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 *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 if (atomic_long_dec_and_test(fsnotify_sb_watched_objects(sb)))
142 wake_up_var(fsnotify_sb_watched_objects(sb));
145 static void fsnotify_get_inode_ref(struct inode *inode)
148 fsnotify_get_sb_watched_objects(inode->i_sb);
151 static void fsnotify_put_inode_ref(struct inode *inode)
153 fsnotify_put_sb_watched_objects(inode->i_sb);
158 * Grab or drop watched objects reference depending on whether the connector
159 * is attached and has any marks attached.
161 static void fsnotify_update_sb_watchers(struct super_block *sb,
162 struct fsnotify_mark_connector *conn)
164 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
165 bool is_watched = conn->flags & FSNOTIFY_CONN_FLAG_IS_WATCHED;
166 struct fsnotify_mark *first_mark = NULL;
167 unsigned int highest_prio = 0;
170 first_mark = hlist_entry_safe(conn->list.first,
171 struct fsnotify_mark, obj_list);
173 highest_prio = first_mark->group->priority;
174 if (WARN_ON(highest_prio >= __FSNOTIFY_PRIO_NUM))
178 * If the highest priority of group watching this object is prio,
179 * then watched object has a reference on counters [0..prio].
180 * Update priority >= 1 watched objects counters.
182 for (unsigned int p = conn->prio + 1; p <= highest_prio; p++)
183 atomic_long_inc(&sbinfo->watched_objects[p]);
184 for (unsigned int p = conn->prio; p > highest_prio; p--)
185 atomic_long_dec(&sbinfo->watched_objects[p]);
186 conn->prio = highest_prio;
188 /* Update priority >= 0 (a.k.a total) watched objects counter */
189 BUILD_BUG_ON(FSNOTIFY_PRIO_NORMAL != 0);
190 if (first_mark && !is_watched) {
191 conn->flags |= FSNOTIFY_CONN_FLAG_IS_WATCHED;
192 fsnotify_get_sb_watched_objects(sb);
193 } else if (!first_mark && is_watched) {
194 conn->flags &= ~FSNOTIFY_CONN_FLAG_IS_WATCHED;
195 fsnotify_put_sb_watched_objects(sb);
200 * Grab or drop inode reference for the connector if needed.
202 * When it's time to drop the reference, we only clear the HAS_IREF flag and
203 * return the inode object. fsnotify_drop_object() will be resonsible for doing
204 * iput() outside of spinlocks. This happens when last mark that wanted iref is
207 static struct inode *fsnotify_update_iref(struct fsnotify_mark_connector *conn,
210 bool has_iref = conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF;
211 struct inode *inode = NULL;
213 if (conn->type != FSNOTIFY_OBJ_TYPE_INODE ||
214 want_iref == has_iref)
218 /* Pin inode if any mark wants inode refcount held */
219 fsnotify_get_inode_ref(fsnotify_conn_inode(conn));
220 conn->flags |= FSNOTIFY_CONN_FLAG_HAS_IREF;
222 /* Unpin inode after detach of last mark that wanted iref */
223 inode = fsnotify_conn_inode(conn);
224 conn->flags &= ~FSNOTIFY_CONN_FLAG_HAS_IREF;
230 static void *__fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
233 bool want_iref = false;
234 struct fsnotify_mark *mark;
236 assert_spin_locked(&conn->lock);
237 /* We can get detached connector here when inode is getting unlinked. */
238 if (!fsnotify_valid_obj_type(conn->type))
240 hlist_for_each_entry(mark, &conn->list, obj_list) {
241 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED))
243 new_mask |= fsnotify_calc_mask(mark);
244 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE &&
245 !(mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
248 *fsnotify_conn_mask_p(conn) = new_mask;
250 return fsnotify_update_iref(conn, want_iref);
254 * Calculate mask of events for a list of marks. The caller must make sure
255 * connector and connector->obj cannot disappear under us. Callers achieve
256 * this by holding a mark->lock or mark->group->mark_mutex for a mark on this
259 void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
264 spin_lock(&conn->lock);
265 __fsnotify_recalc_mask(conn);
266 spin_unlock(&conn->lock);
267 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
268 __fsnotify_update_child_dentry_flags(
269 fsnotify_conn_inode(conn));
272 /* Free all connectors queued for freeing once SRCU period ends */
273 static void fsnotify_connector_destroy_workfn(struct work_struct *work)
275 struct fsnotify_mark_connector *conn, *free;
277 spin_lock(&destroy_lock);
278 conn = connector_destroy_list;
279 connector_destroy_list = NULL;
