xref: /linux/fs/notify/mark.c (revision 9756b9187eebb093b9f6a154ecceb67648e53391)
15444e298SEric Paris /*
25444e298SEric Paris  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
35444e298SEric Paris  *
45444e298SEric Paris  *  This program is free software; you can redistribute it and/or modify
55444e298SEric Paris  *  it under the terms of the GNU General Public License as published by
65444e298SEric Paris  *  the Free Software Foundation; either version 2, or (at your option)
75444e298SEric Paris  *  any later version.
85444e298SEric Paris  *
95444e298SEric Paris  *  This program is distributed in the hope that it will be useful,
105444e298SEric Paris  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
115444e298SEric Paris  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
125444e298SEric Paris  *  GNU General Public License for more details.
135444e298SEric Paris  *
145444e298SEric Paris  *  You should have received a copy of the GNU General Public License
155444e298SEric Paris  *  along with this program; see the file COPYING.  If not, write to
165444e298SEric Paris  *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
175444e298SEric Paris  */
185444e298SEric Paris 
195444e298SEric Paris /*
205444e298SEric Paris  * fsnotify inode mark locking/lifetime/and refcnting
215444e298SEric Paris  *
225444e298SEric Paris  * REFCNT:
23*9756b918SLino Sanfilippo  * The group->recnt and mark->refcnt tell how many "things" in the kernel
24*9756b918SLino Sanfilippo  * currently are referencing the objects. Both kind of objects typically will
25*9756b918SLino Sanfilippo  * live inside the kernel with a refcnt of 2, one for its creation and one for
26*9756b918SLino Sanfilippo  * the reference a group and a mark hold to each other.
27*9756b918SLino Sanfilippo  * If you are holding the appropriate locks, you can take a reference and the
28*9756b918SLino Sanfilippo  * object itself is guaranteed to survive until the reference is dropped.
295444e298SEric Paris  *
305444e298SEric Paris  * LOCKING:
31*9756b918SLino Sanfilippo  * There are 3 locks involved with fsnotify inode marks and they MUST be taken
32*9756b918SLino Sanfilippo  * in order as follows:
335444e298SEric Paris  *
34*9756b918SLino Sanfilippo  * group->mark_mutex
355444e298SEric Paris  * mark->lock
365444e298SEric Paris  * inode->i_lock
375444e298SEric Paris  *
38*9756b918SLino Sanfilippo  * group->mark_mutex protects the marks_list anchored inside a given group and
39*9756b918SLino Sanfilippo  * each mark is hooked via the g_list.  It also protects the groups private
40*9756b918SLino Sanfilippo  * data (i.e group limits).
41*9756b918SLino Sanfilippo 
42*9756b918SLino Sanfilippo  * mark->lock protects the marks attributes like its masks and flags.
43*9756b918SLino Sanfilippo  * Furthermore it protects the access to a reference of the group that the mark
44*9756b918SLino Sanfilippo  * is assigned to as well as the access to a reference of the inode/vfsmount
45*9756b918SLino Sanfilippo  * that is being watched by the mark.
465444e298SEric Paris  *
475444e298SEric Paris  * inode->i_lock protects the i_fsnotify_marks list anchored inside a
485444e298SEric Paris  * given inode and each mark is hooked via the i_list. (and sorta the
495444e298SEric Paris  * free_i_list)
505444e298SEric Paris  *
515444e298SEric Paris  *
525444e298SEric Paris  * LIFETIME:
535444e298SEric Paris  * Inode marks survive between when they are added to an inode and when their
545444e298SEric Paris  * refcnt==0.
555444e298SEric Paris  *
565444e298SEric Paris  * The inode mark can be cleared for a number of different reasons including:
575444e298SEric Paris  * - The inode is unlinked for the last time.  (fsnotify_inode_remove)
585444e298SEric Paris  * - The inode is being evicted from cache. (fsnotify_inode_delete)
595444e298SEric Paris  * - The fs the inode is on is unmounted.  (fsnotify_inode_delete/fsnotify_unmount_inodes)
605444e298SEric Paris  * - Something explicitly requests that it be removed.  (fsnotify_destroy_mark)
615444e298SEric Paris  * - The fsnotify_group associated with the mark is going away and all such marks
625444e298SEric Paris  *   need to be cleaned up. (fsnotify_clear_marks_by_group)
635444e298SEric Paris  *
645444e298SEric Paris  * Worst case we are given an inode and need to clean up all the marks on that
655444e298SEric Paris  * inode.  We take i_lock and walk the i_fsnotify_marks safely.  For each
665444e298SEric Paris  * mark on the list we take a reference (so the mark can't disappear under us).
