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 /* Calculate mask of events for a list of marks */
109 u32 fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
112 struct fsnotify_mark *mark;
117 hlist_for_each_entry(mark, &conn->list, obj_list)
118 new_mask |= mark->mask;
123 * Remove mark from inode / vfsmount list, group list, drop inode reference
126 * Must be called with group->mark_mutex held.
128 void fsnotify_detach_mark(struct fsnotify_mark *mark)
130 struct inode *inode = NULL;
131 struct fsnotify_group *group = mark->group;
133 BUG_ON(!mutex_is_locked(&group->mark_mutex));
135 spin_lock(&mark->lock);
137 /* something else already called this function on this mark */
138 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
139 spin_unlock(&mark->lock);
143 mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
145 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
147 fsnotify_destroy_inode_mark(mark);
148 } else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
149 fsnotify_destroy_vfsmount_mark(mark);
153 * Note that we didn't update flags telling whether inode cares about
154 * what's happening with children. We update these flags from
155 * __fsnotify_parent() lazily when next event happens on one of our
159 list_del_init(&mark->g_list);
161 spin_unlock(&mark->lock);
163 if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
166 atomic_dec(&group->num_marks);
170 * Prepare mark for freeing and add it to the list of marks prepared for
171 * freeing. The actual freeing must happen after SRCU period ends and the
172 * caller is responsible for this.
174 * The function returns true if the mark was added to the list of marks for
175 * freeing. The function returns false if someone else has already called
176 * __fsnotify_free_mark() for the mark.
178 static bool __fsnotify_free_mark(struct fsnotify_mark *mark)
180 struct fsnotify_group *group = mark->group;
182 spin_lock(&mark->lock);
183 /* something else already called this function on this mark */
184 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
185 spin_unlock(&mark->lock);
188 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
189 spin_unlock(&mark->lock);
192 * Some groups like to know that marks are being freed. This is a
193 * callback to the group function to let it know that this mark
196 if (group->ops->freeing_mark)
197 group->ops->freeing_mark(mark, group);
199 spin_lock(&destroy_lock);
200 list_add(&mark->g_list, &destroy_list);
201 spin_unlock(&destroy_lock);
207 * Free fsnotify mark. The freeing is actually happening from a workqueue which
208 * first waits for srcu period end. Caller must have a reference to the mark
209 * or be protected by fsnotify_mark_srcu.
211 void fsnotify_free_mark(struct fsnotify_mark *mark)
213 if (__fsnotify_free_mark(mark)) {
214 queue_delayed_work(system_unbound_wq, &reaper_work,
215 FSNOTIFY_REAPER_DELAY);
219 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
220 struct fsnotify_group *group)
222 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
223 fsnotify_detach_mark(mark);
224 mutex_unlock(&group->mark_mutex);
225 fsnotify_free_mark(mark);
228 void fsnotify_destroy_marks(struct fsnotify_mark_connector *conn,
231 struct fsnotify_mark *mark;
238 * We have to be careful since we can race with e.g.
239 * fsnotify_clear_marks_by_group() and once we drop 'lock',
240 * mark can get removed from the obj_list and destroyed. But
241 * we are holding mark reference so mark cannot be freed and
242 * calling fsnotify_destroy_mark() more than once is fine.
245 if (hlist_empty(&conn->list)) {
249 mark = hlist_entry(conn->list.first, struct fsnotify_mark,
252 * We don't update i_fsnotify_mask / mnt_fsnotify_mask here
253 * since inode / mount is going away anyway. So just remove
254 * mark from the list.
256 hlist_del_init_rcu(&mark->obj_list);
257 fsnotify_get_mark(mark);
259 fsnotify_destroy_mark(mark, mark->group);
260 fsnotify_put_mark(mark);
264 void fsnotify_connector_free(struct fsnotify_mark_connector **connp)
267 kmem_cache_free(fsnotify_mark_connector_cachep, *connp);
272 void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
274 assert_spin_locked(&mark->lock);
278 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
279 fsnotify_set_inode_mark_mask_locked(mark, mask);
282 void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
284 assert_spin_locked(&mark->lock);
286 mark->ignored_mask = mask;
290 * Sorting function for lists of fsnotify marks.
292 * Fanotify supports different notification classes (reflected as priority of
293 * notification group). Events shall be passed to notification groups in
294 * decreasing priority order. To achieve this marks in notification lists for
295 * inodes and vfsmounts are sorted so that priorities of corresponding groups
298 * Furthermore correct handling of the ignore mask requires processing inode
299 * and vfsmount marks of each group together. Using the group address as
300 * further sort criterion provides a unique sorting order and thus we can
301 * merge inode and vfsmount lists of marks in linear time and find groups
302 * present in both lists.
304 * A return value of 1 signifies that b has priority over a.
305 * A return value of 0 signifies that the two marks have to be handled together.
306 * A return value of -1 signifies that a has priority over b.
308 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
316 if (a->priority < b->priority)
318 if (a->priority > b->priority)
325 static int fsnotify_attach_connector_to_object(
326 struct fsnotify_mark_connector **connp)
328 struct fsnotify_mark_connector *conn;
330 conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_ATOMIC);
333 INIT_HLIST_HEAD(&conn->list);
335 * Make sure 'conn' initialization is visible. Matches
336 * lockless_dereference() in fsnotify().
345 * Add mark into proper place in given list of marks. These marks may be used
346 * for the fsnotify backend to determine which event types should be delivered
347 * to which group and for which inodes. These marks are ordered according to
348 * priority, highest number first, and then by the group's location in memory.
350 int fsnotify_add_mark_list(struct fsnotify_mark_connector **connp,
351 struct fsnotify_mark *mark, int allow_dups)
353 struct fsnotify_mark *lmark, *last = NULL;
354 struct fsnotify_mark_connector *conn;
359 err = fsnotify_attach_connector_to_object(connp);
365 /* is mark the first mark? */
366 if (hlist_empty(&conn->list)) {
367 hlist_add_head_rcu(&mark->obj_list, &conn->list);
371 /* should mark be in the middle of the current list? */
372 hlist_for_each_entry(lmark, &conn->list, obj_list) {
375 if ((lmark->group == mark->group) && !allow_dups)
378 cmp = fsnotify_compare_groups(lmark->group, mark->group);
380 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
385 BUG_ON(last == NULL);
386 /* mark should be the last entry. last is the current last entry */
387 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
392 * Attach an initialized mark to a given group and fs object.
393 * These marks may be used for the fsnotify backend to determine which
394 * event types should be delivered to which group.
396 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
397 struct fsnotify_group *group, struct inode *inode,
398 struct vfsmount *mnt, int allow_dups)
402 BUG_ON(inode && mnt);
403 BUG_ON(!inode && !mnt);
404 BUG_ON(!mutex_is_locked(&group->mark_mutex));
412 spin_lock(&mark->lock);
413 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
415 fsnotify_get_group(group);
417 list_add(&mark->g_list, &group->marks_list);
418 atomic_inc(&group->num_marks);
419 fsnotify_get_mark(mark); /* for i_list and g_list */
422 ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
426 ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
433 /* this will pin the object if appropriate */
434 fsnotify_set_mark_mask_locked(mark, mark->mask);
435 spin_unlock(&mark->lock);
438 __fsnotify_update_child_dentry_flags(inode);
442 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
443 list_del_init(&mark->g_list);
444 fsnotify_put_group(group);
446 atomic_dec(&group->num_marks);
448 spin_unlock(&mark->lock);
450 spin_lock(&destroy_lock);
451 list_add(&mark->g_list, &destroy_list);
452 spin_unlock(&destroy_lock);
453 queue_delayed_work(system_unbound_wq, &reaper_work,
454 FSNOTIFY_REAPER_DELAY);
459 int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
460 struct inode *inode, struct vfsmount *mnt, int allow_dups)
463 mutex_lock(&group->mark_mutex);
464 ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
465 mutex_unlock(&group->mark_mutex);
470 * Given a list of marks, find the mark associated with given group. If found
471 * take a reference to that mark and return it, else return NULL.
473 struct fsnotify_mark *fsnotify_find_mark(struct fsnotify_mark_connector *conn,
474 struct fsnotify_group *group)
476 struct fsnotify_mark *mark;
481 hlist_for_each_entry(mark, &conn->list, obj_list) {
482 if (mark->group == group) {
483 fsnotify_get_mark(mark);
491 * clear any marks in a group in which mark->flags & flags is true
493 void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
496 struct fsnotify_mark *lmark, *mark;
500 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
501 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
502 * to_free list so we have to use mark_mutex even when accessing that
503 * list. And freeing mark requires us to drop mark_mutex. So we can
504 * reliably free only the first mark in the list. That's why we first
505 * move marks to free to to_free list in one go and then free marks in
506 * to_free list one by one.
508 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
509 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
510 if (mark->flags & flags)
511 list_move(&mark->g_list, &to_free);
513 mutex_unlock(&group->mark_mutex);
516 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
517 if (list_empty(&to_free)) {
518 mutex_unlock(&group->mark_mutex);
521 mark = list_first_entry(&to_free, struct fsnotify_mark, g_list);
522 fsnotify_get_mark(mark);
523 fsnotify_detach_mark(mark);
524 mutex_unlock(&group->mark_mutex);
525 fsnotify_free_mark(mark);
526 fsnotify_put_mark(mark);
531 * Given a group, prepare for freeing all the marks associated with that group.
532 * The marks are attached to the list of marks prepared for destruction, the
533 * caller is responsible for freeing marks in that list after SRCU period has
536 void fsnotify_detach_group_marks(struct fsnotify_group *group)
538 struct fsnotify_mark *mark;
541 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
542 if (list_empty(&group->marks_list)) {
543 mutex_unlock(&group->mark_mutex);
546 mark = list_first_entry(&group->marks_list,
547 struct fsnotify_mark, g_list);
548 fsnotify_get_mark(mark);
549 fsnotify_detach_mark(mark);
550 mutex_unlock(&group->mark_mutex);
551 __fsnotify_free_mark(mark);
552 fsnotify_put_mark(mark);
557 * Nothing fancy, just initialize lists and locks and counters.
559 void fsnotify_init_mark(struct fsnotify_mark *mark,
560 void (*free_mark)(struct fsnotify_mark *mark))
562 memset(mark, 0, sizeof(*mark));
563 spin_lock_init(&mark->lock);
564 atomic_set(&mark->refcnt, 1);
565 mark->free_mark = free_mark;
569 * Destroy all marks in destroy_list, waits for SRCU period to finish before
570 * actually freeing marks.
572 void fsnotify_mark_destroy_list(void)
574 struct fsnotify_mark *mark, *next;
575 struct list_head private_destroy_list;
577 spin_lock(&destroy_lock);
578 /* exchange the list head */
579 list_replace_init(&destroy_list, &private_destroy_list);
580 spin_unlock(&destroy_lock);
582 synchronize_srcu(&fsnotify_mark_srcu);
584 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
585 list_del_init(&mark->g_list);
586 fsnotify_put_mark(mark);
590 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
592 fsnotify_mark_destroy_list();