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 | ||
c3638b5b AG |
119 | static void fsnotify_get_inode_ref(struct inode *inode) |
120 | { | |
121 | ihold(inode); | |
122 | atomic_long_inc(&inode->i_sb->s_fsnotify_connectors); | |
123 | } | |
124 | ||
125 | /* | |
126 | * Grab or drop inode reference for the connector if needed. | |
127 | * | |
128 | * When it's time to drop the reference, we only clear the HAS_IREF flag and | |
129 | * return the inode object. fsnotify_drop_object() will be resonsible for doing | |
130 | * iput() outside of spinlocks. This happens when last mark that wanted iref is | |
131 | * detached. | |
132 | */ | |
133 | static struct inode *fsnotify_update_iref(struct fsnotify_mark_connector *conn, | |
134 | bool want_iref) | |
135 | { | |
136 | bool has_iref = conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF; | |
137 | struct inode *inode = NULL; | |
138 | ||
139 | if (conn->type != FSNOTIFY_OBJ_TYPE_INODE || | |
140 | want_iref == has_iref) | |
141 | return NULL; | |
142 | ||
143 | if (want_iref) { | |
144 | /* Pin inode if any mark wants inode refcount held */ | |
145 | fsnotify_get_inode_ref(fsnotify_conn_inode(conn)); | |
146 | conn->flags |= FSNOTIFY_CONN_FLAG_HAS_IREF; | |
147 | } else { | |
148 | /* Unpin inode after detach of last mark that wanted iref */ | |
149 | inode = fsnotify_conn_inode(conn); | |
150 | conn->flags &= ~FSNOTIFY_CONN_FLAG_HAS_IREF; | |
151 | } | |
152 | ||
153 | return inode; | |
154 | } | |
155 | ||
156 | static void *__fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) | |
0809ab69 JK |
157 | { |
158 | u32 new_mask = 0; | |
c3638b5b | 159 | bool want_iref = false; |
0809ab69 JK |
160 | struct fsnotify_mark *mark; |
161 | ||
04662cab | 162 | assert_spin_locked(&conn->lock); |
d3bc0fa8 JK |
163 | /* We can get detached connector here when inode is getting unlinked. */ |
164 | if (!fsnotify_valid_obj_type(conn->type)) | |
c3638b5b | 165 | return NULL; |
6b3f05d2 | 166 | hlist_for_each_entry(mark, &conn->list, obj_list) { |
c3638b5b AG |
167 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) |
168 | continue; | |
169 | new_mask |= fsnotify_calc_mask(mark); | |
170 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE && | |
171 | !(mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF)) | |
172 | want_iref = true; | |
6b3f05d2 | 173 | } |
3ac70bfc | 174 | *fsnotify_conn_mask_p(conn) = new_mask; |
c3638b5b AG |
175 | |
176 | return fsnotify_update_iref(conn, want_iref); | |
a242677b JK |
177 | } |
178 | ||
179 | /* | |
180 | * Calculate mask of events for a list of marks. The caller must make sure | |
36f10f55 | 181 | * connector and connector->obj cannot disappear under us. Callers achieve |
6b3f05d2 JK |
182 | * this by holding a mark->lock or mark->group->mark_mutex for a mark on this |
183 | * list. | |
a242677b JK |
184 | */ |
185 | void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn) | |
186 | { | |
187 | if (!conn) | |
188 | return; | |
189 | ||
04662cab | 190 | spin_lock(&conn->lock); |
a242677b | 191 | __fsnotify_recalc_mask(conn); |
04662cab | 192 | spin_unlock(&conn->lock); |
d6f7b98b | 193 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) |
36f10f55 AG |
194 | __fsnotify_update_child_dentry_flags( |
195 | fsnotify_conn_inode(conn)); | |
0809ab69 JK |
196 | } |
197 | ||
08991e83 JK |
198 | /* Free all connectors queued for freeing once SRCU period ends */ |
199 | static void fsnotify_connector_destroy_workfn(struct work_struct *work) | |
200 | { | |
201 | struct fsnotify_mark_connector *conn, *free; | |
202 | ||
203 | spin_lock(&destroy_lock); | |
204 | conn = connector_destroy_list; | |
205 | connector_destroy_list = NULL; | |
206 | spin_unlock(&destroy_lock); | |
207 | ||
208 | synchronize_srcu(&fsnotify_mark_srcu); | |
209 | while (conn) { | |
210 | free = conn; | |
211 | conn = conn->destroy_next; | |
212 | kmem_cache_free(fsnotify_mark_connector_cachep, free); | |
213 | } | |
214 | } | |
215 | ||
11fa333b AG |
216 | static void fsnotify_put_inode_ref(struct inode *inode) |
217 | { | |
218 | struct super_block *sb = inode->i_sb; | |
219 | ||
220 | iput(inode); | |