675444e298SEric Paris  * We remove that mark form the inode's list of marks and we add this mark to a
68*9756b918SLino Sanfilippo  * private list anchored on the stack using i_free_list; we walk i_free_list
69*9756b918SLino Sanfilippo  * and before we destroy the mark we make sure that we dont race with a
70*9756b918SLino Sanfilippo  * concurrent destroy_group by getting a ref to the marks group and taking the
71*9756b918SLino Sanfilippo  * groups mutex.
72*9756b918SLino Sanfilippo 
735444e298SEric Paris  * Very similarly for freeing by group, except we use free_g_list.
745444e298SEric Paris  *
755444e298SEric Paris  * This has the very interesting property of being able to run concurrently with
765444e298SEric Paris  * any (or all) other directions.
775444e298SEric Paris  */
785444e298SEric Paris 
795444e298SEric Paris #include <linux/fs.h>
805444e298SEric Paris #include <linux/init.h>
815444e298SEric Paris #include <linux/kernel.h>
8275c1be48SEric Paris #include <linux/kthread.h>
835444e298SEric Paris #include <linux/module.h>
845444e298SEric Paris #include <linux/mutex.h>
855444e298SEric Paris #include <linux/slab.h>
865444e298SEric Paris #include <linux/spinlock.h>
8775c1be48SEric Paris #include <linux/srcu.h>
885444e298SEric Paris 
8960063497SArun Sharma #include <linux/atomic.h>
905444e298SEric Paris 
915444e298SEric Paris #include <linux/fsnotify_backend.h>
925444e298SEric Paris #include "fsnotify.h"
935444e298SEric Paris 
9475c1be48SEric Paris struct srcu_struct fsnotify_mark_srcu;
9575c1be48SEric Paris static DEFINE_SPINLOCK(destroy_lock);
9675c1be48SEric Paris static LIST_HEAD(destroy_list);
9775c1be48SEric Paris static DECLARE_WAIT_QUEUE_HEAD(destroy_waitq);
9875c1be48SEric Paris 
995444e298SEric Paris void fsnotify_get_mark(struct fsnotify_mark *mark)
1005444e298SEric Paris {
1015444e298SEric Paris 	atomic_inc(&mark->refcnt);
1025444e298SEric Paris }
1035444e298SEric Paris 
1045444e298SEric Paris void fsnotify_put_mark(struct fsnotify_mark *mark)
1055444e298SEric Paris {
10623e964c2SLino Sanfilippo 	if (atomic_dec_and_test(&mark->refcnt)) {
10723e964c2SLino Sanfilippo 		if (mark->group)
10823e964c2SLino Sanfilippo 			fsnotify_put_group(mark->group);
1095444e298SEric Paris 		mark->free_mark(mark);
1105444e298SEric Paris 	}
11123e964c2SLino Sanfilippo }
1125444e298SEric Paris 
1135444e298SEric Paris /*
1145444e298SEric Paris  * Any time a mark is getting freed we end up here.
1155444e298SEric Paris  * The caller had better be holding a reference to this mark so we don't actually
1165444e298SEric Paris  * do the final put under the mark->lock
1175444e298SEric Paris  */
118d5a335b8SLino Sanfilippo void fsnotify_destroy_mark_locked(struct fsnotify_mark *mark,
119e2a29943SLino Sanfilippo 				  struct fsnotify_group *group)
1205444e298SEric Paris {
1210d48b7f0SEric Paris 	struct inode *inode = NULL;
1225444e298SEric Paris 
123d5a335b8SLino Sanfilippo 	BUG_ON(!mutex_is_locked(&group->mark_mutex));
124d5a335b8SLino Sanfilippo 
125104d06f0SLino Sanfilippo 	spin_lock(&mark->lock);
1265444e298SEric Paris 
127700307a2SEric Paris 	/* something else already called this function on this mark */
128700307a2SEric Paris 	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
1295444e298SEric Paris 		spin_unlock(&mark->lock);
130e2a29943SLino Sanfilippo 		return;
1315444e298SEric Paris 	}
1325444e298SEric Paris 
133700307a2SEric Paris 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
134700307a2SEric Paris 
1350d48b7f0SEric Paris 	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
1360d48b7f0SEric Paris 		inode = mark->i.inode;
137b31d397eSEric Paris 		fsnotify_destroy_inode_mark(mark);
1380d48b7f0SEric Paris 	} else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
1390d48b7f0SEric Paris 		fsnotify_destroy_vfsmount_mark(mark);
1405444e298SEric Paris 	else
1415444e298SEric Paris 		BUG();
1425444e298SEric Paris 
1435444e298SEric Paris 	list_del_init(&mark->g_list);
1445444e298SEric Paris 
1455444e298SEric Paris 	spin_unlock(&mark->lock);
146d5a335b8SLino Sanfilippo 
1476960b0d9SLino Sanfilippo 	if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
1486960b0d9SLino Sanfilippo 		iput(inode);
149d5a335b8SLino Sanfilippo 	/* release lock temporarily */
150986ab098SLino Sanfilippo 	mutex_unlock(&group->mark_mutex);
1515444e298SEric Paris 
15275c1be48SEric Paris 	spin_lock(&destroy_lock);
15375c1be48SEric Paris 	list_add(&mark->destroy_list, &destroy_list);
15475c1be48SEric Paris 	spin_unlock(&destroy_lock);
15575c1be48SEric Paris 	wake_up(&destroy_waitq);
1566960b0d9SLino Sanfilippo 	/*
1576960b0d9SLino Sanfilippo 	 * We don't necessarily have a ref on mark from caller so the above destroy
1586960b0d9SLino Sanfilippo 	 * may have actually freed it, unless this group provides a 'freeing_mark'
1596960b0d9SLino Sanfilippo 	 * function which must be holding a reference.
1606960b0d9SLino Sanfilippo 	 */
16175c1be48SEric Paris 
1625444e298SEric Paris 	/*
1635444e298SEric Paris 	 * Some groups like to know that marks are being freed.  This is a
1645444e298SEric Paris 	 * callback to the group function to let it know that this mark
1655444e298SEric Paris 	 * is being freed.
1665444e298SEric Paris 	 */
1675444e298SEric Paris 	if (group->ops->freeing_mark)
1685444e298SEric Paris 		group->ops->freeing_mark(mark, group);
1695444e298SEric Paris 
1705444e298SEric Paris 	/*
1715444e298SEric Paris 	 * __fsnotify_update_child_dentry_flags(inode);
1725444e298SEric Paris 	 *
1735444e298SEric Paris 	 * I really want to call that, but we can't, we have no idea if the inode
1745444e298SEric Paris 	 * still exists the second we drop the mark->lock.
1755444e298SEric Paris 	 *
1765444e298SEric Paris 	 * The next time an event arrive to this inode from one of it's children
1775444e298SEric Paris 	 * __fsnotify_parent will see that the inode doesn't care about it's
1785444e298SEric Paris 	 * children and will update all of these flags then.  So really this
1795444e298SEric Paris 	 * is just a lazy update (and could be a perf win...)
1805444e298SEric Paris 	 */
1815444e298SEric Paris 
18223e964c2SLino Sanfilippo 	atomic_dec(&group->num_marks);
183d5a335b8SLino Sanfilippo 
1846960b0d9SLino Sanfilippo 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
185d5a335b8SLino Sanfilippo }
186d5a335b8SLino Sanfilippo 
187d5a335b8SLino Sanfilippo void fsnotify_destroy_mark(struct fsnotify_mark *mark,
188d5a335b8SLino Sanfilippo 			   struct fsnotify_group *group)
189d5a335b8SLino Sanfilippo {
1906960b0d9SLino Sanfilippo 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
191d5a335b8SLino Sanfilippo 	fsnotify_destroy_mark_locked(mark, group);
192d5a335b8SLino Sanfilippo 	mutex_unlock(&group->mark_mutex);
1935444e298SEric Paris }
1945444e298SEric Paris 
19590b1e7a5SEric Paris void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
19690b1e7a5SEric Paris {
19790b1e7a5SEric Paris 	assert_spin_locked(&mark->lock);
19890b1e7a5SEric Paris 
19990b1e7a5SEric Paris 	mark->mask = mask;
20090b1e7a5SEric Paris 
20190b1e7a5SEric Paris 	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
20290b1e7a5SEric Paris 		fsnotify_set_inode_mark_mask_locked(mark, mask);
20390b1e7a5SEric Paris }
20490b1e7a5SEric Paris 
20533af5e32SEric Paris void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
20633af5e32SEric Paris {
20733af5e32SEric Paris 	assert_spin_locked(&mark->lock);
20833af5e32SEric Paris 
20933af5e32SEric Paris 	mark->ignored_mask = mask;
21033af5e32SEric Paris }
21190b1e7a5SEric Paris 
2125444e298SEric Paris /*
2135444e298SEric Paris  * Attach an initialized mark to a given group and fs object.
2145444e298SEric Paris  * These marks may be used for the fsnotify backend to determine which
2155444e298SEric Paris  * event types should be delivered to which group.
2165444e298SEric Paris  */
217d5a335b8SLino Sanfilippo int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
2185444e298SEric Paris 			     struct fsnotify_group *group, struct inode *inode,
2195444e298SEric Paris 			     struct vfsmount *mnt, int allow_dups)
2205444e298SEric Paris {
2215444e298SEric Paris 	int ret = 0;
2225444e298SEric Paris 
2235444e298SEric Paris 	BUG_ON(inode && mnt);
2245444e298SEric Paris 	BUG_ON(!inode && !mnt);
225d5a335b8SLino Sanfilippo 	BUG_ON(!mutex_is_locked(&group->mark_mutex));
2265444e298SEric Paris 
2275444e298SEric Paris 	/*
2285444e298SEric Paris 	 * LOCKING ORDER!!!!
229986ab098SLino Sanfilippo 	 * group->mark_mutex
230104d06f0SLino Sanfilippo 	 * mark->lock
2315444e298SEric Paris 	 * inode->i_lock
2325444e298SEric Paris 	 */
233104d06f0SLino Sanfilippo 	spin_lock(&mark->lock);
234700307a2SEric Paris 	mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE;
235700307a2SEric Paris 
23623e964c2SLino Sanfilippo 	fsnotify_get_group(group);
2375444e298SEric Paris 	mark->group = group;
2385444e298SEric Paris 	list_add(&mark->g_list, &group->marks_list);
2395444e298SEric Paris 	atomic_inc(&group->num_marks);
2405444e298SEric Paris 	fsnotify_get_mark(mark); /* for i_list and g_list */
2415444e298SEric Paris 
2425444e298SEric Paris 	if (inode) {
2435444e298SEric Paris 		ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
2445444e298SEric Paris 		if (ret)
2455444e298SEric Paris 			goto err;
2460d48b7f0SEric Paris 	} else if (mnt) {
2470d48b7f0SEric Paris 		ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
2480d48b7f0SEric Paris 		if (ret)
2490d48b7f0SEric Paris 			goto err;
2505444e298SEric Paris 	} else {
2515444e298SEric Paris 		BUG();
2525444e298SEric Paris 	}
2535444e298SEric Paris 
25490b1e7a5SEric Paris 	/* this will pin the object if appropriate */
25590b1e7a5SEric Paris 	fsnotify_set_mark_mask_locked(mark, mark->mask);
2565444e298SEric Paris 	spin_unlock(&mark->lock);
2575444e298SEric Paris 
2585444e298SEric Paris 	if (inode)
2595444e298SEric Paris 		__fsnotify_update_child_dentry_flags(inode);
2605444e298SEric Paris 
2615444e298SEric Paris 	return ret;
2625444e298SEric Paris err:
263700307a2SEric Paris 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
2645444e298SEric Paris 	list_del_init(&mark->g_list);
26523e964c2SLino Sanfilippo 	fsnotify_put_group(group);
26675c1be48SEric Paris 	mark->group = NULL;
2675444e298SEric Paris 	atomic_dec(&group->num_marks);
2685444e298SEric Paris 
2695444e298SEric Paris 	spin_unlock(&mark->lock);
2705444e298SEric Paris 
27175c1be48SEric Paris 	spin_lock(&destroy_lock);
27275c1be48SEric Paris 	list_add(&mark->destroy_list, &destroy_list);
27375c1be48SEric Paris 	spin_unlock(&destroy_lock);
27475c1be48SEric Paris 	wake_up(&destroy_waitq);
27575c1be48SEric Paris 
2765444e298SEric Paris 	return ret;
2775444e298SEric Paris }
2785444e298SEric Paris 
279d5a335b8SLino Sanfilippo int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
280d5a335b8SLino Sanfilippo 		      struct inode *inode, struct vfsmount *mnt, int allow_dups)
281d5a335b8SLino Sanfilippo {
282d5a335b8SLino Sanfilippo 	int ret;
283d5a335b8SLino Sanfilippo 	mutex_lock(&group->mark_mutex);
284d5a335b8SLino Sanfilippo 	ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
285d5a335b8SLino Sanfilippo 	mutex_unlock(&group->mark_mutex);
286d5a335b8SLino Sanfilippo 	return ret;
287d5a335b8SLino Sanfilippo }
288d5a335b8SLino Sanfilippo 
2895444e298SEric Paris /*
2904d92604cSEric Paris  * clear any marks in a group in which mark->flags & flags is true
2915444e298SEric Paris  */
2924d92604cSEric Paris void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
2934d92604cSEric Paris 					 unsigned int flags)
2945444e298SEric Paris {
2955444e298SEric Paris 	struct fsnotify_mark *lmark, *mark;
2965444e298SEric Paris 
2976960b0d9SLino Sanfilippo 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
2985444e298SEric Paris 	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
2994d92604cSEric Paris 		if (mark->flags & flags) {
3005444e298SEric Paris 			fsnotify_get_mark(mark);
30164c20d2aSLino Sanfilippo 			fsnotify_destroy_mark_locked(mark, group);
30264c20d2aSLino Sanfilippo 			fsnotify_put_mark(mark);
3035444e298SEric Paris 		}
3044d92604cSEric Paris 	}
305986ab098SLino Sanfilippo 	mutex_unlock(&group->mark_mutex);
3065444e298SEric Paris }
3075444e298SEric Paris 
3084d92604cSEric Paris /*
3094d92604cSEric Paris  * Given a group, destroy all of the marks associated with that group.
3104d92604cSEric Paris  */
3114d92604cSEric Paris void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
3124d92604cSEric Paris {
3134d92604cSEric Paris 	fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
3144d92604cSEric Paris }
3154d92604cSEric Paris 
3165444e298SEric Paris void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
3175444e298SEric Paris {
3185444e298SEric Paris 	assert_spin_locked(&old->lock);
3195444e298SEric Paris 	new->i.inode = old->i.inode;
3205444e298SEric Paris 	new->m.mnt = old->m.mnt;
32123e964c2SLino Sanfilippo 	if (old->group)
32223e964c2SLino Sanfilippo 		fsnotify_get_group(old->group);
3235444e298SEric Paris 	new->group = old->group;
3245444e298SEric Paris 	new->mask = old->mask;
3255444e298SEric Paris 	new->free_mark = old->free_mark;
3265444e298SEric Paris }
3275444e298SEric Paris 
3285444e298SEric Paris /*
3295444e298SEric Paris  * Nothing fancy, just initialize lists and locks and counters.
3305444e298SEric Paris  */
3315444e298SEric Paris void fsnotify_init_mark(struct fsnotify_mark *mark,
3325444e298SEric Paris 			void (*free_mark)(struct fsnotify_mark *mark))
3335444e298SEric Paris {
334ba643f04SEric Paris 	memset(mark, 0, sizeof(*mark));
3355444e298SEric Paris 	spin_lock_init(&mark->lock);
3365444e298SEric Paris 	atomic_set(&mark->refcnt, 1);
3375444e298SEric Paris 	mark->free_mark = free_mark;
3385444e298SEric Paris }
33975c1be48SEric Paris 
34075c1be48SEric Paris static int fsnotify_mark_destroy(void *ignored)
34175c1be48SEric Paris {
34275c1be48SEric Paris 	struct fsnotify_mark *mark, *next;
34375c1be48SEric Paris 	LIST_HEAD(private_destroy_list);
34475c1be48SEric Paris 
34575c1be48SEric Paris 	for (;;) {
34675c1be48SEric Paris 		spin_lock(&destroy_lock);
3478778abb9SAndreas Gruenbacher 		/* exchange the list head */
3488778abb9SAndreas Gruenbacher 		list_replace_init(&destroy_list, &private_destroy_list);
34975c1be48SEric Paris 		spin_unlock(&destroy_lock);
35075c1be48SEric Paris 
35175c1be48SEric Paris 		synchronize_srcu(&fsnotify_mark_srcu);
35275c1be48SEric Paris 
35375c1be48SEric Paris 		list_for_each_entry_safe(mark, next, &private_destroy_list, destroy_list) {
35475c1be48SEric Paris 			list_del_init(&mark->destroy_list);
35575c1be48SEric Paris 			fsnotify_put_mark(mark);
35675c1be48SEric Paris 		}
35775c1be48SEric Paris 
35875c1be48SEric Paris 		wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
35975c1be48SEric Paris 	}
36075c1be48SEric Paris 
36175c1be48SEric Paris 	return 0;
36275c1be48SEric Paris }
36375c1be48SEric Paris 
36475c1be48SEric Paris static int __init fsnotify_mark_init(void)
36575c1be48SEric Paris {
36675c1be48SEric Paris 	struct task_struct *thread;
36775c1be48SEric Paris 
36875c1be48SEric Paris 	thread = kthread_run(fsnotify_mark_destroy, NULL,
36975c1be48SEric Paris 			     "fsnotify_mark");
37075c1be48SEric Paris 	if (IS_ERR(thread))
37175c1be48SEric Paris 		panic("unable to start fsnotify mark destruction thread.");
37275c1be48SEric Paris 
37375c1be48SEric Paris 	return 0;
37475c1be48SEric Paris }
37575c1be48SEric Paris device_initcall(fsnotify_mark_init);
376