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
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 * Worst case we are given an inode and need to clean up all the marks on that
65 * inode. We take i_lock and walk the i_fsnotify_marks safely. For each
66 * mark on the list we take a reference (so the mark can't disappear under us).
67 * We remove that mark form the inode's list of marks and we add this mark to a
68 * private list anchored on the stack using i_free_list; we walk i_free_list
69 * and before we destroy the mark we make sure that we dont race with a
70 * concurrent destroy_group by getting a ref to the marks group and taking the
73 * Very similarly for freeing by group, except we use free_g_list.
75 * This has the very interesting property of being able to run concurrently with
76 * any (or all) other directions.
80 #include <linux/init.h>
81 #include <linux/kernel.h>
82 #include <linux/kthread.h>
83 #include <linux/module.h>
84 #include <linux/mutex.h>
85 #include <linux/slab.h>
86 #include <linux/spinlock.h>
87 #include <linux/srcu.h>
89 #include <linux/atomic.h>
91 #include <linux/fsnotify_backend.h>
94 struct srcu_struct fsnotify_mark_srcu
;
96 void fsnotify_get_mark(struct fsnotify_mark
*mark
)
98 atomic_inc(&mark
->refcnt
);
101 void fsnotify_put_mark(struct fsnotify_mark
*mark
)
103 if (atomic_dec_and_test(&mark
->refcnt
)) {
105 fsnotify_put_group(mark
->group
);
106 mark
->free_mark(mark
);
110 /* Calculate mask of events for a list of marks */
111 u32
fsnotify_recalc_mask(struct hlist_head
*head
)
114 struct fsnotify_mark
*mark
;
116 hlist_for_each_entry(mark
, head
, obj_list
)
117 new_mask
|= mark
->mask
;
122 * Remove mark from inode / vfsmount list, group list, drop inode reference
125 * Must be called with group->mark_mutex held.
127 void fsnotify_detach_mark(struct fsnotify_mark
*mark
)
129 struct inode
*inode
= NULL
;
130 struct fsnotify_group
*group
= mark
->group
;
132 BUG_ON(!mutex_is_locked(&group
->mark_mutex
));
134 spin_lock(&mark
->lock
);
136 /* something else already called this function on this mark */
137 if (!(mark
->flags
& FSNOTIFY_MARK_FLAG_ATTACHED
)) {
138 spin_unlock(&mark
->lock
);
142 mark
->flags
&= ~FSNOTIFY_MARK_FLAG_ATTACHED
;
144 if (mark
->flags
& FSNOTIFY_MARK_FLAG_INODE
) {
146 fsnotify_destroy_inode_mark(mark
);
147 } else if (mark
->flags
& FSNOTIFY_MARK_FLAG_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
);
162 if (inode
&& (mark
->flags
& FSNOTIFY_MARK_FLAG_OBJECT_PINNED
))
165 atomic_dec(&group
->num_marks
);
169 fsnotify_mark_free_rcu(struct rcu_head
*rcu
)
171 struct fsnotify_mark
*mark
;
173 mark
= container_of(rcu
, struct fsnotify_mark
, g_rcu
);
174 fsnotify_put_mark(mark
);
178 * Free fsnotify mark. The freeing is actually happening from a call_srcu
179 * callback. Caller must have a reference to the mark or be protected by
180 * fsnotify_mark_srcu.
182 void fsnotify_free_mark(struct fsnotify_mark
*mark
)
184 struct fsnotify_group
*group
= mark
->group
;
186 spin_lock(&mark
->lock
);
187 /* something else already called this function on this mark */
188 if (!(mark
->flags
& FSNOTIFY_MARK_FLAG_ALIVE
)) {
189 spin_unlock(&mark
->lock
);
192 mark
->flags
&= ~FSNOTIFY_MARK_FLAG_ALIVE
;
193 spin_unlock(&mark
->lock
);
195 call_srcu(&fsnotify_mark_srcu
, &mark
->g_rcu
, fsnotify_mark_free_rcu
);
198 * Some groups like to know that marks are being freed. This is a
199 * callback to the group function to let it know that this mark
202 if (group
->ops
->freeing_mark
)
203 group
->ops
->freeing_mark(mark
, group
);
206 void fsnotify_destroy_mark(struct fsnotify_mark
*mark
,
207 struct fsnotify_group
*group
)
209 mutex_lock_nested(&group
->mark_mutex
, SINGLE_DEPTH_NESTING
);
210 fsnotify_detach_mark(mark
);
211 mutex_unlock(&group
->mark_mutex
);
212 fsnotify_free_mark(mark
);
215 void fsnotify_destroy_marks(struct hlist_head
*head
, spinlock_t
*lock
)
217 struct fsnotify_mark
*mark
;
221 * We have to be careful since we can race with e.g.
222 * fsnotify_clear_marks_by_group() and once we drop 'lock',
223 * mark can get removed from the obj_list and destroyed. But
224 * we are holding mark reference so mark cannot be freed and
225 * calling fsnotify_destroy_mark() more than once is fine.
228 if (hlist_empty(head
)) {
232 mark
= hlist_entry(head
->first
, struct fsnotify_mark
, obj_list
);
234 * We don't update i_fsnotify_mask / mnt_fsnotify_mask here
235 * since inode / mount is going away anyway. So just remove
236 * mark from the list.
238 hlist_del_init_rcu(&mark
->obj_list
);
239 fsnotify_get_mark(mark
);
241 fsnotify_destroy_mark(mark
, mark
->group
);
242 fsnotify_put_mark(mark
);
246 void fsnotify_set_mark_mask_locked(struct fsnotify_mark
*mark
, __u32 mask
)
248 assert_spin_locked(&mark
->lock
);
252 if (mark
->flags
& FSNOTIFY_MARK_FLAG_INODE
)
253 fsnotify_set_inode_mark_mask_locked(mark
, mask
);
256 void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark
*mark
, __u32 mask
)
258 assert_spin_locked(&mark
->lock
);
260 mark
->ignored_mask
= mask
;
264 * Sorting function for lists of fsnotify marks.
266 * Fanotify supports different notification classes (reflected as priority of
267 * notification group). Events shall be passed to notification groups in
268 * decreasing priority order. To achieve this marks in notification lists for
269 * inodes and vfsmounts are sorted so that priorities of corresponding groups
272 * Furthermore correct handling of the ignore mask requires processing inode
273 * and vfsmount marks of each group together. Using the group address as
274 * further sort criterion provides a unique sorting order and thus we can
275 * merge inode and vfsmount lists of marks in linear time and find groups
276 * present in both lists.
278 * A return value of 1 signifies that b has priority over a.
279 * A return value of 0 signifies that the two marks have to be handled together.
280 * A return value of -1 signifies that a has priority over b.
282 int fsnotify_compare_groups(struct fsnotify_group
*a
, struct fsnotify_group
*b
)
290 if (a
->priority
< b
->priority
)
292 if (a
->priority
> b
->priority
)
299 /* Add mark into proper place in given list of marks */
300 int fsnotify_add_mark_list(struct hlist_head
*head
, struct fsnotify_mark
*mark
,
303 struct fsnotify_mark
*lmark
, *last
= NULL
;
306 /* is mark the first mark? */
307 if (hlist_empty(head
)) {
308 hlist_add_head_rcu(&mark
->obj_list
, head
);
312 /* should mark be in the middle of the current list? */
313 hlist_for_each_entry(lmark
, head
, obj_list
) {
316 if ((lmark
->group
== mark
->group
) && !allow_dups
)
319 cmp
= fsnotify_compare_groups(lmark
->group
, mark
->group
);
321 hlist_add_before_rcu(&mark
->obj_list
, &lmark
->obj_list
);
326 BUG_ON(last
== NULL
);
327 /* mark should be the last entry. last is the current last entry */
328 hlist_add_behind_rcu(&mark
->obj_list
, &last
->obj_list
);
333 * Attach an initialized mark to a given group and fs object.
334 * These marks may be used for the fsnotify backend to determine which
335 * event types should be delivered to which group.
337 int fsnotify_add_mark_locked(struct fsnotify_mark
*mark
,
338 struct fsnotify_group
*group
, struct inode
*inode
,
339 struct vfsmount
*mnt
, int allow_dups
)
343 BUG_ON(inode
&& mnt
);
344 BUG_ON(!inode
&& !mnt
);
345 BUG_ON(!mutex_is_locked(&group
->mark_mutex
));
353 spin_lock(&mark
->lock
);
354 mark
->flags
|= FSNOTIFY_MARK_FLAG_ALIVE
| FSNOTIFY_MARK_FLAG_ATTACHED
;
356 fsnotify_get_group(group
);
358 list_add(&mark
->g_list
, &group
->marks_list
);
359 atomic_inc(&group
->num_marks
);
360 fsnotify_get_mark(mark
); /* for i_list and g_list */
363 ret
= fsnotify_add_inode_mark(mark
, group
, inode
, allow_dups
);
367 ret
= fsnotify_add_vfsmount_mark(mark
, group
, mnt
, allow_dups
);
374 /* this will pin the object if appropriate */
375 fsnotify_set_mark_mask_locked(mark
, mark
->mask
);
376 spin_unlock(&mark
->lock
);
379 __fsnotify_update_child_dentry_flags(inode
);
383 mark
->flags
&= ~FSNOTIFY_MARK_FLAG_ALIVE
;
384 list_del_init(&mark
->g_list
);
385 fsnotify_put_group(group
);
387 atomic_dec(&group
->num_marks
);
389 spin_unlock(&mark
->lock
);
391 call_srcu(&fsnotify_mark_srcu
, &mark
->g_rcu
, fsnotify_mark_free_rcu
);
395 int fsnotify_add_mark(struct fsnotify_mark
*mark
, struct fsnotify_group
*group
,
396 struct inode
*inode
, struct vfsmount
*mnt
, int allow_dups
)
399 mutex_lock(&group
->mark_mutex
);
400 ret
= fsnotify_add_mark_locked(mark
, group
, inode
, mnt
, allow_dups
);
401 mutex_unlock(&group
->mark_mutex
);
406 * Given a list of marks, find the mark associated with given group. If found
407 * take a reference to that mark and return it, else return NULL.
409 struct fsnotify_mark
*fsnotify_find_mark(struct hlist_head
*head
,
410 struct fsnotify_group
*group
)
412 struct fsnotify_mark
*mark
;
414 hlist_for_each_entry(mark
, head
, obj_list
) {
415 if (mark
->group
== group
) {
416 fsnotify_get_mark(mark
);
424 * clear any marks in a group in which mark->flags & flags is true
426 void fsnotify_clear_marks_by_group_flags(struct fsnotify_group
*group
,
429 struct fsnotify_mark
*lmark
, *mark
;
433 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
434 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
435 * to_free list so we have to use mark_mutex even when accessing that
436 * list. And freeing mark requires us to drop mark_mutex. So we can
437 * reliably free only the first mark in the list. That's why we first
438 * move marks to free to to_free list in one go and then free marks in
439 * to_free list one by one.
441 mutex_lock_nested(&group
->mark_mutex
, SINGLE_DEPTH_NESTING
);
442 list_for_each_entry_safe(mark
, lmark
, &group
->marks_list
, g_list
) {
443 if (mark
->flags
& flags
)
444 list_move(&mark
->g_list
, &to_free
);
446 mutex_unlock(&group
->mark_mutex
);
449 mutex_lock_nested(&group
->mark_mutex
, SINGLE_DEPTH_NESTING
);
450 if (list_empty(&to_free
)) {
451 mutex_unlock(&group
->mark_mutex
);
454 mark
= list_first_entry(&to_free
, struct fsnotify_mark
, g_list
);
455 fsnotify_get_mark(mark
);
456 fsnotify_detach_mark(mark
);
457 mutex_unlock(&group
->mark_mutex
);
458 fsnotify_free_mark(mark
);
459 fsnotify_put_mark(mark
);
464 * Given a group, destroy all of the marks associated with that group.
466 void fsnotify_clear_marks_by_group(struct fsnotify_group
*group
)
468 fsnotify_clear_marks_by_group_flags(group
, (unsigned int)-1);
471 void fsnotify_duplicate_mark(struct fsnotify_mark
*new, struct fsnotify_mark
*old
)
473 assert_spin_locked(&old
->lock
);
474 new->inode
= old
->inode
;
477 fsnotify_get_group(old
->group
);
478 new->group
= old
->group
;
479 new->mask
= old
->mask
;
480 new->free_mark
= old
->free_mark
;
484 * Nothing fancy, just initialize lists and locks and counters.
486 void fsnotify_init_mark(struct fsnotify_mark
*mark
,
487 void (*free_mark
)(struct fsnotify_mark
*mark
))
489 memset(mark
, 0, sizeof(*mark
));
490 spin_lock_init(&mark
->lock
);
491 atomic_set(&mark
->refcnt
, 1);
492 mark
->free_mark
= free_mark
;