1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
7 * fsnotify inode mark locking/lifetime/and refcnting
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.
18 * There are 3 locks involved with fsnotify inode marks and they MUST be taken
19 * in order as follows:
23 * mark->connector->lock
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).
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.
34 * mark->connector->lock protects the list of marks anchored inside an
35 * inode / vfsmount and each mark is hooked via the i_list.
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
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.
48 * Inode marks survive between when they are added to an inode and when their
49 * refcnt==0. Marks are also protected by fsnotify_mark_srcu.
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
57 * need to be cleaned up. (fsnotify_clear_marks_by_group)
59 * This has the very interesting property of being able to run concurrently with
60 * any (or all) other directions.
64 #include <linux/init.h>
65 #include <linux/kernel.h>
66 #include <linux/kthread.h>
67 #include <linux/module.h>
68 #include <linux/mutex.h>
69 #include <linux/slab.h>
70 #include <linux/spinlock.h>
71 #include <linux/srcu.h>
72 #include <linux/ratelimit.h>
74 #include <linux/atomic.h>
76 #include <linux/fsnotify_backend.h>
79 #define FSNOTIFY_REAPER_DELAY (1) /* 1 jiffy */
81 struct srcu_struct fsnotify_mark_srcu
;
82 struct kmem_cache
*fsnotify_mark_connector_cachep
;
84 static DEFINE_SPINLOCK(destroy_lock
);
85 static LIST_HEAD(destroy_list
);
86 static struct fsnotify_mark_connector
*connector_destroy_list
;
88 static void fsnotify_mark_destroy_workfn(struct work_struct
*work
);
89 static DECLARE_DELAYED_WORK(reaper_work
, fsnotify_mark_destroy_workfn
);
91 static void fsnotify_connector_destroy_workfn(struct work_struct
*work
);
92 static DECLARE_WORK(connector_reaper_work
, fsnotify_connector_destroy_workfn
);
94 void fsnotify_get_mark(struct fsnotify_mark
*mark
)
96 WARN_ON_ONCE(!refcount_read(&mark
->refcnt
));
97 refcount_inc(&mark
->refcnt
);
100 static __u32
*fsnotify_conn_mask_p(struct fsnotify_mark_connector
*conn
)
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
;
106 else if (conn
->type
== FSNOTIFY_OBJ_TYPE_SB
)
107 return &fsnotify_conn_sb(conn
)->s_fsnotify_mask
;
111 __u32
fsnotify_conn_mask(struct fsnotify_mark_connector
*conn
)
113 if (WARN_ON(!fsnotify_valid_obj_type(conn
->type
)))
116 return *fsnotify_conn_mask_p(conn
);
119 static void __fsnotify_recalc_mask(struct fsnotify_mark_connector
*conn
)
122 struct fsnotify_mark
*mark
;
124 assert_spin_locked(&conn
->lock
);
125 /* We can get detached connector here when inode is getting unlinked. */
126 if (!fsnotify_valid_obj_type(conn
->type
))
128 hlist_for_each_entry(mark
, &conn
->list
, obj_list
) {
129 if (mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
)
130 new_mask
|= mark
->mask
;
132 *fsnotify_conn_mask_p(conn
) = new_mask
;
136 * Calculate mask of events for a list of marks. The caller must make sure
137 * connector and connector->obj cannot disappear under us. Callers achieve
138 * this by holding a mark->lock or mark->group->mark_mutex for a mark on this
141 void fsnotify_recalc_mask(struct fsnotify_mark_connector
*conn
)
146 spin_lock(&conn
->lock
);
147 __fsnotify_recalc_mask(conn
);
148 spin_unlock(&conn
->lock
);
149 if (conn
->type
== FSNOTIFY_OBJ_TYPE_INODE
)
150 __fsnotify_update_child_dentry_flags(
151 fsnotify_conn_inode(conn
));
154 /* Free all connectors queued for freeing once SRCU period ends */
155 static void fsnotify_connector_destroy_workfn(struct work_struct
*work
)
157 struct fsnotify_mark_connector
*conn
, *free
;
159 spin_lock(&destroy_lock
);
160 conn
