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_get_inode_ref(struct inode
*inode
)
122 atomic_long_inc(&inode
->i_sb
->s_fsnotify_connectors
);
126 * Grab or drop inode reference for the connector if needed.
128 * When it's time to drop the reference, we only clear the HAS_IREF flag and
129 * return the inode object. fsnotify_drop_object() will be resonsible for doing
130 * iput() outside of spinlocks. This happens when last mark that wanted iref is
133 static struct inode
*fsnotify_update_iref(struct fsnotify_mark_connector
*conn
,
136 bool has_iref
= conn
->flags
& FSNOTIFY_CONN_FLAG_HAS_IREF
;
137 struct inode
*inode
= NULL
;
139 if (conn
->type
!= FSNOTIFY_OBJ_TYPE_INODE
||
140 want_iref
== has_iref
)
144 /* Pin inode if any mark wants inode refcount held */
145 fsnotify_get_inode_ref(fsnotify_conn_inode(conn
));
146 conn
->flags
|= FSNOTIFY_CONN_FLAG_HAS_IREF
;
148 /* Unpin inode after detach of last mark that wanted iref */
149 inode
= fsnotify_conn_inode(conn
);
150 conn
->flags
&= ~FSNOTIFY_CONN_FLAG_HAS_IREF
;
156 static void *__fsnotify_recalc_mask(struct fsnotify_mark_connector
*conn
)
159 bool want_iref
= false;
160 struct fsnotify_mark
*mark
;
162 assert_spin_locked(&conn
->lock
);
163 /* We can get detached connector here when inode is getting unlinked. */
164 if (!fsnotify_valid_obj_type(conn
->type
))
166 hlist_for_each_entry(mark
, &conn
->list
, obj_list
) {
167 if (!(mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
))
169 new_mask
|= fsnotify_calc_mask(mark
);
170 if (conn
->type
== FSNOTIFY_OBJ_TYPE_INODE
&&
171 !(mark
->flags
& FSNOTIFY_MARK_FLAG_NO_IREF
))
174 *fsnotify_conn_mask_p(conn
) = new_mask
;
176 return fsnotify_update_iref(conn
, want_iref
);
180 * Calculate mask of events for a list of marks. The caller must make sure
181 * connector and connector->obj cannot disappear under us. Callers achieve
182 * this by holding a mark->lock or mark->group->mark_mutex for a mark on this
185 void fsnotify_recalc_mask(struct fsnotify_mark_connector
*conn
)
190 spin_lock(&conn
->lock
);
191 __fsnotify_recalc_mask(conn
);
192 spin_unlock(&conn
->lock
);
193 if (conn
->type
== FSNOTIFY_OBJ_TYPE_INODE
)
194 __fsnotify_update_child_dentry_flags(
195 fsnotify_conn_inode(conn
));
198 /* Free all connectors queued for freeing once SRCU period ends */
199 static void fsnotify_connector_destroy_workfn(struct work_struct
*work
)
201 struct fsnotify_mark_connector
*conn
, *free
;
203 spin_lock(&destroy_lock
);
204 conn
= connector_destroy_list
;
205 connector_destroy_list
= NULL
;
206 spin_unlock(&destroy_lock
);
208 synchronize_srcu(&fsnotify_mark_srcu
);
211 conn
= conn
->destroy_next
;
212 kmem_cache_free(fsnotify_mark_connector_cachep
, free
);
216 static void fsnotify_put_inode_ref(struct inode
*inode
)
218 struct super_block
*sb
= inode
->i_sb
;
221 if (atomic_long_dec_and_test(&sb
->s_fsnotify_connectors
))
222 wake_up_var(&sb
->s_fsnotify_connectors
);
225 static void fsnotify_get_sb_connectors(struct fsnotify_mark_connector
*conn
)
227 struct super_block
*sb
= fsnotify_connector_sb(conn
);
230 atomic_long_inc(&sb
->s_fsnotify_connectors
);
233 static void fsnotify_put_sb_connectors(struct fsnotify_mark_connector
*conn
)
235 struct super_block
*sb
= fsnotify_connector_sb(conn
);
237 if (sb
&& atomic_long_dec_and_test(&sb
->s_fsnotify_connectors
))
238 wake_up_var(&sb
->s_fsnotify_connectors
);
241 static void *fsnotify_detach_connector_from_object(
242 struct fsnotify_mark_connector
*conn
,
245 struct inode
*inode
= NULL
;
248 if (conn
->type
== FSNOTIFY_OBJ_TYPE_DETACHED
)
251 if (conn
->type
== FSNOTIFY_OBJ_TYPE_INODE
) {
252 inode
= fsnotify_conn_inode(conn
);
253 inode
->i_fsnotify_mask
= 0;
255 /* Unpin inode when detaching from connector */
256 if (!(conn
->flags
& FSNOTIFY_CONN_FLAG_HAS_IREF
))
258 } else if (conn
->type
== FSNOTIFY_OBJ_TYPE_VFSMOUNT
) {
259 fsnotify_conn_mount(conn
)->mnt_fsnotify_mask
= 0;
260 } else if (conn
->type
== FSNOTIFY_OBJ_TYPE_SB
) {
261 fsnotify_conn_sb(conn
)->s_fsnotify_mask
= 0;
264 fsnotify_put_sb_connectors(conn
);
265 rcu_assign_pointer(*(conn
->obj
), NULL
);
267 conn
->type
= FSNOTIFY_OBJ_TYPE_DETACHED
;
272 static void fsnotify_final_mark_destroy(struct fsnotify_mark
*mark
)
274 struct fsnotify_group
*group
= mark
->group
;
276 if (WARN_ON_ONCE(!group
))
278 group
->ops
->free_mark(mark
);
279 fsnotify_put_group(group
);
282 /* Drop object reference originally held by a connector */
283 static void fsnotify_drop_object(unsigned int type
, void *objp
)
287 /* Currently only inode references are passed to be dropped */
288 if (WARN_ON_ONCE(type
!= FSNOTIFY_OBJ_TYPE_INODE
))
290 fsnotify_put_inode_ref(objp
);
293 void fsnotify_put_mark(struct fsnotify_mark
*mark
)
295 struct fsnotify_mark_connector
*conn
= READ_ONCE(mark
->connector
);
297 unsigned int type
= FSNOTIFY_OBJ_TYPE_DETACHED
;
298 bool free_conn
= false;
300 /* Catch marks that were actually never attached to object */
302 if (refcount_dec_and_test(&mark
->refcnt
))
303 fsnotify_final_mark_destroy(mark
);
308 * We have to be careful so that traversals of obj_list under lock can
309 * safely grab mark reference.
311 if (!refcount_dec_and_lock(&mark
->refcnt
, &conn
->lock
))
314 hlist_del_init_rcu(&mark
->obj_list
);
315 if (hlist_empty(&conn
->list
)) {
316 objp
= fsnotify_detach_connector_from_object(conn
, &type
);
319 objp
= __fsnotify_recalc_mask(conn
);
322 WRITE_ONCE(mark
->connector
, NULL
);
323 spin_unlock(&conn
->lock
);
325 fsnotify_drop_object(type
, objp
);
328 spin_lock(&destroy_lock
);
329 conn
->destroy_next
= connector_destroy_list
;
330 connector_destroy_list
= conn
;
331 spin_unlock(&destroy_lock
);
332 queue_work(system_unbound_wq
, &connector_reaper_work
);
335 * Note that we didn't update flags telling whether inode cares about
336 * what's happening with children. We update these flags from
337 * __fsnotify_parent() lazily when next event happens on one of our
340 spin_lock(&destroy_lock
);
341 list_add(&mark
->g_list
, &destroy_list
);
342 spin_unlock(&destroy_lock
);
343 queue_delayed_work(system_unbound_wq
, &reaper_work
,
344 FSNOTIFY_REAPER_DELAY
);
346 EXPORT_SYMBOL_GPL(fsnotify_put_mark
);
349 * Get mark reference when we found the mark via lockless traversal of object
350 * list. Mark can be already removed from the list by now and on its way to be
351 * destroyed once SRCU period ends.
353 * Also pin the group so it doesn't disappear under us.
355 static bool fsnotify_get_mark_safe(struct fsnotify_mark
*mark
)
360 if (refcount_inc_not_zero(&mark
->refcnt
)) {
361 spin_lock(&mark
->lock
);
362 if (mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
) {
363 /* mark is attached, group is still alive then */
364 atomic_inc(&mark
->group
->user_waits
);
365 spin_unlock(&mark
->lock
);
368 spin_unlock(&mark
->lock
);
369 fsnotify_put_mark(mark
);
375 * Puts marks and wakes up group destruction if necessary.
