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 |= fsnotify_calc_mask(mark);
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_get_inode_ref(struct inode *inode)
175 atomic_long_inc(&inode->i_sb->s_fsnotify_connectors);
178 static void fsnotify_put_inode_ref(struct inode *inode)
180 struct super_block *sb = inode->i_sb;
183 if (atomic_long_dec_and_test(&sb->s_fsnotify_connectors))
184 wake_up_var(&sb->s_fsnotify_connectors);
187 static void fsnotify_get_sb_connectors(struct fsnotify_mark_connector *conn)
189 struct super_block *sb = fsnotify_connector_sb(conn);
192 atomic_long_inc(&sb->s_fsnotify_connectors);
195 static void fsnotify_put_sb_connectors(struct fsnotify_mark_connector *conn)
197 struct super_block *sb = fsnotify_connector_sb(conn);
199 if (sb && atomic_long_dec_and_test(&sb->s_fsnotify_connectors))
200 wake_up_var(&sb->s_fsnotify_connectors);
203 static void *fsnotify_detach_connector_from_object(
204 struct fsnotify_mark_connector *conn,
207 struct inode *inode = NULL;
210 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED)
213 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
214 inode = fsnotify_conn_inode(conn);
215 inode->i_fsnotify_mask = 0;
216 } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) {
217 fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
218 } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
219 fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
222 fsnotify_put_sb_connectors(conn);
223 rcu_assign_pointer(*(conn->obj), NULL);
225 conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
230 static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark)
232 struct fsnotify_group *group = mark->group;
234 if (WARN_ON_ONCE(!group))
236 group->ops->free_mark(mark);
237 fsnotify_put_group(group);
240 /* Drop object reference originally held by a connector */
241 static void fsnotify_drop_object(unsigned int type, void *objp)
245 /* Currently only inode references are passed to be dropped */
246 if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE))
248 fsnotify_put_inode_ref(objp);
251 void fsnotify_put_mark(struct fsnotify_mark *mark)
253 struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector);
255 unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED;
256 bool free_conn = false;
258 /* Catch marks that were actually never attached to object */
260 if (refcount_dec_and_test(&mark->refcnt))
261 fsnotify_final_mark_destroy(mark);
266 * We have to be careful so that traversals of obj_list under lock can
267 * safely grab mark reference.
269 if (!refcount_dec_and_lock(&mark->refcnt, &conn->lock))
272 hlist_del_init_rcu(&mark->obj_list);
273 if (hlist_empty(&conn->list)) {
274 objp = fsnotify_detach_connector_from_object(conn, &type);
277 __fsnotify_recalc_mask(conn);
279 WRITE_ONCE(mark->connector, NULL);
280 spin_unlock(&conn->lock);
282 fsnotify_drop_object(type, objp);
285 spin_lock(&destroy_lock);
286 conn->destroy_next = connector_destroy_list;
287 connector_destroy_list = conn;
288 spin_unlock(&destroy_lock);
289 queue_work(system_unbound_wq, &connector_reaper_work);
292 * Note that we didn't update flags telling whether inode cares about
293 * what's happening with children. We update these flags from
294 * __fsnotify_parent() lazily when next event happens on one of our
297 spin_lock(&destroy_lock);
298 list_add(&mark->g_list, &destroy_list);
299 spin_unlock(&destroy_lock);
300 queue_delayed_work(system_unbound_wq, &reaper_work,
301 FSNOTIFY_REAPER_DELAY);
303 EXPORT_SYMBOL_GPL(fsnotify_put_mark);
306 * Get mark reference when we found the mark via lockless traversal of object
307 * list. Mark can be already removed from the list by now and on its way to be
308 * destroyed once SRCU period ends.
310 * Also pin the group so it doesn't disappear under us.
312 static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark)
317 if (refcount_inc_not_zero(&mark->refcnt)) {
318 spin_lock(&mark->lock);
319 if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) {
320 /* mark is attached, group is still alive then */
321 atomic_inc(&mark->group->user_waits);
322 spin_unlock(&mark->lock);
325 spin_unlock(&mark->lock);
326 fsnotify_put_mark(mark);
332 * Puts marks and wakes up group destruction if necessary.
334 * Pairs with fsnotify_get_mark_safe()
336 static void fsnotify_put_mark_wake(struct fsnotify_mark *mark)
339 struct fsnotify_group *group = mark->group;
341 fsnotify_put_mark(mark);
343 * We abuse notification_waitq on group shutdown for waiting for
344 * all marks pinned when waiting for userspace.
