1 // SPDX-License-Identifier: GPL-2.0-or-later
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 fsnotify_connp_t *fsnotify_object_connp(void *obj,
101 enum fsnotify_obj_type obj_type)
104 case FSNOTIFY_OBJ_TYPE_INODE:
105 return &((struct inode *)obj)->i_fsnotify_marks;
106 case FSNOTIFY_OBJ_TYPE_VFSMOUNT:
107 return &real_mount(obj)->mnt_fsnotify_marks;
108 case FSNOTIFY_OBJ_TYPE_SB:
109 return fsnotify_sb_marks(obj);
115 static __u32 *fsnotify_conn_mask_p(struct fsnotify_mark_connector *conn)
117 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
118 return &fsnotify_conn_inode(conn)->i_fsnotify_mask;
119 else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT)
120 return &fsnotify_conn_mount(conn)->mnt_fsnotify_mask;
121 else if (conn->type == FSNOTIFY_OBJ_TYPE_SB)
122 return &fsnotify_conn_sb(conn)->s_fsnotify_mask;
126 __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn)
128 if (WARN_ON(!fsnotify_valid_obj_type(conn->type)))
131 return READ_ONCE(*fsnotify_conn_mask_p(conn));
134 static void fsnotify_get_sb_watched_objects(struct super_block *sb)
136 atomic_long_inc(fsnotify_sb_watched_objects(sb));
139 static void fsnotify_put_sb_watched_objects(struct super_block *sb)
141 atomic_long_t *watched_objects = fsnotify_sb_watched_objects(sb);
143 /* the superblock can go away after this decrement */
144 if (atomic_long_dec_and_test(watched_objects))
145 wake_up_var(watched_objects);
148 static void fsnotify_get_inode_ref(struct inode *inode)
151 fsnotify_get_sb_watched_objects(inode->i_sb);
154 static void fsnotify_put_inode_ref(struct inode *inode)
156 /* read ->i_sb before the inode can go away */
157 struct super_block *sb = inode->i_sb;
160 fsnotify_put_sb_watched_objects(sb);
164 * Grab or drop watched objects reference depending on whether the connector
165 * is attached and has any marks attached.
167 static void fsnotify_update_sb_watchers(struct super_block *sb,
168 struct fsnotify_mark_connector *conn)
170 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
171 bool is_watched = conn->flags & FSNOTIFY_CONN_FLAG_IS_WATCHED;
172 struct fsnotify_mark *first_mark = NULL;
173 unsigned int highest_prio = 0;
176 first_mark = hlist_entry_safe(conn->list.first,
177 struct fsnotify_mark, obj_list);
179 highest_prio = first_mark->group->priority;
180 if (WARN_ON(highest_prio >= __FSNOTIFY_PRIO_NUM))
184 * If the highest priority of group watching this object is prio,
185 * then watched object has a reference on counters [0..prio].
186 * Update priority >= 1 watched objects counters.
188 for (unsigned int p = conn->prio + 1; p <= highest_prio; p++)
189 atomic_long_inc(&sbinfo->watched_objects[p]);
190 for (unsigned int p = conn->prio; p > highest_prio; p--)
191 atomic_long_dec(&sbinfo->watched_objects[p]);
192 conn->prio = highest_prio;
194 /* Update priority >= 0 (a.k.a total) watched objects counter */
195 BUILD_BUG_ON(FSNOTIFY_PRIO_NORMAL != 0);
196 if (first_mark && !is_watched) {
197 conn->flags |= FSNOTIFY_CONN_FLAG_IS_WATCHED;
198 fsnotify_get_sb_watched_objects(sb);
199 } else if (!first_mark && is_watched) {
200 conn->flags &= ~FSNOTIFY_CONN_FLAG_IS_WATCHED;
201 fsnotify_put_sb_watched_objects(sb);
206 * Grab or drop inode reference for the connector if needed.
208 * When it's time to drop the reference, we only clear the HAS_IREF flag and
209 * return the inode object. fsnotify_drop_object() will be resonsible for doing
210 * iput() outside of spinlocks. This happens when last mark that wanted iref is
213 static struct inode *fsnotify_update_iref(struct fsnotify_mark_connector *conn,
216 bool has_iref = conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF;
217 struct inode *inode = NULL;
219 if (conn->type != FSNOTIFY_OBJ_TYPE_INODE ||
220 want_iref == has_iref)
224 /* Pin inode if any mark wants inode refcount held */
225 fsnotify_get_inode_ref(fsnotify_conn_inode(conn));
226 conn->flags |= FSNOTIFY_CONN_FLAG_HAS_IREF;
228 /* Unpin inode after detach of last mark that wanted iref */
229 inode = fsnotify_conn_inode(conn);
230 conn->flags &= ~FSNOTIFY_CONN_FLAG_HAS_IREF;
236 static void *__fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
239 bool want_iref = false;
240 struct fsnotify_mark *mark;
242 assert_spin_locked(&conn->lock);
243 /* We can get detached connector here when inode is getting unlinked. */
244 if (!fsnotify_valid_obj_type(conn->type))
246 hlist_for_each_entry(mark, &conn->list, obj_list) {
247 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED))
249 new_mask |= fsnotify_calc_mask(mark);
250 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE &&
251 !(mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
255 * We use WRITE_ONCE() to prevent silly compiler optimizations from
256 * confusing readers not holding conn->lock with partial updates.
258 WRITE_ONCE(*fsnotify_conn_mask_p(conn), new_mask);
260 return fsnotify_update_iref(conn, want_iref);
263 static bool fsnotify_conn_watches_children(
264 struct fsnotify_mark_connector *conn)
266 if (conn->type != FSNOTIFY_OBJ_TYPE_INODE)
269 return fsnotify_inode_watches_children(fsnotify_conn_inode(conn));
272 static void fsnotify_conn_set_children_dentry_flags(
273 struct fsnotify_mark_connector *conn)
275 if (conn->type != FSNOTIFY_OBJ_TYPE_INODE)
278 fsnotify_set_children_dentry_flags(fsnotify_conn_inode(conn));
282 * Calculate mask of events for a list of marks. The caller must make sure
283 * connector and connector->obj cannot disappear under us. Callers achieve
284 * this by holding a mark->lock or mark->group->mark_mutex for a mark on this
287 void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
289 bool update_children;
294 spin_lock(&conn->lock);
295 update_children = !fsnotify_conn_watches_children(conn);
296 __fsnotify_recalc_mask(conn);
297 update_children &= fsnotify_conn_watches_children(conn);
298 spin_unlock(&conn->lock);
300 * Set children's PARENT_WATCHED flags only if parent started watching.
301 * When parent stops watching, we clear false positive PARENT_WATCHED
302 * flags lazily in __fsnotify_parent().
305 fsnotify_conn_set_children_dentry_flags(conn);
308 /* Free all connectors queued for freeing once SRCU period ends */
309 static void fsnotify_connector_destroy_workfn(struct work_struct *work)
311 struct fsnotify_mark_connector *conn, *free;
313 spin_lock(&destroy_lock);
314 conn = connector_destroy_list;
315 connector_destroy_list = NULL;
316 spin_unlock(&destroy_lock);
318 synchronize_srcu(&fsnotify_mark_srcu);
321 conn = conn->destroy_next;
322 kmem_cache_free(fsnotify_mark_connector_cachep, free);
326 static void *fsnotify_detach_connector_from_object(
327 struct fsnotify_mark_connector *conn,
330 fsnotify_connp_t *connp = fsnotify_object_connp(conn->obj, conn->type);
331 struct super_block *sb = fsnotify_connector_sb(conn);
332 struct inode *inode = NULL;
335 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED)
338 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
339 inode = fsnotify_conn_inode(conn);
340 inode->i_fsnotify_mask = 0;
342 /* Unpin inode when detaching from connector */
343 if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF))
345 } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) {
346 fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
347 } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
348 fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
351 rcu_assign_pointer(*connp, NULL);
353 conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
354 fsnotify_update_sb_watchers(sb, conn);
359 static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark)
361 struct fsnotify_group *group = mark->group;
363 if (WARN_ON_ONCE(!group))
365 group->ops->free_mark(mark);
366 fsnotify_put_group(group);
369 /* Drop object reference originally held by a connector */
370 static void fsnotify_drop_object(unsigned int type, void *objp)
374 /* Currently only inode references are passed to be dropped */
375 if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE))
377 fsnotify_put_inode_ref(objp);
380 void fsnotify_put_mark(struct fsnotify_mark *mark)
382 struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector);
384 unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED;
385 bool free_conn = false;
387 /* Catch marks that were actually never attached to object */
389 if (refcount_dec_and_test(&mark->refcnt))
390 fsnotify_final_mark_destroy(mark);
395 * We have to be careful so that traversals of obj_list under lock can
396 * safely grab mark reference.