377 * Pairs with fsnotify_get_mark_safe()
379 static void fsnotify_put_mark_wake(struct fsnotify_mark
*mark
)
382 struct fsnotify_group
*group
= mark
->group
;
384 fsnotify_put_mark(mark
);
386 * We abuse notification_waitq on group shutdown for waiting for
387 * all marks pinned when waiting for userspace.
389 if (atomic_dec_and_test(&group
->user_waits
) && group
->shutdown
)
390 wake_up(&group
->notification_waitq
);
394 bool fsnotify_prepare_user_wait(struct fsnotify_iter_info
*iter_info
)
395 __releases(&fsnotify_mark_srcu
)
399 fsnotify_foreach_iter_type(type
) {
400 /* This can fail if mark is being removed */
401 if (!fsnotify_get_mark_safe(iter_info
->marks
[type
])) {
402 __release(&fsnotify_mark_srcu
);
408 * Now that both marks are pinned by refcount in the inode / vfsmount
409 * lists, we can drop SRCU lock, and safely resume the list iteration
410 * once userspace returns.
412 srcu_read_unlock(&fsnotify_mark_srcu
, iter_info
->srcu_idx
);
417 for (type
--; type
>= 0; type
--)
418 fsnotify_put_mark_wake(iter_info
->marks
[type
]);
422 void fsnotify_finish_user_wait(struct fsnotify_iter_info
*iter_info
)
423 __acquires(&fsnotify_mark_srcu
)
427 iter_info
->srcu_idx
= srcu_read_lock(&fsnotify_mark_srcu
);
428 fsnotify_foreach_iter_type(type
)
429 fsnotify_put_mark_wake(iter_info
->marks
[type
]);
433 * Mark mark as detached, remove it from group list. Mark still stays in object
434 * list until its last reference is dropped. Note that we rely on mark being
435 * removed from group list before corresponding reference to it is dropped. In
436 * particular we rely on mark->connector being valid while we hold
437 * group->mark_mutex if we found the mark through g_list.
439 * Must be called with group->mark_mutex held. The caller must either hold
440 * reference to the mark or be protected by fsnotify_mark_srcu.
442 void fsnotify_detach_mark(struct fsnotify_mark
*mark
)
444 fsnotify_group_assert_locked(mark
->group
);
445 WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu
) &&
446 refcount_read(&mark
->refcnt
) < 1 +
447 !!(mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
));
449 spin_lock(&mark
->lock
);
450 /* something else already called this function on this mark */
451 if (!(mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
)) {
452 spin_unlock(&mark
->lock
);
455 mark
->flags
&= ~FSNOTIFY_MARK_FLAG_ATTACHED
;
456 list_del_init(&mark
->g_list
);
457 spin_unlock(&mark
->lock
);
459 /* Drop mark reference acquired in fsnotify_add_mark_locked() */
460 fsnotify_put_mark(mark
);
464 * Free fsnotify mark. The mark is actually only marked as being freed. The
465 * freeing is actually happening only once last reference to the mark is
466 * dropped from a workqueue which first waits for srcu period end.
468 * Caller must have a reference to the mark or be protected by
469 * fsnotify_mark_srcu.
471 void fsnotify_free_mark(struct fsnotify_mark
*mark
)
473 struct fsnotify_group
*group
= mark
->group
;
475 spin_lock(&mark
->lock
);
476 /* something else already called this function on this mark */
477 if (!(mark
->flags
& FSNOTIFY_MARK_FLAG_ALIVE
)) {
478 spin_unlock(&mark
->lock
);
481 mark
->flags
&= ~FSNOTIFY_MARK_FLAG_ALIVE
;
482 spin_unlock(&mark
->lock
);
485 * Some groups like to know that marks are being freed. This is a
486 * callback to the group function to let it know that this mark
489 if (group
->ops
->freeing_mark
)
490 group
->ops
->freeing_mark(mark
, group
);
493 void fsnotify_destroy_mark(struct fsnotify_mark
*mark
,
494 struct fsnotify_group
*group
)
496 fsnotify_group_lock(group
);
497 fsnotify_detach_mark(mark
);
498 fsnotify_group_unlock(group
);
499 fsnotify_free_mark(mark
);
501 EXPORT_SYMBOL_GPL(fsnotify_destroy_mark
);
504 * Sorting function for lists of fsnotify marks.
506 * Fanotify supports different notification classes (reflected as priority of
507 * notification group). Events shall be passed to notification groups in
508 * decreasing priority order. To achieve this marks in notification lists for
509 * inodes and vfsmounts are sorted so that priorities of corresponding groups
512 * Furthermore correct handling of the ignore mask requires processing inode
513 * and vfsmount marks of each group together. Using the group address as
514 * further sort criterion provides a unique sorting order and thus we can
515 * merge inode and vfsmount lists of marks in linear time and find groups
516 * present in both lists.
518 * A return value of 1 signifies that b has priority over a.
519 * A return value of 0 signifies that the two marks have to be handled together.
520 * A return value of -1 signifies that a has priority over b.
522 int fsnotify_compare_groups(struct fsnotify_group
*a
, struct fsnotify_group
*b
)
530 if (a
->priority
< b
->priority
)
532 if (a
->priority
> b
->priority
)
539 static int fsnotify_attach_connector_to_object(fsnotify_connp_t
*connp
,
540 unsigned int obj_type
,
541 __kernel_fsid_t
*fsid
)
543 struct fsnotify_mark_connector
*conn
;
545 conn
= kmem_cache_alloc(fsnotify_mark_connector_cachep
, GFP_KERNEL
);
548 spin_lock_init(&conn
->lock
);
549 INIT_HLIST_HEAD(&conn
->list
);
551 conn
->type
= obj_type
;
553 /* Cache fsid of filesystem containing the object */
556 conn
->flags
= FSNOTIFY_CONN_FLAG_HAS_FSID
;
558 conn
->fsid
.val
[0] = conn
->fsid
.val
[1] = 0;
561 fsnotify_get_sb_connectors(conn
);
564 * cmpxchg() provides the barrier so that readers of *connp can see
565 * only initialized structure
567 if (cmpxchg(connp
, NULL
, conn
)) {
568 /* Someone else created list structure for us */
569 fsnotify_put_sb_connectors(conn
);
570 kmem_cache_free(fsnotify_mark_connector_cachep
, conn
);
577 * Get mark connector, make sure it is alive and return with its lock held.
578 * This is for users that get connector pointer from inode or mount. Users that
579 * hold reference to a mark on the list may directly lock connector->lock as
580 * they are sure list cannot go away under them.
582 static struct fsnotify_mark_connector
*fsnotify_grab_connector(
583 fsnotify_connp_t
*connp
)
585 struct fsnotify_mark_connector
*conn
;
588 idx
= srcu_read_lock(&fsnotify_mark_srcu
);
589 conn
= srcu_dereference(*connp
, &fsnotify_mark_srcu
);
592 spin_lock(&conn
->lock
);
593 if (conn
->type
== FSNOTIFY_OBJ_TYPE_DETACHED
) {
594 spin_unlock(&conn
->lock
);
595 srcu_read_unlock(&fsnotify_mark_srcu
, idx
);
599 srcu_read_unlock(&fsnotify_mark_srcu
, idx
);
604 * Add mark into proper place in given list of marks. These marks may be used
605 * for the fsnotify backend to determine which event types should be delivered
606 * to which group and for which inodes. These marks are ordered according to
607 * priority, highest number first, and then by the group's location in memory.
609 static int fsnotify_add_mark_list(struct fsnotify_mark
*mark
,
610 fsnotify_connp_t
*connp
,
611 unsigned int obj_type
,
612 int add_flags
, __kernel_fsid_t
*fsid
)
614 struct fsnotify_mark
*lmark
, *last
= NULL
;
615 struct fsnotify_mark_connector
*conn
;
619 if (WARN_ON(!fsnotify_valid_obj_type(obj_type
)))
622 /* Backend is expected to check for zero fsid (e.g. tmpfs) */
623 if (fsid
&& WARN_ON_ONCE(!fsid
->val
[0] && !fsid
->val
[1]))
627 spin_lock(&mark
->lock
);
628 conn
= fsnotify_grab_connector(connp
);
630 spin_unlock(&mark
->lock
);
631 err
= fsnotify_attach_connector_to_object(connp
, obj_type
,
636 } else if (fsid
&& !(conn
->flags
& FSNOTIFY_CONN_FLAG_HAS_FSID
)) {
638 /* Pairs with smp_rmb() in fanotify_get_fsid() */
640 conn
->flags
|= FSNOTIFY_CONN_FLAG_HAS_FSID
;
641 } else if (fsid
&& (conn
->flags
& FSNOTIFY_CONN_FLAG_HAS_FSID
) &&
642 (fsid
->val
[0] != conn
->fsid
.val
[0] ||
643 fsid
->val
[1] != conn
->fsid
.val
[1])) {
645 * Backend is expected to check for non uniform fsid
646 * (e.g. btrfs), but maybe we missed something?