346 if (atomic_dec_and_test(&group->user_waits) && group->shutdown)
347 wake_up(&group->notification_waitq);
351 bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info)
352 __releases(&fsnotify_mark_srcu)
356 fsnotify_foreach_iter_type(type) {
357 /* This can fail if mark is being removed */
358 if (!fsnotify_get_mark_safe(iter_info->marks[type])) {
359 __release(&fsnotify_mark_srcu);
365 * Now that both marks are pinned by refcount in the inode / vfsmount
366 * lists, we can drop SRCU lock, and safely resume the list iteration
367 * once userspace returns.
369 srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx);
374 for (type--; type >= 0; type--)
375 fsnotify_put_mark_wake(iter_info->marks[type]);
379 void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info)
380 __acquires(&fsnotify_mark_srcu)
384 iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
385 fsnotify_foreach_iter_type(type)
386 fsnotify_put_mark_wake(iter_info->marks[type]);
390 * Mark mark as detached, remove it from group list. Mark still stays in object
391 * list until its last reference is dropped. Note that we rely on mark being
392 * removed from group list before corresponding reference to it is dropped. In
393 * particular we rely on mark->connector being valid while we hold
394 * group->mark_mutex if we found the mark through g_list.
396 * Must be called with group->mark_mutex held. The caller must either hold
397 * reference to the mark or be protected by fsnotify_mark_srcu.
399 void fsnotify_detach_mark(struct fsnotify_mark *mark)
401 struct fsnotify_group *group = mark->group;
403 WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex));
404 WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) &&
405 refcount_read(&mark->refcnt) < 1 +
406 !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED));
408 spin_lock(&mark->lock);
409 /* something else already called this function on this mark */
410 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
411 spin_unlock(&mark->lock);
414 mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
415 list_del_init(&mark->g_list);
416 spin_unlock(&mark->lock);
418 /* Drop mark reference acquired in fsnotify_add_mark_locked() */
419 fsnotify_put_mark(mark);
423 * Free fsnotify mark. The mark is actually only marked as being freed. The
424 * freeing is actually happening only once last reference to the mark is
425 * dropped from a workqueue which first waits for srcu period end.
427 * Caller must have a reference to the mark or be protected by
428 * fsnotify_mark_srcu.
430 void fsnotify_free_mark(struct fsnotify_mark *mark)
432 struct fsnotify_group *group = mark->group;
434 spin_lock(&mark->lock);
435 /* something else already called this function on this mark */
436 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
437 spin_unlock(&mark->lock);
440 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
441 spin_unlock(&mark->lock);
444 * Some groups like to know that marks are being freed. This is a
445 * callback to the group function to let it know that this mark
448 if (group->ops->freeing_mark)
449 group->ops->freeing_mark(mark, group);
452 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
453 struct fsnotify_group *group)
455 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
456 fsnotify_detach_mark(mark);
457 mutex_unlock(&group->mark_mutex);
458 fsnotify_free_mark(mark);
460 EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
463 * Sorting function for lists of fsnotify marks.
465 * Fanotify supports different notification classes (reflected as priority of
466 * notification group). Events shall be passed to notification groups in
467 * decreasing priority order. To achieve this marks in notification lists for
468 * inodes and vfsmounts are sorted so that priorities of corresponding groups
471 * Furthermore correct handling of the ignore mask requires processing inode
472 * and vfsmount marks of each group together. Using the group address as
473 * further sort criterion provides a unique sorting order and thus we can
474 * merge inode and vfsmount lists of marks in linear time and find groups
475 * present in both lists.
477 * A return value of 1 signifies that b has priority over a.
478 * A return value of 0 signifies that the two marks have to be handled together.
479 * A return value of -1 signifies that a has priority over b.
481 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
489 if (a->priority < b->priority)
491 if (a->priority > b->priority)
498 static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
499 unsigned int obj_type,
500 __kernel_fsid_t *fsid)
502 struct inode *inode = NULL;
503 struct fsnotify_mark_connector *conn;
505 conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL);
508 spin_lock_init(&conn->lock);
509 INIT_HLIST_HEAD(&conn->list);
510 conn->type = obj_type;
512 /* Cache fsid of filesystem containing the object */
515 conn->flags = FSNOTIFY_CONN_FLAG_HAS_FSID;
517 conn->fsid.val[0] = conn->fsid.val[1] = 0;
520 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
521 inode = fsnotify_conn_inode(conn);
522 fsnotify_get_inode_ref(inode);
524 fsnotify_get_sb_connectors(conn);
527 * cmpxchg() provides the barrier so that readers of *connp can see
528 * only initialized structure
530 if (cmpxchg(connp, NULL, conn)) {
531 /* Someone else created list structure for us */
533 fsnotify_put_inode_ref(inode);
534 fsnotify_put_sb_connectors(conn);
535 kmem_cache_free(fsnotify_mark_connector_cachep, conn);
542 * Get mark connector, make sure it is alive and return with its lock held.
