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;
95 static DEFINE_SPINLOCK(destroy_lock);
96 static LIST_HEAD(destroy_list);
97 static DECLARE_WAIT_QUEUE_HEAD(destroy_waitq);
99 void fsnotify_get_mark(struct fsnotify_mark *mark)
101 atomic_inc(&mark->refcnt);
104 void fsnotify_put_mark(struct fsnotify_mark *mark)
106 if (atomic_dec_and_test(&mark->refcnt)) {
108 fsnotify_put_group(mark->group);
109 mark->free_mark(mark);
113 /* Calculate mask of events for a list of marks */
114 u32 fsnotify_recalc_mask(struct hlist_head *head)
117 struct fsnotify_mark *mark;
119 hlist_for_each_entry(mark, head, obj_list)
120 new_mask |= mark->mask;
125 * Any time a mark is getting freed we end up here.
126 * The caller had better be holding a reference to this mark so we don't actually
127 * do the final put under the mark->lock
129 void fsnotify_destroy_mark_locked(struct fsnotify_mark *mark,
130 struct fsnotify_group *group)
132 struct inode *inode = NULL;
134 BUG_ON(!mutex_is_locked(&group->mark_mutex));
136 spin_lock(&mark->lock);
138 /* something else already called this function on this mark */
139 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
140 spin_unlock(&mark->lock);
144 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
146 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
148 fsnotify_destroy_inode_mark(mark);
149 } else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
150 fsnotify_destroy_vfsmount_mark(mark);
154 list_del_init(&mark->g_list);
155 spin_unlock(&mark->lock);
157 if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
160 spin_lock(&destroy_lock);
161 list_add(&mark->g_list, &destroy_list);
162 spin_unlock(&destroy_lock);
163 wake_up(&destroy_waitq);
166 * __fsnotify_update_child_dentry_flags(inode);
168 * I really want to call that, but we can't, we have no idea if the inode
169 * still exists the second we drop the mark->lock.
171 * The next time an event arrive to this inode from one of it's children
172 * __fsnotify_parent will see that the inode doesn't care about it's
173 * children and will update all of these flags then. So really this
174 * is just a lazy update (and could be a perf win...)
177 atomic_dec(&group->num_marks);
180 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
181 struct fsnotify_group *group)
183 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
184 fsnotify_destroy_mark_locked(mark, group);
185 mutex_unlock(&group->mark_mutex);
189 * Destroy all marks in the given list. The marks must be already detached from
190 * the original inode / vfsmount. Note that we can race with
191 * fsnotify_clear_marks_by_group_flags(). However we hold a reference to each
192 * mark so they won't get freed from under us and nobody else touches our
193 * free_list list_head.
195 void fsnotify_destroy_marks(struct list_head *to_free)
197 struct fsnotify_mark *mark, *lmark;
198 struct fsnotify_group *group;
200 list_for_each_entry_safe(mark, lmark, to_free, free_list) {
201 spin_lock(&mark->lock);
202 fsnotify_get_group(mark->group);
204 spin_unlock(&mark->lock);
206 fsnotify_destroy_mark(mark, group);
207 fsnotify_put_mark(mark);
208 fsnotify_put_group(group);
212 void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
214 assert_spin_locked(&mark->lock);
218 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
219 fsnotify_set_inode_mark_mask_locked(mark, mask);
222 void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
224 assert_spin_locked(&mark->lock);
226 mark->ignored_mask = mask;
230 * Sorting function for lists of fsnotify marks.
232 * Fanotify supports different notification classes (reflected as priority of
233 * notification group). Events shall be passed to notification groups in
234 * decreasing priority order. To achieve this marks in notification lists for
235 * inodes and vfsmounts are sorted so that priorities of corresponding groups
238 * Furthermore correct handling of the ignore mask requires processing inode
239 * and vfsmount marks of each group together. Using the group address as
240 * further sort criterion provides a unique sorting order and thus we can
241 * merge inode and vfsmount lists of marks in linear time and find groups
242 * present in both lists.
244 * A return value of 1 signifies that b has priority over a.
245 * A return value of 0 signifies that the two marks have to be handled together.
246 * A return value of -1 signifies that a has priority over b.
248 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
256 if (a->priority < b->priority)
258 if (a->priority > b->priority)
265 /* Add mark into proper place in given list of marks */
266 int fsnotify_add_mark_list(struct hlist_head *head, struct fsnotify_mark *mark,
269 struct fsnotify_mark *lmark, *last = NULL;
272 /* is mark the first mark? */
273 if (hlist_empty(head)) {
274 hlist_add_head_rcu(&mark->obj_list, head);
278 /* should mark be in the middle of the current list? */
279 hlist_for_each_entry(lmark, head, obj_list) {
282 if ((lmark->group == mark->group) && !allow_dups)
285 cmp = fsnotify_compare_groups(lmark->group, mark->group);
287 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
292 BUG_ON(last == NULL);
293 /* mark should be the last entry. last is the current last entry */
294 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
299 * Attach an initialized mark to a given group and fs object.
300 * These marks may be used for the fsnotify backend to determine which
301 * event types should be delivered to which group.
303 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
304 struct fsnotify_group *group, struct inode *inode,
305 struct vfsmount *mnt, int allow_dups)
309 BUG_ON(inode && mnt);
310 BUG_ON(!inode && !mnt);
311 BUG_ON(!mutex_is_locked(&group->mark_mutex));
319 spin_lock(&mark->lock);
320 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE;
322 fsnotify_get_group(group);
324 list_add(&mark->g_list, &group->marks_list);
325 atomic_inc(&group->num_marks);
326 fsnotify_get_mark(mark); /* for i_list and g_list */
329 ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
333 ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
340 /* this will pin the object if appropriate */
341 fsnotify_set_mark_mask_locked(mark, mark->mask);
342 spin_unlock(&mark->lock);
345 __fsnotify_update_child_dentry_flags(inode);
349 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
350 list_del_init(&mark->g_list);
351 fsnotify_put_group(group);
353 atomic_dec(&group->num_marks);
355 spin_unlock(&mark->lock);
357 spin_lock(&destroy_lock);
358 list_add(&mark->g_list, &destroy_list);
359 spin_unlock(&destroy_lock);
360 wake_up(&destroy_waitq);
365 int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
366 struct inode *inode, struct vfsmount *mnt, int allow_dups)
369 mutex_lock(&group->mark_mutex);
370 ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
371 mutex_unlock(&group->mark_mutex);
376 * Given a list of marks, find the mark associated with given group. If found
377 * take a reference to that mark and return it, else return NULL.
379 struct fsnotify_mark *fsnotify_find_mark(struct hlist_head *head,
380 struct fsnotify_group *group)
382 struct fsnotify_mark *mark;
384 hlist_for_each_entry(mark, head, obj_list) {
385 if (mark->group == group) {
386 fsnotify_get_mark(mark);
394 * Clear any marks in a group in which mark->flags & flags is true.
396 void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
399 struct fsnotify_mark *lmark, *mark;
401 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
402 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
403 if (mark->flags & flags) {
404 fsnotify_get_mark(mark);
405 fsnotify_destroy_mark_locked(mark, group);
406 fsnotify_put_mark(mark);
409 mutex_unlock(&group->mark_mutex);
413 * Given a group, destroy all of the marks associated with that group.
415 void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
417 fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
420 void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
422 assert_spin_locked(&old->lock);
423 new->inode = old->inode;
426 fsnotify_get_group(old->group);
427 new->group = old->group;
428 new->mask = old->mask;
429 new->free_mark = old->free_mark;
433 * Nothing fancy, just initialize lists and locks and counters.
435 void fsnotify_init_mark(struct fsnotify_mark *mark,
436 void (*free_mark)(struct fsnotify_mark *mark))
438 memset(mark, 0, sizeof(*mark));
439 spin_lock_init(&mark->lock);
440 atomic_set(&mark->refcnt, 1);
441 mark->free_mark = free_mark;
444 static int fsnotify_mark_destroy(void *ignored)
446 struct fsnotify_mark *mark, *next;
447 struct list_head private_destroy_list;
448 struct fsnotify_group *group;
451 spin_lock(&destroy_lock);
452 /* exchange the list head */
453 list_replace_init(&destroy_list, &private_destroy_list);
454 spin_unlock(&destroy_lock);
456 synchronize_srcu(&fsnotify_mark_srcu);
458 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
459 list_del_init(&mark->g_list);
462 * Some groups like to know that marks are being freed.
463 * This is a callback to the group function to let it
464 * know that this mark is being freed.
466 if (group && group->ops->freeing_mark)
467 group->ops->freeing_mark(mark, group);
468 fsnotify_put_mark(mark);
471 wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
477 static int __init fsnotify_mark_init(void)
479 struct task_struct *thread;
481 thread = kthread_run(fsnotify_mark_destroy, NULL,
484 panic("unable to start fsnotify mark destruction thread.");
488 device_initcall(fsnotify_mark_init);