647 * Only allow setting conn->fsid once to non zero fsid.
648 * inotify and non-fid fanotify groups do not set nor test
651 pr_warn_ratelimited("%s: fsid mismatch on object of type %u: "
652 "%x.%x != %x.%x\n", __func__
, conn
->type
,
653 fsid
->val
[0], fsid
->val
[1],
654 conn
->fsid
.val
[0], conn
->fsid
.val
[1]);
659 /* is mark the first mark? */
660 if (hlist_empty(&conn
->list
)) {
661 hlist_add_head_rcu(&mark
->obj_list
, &conn
->list
);
665 /* should mark be in the middle of the current list? */
666 hlist_for_each_entry(lmark
, &conn
->list
, obj_list
) {
669 if ((lmark
->group
== mark
->group
) &&
670 (lmark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
) &&
671 !(mark
->group
->flags
& FSNOTIFY_GROUP_DUPS
)) {
676 cmp
= fsnotify_compare_groups(lmark
->group
, mark
->group
);
678 hlist_add_before_rcu(&mark
->obj_list
, &lmark
->obj_list
);
683 BUG_ON(last
== NULL
);
684 /* mark should be the last entry. last is the current last entry */
685 hlist_add_behind_rcu(&mark
->obj_list
, &last
->obj_list
);
688 * Since connector is attached to object using cmpxchg() we are
689 * guaranteed that connector initialization is fully visible by anyone
690 * seeing mark->connector set.
692 WRITE_ONCE(mark
->connector
, conn
);
694 spin_unlock(&conn
->lock
);
695 spin_unlock(&mark
->lock
);
700 * Attach an initialized mark to a given group and fs object.
701 * These marks may be used for the fsnotify backend to determine which
702 * event types should be delivered to which group.
704 int fsnotify_add_mark_locked(struct fsnotify_mark
*mark
,
705 fsnotify_connp_t
*connp
, unsigned int obj_type
,
706 int add_flags
, __kernel_fsid_t
*fsid
)
708 struct fsnotify_group
*group
= mark
->group
;
711 fsnotify_group_assert_locked(group
);
717 * mark->connector->lock
719 spin_lock(&mark
->lock
);
720 mark
->flags
|= FSNOTIFY_MARK_FLAG_ALIVE
| FSNOTIFY_MARK_FLAG_ATTACHED
;
722 list_add(&mark
->g_list
, &group
->marks_list
);
723 fsnotify_get_mark(mark
); /* for g_list */
724 spin_unlock(&mark
->lock
);
726 ret
= fsnotify_add_mark_list(mark
, connp
, obj_type
, add_flags
, fsid
);
730 fsnotify_recalc_mask(mark
->connector
);
734 spin_lock(&mark
->lock
);
735 mark
->flags
&= ~(FSNOTIFY_MARK_FLAG_ALIVE
|
736 FSNOTIFY_MARK_FLAG_ATTACHED
);
737 list_del_init(&mark
->g_list
);
738 spin_unlock(&mark
->lock
);
740 fsnotify_put_mark(mark
);
744 int fsnotify_add_mark(struct fsnotify_mark
*mark
, fsnotify_connp_t
*connp
,
745 unsigned int obj_type
, int add_flags
,
746 __kernel_fsid_t
*fsid
)
749 struct fsnotify_group
*group
= mark
->group
;
751 fsnotify_group_lock(group
);
752 ret
= fsnotify_add_mark_locked(mark
, connp
, obj_type
, add_flags
, fsid
);
753 fsnotify_group_unlock(group
);
756 EXPORT_SYMBOL_GPL(fsnotify_add_mark
);
759 * Given a list of marks, find the mark associated with given group. If found
760 * take a reference to that mark and return it, else return NULL.
762 struct fsnotify_mark
*fsnotify_find_mark(fsnotify_connp_t
*connp
,
763 struct fsnotify_group
*group
)
765 struct fsnotify_mark_connector
*conn
;
766 struct fsnotify_mark
*mark
;
768 conn
= fsnotify_grab_connector(connp
);
772 hlist_for_each_entry(mark
, &conn
->list
, obj_list
) {
773 if (mark
->group
== group
&&
774 (mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
)) {
775 fsnotify_get_mark(mark
);
776 spin_unlock(&conn
->lock
);
780 spin_unlock(&conn
->lock
);
783 EXPORT_SYMBOL_GPL(fsnotify_find_mark
);
785 /* Clear any marks in a group with given type mask */
786 void fsnotify_clear_marks_by_group(struct fsnotify_group
*group
,
787 unsigned int obj_type
)
789 struct fsnotify_mark
*lmark
, *mark
;
791 struct list_head
*head
= &to_free
;
793 /* Skip selection step if we want to clear all marks. */
794 if (obj_type
== FSNOTIFY_OBJ_TYPE_ANY
) {
795 head
= &group
->marks_list
;
799 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
800 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
801 * to_free list so we have to use mark_mutex even when accessing that
802 * list. And freeing mark requires us to drop mark_mutex. So we can
803 * reliably free only the first mark in the list. That's why we first
804 * move marks to free to to_free list in one go and then free marks in
805 * to_free list one by one.
807 fsnotify_group_lock(group
);
808 list_for_each_entry_safe(mark
, lmark
, &group
->marks_list
, g_list
) {
809 if (mark
->connector
->type
== obj_type
)
810 list_move(&mark
->g_list
, &to_free
);
812 fsnotify_group_unlock(group
);
816 fsnotify_group_lock(group
);
817 if (list_empty(head
)) {
818 fsnotify_group_unlock(group
);
821 mark
= list_first_entry(head
, struct fsnotify_mark
, g_list
);
822 fsnotify_get_mark(mark
);
823 fsnotify_detach_mark(mark
);
824 fsnotify_group_unlock(group
);
825 fsnotify_free_mark(mark
);
826 fsnotify_put_mark(mark
);
830 /* Destroy all marks attached to an object via connector */
831 void fsnotify_destroy_marks(fsnotify_connp_t
*connp
)
833 struct fsnotify_mark_connector
*conn
;
834 struct fsnotify_mark
*mark
, *old_mark
= NULL
;
838 conn
= fsnotify_grab_connector(connp
);
842 * We have to be careful since we can race with e.g.
843 * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the
844 * list can get modified. However we are holding mark reference and
845 * thus our mark cannot be removed from obj_list so we can continue
846 * iteration after regaining conn->lock.
848 hlist_for_each_entry(mark
, &conn
->list
, obj_list
) {
849 fsnotify_get_mark(mark
);
850 spin_unlock(&conn
->lock
);
852 fsnotify_put_mark(old_mark
);
854 fsnotify_destroy_mark(mark
, mark
->group
);
855 spin_lock(&conn
->lock
);
858 * Detach list from object now so that we don't pin inode until all
859 * mark references get dropped. It would lead to strange results such
860 * as delaying inode deletion or blocking unmount.
862 objp
= fsnotify_detach_connector_from_object(conn
, &type
);
863 spin_unlock(&conn
->lock
);
865 fsnotify_put_mark(old_mark
);
866 fsnotify_drop_object(type
, objp
);
870 * Nothing fancy, just initialize lists and locks and counters.
872 void fsnotify_init_mark(struct fsnotify_mark
*mark
,
873 struct fsnotify_group
*group
)
875 memset(mark
, 0, sizeof(*mark
));
876 spin_lock_init(&mark
->lock
);
877 refcount_set(&mark
->refcnt
, 1);
878 fsnotify_get_group(group
);
880 WRITE_ONCE(mark
->connector
, NULL
);
882 EXPORT_SYMBOL_GPL(fsnotify_init_mark
);
885 * Destroy all marks in destroy_list, waits for SRCU period to finish before
886 * actually freeing marks.
888 static void fsnotify_mark_destroy_workfn(struct work_struct
*work
)
890 struct fsnotify_mark
*mark
, *next
;
891 struct list_head private_destroy_list
;
893 spin_lock(&destroy_lock
);
894 /* exchange the list head */
895 list_replace_init(&destroy_list
, &private_destroy_list
);
896 spin_unlock(&destroy_lock
);
898 synchronize_srcu(&fsnotify_mark_srcu
);
900 list_for_each_entry_safe(mark
, next
, &private_destroy_list
, g_list
) {
901 list_del_init(&mark
->g_list
);
902 fsnotify_final_mark_destroy(mark
);
906 /* Wait for all marks queued for destruction to be actually destroyed */
907 void fsnotify_wait_marks_destroyed(void)
909 flush_delayed_work(&reaper_work
);
911 EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed
);