Commit | Line | Data |
---|---|---|
c82ee6d3 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
5444e298 EP |
2 | /* |
3 | * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com> | |
5444e298 EP |
4 | */ |
5 | ||
6 | /* | |
7 | * fsnotify inode mark locking/lifetime/and refcnting | |
8 | * | |
9 | * REFCNT: | |
9756b918 LS |
10 | * The group->recnt and mark->refcnt tell how many "things" in the kernel |
11 | * currently are referencing the objects. Both kind of objects typically will | |
12 | * live inside the kernel with a refcnt of 2, one for its creation and one for | |
13 | * the reference a group and a mark hold to each other. | |
14 | * If you are holding the appropriate locks, you can take a reference and the | |
15 | * object itself is guaranteed to survive until the reference is dropped. | |
5444e298 EP |
16 | * |
17 | * LOCKING: | |
9756b918 LS |
18 | * There are 3 locks involved with fsnotify inode marks and they MUST be taken |
19 | * in order as follows: | |
5444e298 | 20 | * |
9756b918 | 21 | * group->mark_mutex |
5444e298 | 22 | * mark->lock |
04662cab | 23 | * mark->connector->lock |
5444e298 | 24 | * |
9756b918 LS |
25 | * group->mark_mutex protects the marks_list anchored inside a given group and |
26 | * each mark is hooked via the g_list. It also protects the groups private | |
27 | * data (i.e group limits). | |
28 | ||
29 | * mark->lock protects the marks attributes like its masks and flags. | |
30 | * Furthermore it protects the access to a reference of the group that the mark | |
31 | * is assigned to as well as the access to a reference of the inode/vfsmount | |
32 | * that is being watched by the mark. | |
5444e298 | 33 | * |
04662cab JK |
34 | * mark->connector->lock protects the list of marks anchored inside an |
35 | * inode / vfsmount and each mark is hooked via the i_list. | |
5444e298 | 36 | * |
04662cab JK |
37 | * A list of notification marks relating to inode / mnt is contained in |
38 | * fsnotify_mark_connector. That structure is alive as long as there are any | |
6b3f05d2 JK |
39 | * marks in the list and is also protected by fsnotify_mark_srcu. A mark gets |
40 | * detached from fsnotify_mark_connector when last reference to the mark is | |
41 | * dropped. Thus having mark reference is enough to protect mark->connector | |
42 | * pointer and to make sure fsnotify_mark_connector cannot disappear. Also | |
43 | * because we remove mark from g_list before dropping mark reference associated | |
44 | * with that, any mark found through g_list is guaranteed to have | |
45 | * mark->connector set until we drop group->mark_mutex. | |
5444e298 EP |
46 | * |
47 | * LIFETIME: | |
48 | * Inode marks survive between when they are added to an inode and when their | |
c1f33073 | 49 | * refcnt==0. Marks are also protected by fsnotify_mark_srcu. |
5444e298 EP |
50 | * |
51 | * The inode mark can be cleared for a number of different reasons including: | |
52 | * - The inode is unlinked for the last time. (fsnotify_inode_remove) | |
53 | * - The inode is being evicted from cache. (fsnotify_inode_delete) | |
54 | * - The fs the inode is on is unmounted. (fsnotify_inode_delete/fsnotify_unmount_inodes) | |
55 | * - Something explicitly requests that it be removed. (fsnotify_destroy_mark) | |
56 | * - The fsnotify_group associated with the mark is going away and all such marks | |
2e37c6ca | 57 | * need to be cleaned up. (fsnotify_clear_marks_by_group) |
5444e298 | 58 | * |
5444e298 EP |
59 | * This has the very interesting property of being able to run concurrently with |
60 | * any (or all) other directions. | |
61 | */ | |
62 | ||
63 | #include <linux/fs.h> | |
64 | #include <linux/init.h> | |
65 | #include <linux/kernel.h> | |
75c1be48 | 66 | #include <linux/kthread.h> |
5444e298 EP |
67 | #include <linux/module.h> |
68 | #include <linux/mutex.h> | |
69 | #include <linux/slab.h> | |
70 | #include <linux/spinlock.h> | |
75c1be48 | 71 | #include <linux/srcu.h> |
77115225 | 72 | #include <linux/ratelimit.h> |
5444e298 | 73 | |
60063497 | 74 | #include <linux/atomic.h> |
5444e298 EP |
75 | |
76 | #include <linux/fsnotify_backend.h> | |
77 | #include "fsnotify.h" | |
78 | ||
0918f1c3 JL |
79 | #define FSNOTIFY_REAPER_DELAY (1) /* 1 jiffy */ |
80 | ||
75c1be48 | 81 | struct srcu_struct fsnotify_mark_srcu; |
9dd813c1 JK |
82 | struct kmem_cache *fsnotify_mark_connector_cachep; |
83 | ||
13d34ac6 JL |
84 | static DEFINE_SPINLOCK(destroy_lock); |
85 | static LIST_HEAD(destroy_list); | |
08991e83 | 86 | static struct fsnotify_mark_connector *connector_destroy_list; |
0918f1c3 | 87 | |
35e48176 JK |
88 | static void fsnotify_mark_destroy_workfn(struct work_struct *work); |
89 | static DECLARE_DELAYED_WORK(reaper_work, fsnotify_mark_destroy_workfn); | |
75c1be48 | 90 | |
08991e83 JK |
91 | static void fsnotify_connector_destroy_workfn(struct work_struct *work); |
92 | static DECLARE_WORK(connector_reaper_work, fsnotify_connector_destroy_workfn); | |
93 | ||
5444e298 EP |
94 | void fsnotify_get_mark(struct fsnotify_mark *mark) |
95 | { | |
ab97f873 ER |
96 | WARN_ON_ONCE(!refcount_read(&mark->refcnt)); |
97 | refcount_inc(&mark->refcnt); | |
5444e298 EP |
98 | } |
99 | ||
3ac70bfc AG |
100 | static __u32 *fsnotify_conn_mask_p(struct fsnotify_mark_connector *conn) |
101 | { | |
102 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) | |
103 | return &fsnotify_conn_inode(conn)->i_fsnotify_mask; | |
104 | else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) | |
105 | return &fsnotify_conn_mount(conn)->mnt_fsnotify_mask; | |
1e6cb723 AG |
106 | else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) |
107 | return &fsnotify_conn_sb(conn)->s_fsnotify_mask; | |
3ac70bfc AG |
108 | return NULL; |
109 | } | |
110 | ||
111 | __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn) | |
112 | { | |
113 | if (WARN_ON(!fsnotify_valid_obj_type(conn->type))) | |
114 | return 0; | |
115 | ||
116 | return *fsnotify_conn_mask_p(conn); | |
117 | } | |
118 | ||
a242677b | 119 | static void __fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) |
0809ab69 JK |
120 | { |
121 | u32 new_mask = 0; | |
122 | struct fsnotify_mark *mark; | |
123 | ||
04662cab | 124 | assert_spin_locked(&conn->lock); |
d3bc0fa8 JK |
125 | /* We can get detached connector here when inode is getting unlinked. */ |
126 | if (!fsnotify_valid_obj_type(conn->type)) | |
127 | return; | |
6b3f05d2 JK |
128 | hlist_for_each_entry(mark, &conn->list, obj_list) { |
129 | if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) | |
130 | new_mask |= mark->mask; | |
131 | } | |
3ac70bfc | 132 | *fsnotify_conn_mask_p(conn) = new_mask; |
a242677b JK |
133 | } |
134 | ||
135 | /* | |
136 | * Calculate mask of events for a list of marks. The caller must make sure | |
36f10f55 | 137 | * connector and connector->obj cannot disappear under us. Callers achieve |
6b3f05d2 JK |
138 | * this by holding a mark->lock or mark->group->mark_mutex for a mark on this |
139 | * list. | |
a242677b JK |
140 | */ |
141 | void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) | |
142 | { | |
143 | if (!conn) | |
144 | return; | |
145 | ||
04662cab | 146 | spin_lock(&conn->lock); |
a242677b | 147 | __fsnotify_recalc_mask(conn); |
04662cab | 148 | spin_unlock(&conn->lock); |
d6f7b98b | 149 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) |
36f10f55 AG |
150 | __fsnotify_update_child_dentry_flags( |
151 | fsnotify_conn_inode(conn)); | |
0809ab69 JK |
152 | } |
153 | ||
08991e83 JK |
154 | /* Free all connectors queued for freeing once SRCU period ends */ |
155 | static void fsnotify_connector_destroy_workfn(struct work_struct *work) | |
156 | { | |
157 | struct fsnotify_mark_connector *conn, *free; | |
158 | ||
159 | spin_lock(&destroy_lock); | |
160 | conn = connector_destroy_list; | |
161 | connector_destroy_list = NULL; | |
162 | spin_unlock(&destroy_lock); | |
163 | ||
164 | synchronize_srcu(&fsnotify_mark_srcu); | |
165 | while (conn) { | |
166 | free = conn; | |
167 | conn = conn->destroy_next; | |
168 | kmem_cache_free(fsnotify_mark_connector_cachep, free); | |
169 | } | |
170 | } | |
171 | ||
721fb6fb JK |
172 | static void *fsnotify_detach_connector_from_object( |
173 | struct fsnotify_mark_connector *conn, | |
174 | unsigned int *type) | |
08991e83 JK |
175 | { |
176 | struct inode *inode = NULL; | |
177 | ||
721fb6fb | 178 | *type = conn->type; |
36f10f55 AG |
179 | if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) |
180 | return NULL; | |
181 | ||
d6f7b98b | 182 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) { |
36f10f55 | 183 | inode = fsnotify_conn_inode(conn); |
08991e83 | 184 | inode->i_fsnotify_mask = 0; |
721fb6fb | 185 | atomic_long_inc(&inode->i_sb->s_fsnotify_inode_refs); |
d6f7b98b | 186 | } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) { |
36f10f55 | 187 | fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0; |
1e6cb723 AG |
188 | } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) { |
189 | fsnotify_conn_sb(conn)->s_fsnotify_mask = 0; | |
08991e83 JK |
190 | } |
191 | ||
36f10f55 AG |
192 | rcu_assign_pointer(*(conn->obj), NULL); |
193 | conn->obj = NULL; | |
194 | conn->type = FSNOTIFY_OBJ_TYPE_DETACHED; | |
195 | ||
08991e83 JK |
196 | return inode; |
197 | } | |
198 | ||
6b3f05d2 JK |
199 | static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark) |
200 | { | |
054c636e JK |
201 | struct fsnotify_group *group = mark->group; |
202 | ||
203 | if (WARN_ON_ONCE(!group)) | |
204 | return; | |
205 | group->ops->free_mark(mark); | |
206 | fsnotify_put_group(group); | |
6b3f05d2 JK |
207 | } |
208 | ||
721fb6fb JK |
209 | /* Drop object reference originally held by a connector */ |
210 | static void fsnotify_drop_object(unsigned int type, void *objp) | |
211 | { | |
212 | struct inode *inode; | |
213 | struct super_block *sb; | |
214 | ||
215 | if (!objp) | |
216 | return; | |
217 | /* Currently only inode references are passed to be dropped */ | |
218 | if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE)) | |
219 | return; | |
220 | inode = objp; | |
221 | sb = inode->i_sb; | |
222 | iput(inode); | |
223 | if (atomic_long_dec_and_test(&sb->s_fsnotify_inode_refs)) | |
224 | wake_up_var(&sb->s_fsnotify_inode_refs); | |
225 | } | |
226 | ||
6b3f05d2 | 227 | void fsnotify_put_mark(struct fsnotify_mark *mark) |
8212a609 | 228 | { |
b1da6a51 | 229 | struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector); |
721fb6fb JK |
230 | void *objp = NULL; |
231 | unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED; | |
08991e83 | 232 | bool free_conn = false; |
8212a609 | 233 | |
6b3f05d2 | 234 | /* Catch marks that were actually never attached to object */ |
b1da6a51 | 235 | if (!conn) { |
ab97f873 | 236 | if (refcount_dec_and_test(&mark->refcnt)) |
6b3f05d2 JK |
237 | fsnotify_final_mark_destroy(mark); |
238 | return; | |
239 | } | |
240 | ||
241 | /* | |
242 | * We have to be careful so that traversals of obj_list under lock can | |
243 | * safely grab mark reference. | |
244 | */ | |
b1da6a51 | 245 | if (!refcount_dec_and_lock(&mark->refcnt, &conn->lock)) |
6b3f05d2 JK |
246 | return; |
247 | ||
8212a609 JK |
248 | hlist_del_init_rcu(&mark->obj_list); |
249 | if (hlist_empty(&conn->list)) { | |
721fb6fb | 250 | objp = fsnotify_detach_connector_from_object(conn, &type); |
08991e83 JK |
251 | free_conn = true; |
252 | } else { | |
253 | __fsnotify_recalc_mask(conn); | |
8212a609 | 254 | } |
b1da6a51 | 255 | WRITE_ONCE(mark->connector, NULL); |
04662cab | 256 | spin_unlock(&conn->lock); |
8212a609 | 257 | |
721fb6fb | 258 | fsnotify_drop_object(type, objp); |
6b3f05d2 | 259 | |
08991e83 JK |
260 | if (free_conn) { |
261 | spin_lock(&destroy_lock); | |
262 | conn->destroy_next = connector_destroy_list; | |
263 | connector_destroy_list = conn; | |
264 | spin_unlock(&destroy_lock); | |
265 | queue_work(system_unbound_wq, &connector_reaper_work); | |
266 | } | |
6b3f05d2 JK |
267 | /* |
268 | * Note that we didn't update flags telling whether inode cares about | |
269 | * what's happening with children. We update these flags from | |
270 | * __fsnotify_parent() lazily when next event happens on one of our | |
271 | * children. | |
272 | */ | |
273 | spin_lock(&destroy_lock); | |
274 | list_add(&mark->g_list, &destroy_list); | |
275 | spin_unlock(&destroy_lock); | |
276 | queue_delayed_work(system_unbound_wq, &reaper_work, | |
277 | FSNOTIFY_REAPER_DELAY); | |
8212a609 | 278 | } |
b72679ee | 279 | EXPORT_SYMBOL_GPL(fsnotify_put_mark); |
8212a609 | 280 | |
24c20305 MS |
281 | /* |
282 | * Get mark reference when we found the mark via lockless traversal of object | |
283 | * list. Mark can be already removed from the list by now and on its way to be | |
284 | * destroyed once SRCU period ends. | |
285 | * | |
286 | * Also pin the group so it doesn't disappear under us. | |
287 | */ | |
288 | static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark) | |
abc77577 | 289 | { |
24c20305 MS |
290 | if (!mark) |
291 | return true; | |
abc77577 | 292 | |
ab97f873 | 293 | if (refcount_inc_not_zero(&mark->refcnt)) { |
9a31d7ad MS |
294 | spin_lock(&mark->lock); |
295 | if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) { | |
296 | /* mark is attached, group is still alive then */ | |
297 | atomic_inc(&mark->group->user_waits); | |
298 | spin_unlock(&mark->lock); | |
299 | return true; | |
300 | } | |
301 | spin_unlock(&mark->lock); | |
302 | fsnotify_put_mark(mark); | |
303 | } | |
24c20305 MS |
304 | return false; |
305 | } | |
306 | ||
307 | /* | |
308 | * Puts marks and wakes up group destruction if necessary. | |
309 | * | |
310 | * Pairs with fsnotify_get_mark_safe() | |
311 | */ | |
312 | static void fsnotify_put_mark_wake(struct fsnotify_mark *mark) | |
313 | { | |
314 | if (mark) { | |
315 | struct fsnotify_group *group = mark->group; | |
316 | ||
317 | fsnotify_put_mark(mark); | |
318 | /* | |
319 | * We abuse notification_waitq on group shutdown for waiting for | |
320 | * all marks pinned when waiting for userspace. | |
321 | */ | |
322 | if (atomic_dec_and_test(&group->user_waits) && group->shutdown) | |
323 | wake_up(&group->notification_waitq); | |
abc77577 | 324 | } |
24c20305 MS |
325 | } |
326 | ||
327 | bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info) | |
00e0afb6 | 328 | __releases(&fsnotify_mark_srcu) |
24c20305 | 329 | { |
47d9c7cc AG |
330 | int type; |
331 | ||
332 | fsnotify_foreach_obj_type(type) { | |
333 | /* This can fail if mark is being removed */ | |
00e0afb6 JI |
334 | if (!fsnotify_get_mark_safe(iter_info->marks[type])) { |
335 | __release(&fsnotify_mark_srcu); | |
47d9c7cc | 336 | goto fail; |
00e0afb6 | 337 | } |
abc77577 JK |
338 | } |
339 | ||
340 | /* | |
341 | * Now that both marks are pinned by refcount in the inode / vfsmount | |
342 | * lists, we can drop SRCU lock, and safely resume the list iteration | |
343 | * once userspace returns. | |
344 | */ | |
345 | srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx); | |
346 | ||
347 | return true; | |
47d9c7cc AG |
348 | |
349 | fail: | |
350 | for (type--; type >= 0; type--) | |
351 | fsnotify_put_mark_wake(iter_info->marks[type]); | |
352 | return false; | |
abc77577 JK |
353 | } |
354 | ||
355 | void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info) | |
00e0afb6 | 356 | __acquires(&fsnotify_mark_srcu) |
abc77577 | 357 | { |
47d9c7cc AG |
358 | int type; |
359 | ||
abc77577 | 360 | iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); |
47d9c7cc AG |
361 | fsnotify_foreach_obj_type(type) |
362 | fsnotify_put_mark_wake(iter_info->marks[type]); | |
abc77577 JK |
363 | } |
364 | ||
5444e298 | 365 | /* |
6b3f05d2 JK |
366 | * Mark mark as detached, remove it from group list. Mark still stays in object |
367 | * list until its last reference is dropped. Note that we rely on mark being | |
368 | * removed from group list before corresponding reference to it is dropped. In | |
369 | * particular we rely on mark->connector being valid while we hold | |
370 | * group->mark_mutex if we found the mark through g_list. | |
4712e722 | 371 | * |
11375145 JK |
372 | * Must be called with group->mark_mutex held. The caller must either hold |
373 | * reference to the mark or be protected by fsnotify_mark_srcu. | |
5444e298 | 374 | */ |
4712e722 | 375 | void fsnotify_detach_mark(struct fsnotify_mark *mark) |
5444e298 | 376 | { |
4712e722 | 377 | struct fsnotify_group *group = mark->group; |
5444e298 | 378 | |
11375145 JK |
379 | WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex)); |
380 | WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) && | |
ab97f873 | 381 | refcount_read(&mark->refcnt) < 1 + |
11375145 | 382 | !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)); |
d5a335b8 | 383 | |
104d06f0 | 384 | spin_lock(&mark->lock); |
700307a2 | 385 | /* something else already called this function on this mark */ |
4712e722 | 386 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { |
5444e298 | 387 | spin_unlock(&mark->lock); |
e2a29943 | 388 | return; |
5444e298 | 389 | } |
4712e722 | 390 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED; |
5444e298 | 391 | list_del_init(&mark->g_list); |
5444e298 | 392 | spin_unlock(&mark->lock); |
d5a335b8 | 393 | |
4712e722 | 394 | atomic_dec(&group->num_marks); |
11375145 JK |
395 | |
396 | /* Drop mark reference acquired in fsnotify_add_mark_locked() */ | |
397 | fsnotify_put_mark(mark); | |
4712e722 JK |
398 | } |
399 | ||
400 | /* | |
11375145 JK |
401 | * Free fsnotify mark. The mark is actually only marked as being freed. The |
402 | * freeing is actually happening only once last reference to the mark is | |
403 | * dropped from a workqueue which first waits for srcu period end. | |
35e48176 | 404 | * |
11375145 JK |
405 | * Caller must have a reference to the mark or be protected by |
406 | * fsnotify_mark_srcu. | |
4712e722 | 407 | */ |
11375145 | 408 | void fsnotify_free_mark(struct fsnotify_mark *mark) |
4712e722 JK |
409 | { |
410 | struct fsnotify_group *group = mark->group; | |
411 | ||
412 | spin_lock(&mark->lock); | |
413 | /* something else already called this function on this mark */ | |
414 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) { | |
415 | spin_unlock(&mark->lock); | |
11375145 | 416 | return; |
4712e722 JK |
417 | } |
418 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE; | |
419 | spin_unlock(&mark->lock); | |
5444e298 | 420 | |
d725e66c LT |
421 | /* |
422 | * Some groups like to know that marks are being freed. This is a | |
423 | * callback to the group function to let it know that this mark | |
424 | * is being freed. | |
425 | */ | |
426 | if (group->ops->freeing_mark) | |
427 | group->ops->freeing_mark(mark, group); | |
d5a335b8 LS |
428 | } |
429 | ||
430 | void fsnotify_destroy_mark(struct fsnotify_mark *mark, | |
431 | struct fsnotify_group *group) | |
432 | { | |
6960b0d9 | 433 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); |
4712e722 | 434 | fsnotify_detach_mark(mark); |
d5a335b8 | 435 | mutex_unlock(&group->mark_mutex); |
4712e722 | 436 | fsnotify_free_mark(mark); |
5444e298 | 437 | } |
b72679ee | 438 | EXPORT_SYMBOL_GPL(fsnotify_destroy_mark); |
5444e298 | 439 | |
8edc6e16 JK |
440 | /* |
441 | * Sorting function for lists of fsnotify marks. | |
442 | * | |
443 | * Fanotify supports different notification classes (reflected as priority of | |
444 | * notification group). Events shall be passed to notification groups in | |
445 | * decreasing priority order. To achieve this marks in notification lists for | |
446 | * inodes and vfsmounts are sorted so that priorities of corresponding groups | |
447 | * are descending. | |
448 | * | |
449 | * Furthermore correct handling of the ignore mask requires processing inode | |
450 | * and vfsmount marks of each group together. Using the group address as | |
451 | * further sort criterion provides a unique sorting order and thus we can | |
452 | * merge inode and vfsmount lists of marks in linear time and find groups | |
453 | * present in both lists. | |
454 | * | |
455 | * A return value of 1 signifies that b has priority over a. | |
456 | * A return value of 0 signifies that the two marks have to be handled together. | |
457 | * A return value of -1 signifies that a has priority over b. | |
458 | */ | |
459 | int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b) | |
460 | { | |
461 | if (a == b) | |
462 | return 0; | |
463 | if (!a) | |
464 | return 1; | |
465 | if (!b) | |
466 | return -1; | |
467 | if (a->priority < b->priority) | |
468 | return 1; | |
469 | if (a->priority > b->priority) | |
470 | return -1; | |
471 | if (a < b) | |
472 | return 1; | |
473 | return -1; | |
474 | } | |
475 | ||
9b6e5434 | 476 | static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp, |
77115225 AG |
477 | unsigned int type, |
478 | __kernel_fsid_t *fsid) | |
9dd813c1 | 479 | { |
b812a9f5 | 480 | struct inode *inode = NULL; |
9dd813c1 JK |
481 | struct fsnotify_mark_connector *conn; |
482 | ||
755b5bc6 | 483 | conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL); |
9dd813c1 JK |
484 | if (!conn) |
485 | return -ENOMEM; | |
04662cab | 486 | spin_lock_init(&conn->lock); |
9dd813c1 | 487 | INIT_HLIST_HEAD(&conn->list); |
b812a9f5 | 488 | conn->type = type; |
36f10f55 | 489 | conn->obj = connp; |
77115225 | 490 | /* Cache fsid of filesystem containing the object */ |
c285a2f0 | 491 | if (fsid) { |
77115225 | 492 | conn->fsid = *fsid; |
c285a2f0 AG |
493 | conn->flags = FSNOTIFY_CONN_FLAG_HAS_FSID; |
494 | } else { | |
77115225 | 495 | conn->fsid.val[0] = conn->fsid.val[1] = 0; |
c285a2f0 AG |
496 | conn->flags = 0; |
497 | } | |
b812a9f5 | 498 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) |
36f10f55 | 499 | inode = igrab(fsnotify_conn_inode(conn)); |
9dd813c1 | 500 | /* |
04662cab JK |
501 | * cmpxchg() provides the barrier so that readers of *connp can see |
502 | * only initialized structure | |
9dd813c1 | 503 | */ |
04662cab JK |
504 | if (cmpxchg(connp, NULL, conn)) { |
505 | /* Someone else created list structure for us */ | |
08991e83 JK |
506 | if (inode) |
507 | iput(inode); | |
755b5bc6 | 508 | kmem_cache_free(fsnotify_mark_connector_cachep, conn); |
04662cab | 509 | } |
9dd813c1 JK |
510 | |
511 | return 0; | |
512 | } | |
513 | ||
08991e83 JK |
514 | /* |
515 | * Get mark connector, make sure it is alive and return with its lock held. | |
516 | * This is for users that get connector pointer from inode or mount. Users that | |
517 | * hold reference to a mark on the list may directly lock connector->lock as | |
518 | * they are sure list cannot go away under them. | |
519 | */ | |
520 | static struct fsnotify_mark_connector *fsnotify_grab_connector( | |
9b6e5434 | 521 | fsnotify_connp_t *connp) |
08991e83 JK |
522 | { |
523 | struct fsnotify_mark_connector *conn; | |
524 | int idx; | |
525 | ||
526 | idx = srcu_read_lock(&fsnotify_mark_srcu); | |
527 | conn = srcu_dereference(*connp, &fsnotify_mark_srcu); | |
528 | if (!conn) | |
529 | goto out; | |
530 | spin_lock(&conn->lock); | |
d6f7b98b | 531 | if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) { |
08991e83 JK |
532 | spin_unlock(&conn->lock); |
533 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
534 | return NULL; | |
535 | } | |
536 | out: | |
537 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
538 | return conn; | |
539 | } | |
540 | ||
9dd813c1 JK |
541 | /* |
542 | * Add mark into proper place in given list of marks. These marks may be used | |
543 | * for the fsnotify backend to determine which event types should be delivered | |
544 | * to which group and for which inodes. These marks are ordered according to | |
545 | * priority, highest number first, and then by the group's location in memory. | |
546 | */ | |
755b5bc6 | 547 | static int fsnotify_add_mark_list(struct fsnotify_mark *mark, |
b812a9f5 | 548 | fsnotify_connp_t *connp, unsigned int type, |
77115225 | 549 | int allow_dups, __kernel_fsid_t *fsid) |
0809ab69 JK |
550 | { |
551 | struct fsnotify_mark *lmark, *last = NULL; | |
9dd813c1 | 552 | struct fsnotify_mark_connector *conn; |
0809ab69 | 553 | int cmp; |
755b5bc6 JK |
554 | int err = 0; |
555 | ||
b812a9f5 | 556 | if (WARN_ON(!fsnotify_valid_obj_type(type))) |
755b5bc6 | 557 | return -EINVAL; |
77115225 AG |
558 | |
559 | /* Backend is expected to check for zero fsid (e.g. tmpfs) */ | |
560 | if (fsid && WARN_ON_ONCE(!fsid->val[0] && !fsid->val[1])) | |
561 | return -ENODEV; | |
562 | ||
08991e83 JK |
563 | restart: |
564 | spin_lock(&mark->lock); | |
565 | conn = fsnotify_grab_connector(connp); | |
566 | if (!conn) { | |
567 | spin_unlock(&mark->lock); | |
77115225 | 568 | err = fsnotify_attach_connector_to_object(connp, type, fsid); |
9dd813c1 JK |
569 | if (err) |
570 | return err; | |
08991e83 | 571 | goto restart; |
c285a2f0 AG |
572 | } else if (fsid && !(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID)) { |
573 | conn->fsid = *fsid; | |
574 | /* Pairs with smp_rmb() in fanotify_get_fsid() */ | |
575 | smp_wmb(); | |
576 | conn->flags |= FSNOTIFY_CONN_FLAG_HAS_FSID; | |
577 | } else if (fsid && (conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID) && | |
77115225 AG |
578 | (fsid->val[0] != conn->fsid.val[0] || |
579 | fsid->val[1] != conn->fsid.val[1])) { | |
580 | /* | |
581 | * Backend is expected to check for non uniform fsid | |
582 | * (e.g. btrfs), but maybe we missed something? | |
583 | * Only allow setting conn->fsid once to non zero fsid. | |
584 | * inotify and non-fid fanotify groups do not set nor test | |
585 | * conn->fsid. | |
586 | */ | |
587 | pr_warn_ratelimited("%s: fsid mismatch on object of type %u: " | |
588 | "%x.%x != %x.%x\n", __func__, conn->type, | |
589 | fsid->val[0], fsid->val[1], | |
590 | conn->fsid.val[0], conn->fsid.val[1]); | |
591 | err = -EXDEV; | |
592 | goto out_err; | |
9dd813c1 | 593 | } |
0809ab69 JK |
594 | |
595 | /* is mark the first mark? */ | |
9dd813c1 JK |
596 | if (hlist_empty(&conn->list)) { |
597 | hlist_add_head_rcu(&mark->obj_list, &conn->list); | |
86ffe245 | 598 | goto added; |
0809ab69 JK |
599 | } |
600 | ||
601 | /* should mark be in the middle of the current list? */ | |
9dd813c1 | 602 | hlist_for_each_entry(lmark, &conn->list, obj_list) { |
0809ab69 JK |
603 | last = lmark; |
604 | ||
6b3f05d2 JK |
605 | if ((lmark->group == mark->group) && |
606 | (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) && | |
607 | !allow_dups) { | |
755b5bc6 JK |
608 | err = -EEXIST; |
609 | goto out_err; | |
610 | } | |
0809ab69 JK |
611 | |
612 | cmp = fsnotify_compare_groups(lmark->group, mark->group); | |
613 | if (cmp >= 0) { | |
614 | hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list); | |
86ffe245 | 615 | goto added; |
0809ab69 JK |
616 | } |
617 | } | |
618 | ||
619 | BUG_ON(last == NULL); | |
620 | /* mark should be the last entry. last is the current last entry */ | |
621 | hlist_add_behind_rcu(&mark->obj_list, &last->obj_list); | |
86ffe245 | 622 | added: |
11a6f8e2 JK |
623 | /* |
624 | * Since connector is attached to object using cmpxchg() we are | |
625 | * guaranteed that connector initialization is fully visible by anyone | |
626 | * seeing mark->connector set. | |
627 | */ | |
b1da6a51 | 628 | WRITE_ONCE(mark->connector, conn); |
755b5bc6 | 629 | out_err: |
04662cab | 630 | spin_unlock(&conn->lock); |
755b5bc6 JK |
631 | spin_unlock(&mark->lock); |
632 | return err; | |
0809ab69 JK |
633 | } |
634 | ||
5444e298 EP |
635 | /* |
636 | * Attach an initialized mark to a given group and fs object. | |
637 | * These marks may be used for the fsnotify backend to determine which | |
638 | * event types should be delivered to which group. | |
639 | */ | |
b812a9f5 AG |
640 | int fsnotify_add_mark_locked(struct fsnotify_mark *mark, |
641 | fsnotify_connp_t *connp, unsigned int type, | |
77115225 | 642 | int allow_dups, __kernel_fsid_t *fsid) |
5444e298 | 643 | { |
7b129323 | 644 | struct fsnotify_group *group = mark->group; |
5444e298 EP |
645 | int ret = 0; |
646 | ||
d5a335b8 | 647 | BUG_ON(!mutex_is_locked(&group->mark_mutex)); |
5444e298 | 648 | |
5444e298 EP |
649 | /* |
650 | * LOCKING ORDER!!!! | |
986ab098 | 651 | * group->mark_mutex |
104d06f0 | 652 | * mark->lock |
04662cab | 653 | * mark->connector->lock |
5444e298 | 654 | */ |
104d06f0 | 655 | spin_lock(&mark->lock); |
4712e722 | 656 | mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED; |
700307a2 | 657 | |
5444e298 EP |
658 | list_add(&mark->g_list, &group->marks_list); |
659 | atomic_inc(&group->num_marks); | |
6b3f05d2 | 660 | fsnotify_get_mark(mark); /* for g_list */ |
5444e298 EP |
661 | spin_unlock(&mark->lock); |
662 | ||
77115225 | 663 | ret = fsnotify_add_mark_list(mark, connp, type, allow_dups, fsid); |
755b5bc6 JK |
664 | if (ret) |
665 | goto err; | |
666 | ||
a242677b JK |
667 | if (mark->mask) |
668 | fsnotify_recalc_mask(mark->connector); | |
5444e298 EP |
669 | |
670 | return ret; | |
671 | err: | |
9cf90cef | 672 | spin_lock(&mark->lock); |
11375145 JK |
673 | mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE | |
674 | FSNOTIFY_MARK_FLAG_ATTACHED); | |
5444e298 | 675 | list_del_init(&mark->g_list); |
9cf90cef | 676 | spin_unlock(&mark->lock); |
5444e298 | 677 | atomic_dec(&group->num_marks); |
5444e298 | 678 | |
11375145 | 679 | fsnotify_put_mark(mark); |
5444e298 EP |
680 | return ret; |
681 | } | |
682 | ||
b812a9f5 | 683 | int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp, |
77115225 | 684 | unsigned int type, int allow_dups, __kernel_fsid_t *fsid) |
d5a335b8 LS |
685 | { |
686 | int ret; | |
7b129323 JK |
687 | struct fsnotify_group *group = mark->group; |
688 | ||
d5a335b8 | 689 | mutex_lock(&group->mark_mutex); |
77115225 | 690 | ret = fsnotify_add_mark_locked(mark, connp, type, allow_dups, fsid); |
d5a335b8 LS |
691 | mutex_unlock(&group->mark_mutex); |
692 | return ret; | |
693 | } | |
b72679ee | 694 | EXPORT_SYMBOL_GPL(fsnotify_add_mark); |
d5a335b8 | 695 | |
0809ab69 JK |
696 | /* |
697 | * Given a list of marks, find the mark associated with given group. If found | |
698 | * take a reference to that mark and return it, else return NULL. | |
699 | */ | |
9b6e5434 AG |
700 | struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp, |
701 | struct fsnotify_group *group) | |
0809ab69 | 702 | { |
08991e83 | 703 | struct fsnotify_mark_connector *conn; |
0809ab69 JK |
704 | struct fsnotify_mark *mark; |
705 | ||
08991e83 | 706 | conn = fsnotify_grab_connector(connp); |
9dd813c1 JK |
707 | if (!conn) |
708 | return NULL; | |
709 | ||
710 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
6b3f05d2 JK |
711 | if (mark->group == group && |
712 | (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { | |
0809ab69 | 713 | fsnotify_get_mark(mark); |
04662cab | 714 | spin_unlock(&conn->lock); |
0809ab69 JK |
715 | return mark; |
716 | } | |
717 | } | |
04662cab | 718 | spin_unlock(&conn->lock); |
0809ab69 JK |
719 | return NULL; |
720 | } | |
b72679ee | 721 | EXPORT_SYMBOL_GPL(fsnotify_find_mark); |
0809ab69 | 722 | |
d6f7b98b | 723 | /* Clear any marks in a group with given type mask */ |
18f2e0d3 | 724 | void fsnotify_clear_marks_by_group(struct fsnotify_group *group, |
d6f7b98b | 725 | unsigned int type_mask) |
5444e298 EP |
726 | { |
727 | struct fsnotify_mark *lmark, *mark; | |
8f2f3eb5 | 728 | LIST_HEAD(to_free); |
2e37c6ca | 729 | struct list_head *head = &to_free; |
5444e298 | 730 | |
2e37c6ca | 731 | /* Skip selection step if we want to clear all marks. */ |
d6f7b98b | 732 | if (type_mask == FSNOTIFY_OBJ_ALL_TYPES_MASK) { |
2e37c6ca JK |
733 | head = &group->marks_list; |
734 | goto clear; | |
735 | } | |
8f2f3eb5 JK |
736 | /* |
737 | * We have to be really careful here. Anytime we drop mark_mutex, e.g. | |
738 | * fsnotify_clear_marks_by_inode() can come and free marks. Even in our | |
739 | * to_free list so we have to use mark_mutex even when accessing that | |
740 | * list. And freeing mark requires us to drop mark_mutex. So we can | |
741 | * reliably free only the first mark in the list. That's why we first | |
742 | * move marks to free to to_free list in one go and then free marks in | |
743 | * to_free list one by one. | |
744 | */ | |
6960b0d9 | 745 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); |
5444e298 | 746 | list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) { |
d6f7b98b | 747 | if ((1U << mark->connector->type) & type_mask) |
8f2f3eb5 | 748 | list_move(&mark->g_list, &to_free); |
5444e298 | 749 | } |
986ab098 | 750 | mutex_unlock(&group->mark_mutex); |
8f2f3eb5 | 751 | |
2e37c6ca | 752 | clear: |
8f2f3eb5 JK |
753 | while (1) { |
754 | mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING); | |
2e37c6ca | 755 | if (list_empty(head)) { |
8f2f3eb5 JK |
756 | mutex_unlock(&group->mark_mutex); |
757 | break; | |
758 | } | |
2e37c6ca | 759 | mark = list_first_entry(head, struct fsnotify_mark, g_list); |
8f2f3eb5 | 760 | fsnotify_get_mark(mark); |
4712e722 | 761 | fsnotify_detach_mark(mark); |
8f2f3eb5 | 762 | mutex_unlock(&group->mark_mutex); |
4712e722 | 763 | fsnotify_free_mark(mark); |
8f2f3eb5 JK |
764 | fsnotify_put_mark(mark); |
765 | } | |
5444e298 EP |
766 | } |
767 | ||
9b6e5434 AG |
768 | /* Destroy all marks attached to an object via connector */ |
769 | void fsnotify_destroy_marks(fsnotify_connp_t *connp) | |
0810b4f9 | 770 | { |
08991e83 | 771 | struct fsnotify_mark_connector *conn; |
6b3f05d2 | 772 | struct fsnotify_mark *mark, *old_mark = NULL; |
721fb6fb JK |
773 | void *objp; |
774 | unsigned int type; | |
0810b4f9 | 775 | |
6b3f05d2 JK |
776 | conn = fsnotify_grab_connector(connp); |
777 | if (!conn) | |
778 | return; | |
779 | /* | |
780 | * We have to be careful since we can race with e.g. | |
781 | * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the | |
782 | * list can get modified. However we are holding mark reference and | |
783 | * thus our mark cannot be removed from obj_list so we can continue | |
784 | * iteration after regaining conn->lock. | |
785 | */ | |
786 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
0810b4f9 | 787 | fsnotify_get_mark(mark); |
04662cab | 788 | spin_unlock(&conn->lock); |
6b3f05d2 JK |
789 | if (old_mark) |
790 | fsnotify_put_mark(old_mark); | |
791 | old_mark = mark; | |
0810b4f9 | 792 | fsnotify_destroy_mark(mark, mark->group); |
6b3f05d2 | 793 | spin_lock(&conn->lock); |
0810b4f9 | 794 | } |
6b3f05d2 JK |
795 | /* |
796 | * Detach list from object now so that we don't pin inode until all | |
797 | * mark references get dropped. It would lead to strange results such | |
798 | * as delaying inode deletion or blocking unmount. | |
799 | */ | |
721fb6fb | 800 | objp = fsnotify_detach_connector_from_object(conn, &type); |
6b3f05d2 JK |
801 | spin_unlock(&conn->lock); |
802 | if (old_mark) | |
803 | fsnotify_put_mark(old_mark); | |
721fb6fb | 804 | fsnotify_drop_object(type, objp); |
0810b4f9 JK |
805 | } |
806 | ||
5444e298 EP |
807 | /* |
808 | * Nothing fancy, just initialize lists and locks and counters. | |
809 | */ | |
810 | void fsnotify_init_mark(struct fsnotify_mark *mark, | |
054c636e | 811 | struct fsnotify_group *group) |
5444e298 | 812 | { |
ba643f04 | 813 | memset(mark, 0, sizeof(*mark)); |
5444e298 | 814 | spin_lock_init(&mark->lock); |
ab97f873 | 815 | refcount_set(&mark->refcnt, 1); |
7b129323 JK |
816 | fsnotify_get_group(group); |
817 | mark->group = group; | |
b1da6a51 | 818 | WRITE_ONCE(mark->connector, NULL); |
5444e298 | 819 | } |
b72679ee | 820 | EXPORT_SYMBOL_GPL(fsnotify_init_mark); |
13d34ac6 | 821 | |
35e48176 JK |
822 | /* |
823 | * Destroy all marks in destroy_list, waits for SRCU period to finish before | |
824 | * actually freeing marks. | |
825 | */ | |
f09b04a0 | 826 | static void fsnotify_mark_destroy_workfn(struct work_struct *work) |
13d34ac6 JL |
827 | { |
828 | struct fsnotify_mark *mark, *next; | |
829 | struct list_head private_destroy_list; | |
830 | ||
0918f1c3 JL |
831 | spin_lock(&destroy_lock); |
832 | /* exchange the list head */ | |
833 | list_replace_init(&destroy_list, &private_destroy_list); | |
834 | spin_unlock(&destroy_lock); | |
13d34ac6 | 835 | |
0918f1c3 | 836 | synchronize_srcu(&fsnotify_mark_srcu); |
13d34ac6 | 837 | |
0918f1c3 JL |
838 | list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) { |
839 | list_del_init(&mark->g_list); | |
6b3f05d2 | 840 | fsnotify_final_mark_destroy(mark); |
13d34ac6 | 841 | } |
13d34ac6 | 842 | } |
35e48176 | 843 | |
f09b04a0 JK |
844 | /* Wait for all marks queued for destruction to be actually destroyed */ |
845 | void fsnotify_wait_marks_destroyed(void) | |
35e48176 | 846 | { |
f09b04a0 | 847 | flush_delayed_work(&reaper_work); |
35e48176 | 848 | } |
b72679ee | 849 | EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed); |