ec44610f AG |
221 | if (atomic_long_dec_and_test(&sb->s_fsnotify_connectors)) |
222 | wake_up_var(&sb->s_fsnotify_connectors); | |
223 | } | |
224 | ||
225 | static void fsnotify_get_sb_connectors(struct fsnotify_mark_connector *conn) | |
226 | { | |
227 | struct super_block *sb = fsnotify_connector_sb(conn); | |
228 | ||
229 | if (sb) | |
230 | atomic_long_inc(&sb->s_fsnotify_connectors); | |
231 | } | |
232 | ||
233 | static void fsnotify_put_sb_connectors(struct fsnotify_mark_connector *conn) | |
234 | { | |
235 | struct super_block *sb = fsnotify_connector_sb(conn); | |
236 | ||
237 | if (sb && atomic_long_dec_and_test(&sb->s_fsnotify_connectors)) | |
238 | wake_up_var(&sb->s_fsnotify_connectors); | |
11fa333b AG |
239 | } |
240 | ||
721fb6fb JK |
241 | static void *fsnotify_detach_connector_from_object( |
242 | struct fsnotify_mark_connector *conn, | |
243 | unsigned int *type) | |
08991e83 JK |
244 | { |
245 | struct inode *inode = NULL; | |
246 | ||
721fb6fb | 247 | *type = conn->type; |
36f10f55 AG |
248 | if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) |
249 | return NULL; | |
250 | ||
d6f7b98b | 251 | if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) { |
36f10f55 | 252 | inode = fsnotify_conn_inode(conn); |
08991e83 | 253 | inode->i_fsnotify_mask = 0; |
c3638b5b AG |
254 | |
255 | /* Unpin inode when detaching from connector */ | |
256 | if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF)) | |
257 | inode = NULL; | |
d6f7b98b | 258 | } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) { |
36f10f55 | 259 | fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0; |
1e6cb723 AG |
260 | } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) { |
261 | fsnotify_conn_sb(conn)->s_fsnotify_mask = 0; | |
08991e83 JK |
262 | } |
263 | ||
ec44610f | 264 | fsnotify_put_sb_connectors(conn); |
36f10f55 AG |
265 | rcu_assign_pointer(*(conn->obj), NULL); |
266 | conn->obj = NULL; | |
267 | conn->type = FSNOTIFY_OBJ_TYPE_DETACHED; | |
268 | ||
08991e83 JK |
269 | return inode; |
270 | } | |
271 | ||
6b3f05d2 JK |
272 | static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark) |
273 | { | |
054c636e JK |
274 | struct fsnotify_group *group = mark->group; |
275 | ||
276 | if (WARN_ON_ONCE(!group)) | |
277 | return; | |
278 | group->ops->free_mark(mark); | |
279 | fsnotify_put_group(group); | |
6b3f05d2 JK |
280 | } |
281 | ||
721fb6fb JK |
282 | /* Drop object reference originally held by a connector */ |
283 | static void fsnotify_drop_object(unsigned int type, void *objp) | |
284 | { | |
721fb6fb JK |
285 | if (!objp) |
286 | return; | |
287 | /* Currently only inode references are passed to be dropped */ | |
288 | if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE)) | |
289 | return; | |
11fa333b | 290 | fsnotify_put_inode_ref(objp); |
721fb6fb JK |
291 | } |
292 | ||
6b3f05d2 | 293 | void fsnotify_put_mark(struct fsnotify_mark *mark) |
8212a609 | 294 | { |
b1da6a51 | 295 | struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector); |
721fb6fb JK |
296 | void *objp = NULL; |
297 | unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED; | |
08991e83 | 298 | bool free_conn = false; |
8212a609 | 299 | |
6b3f05d2 | 300 | /* Catch marks that were actually never attached to object */ |
b1da6a51 | 301 | if (!conn) { |
ab97f873 | 302 | if (refcount_dec_and_test(&mark->refcnt)) |
6b3f05d2 JK |
303 | fsnotify_final_mark_destroy(mark); |
304 | return; | |
305 | } | |
306 | ||
307 | /* | |
308 | * We have to be careful so that traversals of obj_list under lock can | |
309 | * safely grab mark reference. | |
310 | */ | |
b1da6a51 | 311 | if (!refcount_dec_and_lock(&mark->refcnt, &conn->lock)) |
6b3f05d2 JK |
312 | return; |
313 | ||
8212a609 JK |
314 | hlist_del_init_rcu(&mark->obj_list); |
315 | if (hlist_empty(&conn->list)) { | |
721fb6fb | 316 | objp = fsnotify_detach_connector_from_object(conn, &type); |
08991e83 JK |
317 | free_conn = true; |
318 | } else { | |
c3638b5b AG |
319 | objp = __fsnotify_recalc_mask(conn); |
320 | type = conn->type; | |
8212a609 | 321 | } |
b1da6a51 | 322 | WRITE_ONCE(mark->connector, NULL); |
04662cab | 323 | spin_unlock(&conn->lock); |
8212a609 | 324 | |
721fb6fb | 325 | fsnotify_drop_object(type, objp); |
6b3f05d2 | 326 | |
08991e83 JK |
327 | if (free_conn) { |
328 | spin_lock(&destroy_lock); | |
329 | conn->destroy_next = connector_destroy_list; | |
330 | connector_destroy_list = conn; | |
331 | spin_unlock(&destroy_lock); | |
332 | queue_work(system_unbound_wq, &connector_reaper_work); | |
333 | } | |
6b3f05d2 JK |
334 | /* |
335 | * Note that we didn't update flags telling whether inode cares about | |
336 | * what's happening with children. We update these flags from | |
337 | * __fsnotify_parent() lazily when next event happens on one of our | |
338 | * children. | |
339 | */ | |
340 | spin_lock(&destroy_lock); | |
341 | list_add(&mark->g_list, &destroy_list); | |
342 | spin_unlock(&destroy_lock); | |
343 | queue_delayed_work(system_unbound_wq, &reaper_work, | |
344 | FSNOTIFY_REAPER_DELAY); | |
8212a609 | 345 | } |
b72679ee | 346 | EXPORT_SYMBOL_GPL(fsnotify_put_mark); |
8212a609 | 347 | |
24c20305 MS |
348 | /* |
349 | * Get mark reference when we found the mark via lockless traversal of object | |
350 | * list. Mark can be already removed from the list by now and on its way to be | |
351 | * destroyed once SRCU period ends. | |
352 | * | |
353 | * Also pin the group so it doesn't disappear under us. | |
354 | */ | |
355 | static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark) | |
abc77577 | 356 | { |
24c20305 MS |
357 | if (!mark) |
358 | return true; | |
abc77577 | 359 | |
ab97f873 | 360 | if (refcount_inc_not_zero(&mark->refcnt)) { |
9a31d7ad MS |
361 | spin_lock(&mark->lock); |
362 | if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) { | |
363 | /* mark is attached, group is still alive then */ | |
364 | atomic_inc(&mark->group->user_waits); | |
365 | spin_unlock(&mark->lock); | |
366 | return true; | |
367 | } | |
368 | spin_unlock(&mark->lock); | |
369 | fsnotify_put_mark(mark); | |
370 | } | |
24c20305 MS |
371 | return false; |
372 | } | |
373 | ||
374 | /* | |
375 | * Puts marks and wakes up group destruction if necessary. | |
376 | * | |
377 | * Pairs with fsnotify_get_mark_safe() | |
378 | */ | |
379 | static void fsnotify_put_mark_wake(struct fsnotify_mark *mark) | |
380 | { | |
381 | if (mark) { | |
382 | struct fsnotify_group *group = mark->group; | |
383 | ||
384 | fsnotify_put_mark(mark); | |
385 | /* | |
386 | * We abuse notification_waitq on group shutdown for waiting for | |
387 | * all marks pinned when waiting for userspace. | |
388 | */ | |
389 | if (atomic_dec_and_test(&group->user_waits) && group->shutdown) | |
390 | wake_up(&group->notification_waitq); | |
abc77577 | 391 | } |
24c20305 MS |
392 | } |
393 | ||
394 | bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info) | |
00e0afb6 | 395 | __releases(&fsnotify_mark_srcu) |
24c20305 | 396 | { |
47d9c7cc AG |
397 | int type; |
398 | ||
1c9007d6 | 399 | fsnotify_foreach_iter_type(type) { |
47d9c7cc | 400 | /* This can fail if mark is being removed */ |
00e0afb6 JI |
401 | if (!fsnotify_get_mark_safe(iter_info->marks[type])) { |
402 | __release(&fsnotify_mark_srcu); | |
47d9c7cc | 403 | goto fail; |
00e0afb6 | 404 | } |
abc77577 JK |
405 | } |
406 | ||
407 | /* | |
408 | * Now that both marks are pinned by refcount in the inode / vfsmount | |
409 | * lists, we can drop SRCU lock, and safely resume the list iteration | |
410 | * once userspace returns. | |
411 | */ | |
412 | srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx); | |
413 | ||
414 | return true; | |
47d9c7cc AG |
415 | |
416 | fail: | |
417 | for (type--; type >= 0; type--) | |
418 | fsnotify_put_mark_wake(iter_info->marks[type]); | |
419 | return false; | |
abc77577 JK |
420 | } |
421 | ||
422 | void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info) | |
00e0afb6 | 423 | __acquires(&fsnotify_mark_srcu) |
abc77577 | 424 | { |
47d9c7cc AG |
425 | int type; |
426 | ||
abc77577 | 427 | iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); |
1c9007d6 | 428 | fsnotify_foreach_iter_type(type) |
47d9c7cc | 429 | fsnotify_put_mark_wake(iter_info->marks[type]); |
abc77577 JK |
430 | } |
431 | ||
5444e298 | 432 | /* |
6b3f05d2 JK |
433 | * Mark mark as detached, remove it from group list. Mark still stays in object |
434 | * list until its last reference is dropped. Note that we rely on mark being | |
435 | * removed from group list before corresponding reference to it is dropped. In | |
436 | * particular we rely on mark->connector being valid while we hold | |
437 | * group->mark_mutex if we found the mark through g_list. | |
4712e722 | 438 | * |
11375145 JK |
439 | * Must be called with group->mark_mutex held. The caller must either hold |
440 | * reference to the mark or be protected by fsnotify_mark_srcu. | |
5444e298 | 441 | */ |
4712e722 | 442 | void fsnotify_detach_mark(struct fsnotify_mark *mark) |
5444e298 | 443 | { |
43b245a7 | 444 | fsnotify_group_assert_locked(mark->group); |
11375145 | 445 | WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) && |
ab97f873 | 446 | refcount_read(&mark->refcnt) < 1 + |
11375145 | 447 | !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)); |
d5a335b8 | 448 | |
104d06f0 | 449 | spin_lock(&mark->lock); |
700307a2 | 450 | /* something else already called this function on this mark */ |
4712e722 | 451 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { |
5444e298 | 452 | spin_unlock(&mark->lock); |
e2a29943 | 453 | return; |
5444e298 | 454 | } |
4712e722 | 455 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED; |
5444e298 | 456 | list_del_init(&mark->g_list); |
5444e298 | 457 | spin_unlock(&mark->lock); |
d5a335b8 | 458 | |
11375145 JK |
459 | /* Drop mark reference acquired in fsnotify_add_mark_locked() */ |
460 | fsnotify_put_mark(mark); | |
4712e722 JK |
461 | } |
462 | ||
463 | /* | |
11375145 JK |
464 | * Free fsnotify mark. The mark is actually only marked as being freed. The |
465 | * freeing is actually happening only once last reference to the mark is | |
466 | * dropped from a workqueue which first waits for srcu period end. | |
35e48176 | 467 | * |
11375145 JK |
468 | * Caller must have a reference to the mark or be protected by |
469 | * fsnotify_mark_srcu. | |
4712e722 | 470 | */ |
11375145 | 471 | void fsnotify_free_mark(struct fsnotify_mark *mark) |
4712e722 JK |
472 | { |
473 | struct fsnotify_group *group = mark->group; | |
474 | ||
475 | spin_lock(&mark->lock); | |
476 | /* something else already called this function on this mark */ | |
477 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) { | |
478 | spin_unlock(&mark->lock); | |
11375145 | 479 | return; |
4712e722 JK |
480 | } |
481 | mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE; | |
482 | spin_unlock(&mark->lock); | |
5444e298 | 483 | |
d725e66c LT |
484 | /* |
485 | * Some groups like to know that marks are being freed. This is a | |
486 | * callback to the group function to let it know that this mark | |
487 | * is being freed. | |
488 | */ | |
489 | if (group->ops->freeing_mark) | |
490 | group->ops->freeing_mark(mark, group); | |
d5a335b8 LS |
491 | } |
492 | ||
493 | void fsnotify_destroy_mark(struct fsnotify_mark *mark, | |
494 | struct fsnotify_group *group) | |
495 | { | |
43b245a7 | 496 | fsnotify_group_lock(group); |
4712e722 | 497 | fsnotify_detach_mark(mark); |
43b245a7 | 498 | fsnotify_group_unlock(group); |
4712e722 | 499 | fsnotify_free_mark(mark); |
5444e298 | 500 | } |
b72679ee | 501 | EXPORT_SYMBOL_GPL(fsnotify_destroy_mark); |
5444e298 | 502 | |
8edc6e16 JK |
503 | /* |
504 | * Sorting function for lists of fsnotify marks. | |
505 | * | |
506 | * Fanotify supports different notification classes (reflected as priority of | |
507 | * notification group). Events shall be passed to notification groups in | |
508 | * decreasing priority order. To achieve this marks in notification lists for | |
509 | * inodes and vfsmounts are sorted so that priorities of corresponding groups | |
510 | * are descending. | |
511 | * | |
512 | * Furthermore correct handling of the ignore mask requires processing inode | |
513 | * and vfsmount marks of each group together. Using the group address as | |
514 | * further sort criterion provides a unique sorting order and thus we can | |
515 | * merge inode and vfsmount lists of marks in linear time and find groups | |
516 | * present in both lists. | |
517 | * | |
518 | * A return value of 1 signifies that b has priority over a. | |
519 | * A return value of 0 signifies that the two marks have to be handled together. | |
520 | * A return value of -1 signifies that a has priority over b. | |
521 | */ | |
522 | int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b) | |
523 | { | |
524 | if (a == b) | |
525 | return 0; | |
526 | if (!a) | |
527 | return 1; | |
528 | if (!b) | |
529 | return -1; | |
530 | if (a->priority < b->priority) | |
531 | return 1; | |
532 | if (a->priority > b->priority) | |
533 | return -1; | |
534 | if (a < b) | |
535 | return 1; | |
536 | return -1; | |
537 | } | |
538 | ||
9b6e5434 | 539 | static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp, |
ad69cd99 | 540 | unsigned int obj_type, |
77115225 | 541 | __kernel_fsid_t *fsid) |
9dd813c1 JK |
542 | { |
543 | struct fsnotify_mark_connector *conn; | |
544 | ||
755b5bc6 | 545 | conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL); |
9dd813c1 JK |
546 | if (!conn) |
547 | return -ENOMEM; | |
04662cab | 548 | spin_lock_init(&conn->lock); |
9dd813c1 | 549 | INIT_HLIST_HEAD(&conn->list); |
c3638b5b | 550 | conn->flags = 0; |
ad69cd99 | 551 | conn->type = obj_type; |
36f10f55 | 552 | conn->obj = connp; |
77115225 | 553 | /* Cache fsid of filesystem containing the object */ |
c285a2f0 | 554 | if (fsid) { |
77115225 | 555 | conn->fsid = *fsid; |
c285a2f0 AG |
556 | conn->flags = FSNOTIFY_CONN_FLAG_HAS_FSID; |
557 | } else { | |
77115225 | 558 | conn->fsid.val[0] = conn->fsid.val[1] = 0; |
c285a2f0 AG |
559 | conn->flags = 0; |
560 | } | |
ec44610f | 561 | fsnotify_get_sb_connectors(conn); |
09ddbe69 | 562 | |
9dd813c1 | 563 | /* |
04662cab JK |
564 | * cmpxchg() provides the barrier so that readers of *connp can see |
565 | * only initialized structure | |
9dd813c1 | 566 | */ |
04662cab JK |
567 | if (cmpxchg(connp, NULL, conn)) { |
568 | /* Someone else created list structure for us */ | |
4396a731 | 569 | fsnotify_put_sb_connectors(conn); |
755b5bc6 | 570 | kmem_cache_free(fsnotify_mark_connector_cachep, conn); |
04662cab | 571 | } |
9dd813c1 JK |
572 | |
573 | return 0; | |
574 | } | |
575 | ||
08991e83 JK |
576 | /* |
577 | * Get mark connector, make sure it is alive and return with its lock held. | |
578 | * This is for users that get connector pointer from inode or mount. Users that | |
579 | * hold reference to a mark on the list may directly lock connector->lock as | |
580 | * they are sure list cannot go away under them. | |
581 | */ | |
582 | static struct fsnotify_mark_connector *fsnotify_grab_connector( | |
9b6e5434 | 583 | fsnotify_connp_t *connp) |
08991e83 JK |
584 | { |
585 | struct fsnotify_mark_connector *conn; | |
586 | int idx; | |
587 | ||
588 | idx = srcu_read_lock(&fsnotify_mark_srcu); | |
589 | conn = srcu_dereference(*connp, &fsnotify_mark_srcu); | |
590 | if (!conn) | |
591 | goto out; | |
592 | spin_lock(&conn->lock); | |
d6f7b98b | 593 | if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) { |
08991e83 JK |
594 | spin_unlock(&conn->lock); |
595 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
596 | return NULL; | |
597 | } | |
598 | out: | |
599 | srcu_read_unlock(&fsnotify_mark_srcu, idx); | |
600 | return conn; | |
601 | } | |
602 | ||
9dd813c1 JK |
603 | /* |
604 | * Add mark into proper place in given list of marks. These marks may be used | |
605 | * for the fsnotify backend to determine which event types should be delivered | |
606 | * to which group and for which inodes. These marks are ordered according to | |
607 | * priority, highest number first, and then by the group's location in memory. | |
608 | */ | |
755b5bc6 | 609 | static int fsnotify_add_mark_list(struct fsnotify_mark *mark, |
ad69cd99 AG |
610 | fsnotify_connp_t *connp, |
611 | unsigned int obj_type, | |
f3010343 | 612 | int add_flags, __kernel_fsid_t *fsid) |
0809ab69 JK |
613 | { |
614 | struct fsnotify_mark *lmark, *last = NULL; | |
9dd813c1 | 615 | struct fsnotify_mark_connector *conn; |
0809ab69 | 616 | int cmp; |
755b5bc6 JK |
617 | int err = 0; |
618 | ||
ad69cd99 | 619 | if (WARN_ON(!fsnotify_valid_obj_type(obj_type))) |
755b5bc6 | 620 | return -EINVAL; |
77115225 AG |
621 | |
622 | /* Backend is expected to check for zero fsid (e.g. tmpfs) */ | |
623 | if (fsid && WARN_ON_ONCE(!fsid->val[0] && !fsid->val[1])) | |
624 | return -ENODEV; | |
625 | ||
08991e83 JK |
626 | restart: |
627 | spin_lock(&mark->lock); | |
628 | conn = fsnotify_grab_connector(connp); | |
629 | if (!conn) { | |
630 | spin_unlock(&mark->lock); | |
ad69cd99 AG |
631 | err = fsnotify_attach_connector_to_object(connp, obj_type, |
632 | fsid); | |
9dd813c1 JK |
633 | if (err) |
634 | return err; | |
08991e83 | 635 | goto restart; |
c285a2f0 AG |
636 | } else if (fsid && !(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID)) { |
637 | conn->fsid = *fsid; | |
638 | /* Pairs with smp_rmb() in fanotify_get_fsid() */ | |
639 | smp_wmb(); | |
640 | conn->flags |= FSNOTIFY_CONN_FLAG_HAS_FSID; | |
641 | } else if (fsid && (conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID) && | |
77115225 AG |
642 | (fsid->val[0] != conn->fsid.val[0] || |
643 | fsid->val[1] != conn->fsid.val[1])) { | |
644 | /* | |
645 | * Backend is expected to check for non uniform fsid | |
646 | * (e.g. btrfs), but maybe we missed something? | |
647 | * Only allow setting conn->fsid once to non zero fsid. | |
648 | * inotify and non-fid fanotify groups do not set nor test | |
649 | * conn->fsid. | |
650 | */ | |
651 | pr_warn_ratelimited("%s: fsid mismatch on object of type %u: " | |
652 | "%x.%x != %x.%x\n", __func__, conn->type, | |
653 | fsid->val[0], fsid->val[1], | |
654 | conn->fsid.val[0], conn->fsid.val[1]); | |
655 | err = -EXDEV; | |
656 | goto out_err; | |
9dd813c1 | 657 | } |
0809ab69 JK |
658 | |
659 | /* is mark the first mark? */ | |
9dd813c1 JK |
660 | if (hlist_empty(&conn->list)) { |
661 | hlist_add_head_rcu(&mark->obj_list, &conn->list); | |
86ffe245 | 662 | goto added; |
0809ab69 JK |
663 | } |
664 | ||
665 | /* should mark be in the middle of the current list? */ | |
9dd813c1 | 666 | hlist_for_each_entry(lmark, &conn->list, obj_list) { |
0809ab69 JK |
667 | last = lmark; |
668 | ||
6b3f05d2 JK |
669 | if ((lmark->group == mark->group) && |
670 | (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) && | |
f3010343 | 671 | !(mark->group->flags & FSNOTIFY_GROUP_DUPS)) { |
755b5bc6 JK |
672 | err = -EEXIST; |
673 | goto out_err; | |
674 | } | |
0809ab69 JK |
675 | |
676 | cmp = fsnotify_compare_groups(lmark->group, mark->group); | |
677 | if (cmp >= 0) { | |
678 | hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list); | |
86ffe245 | 679 | goto added; |
0809ab69 JK |
680 | } |
681 | } | |
682 | ||
683 | BUG_ON(last == NULL); | |
684 | /* mark should be the last entry. last is the current last entry */ | |
685 | hlist_add_behind_rcu(&mark->obj_list, &last->obj_list); | |
86ffe245 | 686 | added: |
11a6f8e2 JK |
687 | /* |
688 | * Since connector is attached to object using cmpxchg() we are | |
689 | * guaranteed that connector initialization is fully visible by anyone | |
690 | * seeing mark->connector set. | |
691 | */ | |
b1da6a51 | 692 | WRITE_ONCE(mark->connector, conn); |
755b5bc6 | 693 | out_err: |
04662cab | 694 | spin_unlock(&conn->lock); |
755b5bc6 JK |
695 | spin_unlock(&mark->lock); |
696 | return err; | |
0809ab69 JK |
697 | } |
698 | ||
5444e298 EP |
699 | /* |
700 | * Attach an initialized mark to a given group and fs object. | |
701 | * These marks may be used for the fsnotify backend to determine which | |
702 | * event types should be delivered to which group. | |
703 | */ | |
b812a9f5 | 704 | int fsnotify_add_mark_locked(struct fsnotify_mark *mark, |
ad69cd99 | 705 | fsnotify_connp_t *connp, unsigned int obj_type, |
f3010343 | 706 | int add_flags, __kernel_fsid_t *fsid) |
5444e298 | 707 | { |
7b129323 | 708 | struct fsnotify_group *group = mark->group; |
5444e298 EP |
709 | int ret = 0; |
710 | ||
43b245a7 | 711 | fsnotify_group_assert_locked(group); |
5444e298 | 712 | |
5444e298 EP |
713 | /* |
714 | * LOCKING ORDER!!!! | |
986ab098 | 715 | * group->mark_mutex |
104d06f0 | 716 | * mark->lock |
04662cab | 717 | * mark->connector->lock |
5444e298 | 718 | */ |
104d06f0 | 719 | spin_lock(&mark->lock); |
4712e722 | 720 | mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED; |
700307a2 | 721 | |
5444e298 | 722 | list_add(&mark->g_list, &group->marks_list); |
6b3f05d2 | 723 | fsnotify_get_mark(mark); /* for g_list */ |
5444e298 EP |
724 | spin_unlock(&mark->lock); |
725 | ||
f3010343 | 726 | ret = fsnotify_add_mark_list(mark, connp, obj_type, add_flags, fsid); |
755b5bc6 JK |
727 | if (ret) |
728 | goto err; | |
729 | ||
c3638b5b | 730 | fsnotify_recalc_mask(mark->connector); |
5444e298 EP |
731 | |
732 | return ret; | |
733 | err: | |
9cf90cef | 734 | spin_lock(&mark->lock); |
11375145 JK |
735 | mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE | |
736 | FSNOTIFY_MARK_FLAG_ATTACHED); | |
5444e298 | 737 | list_del_init(&mark->g_list); |
9cf90cef | 738 | spin_unlock(&mark->lock); |
5444e298 | 739 | |
11375145 | 740 | fsnotify_put_mark(mark); |
5444e298 EP |
741 | return ret; |
742 | } | |
743 | ||
b812a9f5 | 744 | int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp, |
f3010343 | 745 | unsigned int obj_type, int add_flags, |
ad69cd99 | 746 | __kernel_fsid_t *fsid) |
d5a335b8 LS |
747 | { |
748 | int ret; | |
7b129323 JK |
749 | struct fsnotify_group *group = mark->group; |
750 | ||
43b245a7 | 751 | fsnotify_group_lock(group); |
f3010343 | 752 | ret = fsnotify_add_mark_locked(mark, connp, obj_type, add_flags, fsid); |
43b245a7 | 753 | fsnotify_group_unlock(group); |
d5a335b8 LS |
754 | return ret; |
755 | } | |
b72679ee | 756 | EXPORT_SYMBOL_GPL(fsnotify_add_mark); |
d5a335b8 | 757 | |
0809ab69 JK |
758 | /* |
759 | * Given a list of marks, find the mark associated with given group. If found | |
760 | * take a reference to that mark and return it, else return NULL. | |
761 | */ | |
9b6e5434 AG |
762 | struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp, |
763 | struct fsnotify_group *group) | |
0809ab69 | 764 | { |
08991e83 | 765 | struct fsnotify_mark_connector *conn; |
0809ab69 JK |
766 | struct fsnotify_mark *mark; |
767 | ||
08991e83 | 768 | conn = fsnotify_grab_connector(connp); |
9dd813c1 JK |
769 | if (!conn) |
770 | return NULL; | |
771 | ||
772 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
6b3f05d2 JK |
773 | if (mark->group == group && |
774 | (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) { | |
0809ab69 | 775 | fsnotify_get_mark(mark); |
04662cab | 776 | spin_unlock(&conn->lock); |
0809ab69 JK |
777 | return mark; |
778 | } | |
779 | } | |
04662cab | 780 | spin_unlock(&conn->lock); |
0809ab69 JK |
781 | return NULL; |
782 | } | |
b72679ee | 783 | EXPORT_SYMBOL_GPL(fsnotify_find_mark); |
0809ab69 | 784 | |
d6f7b98b | 785 | /* Clear any marks in a group with given type mask */ |
18f2e0d3 | 786 | void fsnotify_clear_marks_by_group(struct fsnotify_group *group, |
ad69cd99 | 787 | unsigned int obj_type) |
5444e298 EP |
788 | { |
789 | struct fsnotify_mark *lmark, *mark; | |
8f2f3eb5 | 790 | LIST_HEAD(to_free); |
2e37c6ca | 791 | struct list_head *head = &to_free; |
5444e298 | 792 | |
2e37c6ca | 793 | /* Skip selection step if we want to clear all marks. */ |
ad69cd99 | 794 | if (obj_type == FSNOTIFY_OBJ_TYPE_ANY) { |
2e37c6ca JK |
795 | head = &group->marks_list; |
796 | goto clear; | |
797 | } | |
8f2f3eb5 JK |
798 | /* |
799 | * We have to be really careful here. Anytime we drop mark_mutex, e.g. | |
800 | * fsnotify_clear_marks_by_inode() can come and free marks. Even in our | |
801 | * to_free list so we have to use mark_mutex even when accessing that | |
802 | * list. And freeing mark requires us to drop mark_mutex. So we can | |
803 | * reliably free only the first mark in the list. That's why we first | |
804 | * move marks to free to to_free list in one go and then free marks in | |
805 | * to_free list one by one. | |
806 | */ | |
43b245a7 | 807 | fsnotify_group_lock(group); |
5444e298 | 808 | list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) { |
ad69cd99 | 809 | if (mark->connector->type == obj_type) |
8f2f3eb5 | 810 | list_move(&mark->g_list, &to_free); |
5444e298 | 811 | } |
43b245a7 | 812 | fsnotify_group_unlock(group); |
8f2f3eb5 | 813 | |
2e37c6ca | 814 | clear: |