= connector_destroy_list
;
161 connector_destroy_list
= NULL
;
162 spin_unlock(&destroy_lock
);
164 synchronize_srcu(&fsnotify_mark_srcu
);
167 conn
= conn
->destroy_next
;
168 kmem_cache_free(fsnotify_mark_connector_cachep
, free
);
172 static void *fsnotify_detach_connector_from_object(
173 struct fsnotify_mark_connector
*conn
,
176 struct inode
*inode
= NULL
;
179 if (conn
->type
== FSNOTIFY_OBJ_TYPE_DETACHED
)
182 if (conn
->type
== FSNOTIFY_OBJ_TYPE_INODE
) {
183 inode
= fsnotify_conn_inode(conn
);
184 inode
->i_fsnotify_mask
= 0;
185 atomic_long_inc(&inode
->i_sb
->s_fsnotify_inode_refs
);
186 } else if (conn
->type
== FSNOTIFY_OBJ_TYPE_VFSMOUNT
) {
187 fsnotify_conn_mount(conn
)->mnt_fsnotify_mask
= 0;
188 } else if (conn
->type
== FSNOTIFY_OBJ_TYPE_SB
) {
189 fsnotify_conn_sb(conn
)->s_fsnotify_mask
= 0;
192 rcu_assign_pointer(*(conn
->obj
), NULL
);
194 conn
->type
= FSNOTIFY_OBJ_TYPE_DETACHED
;
199 static void fsnotify_final_mark_destroy(struct fsnotify_mark
*mark
)
201 struct fsnotify_group
*group
= mark
->group
;
203 if (WARN_ON_ONCE(!group
))
205 group
->ops
->free_mark(mark
);
206 fsnotify_put_group(group
);
209 /* Drop object reference originally held by a connector */
210 static void fsnotify_drop_object(unsigned int type
, void *objp
)
213 struct super_block
*sb
;
217 /* Currently only inode references are passed to be dropped */
218 if (WARN_ON_ONCE(type
!= FSNOTIFY_OBJ_TYPE_INODE
))
223 if (atomic_long_dec_and_test(&sb
->s_fsnotify_inode_refs
))
224 wake_up_var(&sb
->s_fsnotify_inode_refs
);
227 void fsnotify_put_mark(struct fsnotify_mark
*mark
)
229 struct fsnotify_mark_connector
*conn
= READ_ONCE(mark
->connector
);
231 unsigned int type
= FSNOTIFY_OBJ_TYPE_DETACHED
;
232 bool free_conn
= false;
234 /* Catch marks that were actually never attached to object */
236 if (refcount_dec_and_test(&mark
->refcnt
))
237 fsnotify_final_mark_destroy(mark
);
242 * We have to be careful so that traversals of obj_list under lock can
243 * safely grab mark reference.
245 if (!refcount_dec_and_lock(&mark
->refcnt
, &conn
->lock
))
248 hlist_del_init_rcu(&mark
->obj_list
);
249 if (hlist_empty(&conn
->list
)) {
250 objp
= fsnotify_detach_connector_from_object(conn
, &type
);
253 __fsnotify_recalc_mask(conn
);
255 WRITE_ONCE(mark
->connector
, NULL
);
256 spin_unlock(&conn
->lock
);
258 fsnotify_drop_object(type
, objp
);
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
);
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
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
);
279 EXPORT_SYMBOL_GPL(fsnotify_put_mark
);
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.
286 * Also pin the group so it doesn't disappear under us.
288 static bool fsnotify_get_mark_safe(struct fsnotify_mark
*mark
)
293 if (refcount_inc_not_zero(&mark
->refcnt
)) {
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
);
301 spin_unlock(&mark
->lock
);
302 fsnotify_put_mark(mark
);
308 * Puts marks and wakes up group destruction if necessary.
310 * Pairs with fsnotify_get_mark_safe()
312 static void fsnotify_put_mark_wake(struct fsnotify_mark
*mark
)
315 struct fsnotify_group
*group
= mark
->group
;
317 fsnotify_put_mark(mark
);
319 * We abuse notification_waitq on group shutdown for waiting for
320 * all marks pinned when waiting for userspace.
322 if (atomic_dec_and_test(&group
->user_waits
) && group
->shutdown
)
323 wake_up(&group
->notification_waitq
);
327 bool fsnotify_prepare_user_wait(struct fsnotify_iter_info
*iter_info
)
328 __releases(&fsnotify_mark_srcu
)
332 fsnotify_foreach_obj_type(type
) {
333 /* This can fail if mark is being removed */
334 if (!fsnotify_get_mark_safe(iter_info
->marks
[type
])) {
335 __release(&fsnotify_mark_srcu
);
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.
345 srcu_read_unlock(&fsnotify_mark_srcu
, iter_info
->srcu_idx
);
350 for (type
--; type
>= 0; type
--)
351 fsnotify_put_mark_wake(iter_info
->marks
[type
]);
355 void fsnotify_finish_user_wait(struct fsnotify_iter_info
*iter_info
)
356 __acquires(&fsnotify_mark_srcu
)
360 iter_info
->srcu_idx
= srcu_read_lock(&fsnotify_mark_srcu
);
361 fsnotify_foreach_obj_type(type
)
362 fsnotify_put_mark_wake(iter_info
->marks
[type
]);
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.
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.
375 void fsnotify_detach_mark(struct fsnotify_mark
*mark
)
377 struct fsnotify_group
*group
= mark
->group
;
379 WARN_ON_ONCE(!mutex_is_locked(&group
->mark_mutex
));
380 WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu
) &&
381 refcount_read(&mark
->refcnt
) < 1 +
382 !!(mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
));
384 spin_lock(&mark
->lock
);
385 /* something else already called this function on this mark */
386 if (!(mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
)) {
387 spin_unlock(&mark
->lock
);
390 mark
->flags
&= ~FSNOTIFY_MARK_FLAG_ATTACHED
;
391 list_del_init(&mark
->g_list
);
392 spin_unlock(&mark
->lock
);
394 atomic_dec(&group
->num_marks
);
396 /* Drop mark reference acquired in fsnotify_add_mark_locked() */
397 fsnotify_put_mark(mark
);
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.
405 * Caller must have a reference to the mark or be protected by
406 * fsnotify_mark_srcu.
408 void fsnotify_free_mark(struct fsnotify_mark
*mark
)
410 struct fsnotify_group
*group
= mark
->group
;
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
);
418 mark
->flags
&= ~FSNOTIFY_MARK_FLAG_ALIVE
;
419 spin_unlock(&mark
->lock
);
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
426 if (group
->ops
->freeing_mark
)
427 group
->ops
->freeing_mark(mark
, group
);
430 void fsnotify_destroy_mark(struct fsnotify_mark
*mark
,
431 struct fsnotify_group
*group
)
433 mutex_lock_nested(&group
->mark_mutex
, SINGLE_DEPTH_NESTING
);
434 fsnotify_detach_mark(mark
);
435 mutex_unlock(&group
->mark_mutex
);
436 fsnotify_free_mark(mark
);
438 EXPORT_SYMBOL_GPL(fsnotify_destroy_mark
);
441 * Sorting function for lists of fsnotify marks.
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
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.
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.
459 int fsnotify_compare_groups(struct fsnotify_group
*a
, struct fsnotify_group
*b
)
467 if (a
->priority
< b
->priority
)
469 if (a
->priority
> b
->priority
)
476 static int fsnotify_attach_connector_to_object(fsnotify_connp_t
*connp
,
478 __kernel_fsid_t
*fsid
)
480 struct inode
*inode
= NULL
;
481 struct fsnotify_mark_connector
*conn
;
483 conn
= kmem_cache_alloc(fsnotify_mark_connector_cachep
, GFP_KERNEL
);
486 spin_lock_init(&conn
->lock
);
487 INIT_HLIST_HEAD(&conn
->list
);
490 /* Cache fsid of filesystem containing the object */
493 conn
->flags
= FSNOTIFY_CONN_FLAG_HAS_FSID
;
495 conn
->fsid
.val
[0] = conn
->fsid
.val
[1] = 0;
498 if (conn
->type
== FSNOTIFY_OBJ_TYPE_INODE
)
499 inode
= igrab(fsnotify_conn_inode(conn
));
501 * cmpxchg() provides the barrier so that readers of *connp can see
502 * only initialized structure
504 if (cmpxchg(connp
, NULL
, conn
)) {
505 /* Someone else created list structure for us */
508 kmem_cache_free(fsnotify_mark_connector_cachep
, conn
);
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.
520 static struct fsnotify_mark_connector
*fsnotify_grab_connector(
521 fsnotify_connp_t
*connp
)
523 struct fsnotify_mark_connector
*conn
;
526 idx
= srcu_read_lock(&fsnotify_mark_srcu
);
527 conn
= srcu_dereference(*connp
, &fsnotify_mark_srcu
);
530 spin_lock(&conn
->lock
);
531 if (conn
->type
== FSNOTIFY_OBJ_TYPE_DETACHED
) {
532 spin_unlock(&conn
->lock
);
533 srcu_read_unlock(&fsnotify_mark_srcu
, idx
);
537 srcu_read_unlock(&fsnotify_mark_srcu
, idx
);
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.
