Commit | Line | Data |
---|---|---|
df038819 JG |
1 | /* |
2 | * Copyright (C) 2018 - Jérémie Galarneau <jeremie.galarneau@efficios.com> | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms of the GNU General Public License, version 2 only, as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
11 | * more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License along with | |
14 | * this program; if not, write to the Free Software Foundation, Inc., 51 | |
15 | * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
16 | */ | |
17 | ||
18 | #include <urcu.h> | |
19 | #include <urcu/list.h> | |
20 | #include <urcu/rculfhash.h> | |
21 | ||
df038819 | 22 | #include <fcntl.h> |
df038819 | 23 | #include <inttypes.h> |
5c1f54d1 JG |
24 | #include <pthread.h> |
25 | #include <stdbool.h> | |
26 | #include <sys/stat.h> | |
27 | #include <sys/types.h> | |
df038819 | 28 | |
df038819 | 29 | #include "common/defaults.h" |
5c1f54d1 | 30 | #include "common/error.h" |
df038819 | 31 | #include "common/hashtable/hashtable.h" |
5c1f54d1 JG |
32 | #include "common/hashtable/utils.h" |
33 | #include "common/macros.h" | |
df038819 JG |
34 | |
35 | #include "fd-tracker.h" | |
874a3441 | 36 | #include "inode.h" |
df038819 JG |
37 | |
38 | /* Tracker lock must be taken by the user. */ | |
5c1f54d1 JG |
39 | #define TRACKED_COUNT(tracker) \ |
40 | (tracker->count.suspendable.active + \ | |
41 | tracker->count.suspendable.suspended + \ | |
42 | tracker->count.unsuspendable) | |
df038819 JG |
43 | |
44 | /* Tracker lock must be taken by the user. */ | |
5c1f54d1 JG |
45 | #define ACTIVE_COUNT(tracker) \ |
46 | (tracker->count.suspendable.active + tracker->count.unsuspendable) | |
df038819 JG |
47 | |
48 | /* Tracker lock must be taken by the user. */ | |
5c1f54d1 | 49 | #define SUSPENDED_COUNT(tracker) (tracker->count.suspendable.suspended) |
df038819 JG |
50 | |
51 | /* Tracker lock must be taken by the user. */ | |
5c1f54d1 JG |
52 | #define SUSPENDABLE_COUNT(tracker) \ |
53 | (tracker->count.suspendable.active + \ | |
54 | tracker->count.suspendable.suspended) | |
df038819 JG |
55 | |
56 | /* Tracker lock must be taken by the user. */ | |
5c1f54d1 | 57 | #define UNSUSPENDABLE_COUNT(tracker) (tracker->count.unsuspendable) |
df038819 JG |
58 | |
59 | struct fd_tracker { | |
60 | pthread_mutex_t lock; | |
61 | struct { | |
5c1f54d1 | 62 | struct { |
df038819 JG |
63 | unsigned int active; |
64 | unsigned int suspended; | |
65 | } suspendable; | |
66 | unsigned int unsuspendable; | |
67 | } count; | |
68 | unsigned int capacity; | |
69 | struct { | |
70 | uint64_t uses; | |
71 | uint64_t misses; | |
72 | /* Failures to suspend or restore fs handles. */ | |
73 | uint64_t errors; | |
74 | } stats; | |
75 | /* | |
76 | * The head of the active_handles list is always the least recently | |
77 | * used active handle. When an handle is used, it is removed from the | |
78 | * list and added to the end. When a file has to be suspended, the | |
79 | * first element in the list is "popped", suspended, and added to the | |
80 | * list of suspended handles. | |
81 | */ | |
82 | struct cds_list_head active_handles; | |
83 | struct cds_list_head suspended_handles; | |
84 | struct cds_lfht *unsuspendable_fds; | |
874a3441 | 85 | struct lttng_inode_registry *inode_registry; |
df038819 JG |
86 | }; |
87 | ||
88 | struct open_properties { | |
df038819 JG |
89 | int flags; |
90 | struct { | |
91 | bool is_set; | |
92 | mode_t value; | |
93 | } mode; | |
94 | }; | |
95 | ||
96 | /* | |
97 | * A fs_handle is not ref-counted. Therefore, it is assumed that a | |
98 | * handle is never in-use while it is being reclaimed. It can be | |
99 | * shared by multiple threads, but external synchronization is required | |
100 | * to ensure it is not still being used when it is reclaimed (close method). | |
101 | * In this respect, it is not different from a regular file descriptor. | |
102 | * | |
103 | * The fs_handle lock always nests _within_ the tracker's lock. | |
104 | */ | |
105 | struct fs_handle { | |
106 | pthread_mutex_t lock; | |
107 | /* | |
108 | * Weak reference to the tracker. All fs_handles are assumed to have | |
109 | * been closed at the moment of the destruction of the fd_tracker. | |
110 | */ | |
111 | struct fd_tracker *tracker; | |
112 | struct open_properties properties; | |
874a3441 | 113 | struct lttng_inode *inode; |
df038819 JG |
114 | int fd; |
115 | /* inode number of the file at the time of the handle's creation. */ | |
116 | uint64_t ino; | |
117 | bool in_use; | |
118 | /* Offset to which the file should be restored. */ | |
119 | off_t offset; | |
120 | struct cds_list_head handles_list_node; | |
121 | }; | |
122 | ||
123 | struct unsuspendable_fd { | |
124 | /* | |
125 | * Accesses are only performed through the tracker, which is protected | |
126 | * by its own lock. | |
127 | */ | |
128 | int fd; | |
129 | char *name; | |
130 | struct cds_lfht_node tracker_node; | |
131 | struct rcu_head rcu_head; | |
132 | }; | |
133 | ||
134 | static struct { | |
135 | pthread_mutex_t lock; | |
136 | bool initialized; | |
137 | unsigned long value; | |
138 | } seed = { | |
139 | .lock = PTHREAD_MUTEX_INITIALIZER, | |
140 | }; | |
141 | ||
142 | static int match_fd(struct cds_lfht_node *node, const void *key); | |
143 | static void unsuspendable_fd_destroy(struct unsuspendable_fd *entry); | |
5c1f54d1 JG |
144 | static struct unsuspendable_fd *unsuspendable_fd_create( |
145 | const char *name, int fd); | |
146 | static int open_from_properties( | |
147 | const char *path, struct open_properties *properties); | |
df038819 JG |
148 | |
149 | static void fs_handle_log(struct fs_handle *handle); | |
150 | static int fs_handle_suspend(struct fs_handle *handle); | |
151 | static int fs_handle_restore(struct fs_handle *handle); | |
152 | ||
5c1f54d1 JG |
153 | static void fd_tracker_track( |
154 | struct fd_tracker *tracker, struct fs_handle *handle); | |
155 | static void fd_tracker_untrack( | |
156 | struct fd_tracker *tracker, struct fs_handle *handle); | |
157 | static int fd_tracker_suspend_handles( | |
158 | struct fd_tracker *tracker, unsigned int count); | |
159 | static int fd_tracker_restore_handle( | |
160 | struct fd_tracker *tracker, struct fs_handle *handle); | |
df038819 JG |
161 | |
162 | /* Match function of the tracker's unsuspendable_fds hash table. */ | |
5c1f54d1 | 163 | static int match_fd(struct cds_lfht_node *node, const void *key) |
df038819 | 164 | { |
5c1f54d1 JG |
165 | struct unsuspendable_fd *entry = caa_container_of( |
166 | node, struct unsuspendable_fd, tracker_node); | |
df038819 | 167 | |
5c1f54d1 JG |
168 | return hash_match_key_ulong( |
169 | (void *) (unsigned long) entry->fd, (void *) key); | |
df038819 JG |
170 | } |
171 | ||
5c1f54d1 | 172 | static void delete_unsuspendable_fd(struct rcu_head *head) |
df038819 | 173 | { |
5c1f54d1 JG |
174 | struct unsuspendable_fd *fd = caa_container_of( |
175 | head, struct unsuspendable_fd, rcu_head); | |
df038819 JG |
176 | |
177 | free(fd->name); | |
178 | free(fd); | |
179 | } | |
180 | ||
5c1f54d1 | 181 | static void unsuspendable_fd_destroy(struct unsuspendable_fd *entry) |
df038819 JG |
182 | { |
183 | if (!entry) { | |
184 | return; | |
185 | } | |
186 | call_rcu(&entry->rcu_head, delete_unsuspendable_fd); | |
187 | } | |
188 | ||
5c1f54d1 JG |
189 | static struct unsuspendable_fd *unsuspendable_fd_create( |
190 | const char *name, int fd) | |
df038819 | 191 | { |
5c1f54d1 | 192 | struct unsuspendable_fd *entry = zmalloc(sizeof(*entry)); |
df038819 JG |
193 | |
194 | if (!entry) { | |
195 | goto error; | |
196 | } | |
197 | if (name) { | |
198 | entry->name = strdup(name); | |
199 | if (!entry->name) { | |
200 | goto error; | |
201 | } | |
202 | } | |
203 | cds_lfht_node_init(&entry->tracker_node); | |
204 | entry->fd = fd; | |
205 | return entry; | |
206 | error: | |
207 | unsuspendable_fd_destroy(entry); | |
208 | return NULL; | |
209 | } | |
210 | ||
5c1f54d1 | 211 | static void fs_handle_log(struct fs_handle *handle) |
df038819 | 212 | { |
874a3441 JG |
213 | const char *path; |
214 | ||
df038819 | 215 | pthread_mutex_lock(&handle->lock); |
874a3441 JG |
216 | path = lttng_inode_get_path(handle->inode); |
217 | ||
df038819 | 218 | if (handle->fd >= 0) { |
5c1f54d1 | 219 | DBG_NO_LOC(" %s [active, fd %d%s]", path, handle->fd, |
df038819 JG |
220 | handle->in_use ? ", in use" : ""); |
221 | } else { | |
874a3441 | 222 | DBG_NO_LOC(" %s [suspended]", path); |
df038819 JG |
223 | } |
224 | pthread_mutex_unlock(&handle->lock); | |
225 | } | |
226 | ||
874a3441 | 227 | /* Tracker lock must be held by the caller. */ |
5c1f54d1 | 228 | static int fs_handle_suspend(struct fs_handle *handle) |
df038819 JG |
229 | { |
230 | int ret = 0; | |
231 | struct stat fs_stat; | |
874a3441 | 232 | const char *path; |
df038819 JG |
233 | |
234 | pthread_mutex_lock(&handle->lock); | |
874a3441 | 235 | path = lttng_inode_get_path(handle->inode); |
df038819 JG |
236 | assert(handle->fd >= 0); |
237 | if (handle->in_use) { | |
238 | /* This handle can't be suspended as it is currently in use. */ | |
239 | ret = -EAGAIN; | |
240 | goto end; | |
241 | } | |
242 | ||
874a3441 | 243 | ret = stat(path, &fs_stat); |
df038819 | 244 | if (ret) { |
5c1f54d1 | 245 | PERROR("Filesystem handle to %s cannot be suspended as stat() failed", |
874a3441 | 246 | path); |
df038819 JG |
247 | ret = -errno; |
248 | goto end; | |
249 | } | |
250 | ||
251 | if (fs_stat.st_ino != handle->ino) { | |
252 | /* Don't suspend as the handle would not be restorable. */ | |
253 | WARN("Filesystem handle to %s cannot be suspended as its inode changed", | |
874a3441 | 254 | path); |
df038819 JG |
255 | ret = -ENOENT; |
256 | goto end; | |
257 | } | |
258 | ||
5c1f54d1 | 259 | handle->offset = lseek(handle->fd, 0, SEEK_CUR); |
df038819 JG |
260 | if (handle->offset == -1) { |
261 | WARN("Filesystem handle to %s cannot be suspended as lseek() failed to sample its current position", | |
874a3441 | 262 | path); |
df038819 JG |
263 | ret = -errno; |
264 | goto end; | |
265 | } | |
266 | ||
267 | ret = close(handle->fd); | |
268 | if (ret) { | |
5c1f54d1 | 269 | PERROR("Filesystem handle to %s cannot be suspended as close() failed", |
874a3441 | 270 | path); |
df038819 JG |
271 | ret = -errno; |
272 | goto end; | |
273 | } | |
274 | DBG("Suspended filesystem handle to %s (fd %i) at position %" PRId64, | |
874a3441 | 275 | path, handle->fd, handle->offset); |
df038819 JG |
276 | handle->fd = -1; |
277 | end: | |
278 | if (ret) { | |
279 | handle->tracker->stats.errors++; | |
280 | } | |
281 | pthread_mutex_unlock(&handle->lock); | |
282 | return ret; | |
283 | } | |
284 | ||
285 | /* Caller must hold the tracker and handle's locks. */ | |
5c1f54d1 | 286 | static int fs_handle_restore(struct fs_handle *handle) |
df038819 JG |
287 | { |
288 | int ret, fd = -1; | |
874a3441 | 289 | const char *path = lttng_inode_get_path(handle->inode); |
df038819 JG |
290 | |
291 | assert(handle->fd == -1); | |
874a3441 | 292 | assert(path); |
5c1f54d1 | 293 | ret = open_from_properties(path, &handle->properties); |
df038819 | 294 | if (ret < 0) { |
5c1f54d1 | 295 | PERROR("Failed to restore filesystem handle to %s, open() failed", |
874a3441 | 296 | path); |
df038819 JG |
297 | ret = -errno; |
298 | goto end; | |
299 | } | |
300 | fd = ret; | |
301 | ||
302 | ret = lseek(fd, handle->offset, SEEK_SET); | |
303 | if (ret < 0) { | |
5c1f54d1 | 304 | PERROR("Failed to restore filesystem handle to %s, lseek() failed", |
874a3441 | 305 | path); |
df038819 JG |
306 | ret = -errno; |
307 | goto end; | |
308 | } | |
309 | DBG("Restored filesystem handle to %s (fd %i) at position %" PRId64, | |
874a3441 | 310 | path, fd, handle->offset); |
df038819 JG |
311 | ret = 0; |
312 | handle->fd = fd; | |
313 | fd = -1; | |
314 | end: | |
315 | if (fd >= 0) { | |
316 | (void) close(fd); | |
317 | } | |
318 | return ret; | |
319 | } | |
320 | ||
5c1f54d1 JG |
321 | static int open_from_properties( |
322 | const char *path, struct open_properties *properties) | |
df038819 JG |
323 | { |
324 | int ret; | |
325 | ||
326 | /* | |
327 | * open() ignores the 'flags' parameter unless the O_CREAT or O_TMPFILE | |
328 | * flags are set. O_TMPFILE would not make sense in the context of a | |
329 | * suspendable fs_handle as it would not be restorable (see OPEN(2)), | |
330 | * thus it is ignored here. | |
331 | */ | |
332 | if ((properties->flags & O_CREAT) && properties->mode.is_set) { | |
5c1f54d1 | 333 | ret = open(path, properties->flags, properties->mode.value); |
df038819 | 334 | } else { |
874a3441 | 335 | ret = open(path, properties->flags); |
df038819 JG |
336 | } |
337 | /* | |
338 | * Some flags should not be used beyond the initial open() of a | |
339 | * restorable file system handle. O_CREAT and O_TRUNC must | |
340 | * be cleared since it would be unexpected to re-use them | |
341 | * when the handle is retored: | |
342 | * - O_CREAT should not be needed as the file has been created | |
343 | * on the initial call to open(), | |
344 | * - O_TRUNC would destroy the file's contents by truncating it | |
345 | * to length 0. | |
346 | */ | |
347 | properties->flags &= ~(O_CREAT | O_TRUNC); | |
348 | if (ret < 0) { | |
349 | ret = -errno; | |
350 | goto end; | |
351 | } | |
352 | end: | |
353 | return ret; | |
354 | } | |
355 | ||
356 | struct fd_tracker *fd_tracker_create(unsigned int capacity) | |
357 | { | |
358 | struct fd_tracker *tracker = zmalloc(sizeof(struct fd_tracker)); | |
359 | ||
360 | if (!tracker) { | |
361 | goto end; | |
362 | } | |
363 | ||
364 | pthread_mutex_lock(&seed.lock); | |
365 | if (!seed.initialized) { | |
366 | seed.value = (unsigned long) time(NULL); | |
367 | seed.initialized = true; | |
368 | } | |
369 | pthread_mutex_unlock(&seed.lock); | |
370 | ||
371 | CDS_INIT_LIST_HEAD(&tracker->active_handles); | |
372 | CDS_INIT_LIST_HEAD(&tracker->suspended_handles); | |
373 | tracker->capacity = capacity; | |
374 | tracker->unsuspendable_fds = cds_lfht_new(DEFAULT_HT_SIZE, 1, 0, | |
375 | CDS_LFHT_AUTO_RESIZE | CDS_LFHT_ACCOUNTING, NULL); | |
874a3441 JG |
376 | if (!tracker->unsuspendable_fds) { |
377 | ERR("Failed to create fd-tracker's unsuspendable_fds hash table"); | |
378 | goto error; | |
379 | } | |
380 | tracker->inode_registry = lttng_inode_registry_create(); | |
381 | if (!tracker->inode_registry) { | |
382 | ERR("Failed to create fd-tracker's inode registry"); | |
383 | goto error; | |
384 | } | |
df038819 JG |
385 | DBG("File descriptor tracker created with a limit of %u simultaneously-opened FDs", |
386 | capacity); | |
387 | end: | |
388 | return tracker; | |
874a3441 JG |
389 | error: |
390 | fd_tracker_destroy(tracker); | |
391 | return NULL; | |
df038819 JG |
392 | } |
393 | ||
394 | void fd_tracker_log(struct fd_tracker *tracker) | |
395 | { | |
396 | struct fs_handle *handle; | |
397 | struct unsuspendable_fd *unsuspendable_fd; | |
398 | struct cds_lfht_iter iter; | |
399 | ||
400 | pthread_mutex_lock(&tracker->lock); | |
401 | DBG_NO_LOC("File descriptor tracker"); | |
402 | DBG_NO_LOC(" Stats:"); | |
403 | DBG_NO_LOC(" uses: %" PRIu64, tracker->stats.uses); | |
404 | DBG_NO_LOC(" misses: %" PRIu64, tracker->stats.misses); | |
405 | DBG_NO_LOC(" errors: %" PRIu64, tracker->stats.errors); | |
406 | DBG_NO_LOC(" Tracked: %u", TRACKED_COUNT(tracker)); | |
407 | DBG_NO_LOC(" active: %u", ACTIVE_COUNT(tracker)); | |
408 | DBG_NO_LOC(" suspendable: %u", SUSPENDABLE_COUNT(tracker)); | |
409 | DBG_NO_LOC(" unsuspendable: %u", UNSUSPENDABLE_COUNT(tracker)); | |
410 | DBG_NO_LOC(" suspended: %u", SUSPENDED_COUNT(tracker)); | |
411 | DBG_NO_LOC(" capacity: %u", tracker->capacity); | |
412 | ||
413 | DBG_NO_LOC(" Tracked suspendable file descriptors"); | |
5c1f54d1 JG |
414 | cds_list_for_each_entry ( |
415 | handle, &tracker->active_handles, handles_list_node) { | |
df038819 JG |
416 | fs_handle_log(handle); |
417 | } | |
5c1f54d1 | 418 | cds_list_for_each_entry (handle, &tracker->suspended_handles, |
df038819 JG |
419 | handles_list_node) { |
420 | fs_handle_log(handle); | |
421 | } | |
422 | if (!SUSPENDABLE_COUNT(tracker)) { | |
423 | DBG_NO_LOC(" None"); | |
424 | } | |
425 | ||
426 | DBG_NO_LOC(" Tracked unsuspendable file descriptors"); | |
427 | rcu_read_lock(); | |
5c1f54d1 | 428 | cds_lfht_for_each_entry (tracker->unsuspendable_fds, &iter, |
df038819 | 429 | unsuspendable_fd, tracker_node) { |
5c1f54d1 JG |
430 | DBG_NO_LOC(" %s [active, fd %d]", |
431 | unsuspendable_fd->name ?: "Unnamed", | |
df038819 JG |
432 | unsuspendable_fd->fd); |
433 | } | |
434 | rcu_read_unlock(); | |
435 | if (!UNSUSPENDABLE_COUNT(tracker)) { | |
436 | DBG_NO_LOC(" None"); | |
437 | } | |
438 | ||
439 | pthread_mutex_unlock(&tracker->lock); | |
440 | } | |
441 | ||
442 | int fd_tracker_destroy(struct fd_tracker *tracker) | |
443 | { | |
444 | int ret = 0; | |
445 | ||
446 | /* | |
447 | * Refuse to destroy the tracker as fs_handles may still old | |
448 | * weak references to the tracker. | |
449 | */ | |
450 | pthread_mutex_lock(&tracker->lock); | |
451 | if (TRACKED_COUNT(tracker)) { | |
452 | ERR("A file descriptor leak has been detected: %u tracked file descriptors are still being tracked", | |
453 | TRACKED_COUNT(tracker)); | |
454 | pthread_mutex_unlock(&tracker->lock); | |
455 | fd_tracker_log(tracker); | |
456 | ret = -1; | |
457 | goto end; | |
458 | } | |
459 | pthread_mutex_unlock(&tracker->lock); | |
460 | ||
874a3441 JG |
461 | if (tracker->unsuspendable_fds) { |
462 | ret = cds_lfht_destroy(tracker->unsuspendable_fds, NULL); | |
463 | assert(!ret); | |
464 | } | |
660a216c JG |
465 | |
466 | lttng_inode_registry_destroy(tracker->inode_registry); | |
df038819 JG |
467 | pthread_mutex_destroy(&tracker->lock); |
468 | free(tracker); | |
469 | end: | |
470 | return ret; | |
471 | } | |
472 | ||
473 | struct fs_handle *fd_tracker_open_fs_handle(struct fd_tracker *tracker, | |
5c1f54d1 JG |
474 | const char *path, |
475 | int flags, | |
476 | mode_t *mode) | |
df038819 JG |
477 | { |
478 | int ret; | |
479 | struct fs_handle *handle = NULL; | |
480 | struct stat fd_stat; | |
481 | struct open_properties properties = { | |
df038819 JG |
482 | .flags = flags, |
483 | .mode.is_set = !!mode, | |
484 | .mode.value = mode ? *mode : 0, | |
485 | }; | |
486 | ||
df038819 JG |
487 | pthread_mutex_lock(&tracker->lock); |
488 | if (ACTIVE_COUNT(tracker) == tracker->capacity) { | |
489 | if (tracker->count.suspendable.active > 0) { | |
490 | ret = fd_tracker_suspend_handles(tracker, 1); | |
491 | if (ret) { | |
19380ea8 | 492 | goto end; |
df038819 JG |
493 | } |
494 | } else { | |
495 | /* | |
496 | * There are not enough active suspendable file | |
19380ea8 JG |
497 | * descriptors to open a new fd and still accommodate |
498 | * the tracker's capacity. | |
df038819 JG |
499 | */ |
500 | WARN("Cannot open file system handle, too many unsuspendable file descriptors are opened (%u)", | |
501 | tracker->count.unsuspendable); | |
19380ea8 | 502 | goto end; |
df038819 JG |
503 | } |
504 | } | |
505 | ||
506 | handle = zmalloc(sizeof(*handle)); | |
507 | if (!handle) { | |
508 | goto end; | |
509 | } | |
660a216c | 510 | handle->tracker = tracker; |
df038819 JG |
511 | |
512 | ret = pthread_mutex_init(&handle->lock, NULL); | |
513 | if (ret) { | |
514 | PERROR("Failed to initialize handle mutex while creating fs handle"); | |
19380ea8 | 515 | goto error_mutex_init; |
df038819 JG |
516 | } |
517 | ||
874a3441 | 518 | handle->fd = open_from_properties(path, &properties); |
df038819 JG |
519 | if (handle->fd < 0) { |
520 | PERROR("Failed to open fs handle to %s, open() returned", path); | |
19380ea8 | 521 | goto error; |
df038819 JG |
522 | } |
523 | ||
524 | handle->properties = properties; | |
874a3441 | 525 | |
5c1f54d1 JG |
526 | handle->inode = lttng_inode_registry_get_inode( |
527 | tracker->inode_registry, handle->fd, path); | |
874a3441 | 528 | if (!handle->inode) { |
5c1f54d1 | 529 | ERR("Failed to get lttng_inode corresponding to file %s", path); |
19380ea8 | 530 | goto error; |
874a3441 | 531 | } |
df038819 JG |
532 | |
533 | if (fstat(handle->fd, &fd_stat)) { | |
534 | PERROR("Failed to retrieve file descriptor inode while creating fs handle, fstat() returned"); | |
19380ea8 | 535 | goto error; |
df038819 JG |
536 | } |
537 | handle->ino = fd_stat.st_ino; | |
538 | ||
539 | fd_tracker_track(tracker, handle); | |
df038819 | 540 | end: |
19380ea8 | 541 | pthread_mutex_unlock(&tracker->lock); |
df038819 | 542 | return handle; |
19380ea8 | 543 | error: |
660a216c JG |
544 | if (handle->inode) { |
545 | lttng_inode_put(handle->inode); | |
546 | } | |
19380ea8 JG |
547 | pthread_mutex_destroy(&handle->lock); |
548 | error_mutex_init: | |
660a216c | 549 | free(handle); |
df038819 JG |
550 | handle = NULL; |
551 | goto end; | |
552 | } | |
553 | ||
554 | /* Caller must hold the tracker's lock. */ | |
5c1f54d1 JG |
555 | static int fd_tracker_suspend_handles( |
556 | struct fd_tracker *tracker, unsigned int count) | |
df038819 JG |
557 | { |
558 | unsigned int left_to_close = count; | |
559 | struct fs_handle *handle, *tmp; | |
560 | ||
5c1f54d1 | 561 | cds_list_for_each_entry_safe (handle, tmp, &tracker->active_handles, |
df038819 JG |
562 | handles_list_node) { |
563 | int ret; | |
564 | ||
565 | fd_tracker_untrack(tracker, handle); | |
566 | ret = fs_handle_suspend(handle); | |
567 | fd_tracker_track(tracker, handle); | |
568 | if (!ret) { | |
569 | left_to_close--; | |
570 | } | |
571 | ||
572 | if (!left_to_close) { | |
573 | break; | |
574 | } | |
575 | } | |
576 | return left_to_close ? -EMFILE : 0; | |
577 | } | |
578 | ||
579 | int fd_tracker_open_unsuspendable_fd(struct fd_tracker *tracker, | |
5c1f54d1 JG |
580 | int *out_fds, |
581 | const char **names, | |
582 | unsigned int fd_count, | |
583 | fd_open_cb open, | |
584 | void *user_data) | |
df038819 JG |
585 | { |
586 | int ret, user_ret, i, fds_to_suspend; | |
587 | unsigned int active_fds; | |
588 | struct unsuspendable_fd *entries[fd_count]; | |
589 | ||
590 | memset(entries, 0, sizeof(entries)); | |
591 | ||
592 | pthread_mutex_lock(&tracker->lock); | |
593 | ||
594 | active_fds = ACTIVE_COUNT(tracker); | |
5c1f54d1 JG |
595 | fds_to_suspend = (int) active_fds + (int) fd_count - |
596 | (int) tracker->capacity; | |
df038819 JG |
597 | if (fds_to_suspend > 0) { |
598 | if (fds_to_suspend <= tracker->count.suspendable.active) { | |
5c1f54d1 JG |
599 | ret = fd_tracker_suspend_handles( |
600 | tracker, fds_to_suspend); | |
df038819 JG |
601 | if (ret) { |
602 | goto end; | |
603 | } | |
604 | } else { | |
605 | /* | |
606 | * There are not enough active suspendable file | |
607 | * descriptors to open a new fd and still accomodate the | |
608 | * tracker's capacity. | |
609 | */ | |
610 | WARN("Cannot open unsuspendable fd, too many unsuspendable file descriptors are opened (%u)", | |
611 | tracker->count.unsuspendable); | |
612 | ret = -EMFILE; | |
613 | goto end; | |
614 | } | |
615 | } | |
616 | ||
617 | user_ret = open(user_data, out_fds); | |
618 | if (user_ret) { | |
619 | ret = user_ret; | |
620 | goto end; | |
621 | } | |
622 | ||
623 | /* | |
624 | * Add the fds returned by the user's callback to the hashtable | |
625 | * of unsuspendable fds. | |
626 | */ | |
627 | for (i = 0; i < fd_count; i++) { | |
5c1f54d1 JG |
628 | struct unsuspendable_fd *entry = unsuspendable_fd_create( |
629 | names ? names[i] : NULL, out_fds[i]); | |
df038819 JG |
630 | |
631 | if (!entry) { | |
632 | ret = -1; | |
633 | goto end_free_entries; | |
634 | } | |
635 | entries[i] = entry; | |
636 | } | |
637 | ||
638 | rcu_read_lock(); | |
639 | for (i = 0; i < fd_count; i++) { | |
640 | struct cds_lfht_node *node; | |
641 | struct unsuspendable_fd *entry = entries[i]; | |
642 | ||
5c1f54d1 JG |
643 | node = cds_lfht_add_unique(tracker->unsuspendable_fds, |
644 | hash_key_ulong((void *) (unsigned long) | |
645 | out_fds[i], | |
df038819 | 646 | seed.value), |
5c1f54d1 | 647 | match_fd, (void *) (unsigned long) out_fds[i], |
df038819 JG |
648 | &entry->tracker_node); |
649 | ||
650 | if (node != &entry->tracker_node) { | |
651 | ret = -EEXIST; | |
652 | rcu_read_unlock(); | |
653 | goto end_free_entries; | |
654 | } | |
655 | entries[i] = NULL; | |
656 | } | |
657 | tracker->count.unsuspendable += fd_count; | |
658 | rcu_read_unlock(); | |
659 | ret = user_ret; | |
660 | end: | |
661 | pthread_mutex_unlock(&tracker->lock); | |
662 | return ret; | |
663 | end_free_entries: | |
664 | for (i = 0; i < fd_count; i++) { | |
665 | unsuspendable_fd_destroy(entries[i]); | |
666 | } | |
667 | goto end; | |
668 | } | |
669 | ||
670 | int fd_tracker_close_unsuspendable_fd(struct fd_tracker *tracker, | |
5c1f54d1 JG |
671 | int *fds_in, |
672 | unsigned int fd_count, | |
673 | fd_close_cb close, | |
df038819 JG |
674 | void *user_data) |
675 | { | |
676 | int i, ret, user_ret; | |
677 | int fds[fd_count]; | |
678 | ||
679 | /* | |
680 | * Maintain a local copy of fds_in as the user's callback may modify its | |
681 | * contents (e.g. setting the fd(s) to -1 after close). | |
682 | */ | |
683 | memcpy(fds, fds_in, sizeof(*fds) * fd_count); | |
684 | ||
685 | pthread_mutex_lock(&tracker->lock); | |
686 | rcu_read_lock(); | |
687 | ||
688 | /* Let the user close the file descriptors. */ | |
689 | user_ret = close(user_data, fds_in); | |
690 | if (user_ret) { | |
691 | ret = user_ret; | |
692 | goto end; | |
693 | } | |
694 | ||
695 | /* Untrack the fds that were just closed by the user's callback. */ | |
696 | for (i = 0; i < fd_count; i++) { | |
697 | struct cds_lfht_node *node; | |
698 | struct cds_lfht_iter iter; | |
699 | struct unsuspendable_fd *entry; | |
700 | ||
701 | cds_lfht_lookup(tracker->unsuspendable_fds, | |
702 | hash_key_ulong((void *) (unsigned long) fds[i], | |
703 | seed.value), | |
5c1f54d1 | 704 | match_fd, (void *) (unsigned long) fds[i], |
df038819 JG |
705 | &iter); |
706 | node = cds_lfht_iter_get_node(&iter); | |
707 | if (!node) { | |
708 | /* Unknown file descriptor. */ | |
709 | WARN("Untracked file descriptor %d passed to fd_tracker_close_unsuspendable_fd()", | |
710 | fds[i]); | |
711 | ret = -EINVAL; | |
712 | goto end; | |
713 | } | |
5c1f54d1 JG |
714 | entry = caa_container_of( |
715 | node, struct unsuspendable_fd, tracker_node); | |
df038819 JG |
716 | |
717 | cds_lfht_del(tracker->unsuspendable_fds, node); | |
718 | unsuspendable_fd_destroy(entry); | |
719 | fds[i] = -1; | |
720 | } | |
721 | ||
722 | tracker->count.unsuspendable -= fd_count; | |
723 | ret = 0; | |
724 | end: | |
725 | rcu_read_unlock(); | |
726 | pthread_mutex_unlock(&tracker->lock); | |
727 | return ret; | |
728 | } | |
729 | ||
730 | /* Caller must have taken the tracker's and handle's locks. */ | |
5c1f54d1 JG |
731 | static void fd_tracker_track( |
732 | struct fd_tracker *tracker, struct fs_handle *handle) | |
df038819 JG |
733 | { |
734 | if (handle->fd >= 0) { | |
735 | tracker->count.suspendable.active++; | |
736 | cds_list_add_tail(&handle->handles_list_node, | |
737 | &tracker->active_handles); | |
738 | } else { | |
739 | tracker->count.suspendable.suspended++; | |
740 | cds_list_add_tail(&handle->handles_list_node, | |
741 | &tracker->suspended_handles); | |
742 | } | |
743 | } | |
744 | ||
745 | /* Caller must have taken the tracker's and handle's locks. */ | |
5c1f54d1 JG |
746 | static void fd_tracker_untrack( |
747 | struct fd_tracker *tracker, struct fs_handle *handle) | |
df038819 JG |
748 | { |
749 | if (handle->fd >= 0) { | |
750 | tracker->count.suspendable.active--; | |
751 | } else { | |
752 | tracker->count.suspendable.suspended--; | |
753 | } | |
754 | cds_list_del(&handle->handles_list_node); | |
755 | } | |
756 | ||
757 | /* Caller must have taken the tracker's and handle's locks. */ | |
5c1f54d1 JG |
758 | static int fd_tracker_restore_handle( |
759 | struct fd_tracker *tracker, struct fs_handle *handle) | |
df038819 JG |
760 | { |
761 | int ret; | |
762 | ||
763 | fd_tracker_untrack(tracker, handle); | |
764 | if (ACTIVE_COUNT(tracker) >= tracker->capacity) { | |
765 | ret = fd_tracker_suspend_handles(tracker, 1); | |
766 | if (ret) { | |
767 | goto end; | |
768 | } | |
769 | } | |
770 | ret = fs_handle_restore(handle); | |
771 | end: | |
772 | fd_tracker_track(tracker, handle); | |
773 | return ret ? ret : handle->fd; | |
774 | } | |
775 | ||
776 | int fs_handle_get_fd(struct fs_handle *handle) | |
777 | { | |
778 | int ret; | |
779 | ||
780 | /* | |
781 | * TODO This should be optimized as it is a fairly hot path. | |
782 | * The fd-tracker's lock should only be taken when a fs_handle is | |
783 | * restored (slow path). On the fast path (fs_handle is active), | |
784 | * the only effect on the fd_tracker is marking the handle as the | |
785 | * most recently used. Currently, it is done by a call to the | |
786 | * track/untrack helpers, but it should be done atomically. | |
787 | * | |
788 | * Note that the lock's nesting order must still be respected here. | |
789 | * The handle's lock nests inside the tracker's lock. | |
790 | */ | |
791 | pthread_mutex_lock(&handle->tracker->lock); | |
792 | pthread_mutex_lock(&handle->lock); | |
793 | assert(!handle->in_use); | |
794 | ||
795 | handle->tracker->stats.uses++; | |
796 | if (handle->fd >= 0) { | |
797 | ret = handle->fd; | |
798 | /* Mark as most recently used. */ | |
799 | fd_tracker_untrack(handle->tracker, handle); | |
800 | fd_tracker_track(handle->tracker, handle); | |
801 | } else { | |
802 | handle->tracker->stats.misses++; | |
803 | ret = fd_tracker_restore_handle(handle->tracker, handle); | |
804 | if (ret < 0) { | |
805 | handle->tracker->stats.errors++; | |
806 | goto end; | |
807 | } | |
808 | } | |
809 | handle->in_use = true; | |
810 | end: | |
811 | pthread_mutex_unlock(&handle->lock); | |
812 | pthread_mutex_unlock(&handle->tracker->lock); | |
813 | return ret; | |
814 | } | |
815 | ||
816 | void fs_handle_put_fd(struct fs_handle *handle) | |
817 | { | |
818 | pthread_mutex_lock(&handle->lock); | |
819 | handle->in_use = false; | |
820 | pthread_mutex_unlock(&handle->lock); | |
821 | } | |
822 | ||
9d16fc7f JG |
823 | int fs_handle_unlink(struct fs_handle *handle) |
824 | { | |
825 | int ret; | |
826 | ||
827 | pthread_mutex_lock(&handle->tracker->lock); | |
828 | pthread_mutex_lock(&handle->lock); | |
829 | ret = lttng_inode_defer_unlink(handle->inode); | |
830 | pthread_mutex_unlock(&handle->lock); | |
831 | pthread_mutex_unlock(&handle->tracker->lock); | |
832 | return ret; | |
833 | } | |
834 | ||
df038819 JG |
835 | int fs_handle_close(struct fs_handle *handle) |
836 | { | |
837 | int ret = 0; | |
660a216c | 838 | const char *path = NULL; |
df038819 JG |
839 | |
840 | if (!handle) { | |
841 | ret = -EINVAL; | |
842 | goto end; | |
843 | } | |
844 | ||
845 | pthread_mutex_lock(&handle->tracker->lock); | |
846 | pthread_mutex_lock(&handle->lock); | |
660a216c JG |
847 | if (handle->inode) { |
848 | path = lttng_inode_get_path(handle->inode); | |
849 | } | |
df038819 JG |
850 | fd_tracker_untrack(handle->tracker, handle); |
851 | if (handle->fd >= 0) { | |
852 | /* | |
853 | * The return value of close() is not propagated as there | |
854 | * isn't much the user can do about it. | |
855 | */ | |
856 | if (close(handle->fd)) { | |
857 | PERROR("Failed to close the file descritptor (%d) of fs handle to %s, close() returned", | |
660a216c | 858 | handle->fd, path ? path : "Unknown"); |
df038819 JG |
859 | } |
860 | handle->fd = -1; | |
861 | } | |
4c372c54 JG |
862 | if (handle->inode) { |
863 | lttng_inode_put(handle->inode); | |
864 | } | |
df038819 JG |
865 | pthread_mutex_unlock(&handle->lock); |
866 | pthread_mutex_destroy(&handle->lock); | |
867 | pthread_mutex_unlock(&handle->tracker->lock); | |
df038819 JG |
868 | free(handle); |
869 | end: | |
870 | return ret; | |
871 | } |