543 * This is for users that get connector pointer from inode or mount. Users that
544 * hold reference to a mark on the list may directly lock connector->lock as
545 * they are sure list cannot go away under them.
547 static struct fsnotify_mark_connector *fsnotify_grab_connector(
548 fsnotify_connp_t *connp)
550 struct fsnotify_mark_connector *conn;
553 idx = srcu_read_lock(&fsnotify_mark_srcu);
554 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
557 spin_lock(&conn->lock);
558 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) {
559 spin_unlock(&conn->lock);
560 srcu_read_unlock(&fsnotify_mark_srcu, idx);
564 srcu_read_unlock(&fsnotify_mark_srcu, idx);
569 * Add mark into proper place in given list of marks. These marks may be used
570 * for the fsnotify backend to determine which event types should be delivered
571 * to which group and for which inodes. These marks are ordered according to
572 * priority, highest number first, and then by the group's location in memory.
574 static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
575 fsnotify_connp_t *connp,
576 unsigned int obj_type,
577 int allow_dups, __kernel_fsid_t *fsid)
579 struct fsnotify_mark *lmark, *last = NULL;
580 struct fsnotify_mark_connector *conn;
584 if (WARN_ON(!fsnotify_valid_obj_type(obj_type)))
587 /* Backend is expected to check for zero fsid (e.g. tmpfs) */
588 if (fsid && WARN_ON_ONCE(!fsid->val[0] && !fsid->val[1]))
592 spin_lock(&mark->lock);
593 conn = fsnotify_grab_connector(connp);
595 spin_unlock(&mark->lock);
596 err = fsnotify_attach_connector_to_object(connp, obj_type,
601 } else if (fsid && !(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID)) {
603 /* Pairs with smp_rmb() in fanotify_get_fsid() */
605 conn->flags |= FSNOTIFY_CONN_FLAG_HAS_FSID;
606 } else if (fsid && (conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID) &&
607 (fsid->val[0] != conn->fsid.val[0] ||
608 fsid->val[1] != conn->fsid.val[1])) {
610 * Backend is expected to check for non uniform fsid
611 * (e.g. btrfs), but maybe we missed something?
612 * Only allow setting conn->fsid once to non zero fsid.
613 * inotify and non-fid fanotify groups do not set nor test
616 pr_warn_ratelimited("%s: fsid mismatch on object of type %u: "
617 "%x.%x != %x.%x\n", __func__, conn->type,
618 fsid->val[0], fsid->val[1],
619 conn->fsid.val[0], conn->fsid.val[1]);
624 /* is mark the first mark? */
625 if (hlist_empty(&conn->list)) {
626 hlist_add_head_rcu(&mark->obj_list, &conn->list);
630 /* should mark be in the middle of the current list? */
631 hlist_for_each_entry(lmark, &conn->list, obj_list) {
634 if ((lmark->group == mark->group) &&
635 (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) &&
641 cmp = fsnotify_compare_groups(lmark->group, mark->group);
643 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
648 BUG_ON(last == NULL);
649 /* mark should be the last entry. last is the current last entry */
650 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
653 * Since connector is attached to object using cmpxchg() we are
654 * guaranteed that connector initialization is fully visible by anyone
655 * seeing mark->connector set.
657 WRITE_ONCE(mark->connector, conn);
659 spin_unlock(&conn->lock);
660 spin_unlock(&mark->lock);
665 * Attach an initialized mark to a given group and fs object.
666 * These marks may be used for the fsnotify backend to determine which
667 * event types should be delivered to which group.
669 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
670 fsnotify_connp_t *connp, unsigned int obj_type,
671 int allow_dups, __kernel_fsid_t *fsid)
673 struct fsnotify_group *group = mark->group;
676 BUG_ON(!mutex_is_locked(&group->mark_mutex));
682 * mark->connector->lock
684 spin_lock(&mark->lock);
685 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
687 list_add(&mark->g_list, &group->marks_list);
688 fsnotify_get_mark(mark); /* for g_list */
689 spin_unlock(&mark->lock);
691 ret = fsnotify_add_mark_list(mark, connp, obj_type, allow_dups, fsid);
695 if (mark->mask || mark->ignored_mask)
696 fsnotify_recalc_mask(mark->connector);
700 spin_lock(&mark->lock);
701 mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE |
702 FSNOTIFY_MARK_FLAG_ATTACHED);
703 list_del_init(&mark->g_list);
704 spin_unlock(&mark->lock);
706 fsnotify_put_mark(mark);
710 int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp,
711 unsigned int obj_type, int allow_dups,
712 __kernel_fsid_t *fsid)
715 struct fsnotify_group *group = mark->group;
717 mutex_lock(&group->mark_mutex);
718 ret = fsnotify_add_mark_locked(mark, connp, obj_type, allow_dups, fsid);
719 mutex_unlock(&group->mark_mutex);
722 EXPORT_SYMBOL_GPL(fsnotify_add_mark);
725 * Given a list of marks, find the mark associated with given group. If found
726 * take a reference to that mark and return it, else return NULL.
