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->connector->flags & FSNOTIFY_OBJ_TYPE_INODE)
146 inode = fsnotify_destroy_inode_mark(mark);
147 else if (mark->connector->flags & FSNOTIFY_OBJ_TYPE_VFSMOUNT)
148 fsnotify_destroy_vfsmount_mark(mark);
152 * Note that we didn't update flags telling whether inode cares about
153 * what's happening with children. We update these flags from
154 * __fsnotify_parent() lazily when next event happens on one of our
158 list_del_init(&mark->g_list);
160 spin_unlock(&mark->lock);
165 atomic_dec(&group->num_marks);
169 * Prepare mark for freeing and add it to the list of marks prepared for
170 * freeing. The actual freeing must happen after SRCU period ends and the
171 * caller is responsible for this.
173 * The function returns true if the mark was added to the list of marks for
174 * freeing. The function returns false if someone else has already called
175 * __fsnotify_free_mark() for the mark.
177 static bool __fsnotify_free_mark(struct fsnotify_mark *mark)
179 struct fsnotify_group *group = mark->group;
181 spin_lock(&mark->lock);
182 /* something else already called this function on this mark */
183 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
184 spin_unlock(&mark->lock);
187 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
188 spin_unlock(&mark->lock);
191 * Some groups like to know that marks are being freed. This is a
192 * callback to the group function to let it know that this mark
195 if (group->ops->freeing_mark)
196 group->ops->freeing_mark(mark, group);
198 spin_lock(&destroy_lock);
199 list_add(&mark->g_list, &destroy_list);
200 spin_unlock(&destroy_lock);
206 * Free fsnotify mark. The freeing is actually happening from a workqueue which
207 * first waits for srcu period end. Caller must have a reference to the mark
208 * or be protected by fsnotify_mark_srcu.
210 void fsnotify_free_mark(struct fsnotify_mark *mark)
212 if (__fsnotify_free_mark(mark)) {
213 queue_delayed_work(system_unbound_wq, &reaper_work,
214 FSNOTIFY_REAPER_DELAY);
218 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
219 struct fsnotify_group *group)
221 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
222 fsnotify_detach_mark(mark);
223 mutex_unlock(&group->mark_mutex);
224 fsnotify_free_mark(mark);
227 void fsnotify_connector_free(struct fsnotify_mark_connector **connp)
230 kmem_cache_free(fsnotify_mark_connector_cachep, *connp);
235 void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
237 assert_spin_locked(&mark->lock);
242 void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
244 assert_spin_locked(&mark->lock);
246 mark->ignored_mask = mask;
250 * Sorting function for lists of fsnotify marks.
252 * Fanotify supports different notification classes (reflected as priority of
253 * notification group). Events shall be passed to notification groups in
254 * decreasing priority order. To achieve this marks in notification lists for
255 * inodes and vfsmounts are sorted so that priorities of corresponding groups
258 * Furthermore correct handling of the ignore mask requires processing inode
259 * and vfsmount marks of each group together. Using the group address as
260 * further sort criterion provides a unique sorting order and thus we can
261 * merge inode and vfsmount lists of marks in linear time and find groups
262 * present in both lists.
264 * A return value of 1 signifies that b has priority over a.
265 * A return value of 0 signifies that the two marks have to be handled together.
266 * A return value of -1 signifies that a has priority over b.
268 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
276 if (a->priority < b->priority)
278 if (a->priority > b->priority)
285 static int fsnotify_attach_connector_to_object(
286 struct fsnotify_mark_connector **connp,
289 struct vfsmount *mnt)
291 struct fsnotify_mark_connector *conn;
293 conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL);
296 INIT_HLIST_HEAD(&conn->list);
298 conn->flags = FSNOTIFY_OBJ_TYPE_INODE;
301 conn->flags = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
305 * Make sure 'conn' initialization is visible. Matches
306 * lockless_dereference() in fsnotify().
313 kmem_cache_free(fsnotify_mark_connector_cachep, conn);
320 * Add mark into proper place in given list of marks. These marks may be used
321 * for the fsnotify backend to determine which event types should be delivered
322 * to which group and for which inodes. These marks are ordered according to
323 * priority, highest number first, and then by the group's location in memory.
325 static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
326 struct inode *inode, struct vfsmount *mnt,
329 struct fsnotify_mark *lmark, *last = NULL;
330 struct fsnotify_mark_connector *conn;
331 struct fsnotify_mark_connector **connp;
336 if (WARN_ON(!inode && !mnt))
339 connp = &inode->i_fsnotify_marks;
340 lock = &inode->i_lock;
342 connp = &real_mount(mnt)->mnt_fsnotify_marks;
343 lock = &mnt->mnt_root->d_lock;
347 err = fsnotify_attach_connector_to_object(connp, lock,
352 spin_lock(&mark->lock);
356 /* is mark the first mark? */
357 if (hlist_empty(&conn->list)) {
358 hlist_add_head_rcu(&mark->obj_list, &conn->list);
364 /* should mark be in the middle of the current list? */
365 hlist_for_each_entry(lmark, &conn->list, obj_list) {
368 if ((lmark->group == mark->group) && !allow_dups) {
373 cmp = fsnotify_compare_groups(lmark->group, mark->group);
375 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
380 BUG_ON(last == NULL);
381 /* mark should be the last entry. last is the current last entry */
382 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
384 mark->connector = conn;
387 spin_unlock(&mark->lock);
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 */
420 spin_unlock(&mark->lock);
422 ret = fsnotify_add_mark_list(mark, inode, mnt, allow_dups);
427 fsnotify_recalc_inode_mask(inode);
429 fsnotify_recalc_vfsmount_mask(mnt);
433 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
434 list_del_init(&mark->g_list);
435 fsnotify_put_group(group);
437 atomic_dec(&group->num_marks);
439 spin_unlock(&mark->lock);
441 spin_lock(&destroy_lock);
442 list_add(&mark->g_list, &destroy_list);
443 spin_unlock(&destroy_lock);
444 queue_delayed_work(system_unbound_wq, &reaper_work,
445 FSNOTIFY_REAPER_DELAY);
450 int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
451 struct inode *inode, struct vfsmount *mnt, int allow_dups)
454 mutex_lock(&group->mark_mutex);
455 ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
456 mutex_unlock(&group->mark_mutex);
461 * Given a list of marks, find the mark associated with given group. If found
462 * take a reference to that mark and return it, else return NULL.
464 struct fsnotify_mark *fsnotify_find_mark(struct fsnotify_mark_connector *conn,
465 struct fsnotify_group *group)
467 struct fsnotify_mark *mark;
472 hlist_for_each_entry(mark, &conn->list, obj_list) {
473 if (mark->group == group) {
474 fsnotify_get_mark(mark);
482 * clear any marks in a group in which mark->flags & flags is true
484 void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
487 struct fsnotify_mark *lmark, *mark;
491 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
492 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
493 * to_free list so we have to use mark_mutex even when accessing that
494 * list. And freeing mark requires us to drop mark_mutex. So we can
495 * reliably free only the first mark in the list. That's why we first
496 * move marks to free to to_free list in one go and then free marks in
497 * to_free list one by one.
499 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
500 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
501 if (mark->connector->flags & flags)
502 list_move(&mark->g_list, &to_free);
504 mutex_unlock(&group->mark_mutex);
507 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
508 if (list_empty(&to_free)) {
509 mutex_unlock(&group->mark_mutex);
512 mark = list_first_entry(&to_free, struct fsnotify_mark, g_list);
513 fsnotify_get_mark(mark);
514 fsnotify_detach_mark(mark);
515 mutex_unlock(&group->mark_mutex);
516 fsnotify_free_mark(mark);
517 fsnotify_put_mark(mark);
522 * Given a group, prepare for freeing all the marks associated with that group.
523 * The marks are attached to the list of marks prepared for destruction, the
524 * caller is responsible for freeing marks in that list after SRCU period has
527 void fsnotify_detach_group_marks(struct fsnotify_group *group)
529 struct fsnotify_mark *mark;
532 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
533 if (list_empty(&group->marks_list)) {
534 mutex_unlock(&group->mark_mutex);
537 mark = list_first_entry(&group->marks_list,
538 struct fsnotify_mark, g_list);
539 fsnotify_get_mark(mark);
540 fsnotify_detach_mark(mark);
541 mutex_unlock(&group->mark_mutex);
542 __fsnotify_free_mark(mark);
543 fsnotify_put_mark(mark);
547 void fsnotify_destroy_marks(struct fsnotify_mark_connector *conn,
550 struct fsnotify_mark *mark;
557 * We have to be careful since we can race with e.g.
558 * fsnotify_clear_marks_by_group() and once we drop 'lock',
559 * mark can get removed from the obj_list and destroyed. But
560 * we are holding mark reference so mark cannot be freed and
561 * calling fsnotify_destroy_mark() more than once is fine.
564 if (hlist_empty(&conn->list)) {
568 mark = hlist_entry(conn->list.first, struct fsnotify_mark,
571 * We don't update i_fsnotify_mask / mnt_fsnotify_mask here
572 * since inode / mount is going away anyway. So just remove
573 * mark from the list.
575 hlist_del_init_rcu(&mark->obj_list);
576 fsnotify_get_mark(mark);
578 fsnotify_destroy_mark(mark, mark->group);
579 fsnotify_put_mark(mark);
584 * Nothing fancy, just initialize lists and locks and counters.
586 void fsnotify_init_mark(struct fsnotify_mark *mark,
587 void (*free_mark)(struct fsnotify_mark *mark))
589 memset(mark, 0, sizeof(*mark));
590 spin_lock_init(&mark->lock);
591 atomic_set(&mark->refcnt, 1);
592 mark->free_mark = free_mark;
596 * Destroy all marks in destroy_list, waits for SRCU period to finish before
597 * actually freeing marks.
599 void fsnotify_mark_destroy_list(void)
601 struct fsnotify_mark *mark, *next;
602 struct list_head private_destroy_list;
604 spin_lock(&destroy_lock);
605 /* exchange the list head */
606 list_replace_init(&destroy_list, &private_destroy_list);
607 spin_unlock(&destroy_lock);
609 synchronize_srcu(&fsnotify_mark_srcu);
611 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
612 list_del_init(&mark->g_list);
613 fsnotify_put_mark(mark);
617 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
619 fsnotify_mark_destroy_list();