398 if (!refcount_dec_and_lock(&mark->refcnt, &conn->lock))
401 hlist_del_init_rcu(&mark->obj_list);
402 if (hlist_empty(&conn->list)) {
403 objp = fsnotify_detach_connector_from_object(conn, &type);
406 struct super_block *sb = fsnotify_connector_sb(conn);
408 /* Update watched objects after detaching mark */
410 fsnotify_update_sb_watchers(sb, conn);
411 objp = __fsnotify_recalc_mask(conn);
414 WRITE_ONCE(mark->connector, NULL);
415 spin_unlock(&conn->lock);
417 fsnotify_drop_object(type, objp);
420 spin_lock(&destroy_lock);
421 conn->destroy_next = connector_destroy_list;
422 connector_destroy_list = conn;
423 spin_unlock(&destroy_lock);
424 queue_work(system_unbound_wq, &connector_reaper_work);
427 * Note that we didn't update flags telling whether inode cares about
428 * what's happening with children. We update these flags from
429 * __fsnotify_parent() lazily when next event happens on one of our
432 spin_lock(&destroy_lock);
433 list_add(&mark->g_list, &destroy_list);
434 spin_unlock(&destroy_lock);
435 queue_delayed_work(system_unbound_wq, &reaper_work,
436 FSNOTIFY_REAPER_DELAY);
438 EXPORT_SYMBOL_GPL(fsnotify_put_mark);
441 * Get mark reference when we found the mark via lockless traversal of object
442 * list. Mark can be already removed from the list by now and on its way to be
443 * destroyed once SRCU period ends.
445 * Also pin the group so it doesn't disappear under us.
447 static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark)
452 if (refcount_inc_not_zero(&mark->refcnt)) {
453 spin_lock(&mark->lock);
454 if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) {
455 /* mark is attached, group is still alive then */
456 atomic_inc(&mark->group->user_waits);
457 spin_unlock(&mark->lock);
460 spin_unlock(&mark->lock);
461 fsnotify_put_mark(mark);
467 * Puts marks and wakes up group destruction if necessary.
469 * Pairs with fsnotify_get_mark_safe()
471 static void fsnotify_put_mark_wake(struct fsnotify_mark *mark)
474 struct fsnotify_group *group = mark->group;
476 fsnotify_put_mark(mark);
478 * We abuse notification_waitq on group shutdown for waiting for
479 * all marks pinned when waiting for userspace.
481 if (atomic_dec_and_test(&group->user_waits) && group->shutdown)
482 wake_up(&group->notification_waitq);
486 bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info)
487 __releases(&fsnotify_mark_srcu)
491 fsnotify_foreach_iter_type(type) {
492 /* This can fail if mark is being removed */
493 if (!fsnotify_get_mark_safe(iter_info->marks[type])) {
494 __release(&fsnotify_mark_srcu);
500 * Now that both marks are pinned by refcount in the inode / vfsmount
501 * lists, we can drop SRCU lock, and safely resume the list iteration
502 * once userspace returns.
504 srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx);
509 for (type--; type >= 0; type--)
510 fsnotify_put_mark_wake(iter_info->marks[type]);
514 void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info)
515 __acquires(&fsnotify_mark_srcu)
519 iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
520 fsnotify_foreach_iter_type(type)
521 fsnotify_put_mark_wake(iter_info->marks[type]);
525 * Mark mark as detached, remove it from group list. Mark still stays in object
526 * list until its last reference is dropped. Note that we rely on mark being
527 * removed from group list before corresponding reference to it is dropped. In
528 * particular we rely on mark->connector being valid while we hold
529 * group->mark_mutex if we found the mark through g_list.
531 * Must be called with group->mark_mutex held. The caller must either hold
532 * reference to the mark or be protected by fsnotify_mark_srcu.
534 void fsnotify_detach_mark(struct fsnotify_mark *mark)
536 fsnotify_group_assert_locked(mark->group);
537 WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) &&
538 refcount_read(&mark->refcnt) < 1 +
539 !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED));
541 spin_lock(&mark->lock);
542 /* something else already called this function on this mark */
543 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
544 spin_unlock(&mark->lock);
547 mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
548 list_del_init(&mark->g_list);
549 spin_unlock(&mark->lock);
551 /* Drop mark reference acquired in fsnotify_add_mark_locked() */
552 fsnotify_put_mark(mark);
556 * Free fsnotify mark. The mark is actually only marked as being freed. The
557 * freeing is actually happening only once last reference to the mark is
558 * dropped from a workqueue which first waits for srcu period end.
560 * Caller must have a reference to the mark or be protected by
561 * fsnotify_mark_srcu.
563 void fsnotify_free_mark(struct fsnotify_mark *mark)
565 struct fsnotify_group *group = mark->group;
567 spin_lock(&mark->lock);
568 /* something else already called this function on this mark */
569 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
570 spin_unlock(&mark->lock);
573 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
574 spin_unlock(&mark->lock);
577 * Some groups like to know that marks are being freed. This is a
578 * callback to the group function to let it know that this mark
581 if (group->ops->freeing_mark)
582 group->ops->freeing_mark(mark, group);
585 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
586 struct fsnotify_group *group)
588 fsnotify_group_lock(group);
589 fsnotify_detach_mark(mark);
590 fsnotify_group_unlock(group);
591 fsnotify_free_mark(mark);
593 EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
596 * Sorting function for lists of fsnotify marks.
598 * Fanotify supports different notification classes (reflected as priority of
599 * notification group). Events shall be passed to notification groups in
600 * decreasing priority order. To achieve this marks in notification lists for
601 * inodes and vfsmounts are sorted so that priorities of corresponding groups
604 * Furthermore correct handling of the ignore mask requires processing inode
605 * and vfsmount marks of each group together. Using the group address as
606 * further sort criterion provides a unique sorting order and thus we can
607 * merge inode and vfsmount lists of marks in linear time and find groups
608 * present in both lists.
610 * A return value of 1 signifies that b has priority over a.
611 * A return value of 0 signifies that the two marks have to be handled together.
612 * A return value of -1 signifies that a has priority over b.
614 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
622 if (a->priority < b->priority)
624 if (a->priority > b->priority)
631 static int fsnotify_attach_info_to_sb(struct super_block *sb)
633 struct fsnotify_sb_info *sbinfo;
635 /* sb info is freed on fsnotify_sb_delete() */
636 sbinfo = kzalloc(sizeof(*sbinfo), GFP_KERNEL);
641 * cmpxchg() provides the barrier so that callers of fsnotify_sb_info()
642 * will observe an initialized structure
644 if (cmpxchg(&sb->s_fsnotify_info, NULL, sbinfo)) {
645 /* Someone else created sbinfo for us */
651 static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
652 void *obj, unsigned int obj_type)
654 struct fsnotify_mark_connector *conn;
656 conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL);
659 spin_lock_init(&conn->lock);
660 INIT_HLIST_HEAD(&conn->list);
663 conn->type = obj_type;
667 * cmpxchg() provides the barrier so that readers of *connp can see
668 * only initialized structure
670 if (cmpxchg(connp, NULL, conn)) {
671 /* Someone else created list structure for us */
672 kmem_cache_free(fsnotify_mark_connector_cachep, conn);
678 * Get mark connector, make sure it is alive and return with its lock held.
679 * This is for users that get connector pointer from inode or mount. Users that
680 * hold reference to a mark on the list may directly lock connector->lock as
681 * they are sure list cannot go away under them.
683 static struct fsnotify_mark_connector *fsnotify_grab_connector(
684 fsnotify_connp_t *connp)
686 struct fsnotify_mark_connector *conn;
689 idx = srcu_read_lock(&fsnotify_mark_srcu);
690 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
693 spin_lock(&conn->lock);
694 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) {
695 spin_unlock(&conn->lock);
696 srcu_read_unlock(&fsnotify_mark_srcu, idx);
700 srcu_read_unlock(&fsnotify_mark_srcu, idx);
705 * Add mark into proper place in given list of marks. These marks may be used
706 * for the fsnotify backend to determine which event types should be delivered
707 * to which group and for which inodes. These marks are ordered according to
708 * priority, highest number first, and then by the group's location in memory.
710 static int fsnotify_add_mark_list(struct fsnotify_mark *mark, void *obj,
711 unsigned int obj_type, int add_flags)
713 struct super_block *sb = fsnotify_object_sb(obj, obj_type);
714 struct fsnotify_mark *lmark, *last = NULL;
715 struct fsnotify_mark_connector *conn;
716 fsnotify_connp_t *connp;
720 if (WARN_ON(!fsnotify_valid_obj_type(obj_type)))
724 * Attach the sb info before attaching a connector to any object on sb.
725 * The sb info will remain attached as long as sb lives.
727 if (!fsnotify_sb_info(sb)) {
728 err = fsnotify_attach_info_to_sb(sb);
733 connp = fsnotify_object_connp(obj, obj_type);
735 spin_lock(&mark->lock);
736 conn = fsnotify_grab_connector(connp);
738 spin_unlock(&mark->lock);
739 err = fsnotify_attach_connector_to_object(connp, obj, obj_type);
745 /* is mark the first mark? */
746 if (hlist_empty(&conn->list)) {
747 hlist_add_head_rcu(&mark->obj_list, &conn->list);
751 /* should mark be in the middle of the current list? */
752 hlist_for_each_entry(lmark, &conn->list, obj_list) {
755 if ((lmark->group == mark->group) &&
756 (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) &&
757 !(mark->group->flags & FSNOTIFY_GROUP_DUPS)) {
762 cmp = fsnotify_compare_groups(lmark->group, mark->group);
764 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
769 BUG_ON(last == NULL);
770 /* mark should be the last entry. last is the current last entry */
771 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
773 fsnotify_update_sb_watchers(sb, conn);
775 * Since connector is attached to object using cmpxchg() we are
776 * guaranteed that connector initialization is fully visible by anyone
777 * seeing mark->connector set.
779 WRITE_ONCE(mark->connector, conn);
781 spin_unlock(&conn->lock);
782 spin_unlock(&mark->lock);
787 * Attach an initialized mark to a given group and fs object.
788 * These marks may be used for the fsnotify backend to determine which
789 * event types should be delivered to which group.
791 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
792 void *obj, unsigned int obj_type,
795 struct fsnotify_group *group = mark->group;
798 fsnotify_group_assert_locked(group);
804 * mark->connector->lock
806 spin_lock(&mark->lock);
807 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
809 list_add(&mark->g_list, &group->marks_list);
810 fsnotify_get_mark(mark); /* for g_list */
811 spin_unlock(&mark->lock);
813 ret = fsnotify_add_mark_list(mark, obj, obj_type, add_flags);
817 fsnotify_recalc_mask(mark->connector);
821 spin_lock(&mark->lock);
822 mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE |
823 FSNOTIFY_MARK_FLAG_ATTACHED);
824 list_del_init(&mark->g_list);
825 spin_unlock(&mark->lock);
827 fsnotify_put_mark(mark);
831 int fsnotify_add_mark(struct fsnotify_mark *mark, void *obj,
832 unsigned int obj_type, int add_flags)
835 struct fsnotify_group *group = mark->group;
837 fsnotify_group_lock(group);
838 ret = fsnotify_add_mark_locked(mark, obj, obj_type, add_flags);
839 fsnotify_group_unlock(group);
842 EXPORT_SYMBOL_GPL(fsnotify_add_mark);
845 * Given a list of marks, find the mark associated with given group. If found
846 * take a reference to that mark and return it, else return NULL.
848 struct fsnotify_mark *fsnotify_find_mark(void *obj, unsigned int obj_type,
849 struct fsnotify_group *group)
851 fsnotify_connp_t *connp = fsnotify_object_connp(obj, obj_type);
852 struct fsnotify_mark_connector *conn;
853 struct fsnotify_mark *mark;
858 conn = fsnotify_grab_connector(connp);
862 hlist_for_each_entry(mark, &conn->list, obj_list) {
863 if (mark->group == group &&
864 (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
865 fsnotify_get_mark(mark);
866 spin_unlock(&conn->lock);
870 spin_unlock(&conn->lock);
873 EXPORT_SYMBOL_GPL(fsnotify_find_mark);
875 /* Clear any marks in a group with given type mask */
876 void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
877 unsigned int obj_type)
879 struct fsnotify_mark *lmark, *mark;
881 struct list_head *head = &to_free;
883 /* Skip selection step if we want to clear all marks. */
884 if (obj_type == FSNOTIFY_OBJ_TYPE_ANY) {
885 head = &group->marks_list;
889 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
890 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
891 * to_free list so we have to use mark_mutex even when accessing that
892 * list. And freeing mark requires us to drop mark_mutex. So we can
893 * reliably free only the first mark in the list. That's why we first
894 * move marks to free to to_free list in one go and then free marks in
895 * to_free list one by one.
897 fsnotify_group_lock(group);
898 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
899 if (mark->connector->type == obj_type)
900 list_move(&mark->g_list, &to_free);
902 fsnotify_group_unlock(group);
906 fsnotify_group_lock(group);
907 if (list_empty(head)) {
908 fsnotify_group_unlock(group);
911 mark = list_first_entry(head, struct fsnotify_mark, g_list);
912 fsnotify_get_mark(mark);
913 fsnotify_detach_mark(mark);
914 fsnotify_group_unlock(group);
915 fsnotify_free_mark(mark);
916 fsnotify_put_mark(mark);
920 /* Destroy all marks attached to an object via connector */
921 void fsnotify_destroy_marks(fsnotify_connp_t *connp)
923 struct fsnotify_mark_connector *conn;
924 struct fsnotify_mark *mark, *old_mark = NULL;
928 conn = fsnotify_grab_connector(connp);
932 * We have to be careful since we can race with e.g.
933 * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the
934 * list can get modified. However we are holding mark reference and
935 * thus our mark cannot be removed from obj_list so we can continue
936 * iteration after regaining conn->lock.
938 hlist_for_each_entry(mark, &conn->list, obj_list) {
939 fsnotify_get_mark(mark);
940 spin_unlock(&conn->lock);
942 fsnotify_put_mark(old_mark);
944 fsnotify_destroy_mark(mark, mark->group);
945 spin_lock(&conn->lock);
948 * Detach list from object now so that we don't pin inode until all
949 * mark references get dropped. It would lead to strange results such
950 * as delaying inode deletion or blocking unmount.
952 objp = fsnotify_detach_connector_from_object(conn, &type);
953 spin_unlock(&conn->lock);
955 fsnotify_put_mark(old_mark);
956 fsnotify_drop_object(type, objp);
960 * Nothing fancy, just initialize lists and locks and counters.
962 void fsnotify_init_mark(struct fsnotify_mark *mark,
963 struct fsnotify_group *group)
965 memset(mark, 0, sizeof(*mark));
966 spin_lock_init(&mark->lock);
967 refcount_set(&mark->refcnt, 1);
968 fsnotify_get_group(group);
970 WRITE_ONCE(mark->connector, NULL);
972 EXPORT_SYMBOL_GPL(fsnotify_init_mark);
975 * Destroy all marks in destroy_list, waits for SRCU period to finish before
976 * actually freeing marks.
978 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
980 struct fsnotify_mark *mark, *next;
981 struct list_head private_destroy_list;
983 spin_lock(&destroy_lock);
984 /* exchange the list head */
985 list_replace_init(&destroy_list, &private_destroy_list);
986 spin_unlock(&destroy_lock);
988 synchronize_srcu(&fsnotify_mark_srcu);
990 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
991 list_del_init(&mark->g_list);
992 fsnotify_final_mark_destroy(mark);
996 /* Wait for all marks queued for destruction to be actually destroyed */
997 void fsnotify_wait_marks_destroyed(void)
999 flush_delayed_work(&reaper_work);
1001 EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed);