2 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2, or (at your option)
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; see the file COPYING. If not, write to
16 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
20 * fsnotify inode mark locking/lifetime/and refcnting
23 * The group->recnt and mark->refcnt tell how many "things" in the kernel
24 * currently are referencing the objects. Both kind of objects typically will
25 * live inside the kernel with a refcnt of 2, one for its creation and one for
26 * the reference a group and a mark hold to each other.
27 * If you are holding the appropriate locks, you can take a reference and the
28 * object itself is guaranteed to survive until the reference is dropped.
31 * There are 3 locks involved with fsnotify inode marks and they MUST be taken
32 * in order as follows:
38 * group->mark_mutex protects the marks_list anchored inside a given group and
39 * each mark is hooked via the g_list. It also protects the groups private
40 * data (i.e group limits).
42 * mark->lock protects the marks attributes like its masks and flags.
43 * Furthermore it protects the access to a reference of the group that the mark
44 * is assigned to as well as the access to a reference of the inode/vfsmount
45 * that is being watched by the mark.
47 * inode->i_lock protects the i_fsnotify_marks list anchored inside a
48 * given inode and each mark is hooked via the i_list. (and sorta the
53 * Inode marks survive between when they are added to an inode and when their
54 * refcnt==0. Marks are also protected by fsnotify_mark_srcu.
56 * The inode mark can be cleared for a number of different reasons including:
57 * - The inode is unlinked for the last time. (fsnotify_inode_remove)
58 * - The inode is being evicted from cache. (fsnotify_inode_delete)
59 * - The fs the inode is on is unmounted. (fsnotify_inode_delete/fsnotify_unmount_inodes)
60 * - Something explicitly requests that it be removed. (fsnotify_destroy_mark)
61 * - The fsnotify_group associated with the mark is going away and all such marks
62 * need to be cleaned up. (fsnotify_clear_marks_by_group)
64 * This has the very interesting property of being able to run concurrently with
65 * any (or all) other directions.
69 #include <linux/init.h>
70 #include <linux/kernel.h>
71 #include <linux/kthread.h>
72 #include <linux/module.h>
73 #include <linux/mutex.h>
74 #include <linux/slab.h>
75 #include <linux/spinlock.h>
76 #include <linux/srcu.h>
78 #include <linux/atomic.h>
80 #include <linux/fsnotify_backend.h>
83 #define FSNOTIFY_REAPER_DELAY (1) /* 1 jiffy */
85 struct srcu_struct fsnotify_mark_srcu;
86 struct kmem_cache *fsnotify_mark_connector_cachep;
88 static DEFINE_SPINLOCK(destroy_lock);
89 static LIST_HEAD(destroy_list);
91 static void fsnotify_mark_destroy_workfn(struct work_struct *work);
92 static DECLARE_DELAYED_WORK(reaper_work, fsnotify_mark_destroy_workfn);
94 void fsnotify_get_mark(struct fsnotify_mark *mark)
96 atomic_inc(&mark->refcnt);
99 void fsnotify_put_mark(struct fsnotify_mark *mark)
101 if (atomic_dec_and_test(&mark->refcnt)) {
103 fsnotify_put_group(mark->group);
104 mark->free_mark(mark);
108 static void __fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
111 struct fsnotify_mark *mark;
113 hlist_for_each_entry(mark, &conn->list, obj_list)
114 new_mask |= mark->mask;
115 if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE)
116 conn->inode->i_fsnotify_mask = new_mask;
117 else if (conn->flags & FSNOTIFY_OBJ_TYPE_VFSMOUNT)
118 real_mount(conn->mnt)->mnt_fsnotify_mask = new_mask;
122 * Calculate mask of events for a list of marks. The caller must make sure
123 * connector cannot disappear under us (usually by holding a mark->lock or
124 * mark->group->mark_mutex for a mark on this list).
126 void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
131 if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE)
132 spin_lock(&conn->inode->i_lock);
134 spin_lock(&conn->mnt->mnt_root->d_lock);
135 __fsnotify_recalc_mask(conn);
136 if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE) {
137 spin_unlock(&conn->inode->i_lock);
138 __fsnotify_update_child_dentry_flags(conn->inode);
140 spin_unlock(&conn->mnt->mnt_root->d_lock);
144 static struct inode *fsnotify_detach_from_object(struct fsnotify_mark *mark)
146 struct fsnotify_mark_connector *conn;
147 struct inode *inode = NULL;
150 conn = mark->connector;
151 if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE)
152 lock = &conn->inode->i_lock;
154 lock = &conn->mnt->mnt_root->d_lock;
156 hlist_del_init_rcu(&mark->obj_list);
157 if (hlist_empty(&conn->list)) {
158 if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE)
161 __fsnotify_recalc_mask(conn);
162 mark->connector = NULL;
169 * Remove mark from inode / vfsmount list, group list, drop inode reference
172 * Must be called with group->mark_mutex held.
174 void fsnotify_detach_mark(struct fsnotify_mark *mark)
176 struct inode *inode = NULL;
177 struct fsnotify_group *group = mark->group;
179 BUG_ON(!mutex_is_locked(&group->mark_mutex));
181 spin_lock(&mark->lock);
183 /* something else already called this function on this mark */
184 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
185 spin_unlock(&mark->lock);
189 mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
191 inode = fsnotify_detach_from_object(mark);
194 * Note that we didn't update flags telling whether inode cares about
195 * what's happening with children. We update these flags from
196 * __fsnotify_parent() lazily when next event happens on one of our
200 list_del_init(&mark->g_list);
202 spin_unlock(&mark->lock);
207 atomic_dec(&group->num_marks);
211 * Prepare mark for freeing and add it to the list of marks prepared for
212 * freeing. The actual freeing must happen after SRCU period ends and the
213 * caller is responsible for this.
215 * The function returns true if the mark was added to the list of marks for
216 * freeing. The function returns false if someone else has already called
217 * __fsnotify_free_mark() for the mark.
219 static bool __fsnotify_free_mark(struct fsnotify_mark *mark)
221 struct fsnotify_group *group = mark->group;
223 spin_lock(&mark->lock);
224 /* something else already called this function on this mark */
225 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
226 spin_unlock(&mark->lock);
229 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
230 spin_unlock(&mark->lock);
233 * Some groups like to know that marks are being freed. This is a
234 * callback to the group function to let it know that this mark
237 if (group->ops->freeing_mark)
238 group->ops->freeing_mark(mark, group);
240 spin_lock(&destroy_lock);
241 list_add(&mark->g_list, &destroy_list);
242 spin_unlock(&destroy_lock);
248 * Free fsnotify mark. The freeing is actually happening from a workqueue which
249 * first waits for srcu period end. Caller must have a reference to the mark
250 * or be protected by fsnotify_mark_srcu.
252 void fsnotify_free_mark(struct fsnotify_mark *mark)
254 if (__fsnotify_free_mark(mark)) {
255 queue_delayed_work(system_unbound_wq, &reaper_work,
256 FSNOTIFY_REAPER_DELAY);
260 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
261 struct fsnotify_group *group)
263 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
264 fsnotify_detach_mark(mark);
265 mutex_unlock(&group->mark_mutex);
266 fsnotify_free_mark(mark);
269 void fsnotify_connector_free(struct fsnotify_mark_connector **connp)
272 kmem_cache_free(fsnotify_mark_connector_cachep, *connp);
277 void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
279 assert_spin_locked(&mark->lock);
284 void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
286 assert_spin_locked(&mark->lock);
288 mark->ignored_mask = mask;
292 * Sorting function for lists of fsnotify marks.
294 * Fanotify supports different notification classes (reflected as priority of
295 * notification group). Events shall be passed to notification groups in
296 * decreasing priority order. To achieve this marks in notification lists for
297 * inodes and vfsmounts are sorted so that priorities of corresponding groups
300 * Furthermore correct handling of the ignore mask requires processing inode
301 * and vfsmount marks of each group together. Using the group address as
302 * further sort criterion provides a unique sorting order and thus we can
303 * merge inode and vfsmount lists of marks in linear time and find groups
304 * present in both lists.
306 * A return value of 1 signifies that b has priority over a.
307 * A return value of 0 signifies that the two marks have to be handled together.
308 * A return value of -1 signifies that a has priority over b.
310 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
318 if (a->priority < b->priority)
320 if (a->priority > b->priority)
327 static int fsnotify_attach_connector_to_object(
328 struct fsnotify_mark_connector **connp,
331 struct vfsmount *mnt)
333 struct fsnotify_mark_connector *conn;
335 conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL);
338 INIT_HLIST_HEAD(&conn->list);
340 conn->flags = FSNOTIFY_OBJ_TYPE_INODE;
343 conn->flags = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
347 * Make sure 'conn' initialization is visible. Matches
348 * lockless_dereference() in fsnotify().
355 kmem_cache_free(fsnotify_mark_connector_cachep, conn);
362 * Add mark into proper place in given list of marks. These marks may be used
363 * for the fsnotify backend to determine which event types should be delivered
364 * to which group and for which inodes. These marks are ordered according to
365 * priority, highest number first, and then by the group's location in memory.
367 static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
368 struct inode *inode, struct vfsmount *mnt,
371 struct fsnotify_mark *lmark, *last = NULL;
372 struct fsnotify_mark_connector *conn;
373 struct fsnotify_mark_connector **connp;
378 if (WARN_ON(!inode && !mnt))
381 connp = &inode->i_fsnotify_marks;
382 lock = &inode->i_lock;
384 connp = &real_mount(mnt)->mnt_fsnotify_marks;
385 lock = &mnt->mnt_root->d_lock;
389 err = fsnotify_attach_connector_to_object(connp, lock,
394 spin_lock(&mark->lock);
398 /* is mark the first mark? */
399 if (hlist_empty(&conn->list)) {
400 hlist_add_head_rcu(&mark->obj_list, &conn->list);
406 /* should mark be in the middle of the current list? */
407 hlist_for_each_entry(lmark, &conn->list, obj_list) {
410 if ((lmark->group == mark->group) && !allow_dups) {
415 cmp = fsnotify_compare_groups(lmark->group, mark->group);
417 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
422 BUG_ON(last == NULL);
423 /* mark should be the last entry. last is the current last entry */
424 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
426 mark->connector = conn;
429 spin_unlock(&mark->lock);
434 * Attach an initialized mark to a given group and fs object.
435 * These marks may be used for the fsnotify backend to determine which
436 * event types should be delivered to which group.
438 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
439 struct fsnotify_group *group, struct inode *inode,
440 struct vfsmount *mnt, int allow_dups)
444 BUG_ON(inode && mnt);
445 BUG_ON(!inode && !mnt);
446 BUG_ON(!mutex_is_locked(&group->mark_mutex));
454 spin_lock(&mark->lock);
455 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
457 fsnotify_get_group(group);
459 list_add(&mark->g_list, &group->marks_list);
460 atomic_inc(&group->num_marks);
461 fsnotify_get_mark(mark); /* for i_list and g_list */
462 spin_unlock(&mark->lock);
464 ret = fsnotify_add_mark_list(mark, inode, mnt, allow_dups);
469 fsnotify_recalc_mask(mark->connector);
473 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
474 list_del_init(&mark->g_list);
475 fsnotify_put_group(group);
477 atomic_dec(&group->num_marks);
479 spin_unlock(&mark->lock);
481 spin_lock(&destroy_lock);
482 list_add(&mark->g_list, &destroy_list);
483 spin_unlock(&destroy_lock);
484 queue_delayed_work(system_unbound_wq, &reaper_work,
485 FSNOTIFY_REAPER_DELAY);
490 int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
491 struct inode *inode, struct vfsmount *mnt, int allow_dups)
494 mutex_lock(&group->mark_mutex);
495 ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
496 mutex_unlock(&group->mark_mutex);
501 * Given a list of marks, find the mark associated with given group. If found
502 * take a reference to that mark and return it, else return NULL.
504 struct fsnotify_mark *fsnotify_find_mark(struct fsnotify_mark_connector *conn,
505 struct fsnotify_group *group)
507 struct fsnotify_mark *mark;
513 if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE)
514 lock = &conn->inode->i_lock;
516 lock = &conn->mnt->mnt_root->d_lock;
518 hlist_for_each_entry(mark, &conn->list, obj_list) {
519 if (mark->group == group) {
520 fsnotify_get_mark(mark);
530 * clear any marks in a group in which mark->flags & flags is true
532 void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
535 struct fsnotify_mark *lmark, *mark;
539 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
540 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
541 * to_free list so we have to use mark_mutex even when accessing that
542 * list. And freeing mark requires us to drop mark_mutex. So we can
543 * reliably free only the first mark in the list. That's why we first
544 * move marks to free to to_free list in one go and then free marks in
545 * to_free list one by one.
547 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
548 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
549 if (mark->connector->flags & flags)
550 list_move(&mark->g_list, &to_free);
552 mutex_unlock(&group->mark_mutex);
555 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
556 if (list_empty(&to_free)) {
557 mutex_unlock(&group->mark_mutex);
560 mark = list_first_entry(&to_free, struct fsnotify_mark, g_list);
561 fsnotify_get_mark(mark);
562 fsnotify_detach_mark(mark);
563 mutex_unlock(&group->mark_mutex);
564 fsnotify_free_mark(mark);
565 fsnotify_put_mark(mark);
570 * Given a group, prepare for freeing all the marks associated with that group.
571 * The marks are attached to the list of marks prepared for destruction, the
572 * caller is responsible for freeing marks in that list after SRCU period has
575 void fsnotify_detach_group_marks(struct fsnotify_group *group)
577 struct fsnotify_mark *mark;
580 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
581 if (list_empty(&group->marks_list)) {
582 mutex_unlock(&group->mark_mutex);
585 mark = list_first_entry(&group->marks_list,
586 struct fsnotify_mark, g_list);
587 fsnotify_get_mark(mark);
588 fsnotify_detach_mark(mark);
589 mutex_unlock(&group->mark_mutex);
590 __fsnotify_free_mark(mark);
591 fsnotify_put_mark(mark);
595 void fsnotify_destroy_marks(struct fsnotify_mark_connector *conn)
597 struct fsnotify_mark *mark;
603 if (conn->flags & FSNOTIFY_OBJ_TYPE_INODE)
604 lock = &conn->inode->i_lock;
606 lock = &conn->mnt->mnt_root->d_lock;
610 * We have to be careful since we can race with e.g.
611 * fsnotify_clear_marks_by_group() and once we drop 'lock',
612 * mark can get removed from the obj_list and destroyed. But
613 * we are holding mark reference so mark cannot be freed and
614 * calling fsnotify_destroy_mark() more than once is fine.
617 if (hlist_empty(&conn->list)) {
621 mark = hlist_entry(conn->list.first, struct fsnotify_mark,
623 fsnotify_get_mark(mark);
625 fsnotify_destroy_mark(mark, mark->group);
626 fsnotify_put_mark(mark);
631 * Nothing fancy, just initialize lists and locks and counters.
633 void fsnotify_init_mark(struct fsnotify_mark *mark,
634 void (*free_mark)(struct fsnotify_mark *mark))
636 memset(mark, 0, sizeof(*mark));
637 spin_lock_init(&mark->lock);
638 atomic_set(&mark->refcnt, 1);
639 mark->free_mark = free_mark;
643 * Destroy all marks in destroy_list, waits for SRCU period to finish before
644 * actually freeing marks.
646 void fsnotify_mark_destroy_list(void)
648 struct fsnotify_mark *mark, *next;
649 struct list_head private_destroy_list;
651 spin_lock(&destroy_lock);
652 /* exchange the list head */
653 list_replace_init(&destroy_list, &private_destroy_list);
654 spin_unlock(&destroy_lock);
656 synchronize_srcu(&fsnotify_mark_srcu);
658 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
659 list_del_init(&mark->g_list);
660 fsnotify_put_mark(mark);
664 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
666 fsnotify_mark_destroy_list();