8f2f3eb5 | 815 | while (1) { |
43b245a7 | 816 | fsnotify_group_lock(group); |
2e37c6ca | 817 | if (list_empty(head)) { |
43b245a7 | 818 | fsnotify_group_unlock(group); |
8f2f3eb5 JK |
819 | break; |
820 | } | |
2e37c6ca | 821 | mark = list_first_entry(head, struct fsnotify_mark, g_list); |
8f2f3eb5 | 822 | fsnotify_get_mark(mark); |
4712e722 | 823 | fsnotify_detach_mark(mark); |
43b245a7 | 824 | fsnotify_group_unlock(group); |
4712e722 | 825 | fsnotify_free_mark(mark); |
8f2f3eb5 JK |
826 | fsnotify_put_mark(mark); |
827 | } | |
5444e298 EP |
828 | } |
829 | ||
9b6e5434 AG |
830 | /* Destroy all marks attached to an object via connector */ |
831 | void fsnotify_destroy_marks(fsnotify_connp_t *connp) | |
0810b4f9 | 832 | { |
08991e83 | 833 | struct fsnotify_mark_connector *conn; |
6b3f05d2 | 834 | struct fsnotify_mark *mark, *old_mark = NULL; |
721fb6fb JK |
835 | void *objp; |
836 | unsigned int type; | |
0810b4f9 | 837 | |
6b3f05d2 JK |
838 | conn = fsnotify_grab_connector(connp); |
839 | if (!conn) | |
840 | return; | |
841 | /* | |
842 | * We have to be careful since we can race with e.g. | |
843 | * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the | |
844 | * list can get modified. However we are holding mark reference and | |
845 | * thus our mark cannot be removed from obj_list so we can continue | |
846 | * iteration after regaining conn->lock. | |
847 | */ | |
848 | hlist_for_each_entry(mark, &conn->list, obj_list) { | |
0810b4f9 | 849 | fsnotify_get_mark(mark); |
04662cab | 850 | spin_unlock(&conn->lock); |
6b3f05d2 JK |
851 | if (old_mark) |
852 | fsnotify_put_mark(old_mark); | |
853 | old_mark = mark; | |
0810b4f9 | 854 | fsnotify_destroy_mark(mark, mark->group); |
6b3f05d2 | 855 | spin_lock(&conn->lock); |
0810b4f9 | 856 | } |
6b3f05d2 JK |
857 | /* |
858 | * Detach list from object now so that we don't pin inode until all | |
859 | * mark references get dropped. It would lead to strange results such | |
860 | * as delaying inode deletion or blocking unmount. | |
861 | */ | |
721fb6fb | 862 | objp = fsnotify_detach_connector_from_object(conn, &type); |
6b3f05d2 JK |
863 | spin_unlock(&conn->lock); |
864 | if (old_mark) | |
865 | fsnotify_put_mark(old_mark); | |
721fb6fb | 866 | fsnotify_drop_object(type, objp); |
0810b4f9 JK |
867 | } |
868 | ||
5444e298 EP |
869 | /* |
870 | * Nothing fancy, just initialize lists and locks and counters. | |
871 | */ | |
872 | void fsnotify_init_mark(struct fsnotify_mark *mark, | |
054c636e | 873 | struct fsnotify_group *group) |
5444e298 | 874 | { |
ba643f04 | 875 | memset(mark, 0, sizeof(*mark)); |
5444e298 | 876 | spin_lock_init(&mark->lock); |
ab97f873 | 877 | refcount_set(&mark->refcnt, 1); |
7b129323 JK |
878 | fsnotify_get_group(group); |
879 | mark->group = group; | |
b1da6a51 | 880 | WRITE_ONCE(mark->connector, NULL); |
5444e298 | 881 | } |
b72679ee | 882 | EXPORT_SYMBOL_GPL(fsnotify_init_mark); |
13d34ac6 | 883 | |
35e48176 JK |
884 | /* |
885 | * Destroy all marks in destroy_list, waits for SRCU period to finish before | |
886 | * actually freeing marks. | |
887 | */ | |
f09b04a0 | 888 | static void fsnotify_mark_destroy_workfn(struct work_struct *work) |
13d34ac6 JL |
889 | { |
890 | struct fsnotify_mark *mark, *next; | |
891 | struct list_head private_destroy_list; | |
892 | ||
0918f1c3 JL |
893 | spin_lock(&destroy_lock); |
894 | /* exchange the list head */ | |
895 | list_replace_init(&destroy_list, &private_destroy_list); | |
896 | spin_unlock(&destroy_lock); | |
13d34ac6 | 897 | |
0918f1c3 | 898 | synchronize_srcu(&fsnotify_mark_srcu); |
13d34ac6 | 899 | |
0918f1c3 JL |
900 | list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) { |
901 | list_del_init(&mark->g_list); | |
6b3f05d2 | 902 | fsnotify_final_mark_destroy(mark); |
13d34ac6 | 903 | } |
13d34ac6 | 904 | } |
35e48176 | 905 | |
f09b04a0 JK |
906 | /* Wait for all marks queued for destruction to be actually destroyed */ |
907 | void fsnotify_wait_marks_destroyed(void) | |
35e48176 | 908 | { |
f09b04a0 | 909 | flush_delayed_work(&reaper_work); |
35e48176 | 910 | } |
b72679ee | 911 | EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed); |