547 static int fsnotify_add_mark_list(struct fsnotify_mark
*mark
,
548 fsnotify_connp_t
*connp
, unsigned int type
,
549 int allow_dups
, __kernel_fsid_t
*fsid
)
551 struct fsnotify_mark
*lmark
, *last
= NULL
;
552 struct fsnotify_mark_connector
*conn
;
556 if (WARN_ON(!fsnotify_valid_obj_type(type
)))
559 /* Backend is expected to check for zero fsid (e.g. tmpfs) */
560 if (fsid
&& WARN_ON_ONCE(!fsid
->val
[0] && !fsid
->val
[1]))
564 spin_lock(&mark
->lock
);
565 conn
= fsnotify_grab_connector(connp
);
567 spin_unlock(&mark
->lock
);
568 err
= fsnotify_attach_connector_to_object(connp
, type
, fsid
);
572 } else if (fsid
&& !(conn
->flags
& FSNOTIFY_CONN_FLAG_HAS_FSID
)) {
574 /* Pairs with smp_rmb() in fanotify_get_fsid() */
576 conn
->flags
|= FSNOTIFY_CONN_FLAG_HAS_FSID
;
577 } else if (fsid
&& (conn
->flags
& FSNOTIFY_CONN_FLAG_HAS_FSID
) &&
578 (fsid
->val
[0] != conn
->fsid
.val
[0] ||
579 fsid
->val
[1] != conn
->fsid
.val
[1])) {
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
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]);
595 /* is mark the first mark? */
596 if (hlist_empty(&conn
->list
)) {
597 hlist_add_head_rcu(&mark
->obj_list
, &conn
->list
);
601 /* should mark be in the middle of the current list? */
602 hlist_for_each_entry(lmark
, &conn
->list
, obj_list
) {
605 if ((lmark
->group
== mark
->group
) &&
606 (lmark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
) &&
612 cmp
= fsnotify_compare_groups(lmark
->group
, mark
->group
);
614 hlist_add_before_rcu(&mark
->obj_list
, &lmark
->obj_list
);
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
);
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.
628 WRITE_ONCE(mark
->connector
, conn
);
630 spin_unlock(&conn
->lock
);
631 spin_unlock(&mark
->lock
);
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.
640 int fsnotify_add_mark_locked(struct fsnotify_mark
*mark
,
641 fsnotify_connp_t
*connp
, unsigned int type
,
642 int allow_dups
, __kernel_fsid_t
*fsid
)
644 struct fsnotify_group
*group
= mark
->group
;
647 BUG_ON(!mutex_is_locked(&group
->mark_mutex
));
653 * mark->connector->lock
655 spin_lock(&mark
->lock
);
656 mark
->flags
|= FSNOTIFY_MARK_FLAG_ALIVE
| FSNOTIFY_MARK_FLAG_ATTACHED
;
658 list_add(&mark
->g_list
, &group
->marks_list
);
659 atomic_inc(&group
->num_marks
);
660 fsnotify_get_mark(mark
); /* for g_list */
661 spin_unlock(&mark
->lock
);
663 ret
= fsnotify_add_mark_list(mark
, connp
, type
, allow_dups
, fsid
);
668 fsnotify_recalc_mask(mark
->connector
);
672 spin_lock(&mark
->lock
);
673 mark
->flags
&= ~(FSNOTIFY_MARK_FLAG_ALIVE
|
674 FSNOTIFY_MARK_FLAG_ATTACHED
);
675 list_del_init(&mark
->g_list
);
676 spin_unlock(&mark
->lock
);
677 atomic_dec(&group
->num_marks
);
679 fsnotify_put_mark(mark
);
683 int fsnotify_add_mark(struct fsnotify_mark
*mark
, fsnotify_connp_t
*connp
,
684 unsigned int type
, int allow_dups
, __kernel_fsid_t
*fsid
)
687 struct fsnotify_group
*group
= mark
->group
;
689 mutex_lock(&group
->mark_mutex
);
690 ret
= fsnotify_add_mark_locked(mark
, connp
, type
, allow_dups
, fsid
);
691 mutex_unlock(&group
->mark_mutex
);
694 EXPORT_SYMBOL_GPL(fsnotify_add_mark
);
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.
700 struct fsnotify_mark
*fsnotify_find_mark(fsnotify_connp_t
*connp
,
701 struct fsnotify_group
*group
)
703 struct fsnotify_mark_connector
*conn
;
704 struct fsnotify_mark
*mark
;
706 conn
= fsnotify_grab_connector(connp
);
710 hlist_for_each_entry(mark
, &conn
->list
, obj_list
) {
711 if (mark
->group
== group
&&
712 (mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
)) {
713 fsnotify_get_mark(mark
);
714 spin_unlock(&conn
->lock
);
718 spin_unlock(&conn
->lock
);
721 EXPORT_SYMBOL_GPL(fsnotify_find_mark
);
723 /* Clear any marks in a group with given type mask */
724 void fsnotify_clear_marks_by_group(struct fsnotify_group
*group
,
725 unsigned int type_mask
)
727 struct fsnotify_mark
*lmark
, *mark
;
729 struct list_head
*head
= &to_free
;
731 /* Skip selection step if we want to clear all marks. */
732 if (type_mask
== FSNOTIFY_OBJ_ALL_TYPES_MASK
) {
733 head
= &group
->marks_list
;
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.
745 mutex_lock_nested(&group
->mark_mutex
, SINGLE_DEPTH_NESTING
);
746 list_for_each_entry_safe(mark
, lmark
, &group
->marks_list
, g_list
) {
747 if ((1U << mark
->connector
->type
) & type_mask
)
748 list_move(&mark
->g_list
, &to_free
);
750 mutex_unlock(&group
->mark_mutex
);
754 mutex_lock_nested(&group
->mark_mutex
, SINGLE_DEPTH_NESTING
);
755 if (list_empty(head
)) {
756 mutex_unlock(&group
->mark_mutex
);
759 mark
= list_first_entry(head
, struct fsnotify_mark
, g_list
);
760 fsnotify_get_mark(mark
);
761 fsnotify_detach_mark(mark
);
762 mutex_unlock(&group
->mark_mutex
);
763 fsnotify_free_mark(mark
);
764 fsnotify_put_mark(mark
);
768 /* Destroy all marks attached to an object via connector */
769 void fsnotify_destroy_marks(fsnotify_connp_t
*connp
)
771 struct fsnotify_mark_connector
*conn
;
772 struct fsnotify_mark
*mark
, *old_mark
= NULL
;
776 conn
= fsnotify_grab_connector(connp
);
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.
786 hlist_for_each_entry(mark
, &conn
->list
, obj_list
) {
787 fsnotify_get_mark(mark
);
788 spin_unlock(&conn
->lock
);
790 fsnotify_put_mark(old_mark
);
792 fsnotify_destroy_mark(mark
, mark
->group
);
793 spin_lock(&conn
->lock
);
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.
800 objp
= fsnotify_detach_connector_from_object(conn
, &type
);
801 spin_unlock(&conn
->lock
);
803 fsnotify_put_mark(old_mark
);
804 fsnotify_drop_object(type
, objp
);
808 * Nothing fancy, just initialize lists and locks and counters.
810 void fsnotify_init_mark(struct fsnotify_mark
*mark
,
811 struct fsnotify_group
*group
)
813 memset(mark
, 0, sizeof(*mark
));
814 spin_lock_init(&mark
->lock
);
815 refcount_set(&mark
->refcnt
, 1);
816 fsnotify_get_group(group
);
818 WRITE_ONCE(mark
->connector
, NULL
);
820 EXPORT_SYMBOL_GPL(fsnotify_init_mark
);
823 * Destroy all marks in destroy_list, waits for SRCU period to finish before
824 * actually freeing marks.
826 static void fsnotify_mark_destroy_workfn(struct work_struct
*work
)
828 struct fsnotify_mark
*mark
, *next
;
829 struct list_head private_destroy_list
;
831 spin_lock(&destroy_lock
);
832 /* exchange the list head */
833 list_replace_init(&destroy_list
, &private_destroy_list
);
834 spin_unlock(&destroy_lock
);
836 synchronize_srcu(&fsnotify_mark_srcu
);
838 list_for_each_entry_safe(mark
, next
, &private_destroy_list
, g_list
) {
839 list_del_init(&mark
->g_list
);
840 fsnotify_final_mark_destroy(mark
);
844 /* Wait for all marks queued for destruction to be actually destroyed */
845 void fsnotify_wait_marks_destroyed(void)
847 flush_delayed_work(&reaper_work
);
849 EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed
);