Commit | Line | Data |
---|---|---|
5444e298 EP |
1 | /* |
2 | * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com> | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2, or (at your option) | |
7 | * any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; see the file COPYING. If not, write to | |
16 | * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
17 | */ | |
18 | ||
19 | /* | |
20 | * fsnotify inode mark locking/lifetime/and refcnting | |
21 | * | |
22 | * REFCNT: | |
9756b918 LS |
23 | * The group->recnt and mark->refcnt tell how many "things" in the kernel |
24 | * currently are referencing the objects. Both kind of objects typically will | |
25 | * live inside the kernel with a refcnt of 2, one for its creation and one for | |
26 | * the reference a group and a mark hold to each other. | |
27 | * If you are holding the appropriate locks, you can take a reference and the | |
28 | * object itself is guaranteed to survive until the reference is dropped. | |
5444e298 EP |
29 | * |
30 | * LOCKING: | |
9756b918 LS |
31 | * There are 3 locks involved with fsnotify inode marks and they MUST be taken |
32 | * in order as follows: | |
5444e298 | 33 | * |
9756b918 | 34 | * group->mark_mutex |
5444e298 | 35 | * mark->lock |
04662cab | 36 | * mark->connector->lock |
5444e298 | 37 | * |
9756b918 LS |
38 | * group->mark_mutex protects the marks_list anchored inside a given group and |
39 | * each mark is hooked via the g_list. It also protects the groups private | |
40 | * data (i.e group limits). | |
41 | ||
42 | * mark->lock protects the marks attributes like its masks and flags. | |
43 | * Furthermore it protects the access to a reference of the group that the mark | |
44 | * is assigned to as well as the access to a reference of the inode/vfsmount | |
45 | * that is being watched by the mark. | |
5444e298 | 46 | * |
04662cab JK |
47 | * mark->connector->lock protects the list of marks anchored inside an |
48 | * inode / vfsmount and each mark is hooked via the i_list. | |
5444e298 | 49 | * |
04662cab JK |
50 | * A list of notification marks relating to inode / mnt is contained in |
51 | * fsnotify_mark_connector. That structure is alive as long as there are any | |
6b3f05d2 JK |
52 | * marks in the list and is also protected by fsnotify_mark_srcu. A mark gets |
53 | * detached from fsnotify_mark_connector when last reference to the mark is | |
54 | * dropped. Thus having mark reference is enough to protect mark->connector | |
55 | * pointer and to make sure fsnotify_mark_connector cannot disappear. Also | |
56 | * because we remove mark from g_list before dropping mark reference associated | |
57 | * with that, any mark found through g_list is guaranteed to have | |
58 | * mark->connector set until we drop group->mark_mutex. | |
5444e298 EP |
59 | * |
60 | * LIFETIME: | |
61 | * Inode marks survive between when they are added to an inode and when their | |
c1f33073 | 62 | * refcnt==0. Marks are also protected by fsnotify_mark_srcu. |
5444e298 EP |
63 | * |
64 | * The inode mark can be cleared for a number of different reasons including: | |
65 | * - The inode is unlinked for the last time. (fsnotify_inode_remove) | |
66 | * - The inode is being evicted from cache. (fsnotify_inode_delete) | |
67 | * - The fs the inode is on is unmounted. (fsnotify_inode_delete/fsnotify_unmount_inodes) | |
68 | * - Something explicitly requests that it be removed. (fsnotify_destroy_mark) | |
69 | * - The fsnotify_group associated with the mark is going away and all such marks | |
2e37c6ca | 70 | * need to be cleaned up. (fsnotify_clear_marks_by_group) |
5444e298 | 71 | * |
5444e298 EP |
72 | * This has the very interesting property of being able to run concurrently with |
73 | * any (or all) other directions. | |
74 | */ | |
75 | ||
76 | #include <linux/fs.h> | |
77 | #include <linux/init.h> | |
78 | #include <linux/kernel.h> | |
75c1be48 | 79 | #include <linux/kthread.h> |
5444e298 EP |
80 | #include <linux/module.h> |
81 | #include <linux/mutex.h> | |
82 | #include <linux/slab.h> | |
83 | #include <linux/spinlock.h> | |
75c1be48 | 84 | #include <linux/srcu.h> |
5444e298 | 85 | |
60063497 | 86 | #include <linux/atomic.h> |
5444e298 EP |
87 | |
88 | #include <linux/fsnotify_backend.h> | |
89 | #include "fsnotify.h" | |
90 | ||
0918f1c3 JL |
91 | #define FSNOTIFY_REAPER_DELAY (1) /* 1 jiffy */ |
92 | ||
75c1be48 | 93 | struct srcu_struct fsnotify_mark_srcu; |
9dd813c1 JK |
94 | struct kmem_cache *fsnotify_mark_connector_cachep; |
95 | ||
13d34ac6 JL |
96 | static DEFINE_SPINLOCK(destroy_lock); |
97 | static LIST_HEAD(destroy_list); | |
08991e83 | 98 | static struct fsnotify_mark_connector *connector_destroy_list; |
0918f1c3 | 99 | |
35e48176 JK |
100 | static void fsnotify_mark_destroy_workfn(struct work_struct *work); |
101 | static DECLARE_DELAYED_WORK(reaper_work, fsnotify_mark_destroy_workfn); | |
75c1be48 | 102 | |
08991e83 JK |
103 | static void fsnotify_connector_destroy_workfn(struct work_struct *work); |
104 | static DECLARE_WORK(connector_reaper_work, fsnotify_connector_destroy_workfn); | |
105 | ||
5444e298 EP |
106 | void fsnotify_get_mark(struct fsnotify_mark *mark) |
107 | { | |
11375145 | 108 | WARN_ON_ONCE(!atomic_read(&mark->refcnt)); |
5444e298 EP |
109 | atomic_inc(&mark->refcnt); |
110 | } | |
111 | ||
a242677b | 112 | static void __fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) |
0809ab69 JK |
113 | { |
114 | u32 new_mask = 0; | |
115 | struct fsnotify_mark *mark; | |
116 | ||
04662cab | 117 | assert_spin_locked(&conn->lock); |
6b3f05d2 JK |
118 | hlist_for_each_entry(mark, &conn->list, obj_list) { |
119 | if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) | |
120 | new_mask |= mark->mask; | |
121 | } | |
a242677b JK |
122 | if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE) |
123 | conn->inode->i_fsnotify_mask = new_mask; | |
124 | else if (conn->flags & FSNOTIFY_OBJ_TYPE_VFSMOUNT) | |
125 | real_mount(conn->mnt)->mnt_fsnotify_mask = new_mask; | |
126 | } | |
127 | ||
128 | /* | |
129 | * Calculate mask of events for a list of marks. The caller must make sure | |
6b3f05d2 JK |
130 | * connector and connector->inode cannot disappear under us. Callers achieve |
131 | * this by holding a mark->lock or mark->group->mark_mutex for a mark on this | |
132 | * list. | |
a242677b JK |
133 | */ |
134 | void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) | |
135 | { | |
136 | if (!conn) | |
137 | return; | |
138 | ||
04662cab | 139 | spin_lock(&conn->lock); |
a242677b | 140 | __fsnotify_recalc_mask(conn); |
04662cab JK |
141 | spin_unlock(&conn->lock); |
142 | if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE) | |
a242677b | 143 | __fsnotify_update_child_dentry_flags(conn->inode); |
0809ab69 JK |
144 | } |
145 | ||
08991e83 JK |
146 | /* Free all connectors queued for freeing once SRCU period ends */ |
147 | static void fsnotify_connector_destroy_workfn(struct work_struct *work) | |
148 | { | |
149 | struct fsnotify_mark_connector *conn, *free; | |
150 | ||
151 | spin_lock(&destroy_lock); | |
152 | conn = connector_destroy_list; | |
153 | connector_destroy_list = NULL; | |
154 | spin_unlock(&destroy_lock); | |
155 | ||
156 | synchronize_srcu(&fsnotify_mark_srcu); | |
157 | while (conn) { | |
158 | free = conn; | |
159 | conn = conn->destroy_next; | |
160 | kmem_cache_free(fsnotify_mark_connector_cachep, free); | |
161 | } | |
162 | } | |
163 | ||
08991e83 JK |
164 | static struct inode *fsnotify_detach_connector_from_object( |
165 | struct fsnotify_mark_connector *conn) | |
166 | { | |
167 | struct inode *inode = NULL; | |
168 | ||
169 | if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE) { | |
170 | inode = conn->inode; | |
171 | rcu_assign_pointer(inode->i_fsnotify_marks, NULL); | |
172 | inode->i_fsnotify_mask = 0; | |
173 | conn->inode = NULL; | |
174 | conn->flags &= ~FSNOTIFY_OBJ_TYPE_INODE; | |
175 | } else if (conn->flags & FSNOTIFY_OBJ_TYPE_VFSMOUNT) { | |
176 | rcu_assign_pointer(real_mount(conn->mnt)->mnt_fsnotify_marks, | |
177 | NULL); | |
178 | real_mount(conn->mnt)->mnt_fsnotify_mask = 0; | |
179 | conn->mnt = NULL; | |
180 | conn->flags &= ~FSNOTIFY_OBJ_TYPE_VFSMOUNT; | |
181 | } | |
182 | ||
183 | return inode; | |
184 | } | |
185 | ||
6b3f05d2 JK |
186 | static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark) |
187 | { | |
054c636e JK |
188 | struct fsnotify_group *group = mark->group; |
189 | ||
190 | if (WARN_ON_ONCE(!group)) | |
191 | return; | |
192 | group->ops->free_mark(mark); | |
193 | fsnotify_put_group(group); | |
6b3f05d2 JK |
194 | } |
195 | ||
196 | void fsnotify_put_mark(struct fsnotify_mark *mark) | |
8212a609 JK |
197 | { |
198 | struct fsnotify_mark_connector *conn; | |
199 | struct inode *inode = NULL; | |
08991e83 | 200 | bool free_conn = false; |
8212a609 | 201 | |
6b3f05d2 JK |
202 | /* Catch marks that were actually never attached to object */ |
203 | if (!mark->connector) { | |
204 | if (atomic_dec_and_test(&mark->refcnt)) | |
205 | fsnotify_final_mark_destroy(mark); | |
206 | return; | |
207 | } | |
208 | ||
209 | /* | |
210 | * We have to be careful so that traversals of obj_list under lock can | |
211 | * safely grab mark reference. | |
212 | */ | |
213 | if (!atomic_dec_and_lock(&mark->refcnt, &mark->connector->lock)) | |
214 | return; | |
215 | ||
8212a609 | 216 | conn = mark->connector; |
8212a609 JK |
217 | hlist_del_init_rcu(&mark->obj_list); |
218 | if (hlist_empty(&conn->list)) { | |
08991e83 JK |
219 | inode = fsnotify_detach_connector_from_object(conn); |
220 | free_conn = true; | |
221 | } else { | |
222 | __fsnotify_recalc_mask(conn); | |
8212a609 JK |
223 | } |
224 | mark->connector = NULL; | |
04662cab | 225 | spin_unlock(&conn->lock); |
8212a609 | 226 | |
6b3f05d2 JK |
227 | iput(inode); |
228 | ||
08991e83 JK |
229 | if (free_conn) { |
230 | spin_lock(&destroy_lock); | |
231 | conn->destroy_next = connector_destroy_list; | |
232 | connector_destroy_list = conn; | |
233 | spin_unlock(&destroy_lock); | |
234 | queue_work(system_unbound_wq, &connector_reaper_work); | |
235 | } | |
6b3f05d2 JK |
236 | /* |
237 | * Note that we didn't update flags telling whether inode cares about | |
238 | * what's happening with children. We update these flags from | |
239 | * __fsnotify_parent() lazily when next event happens on one of our | |
240 | * children. | |
241 | */ | |
242 | spin_lock(&destroy_lock); | |
243 | list_add(&mark->g_list, &destroy_list); | |
244 | spin_unlock(&destroy_lock); | |
245 | queue_delayed_work(system_unbound_wq, &reaper_work, | |
246 | FSNOTIFY_REAPER_DELAY); | |
8212a609 JK |
247 | } |
248 | ||
24c20305 MS |
249 | /* |
250 | * Get mark reference when we found the mark via lockless traversal of object | |
251 | * list. Mark can be already removed from the list by now and on its way to be | |
252 | * destroyed once SRCU period ends. | |
253 | * | |
254 | * Also pin the group so it doesn't disappear under us. | |
255 | */ | |
256 | static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark) | |
abc77577 | 257 | { |
24c20305 MS |
258 | if (!mark) |
259 | return true; | |
abc77577 | 260 | |
9a31d7ad MS |
261 | if (atomic_inc_not_zero(&mark->refcnt)) { |
262 | spin_lock(&mark->lock); | |
263 | if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) { | |
264 | /* mark is attached, group is still alive then */ | |
265 | atomic_inc(&mark->group->user_waits); | |
266 | spin_unlock(&mark->lock); | |
267 | return true; | |
268 | } | |
269 | spin_unlock(&mark->lock); | |
270 | fsnotify_put_mark(mark); | |
271 | } | |
24c20305 MS |
272 | return false; |
273 | } | |
274 | ||
275 | /* | |
276 | * Puts marks and wakes up group destruction if necessary. | |
277 | * | |
278 | * Pairs with fsnotify_get_mark_safe() | |
279 | */ | |
280 | static void fsnotify_put_mark_wake(struct fsnotify_mark *mark) | |
281 | { | |
282 | if (mark) { | |
283 | struct fsnotify_group *group = mark->group; | |
284 | ||
285 | fsnotify_put_mark(mark); | |
286 | /* | |
287 | * We abuse notification_waitq on group shutdown for waiting for | |
288 | * all marks pinned when waiting for userspace. | |
289 | */ | |
290 | if (atomic_dec_and_test(&group->user_waits) && group->shutdown) | |
291 | wake_up(&group->notification_waitq); | |
abc77577 | 292 | } |
24c20305 MS |
293 | } |
294 | ||
295 | bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info) | |
296 | { | |
297 | /* This can fail if mark is being removed */ | |
298 | if (!fsnotify_get_mark_safe(iter_info->inode_mark)) | |
299 | return false; | |
300 | if (!fsnotify_get_mark_safe(iter_info->vfsmount_mark)) { | |
301 | fsnotify_put_mark_wake(iter_info->inode_mark); | |
302 | return false; | |
abc77577 JK |
303 | } |
304 | ||
305 | /* | |
306 | * Now that both marks are pinned by refcount in the inode / vfsmount | |
307 | * lists, we can drop SRCU lock, and safely resume the list iteration | |
308 | * once userspace returns. | |
309 | */ | |
310 | srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx); | |
311 | ||
312 | return true; | |
abc77577 JK |
313 | } |
314 | ||
315 | void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info) | |
316 | { | |
abc77577 | 317 | iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); |
24c20305 MS |
318 | fsnotify_put_mark_wake(iter_info->inode_mark); |
319 | fsnotify_put_mark_wake(iter_info->vfsmount_mark); | |
abc77577 JK |
320 | } |
321 | ||
5444e298 | 322 | /* |
6b3f05d2 JK |
323 | * Mark mark as detached, remove it from group list. Mark still stays in object |
324 | * list until its last reference is dropped. Note that we rely on mark being | |
325 | * removed from group list before corresponding reference to it is dropped. In | |
326 | * particular we rely on mark->connector being valid while we hold | |
327 | * group->mark_mutex if we found the mark through g_list. | |
4712e722 | 328 | * |
11375145 JK |
329 | * Must be called with group->mark_mutex held. The caller must either hold |
330 | * reference to the mark or be protected by fsnotify_mark_srcu. | |
5444e298 | 331 | */ |
4712e722 | 332 | void fsnotify_detach_mark(struct fsnotify_mark *mark) |
5444e298 | 333 | { |
4712e722 | 334 | struct fsnotify_group *group = mark->group; |
5444e298 | 335 | |
11375145 JK |
336 | WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex)); |
337 | WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) && | |
338 | atomic_read(&mark->refcnt) < 1 + | |
339 | !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)); | |
d5a335b8 | 340 | |
104d06f0 | 341 | spin_lock(&mark->lock); |
700307a2 | 342 | /* something else already called this function on this mark */ |
4712e722 | 343 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { |
5444e298 | 344 | spin_unlock(&mark->lock); |
e2a29943 | 345 | return; |
5444e298 | 346 | } |
4712e722 | 347 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED; |
5444e298 | 348 | list_del_init(&mark->g_list); |
5444e298 | 349 | spin_unlock(&mark->lock); |
d5a335b8 | 350 | |
4712e722 | 351 | atomic_dec(&group->num_marks); |
11375145 JK |
352 | |
353 | /* Drop mark reference acquired in fsnotify_add_mark_locked() */ | |
354 | fsnotify_put_mark(mark); | |
4712e722 JK |
355 | } |
356 | ||
357 | /* | |
11375145 JK |
358 | * Free fsnotify mark. The mark is actually only marked as being freed. The |
359 | * freeing is actually happening only once last reference to the mark is | |
360 | * dropped from a workqueue which first waits for srcu period end. | |
35e48176 | 361 | * |
11375145 JK |
362 | * Caller must have a reference to the mark or be protected by |
363 | * fsnotify_mark_srcu. | |
4712e722 | 364 | */ |
11375145 | 365 | void fsnotify_free_mark(struct fsnotify_mark *mark) |
4712e722 JK |
366 | { |
367 | struct fsnotify_group *group = mark->group; | |
368 | ||
369 | spin_lock(&mark->lock); | |
370 | /* something else already called this function on this mark */ | |
371 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) { | |
372 | spin_unlock(&mark->lock); | |
11375145 | 373 | return; |
4712e722 JK |
374 | } |
375 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE; | |
376 | spin_unlock(&mark->lock); | |
5444e298 | 377 | |
d725e66c LT |
378 | /* |
379 | * Some groups like to know that marks are being freed. This is a | |
380 | * callback to the group function to let it know that this mark | |
381 | * is being freed. | |
382 | */ | |
383 | if (group->ops->freeing_mark) | |
384 | group->ops->freeing_mark(mark, group); | |
d5a335b8 LS |
385 | } |
386 | ||
387 | void fsnotify_destroy_mark(struct fsnotify_mark *mark, | |
388 | struct fsnotify_group *group) | |
389 | { | |
6960b0d9 | 390 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); |
4712e722 | 391 | fsnotify_detach_mark(mark); |
d5a335b8 | 392 | mutex_unlock(&group->mark_mutex); |
4712e722 | 393 | fsnotify_free_mark(mark); |
5444e298 EP |
394 | } |
395 | ||
8edc6e16 JK |
396 | /* |
397 | * Sorting function for lists of fsnotify marks. | |
398 | * | |
399 | * Fanotify supports different notification classes (reflected as priority of | |
400 | * notification group). Events shall be passed to notification groups in | |
401 | * decreasing priority order. To achieve this marks in notification lists for | |
402 | * inodes and vfsmounts are sorted so that priorities of corresponding groups | |
403 | * are descending. | |
404 | * | |
405 | * Furthermore correct handling of the ignore mask requires processing inode | |
406 | * and vfsmount marks of each group together. Using the group address as | |
407 | * further sort criterion provides a unique sorting order and thus we can | |
408 | * merge inode and vfsmount lists of marks in linear time and find groups | |
409 | * present in both lists. | |
410 | * | |
411 | * A return value of 1 signifies that b has priority over a. | |
412 | * A return value of 0 signifies that the two marks have to be handled together. | |
413 | * A return value of -1 signifies that a has priority over b. | |
414 | */ | |
415 | int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b) | |
416 | { | |
417 | if (a == b) | |
418 | return 0; | |
419 | if (!a) | |
420 | return 1; | |
421 | if (!b) | |
422 | return -1; | |
423 | if (a->priority < b->priority) | |
424 | return 1; | |
425 | if (a->priority > b->priority) | |
426 | return -1; | |
427 | if (a < b) | |
428 | return 1; | |
429 | return -1; | |
430 | } | |
431 | ||
9dd813c1 | 432 | static int fsnotify_attach_connector_to_object( |
08991e83 JK |
433 | struct fsnotify_mark_connector __rcu **connp, |
434 | struct inode *inode, | |
435 | struct vfsmount *mnt) | |
9dd813c1 JK |
436 | { |
437 | struct fsnotify_mark_connector *conn; | |
438 | ||
755b5bc6 | 439 | conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL); |
9dd813c1 JK |
440 | if (!conn) |
441 | return -ENOMEM; | |
04662cab | 442 | spin_lock_init(&conn->lock); |
9dd813c1 | 443 | INIT_HLIST_HEAD(&conn->list); |
86ffe245 JK |
444 | if (inode) { |
445 | conn->flags = FSNOTIFY_OBJ_TYPE_INODE; | |
08991e83 | 446 | conn->inode = igrab(inode); |
86ffe245 JK |
447 | } else { |
448 | conn->flags = FSNOTIFY_OBJ_TYPE_VFSMOUNT; | |
449 | conn->mnt = mnt; | |
450 | } | |
9dd813c1 | 451 | /* |
04662cab JK |
452 | * cmpxchg() provides the barrier so that readers of *connp can see |
453 | * only initialized structure | |
9dd813c1 | 454 | */ |
04662cab JK |
455 | if (cmpxchg(connp, NULL, conn)) { |
456 | /* Someone else created list structure for us */ | |
08991e83 JK |
457 | if (inode) |
458 | iput(inode); | |
755b5bc6 | 459 | kmem_cache_free(fsnotify_mark_connector_cachep, conn); |
04662cab | 460 | } |
9dd813c1 JK |
461 | |
462 | return 0; | |
463 | } | |
464 | ||
08991e83 JK |
465 | /* |
466 | * Get mark connector, make sure it is alive and return with its lock held. | |
467 | * This is for users that get connector pointer from inode or mount. Users that | |
468 | * hold reference to a mark on the list may directly lock connector->lock as | |
469 | * they are sure list cannot go away under them. | |
470 | */ | |
471 | static struct fsnotify_mark_connector *fsnotify_grab_connector( | |
472 | struct fsnotify_mark_connector __rcu **connp) | |
473 | { | |
474 | struct fsnotify_mark_connector *conn; | |
475 | int idx; | |
476 | ||
477 | idx = srcu_read_lock(&fsnotify_mark_srcu); | |
478 | conn = srcu_dereference(*connp, &fsnotify_mark_srcu); | |
479 | if (!conn) | |
480 | goto out; | |
481 | spin_lock(&conn->lock); | |
482 | if (!(conn->flags & (FSNOTIFY_OBJ_TYPE_INODE | | |
483 | FSNOTIFY_OBJ_TYPE_VFSMOUNT))) { | |
484 | spin_unlock(&conn->lock); | |
485 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
486 | return NULL; | |
487 | } | |
488 | out: | |
489 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
490 | return conn; | |
491 | } | |
492 | ||
9dd813c1 JK |
493 | /* |
494 | * Add mark into proper place in given list of marks. These marks may be used | |
495 | * for the fsnotify backend to determine which event types should be delivered | |
496 | * to which group and for which inodes. These marks are ordered according to | |
497 | * priority, highest number first, and then by the group's location in memory. | |
498 | */ | |
755b5bc6 JK |
499 | static int fsnotify_add_mark_list(struct fsnotify_mark *mark, |
500 | struct inode *inode, struct vfsmount *mnt, | |
501 | int allow_dups) | |
0809ab69 JK |
502 | { |
503 | struct fsnotify_mark *lmark, *last = NULL; | |
9dd813c1 | 504 | struct fsnotify_mark_connector *conn; |
08991e83 | 505 | struct fsnotify_mark_connector __rcu **connp; |
0809ab69 | 506 | int cmp; |
755b5bc6 JK |
507 | int err = 0; |
508 | ||
509 | if (WARN_ON(!inode && !mnt)) | |
510 | return -EINVAL; | |
04662cab | 511 | if (inode) |
755b5bc6 | 512 | connp = &inode->i_fsnotify_marks; |
04662cab | 513 | else |
755b5bc6 | 514 | connp = &real_mount(mnt)->mnt_fsnotify_marks; |
08991e83 JK |
515 | restart: |
516 | spin_lock(&mark->lock); | |
517 | conn = fsnotify_grab_connector(connp); | |
518 | if (!conn) { | |
519 | spin_unlock(&mark->lock); | |
04662cab | 520 | err = fsnotify_attach_connector_to_object(connp, inode, mnt); |
9dd813c1 JK |
521 | if (err) |
522 | return err; | |
08991e83 | 523 | goto restart; |
9dd813c1 | 524 | } |
0809ab69 JK |
525 | |
526 | /* is mark the first mark? */ | |
9dd813c1 JK |
527 | if (hlist_empty(&conn->list)) { |
528 | hlist_add_head_rcu(&mark->obj_list, &conn->list); | |
86ffe245 | 529 | goto added; |
0809ab69 JK |
530 | } |
531 | ||
532 | /* should mark be in the middle of the current list? */ | |
9dd813c1 | 533 | hlist_for_each_entry(lmark, &conn->list, obj_list) { |
0809ab69 JK |
534 | last = lmark; |
535 | ||
6b3f05d2 JK |
536 | if ((lmark->group == mark->group) && |
537 | (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) && | |
538 | !allow_dups) { | |
755b5bc6 JK |
539 | err = -EEXIST; |
540 | goto out_err; | |
541 | } | |
0809ab69 JK |
542 | |
543 | cmp = fsnotify_compare_groups(lmark->group, mark->group); | |
544 | if (cmp >= 0) { | |
545 | hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list); | |
86ffe245 | 546 | goto added; |
0809ab69 JK |
547 | } |
548 | } | |
549 | ||
550 | BUG_ON(last == NULL); | |
551 | /* mark should be the last entry. last is the current last entry */ | |
552 | hlist_add_behind_rcu(&mark->obj_list, &last->obj_list); | |
86ffe245 JK |
553 | added: |
554 | mark->connector = conn; | |
755b5bc6 | 555 | out_err: |
04662cab | 556 | spin_unlock(&conn->lock); |
755b5bc6 JK |
557 | spin_unlock(&mark->lock); |
558 | return err; | |
0809ab69 JK |
559 | } |
560 | ||
5444e298 EP |
561 | /* |
562 | * Attach an initialized mark to a given group and fs object. | |
563 | * These marks may be used for the fsnotify backend to determine which | |
564 | * event types should be delivered to which group. | |
565 | */ | |
7b129323 | 566 | int fsnotify_add_mark_locked(struct fsnotify_mark *mark, struct inode *inode, |
d5a335b8 | 567 | struct vfsmount *mnt, int allow_dups) |
5444e298 | 568 | { |
7b129323 | 569 | struct fsnotify_group *group = mark->group; |
5444e298 EP |
570 | int ret = 0; |
571 | ||
5444e298 EP |
572 | BUG_ON(inode && mnt); |
573 | BUG_ON(!inode && !mnt); | |
d5a335b8 | 574 | BUG_ON(!mutex_is_locked(&group->mark_mutex)); |
5444e298 | 575 | |
5444e298 EP |
576 | /* |
577 | * LOCKING ORDER!!!! | |
986ab098 | 578 | * group->mark_mutex |
104d06f0 | 579 | * mark->lock |
04662cab | 580 | * mark->connector->lock |
5444e298 | 581 | */ |
104d06f0 | 582 | spin_lock(&mark->lock); |
4712e722 | 583 | mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED; |
700307a2 | 584 | |
5444e298 EP |
585 | list_add(&mark->g_list, &group->marks_list); |
586 | atomic_inc(&group->num_marks); | |
6b3f05d2 | 587 | fsnotify_get_mark(mark); /* for g_list */ |
5444e298 EP |
588 | spin_unlock(&mark->lock); |
589 | ||
755b5bc6 JK |
590 | ret = fsnotify_add_mark_list(mark, inode, mnt, allow_dups); |
591 | if (ret) | |
592 | goto err; | |
593 | ||
a242677b JK |
594 | if (mark->mask) |
595 | fsnotify_recalc_mask(mark->connector); | |
5444e298 EP |
596 | |
597 | return ret; | |
598 | err: | |
9cf90cef | 599 | spin_lock(&mark->lock); |
11375145 JK |
600 | mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE | |
601 | FSNOTIFY_MARK_FLAG_ATTACHED); | |
5444e298 | 602 | list_del_init(&mark->g_list); |
9cf90cef | 603 | spin_unlock(&mark->lock); |
5444e298 | 604 | atomic_dec(&group->num_marks); |
5444e298 | 605 | |
11375145 | 606 | fsnotify_put_mark(mark); |
5444e298 EP |
607 | return ret; |
608 | } | |
609 | ||
7b129323 JK |
610 | int fsnotify_add_mark(struct fsnotify_mark *mark, struct inode *inode, |
611 | struct vfsmount *mnt, int allow_dups) | |
d5a335b8 LS |
612 | { |
613 | int ret; | |
7b129323 JK |
614 | struct fsnotify_group *group = mark->group; |
615 | ||
d5a335b8 | 616 | mutex_lock(&group->mark_mutex); |
7b129323 | 617 | ret = fsnotify_add_mark_locked(mark, inode, mnt, allow_dups); |
d5a335b8 LS |
618 | mutex_unlock(&group->mark_mutex); |
619 | return ret; | |
620 | } | |
621 | ||
0809ab69 JK |
622 | /* |
623 | * Given a list of marks, find the mark associated with given group. If found | |
624 | * take a reference to that mark and return it, else return NULL. | |
625 | */ | |
08991e83 JK |
626 | struct fsnotify_mark *fsnotify_find_mark( |
627 | struct fsnotify_mark_connector __rcu **connp, | |
628 | struct fsnotify_group *group) | |
0809ab69 | 629 | { |
08991e83 | 630 | struct fsnotify_mark_connector *conn; |
0809ab69 JK |
631 | struct fsnotify_mark *mark; |
632 | ||
08991e83 | 633 | conn = fsnotify_grab_connector(connp); |
9dd813c1 JK |
634 | if (!conn) |
635 | return NULL; | |
636 | ||
637 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
6b3f05d2 JK |
638 | if (mark->group == group && |
639 | (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { | |
0809ab69 | 640 | fsnotify_get_mark(mark); |
04662cab | 641 | spin_unlock(&conn->lock); |
0809ab69 JK |
642 | return mark; |
643 | } | |
644 | } | |
04662cab | 645 | spin_unlock(&conn->lock); |
0809ab69 JK |
646 | return NULL; |
647 | } | |
648 | ||
18f2e0d3 JK |
649 | /* Clear any marks in a group with given type */ |
650 | void fsnotify_clear_marks_by_group(struct fsnotify_group *group, | |
651 | unsigned int type) | |
5444e298 EP |
652 | { |
653 | struct fsnotify_mark *lmark, *mark; | |
8f2f3eb5 | 654 | LIST_HEAD(to_free); |
2e37c6ca | 655 | struct list_head *head = &to_free; |
5444e298 | 656 | |
2e37c6ca JK |
657 | /* Skip selection step if we want to clear all marks. */ |
658 | if (type == FSNOTIFY_OBJ_ALL_TYPES) { | |
659 | head = &group->marks_list; | |
660 | goto clear; | |
661 | } | |
8f2f3eb5 JK |
662 | /* |
663 | * We have to be really careful here. Anytime we drop mark_mutex, e.g. | |
664 | * fsnotify_clear_marks_by_inode() can come and free marks. Even in our | |
665 | * to_free list so we have to use mark_mutex even when accessing that | |
666 | * list. And freeing mark requires us to drop mark_mutex. So we can | |
667 | * reliably free only the first mark in the list. That's why we first | |
668 | * move marks to free to to_free list in one go and then free marks in | |
669 | * to_free list one by one. | |
670 | */ | |
6960b0d9 | 671 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); |
5444e298 | 672 | list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) { |
18f2e0d3 | 673 | if (mark->connector->flags & type) |
8f2f3eb5 | 674 | list_move(&mark->g_list, &to_free); |
5444e298 | 675 | } |
986ab098 | 676 | mutex_unlock(&group->mark_mutex); |
8f2f3eb5 | 677 | |
2e37c6ca | 678 | clear: |
8f2f3eb5 JK |
679 | while (1) { |
680 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); | |
2e37c6ca | 681 | if (list_empty(head)) { |
8f2f3eb5 JK |
682 | mutex_unlock(&group->mark_mutex); |
683 | break; | |
684 | } | |
2e37c6ca | 685 | mark = list_first_entry(head, struct fsnotify_mark, g_list); |
8f2f3eb5 | 686 | fsnotify_get_mark(mark); |
4712e722 | 687 | fsnotify_detach_mark(mark); |
8f2f3eb5 | 688 | mutex_unlock(&group->mark_mutex); |
4712e722 | 689 | fsnotify_free_mark(mark); |
8f2f3eb5 JK |
690 | fsnotify_put_mark(mark); |
691 | } | |
5444e298 EP |
692 | } |
693 | ||
08991e83 JK |
694 | /* Destroy all marks attached to inode / vfsmount */ |
695 | void fsnotify_destroy_marks(struct fsnotify_mark_connector __rcu **connp) | |
0810b4f9 | 696 | { |
08991e83 | 697 | struct fsnotify_mark_connector *conn; |
6b3f05d2 JK |
698 | struct fsnotify_mark *mark, *old_mark = NULL; |
699 | struct inode *inode; | |
0810b4f9 | 700 | |
6b3f05d2 JK |
701 | conn = fsnotify_grab_connector(connp); |
702 | if (!conn) | |
703 | return; | |
704 | /* | |
705 | * We have to be careful since we can race with e.g. | |
706 | * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the | |
707 | * list can get modified. However we are holding mark reference and | |
708 | * thus our mark cannot be removed from obj_list so we can continue | |
709 | * iteration after regaining conn->lock. | |
710 | */ | |
711 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
0810b4f9 | 712 | fsnotify_get_mark(mark); |
04662cab | 713 | spin_unlock(&conn->lock); |
6b3f05d2 JK |
714 | if (old_mark) |
715 | fsnotify_put_mark(old_mark); | |
716 | old_mark = mark; | |
0810b4f9 | 717 | fsnotify_destroy_mark(mark, mark->group); |
6b3f05d2 | 718 | spin_lock(&conn->lock); |
0810b4f9 | 719 | } |
6b3f05d2 JK |
720 | /* |
721 | * Detach list from object now so that we don't pin inode until all | |
722 | * mark references get dropped. It would lead to strange results such | |
723 | * as delaying inode deletion or blocking unmount. | |
724 | */ | |
725 | inode = fsnotify_detach_connector_from_object(conn); | |
726 | spin_unlock(&conn->lock); | |
727 | if (old_mark) | |
728 | fsnotify_put_mark(old_mark); | |
729 | iput(inode); | |
0810b4f9 JK |
730 | } |
731 | ||
5444e298 EP |
732 | /* |
733 | * Nothing fancy, just initialize lists and locks and counters. | |
734 | */ | |
735 | void fsnotify_init_mark(struct fsnotify_mark *mark, | |
054c636e | 736 | struct fsnotify_group *group) |
5444e298 | 737 | { |
ba643f04 | 738 | memset(mark, 0, sizeof(*mark)); |
5444e298 EP |
739 | spin_lock_init(&mark->lock); |
740 | atomic_set(&mark->refcnt, 1); | |
7b129323 JK |
741 | fsnotify_get_group(group); |
742 | mark->group = group; | |
5444e298 | 743 | } |
13d34ac6 | 744 | |
35e48176 JK |
745 | /* |
746 | * Destroy all marks in destroy_list, waits for SRCU period to finish before | |
747 | * actually freeing marks. | |
748 | */ | |
f09b04a0 | 749 | static void fsnotify_mark_destroy_workfn(struct work_struct *work) |
13d34ac6 JL |
750 | { |
751 | struct fsnotify_mark *mark, *next; | |
752 | struct list_head private_destroy_list; | |
753 | ||
0918f1c3 JL |
754 | spin_lock(&destroy_lock); |
755 | /* exchange the list head */ | |
756 | list_replace_init(&destroy_list, &private_destroy_list); | |
757 | spin_unlock(&destroy_lock); | |
13d34ac6 | 758 | |
0918f1c3 | 759 | synchronize_srcu(&fsnotify_mark_srcu); |
13d34ac6 | 760 | |
0918f1c3 JL |
761 | list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) { |
762 | list_del_init(&mark->g_list); | |
6b3f05d2 | 763 | fsnotify_final_mark_destroy(mark); |
13d34ac6 | 764 | } |
13d34ac6 | 765 | } |
35e48176 | 766 | |
f09b04a0 JK |
767 | /* Wait for all marks queued for destruction to be actually destroyed */ |
768 | void fsnotify_wait_marks_destroyed(void) | |
35e48176 | 769 | { |
f09b04a0 | 770 | flush_delayed_work(&reaper_work); |
35e48176 | 771 | } |