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 | { | |
ab97f873 ER |
108 | WARN_ON_ONCE(!refcount_read(&mark->refcnt)); |
109 | refcount_inc(&mark->refcnt); | |
5444e298 EP |
110 | } |
111 | ||
3ac70bfc AG |
112 | static __u32 *fsnotify_conn_mask_p(struct fsnotify_mark_connector *conn) |
113 | { | |
114 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) | |
115 | return &fsnotify_conn_inode(conn)->i_fsnotify_mask; | |
116 | else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) | |
117 | return &fsnotify_conn_mount(conn)->mnt_fsnotify_mask; | |
1e6cb723 AG |
118 | else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) |
119 | return &fsnotify_conn_sb(conn)->s_fsnotify_mask; | |
3ac70bfc AG |
120 | return NULL; |
121 | } | |
122 | ||
123 | __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn) | |
124 | { | |
125 | if (WARN_ON(!fsnotify_valid_obj_type(conn->type))) | |
126 | return 0; | |
127 | ||
128 | return *fsnotify_conn_mask_p(conn); | |
129 | } | |
130 | ||
a242677b | 131 | static void __fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) |
0809ab69 JK |
132 | { |
133 | u32 new_mask = 0; | |
134 | struct fsnotify_mark *mark; | |
135 | ||
04662cab | 136 | assert_spin_locked(&conn->lock); |
d3bc0fa8 JK |
137 | /* We can get detached connector here when inode is getting unlinked. */ |
138 | if (!fsnotify_valid_obj_type(conn->type)) | |
139 | return; | |
6b3f05d2 JK |
140 | hlist_for_each_entry(mark, &conn->list, obj_list) { |
141 | if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) | |
142 | new_mask |= mark->mask; | |
143 | } | |
3ac70bfc | 144 | *fsnotify_conn_mask_p(conn) = new_mask; |
a242677b JK |
145 | } |
146 | ||
147 | /* | |
148 | * Calculate mask of events for a list of marks. The caller must make sure | |
36f10f55 | 149 | * connector and connector->obj cannot disappear under us. Callers achieve |
6b3f05d2 JK |
150 | * this by holding a mark->lock or mark->group->mark_mutex for a mark on this |
151 | * list. | |
a242677b JK |
152 | */ |
153 | void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) | |
154 | { | |
155 | if (!conn) | |
156 | return; | |
157 | ||
04662cab | 158 | spin_lock(&conn->lock); |
a242677b | 159 | __fsnotify_recalc_mask(conn); |
04662cab | 160 | spin_unlock(&conn->lock); |
d6f7b98b | 161 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) |
36f10f55 AG |
162 | __fsnotify_update_child_dentry_flags( |
163 | fsnotify_conn_inode(conn)); | |
0809ab69 JK |
164 | } |
165 | ||
08991e83 JK |
166 | /* Free all connectors queued for freeing once SRCU period ends */ |
167 | static void fsnotify_connector_destroy_workfn(struct work_struct *work) | |
168 | { | |
169 | struct fsnotify_mark_connector *conn, *free; | |
170 | ||
171 | spin_lock(&destroy_lock); | |
172 | conn = connector_destroy_list; | |
173 | connector_destroy_list = NULL; | |
174 | spin_unlock(&destroy_lock); | |
175 | ||
176 | synchronize_srcu(&fsnotify_mark_srcu); | |
177 | while (conn) { | |
178 | free = conn; | |
179 | conn = conn->destroy_next; | |
180 | kmem_cache_free(fsnotify_mark_connector_cachep, free); | |
181 | } | |
182 | } | |
183 | ||
721fb6fb JK |
184 | static void *fsnotify_detach_connector_from_object( |
185 | struct fsnotify_mark_connector *conn, | |
186 | unsigned int *type) | |
08991e83 JK |
187 | { |
188 | struct inode *inode = NULL; | |
189 | ||
721fb6fb | 190 | *type = conn->type; |
36f10f55 AG |
191 | if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) |
192 | return NULL; | |
193 | ||
d6f7b98b | 194 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) { |
36f10f55 | 195 | inode = fsnotify_conn_inode(conn); |
08991e83 | 196 | inode->i_fsnotify_mask = 0; |
721fb6fb | 197 | atomic_long_inc(&inode->i_sb->s_fsnotify_inode_refs); |
d6f7b98b | 198 | } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) { |
36f10f55 | 199 | fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0; |
1e6cb723 AG |
200 | } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) { |
201 | fsnotify_conn_sb(conn)->s_fsnotify_mask = 0; | |
08991e83 JK |
202 | } |
203 | ||
36f10f55 AG |
204 | rcu_assign_pointer(*(conn->obj), NULL); |
205 | conn->obj = NULL; | |
206 | conn->type = FSNOTIFY_OBJ_TYPE_DETACHED; | |
207 | ||
08991e83 JK |
208 | return inode; |
209 | } | |
210 | ||
6b3f05d2 JK |
211 | static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark) |
212 | { | |
054c636e JK |
213 | struct fsnotify_group *group = mark->group; |
214 | ||
215 | if (WARN_ON_ONCE(!group)) | |
216 | return; | |
217 | group->ops->free_mark(mark); | |
218 | fsnotify_put_group(group); | |
6b3f05d2 JK |
219 | } |
220 | ||
721fb6fb JK |
221 | /* Drop object reference originally held by a connector */ |
222 | static void fsnotify_drop_object(unsigned int type, void *objp) | |
223 | { | |
224 | struct inode *inode; | |
225 | struct super_block *sb; | |
226 | ||
227 | if (!objp) | |
228 | return; | |
229 | /* Currently only inode references are passed to be dropped */ | |
230 | if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE)) | |
231 | return; | |
232 | inode = objp; | |
233 | sb = inode->i_sb; | |
234 | iput(inode); | |
235 | if (atomic_long_dec_and_test(&sb->s_fsnotify_inode_refs)) | |
236 | wake_up_var(&sb->s_fsnotify_inode_refs); | |
237 | } | |
238 | ||
6b3f05d2 | 239 | void fsnotify_put_mark(struct fsnotify_mark *mark) |
8212a609 JK |
240 | { |
241 | struct fsnotify_mark_connector *conn; | |
721fb6fb JK |
242 | void *objp = NULL; |
243 | unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED; | |
08991e83 | 244 | bool free_conn = false; |
8212a609 | 245 | |
6b3f05d2 JK |
246 | /* Catch marks that were actually never attached to object */ |
247 | if (!mark->connector) { | |
ab97f873 | 248 | if (refcount_dec_and_test(&mark->refcnt)) |
6b3f05d2 JK |
249 | fsnotify_final_mark_destroy(mark); |
250 | return; | |
251 | } | |
252 | ||
253 | /* | |
254 | * We have to be careful so that traversals of obj_list under lock can | |
255 | * safely grab mark reference. | |
256 | */ | |
ab97f873 | 257 | if (!refcount_dec_and_lock(&mark->refcnt, &mark->connector->lock)) |
6b3f05d2 JK |
258 | return; |
259 | ||
8212a609 | 260 | conn = mark->connector; |
8212a609 JK |
261 | hlist_del_init_rcu(&mark->obj_list); |
262 | if (hlist_empty(&conn->list)) { | |
721fb6fb | 263 | objp = fsnotify_detach_connector_from_object(conn, &type); |
08991e83 JK |
264 | free_conn = true; |
265 | } else { | |
266 | __fsnotify_recalc_mask(conn); | |
8212a609 JK |
267 | } |
268 | mark->connector = NULL; | |
04662cab | 269 | spin_unlock(&conn->lock); |
8212a609 | 270 | |
721fb6fb | 271 | fsnotify_drop_object(type, objp); |
6b3f05d2 | 272 | |
08991e83 JK |
273 | if (free_conn) { |
274 | spin_lock(&destroy_lock); | |
275 | conn->destroy_next = connector_destroy_list; | |
276 | connector_destroy_list = conn; | |
277 | spin_unlock(&destroy_lock); | |
278 | queue_work(system_unbound_wq, &connector_reaper_work); | |
279 | } | |
6b3f05d2 JK |
280 | /* |
281 | * Note that we didn't update flags telling whether inode cares about | |
282 | * what's happening with children. We update these flags from | |
283 | * __fsnotify_parent() lazily when next event happens on one of our | |
284 | * children. | |
285 | */ | |
286 | spin_lock(&destroy_lock); | |
287 | list_add(&mark->g_list, &destroy_list); | |
288 | spin_unlock(&destroy_lock); | |
289 | queue_delayed_work(system_unbound_wq, &reaper_work, | |
290 | FSNOTIFY_REAPER_DELAY); | |
8212a609 JK |
291 | } |
292 | ||
24c20305 MS |
293 | /* |
294 | * Get mark reference when we found the mark via lockless traversal of object | |
295 | * list. Mark can be already removed from the list by now and on its way to be | |
296 | * destroyed once SRCU period ends. | |
297 | * | |
298 | * Also pin the group so it doesn't disappear under us. | |
299 | */ | |
300 | static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark) | |
abc77577 | 301 | { |
24c20305 MS |
302 | if (!mark) |
303 | return true; | |
abc77577 | 304 | |
ab97f873 | 305 | if (refcount_inc_not_zero(&mark->refcnt)) { |
9a31d7ad MS |
306 | spin_lock(&mark->lock); |
307 | if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) { | |
308 | /* mark is attached, group is still alive then */ | |
309 | atomic_inc(&mark->group->user_waits); | |
310 | spin_unlock(&mark->lock); | |
311 | return true; | |
312 | } | |
313 | spin_unlock(&mark->lock); | |
314 | fsnotify_put_mark(mark); | |
315 | } | |
24c20305 MS |
316 | return false; |
317 | } | |
318 | ||
319 | /* | |
320 | * Puts marks and wakes up group destruction if necessary. | |
321 | * | |
322 | * Pairs with fsnotify_get_mark_safe() | |
323 | */ | |
324 | static void fsnotify_put_mark_wake(struct fsnotify_mark *mark) | |
325 | { | |
326 | if (mark) { | |
327 | struct fsnotify_group *group = mark->group; | |
328 | ||
329 | fsnotify_put_mark(mark); | |
330 | /* | |
331 | * We abuse notification_waitq on group shutdown for waiting for | |
332 | * all marks pinned when waiting for userspace. | |
333 | */ | |
334 | if (atomic_dec_and_test(&group->user_waits) && group->shutdown) | |
335 | wake_up(&group->notification_waitq); | |
abc77577 | 336 | } |
24c20305 MS |
337 | } |
338 | ||
339 | bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info) | |
340 | { | |
47d9c7cc AG |
341 | int type; |
342 | ||
343 | fsnotify_foreach_obj_type(type) { | |
344 | /* This can fail if mark is being removed */ | |
345 | if (!fsnotify_get_mark_safe(iter_info->marks[type])) | |
346 | goto fail; | |
abc77577 JK |
347 | } |
348 | ||
349 | /* | |
350 | * Now that both marks are pinned by refcount in the inode / vfsmount | |
351 | * lists, we can drop SRCU lock, and safely resume the list iteration | |
352 | * once userspace returns. | |
353 | */ | |
354 | srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx); | |
355 | ||
356 | return true; | |
47d9c7cc AG |
357 | |
358 | fail: | |
359 | for (type--; type >= 0; type--) | |
360 | fsnotify_put_mark_wake(iter_info->marks[type]); | |
361 | return false; | |
abc77577 JK |
362 | } |
363 | ||
364 | void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info) | |
365 | { | |
47d9c7cc AG |
366 | int type; |
367 | ||
abc77577 | 368 | iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); |
47d9c7cc AG |
369 | fsnotify_foreach_obj_type(type) |
370 | fsnotify_put_mark_wake(iter_info->marks[type]); | |
abc77577 JK |
371 | } |
372 | ||
5444e298 | 373 | /* |
6b3f05d2 JK |
374 | * Mark mark as detached, remove it from group list. Mark still stays in object |
375 | * list until its last reference is dropped. Note that we rely on mark being | |
376 | * removed from group list before corresponding reference to it is dropped. In | |
377 | * particular we rely on mark->connector being valid while we hold | |
378 | * group->mark_mutex if we found the mark through g_list. | |
4712e722 | 379 | * |
11375145 JK |
380 | * Must be called with group->mark_mutex held. The caller must either hold |
381 | * reference to the mark or be protected by fsnotify_mark_srcu. | |
5444e298 | 382 | */ |
4712e722 | 383 | void fsnotify_detach_mark(struct fsnotify_mark *mark) |
5444e298 | 384 | { |
4712e722 | 385 | struct fsnotify_group *group = mark->group; |
5444e298 | 386 | |
11375145 JK |
387 | WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex)); |
388 | WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) && | |
ab97f873 | 389 | refcount_read(&mark->refcnt) < 1 + |
11375145 | 390 | !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)); |
d5a335b8 | 391 | |
104d06f0 | 392 | spin_lock(&mark->lock); |
700307a2 | 393 | /* something else already called this function on this mark */ |
4712e722 | 394 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { |
5444e298 | 395 | spin_unlock(&mark->lock); |
e2a29943 | 396 | return; |
5444e298 | 397 | } |
4712e722 | 398 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED; |
5444e298 | 399 | list_del_init(&mark->g_list); |
5444e298 | 400 | spin_unlock(&mark->lock); |
d5a335b8 | 401 | |
4712e722 | 402 | atomic_dec(&group->num_marks); |
11375145 JK |
403 | |
404 | /* Drop mark reference acquired in fsnotify_add_mark_locked() */ | |
405 | fsnotify_put_mark(mark); | |
4712e722 JK |
406 | } |
407 | ||
408 | /* | |
11375145 JK |
409 | * Free fsnotify mark. The mark is actually only marked as being freed. The |
410 | * freeing is actually happening only once last reference to the mark is | |
411 | * dropped from a workqueue which first waits for srcu period end. | |
35e48176 | 412 | * |
11375145 JK |
413 | * Caller must have a reference to the mark or be protected by |
414 | * fsnotify_mark_srcu. | |
4712e722 | 415 | */ |
11375145 | 416 | void fsnotify_free_mark(struct fsnotify_mark *mark) |
4712e722 JK |
417 | { |
418 | struct fsnotify_group *group = mark->group; | |
419 | ||
420 | spin_lock(&mark->lock); | |
421 | /* something else already called this function on this mark */ | |
422 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) { | |
423 | spin_unlock(&mark->lock); | |
11375145 | 424 | return; |
4712e722 JK |
425 | } |
426 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE; | |
427 | spin_unlock(&mark->lock); | |
5444e298 | 428 | |
d725e66c LT |
429 | /* |
430 | * Some groups like to know that marks are being freed. This is a | |
431 | * callback to the group function to let it know that this mark | |
432 | * is being freed. | |
433 | */ | |
434 | if (group->ops->freeing_mark) | |
435 | group->ops->freeing_mark(mark, group); | |
d5a335b8 LS |
436 | } |
437 | ||
438 | void fsnotify_destroy_mark(struct fsnotify_mark *mark, | |
439 | struct fsnotify_group *group) | |
440 | { | |
6960b0d9 | 441 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); |
4712e722 | 442 | fsnotify_detach_mark(mark); |
d5a335b8 | 443 | mutex_unlock(&group->mark_mutex); |
4712e722 | 444 | fsnotify_free_mark(mark); |
5444e298 EP |
445 | } |
446 | ||
8edc6e16 JK |
447 | /* |
448 | * Sorting function for lists of fsnotify marks. | |
449 | * | |
450 | * Fanotify supports different notification classes (reflected as priority of | |
451 | * notification group). Events shall be passed to notification groups in | |
452 | * decreasing priority order. To achieve this marks in notification lists for | |
453 | * inodes and vfsmounts are sorted so that priorities of corresponding groups | |
454 | * are descending. | |
455 | * | |
456 | * Furthermore correct handling of the ignore mask requires processing inode | |
457 | * and vfsmount marks of each group together. Using the group address as | |
458 | * further sort criterion provides a unique sorting order and thus we can | |
459 | * merge inode and vfsmount lists of marks in linear time and find groups | |
460 | * present in both lists. | |
461 | * | |
462 | * A return value of 1 signifies that b has priority over a. | |
463 | * A return value of 0 signifies that the two marks have to be handled together. | |
464 | * A return value of -1 signifies that a has priority over b. | |
465 | */ | |
466 | int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b) | |
467 | { | |
468 | if (a == b) | |
469 | return 0; | |
470 | if (!a) | |
471 | return 1; | |
472 | if (!b) | |
473 | return -1; | |
474 | if (a->priority < b->priority) | |
475 | return 1; | |
476 | if (a->priority > b->priority) | |
477 | return -1; | |
478 | if (a < b) | |
479 | return 1; | |
480 | return -1; | |
481 | } | |
482 | ||
9b6e5434 | 483 | static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp, |
b812a9f5 | 484 | unsigned int type) |
9dd813c1 | 485 | { |
b812a9f5 | 486 | struct inode *inode = NULL; |
9dd813c1 JK |
487 | struct fsnotify_mark_connector *conn; |
488 | ||
755b5bc6 | 489 | conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL); |
9dd813c1 JK |
490 | if (!conn) |
491 | return -ENOMEM; | |
04662cab | 492 | spin_lock_init(&conn->lock); |
9dd813c1 | 493 | INIT_HLIST_HEAD(&conn->list); |
b812a9f5 | 494 | conn->type = type; |
36f10f55 | 495 | conn->obj = connp; |
b812a9f5 | 496 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) |
36f10f55 | 497 | inode = igrab(fsnotify_conn_inode(conn)); |
9dd813c1 | 498 | /* |
04662cab JK |
499 | * cmpxchg() provides the barrier so that readers of *connp can see |
500 | * only initialized structure | |
9dd813c1 | 501 | */ |
04662cab JK |
502 | if (cmpxchg(connp, NULL, conn)) { |
503 | /* Someone else created list structure for us */ | |
08991e83 JK |
504 | if (inode) |
505 | iput(inode); | |
755b5bc6 | 506 | kmem_cache_free(fsnotify_mark_connector_cachep, conn); |
04662cab | 507 | } |
9dd813c1 JK |
508 | |
509 | return 0; | |
510 | } | |
511 | ||
08991e83 JK |
512 | /* |
513 | * Get mark connector, make sure it is alive and return with its lock held. | |
514 | * This is for users that get connector pointer from inode or mount. Users that | |
515 | * hold reference to a mark on the list may directly lock connector->lock as | |
516 | * they are sure list cannot go away under them. | |
517 | */ | |
518 | static struct fsnotify_mark_connector *fsnotify_grab_connector( | |
9b6e5434 | 519 | fsnotify_connp_t *connp) |
08991e83 JK |
520 | { |
521 | struct fsnotify_mark_connector *conn; | |
522 | int idx; | |
523 | ||
524 | idx = srcu_read_lock(&fsnotify_mark_srcu); | |
525 | conn = srcu_dereference(*connp, &fsnotify_mark_srcu); | |
526 | if (!conn) | |
527 | goto out; | |
528 | spin_lock(&conn->lock); | |
d6f7b98b | 529 | if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) { |
08991e83 JK |
530 | spin_unlock(&conn->lock); |
531 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
532 | return NULL; | |
533 | } | |
534 | out: | |
535 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
536 | return conn; | |
537 | } | |
538 | ||
9dd813c1 JK |
539 | /* |
540 | * Add mark into proper place in given list of marks. These marks may be used | |
541 | * for the fsnotify backend to determine which event types should be delivered | |
542 | * to which group and for which inodes. These marks are ordered according to | |
543 | * priority, highest number first, and then by the group's location in memory. | |
544 | */ | |
755b5bc6 | 545 | static int fsnotify_add_mark_list(struct fsnotify_mark *mark, |
b812a9f5 | 546 | fsnotify_connp_t *connp, unsigned int type, |
755b5bc6 | 547 | int allow_dups) |
0809ab69 JK |
548 | { |
549 | struct fsnotify_mark *lmark, *last = NULL; | |
9dd813c1 | 550 | struct fsnotify_mark_connector *conn; |
0809ab69 | 551 | int cmp; |
755b5bc6 JK |
552 | int err = 0; |
553 | ||
b812a9f5 | 554 | if (WARN_ON(!fsnotify_valid_obj_type(type))) |
755b5bc6 | 555 | return -EINVAL; |
08991e83 JK |
556 | restart: |
557 | spin_lock(&mark->lock); | |
558 | conn = fsnotify_grab_connector(connp); | |
559 | if (!conn) { | |
560 | spin_unlock(&mark->lock); | |
b812a9f5 | 561 | err = fsnotify_attach_connector_to_object(connp, type); |
9dd813c1 JK |
562 | if (err) |
563 | return err; | |
08991e83 | 564 | goto restart; |
9dd813c1 | 565 | } |
0809ab69 JK |
566 | |
567 | /* is mark the first mark? */ | |
9dd813c1 JK |
568 | if (hlist_empty(&conn->list)) { |
569 | hlist_add_head_rcu(&mark->obj_list, &conn->list); | |
86ffe245 | 570 | goto added; |
0809ab69 JK |
571 | } |
572 | ||
573 | /* should mark be in the middle of the current list? */ | |
9dd813c1 | 574 | hlist_for_each_entry(lmark, &conn->list, obj_list) { |
0809ab69 JK |
575 | last = lmark; |
576 | ||
6b3f05d2 JK |
577 | if ((lmark->group == mark->group) && |
578 | (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) && | |
579 | !allow_dups) { | |
755b5bc6 JK |
580 | err = -EEXIST; |
581 | goto out_err; | |
582 | } | |
0809ab69 JK |
583 | |
584 | cmp = fsnotify_compare_groups(lmark->group, mark->group); | |
585 | if (cmp >= 0) { | |
586 | hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list); | |
86ffe245 | 587 | goto added; |
0809ab69 JK |
588 | } |
589 | } | |
590 | ||
591 | BUG_ON(last == NULL); | |
592 | /* mark should be the last entry. last is the current last entry */ | |
593 | hlist_add_behind_rcu(&mark->obj_list, &last->obj_list); | |
86ffe245 JK |
594 | added: |
595 | mark->connector = conn; | |
755b5bc6 | 596 | out_err: |
04662cab | 597 | spin_unlock(&conn->lock); |
755b5bc6 JK |
598 | spin_unlock(&mark->lock); |
599 | return err; | |
0809ab69 JK |
600 | } |
601 | ||
5444e298 EP |
602 | /* |
603 | * Attach an initialized mark to a given group and fs object. | |
604 | * These marks may be used for the fsnotify backend to determine which | |
605 | * event types should be delivered to which group. | |
606 | */ | |
b812a9f5 AG |
607 | int fsnotify_add_mark_locked(struct fsnotify_mark *mark, |
608 | fsnotify_connp_t *connp, unsigned int type, | |
609 | int allow_dups) | |
5444e298 | 610 | { |
7b129323 | 611 | struct fsnotify_group *group = mark->group; |
5444e298 EP |
612 | int ret = 0; |
613 | ||
d5a335b8 | 614 | BUG_ON(!mutex_is_locked(&group->mark_mutex)); |
5444e298 | 615 | |
5444e298 EP |
616 | /* |
617 | * LOCKING ORDER!!!! | |
986ab098 | 618 | * group->mark_mutex |
104d06f0 | 619 | * mark->lock |
04662cab | 620 | * mark->connector->lock |
5444e298 | 621 | */ |
104d06f0 | 622 | spin_lock(&mark->lock); |
4712e722 | 623 | mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED; |
700307a2 | 624 | |
5444e298 EP |
625 | list_add(&mark->g_list, &group->marks_list); |
626 | atomic_inc(&group->num_marks); | |
6b3f05d2 | 627 | fsnotify_get_mark(mark); /* for g_list */ |
5444e298 EP |
628 | spin_unlock(&mark->lock); |
629 | ||
b812a9f5 | 630 | ret = fsnotify_add_mark_list(mark, connp, type, allow_dups); |
755b5bc6 JK |
631 | if (ret) |
632 | goto err; | |
633 | ||
a242677b JK |
634 | if (mark->mask) |
635 | fsnotify_recalc_mask(mark->connector); | |
5444e298 EP |
636 | |
637 | return ret; | |
638 | err: | |
9cf90cef | 639 | spin_lock(&mark->lock); |
11375145 JK |
640 | mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE | |
641 | FSNOTIFY_MARK_FLAG_ATTACHED); | |
5444e298 | 642 | list_del_init(&mark->g_list); |
9cf90cef | 643 | spin_unlock(&mark->lock); |
5444e298 | 644 | atomic_dec(&group->num_marks); |
5444e298 | 645 | |
11375145 | 646 | fsnotify_put_mark(mark); |
5444e298 EP |
647 | return ret; |
648 | } | |
649 | ||
b812a9f5 AG |
650 | int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp, |
651 | unsigned int type, int allow_dups) | |
d5a335b8 LS |
652 | { |
653 | int ret; | |
7b129323 JK |
654 | struct fsnotify_group *group = mark->group; |
655 | ||
d5a335b8 | 656 | mutex_lock(&group->mark_mutex); |
b812a9f5 | 657 | ret = fsnotify_add_mark_locked(mark, connp, type, allow_dups); |
d5a335b8 LS |
658 | mutex_unlock(&group->mark_mutex); |
659 | return ret; | |
660 | } | |
661 | ||
0809ab69 JK |
662 | /* |
663 | * Given a list of marks, find the mark associated with given group. If found | |
664 | * take a reference to that mark and return it, else return NULL. | |
665 | */ | |
9b6e5434 AG |
666 | struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp, |
667 | struct fsnotify_group *group) | |
0809ab69 | 668 | { |
08991e83 | 669 | struct fsnotify_mark_connector *conn; |
0809ab69 JK |
670 | struct fsnotify_mark *mark; |
671 | ||
08991e83 | 672 | conn = fsnotify_grab_connector(connp); |
9dd813c1 JK |
673 | if (!conn) |
674 | return NULL; | |
675 | ||
676 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
6b3f05d2 JK |
677 | if (mark->group == group && |
678 | (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { | |
0809ab69 | 679 | fsnotify_get_mark(mark); |
04662cab | 680 | spin_unlock(&conn->lock); |
0809ab69 JK |
681 | return mark; |
682 | } | |
683 | } | |
04662cab | 684 | spin_unlock(&conn->lock); |
0809ab69 JK |
685 | return NULL; |
686 | } | |
687 | ||
d6f7b98b | 688 | /* Clear any marks in a group with given type mask */ |
18f2e0d3 | 689 | void fsnotify_clear_marks_by_group(struct fsnotify_group *group, |
d6f7b98b | 690 | unsigned int type_mask) |
5444e298 EP |
691 | { |
692 | struct fsnotify_mark *lmark, *mark; | |
8f2f3eb5 | 693 | LIST_HEAD(to_free); |
2e37c6ca | 694 | struct list_head *head = &to_free; |
5444e298 | 695 | |
2e37c6ca | 696 | /* Skip selection step if we want to clear all marks. */ |
d6f7b98b | 697 | if (type_mask == FSNOTIFY_OBJ_ALL_TYPES_MASK) { |
2e37c6ca JK |
698 | head = &group->marks_list; |
699 | goto clear; | |
700 | } | |
8f2f3eb5 JK |
701 | /* |
702 | * We have to be really careful here. Anytime we drop mark_mutex, e.g. | |
703 | * fsnotify_clear_marks_by_inode() can come and free marks. Even in our | |
704 | * to_free list so we have to use mark_mutex even when accessing that | |
705 | * list. And freeing mark requires us to drop mark_mutex. So we can | |
706 | * reliably free only the first mark in the list. That's why we first | |
707 | * move marks to free to to_free list in one go and then free marks in | |
708 | * to_free list one by one. | |
709 | */ | |
6960b0d9 | 710 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); |
5444e298 | 711 | list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) { |
d6f7b98b | 712 | if ((1U << mark->connector->type) & type_mask) |
8f2f3eb5 | 713 | list_move(&mark->g_list, &to_free); |
5444e298 | 714 | } |
986ab098 | 715 | mutex_unlock(&group->mark_mutex); |
8f2f3eb5 | 716 | |
2e37c6ca | 717 | clear: |
8f2f3eb5 JK |
718 | while (1) { |
719 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); | |
2e37c6ca | 720 | if (list_empty(head)) { |
8f2f3eb5 JK |
721 | mutex_unlock(&group->mark_mutex); |
722 | break; | |
723 | } | |
2e37c6ca | 724 | mark = list_first_entry(head, struct fsnotify_mark, g_list); |
8f2f3eb5 | 725 | fsnotify_get_mark(mark); |
4712e722 | 726 | fsnotify_detach_mark(mark); |
8f2f3eb5 | 727 | mutex_unlock(&group->mark_mutex); |
4712e722 | 728 | fsnotify_free_mark(mark); |
8f2f3eb5 JK |
729 | fsnotify_put_mark(mark); |
730 | } | |
5444e298 EP |
731 | } |
732 | ||
9b6e5434 AG |
733 | /* Destroy all marks attached to an object via connector */ |
734 | void fsnotify_destroy_marks(fsnotify_connp_t *connp) | |
0810b4f9 | 735 | { |
08991e83 | 736 | struct fsnotify_mark_connector *conn; |
6b3f05d2 | 737 | struct fsnotify_mark *mark, *old_mark = NULL; |
721fb6fb JK |
738 | void *objp; |
739 | unsigned int type; | |
0810b4f9 | 740 | |
6b3f05d2 JK |
741 | conn = fsnotify_grab_connector(connp); |
742 | if (!conn) | |
743 | return; | |
744 | /* | |
745 | * We have to be careful since we can race with e.g. | |
746 | * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the | |
747 | * list can get modified. However we are holding mark reference and | |
748 | * thus our mark cannot be removed from obj_list so we can continue | |
749 | * iteration after regaining conn->lock. | |
750 | */ | |
751 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
0810b4f9 | 752 | fsnotify_get_mark(mark); |
04662cab | 753 | spin_unlock(&conn->lock); |
6b3f05d2 JK |
754 | if (old_mark) |
755 | fsnotify_put_mark(old_mark); | |
756 | old_mark = mark; | |
0810b4f9 | 757 | fsnotify_destroy_mark(mark, mark->group); |
6b3f05d2 | 758 | spin_lock(&conn->lock); |
0810b4f9 | 759 | } |
6b3f05d2 JK |
760 | /* |
761 | * Detach list from object now so that we don't pin inode until all | |
762 | * mark references get dropped. It would lead to strange results such | |
763 | * as delaying inode deletion or blocking unmount. | |
764 | */ | |
721fb6fb | 765 | objp = fsnotify_detach_connector_from_object(conn, &type); |
6b3f05d2 JK |
766 | spin_unlock(&conn->lock); |
767 | if (old_mark) | |
768 | fsnotify_put_mark(old_mark); | |
721fb6fb | 769 | fsnotify_drop_object(type, objp); |
0810b4f9 JK |
770 | } |
771 | ||
5444e298 EP |
772 | /* |
773 | * Nothing fancy, just initialize lists and locks and counters. | |
774 | */ | |
775 | void fsnotify_init_mark(struct fsnotify_mark *mark, | |
054c636e | 776 | struct fsnotify_group *group) |
5444e298 | 777 | { |
ba643f04 | 778 | memset(mark, 0, sizeof(*mark)); |
5444e298 | 779 | spin_lock_init(&mark->lock); |
ab97f873 | 780 | refcount_set(&mark->refcnt, 1); |
7b129323 JK |
781 | fsnotify_get_group(group); |
782 | mark->group = group; | |
5444e298 | 783 | } |
13d34ac6 | 784 | |
35e48176 JK |
785 | /* |
786 | * Destroy all marks in destroy_list, waits for SRCU period to finish before | |
787 | * actually freeing marks. | |
788 | */ | |
f09b04a0 | 789 | static void fsnotify_mark_destroy_workfn(struct work_struct *work) |
13d34ac6 JL |
790 | { |
791 | struct fsnotify_mark *mark, *next; | |
792 | struct list_head private_destroy_list; | |
793 | ||
0918f1c3 JL |
794 | spin_lock(&destroy_lock); |
795 | /* exchange the list head */ | |
796 | list_replace_init(&destroy_list, &private_destroy_list); | |
797 | spin_unlock(&destroy_lock); | |
13d34ac6 | 798 | |
0918f1c3 | 799 | synchronize_srcu(&fsnotify_mark_srcu); |
13d34ac6 | 800 | |
0918f1c3 JL |
801 | list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) { |
802 | list_del_init(&mark->g_list); | |
6b3f05d2 | 803 | fsnotify_final_mark_destroy(mark); |
13d34ac6 | 804 | } |
13d34ac6 | 805 | } |
35e48176 | 806 | |
f09b04a0 JK |
807 | /* Wait for all marks queued for destruction to be actually destroyed */ |
808 | void fsnotify_wait_marks_destroyed(void) | |
35e48176 | 809 | { |
f09b04a0 | 810 | flush_delayed_work(&reaper_work); |
35e48176 | 811 | } |