280 spin_unlock(&destroy_lock);
282 synchronize_srcu(&fsnotify_mark_srcu);
285 conn = conn->destroy_next;
286 kmem_cache_free(fsnotify_mark_connector_cachep, free);
290 static void *fsnotify_detach_connector_from_object(
291 struct fsnotify_mark_connector *conn,
294 fsnotify_connp_t *connp = fsnotify_object_connp(conn->obj, conn->type);
295 struct super_block *sb = fsnotify_connector_sb(conn);
296 struct inode *inode = NULL;
299 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED)
302 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
303 inode = fsnotify_conn_inode(conn);
304 inode->i_fsnotify_mask = 0;
306 /* Unpin inode when detaching from connector */
307 if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF))
309 } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) {
310 fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
311 } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
312 fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
315 rcu_assign_pointer(*connp, NULL);
317 conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
318 fsnotify_update_sb_watchers(sb, conn);
323 static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark)
325 struct fsnotify_group *group = mark->group;
327 if (WARN_ON_ONCE(!group))
329 group->ops->free_mark(mark);
330 fsnotify_put_group(group);
333 /* Drop object reference originally held by a connector */
334 static void fsnotify_drop_object(unsigned int type, void *objp)
338 /* Currently only inode references are passed to be dropped */
339 if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE))
341 fsnotify_put_inode_ref(objp);
344 void fsnotify_put_mark(struct fsnotify_mark *mark)
346 struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector);
348 unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED;
349 bool free_conn = false;
351 /* Catch marks that were actually never attached to object */
353 if (refcount_dec_and_test(&mark->refcnt))
354 fsnotify_final_mark_destroy(mark);
359 * We have to be careful so that traversals of obj_list under lock can
360 * safely grab mark reference.
362 if (!refcount_dec_and_lock(&mark->refcnt, &conn->lock))
365 hlist_del_init_rcu(&mark->obj_list);
366 if (hlist_empty(&conn->list)) {
367 objp = fsnotify_detach_connector_from_object(conn, &type);
370 struct super_block *sb = fsnotify_connector_sb(conn);
372 /* Update watched objects after detaching mark */
374 fsnotify_update_sb_watchers(sb, conn);
375 objp = __fsnotify_recalc_mask(conn);
378 WRITE_ONCE(mark->connector, NULL);
379 spin_unlock(&conn->lock);
381 fsnotify_drop_object(type, objp);
384 spin_lock(&destroy_lock);
385 conn->destroy_next = connector_destroy_list;
386 connector_destroy_list = conn;
387 spin_unlock(&destroy_lock);
388 queue_work(system_unbound_wq, &connector_reaper_work);
391 * Note that we didn't update flags telling whether inode cares about
392 * what's happening with children. We update these flags from
393 * __fsnotify_parent() lazily when next event happens on one of our
396 spin_lock(&destroy_lock);
397 list_add(&mark->g_list, &destroy_list);
398 spin_unlock(&destroy_lock);
399 queue_delayed_work(system_unbound_wq, &reaper_work,
400 FSNOTIFY_REAPER_DELAY);
402 EXPORT_SYMBOL_GPL(fsnotify_put_mark);
405 * Get mark reference when we found the mark via lockless traversal of object
406 * list. Mark can be already removed from the list by now and on its way to be
407 * destroyed once SRCU period ends.
409 * Also pin the group so it doesn't disappear under us.
411 static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark)
416 if (refcount_inc_not_zero(&mark->refcnt)) {
417 spin_lock(&mark->lock);
418 if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) {
419 /* mark is attached, group is still alive then */
420 atomic_inc(&mark->group->user_waits);
421 spin_unlock(&mark->lock);
424 spin_unlock(&mark->lock);
425 fsnotify_put_mark(mark);
431 * Puts marks and wakes up group destruction if necessary.
433 * Pairs with fsnotify_get_mark_safe()
435 static void fsnotify_put_mark_wake(struct fsnotify_mark *mark)
438 struct fsnotify_group *group = mark->group;
440 fsnotify_put_mark(mark);
442 * We abuse notification_waitq on group shutdown for waiting for
443 * all marks pinned when waiting for userspace.
445 if (atomic_dec_and_test(&group->user_waits) && group->shutdown)
446 wake_up(&group->notification_waitq);
450 bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info)
451 __releases(&fsnotify_mark_srcu)
455 fsnotify_foreach_iter_type(type) {
456 /* This can fail if mark is being removed */
457 if (!fsnotify_get_mark_safe(iter_info->marks[type])) {
458 __release(&fsnotify_mark_srcu);
464 * Now that both marks are pinned by refcount in the inode / vfsmount
465 * lists, we can drop SRCU lock, and safely resume the list iteration
466 * once userspace returns.
468 srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx);
473 for (type--; type >= 0; type--)
474 fsnotify_put_mark_wake(iter_info->marks[type]);
478 void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info)
479 __acquires(&fsnotify_mark_srcu)
483 iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
484 fsnotify_foreach_iter_type(type)
485 fsnotify_put_mark_wake(iter_info->marks[type]);
489 * Mark mark as detached, remove it from group list. Mark still stays in object
490 * list until its last reference is dropped. Note that we rely on mark being
491 * removed from group list before corresponding reference to it is dropped. In
492 * particular we rely on mark->connector being valid while we hold
493 * group->mark_mutex if we found the mark through g_list.
495 * Must be called with group->mark_mutex held. The caller must either hold
496 * reference to the mark or be protected by fsnotify_mark_srcu.
498 void fsnotify_detach_mark(struct fsnotify_mark *mark)
500 fsnotify_group_assert_locked(mark->group);
501 WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) &&
502 refcount_read(&mark->refcnt) < 1 +
503 !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED));
505 spin_lock(&mark->lock);
506 /* something else already called this function on this mark */
507 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
508 spin_unlock(&mark->lock);
511 mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
512 list_del_init(&mark->g_list);
513 spin_unlock(&mark->lock);
515 /* Drop mark reference acquired in fsnotify_add_mark_locked() */
516 fsnotify_put_mark(mark);
520 * Free fsnotify mark. The mark is actually only marked as being freed. The
521 * freeing is actually happening only once last reference to the mark is
522 * dropped from a workqueue which first waits for srcu period end.
524 * Caller must have a reference to the mark or be protected by
525 * fsnotify_mark_srcu.
527 void fsnotify_free_mark(struct fsnotify_mark *mark)
529 struct fsnotify_group *group = mark->group;
531 spin_lock(&mark->lock);
532 /* something else already called this function on this mark */
533 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
534 spin_unlock(&mark->lock);
537 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
538 spin_unlock(&mark->lock);
541 * Some groups like to know that marks are being freed. This is a
542 * callback to the group function to let it know that this mark
545 if (group->ops->freeing_mark)
546 group->ops->freeing_mark(mark, group);
549 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
550 struct fsnotify_group *group)
552 fsnotify_group_lock(group);
553 fsnotify_detach_mark(mark);
554 fsnotify_group_unlock(group);
555 fsnotify_free_mark(mark);
557 EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
560 * Sorting function for lists of fsnotify marks.
562 * Fanotify supports different notification classes (reflected as priority of
563 * notification group). Events shall be passed to notification groups in
564 * decreasing priority order. To achieve this marks in notification lists for
565 * inodes and vfsmounts are sorted so that priorities of corresponding groups
568 * Furthermore correct handling of the ignore mask requires processing inode
569 * and vfsmount marks of each group together. Using the group address as
570 * further sort criterion provides a unique sorting order and thus we can
571 * merge inode and vfsmount lists of marks in linear time and find groups
572 * present in both lists.
574 * A return value of 1 signifies that b has priority over a.
575 * A return value of 0 signifies that the two marks have to be handled together.
576 * A return value of -1 signifies that a has priority over b.
578 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
586 if (a->priority < b->priority)
588 if (a->priority > b->priority)
595 static int fsnotify_attach_info_to_sb(struct super_block *sb)
597 struct fsnotify_sb_info *sbinfo;
599 /* sb info is freed on fsnotify_sb_delete() */
600 sbinfo = kzalloc(sizeof(*sbinfo), GFP_KERNEL);
605 * cmpxchg() provides the barrier so that callers of fsnotify_sb_info()
606 * will observe an initialized structure
608 if (cmpxchg(&sb->s_fsnotify_info, NULL, sbinfo)) {
609 /* Someone else created sbinfo for us */
615 static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
616 void *obj, unsigned int obj_type)
618 struct fsnotify_mark_connector *conn;
620 conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL);
623 spin_lock_init(&conn->lock);
624 INIT_HLIST_HEAD(&conn->list);
627 conn->type = obj_type;
631 * cmpxchg() provides the barrier so that readers of *connp can see
632 * only initialized structure
634 if (cmpxchg(connp, NULL, conn)) {
635 /* Someone else created list structure for us */
636 kmem_cache_free(fsnotify_mark_connector_cachep, conn);
642 * Get mark connector, make sure it is alive and return with its lock held.
643 * This is for users that get connector pointer from inode or mount. Users that
644 * hold reference to a mark on the list may directly lock connector->lock as
645 * they are sure list cannot go away under them.
647 static struct fsnotify_mark_connector *fsnotify_grab_connector(
648 fsnotify_connp_t *connp)
650 struct fsnotify_mark_connector *conn;
653 idx = srcu_read_lock(&fsnotify_mark_srcu);
654 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
657 spin_lock(&conn->lock);
658 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) {
659 spin_unlock(&conn->lock);
660 srcu_read_unlock(&fsnotify_mark_srcu, idx);
664 srcu_read_unlock(&fsnotify_mark_srcu, idx);
669 * Add mark into proper place in given list of marks. These marks may be used
670 * for the fsnotify backend to determine which event types should be delivered
671 * to which group and for which inodes. These marks are ordered according to
672 * priority, highest number first, and then by the group's location in memory.
674 static int fsnotify_add_mark_list(struct fsnotify_mark *mark, void *obj,
675 unsigned int obj_type, int add_flags)
677 struct super_block *sb = fsnotify_object_sb(obj, obj_type);
678 struct fsnotify_mark *lmark, *last = NULL;
679 struct fsnotify_mark_connector *conn;
680 fsnotify_connp_t *connp;
684 if (WARN_ON(!fsnotify_valid_obj_type(obj_type)))
688 * Attach the sb info before attaching a connector to any object on sb.
689 * The sb info will remain attached as long as sb lives.
691 if (!fsnotify_sb_info(sb)) {
692 err = fsnotify_attach_info_to_sb(sb);
697 connp = fsnotify_object_connp(obj, obj_type);
699 spin_lock(&mark->lock);
700 conn = fsnotify_grab_connector(connp);
702 spin_unlock(&mark->lock);
703 err = fsnotify_attach_connector_to_object(connp, obj, obj_type);
709 /* is mark the first mark? */
710 if (hlist_empty(&conn->list)) {
711 hlist_add_head_rcu(&mark->obj_list, &conn->list);
715 /* should mark be in the middle of the current list? */
716 hlist_for_each_entry(lmark, &conn->list, obj_list) {
719 if ((lmark->group == mark->group) &&
720 (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) &&
721 !(mark->group->flags & FSNOTIFY_GROUP_DUPS)) {
726 cmp = fsnotify_compare_groups(lmark->group, mark->group);
728 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
733 BUG_ON(last == NULL);
734 /* mark should be the last entry. last is the current last entry */
735 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
737 fsnotify_update_sb_watchers(sb, conn);
739 * Since connector is attached to object using cmpxchg() we are
740 * guaranteed that connector initialization is fully visible by anyone
741 * seeing mark->connector set.
743 WRITE_ONCE(mark->connector, conn);
745 spin_unlock(&conn->lock);
746 spin_unlock(&mark->lock);
751 * Attach an initialized mark to a given group and fs object.
752 * These marks may be used for the fsnotify backend to determine which
753 * event types should be delivered to which group.
755 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
756 void *obj, unsigned int obj_type,
759 struct fsnotify_group *group = mark->group;
762 fsnotify_group_assert_locked(group);
768 * mark->connector->lock
770 spin_lock(&mark->lock);
771 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
773 list_add(&mark->g_list, &group->marks_list);
774 fsnotify_get_mark(mark); /* for g_list */
775 spin_unlock(&mark->lock);
777 ret = fsnotify_add_mark_list(mark, obj, obj_type, add_flags);
781 fsnotify_recalc_mask(mark->connector);
785 spin_lock(&mark->lock);
786 mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE |
787 FSNOTIFY_MARK_FLAG_ATTACHED);
788 list_del_init(&mark->g_list);
789 spin_unlock(&mark->lock);
791 fsnotify_put_mark(mark);
795 int fsnotify_add_mark(struct fsnotify_mark *mark, void *obj,
796 unsigned int obj_type, int add_flags)
799 struct fsnotify_group *group = mark->group;
801 fsnotify_group_lock(group);
802 ret = fsnotify_add_mark_locked(mark, obj, obj_type, add_flags);
803 fsnotify_group_unlock(group);
806 EXPORT_SYMBOL_GPL(fsnotify_add_mark);
809 * Given a list of marks, find the mark associated with given group. If found
810 * take a reference to that mark and return it, else return NULL.