728 struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
729 struct fsnotify_group *group)
731 struct fsnotify_mark_connector *conn;
732 struct fsnotify_mark *mark;
734 conn = fsnotify_grab_connector(connp);
738 hlist_for_each_entry(mark, &conn->list, obj_list) {
739 if (mark->group == group &&
740 (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
741 fsnotify_get_mark(mark);
742 spin_unlock(&conn->lock);
746 spin_unlock(&conn->lock);
749 EXPORT_SYMBOL_GPL(fsnotify_find_mark);
751 /* Clear any marks in a group with given type mask */
752 void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
753 unsigned int obj_type)
755 struct fsnotify_mark *lmark, *mark;
757 struct list_head *head = &to_free;
759 /* Skip selection step if we want to clear all marks. */
760 if (obj_type == FSNOTIFY_OBJ_TYPE_ANY) {
761 head = &group->marks_list;
765 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
766 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
767 * to_free list so we have to use mark_mutex even when accessing that
768 * list. And freeing mark requires us to drop mark_mutex. So we can
769 * reliably free only the first mark in the list. That's why we first
770 * move marks to free to to_free list in one go and then free marks in
771 * to_free list one by one.
773 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
774 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
775 if (mark->connector->type == obj_type)
776 list_move(&mark->g_list, &to_free);
778 mutex_unlock(&group->mark_mutex);
782 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
783 if (list_empty(head)) {
784 mutex_unlock(&group->mark_mutex);
787 mark = list_first_entry(head, struct fsnotify_mark, g_list);
788 fsnotify_get_mark(mark);
789 fsnotify_detach_mark(mark);
790 mutex_unlock(&group->mark_mutex);
791 fsnotify_free_mark(mark);
792 fsnotify_put_mark(mark);
796 /* Destroy all marks attached to an object via connector */
797 void fsnotify_destroy_marks(fsnotify_connp_t *connp)
799 struct fsnotify_mark_connector *conn;
800 struct fsnotify_mark *mark, *old_mark = NULL;
804 conn = fsnotify_grab_connector(connp);
808 * We have to be careful since we can race with e.g.
809 * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the
810 * list can get modified. However we are holding mark reference and
811 * thus our mark cannot be removed from obj_list so we can continue
812 * iteration after regaining conn->lock.
814 hlist_for_each_entry(mark, &conn->list, obj_list) {
815 fsnotify_get_mark(mark);
816 spin_unlock(&conn->lock);
818 fsnotify_put_mark(old_mark);
820 fsnotify_destroy_mark(mark, mark->group);
821 spin_lock(&conn->lock);
824 * Detach list from object now so that we don't pin inode until all
825 * mark references get dropped. It would lead to strange results such
826 * as delaying inode deletion or blocking unmount.
828 objp = fsnotify_detach_connector_from_object(conn, &type);
829 spin_unlock(&conn->lock);
831 fsnotify_put_mark(old_mark);
832 fsnotify_drop_object(type, objp);
836 * Nothing fancy, just initialize lists and locks and counters.
838 void fsnotify_init_mark(struct fsnotify_mark *mark,
839 struct fsnotify_group *group)
841 memset(mark, 0, sizeof(*mark));
842 spin_lock_init(&mark->lock);
843 refcount_set(&mark->refcnt, 1);
844 fsnotify_get_group(group);
846 WRITE_ONCE(mark->connector, NULL);
848 EXPORT_SYMBOL_GPL(fsnotify_init_mark);
851 * Destroy all marks in destroy_list, waits for SRCU period to finish before
852 * actually freeing marks.
854 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
856 struct fsnotify_mark *mark, *next;
857 struct list_head private_destroy_list;
859 spin_lock(&destroy_lock);
860 /* exchange the list head */
861 list_replace_init(&destroy_list, &private_destroy_list);
862 spin_unlock(&destroy_lock);
864 synchronize_srcu(&fsnotify_mark_srcu);
866 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
867 list_del_init(&mark->g_list);
868 fsnotify_final_mark_destroy(mark);
872 /* Wait for all marks queued for destruction to be actually destroyed */
873 void fsnotify_wait_marks_destroyed(void)
875 flush_delayed_work(&reaper_work);
877 EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed);