13741 | int Client::_link(Inode *in, Inode *dir, const char *newname, const UserPerm& perm, std::string alternate_name, InodeRef *inp) |
7c673cae | 13742 | { |
1adf2230 | 13743 | ldout(cct, 8) << "_link(" << in->ino << " to " << dir->ino << " " << newname |
7c673cae FG |
13744 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" << dendl; |
13745 | ||
13746 | if (strlen(newname) > NAME_MAX) | |
f67539c2 | 13747 | return -CEPHFS_ENAMETOOLONG; |
7c673cae FG |
13748 | |
13749 | if (in->snapid != CEPH_NOSNAP || dir->snapid != CEPH_NOSNAP) { | |
f67539c2 | 13750 | return -CEPHFS_EROFS; |
7c673cae FG |
13751 | } |
13752 | if (is_quota_files_exceeded(dir, perm)) { | |
f67539c2 | 13753 | return -CEPHFS_EDQUOT; |
7c673cae FG |
13754 | } |
13755 | ||
b32b8144 | 13756 | in->break_all_delegs(); |
7c673cae FG |
13757 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LINK); |
13758 | ||
13759 | filepath path(newname, dir->ino); | |
13760 | req->set_filepath(path); | |
f67539c2 | 13761 | req->set_alternate_name(std::move(alternate_name)); |
7c673cae FG |
13762 | filepath existing(in->ino); |
13763 | req->set_filepath2(existing); | |
13764 | ||
13765 | req->set_inode(dir); | |
13766 | req->inode_drop = CEPH_CAP_FILE_SHARED; | |
13767 | req->inode_unless = CEPH_CAP_FILE_EXCL; | |
13768 | ||
13769 | Dentry *de; | |
13770 | int res = get_or_create(dir, newname, &de); | |
13771 | if (res < 0) | |
13772 | goto fail; | |
13773 | req->set_dentry(de); | |
13774 | ||
13775 | res = make_request(req, perm, inp); | |
13776 | ldout(cct, 10) << "link result is " << res << dendl; | |
13777 | ||
13778 | trim_cache(); | |
1adf2230 | 13779 | ldout(cct, 8) << "link(" << existing << ", " << path << ") = " << res << dendl; |
7c673cae FG |
13780 | return res; |
13781 | ||
13782 | fail: | |
13783 | put_request(req); | |
13784 | return res; | |
13785 | } | |
13786 | ||
13787 | int Client::ll_link(Inode *in, Inode *newparent, const char *newname, | |
13788 | const UserPerm& perm) | |
13789 | { | |
f67539c2 TL |
13790 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13791 | if (!mref_reader.is_state_satisfied()) | |
13792 | return -CEPHFS_ENOTCONN; | |
181888fb | 13793 | |
7c673cae FG |
13794 | vinodeno_t vino = _get_vino(in); |
13795 | vinodeno_t vnewparent = _get_vino(newparent); | |
13796 | ||
31f18b77 | 13797 | ldout(cct, 3) << "ll_link " << vino << " to " << vnewparent << " " << |
7c673cae FG |
13798 | newname << dendl; |
13799 | tout(cct) << "ll_link" << std::endl; | |
13800 | tout(cct) << vino.ino.val << std::endl; | |
13801 | tout(cct) << vnewparent << std::endl; | |
13802 | tout(cct) << newname << std::endl; | |
13803 | ||
7c673cae FG |
13804 | InodeRef target; |
13805 | ||
f67539c2 TL |
13806 | std::scoped_lock lock(client_lock); |
13807 | ||
11fdf7f2 | 13808 | if (!fuse_default_permissions) { |
7c673cae | 13809 | if (S_ISDIR(in->mode)) |
f67539c2 | 13810 | return -CEPHFS_EPERM; |
7c673cae | 13811 | |
11fdf7f2 | 13812 | int r = may_hardlink(in, perm); |
7c673cae FG |
13813 | if (r < 0) |
13814 | return r; | |
13815 | ||
13816 | r = may_create(newparent, perm); | |
13817 | if (r < 0) | |
13818 | return r; | |
13819 | } | |
13820 | ||
f67539c2 | 13821 | return _link(in, newparent, newname, perm, "", &target); |
7c673cae FG |
13822 | } |
13823 | ||
13824 | int Client::ll_num_osds(void) | |
13825 | { | |
f67539c2 | 13826 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13827 | return objecter->with_osdmap(std::mem_fn(&OSDMap::get_num_osds)); |
13828 | } | |
13829 | ||
13830 | int Client::ll_osdaddr(int osd, uint32_t *addr) | |
13831 | { | |
f67539c2 | 13832 | std::scoped_lock lock(client_lock); |
181888fb | 13833 | |
7c673cae FG |
13834 | entity_addr_t g; |
13835 | bool exists = objecter->with_osdmap([&](const OSDMap& o) { | |
13836 | if (!o.exists(osd)) | |
13837 | return false; | |
11fdf7f2 | 13838 | g = o.get_addrs(osd).front(); |
7c673cae FG |
13839 | return true; |
13840 | }); | |
13841 | if (!exists) | |
13842 | return -1; | |
13843 | uint32_t nb_addr = (g.in4_addr()).sin_addr.s_addr; | |
13844 | *addr = ntohl(nb_addr); | |
13845 | return 0; | |
13846 | } | |
181888fb | 13847 | |
7c673cae FG |
13848 | uint32_t Client::ll_stripe_unit(Inode *in) |
13849 | { | |
f67539c2 | 13850 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13851 | return in->layout.stripe_unit; |
13852 | } | |
13853 | ||
13854 | uint64_t Client::ll_snap_seq(Inode *in) | |
13855 | { | |
f67539c2 | 13856 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13857 | return in->snaprealm->seq; |
13858 | } | |
13859 | ||
13860 | int Client::ll_file_layout(Inode *in, file_layout_t *layout) | |
13861 | { | |
f67539c2 | 13862 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13863 | *layout = in->layout; |
13864 | return 0; | |
13865 | } | |
13866 | ||
13867 | int Client::ll_file_layout(Fh *fh, file_layout_t *layout) | |
13868 | { | |
13869 | return ll_file_layout(fh->inode.get(), layout); | |
13870 | } | |
13871 | ||
13872 | /* Currently we cannot take advantage of redundancy in reads, since we | |
13873 | would have to go through all possible placement groups (a | |
13874 | potentially quite large number determined by a hash), and use CRUSH | |
13875 | to calculate the appropriate set of OSDs for each placement group, | |
13876 | then index into that. An array with one entry per OSD is much more | |
13877 | tractable and works for demonstration purposes. */ | |
13878 | ||
13879 | int Client::ll_get_stripe_osd(Inode *in, uint64_t blockno, | |
13880 | file_layout_t* layout) | |
13881 | { | |
f67539c2 | 13882 | std::scoped_lock lock(client_lock); |
181888fb | 13883 | |
28e407b8 | 13884 | inodeno_t ino = in->ino; |
7c673cae FG |
13885 | uint32_t object_size = layout->object_size; |
13886 | uint32_t su = layout->stripe_unit; | |
13887 | uint32_t stripe_count = layout->stripe_count; | |
13888 | uint64_t stripes_per_object = object_size / su; | |
11fdf7f2 | 13889 | uint64_t stripeno = 0, stripepos = 0; |
7c673cae | 13890 | |
11fdf7f2 TL |
13891 | if(stripe_count) { |
13892 | stripeno = blockno / stripe_count; // which horizontal stripe (Y) | |
13893 | stripepos = blockno % stripe_count; // which object in the object set (X) | |
13894 | } | |
7c673cae FG |
13895 | uint64_t objectsetno = stripeno / stripes_per_object; // which object set |
13896 | uint64_t objectno = objectsetno * stripe_count + stripepos; // object id | |
13897 | ||
13898 | object_t oid = file_object_t(ino, objectno); | |
13899 | return objecter->with_osdmap([&](const OSDMap& o) { | |
13900 | ceph_object_layout olayout = | |
13901 | o.file_to_object_layout(oid, *layout); | |
13902 | pg_t pg = (pg_t)olayout.ol_pgid; | |
13903 | vector<int> osds; | |
13904 | int primary; | |
13905 | o.pg_to_acting_osds(pg, &osds, &primary); | |
13906 | return primary; | |
13907 | }); | |
13908 | } | |
13909 | ||
13910 | /* Return the offset of the block, internal to the object */ | |
13911 | ||
13912 | uint64_t Client::ll_get_internal_offset(Inode *in, uint64_t blockno) | |
13913 | { | |
f67539c2 | 13914 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13915 | file_layout_t *layout=&(in->layout); |
13916 | uint32_t object_size = layout->object_size; | |
13917 | uint32_t su = layout->stripe_unit; | |
13918 | uint64_t stripes_per_object = object_size / su; | |
13919 | ||
13920 | return (blockno % stripes_per_object) * su; | |
13921 | } | |
13922 | ||
13923 | int Client::ll_opendir(Inode *in, int flags, dir_result_t** dirpp, | |
13924 | const UserPerm& perms) | |
13925 | { | |
f67539c2 TL |
13926 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13927 | if (!mref_reader.is_state_satisfied()) | |
13928 | return -CEPHFS_ENOTCONN; | |
181888fb | 13929 | |
7c673cae FG |
13930 | vinodeno_t vino = _get_vino(in); |
13931 | ||
13932 | ldout(cct, 3) << "ll_opendir " << vino << dendl; | |
13933 | tout(cct) << "ll_opendir" << std::endl; | |
13934 | tout(cct) << vino.ino.val << std::endl; | |
13935 | ||
f67539c2 TL |
13936 | std::scoped_lock lock(client_lock); |
13937 | ||
11fdf7f2 | 13938 | if (!fuse_default_permissions) { |
7c673cae FG |
13939 | int r = may_open(in, flags, perms); |
13940 | if (r < 0) | |
13941 | return r; | |
13942 | } | |
13943 | ||
13944 | int r = _opendir(in, dirpp, perms); | |
f67539c2 | 13945 | tout(cct) << (uintptr_t)*dirpp << std::endl; |
7c673cae FG |
13946 | |
13947 | ldout(cct, 3) << "ll_opendir " << vino << " = " << r << " (" << *dirpp << ")" | |
13948 | << dendl; | |
13949 | return r; | |
13950 | } | |
13951 | ||
13952 | int Client::ll_releasedir(dir_result_t *dirp) | |
13953 | { | |
f67539c2 TL |
13954 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13955 | if (!mref_reader.is_state_satisfied()) | |
13956 | return -CEPHFS_ENOTCONN; | |
13957 | ||
7c673cae FG |
13958 | ldout(cct, 3) << "ll_releasedir " << dirp << dendl; |
13959 | tout(cct) << "ll_releasedir" << std::endl; | |
f67539c2 | 13960 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 13961 | |
f67539c2 | 13962 | std::scoped_lock lock(client_lock); |
181888fb | 13963 | |
7c673cae FG |
13964 | _closedir(dirp); |
13965 | return 0; | |
13966 | } | |
13967 | ||
13968 | int Client::ll_fsyncdir(dir_result_t *dirp) | |
13969 | { | |
f67539c2 TL |
13970 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13971 | if (!mref_reader.is_state_satisfied()) | |
13972 | return -CEPHFS_ENOTCONN; | |
13973 | ||
7c673cae FG |
13974 | ldout(cct, 3) << "ll_fsyncdir " << dirp << dendl; |
13975 | tout(cct) << "ll_fsyncdir" << std::endl; | |
f67539c2 | 13976 | tout(cct) << (uintptr_t)dirp << std::endl; |
181888fb | 13977 | |
f67539c2 | 13978 | std::scoped_lock lock(client_lock); |
7c673cae FG |
13979 | return _fsync(dirp->inode.get(), false); |
13980 | } | |
13981 | ||
13982 | int Client::ll_open(Inode *in, int flags, Fh **fhp, const UserPerm& perms) | |
13983 | { | |
11fdf7f2 | 13984 | ceph_assert(!(flags & O_CREAT)); |
7c673cae | 13985 | |
f67539c2 TL |
13986 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
13987 | if (!mref_reader.is_state_satisfied()) | |
13988 | return -CEPHFS_ENOTCONN; | |
181888fb | 13989 | |
7c673cae FG |
13990 | vinodeno_t vino = _get_vino(in); |
13991 | ||
13992 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << dendl; | |
13993 | tout(cct) << "ll_open" << std::endl; | |
13994 | tout(cct) << vino.ino.val << std::endl; | |
13995 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
13996 | ||
f67539c2 TL |
13997 | std::scoped_lock lock(client_lock); |
13998 | ||
7c673cae | 13999 | int r; |
11fdf7f2 | 14000 | if (!fuse_default_permissions) { |
7c673cae FG |
14001 | r = may_open(in, flags, perms); |
14002 | if (r < 0) | |
14003 | goto out; | |
14004 | } | |
14005 | ||
14006 | r = _open(in, flags, 0, fhp /* may be NULL */, perms); | |
14007 | ||
14008 | out: | |
14009 | Fh *fhptr = fhp ? *fhp : NULL; | |
14010 | if (fhptr) { | |
14011 | ll_unclosed_fh_set.insert(fhptr); | |
14012 | } | |
f67539c2 | 14013 | tout(cct) << (uintptr_t)fhptr << std::endl; |
7c673cae FG |
14014 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << |
14015 | " = " << r << " (" << fhptr << ")" << dendl; | |
14016 | return r; | |
14017 | } | |
14018 | ||
14019 | int Client::_ll_create(Inode *parent, const char *name, mode_t mode, | |
14020 | int flags, InodeRef *in, int caps, Fh **fhp, | |
14021 | const UserPerm& perms) | |
14022 | { | |
14023 | *fhp = NULL; | |
14024 | ||
14025 | vinodeno_t vparent = _get_vino(parent); | |
14026 | ||
1adf2230 | 14027 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14028 | mode << dec << " " << ceph_flags_sys2wire(flags) << ", uid " << perms.uid() |
14029 | << ", gid " << perms.gid() << dendl; | |
14030 | tout(cct) << "ll_create" << std::endl; | |
14031 | tout(cct) << vparent.ino.val << std::endl; | |
14032 | tout(cct) << name << std::endl; | |
14033 | tout(cct) << mode << std::endl; | |
14034 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
14035 | ||
14036 | bool created = false; | |
14037 | int r = _lookup(parent, name, caps, in, perms); | |
14038 | ||
14039 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
f67539c2 | 14040 | return -CEPHFS_EEXIST; |
7c673cae | 14041 | |
f67539c2 | 14042 | if (r == -CEPHFS_ENOENT && (flags & O_CREAT)) { |
11fdf7f2 | 14043 | if (!fuse_default_permissions) { |
7c673cae FG |
14044 | r = may_create(parent, perms); |
14045 | if (r < 0) | |
14046 | goto out; | |
14047 | } | |
14048 | r = _create(parent, name, flags, mode, in, fhp, 0, 0, 0, NULL, &created, | |
f67539c2 | 14049 | perms, ""); |
7c673cae FG |
14050 | if (r < 0) |
14051 | goto out; | |
14052 | } | |
14053 | ||
14054 | if (r < 0) | |
14055 | goto out; | |
14056 | ||
11fdf7f2 | 14057 | ceph_assert(*in); |
7c673cae FG |
14058 | |
14059 | ldout(cct, 20) << "_ll_create created = " << created << dendl; | |
14060 | if (!created) { | |
11fdf7f2 | 14061 | if (!fuse_default_permissions) { |
7c673cae FG |
14062 | r = may_open(in->get(), flags, perms); |
14063 | if (r < 0) { | |
14064 | if (*fhp) { | |
14065 | int release_r = _release_fh(*fhp); | |
11fdf7f2 | 14066 | ceph_assert(release_r == 0); // during create, no async data ops should have happened |
7c673cae FG |
14067 | } |
14068 | goto out; | |
14069 | } | |
14070 | } | |
14071 | if (*fhp == NULL) { | |
14072 | r = _open(in->get(), flags, mode, fhp, perms); | |
14073 | if (r < 0) | |
14074 | goto out; | |
14075 | } | |
14076 | } | |
14077 | ||
14078 | out: | |
14079 | if (*fhp) { | |
14080 | ll_unclosed_fh_set.insert(*fhp); | |
14081 | } | |
14082 | ||
14083 | ino_t ino = 0; | |
14084 | if (r >= 0) { | |
14085 | Inode *inode = in->get(); | |
14086 | if (use_faked_inos()) | |
14087 | ino = inode->faked_ino; | |
14088 | else | |
14089 | ino = inode->ino; | |
14090 | } | |
14091 | ||
f67539c2 | 14092 | tout(cct) << (uintptr_t)*fhp << std::endl; |
7c673cae | 14093 | tout(cct) << ino << std::endl; |
1adf2230 | 14094 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
14095 | mode << dec << " " << ceph_flags_sys2wire(flags) << " = " << r << " (" << |
14096 | *fhp << " " << hex << ino << dec << ")" << dendl; | |
14097 | ||
14098 | return r; | |
14099 | } | |
14100 | ||
14101 | int Client::ll_create(Inode *parent, const char *name, mode_t mode, | |
14102 | int flags, struct stat *attr, Inode **outp, Fh **fhp, | |
14103 | const UserPerm& perms) | |
14104 | { | |
f67539c2 TL |
14105 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14106 | if (!mref_reader.is_state_satisfied()) | |
14107 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14108 | |
f67539c2 TL |
14109 | std::scoped_lock lock(client_lock); |
14110 | InodeRef in; | |
181888fb | 14111 | |
7c673cae FG |
14112 | int r = _ll_create(parent, name, mode, flags, &in, CEPH_STAT_CAP_INODE_ALL, |
14113 | fhp, perms); | |
14114 | if (r >= 0) { | |
11fdf7f2 | 14115 | ceph_assert(in); |
7c673cae FG |
14116 | |
14117 | // passing an Inode in outp requires an additional ref | |
14118 | if (outp) { | |
14119 | _ll_get(in.get()); | |
14120 | *outp = in.get(); | |
14121 | } | |
14122 | fill_stat(in, attr); | |
14123 | } else { | |
14124 | attr->st_ino = 0; | |
14125 | } | |
14126 | ||
14127 | return r; | |
14128 | } | |
14129 | ||
14130 | int Client::ll_createx(Inode *parent, const char *name, mode_t mode, | |
14131 | int oflags, Inode **outp, Fh **fhp, | |
14132 | struct ceph_statx *stx, unsigned want, unsigned lflags, | |
14133 | const UserPerm& perms) | |
14134 | { | |
14135 | unsigned caps = statx_to_mask(lflags, want); | |
f67539c2 TL |
14136 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14137 | if (!mref_reader.is_state_satisfied()) | |
14138 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14139 | |
f67539c2 TL |
14140 | std::scoped_lock lock(client_lock); |
14141 | InodeRef in; | |
7c673cae FG |
14142 | |
14143 | int r = _ll_create(parent, name, mode, oflags, &in, caps, fhp, perms); | |
14144 | if (r >= 0) { | |
11fdf7f2 | 14145 | ceph_assert(in); |
7c673cae FG |
14146 | |
14147 | // passing an Inode in outp requires an additional ref | |
14148 | if (outp) { | |
14149 | _ll_get(in.get()); | |
14150 | *outp = in.get(); | |
14151 | } | |
14152 | fill_statx(in, caps, stx); | |
14153 | } else { | |
14154 | stx->stx_ino = 0; | |
14155 | stx->stx_mask = 0; | |
14156 | } | |
14157 | ||
14158 | return r; | |
14159 | } | |
14160 | ||
14161 | loff_t Client::ll_lseek(Fh *fh, loff_t offset, int whence) | |
14162 | { | |
f67539c2 TL |
14163 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14164 | if (!mref_reader.is_state_satisfied()) | |
14165 | return -CEPHFS_ENOTCONN; | |
14166 | ||
7c673cae FG |
14167 | tout(cct) << "ll_lseek" << std::endl; |
14168 | tout(cct) << offset << std::endl; | |
14169 | tout(cct) << whence << std::endl; | |
14170 | ||
f67539c2 | 14171 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14172 | return _lseek(fh, offset, whence); |
14173 | } | |
14174 | ||
14175 | int Client::ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl) | |
14176 | { | |
f67539c2 TL |
14177 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14178 | if (!mref_reader.is_state_satisfied()) | |
14179 | return -CEPHFS_ENOTCONN; | |
14180 | ||
7c673cae FG |
14181 | ldout(cct, 3) << "ll_read " << fh << " " << fh->inode->ino << " " << " " << off << "~" << len << dendl; |
14182 | tout(cct) << "ll_read" << std::endl; | |
f67539c2 | 14183 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14184 | tout(cct) << off << std::endl; |
14185 | tout(cct) << len << std::endl; | |
14186 | ||
11fdf7f2 TL |
14187 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14188 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14189 | std::scoped_lock lock(client_lock); |
14190 | ||
f6b5b4d7 TL |
14191 | int r = _read(fh, off, len, bl); |
14192 | ldout(cct, 3) << "ll_read " << fh << " " << off << "~" << len << " = " << r | |
14193 | << dendl; | |
14194 | return r; | |
7c673cae FG |
14195 | } |
14196 | ||
14197 | int Client::ll_read_block(Inode *in, uint64_t blockid, | |
14198 | char *buf, | |
14199 | uint64_t offset, | |
14200 | uint64_t length, | |
14201 | file_layout_t* layout) | |
14202 | { | |
f67539c2 TL |
14203 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14204 | if (!mref_reader.is_state_satisfied()) | |
14205 | return -CEPHFS_ENOTCONN; | |
181888fb | 14206 | |
b32b8144 | 14207 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14208 | object_t oid = file_object_t(vino.ino, blockid); |
14209 | C_SaferCond onfinish; | |
14210 | bufferlist bl; | |
14211 | ||
14212 | objecter->read(oid, | |
14213 | object_locator_t(layout->pool_id), | |
14214 | offset, | |
14215 | length, | |
14216 | vino.snapid, | |
14217 | &bl, | |
14218 | CEPH_OSD_FLAG_READ, | |
14219 | &onfinish); | |
14220 | ||
7c673cae | 14221 | int r = onfinish.wait(); |
7c673cae | 14222 | if (r >= 0) { |
9f95a23c | 14223 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
14224 | r = bl.length(); |
14225 | } | |
14226 | ||
14227 | return r; | |
14228 | } | |
14229 | ||
14230 | /* It appears that the OSD doesn't return success unless the entire | |
14231 | buffer was written, return the write length on success. */ | |
14232 | ||
14233 | int Client::ll_write_block(Inode *in, uint64_t blockid, | |
14234 | char* buf, uint64_t offset, | |
14235 | uint64_t length, file_layout_t* layout, | |
14236 | uint64_t snapseq, uint32_t sync) | |
14237 | { | |
7c673cae | 14238 | vinodeno_t vino = ll_get_vino(in); |
7c673cae | 14239 | int r = 0; |
11fdf7f2 | 14240 | std::unique_ptr<C_SaferCond> onsafe = nullptr; |
f67539c2 TL |
14241 | |
14242 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); | |
14243 | if (!mref_reader.is_state_satisfied()) | |
14244 | return -CEPHFS_ENOTCONN; | |
14245 | ||
7c673cae | 14246 | if (length == 0) { |
f67539c2 | 14247 | return -CEPHFS_EINVAL; |
7c673cae FG |
14248 | } |
14249 | if (true || sync) { | |
14250 | /* if write is stable, the epilogue is waiting on | |
14251 | * flock */ | |
11fdf7f2 | 14252 | onsafe.reset(new C_SaferCond("Client::ll_write_block flock")); |
7c673cae FG |
14253 | } |
14254 | object_t oid = file_object_t(vino.ino, blockid); | |
14255 | SnapContext fakesnap; | |
11fdf7f2 TL |
14256 | ceph::bufferlist bl; |
14257 | if (length > 0) { | |
14258 | bl.push_back(buffer::copy(buf, length)); | |
14259 | } | |
7c673cae FG |
14260 | |
14261 | ldout(cct, 1) << "ll_block_write for " << vino.ino << "." << blockid | |
14262 | << dendl; | |
14263 | ||
14264 | fakesnap.seq = snapseq; | |
14265 | ||
14266 | /* lock just in time */ | |
7c673cae FG |
14267 | objecter->write(oid, |
14268 | object_locator_t(layout->pool_id), | |
14269 | offset, | |
14270 | length, | |
14271 | fakesnap, | |
14272 | bl, | |
14273 | ceph::real_clock::now(), | |
14274 | 0, | |
11fdf7f2 | 14275 | onsafe.get()); |
7c673cae | 14276 | |
11fdf7f2 TL |
14277 | if (nullptr != onsafe) { |
14278 | r = onsafe->wait(); | |
7c673cae FG |
14279 | } |
14280 | ||
14281 | if (r < 0) { | |
14282 | return r; | |
14283 | } else { | |
14284 | return length; | |
14285 | } | |
14286 | } | |
14287 | ||
14288 | int Client::ll_commit_blocks(Inode *in, | |
14289 | uint64_t offset, | |
14290 | uint64_t length) | |
14291 | { | |
7c673cae FG |
14292 | /* |
14293 | BarrierContext *bctx; | |
b32b8144 | 14294 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
14295 | uint64_t ino = vino.ino; |
14296 | ||
14297 | ldout(cct, 1) << "ll_commit_blocks for " << vino.ino << " from " | |
14298 | << offset << " to " << length << dendl; | |
14299 | ||
14300 | if (length == 0) { | |
f67539c2 | 14301 | return -CEPHFS_EINVAL; |
7c673cae FG |
14302 | } |
14303 | ||
f67539c2 | 14304 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14305 | map<uint64_t, BarrierContext*>::iterator p = barriers.find(ino); |
14306 | if (p != barriers.end()) { | |
14307 | barrier_interval civ(offset, offset + length); | |
14308 | p->second->commit_barrier(civ); | |
14309 | } | |
14310 | */ | |
14311 | return 0; | |
14312 | } | |
14313 | ||
14314 | int Client::ll_write(Fh *fh, loff_t off, loff_t len, const char *data) | |
14315 | { | |
7c673cae FG |
14316 | ldout(cct, 3) << "ll_write " << fh << " " << fh->inode->ino << " " << off << |
14317 | "~" << len << dendl; | |
14318 | tout(cct) << "ll_write" << std::endl; | |
f67539c2 | 14319 | tout(cct) << (uintptr_t)fh << std::endl; |
7c673cae FG |
14320 | tout(cct) << off << std::endl; |
14321 | tout(cct) << len << std::endl; | |
14322 | ||
f67539c2 TL |
14323 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14324 | if (!mref_reader.is_state_satisfied()) | |
14325 | return -CEPHFS_ENOTCONN; | |
181888fb | 14326 | |
11fdf7f2 TL |
14327 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
14328 | len = std::min(len, (loff_t)INT_MAX); | |
f67539c2 TL |
14329 | std::scoped_lock lock(client_lock); |
14330 | ||
7c673cae FG |
14331 | int r = _write(fh, off, len, data, NULL, 0); |
14332 | ldout(cct, 3) << "ll_write " << fh << " " << off << "~" << len << " = " << r | |
14333 | << dendl; | |
14334 | return r; | |
14335 | } | |
14336 | ||
11fdf7f2 TL |
14337 | int64_t Client::ll_writev(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) |
14338 | { | |
f67539c2 TL |
14339 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14340 | if (!mref_reader.is_state_satisfied()) | |
14341 | return -CEPHFS_ENOTCONN; | |
14342 | ||
14343 | std::unique_lock cl(client_lock); | |
14344 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, true, false, cl); | |
11fdf7f2 TL |
14345 | } |
14346 | ||
14347 | int64_t Client::ll_readv(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) | |
14348 | { | |
f67539c2 TL |
14349 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14350 | if (!mref_reader.is_state_satisfied()) | |
14351 | return -CEPHFS_ENOTCONN; | |
14352 | ||
14353 | std::unique_lock cl(client_lock); | |
14354 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, false, false, cl); | |
11fdf7f2 TL |
14355 | } |
14356 | ||
7c673cae FG |
14357 | int Client::ll_flush(Fh *fh) |
14358 | { | |
f67539c2 TL |
14359 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14360 | if (!mref_reader.is_state_satisfied()) | |
14361 | return -CEPHFS_ENOTCONN; | |
14362 | ||
7c673cae FG |
14363 | ldout(cct, 3) << "ll_flush " << fh << " " << fh->inode->ino << " " << dendl; |
14364 | tout(cct) << "ll_flush" << std::endl; | |
f67539c2 | 14365 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14366 | |
f67539c2 | 14367 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14368 | return _flush(fh); |
14369 | } | |
14370 | ||
14371 | int Client::ll_fsync(Fh *fh, bool syncdataonly) | |
14372 | { | |
f67539c2 TL |
14373 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14374 | if (!mref_reader.is_state_satisfied()) | |
14375 | return -CEPHFS_ENOTCONN; | |
14376 | ||
7c673cae FG |
14377 | ldout(cct, 3) << "ll_fsync " << fh << " " << fh->inode->ino << " " << dendl; |
14378 | tout(cct) << "ll_fsync" << std::endl; | |
f67539c2 | 14379 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14380 | |
f67539c2 | 14381 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14382 | int r = _fsync(fh, syncdataonly); |
14383 | if (r) { | |
14384 | // If we're returning an error, clear it from the FH | |
14385 | fh->take_async_err(); | |
14386 | } | |
14387 | return r; | |
14388 | } | |
14389 | ||
28e407b8 AA |
14390 | int Client::ll_sync_inode(Inode *in, bool syncdataonly) |
14391 | { | |
f67539c2 TL |
14392 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14393 | if (!mref_reader.is_state_satisfied()) | |
14394 | return -CEPHFS_ENOTCONN; | |
14395 | ||
28e407b8 AA |
14396 | ldout(cct, 3) << "ll_sync_inode " << *in << " " << dendl; |
14397 | tout(cct) << "ll_sync_inode" << std::endl; | |
f67539c2 | 14398 | tout(cct) << (uintptr_t)in << std::endl; |
28e407b8 | 14399 | |
f67539c2 | 14400 | std::scoped_lock lock(client_lock); |
28e407b8 AA |
14401 | return _fsync(in, syncdataonly); |
14402 | } | |
14403 | ||
7c673cae FG |
14404 | #ifdef FALLOC_FL_PUNCH_HOLE |
14405 | ||
14406 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) | |
14407 | { | |
f67539c2 TL |
14408 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
14409 | ||
7c673cae | 14410 | if (offset < 0 || length <= 0) |
f67539c2 | 14411 | return -CEPHFS_EINVAL; |
7c673cae FG |
14412 | |
14413 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
f67539c2 | 14414 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14415 | |
14416 | if ((mode & FALLOC_FL_PUNCH_HOLE) && !(mode & FALLOC_FL_KEEP_SIZE)) | |
f67539c2 | 14417 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14418 | |
14419 | Inode *in = fh->inode.get(); | |
14420 | ||
14421 | if (objecter->osdmap_pool_full(in->layout.pool_id) && | |
14422 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
f67539c2 | 14423 | return -CEPHFS_ENOSPC; |
7c673cae FG |
14424 | } |
14425 | ||
14426 | if (in->snapid != CEPH_NOSNAP) | |
f67539c2 | 14427 | return -CEPHFS_EROFS; |
7c673cae FG |
14428 | |
14429 | if ((fh->mode & CEPH_FILE_MODE_WR) == 0) | |
f67539c2 | 14430 | return -CEPHFS_EBADF; |
7c673cae FG |
14431 | |
14432 | uint64_t size = offset + length; | |
14433 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && | |
14434 | size > in->size && | |
11fdf7f2 | 14435 | is_quota_bytes_exceeded(in, size - in->size, fh->actor_perms)) { |
f67539c2 | 14436 | return -CEPHFS_EDQUOT; |
7c673cae FG |
14437 | } |
14438 | ||
14439 | int have; | |
f6b5b4d7 | 14440 | int r = get_caps(fh, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER, &have, -1); |
7c673cae FG |
14441 | if (r < 0) |
14442 | return r; | |
14443 | ||
11fdf7f2 | 14444 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
7c673cae FG |
14445 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
14446 | if (in->inline_version < CEPH_INLINE_NONE && | |
14447 | (have & CEPH_CAP_FILE_BUFFER)) { | |
14448 | bufferlist bl; | |
9f95a23c | 14449 | auto inline_iter = in->inline_data.cbegin(); |
7c673cae FG |
14450 | int len = in->inline_data.length(); |
14451 | if (offset < len) { | |
14452 | if (offset > 0) | |
9f95a23c | 14453 | inline_iter.copy(offset, bl); |
7c673cae FG |
14454 | int size = length; |
14455 | if (offset + size > len) | |
14456 | size = len - offset; | |
14457 | if (size > 0) | |
14458 | bl.append_zero(size); | |
9f95a23c TL |
14459 | if (offset + size < len) { |
14460 | inline_iter += size; | |
14461 | inline_iter.copy(len - offset - size, bl); | |
14462 | } | |
7c673cae FG |
14463 | in->inline_data = bl; |
14464 | in->inline_version++; | |
14465 | } | |
91327a77 | 14466 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14467 | in->change_attr++; |
28e407b8 | 14468 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14469 | } else { |
14470 | if (in->inline_version < CEPH_INLINE_NONE) { | |
11fdf7f2 TL |
14471 | onuninline.reset(new C_SaferCond("Client::_fallocate_uninline_data flock")); |
14472 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
14473 | } |
14474 | ||
11fdf7f2 | 14475 | C_SaferCond onfinish("Client::_punch_hole flock"); |
7c673cae | 14476 | |
7c673cae FG |
14477 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
14478 | ||
14479 | _invalidate_inode_cache(in, offset, length); | |
14480 | filer->zero(in->ino, &in->layout, | |
14481 | in->snaprealm->get_snap_context(), | |
14482 | offset, length, | |
14483 | ceph::real_clock::now(), | |
11fdf7f2 | 14484 | 0, true, &onfinish); |
91327a77 | 14485 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14486 | in->change_attr++; |
28e407b8 | 14487 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14488 | |
9f95a23c | 14489 | client_lock.unlock(); |
11fdf7f2 | 14490 | onfinish.wait(); |
9f95a23c | 14491 | client_lock.lock(); |
f67539c2 | 14492 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); |
7c673cae FG |
14493 | } |
14494 | } else if (!(mode & FALLOC_FL_KEEP_SIZE)) { | |
14495 | uint64_t size = offset + length; | |
14496 | if (size > in->size) { | |
14497 | in->size = size; | |
91327a77 | 14498 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 14499 | in->change_attr++; |
28e407b8 | 14500 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 14501 | |
11fdf7f2 | 14502 | if (is_quota_bytes_approaching(in, fh->actor_perms)) { |
7c673cae | 14503 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
14504 | } else if (is_max_size_approaching(in)) { |
14505 | check_caps(in, 0); | |
7c673cae FG |
14506 | } |
14507 | } | |
14508 | } | |
14509 | ||
11fdf7f2 | 14510 | if (nullptr != onuninline) { |
9f95a23c | 14511 | client_lock.unlock(); |
11fdf7f2 | 14512 | int ret = onuninline->wait(); |
9f95a23c | 14513 | client_lock.lock(); |
7c673cae | 14514 | |
f67539c2 | 14515 | if (ret >= 0 || ret == -CEPHFS_ECANCELED) { |
7c673cae FG |
14516 | in->inline_data.clear(); |
14517 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 14518 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
14519 | check_caps(in, 0); |
14520 | } else | |
11fdf7f2 | 14521 | r = ret; |
7c673cae FG |
14522 | } |
14523 | ||
14524 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
14525 | return r; | |
14526 | } | |
14527 | #else | |
14528 | ||
14529 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) | |
14530 | { | |
f67539c2 | 14531 | return -CEPHFS_EOPNOTSUPP; |
7c673cae FG |
14532 | } |
14533 | ||
14534 | #endif | |
14535 | ||
14536 | ||
11fdf7f2 | 14537 | int Client::ll_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
7c673cae | 14538 | { |
f67539c2 TL |
14539 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14540 | if (!mref_reader.is_state_satisfied()) | |
14541 | return -CEPHFS_ENOTCONN; | |
14542 | ||
11fdf7f2 TL |
14543 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << dendl; |
14544 | tout(cct) << __func__ << " " << mode << " " << offset << " " << length << std::endl; | |
f67539c2 | 14545 | tout(cct) << (uintptr_t)fh << std::endl; |
181888fb | 14546 | |
f67539c2 | 14547 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14548 | return _fallocate(fh, mode, offset, length); |
14549 | } | |
14550 | ||
14551 | int Client::fallocate(int fd, int mode, loff_t offset, loff_t length) | |
14552 | { | |
f67539c2 TL |
14553 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14554 | if (!mref_reader.is_state_satisfied()) | |
14555 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14556 | |
f67539c2 | 14557 | tout(cct) << __func__ << " " << " " << fd << mode << " " << offset << " " << length << std::endl; |
181888fb | 14558 | |
f67539c2 | 14559 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14560 | Fh *fh = get_filehandle(fd); |
14561 | if (!fh) | |
f67539c2 | 14562 | return -CEPHFS_EBADF; |
7c673cae FG |
14563 | #if defined(__linux__) && defined(O_PATH) |
14564 | if (fh->flags & O_PATH) | |
f67539c2 | 14565 | return -CEPHFS_EBADF; |
7c673cae FG |
14566 | #endif |
14567 | return _fallocate(fh, mode, offset, length); | |
14568 | } | |
14569 | ||
14570 | int Client::ll_release(Fh *fh) | |
14571 | { | |
f67539c2 TL |
14572 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14573 | if (!mref_reader.is_state_satisfied()) | |
14574 | return -CEPHFS_ENOTCONN; | |
91327a77 | 14575 | |
11fdf7f2 | 14576 | ldout(cct, 3) << __func__ << " (fh)" << fh << " " << fh->inode->ino << " " << |
7c673cae | 14577 | dendl; |
11fdf7f2 | 14578 | tout(cct) << __func__ << " (fh)" << std::endl; |
f67539c2 TL |
14579 | tout(cct) << (uintptr_t)fh << std::endl; |
14580 | ||
14581 | std::scoped_lock lock(client_lock); | |
7c673cae FG |
14582 | |
14583 | if (ll_unclosed_fh_set.count(fh)) | |
14584 | ll_unclosed_fh_set.erase(fh); | |
14585 | return _release_fh(fh); | |
14586 | } | |
14587 | ||
14588 | int Client::ll_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
14589 | { | |
f67539c2 TL |
14590 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14591 | if (!mref_reader.is_state_satisfied()) | |
14592 | return -CEPHFS_ENOTCONN; | |
7c673cae FG |
14593 | |
14594 | ldout(cct, 3) << "ll_getlk (fh)" << fh << " " << fh->inode->ino << dendl; | |
f67539c2 | 14595 | tout(cct) << "ll_getk (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14596 | |
f67539c2 | 14597 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14598 | return _getlk(fh, fl, owner); |
14599 | } | |
14600 | ||
14601 | int Client::ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
14602 | { | |
f67539c2 TL |
14603 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14604 | if (!mref_reader.is_state_satisfied()) | |
14605 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14606 | |
11fdf7f2 | 14607 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14608 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14609 | |
f67539c2 | 14610 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14611 | return _setlk(fh, fl, owner, sleep); |
14612 | } | |
14613 | ||
14614 | int Client::ll_flock(Fh *fh, int cmd, uint64_t owner) | |
14615 | { | |
f67539c2 TL |
14616 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14617 | if (!mref_reader.is_state_satisfied()) | |
14618 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14619 | |
11fdf7f2 | 14620 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
f67539c2 | 14621 | tout(cct) << __func__ << " (fh)" << (uintptr_t)fh << std::endl; |
181888fb | 14622 | |
f67539c2 | 14623 | std::scoped_lock lock(client_lock); |
7c673cae FG |
14624 | return _flock(fh, cmd, owner); |
14625 | } | |
14626 | ||
b32b8144 FG |
14627 | int Client::set_deleg_timeout(uint32_t timeout) |
14628 | { | |
f67539c2 | 14629 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14630 | |
14631 | /* | |
f67539c2 | 14632 | * The whole point is to prevent blocklisting so we must time out the |
b32b8144 FG |
14633 | * delegation before the session autoclose timeout kicks in. |
14634 | */ | |
14635 | if (timeout >= mdsmap->get_session_autoclose()) | |
f67539c2 | 14636 | return -CEPHFS_EINVAL; |
b32b8144 FG |
14637 | |
14638 | deleg_timeout = timeout; | |
14639 | return 0; | |
14640 | } | |
14641 | ||
14642 | int Client::ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv) | |
14643 | { | |
f67539c2 | 14644 | int ret = -CEPHFS_EINVAL; |
b32b8144 | 14645 | |
f67539c2 TL |
14646 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14647 | if (!mref_reader.is_state_satisfied()) | |
14648 | return -CEPHFS_ENOTCONN; | |
b32b8144 | 14649 | |
f67539c2 | 14650 | std::scoped_lock lock(client_lock); |
b32b8144 FG |
14651 | |
14652 | Inode *inode = fh->inode.get(); | |
14653 | ||
14654 | switch(cmd) { | |
14655 | case CEPH_DELEGATION_NONE: | |
14656 | inode->unset_deleg(fh); | |
14657 | ret = 0; | |
14658 | break; | |
14659 | default: | |
14660 | try { | |
14661 | ret = inode->set_deleg(fh, cmd, cb, priv); | |
11fdf7f2 | 14662 | } catch (std::bad_alloc&) { |
f67539c2 | 14663 | ret = -CEPHFS_ENOMEM; |
b32b8144 FG |
14664 | } |
14665 | break; | |
14666 | } | |
14667 | return ret; | |
14668 | } | |
14669 | ||
7c673cae FG |
14670 | class C_Client_RequestInterrupt : public Context { |
14671 | private: | |
14672 | Client *client; | |
14673 | MetaRequest *req; | |
14674 | public: | |
14675 | C_Client_RequestInterrupt(Client *c, MetaRequest *r) : client(c), req(r) { | |
14676 | req->get(); | |
14677 | } | |
14678 | void finish(int r) override { | |
f67539c2 | 14679 | std::scoped_lock l(client->client_lock); |
11fdf7f2 | 14680 | ceph_assert(req->head.op == CEPH_MDS_OP_SETFILELOCK); |
7c673cae FG |
14681 | client->_interrupt_filelock(req); |
14682 | client->put_request(req); | |
14683 | } | |
14684 | }; | |
14685 | ||
14686 | void Client::ll_interrupt(void *d) | |
14687 | { | |
14688 | MetaRequest *req = static_cast<MetaRequest*>(d); | |
11fdf7f2 TL |
14689 | ldout(cct, 3) << __func__ << " tid " << req->get_tid() << dendl; |
14690 | tout(cct) << __func__ << " tid " << req->get_tid() << std::endl; | |
7c673cae FG |
14691 | interrupt_finisher.queue(new C_Client_RequestInterrupt(this, req)); |
14692 | } | |
14693 | ||
14694 | // ========================================= | |
14695 | // layout | |
14696 | ||
14697 | // expose file layouts | |
14698 | ||
14699 | int Client::describe_layout(const char *relpath, file_layout_t *lp, | |
14700 | const UserPerm& perms) | |
14701 | { | |
f67539c2 TL |
14702 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14703 | if (!mref_reader.is_state_satisfied()) | |
14704 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14705 | |
f67539c2 | 14706 | std::scoped_lock lock(client_lock); |
181888fb | 14707 | |
7c673cae FG |
14708 | filepath path(relpath); |
14709 | InodeRef in; | |
14710 | int r = path_walk(path, &in, perms); | |
14711 | if (r < 0) | |
14712 | return r; | |
14713 | ||
14714 | *lp = in->layout; | |
14715 | ||
11fdf7f2 | 14716 | ldout(cct, 3) << __func__ << "(" << relpath << ") = 0" << dendl; |
7c673cae FG |
14717 | return 0; |
14718 | } | |
14719 | ||
14720 | int Client::fdescribe_layout(int fd, file_layout_t *lp) | |
14721 | { | |
f67539c2 TL |
14722 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14723 | if (!mref_reader.is_state_satisfied()) | |
14724 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14725 | |
f67539c2 | 14726 | std::scoped_lock lock(client_lock); |
181888fb | 14727 | |
7c673cae FG |
14728 | Fh *f = get_filehandle(fd); |
14729 | if (!f) | |
f67539c2 | 14730 | return -CEPHFS_EBADF; |
7c673cae FG |
14731 | Inode *in = f->inode.get(); |
14732 | ||
14733 | *lp = in->layout; | |
14734 | ||
11fdf7f2 | 14735 | ldout(cct, 3) << __func__ << "(" << fd << ") = 0" << dendl; |
7c673cae FG |
14736 | return 0; |
14737 | } | |
14738 | ||
d2e6a577 FG |
14739 | int64_t Client::get_default_pool_id() |
14740 | { | |
f67539c2 TL |
14741 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14742 | if (!mref_reader.is_state_satisfied()) | |
14743 | return -CEPHFS_ENOTCONN; | |
181888fb | 14744 | |
f67539c2 | 14745 | std::scoped_lock lock(client_lock); |
181888fb | 14746 | |
d2e6a577 FG |
14747 | /* first data pool is the default */ |
14748 | return mdsmap->get_first_data_pool(); | |
14749 | } | |
7c673cae FG |
14750 | |
14751 | // expose osdmap | |
14752 | ||
14753 | int64_t Client::get_pool_id(const char *pool_name) | |
14754 | { | |
f67539c2 TL |
14755 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14756 | if (!mref_reader.is_state_satisfied()) | |
14757 | return -CEPHFS_ENOTCONN; | |
181888fb | 14758 | |
f67539c2 | 14759 | std::scoped_lock lock(client_lock); |
181888fb | 14760 | |
7c673cae FG |
14761 | return objecter->with_osdmap(std::mem_fn(&OSDMap::lookup_pg_pool_name), |
14762 | pool_name); | |
14763 | } | |
14764 | ||
14765 | string Client::get_pool_name(int64_t pool) | |
14766 | { | |
f67539c2 TL |
14767 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14768 | if (!mref_reader.is_state_satisfied()) | |
181888fb FG |
14769 | return string(); |
14770 | ||
f67539c2 TL |
14771 | std::scoped_lock lock(client_lock); |
14772 | ||
7c673cae FG |
14773 | return objecter->with_osdmap([pool](const OSDMap& o) { |
14774 | return o.have_pg_pool(pool) ? o.get_pool_name(pool) : string(); | |
14775 | }); | |
14776 | } | |
14777 | ||
14778 | int Client::get_pool_replication(int64_t pool) | |
14779 | { | |
f67539c2 TL |
14780 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14781 | if (!mref_reader.is_state_satisfied()) | |
14782 | return -CEPHFS_ENOTCONN; | |
181888fb | 14783 | |
f67539c2 | 14784 | std::scoped_lock lock(client_lock); |
181888fb | 14785 | |
7c673cae | 14786 | return objecter->with_osdmap([pool](const OSDMap& o) { |
f67539c2 | 14787 | return o.have_pg_pool(pool) ? o.get_pg_pool(pool)->get_size() : -CEPHFS_ENOENT; |
7c673cae FG |
14788 | }); |
14789 | } | |
14790 | ||
14791 | int Client::get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds) | |
14792 | { | |
f67539c2 TL |
14793 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14794 | if (!mref_reader.is_state_satisfied()) | |
14795 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14796 | |
f67539c2 | 14797 | std::scoped_lock lock(client_lock); |
181888fb | 14798 | |
7c673cae FG |
14799 | Fh *f = get_filehandle(fd); |
14800 | if (!f) | |
f67539c2 | 14801 | return -CEPHFS_EBADF; |
7c673cae FG |
14802 | Inode *in = f->inode.get(); |
14803 | ||
14804 | vector<ObjectExtent> extents; | |
14805 | Striper::file_to_extents(cct, in->ino, &in->layout, off, 1, in->truncate_size, extents); | |
11fdf7f2 | 14806 | ceph_assert(extents.size() == 1); |
7c673cae FG |
14807 | |
14808 | objecter->with_osdmap([&](const OSDMap& o) { | |
14809 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
14810 | o.pg_to_acting_osds(pg, osds); | |
14811 | }); | |
14812 | ||
14813 | if (osds.empty()) | |
f67539c2 | 14814 | return -CEPHFS_EINVAL; |
7c673cae FG |
14815 | |
14816 | /* | |
14817 | * Return the remainder of the extent (stripe unit) | |
14818 | * | |
14819 | * If length = 1 is passed to Striper::file_to_extents we get a single | |
14820 | * extent back, but its length is one so we still need to compute the length | |
14821 | * to the end of the stripe unit. | |
14822 | * | |
14823 | * If length = su then we may get 1 or 2 objects back in the extents vector | |
14824 | * which would have to be examined. Even then, the offsets are local to the | |
14825 | * object, so matching up to the file offset is extra work. | |
14826 | * | |
14827 | * It seems simpler to stick with length = 1 and manually compute the | |
14828 | * remainder. | |
14829 | */ | |
14830 | if (len) { | |
14831 | uint64_t su = in->layout.stripe_unit; | |
14832 | *len = su - (off % su); | |
14833 | } | |
14834 | ||
14835 | return 0; | |
14836 | } | |
14837 | ||
14838 | int Client::get_osd_crush_location(int id, vector<pair<string, string> >& path) | |
14839 | { | |
f67539c2 TL |
14840 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14841 | if (!mref_reader.is_state_satisfied()) | |
14842 | return -CEPHFS_ENOTCONN; | |
181888fb | 14843 | |
f67539c2 | 14844 | std::scoped_lock lock(client_lock); |
181888fb | 14845 | |
7c673cae | 14846 | if (id < 0) |
f67539c2 | 14847 | return -CEPHFS_EINVAL; |
7c673cae FG |
14848 | return objecter->with_osdmap([&](const OSDMap& o) { |
14849 | return o.crush->get_full_location_ordered(id, path); | |
14850 | }); | |
14851 | } | |
14852 | ||
14853 | int Client::get_file_stripe_address(int fd, loff_t offset, | |
14854 | vector<entity_addr_t>& address) | |
14855 | { | |
f67539c2 TL |
14856 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14857 | if (!mref_reader.is_state_satisfied()) | |
14858 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14859 | |
f67539c2 | 14860 | std::scoped_lock lock(client_lock); |
181888fb | 14861 | |
7c673cae FG |
14862 | Fh *f = get_filehandle(fd); |
14863 | if (!f) | |
f67539c2 | 14864 | return -CEPHFS_EBADF; |
7c673cae FG |
14865 | Inode *in = f->inode.get(); |
14866 | ||
14867 | // which object? | |
14868 | vector<ObjectExtent> extents; | |
14869 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, 1, | |
14870 | in->truncate_size, extents); | |
11fdf7f2 | 14871 | ceph_assert(extents.size() == 1); |
7c673cae FG |
14872 | |
14873 | // now we have the object and its 'layout' | |
14874 | return objecter->with_osdmap([&](const OSDMap& o) { | |
14875 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
14876 | vector<int> osds; | |
14877 | o.pg_to_acting_osds(pg, osds); | |
14878 | if (osds.empty()) | |
f67539c2 | 14879 | return -CEPHFS_EINVAL; |
7c673cae | 14880 | for (unsigned i = 0; i < osds.size(); i++) { |
11fdf7f2 | 14881 | entity_addr_t addr = o.get_addrs(osds[i]).front(); |
7c673cae FG |
14882 | address.push_back(addr); |
14883 | } | |
14884 | return 0; | |
14885 | }); | |
14886 | } | |
14887 | ||
14888 | int Client::get_osd_addr(int osd, entity_addr_t& addr) | |
14889 | { | |
f67539c2 TL |
14890 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14891 | if (!mref_reader.is_state_satisfied()) | |
14892 | return -CEPHFS_ENOTCONN; | |
181888fb | 14893 | |
f67539c2 | 14894 | std::scoped_lock lock(client_lock); |
181888fb | 14895 | |
7c673cae FG |
14896 | return objecter->with_osdmap([&](const OSDMap& o) { |
14897 | if (!o.exists(osd)) | |
f67539c2 | 14898 | return -CEPHFS_ENOENT; |
7c673cae | 14899 | |
11fdf7f2 | 14900 | addr = o.get_addrs(osd).front(); |
7c673cae FG |
14901 | return 0; |
14902 | }); | |
14903 | } | |
14904 | ||
14905 | int Client::enumerate_layout(int fd, vector<ObjectExtent>& result, | |
14906 | loff_t length, loff_t offset) | |
14907 | { | |
f67539c2 TL |
14908 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14909 | if (!mref_reader.is_state_satisfied()) | |
14910 | return -CEPHFS_ENOTCONN; | |
7c673cae | 14911 | |
f67539c2 | 14912 | std::scoped_lock lock(client_lock); |
181888fb | 14913 | |
7c673cae FG |
14914 | Fh *f = get_filehandle(fd); |
14915 | if (!f) | |
f67539c2 | 14916 | return -CEPHFS_EBADF; |
7c673cae FG |
14917 | Inode *in = f->inode.get(); |
14918 | ||
14919 | // map to a list of extents | |
14920 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, length, in->truncate_size, result); | |
14921 | ||
11fdf7f2 | 14922 | ldout(cct, 3) << __func__ << "(" << fd << ", " << length << ", " << offset << ") = 0" << dendl; |
7c673cae FG |
14923 | return 0; |
14924 | } | |
14925 | ||
14926 | ||
f67539c2 | 14927 | /* find an osd with the same ip. -CEPHFS_ENXIO if none. */ |
7c673cae FG |
14928 | int Client::get_local_osd() |
14929 | { | |
f67539c2 TL |
14930 | RWRef_t mref_reader(mount_state, CLIENT_MOUNTING); |
14931 | if (!mref_reader.is_state_satisfied()) | |
14932 | return -CEPHFS_ENOTCONN; | |
181888fb | 14933 | |
f67539c2 | 14934 | std::scoped_lock lock(client_lock); |
181888fb | 14935 | |
7c673cae FG |
14936 | objecter->with_osdmap([this](const OSDMap& o) { |
14937 | if (o.get_epoch() != local_osd_epoch) { | |
11fdf7f2 | 14938 | local_osd = o.find_osd_on_ip(messenger->get_myaddrs().front()); |
7c673cae FG |
14939 | local_osd_epoch = o.get_epoch(); |
14940 | } | |
14941 | }); | |
14942 | return local_osd; | |
14943 | } | |
14944 | ||
14945 | ||
14946 | ||
14947 | ||
14948 | ||
14949 | ||
14950 | // =============================== | |
14951 | ||
14952 | void Client::ms_handle_connect(Connection *con) | |
14953 | { | |
11fdf7f2 | 14954 | ldout(cct, 10) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
14955 | } |
14956 | ||
14957 | bool Client::ms_handle_reset(Connection *con) | |
14958 | { | |
11fdf7f2 | 14959 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
14960 | return false; |
14961 | } | |
14962 | ||
14963 | void Client::ms_handle_remote_reset(Connection *con) | |
14964 | { | |
f67539c2 | 14965 | std::scoped_lock lock(client_lock); |
11fdf7f2 | 14966 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
14967 | switch (con->get_peer_type()) { |
14968 | case CEPH_ENTITY_TYPE_MDS: | |
14969 | { | |
14970 | // kludge to figure out which mds this is; fixme with a Connection* state | |
14971 | mds_rank_t mds = MDS_RANK_NONE; | |
14972 | MetaSession *s = NULL; | |
11fdf7f2 | 14973 | for (auto &p : mds_sessions) { |
b3b6e05e | 14974 | if (mdsmap->have_inst(p.first) && mdsmap->get_addrs(p.first) == con->get_peer_addrs()) { |
11fdf7f2 TL |
14975 | mds = p.first; |
14976 | s = &p.second; | |
7c673cae FG |
14977 | } |
14978 | } | |
14979 | if (mds >= 0) { | |
d2e6a577 | 14980 | assert (s != NULL); |
7c673cae FG |
14981 | switch (s->state) { |
14982 | case MetaSession::STATE_CLOSING: | |
14983 | ldout(cct, 1) << "reset from mds we were closing; we'll call that closed" << dendl; | |
14984 | _closed_mds_session(s); | |
14985 | break; | |
14986 | ||
14987 | case MetaSession::STATE_OPENING: | |
14988 | { | |
14989 | ldout(cct, 1) << "reset from mds we were opening; retrying" << dendl; | |
14990 | list<Context*> waiters; | |
14991 | waiters.swap(s->waiting_for_open); | |
14992 | _closed_mds_session(s); | |
14993 | MetaSession *news = _get_or_open_mds_session(mds); | |
14994 | news->waiting_for_open.swap(waiters); | |
14995 | } | |
14996 | break; | |
14997 | ||
14998 | case MetaSession::STATE_OPEN: | |
14999 | { | |
f67539c2 | 15000 | objecter->maybe_request_map(); /* to check if we are blocklisted */ |
f6b5b4d7 | 15001 | if (cct->_conf.get_val<bool>("client_reconnect_stale")) { |
7c673cae FG |
15002 | ldout(cct, 1) << "reset from mds we were open; close mds session for reconnect" << dendl; |
15003 | _closed_mds_session(s); | |
15004 | } else { | |
15005 | ldout(cct, 1) << "reset from mds we were open; mark session as stale" << dendl; | |
15006 | s->state = MetaSession::STATE_STALE; | |
15007 | } | |
15008 | } | |
15009 | break; | |
15010 | ||
15011 | case MetaSession::STATE_NEW: | |
15012 | case MetaSession::STATE_CLOSED: | |
15013 | default: | |
15014 | break; | |
15015 | } | |
15016 | } | |
15017 | } | |
15018 | break; | |
15019 | } | |
15020 | } | |
15021 | ||
15022 | bool Client::ms_handle_refused(Connection *con) | |
15023 | { | |
11fdf7f2 | 15024 | ldout(cct, 1) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
15025 | return false; |
15026 | } | |
15027 | ||
7c673cae FG |
15028 | Inode *Client::get_quota_root(Inode *in, const UserPerm& perms) |
15029 | { | |
11fdf7f2 TL |
15030 | Inode *quota_in = root_ancestor; |
15031 | SnapRealm *realm = in->snaprealm; | |
15032 | while (realm) { | |
15033 | ldout(cct, 10) << __func__ << " realm " << realm->ino << dendl; | |
15034 | if (realm->ino != in->ino) { | |
15035 | auto p = inode_map.find(vinodeno_t(realm->ino, CEPH_NOSNAP)); | |
15036 | if (p == inode_map.end()) | |
15037 | break; | |
7c673cae | 15038 | |
11fdf7f2 TL |
15039 | if (p->second->quota.is_enable()) { |
15040 | quota_in = p->second; | |
15041 | break; | |
7c673cae | 15042 | } |
7c673cae | 15043 | } |
11fdf7f2 | 15044 | realm = realm->pparent; |
7c673cae | 15045 | } |
11fdf7f2 TL |
15046 | ldout(cct, 10) << __func__ << " " << in->vino() << " -> " << quota_in->vino() << dendl; |
15047 | return quota_in; | |
7c673cae FG |
15048 | } |
15049 | ||
15050 | /** | |
15051 | * Traverse quota ancestors of the Inode, return true | |
15052 | * if any of them passes the passed function | |
15053 | */ | |
15054 | bool Client::check_quota_condition(Inode *in, const UserPerm& perms, | |
15055 | std::function<bool (const Inode &in)> test) | |
15056 | { | |
15057 | while (true) { | |
11fdf7f2 | 15058 | ceph_assert(in != NULL); |
7c673cae FG |
15059 | if (test(*in)) { |
15060 | return true; | |
15061 | } | |
15062 | ||
15063 | if (in == root_ancestor) { | |
15064 | // We're done traversing, drop out | |
15065 | return false; | |
15066 | } else { | |
15067 | // Continue up the tree | |
15068 | in = get_quota_root(in, perms); | |
15069 | } | |
15070 | } | |
15071 | ||
15072 | return false; | |
15073 | } | |
15074 | ||
15075 | bool Client::is_quota_files_exceeded(Inode *in, const UserPerm& perms) | |
15076 | { | |
15077 | return check_quota_condition(in, perms, | |
15078 | [](const Inode &in) { | |
15079 | return in.quota.max_files && in.rstat.rsize() >= in.quota.max_files; | |
15080 | }); | |
15081 | } | |
15082 | ||
15083 | bool Client::is_quota_bytes_exceeded(Inode *in, int64_t new_bytes, | |
11fdf7f2 | 15084 | const UserPerm& perms) |
7c673cae FG |
15085 | { |
15086 | return check_quota_condition(in, perms, | |
11fdf7f2 | 15087 | [&new_bytes](const Inode &in) { |
7c673cae FG |
15088 | return in.quota.max_bytes && (in.rstat.rbytes + new_bytes) |
15089 | > in.quota.max_bytes; | |
15090 | }); | |
15091 | } | |
15092 | ||
11fdf7f2 | 15093 | bool Client::is_quota_bytes_approaching(Inode *in, const UserPerm& perms) |
7c673cae | 15094 | { |
9f95a23c TL |
15095 | ceph_assert(in->size >= in->reported_size); |
15096 | const uint64_t size = in->size - in->reported_size; | |
11fdf7f2 | 15097 | return check_quota_condition(in, perms, |
9f95a23c | 15098 | [&size](const Inode &in) { |
11fdf7f2 TL |
15099 | if (in.quota.max_bytes) { |
15100 | if (in.rstat.rbytes >= in.quota.max_bytes) { | |
15101 | return true; | |
15102 | } | |
15103 | ||
11fdf7f2 | 15104 | const uint64_t space = in.quota.max_bytes - in.rstat.rbytes; |
11fdf7f2 TL |
15105 | return (space >> 4) < size; |
15106 | } else { | |
15107 | return false; | |
15108 | } | |
15109 | }); | |
7c673cae FG |
15110 | } |
15111 | ||
15112 | enum { | |
15113 | POOL_CHECKED = 1, | |
15114 | POOL_CHECKING = 2, | |
15115 | POOL_READ = 4, | |
15116 | POOL_WRITE = 8, | |
15117 | }; | |
15118 | ||
15119 | int Client::check_pool_perm(Inode *in, int need) | |
15120 | { | |
f67539c2 TL |
15121 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
15122 | ||
7c673cae FG |
15123 | if (!cct->_conf->client_check_pool_perm) |
15124 | return 0; | |
15125 | ||
f67539c2 TL |
15126 | /* Only need to do this for regular files */ |
15127 | if (!in->is_file()) | |
15128 | return 0; | |
15129 | ||
7c673cae FG |
15130 | int64_t pool_id = in->layout.pool_id; |
15131 | std::string pool_ns = in->layout.pool_ns; | |
15132 | std::pair<int64_t, std::string> perm_key(pool_id, pool_ns); | |
15133 | int have = 0; | |
15134 | while (true) { | |
15135 | auto it = pool_perms.find(perm_key); | |
15136 | if (it == pool_perms.end()) | |
15137 | break; | |
15138 | if (it->second == POOL_CHECKING) { | |
15139 | // avoid concurrent checkings | |
15140 | wait_on_list(waiting_for_pool_perm); | |
15141 | } else { | |
15142 | have = it->second; | |
11fdf7f2 | 15143 | ceph_assert(have & POOL_CHECKED); |
7c673cae FG |
15144 | break; |
15145 | } | |
15146 | } | |
15147 | ||
15148 | if (!have) { | |
15149 | if (in->snapid != CEPH_NOSNAP) { | |
15150 | // pool permission check needs to write to the first object. But for snapshot, | |
15151 | // head of the first object may have alread been deleted. To avoid creating | |
15152 | // orphan object, skip the check for now. | |
15153 | return 0; | |
15154 | } | |
15155 | ||
15156 | pool_perms[perm_key] = POOL_CHECKING; | |
15157 | ||
15158 | char oid_buf[32]; | |
15159 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (unsigned long long)in->ino); | |
15160 | object_t oid = oid_buf; | |
15161 | ||
15162 | SnapContext nullsnapc; | |
15163 | ||
15164 | C_SaferCond rd_cond; | |
15165 | ObjectOperation rd_op; | |
f67539c2 | 15166 | rd_op.stat(nullptr, nullptr, nullptr); |
7c673cae FG |
15167 | |
15168 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), rd_op, | |
15169 | nullsnapc, ceph::real_clock::now(), 0, &rd_cond); | |
15170 | ||
15171 | C_SaferCond wr_cond; | |
15172 | ObjectOperation wr_op; | |
15173 | wr_op.create(true); | |
15174 | ||
15175 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), wr_op, | |
15176 | nullsnapc, ceph::real_clock::now(), 0, &wr_cond); | |
15177 | ||
9f95a23c | 15178 | client_lock.unlock(); |
7c673cae FG |
15179 | int rd_ret = rd_cond.wait(); |
15180 | int wr_ret = wr_cond.wait(); | |
9f95a23c | 15181 | client_lock.lock(); |
7c673cae FG |
15182 | |
15183 | bool errored = false; | |
15184 | ||
f67539c2 | 15185 | if (rd_ret == 0 || rd_ret == -CEPHFS_ENOENT) |
7c673cae | 15186 | have |= POOL_READ; |
f67539c2 | 15187 | else if (rd_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15188 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15189 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15190 | errored = true; | |
15191 | } | |
15192 | ||
f67539c2 | 15193 | if (wr_ret == 0 || wr_ret == -CEPHFS_EEXIST) |
7c673cae | 15194 | have |= POOL_WRITE; |
f67539c2 | 15195 | else if (wr_ret != -CEPHFS_EPERM) { |
11fdf7f2 | 15196 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
15197 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
15198 | errored = true; | |
15199 | } | |
15200 | ||
15201 | if (errored) { | |
15202 | // Indeterminate: erase CHECKING state so that subsequent calls re-check. | |
15203 | // Raise EIO because actual error code might be misleading for | |
15204 | // userspace filesystem user. | |
15205 | pool_perms.erase(perm_key); | |
15206 | signal_cond_list(waiting_for_pool_perm); | |
f67539c2 | 15207 | return -CEPHFS_EIO; |
7c673cae FG |
15208 | } |
15209 | ||
15210 | pool_perms[perm_key] = have | POOL_CHECKED; | |
15211 | signal_cond_list(waiting_for_pool_perm); | |
15212 | } | |
15213 | ||
15214 | if ((need & CEPH_CAP_FILE_RD) && !(have & POOL_READ)) { | |
11fdf7f2 | 15215 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15216 | << " need " << ccap_string(need) << ", but no read perm" << dendl; |
f67539c2 | 15217 | return -CEPHFS_EPERM; |
7c673cae FG |
15218 | } |
15219 | if ((need & CEPH_CAP_FILE_WR) && !(have & POOL_WRITE)) { | |
11fdf7f2 | 15220 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae | 15221 | << " need " << ccap_string(need) << ", but no write perm" << dendl; |
f67539c2 | 15222 | return -CEPHFS_EPERM; |
7c673cae FG |
15223 | } |
15224 | ||
15225 | return 0; | |
15226 | } | |
15227 | ||
15228 | int Client::_posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want) | |
15229 | { | |
15230 | if (acl_type == POSIX_ACL) { | |
15231 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15232 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15233 | ||
15234 | return posix_acl_permits(access_acl, in->uid, in->gid, perms, want); | |
15235 | } | |
15236 | } | |
f67539c2 | 15237 | return -CEPHFS_EAGAIN; |
7c673cae FG |
15238 | } |
15239 | ||
15240 | int Client::_posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms) | |
15241 | { | |
15242 | if (acl_type == NO_ACL) | |
15243 | return 0; | |
15244 | ||
15245 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
15246 | if (r < 0) | |
15247 | goto out; | |
15248 | ||
15249 | if (acl_type == POSIX_ACL) { | |
15250 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
15251 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
15252 | bufferptr acl(access_acl.c_str(), access_acl.length()); | |
15253 | r = posix_acl_access_chmod(acl, mode); | |
15254 | if (r < 0) | |
15255 | goto out; | |
15256 | r = _do_setxattr(in, ACL_EA_ACCESS, acl.c_str(), acl.length(), 0, perms); | |
15257 | } else { | |
15258 | r = 0; | |
15259 | } | |
15260 | } | |
15261 | out: | |
15262 | ldout(cct, 10) << __func__ << " ino " << in->ino << " result=" << r << dendl; | |
15263 | return r; | |
15264 | } | |
15265 | ||
15266 | int Client::_posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl, | |
15267 | const UserPerm& perms) | |
15268 | { | |
15269 | if (acl_type == NO_ACL) | |
15270 | return 0; | |
15271 | ||
15272 | if (S_ISLNK(*mode)) | |
15273 | return 0; | |
15274 | ||
15275 | int r = _getattr(dir, CEPH_STAT_CAP_XATTR, perms, dir->xattr_version == 0); | |
15276 | if (r < 0) | |
15277 | goto out; | |
15278 | ||
15279 | if (acl_type == POSIX_ACL) { | |
15280 | if (dir->xattrs.count(ACL_EA_DEFAULT)) { | |
15281 | map<string, bufferptr> xattrs; | |
15282 | ||
15283 | const bufferptr& default_acl = dir->xattrs[ACL_EA_DEFAULT]; | |
15284 | bufferptr acl(default_acl.c_str(), default_acl.length()); | |
15285 | r = posix_acl_inherit_mode(acl, mode); | |
15286 | if (r < 0) | |
15287 | goto out; | |
15288 | ||
15289 | if (r > 0) { | |
15290 | r = posix_acl_equiv_mode(acl.c_str(), acl.length(), mode); | |
15291 | if (r < 0) | |
15292 | goto out; | |
15293 | if (r > 0) | |
15294 | xattrs[ACL_EA_ACCESS] = acl; | |
15295 | } | |
15296 | ||
15297 | if (S_ISDIR(*mode)) | |
15298 | xattrs[ACL_EA_DEFAULT] = dir->xattrs[ACL_EA_DEFAULT]; | |
15299 | ||
15300 | r = xattrs.size(); | |
15301 | if (r > 0) | |
11fdf7f2 | 15302 | encode(xattrs, xattrs_bl); |
7c673cae FG |
15303 | } else { |
15304 | if (umask_cb) | |
15305 | *mode &= ~umask_cb(callback_handle); | |
15306 | r = 0; | |
15307 | } | |
15308 | } | |
15309 | out: | |
15310 | ldout(cct, 10) << __func__ << " dir ino " << dir->ino << " result=" << r << dendl; | |
15311 | return r; | |
15312 | } | |
15313 | ||
15314 | void Client::set_filer_flags(int flags) | |
15315 | { | |
f67539c2 | 15316 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15317 | ceph_assert(flags == 0 || |
7c673cae FG |
15318 | flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
15319 | objecter->add_global_op_flags(flags); | |
15320 | } | |
15321 | ||
15322 | void Client::clear_filer_flags(int flags) | |
15323 | { | |
f67539c2 | 15324 | std::scoped_lock l(client_lock); |
11fdf7f2 | 15325 | ceph_assert(flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
7c673cae FG |
15326 | objecter->clear_global_op_flag(flags); |
15327 | } | |
15328 | ||
11fdf7f2 TL |
15329 | // called before mount |
15330 | void Client::set_uuid(const std::string& uuid) | |
15331 | { | |
f67539c2 TL |
15332 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15333 | ceph_assert(iref_reader.is_state_satisfied()); | |
15334 | ||
15335 | std::scoped_lock l(client_lock); | |
11fdf7f2 TL |
15336 | assert(!uuid.empty()); |
15337 | ||
15338 | metadata["uuid"] = uuid; | |
15339 | _close_sessions(); | |
15340 | } | |
15341 | ||
15342 | // called before mount. 0 means infinite | |
15343 | void Client::set_session_timeout(unsigned timeout) | |
15344 | { | |
f67539c2 TL |
15345 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15346 | ceph_assert(iref_reader.is_state_satisfied()); | |
15347 | ||
15348 | std::scoped_lock l(client_lock); | |
11fdf7f2 TL |
15349 | |
15350 | metadata["timeout"] = stringify(timeout); | |
15351 | } | |
15352 | ||
15353 | // called before mount | |
15354 | int Client::start_reclaim(const std::string& uuid, unsigned flags, | |
15355 | const std::string& fs_name) | |
15356 | { | |
f67539c2 TL |
15357 | RWRef_t iref_reader(initialize_state, CLIENT_INITIALIZED); |
15358 | if (!iref_reader.is_state_satisfied()) | |
15359 | return -CEPHFS_ENOTCONN; | |
11fdf7f2 TL |
15360 | |
15361 | if (uuid.empty()) | |
f67539c2 | 15362 | return -CEPHFS_EINVAL; |
11fdf7f2 | 15363 | |
f67539c2 | 15364 | std::unique_lock l(client_lock); |
11fdf7f2 TL |
15365 | { |
15366 | auto it = metadata.find("uuid"); | |
15367 | if (it != metadata.end() && it->second == uuid) | |
f67539c2 | 15368 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15369 | } |
15370 | ||
15371 | int r = subscribe_mdsmap(fs_name); | |
15372 | if (r < 0) { | |
15373 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
15374 | return r; | |
15375 | } | |
15376 | ||
15377 | if (metadata.empty()) | |
15378 | populate_metadata(""); | |
15379 | ||
15380 | while (mdsmap->get_epoch() == 0) | |
15381 | wait_on_list(waiting_for_mdsmap); | |
15382 | ||
15383 | reclaim_errno = 0; | |
15384 | for (unsigned mds = 0; mds < mdsmap->get_num_in_mds(); ) { | |
15385 | if (!mdsmap->is_up(mds)) { | |
15386 | ldout(cct, 10) << "mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
15387 | wait_on_list(waiting_for_mdsmap); | |
15388 | continue; | |
15389 | } | |
15390 | ||
15391 | MetaSession *session; | |
15392 | if (!have_open_session(mds)) { | |
15393 | session = _get_or_open_mds_session(mds); | |
f6b5b4d7 | 15394 | if (session->state == MetaSession::STATE_REJECTED) |
f67539c2 | 15395 | return -CEPHFS_EPERM; |
11fdf7f2 TL |
15396 | if (session->state != MetaSession::STATE_OPENING) { |
15397 | // umounting? | |
f67539c2 | 15398 | return -CEPHFS_EINVAL; |
11fdf7f2 TL |
15399 | } |
15400 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
15401 | wait_on_context_list(session->waiting_for_open); | |
11fdf7f2 TL |
15402 | continue; |
15403 | } | |
15404 | ||
15405 | session = &mds_sessions.at(mds); | |
15406 | if (!session->mds_features.test(CEPHFS_FEATURE_RECLAIM_CLIENT)) | |
f67539c2 | 15407 | return -CEPHFS_EOPNOTSUPP; |
11fdf7f2 TL |
15408 | |
15409 | if (session->reclaim_state == MetaSession::RECLAIM_NULL || | |
15410 | session->reclaim_state == MetaSession::RECLAIMING) { | |
15411 | session->reclaim_state = MetaSession::RECLAIMING; | |
9f95a23c | 15412 | auto m = make_message<MClientReclaim>(uuid, flags); |
11fdf7f2 TL |
15413 | session->con->send_message2(std::move(m)); |
15414 | wait_on_list(waiting_for_reclaim); | |
15415 | } else if (session->reclaim_state == MetaSession::RECLAIM_FAIL) { | |
f67539c2 | 15416 | return reclaim_errno ? : -CEPHFS_ENOTRECOVERABLE; |
11fdf7f2 TL |
15417 | } else { |
15418 | mds++; | |
15419 | } | |
15420 | } | |
15421 | ||
15422 | // didn't find target session in any mds | |
15423 | if (reclaim_target_addrs.empty()) { | |
15424 | if (flags & CEPH_RECLAIM_RESET) | |
f67539c2 TL |
15425 | return -CEPHFS_ENOENT; |
15426 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15427 | } |
15428 | ||
15429 | if (flags & CEPH_RECLAIM_RESET) | |
15430 | return 0; | |
15431 | ||
f67539c2 TL |
15432 | // use blocklist to check if target session was killed |
15433 | // (config option mds_session_blocklist_on_evict needs to be true) | |
15434 | ldout(cct, 10) << __func__ << ": waiting for OSD epoch " << reclaim_osd_epoch << dendl; | |
15435 | bs::error_code ec; | |
15436 | l.unlock(); | |
15437 | objecter->wait_for_map(reclaim_osd_epoch, ca::use_blocked[ec]); | |
15438 | l.lock(); | |
11fdf7f2 | 15439 | |
f67539c2 TL |
15440 | if (ec) |
15441 | return ceph::from_error_code(ec); | |
15442 | ||
15443 | bool blocklisted = objecter->with_osdmap( | |
11fdf7f2 | 15444 | [this](const OSDMap &osd_map) -> bool { |
f67539c2 | 15445 | return osd_map.is_blocklisted(reclaim_target_addrs); |
11fdf7f2 | 15446 | }); |
f67539c2 TL |
15447 | if (blocklisted) |
15448 | return -CEPHFS_ENOTRECOVERABLE; | |
11fdf7f2 TL |
15449 | |
15450 | metadata["reclaiming_uuid"] = uuid; | |
15451 | return 0; | |
15452 | } | |
15453 | ||
15454 | void Client::finish_reclaim() | |
15455 | { | |
15456 | auto it = metadata.find("reclaiming_uuid"); | |
15457 | if (it == metadata.end()) { | |
15458 | for (auto &p : mds_sessions) | |
15459 | p.second.reclaim_state = MetaSession::RECLAIM_NULL; | |
15460 | return; | |
15461 | } | |
15462 | ||
15463 | for (auto &p : mds_sessions) { | |
15464 | p.second.reclaim_state = MetaSession::RECLAIM_NULL; | |
9f95a23c | 15465 | auto m = make_message<MClientReclaim>("", MClientReclaim::FLAG_FINISH); |
11fdf7f2 TL |
15466 | p.second.con->send_message2(std::move(m)); |
15467 | } | |
15468 | ||
15469 | metadata["uuid"] = it->second; | |
15470 | metadata.erase(it); | |
15471 | } | |
15472 | ||
15473 | void Client::handle_client_reclaim_reply(const MConstRef<MClientReclaimReply>& reply) | |
15474 | { | |
15475 | mds_rank_t from = mds_rank_t(reply->get_source().num()); | |
15476 | ldout(cct, 10) << __func__ << " " << *reply << " from mds." << from << dendl; | |
15477 | ||
f67539c2 | 15478 | std::scoped_lock cl(client_lock); |
11fdf7f2 TL |
15479 | MetaSession *session = _get_mds_session(from, reply->get_connection().get()); |
15480 | if (!session) { | |
15481 | ldout(cct, 10) << " discarding reclaim reply from sessionless mds." << from << dendl; | |
15482 | return; | |
15483 | } | |
15484 | ||
15485 | if (reply->get_result() >= 0) { | |
15486 | session->reclaim_state = MetaSession::RECLAIM_OK; | |
15487 | if (reply->get_epoch() > reclaim_osd_epoch) | |
15488 | reclaim_osd_epoch = reply->get_epoch(); | |
15489 | if (!reply->get_addrs().empty()) | |
15490 | reclaim_target_addrs = reply->get_addrs(); | |
15491 | } else { | |
15492 | session->reclaim_state = MetaSession::RECLAIM_FAIL; | |
15493 | reclaim_errno = reply->get_result(); | |
15494 | } | |
15495 | ||
15496 | signal_cond_list(waiting_for_reclaim); | |
15497 | } | |
15498 | ||
7c673cae FG |
15499 | /** |
15500 | * This is included in cap release messages, to cause | |
15501 | * the MDS to wait until this OSD map epoch. It is necessary | |
15502 | * in corner cases where we cancel RADOS ops, so that | |
15503 | * nobody else tries to do IO to the same objects in | |
15504 | * the same epoch as the cancelled ops. | |
15505 | */ | |
15506 | void Client::set_cap_epoch_barrier(epoch_t e) | |
15507 | { | |
15508 | ldout(cct, 5) << __func__ << " epoch = " << e << dendl; | |
15509 | cap_epoch_barrier = e; | |
15510 | } | |
15511 | ||
15512 | const char** Client::get_tracked_conf_keys() const | |
15513 | { | |
15514 | static const char* keys[] = { | |
15515 | "client_cache_size", | |
15516 | "client_cache_mid", | |
15517 | "client_acl_type", | |
b32b8144 FG |
15518 | "client_deleg_timeout", |
15519 | "client_deleg_break_on_open", | |
f67539c2 TL |
15520 | "client_oc_size", |
15521 | "client_oc_max_objects", | |
15522 | "client_oc_max_dirty", | |
15523 | "client_oc_target_dirty", | |
15524 | "client_oc_max_dirty_age", | |
7c673cae FG |
15525 | NULL |
15526 | }; | |
15527 | return keys; | |
15528 | } | |
15529 | ||
11fdf7f2 | 15530 | void Client::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
15531 | const std::set <std::string> &changed) |
15532 | { | |
f67539c2 | 15533 | std::scoped_lock lock(client_lock); |
7c673cae | 15534 | |
181888fb | 15535 | if (changed.count("client_cache_mid")) { |
7c673cae FG |
15536 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
15537 | } | |
15538 | if (changed.count("client_acl_type")) { | |
15539 | acl_type = NO_ACL; | |
15540 | if (cct->_conf->client_acl_type == "posix_acl") | |
15541 | acl_type = POSIX_ACL; | |
15542 | } | |
f67539c2 TL |
15543 | if (changed.count("client_oc_size")) { |
15544 | objectcacher->set_max_size(cct->_conf->client_oc_size); | |
15545 | } | |
15546 | if (changed.count("client_oc_max_objects")) { | |
15547 | objectcacher->set_max_objects(cct->_conf->client_oc_max_objects); | |
15548 | } | |
15549 | if (changed.count("client_oc_max_dirty")) { | |
15550 | objectcacher->set_max_dirty(cct->_conf->client_oc_max_dirty); | |
15551 | } | |
15552 | if (changed.count("client_oc_target_dirty")) { | |
15553 | objectcacher->set_target_dirty(cct->_conf->client_oc_target_dirty); | |
15554 | } | |
15555 | if (changed.count("client_oc_max_dirty_age")) { | |
15556 | objectcacher->set_max_dirty_age(cct->_conf->client_oc_max_dirty_age); | |
15557 | } | |
7c673cae FG |
15558 | } |
15559 | ||
7c673cae FG |
15560 | void intrusive_ptr_add_ref(Inode *in) |
15561 | { | |
b3b6e05e | 15562 | in->iget(); |
7c673cae | 15563 | } |
f67539c2 | 15564 | |
7c673cae FG |
15565 | void intrusive_ptr_release(Inode *in) |
15566 | { | |
15567 | in->client->put_inode(in); | |
15568 | } | |
15569 | ||
15570 | mds_rank_t Client::_get_random_up_mds() const | |
15571 | { | |
9f95a23c | 15572 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
15573 | |
15574 | std::set<mds_rank_t> up; | |
15575 | mdsmap->get_up_mds_set(up); | |
15576 | ||
15577 | if (up.empty()) | |
15578 | return MDS_RANK_NONE; | |
15579 | std::set<mds_rank_t>::const_iterator p = up.begin(); | |
15580 | for (int n = rand() % up.size(); n; n--) | |
15581 | ++p; | |
15582 | return *p; | |
15583 | } | |
15584 | ||
15585 | ||
f67539c2 TL |
15586 | StandaloneClient::StandaloneClient(Messenger *m, MonClient *mc, |
15587 | boost::asio::io_context& ictx) | |
15588 | : Client(m, mc, new Objecter(m->cct, m, mc, ictx)) | |
7c673cae FG |
15589 | { |
15590 | monclient->set_messenger(m); | |
15591 | objecter->set_client_incarnation(0); | |
15592 | } | |
15593 | ||
15594 | StandaloneClient::~StandaloneClient() | |
15595 | { | |
15596 | delete objecter; | |
15597 | objecter = nullptr; | |
15598 | } | |
15599 | ||
15600 | int StandaloneClient::init() | |
15601 | { | |
f67539c2 TL |
15602 | RWRef_t iref_writer(initialize_state, CLIENT_INITIALIZING, false); |
15603 | ceph_assert(iref_writer.is_first_writer()); | |
15604 | ||
e306af50 | 15605 | _pre_init(); |
7c673cae FG |
15606 | objecter->init(); |
15607 | ||
9f95a23c | 15608 | client_lock.lock(); |
7c673cae FG |
15609 | |
15610 | messenger->add_dispatcher_tail(objecter); | |
15611 | messenger->add_dispatcher_tail(this); | |
15612 | ||
15613 | monclient->set_want_keys(CEPH_ENTITY_TYPE_MDS | CEPH_ENTITY_TYPE_OSD); | |
15614 | int r = monclient->init(); | |
15615 | if (r < 0) { | |
15616 | // need to do cleanup because we're in an intermediate init state | |
f67539c2 TL |
15617 | { |
15618 | std::scoped_lock l(timer_lock); | |
15619 | timer.shutdown(); | |
15620 | } | |
15621 | ||
9f95a23c | 15622 | client_lock.unlock(); |
7c673cae FG |
15623 | objecter->shutdown(); |
15624 | objectcacher->stop(); | |
15625 | monclient->shutdown(); | |
15626 | return r; | |
15627 | } | |
15628 | objecter->start(); | |
15629 | ||
9f95a23c | 15630 | client_lock.unlock(); |
7c673cae | 15631 | _finish_init(); |
f67539c2 | 15632 | iref_writer.update_state(CLIENT_INITIALIZED); |
7c673cae FG |
15633 | |
15634 | return 0; | |
15635 | } | |
15636 | ||
15637 | void StandaloneClient::shutdown() | |
15638 | { | |
15639 | Client::shutdown(); | |
15640 | objecter->shutdown(); | |
15641 | monclient->shutdown(); | |
15642 | } |