812 struct fsnotify_mark *fsnotify_find_mark(void *obj, unsigned int obj_type,
813 struct fsnotify_group *group)
815 fsnotify_connp_t *connp = fsnotify_object_connp(obj, obj_type);
816 struct fsnotify_mark_connector *conn;
817 struct fsnotify_mark *mark;
822 conn = fsnotify_grab_connector(connp);
826 hlist_for_each_entry(mark, &conn->list, obj_list) {
827 if (mark->group == group &&
828 (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
829 fsnotify_get_mark(mark);
830 spin_unlock(&conn->lock);
834 spin_unlock(&conn->lock);
837 EXPORT_SYMBOL_GPL(fsnotify_find_mark);
839 /* Clear any marks in a group with given type mask */
840 void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
841 unsigned int obj_type)
843 struct fsnotify_mark *lmark, *mark;
845 struct list_head *head = &to_free;
847 /* Skip selection step if we want to clear all marks. */
848 if (obj_type == FSNOTIFY_OBJ_TYPE_ANY) {
849 head = &group->marks_list;
853 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
854 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
855 * to_free list so we have to use mark_mutex even when accessing that
856 * list. And freeing mark requires us to drop mark_mutex. So we can
857 * reliably free only the first mark in the list. That's why we first
858 * move marks to free to to_free list in one go and then free marks in
859 * to_free list one by one.
861 fsnotify_group_lock(group);
862 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
863 if (mark->connector->type == obj_type)
864 list_move(&mark->g_list, &to_free);
866 fsnotify_group_unlock(group);
870 fsnotify_group_lock(group);
871 if (list_empty(head)) {
872 fsnotify_group_unlock(group);
875 mark = list_first_entry(head, struct fsnotify_mark, g_list);
876 fsnotify_get_mark(mark);
877 fsnotify_detach_mark(mark);
878 fsnotify_group_unlock(group);
879 fsnotify_free_mark(mark);
880 fsnotify_put_mark(mark);
884 /* Destroy all marks attached to an object via connector */
885 void fsnotify_destroy_marks(fsnotify_connp_t *connp)
887 struct fsnotify_mark_connector *conn;
888 struct fsnotify_mark *mark, *old_mark = NULL;
892 conn = fsnotify_grab_connector(connp);
896 * We have to be careful since we can race with e.g.
897 * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the
898 * list can get modified. However we are holding mark reference and
899 * thus our mark cannot be removed from obj_list so we can continue
900 * iteration after regaining conn->lock.
902 hlist_for_each_entry(mark, &conn->list, obj_list) {
903 fsnotify_get_mark(mark);
904 spin_unlock(&conn->lock);
906 fsnotify_put_mark(old_mark);
908 fsnotify_destroy_mark(mark, mark->group);
909 spin_lock(&conn->lock);
912 * Detach list from object now so that we don't pin inode until all
913 * mark references get dropped. It would lead to strange results such
914 * as delaying inode deletion or blocking unmount.
916 objp = fsnotify_detach_connector_from_object(conn, &type);
917 spin_unlock(&conn->lock);
919 fsnotify_put_mark(old_mark);
920 fsnotify_drop_object(type, objp);
924 * Nothing fancy, just initialize lists and locks and counters.
926 void fsnotify_init_mark(struct fsnotify_mark *mark,
927 struct fsnotify_group *group)
929 memset(mark, 0, sizeof(*mark));
930 spin_lock_init(&mark->lock);
931 refcount_set(&mark->refcnt, 1);
932 fsnotify_get_group(group);
934 WRITE_ONCE(mark->connector, NULL);
936 EXPORT_SYMBOL_GPL(fsnotify_init_mark);
939 * Destroy all marks in destroy_list, waits for SRCU period to finish before
940 * actually freeing marks.
942 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
944 struct fsnotify_mark *mark, *next;
945 struct list_head private_destroy_list;
947 spin_lock(&destroy_lock);
948 /* exchange the list head */
949 list_replace_init(&destroy_list, &private_destroy_list);
950 spin_unlock(&destroy_lock);
952 synchronize_srcu(&fsnotify_mark_srcu);
954 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
955 list_del_init(&mark->g_list);
956 fsnotify_final_mark_destroy(mark);
960 /* Wait for all marks queued for destruction to be actually destroyed */
961 void fsnotify_wait_marks_destroyed(void)
963 flush_delayed_work(&reaper_work);
965 EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed);