fs: use kmem_cache_zalloc instead
[pv_ops_mirror.git] / fs / inotify.c
blob7457501b95656087cfe27339b94d3bace7585782
1 /*
2 * fs/inotify.c - inode-based file event notifications
4 * Authors:
5 * John McCutchan <ttb@tentacle.dhs.org>
6 * Robert Love <rml@novell.com>
8 * Kernel API added by: Amy Griffis <amy.griffis@hp.com>
10 * Copyright (C) 2005 John McCutchan
11 * Copyright 2006 Hewlett-Packard Development Company, L.P.
13 * This program is free software; you can redistribute it and/or modify it
14 * under the terms of the GNU General Public License as published by the
15 * Free Software Foundation; either version 2, or (at your option) any
16 * later version.
18 * This program is distributed in the hope that it will be useful, but
19 * WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * General Public License for more details.
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/spinlock.h>
27 #include <linux/idr.h>
28 #include <linux/slab.h>
29 #include <linux/fs.h>
30 #include <linux/sched.h>
31 #include <linux/init.h>
32 #include <linux/list.h>
33 #include <linux/writeback.h>
34 #include <linux/inotify.h>
36 static atomic_t inotify_cookie;
39 * Lock ordering:
41 * dentry->d_lock (used to keep d_move() away from dentry->d_parent)
42 * iprune_mutex (synchronize shrink_icache_memory())
43 * inode_lock (protects the super_block->s_inodes list)
44 * inode->inotify_mutex (protects inode->inotify_watches and watches->i_list)
45 * inotify_handle->mutex (protects inotify_handle and watches->h_list)
47 * The inode->inotify_mutex and inotify_handle->mutex and held during execution
48 * of a caller's event handler. Thus, the caller must not hold any locks
49 * taken in their event handler while calling any of the published inotify
50 * interfaces.
54 * Lifetimes of the three main data structures--inotify_handle, inode, and
55 * inotify_watch--are managed by reference count.
57 * inotify_handle: Lifetime is from inotify_init() to inotify_destroy().
58 * Additional references can bump the count via get_inotify_handle() and drop
59 * the count via put_inotify_handle().
61 * inotify_watch: for inotify's purposes, lifetime is from inotify_add_watch()
62 * to remove_watch_no_event(). Additional references can bump the count via
63 * get_inotify_watch() and drop the count via put_inotify_watch(). The caller
64 * is reponsible for the final put after receiving IN_IGNORED, or when using
65 * IN_ONESHOT after receiving the first event. Inotify does the final put if
66 * inotify_destroy() is called.
68 * inode: Pinned so long as the inode is associated with a watch, from
69 * inotify_add_watch() to the final put_inotify_watch().
73 * struct inotify_handle - represents an inotify instance
75 * This structure is protected by the mutex 'mutex'.
77 struct inotify_handle {
78 struct idr idr; /* idr mapping wd -> watch */
79 struct mutex mutex; /* protects this bad boy */
80 struct list_head watches; /* list of watches */
81 atomic_t count; /* reference count */
82 u32 last_wd; /* the last wd allocated */
83 const struct inotify_operations *in_ops; /* inotify caller operations */
86 static inline void get_inotify_handle(struct inotify_handle *ih)
88 atomic_inc(&ih->count);
91 static inline void put_inotify_handle(struct inotify_handle *ih)
93 if (atomic_dec_and_test(&ih->count)) {
94 idr_destroy(&ih->idr);
95 kfree(ih);
99 /**
100 * get_inotify_watch - grab a reference to an inotify_watch
101 * @watch: watch to grab
103 void get_inotify_watch(struct inotify_watch *watch)
105 atomic_inc(&watch->count);
107 EXPORT_SYMBOL_GPL(get_inotify_watch);
110 * put_inotify_watch - decrements the ref count on a given watch. cleans up
111 * watch references if the count reaches zero. inotify_watch is freed by
112 * inotify callers via the destroy_watch() op.
113 * @watch: watch to release
115 void put_inotify_watch(struct inotify_watch *watch)
117 if (atomic_dec_and_test(&watch->count)) {
118 struct inotify_handle *ih = watch->ih;
120 iput(watch->inode);
121 ih->in_ops->destroy_watch(watch);
122 put_inotify_handle(ih);
125 EXPORT_SYMBOL_GPL(put_inotify_watch);
128 * inotify_handle_get_wd - returns the next WD for use by the given handle
130 * Callers must hold ih->mutex. This function can sleep.
132 static int inotify_handle_get_wd(struct inotify_handle *ih,
133 struct inotify_watch *watch)
135 int ret;
137 do {
138 if (unlikely(!idr_pre_get(&ih->idr, GFP_KERNEL)))
139 return -ENOSPC;
140 ret = idr_get_new_above(&ih->idr, watch, ih->last_wd+1, &watch->wd);
141 } while (ret == -EAGAIN);
143 if (likely(!ret))
144 ih->last_wd = watch->wd;
146 return ret;
150 * inotify_inode_watched - returns nonzero if there are watches on this inode
151 * and zero otherwise. We call this lockless, we do not care if we race.
153 static inline int inotify_inode_watched(struct inode *inode)
155 return !list_empty(&inode->inotify_watches);
159 * Get child dentry flag into synch with parent inode.
160 * Flag should always be clear for negative dentrys.
162 static void set_dentry_child_flags(struct inode *inode, int watched)
164 struct dentry *alias;
166 spin_lock(&dcache_lock);
167 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
168 struct dentry *child;
170 list_for_each_entry(child, &alias->d_subdirs, d_u.d_child) {
171 if (!child->d_inode) {
172 WARN_ON(child->d_flags & DCACHE_INOTIFY_PARENT_WATCHED);
173 continue;
175 spin_lock(&child->d_lock);
176 if (watched) {
177 WARN_ON(child->d_flags &
178 DCACHE_INOTIFY_PARENT_WATCHED);
179 child->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED;
180 } else {
181 WARN_ON(!(child->d_flags &
182 DCACHE_INOTIFY_PARENT_WATCHED));
183 child->d_flags&=~DCACHE_INOTIFY_PARENT_WATCHED;
185 spin_unlock(&child->d_lock);
188 spin_unlock(&dcache_lock);
192 * inotify_find_handle - find the watch associated with the given inode and
193 * handle
195 * Callers must hold inode->inotify_mutex.
197 static struct inotify_watch *inode_find_handle(struct inode *inode,
198 struct inotify_handle *ih)
200 struct inotify_watch *watch;
202 list_for_each_entry(watch, &inode->inotify_watches, i_list) {
203 if (watch->ih == ih)
204 return watch;
207 return NULL;
211 * remove_watch_no_event - remove watch without the IN_IGNORED event.
213 * Callers must hold both inode->inotify_mutex and ih->mutex.
215 static void remove_watch_no_event(struct inotify_watch *watch,
216 struct inotify_handle *ih)
218 list_del(&watch->i_list);
219 list_del(&watch->h_list);
221 if (!inotify_inode_watched(watch->inode))
222 set_dentry_child_flags(watch->inode, 0);
224 idr_remove(&ih->idr, watch->wd);
228 * inotify_remove_watch_locked - Remove a watch from both the handle and the
229 * inode. Sends the IN_IGNORED event signifying that the inode is no longer
230 * watched. May be invoked from a caller's event handler.
231 * @ih: inotify handle associated with watch
232 * @watch: watch to remove
234 * Callers must hold both inode->inotify_mutex and ih->mutex.
236 void inotify_remove_watch_locked(struct inotify_handle *ih,
237 struct inotify_watch *watch)
239 remove_watch_no_event(watch, ih);
240 ih->in_ops->handle_event(watch, watch->wd, IN_IGNORED, 0, NULL, NULL);
242 EXPORT_SYMBOL_GPL(inotify_remove_watch_locked);
244 /* Kernel API for producing events */
247 * inotify_d_instantiate - instantiate dcache entry for inode
249 void inotify_d_instantiate(struct dentry *entry, struct inode *inode)
251 struct dentry *parent;
253 if (!inode)
254 return;
256 WARN_ON(entry->d_flags & DCACHE_INOTIFY_PARENT_WATCHED);
257 spin_lock(&entry->d_lock);
258 parent = entry->d_parent;
259 if (parent->d_inode && inotify_inode_watched(parent->d_inode))
260 entry->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED;
261 spin_unlock(&entry->d_lock);
265 * inotify_d_move - dcache entry has been moved
267 void inotify_d_move(struct dentry *entry)
269 struct dentry *parent;
271 parent = entry->d_parent;
272 if (inotify_inode_watched(parent->d_inode))
273 entry->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED;
274 else
275 entry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
279 * inotify_inode_queue_event - queue an event to all watches on this inode
280 * @inode: inode event is originating from
281 * @mask: event mask describing this event
282 * @cookie: cookie for synchronization, or zero
283 * @name: filename, if any
284 * @n_inode: inode associated with name
286 void inotify_inode_queue_event(struct inode *inode, u32 mask, u32 cookie,
287 const char *name, struct inode *n_inode)
289 struct inotify_watch *watch, *next;
291 if (!inotify_inode_watched(inode))
292 return;
294 mutex_lock(&inode->inotify_mutex);
295 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
296 u32 watch_mask = watch->mask;
297 if (watch_mask & mask) {
298 struct inotify_handle *ih= watch->ih;
299 mutex_lock(&ih->mutex);
300 if (watch_mask & IN_ONESHOT)
301 remove_watch_no_event(watch, ih);
302 ih->in_ops->handle_event(watch, watch->wd, mask, cookie,
303 name, n_inode);
304 mutex_unlock(&ih->mutex);
307 mutex_unlock(&inode->inotify_mutex);
309 EXPORT_SYMBOL_GPL(inotify_inode_queue_event);
312 * inotify_dentry_parent_queue_event - queue an event to a dentry's parent
313 * @dentry: the dentry in question, we queue against this dentry's parent
314 * @mask: event mask describing this event
315 * @cookie: cookie for synchronization, or zero
316 * @name: filename, if any
318 void inotify_dentry_parent_queue_event(struct dentry *dentry, u32 mask,
319 u32 cookie, const char *name)
321 struct dentry *parent;
322 struct inode *inode;
324 if (!(dentry->d_flags & DCACHE_INOTIFY_PARENT_WATCHED))
325 return;
327 spin_lock(&dentry->d_lock);
328 parent = dentry->d_parent;
329 inode = parent->d_inode;
331 if (inotify_inode_watched(inode)) {
332 dget(parent);
333 spin_unlock(&dentry->d_lock);
334 inotify_inode_queue_event(inode, mask, cookie, name,
335 dentry->d_inode);
336 dput(parent);
337 } else
338 spin_unlock(&dentry->d_lock);
340 EXPORT_SYMBOL_GPL(inotify_dentry_parent_queue_event);
343 * inotify_get_cookie - return a unique cookie for use in synchronizing events.
345 u32 inotify_get_cookie(void)
347 return atomic_inc_return(&inotify_cookie);
349 EXPORT_SYMBOL_GPL(inotify_get_cookie);
352 * inotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
353 * @list: list of inodes being unmounted (sb->s_inodes)
355 * Called with inode_lock held, protecting the unmounting super block's list
356 * of inodes, and with iprune_mutex held, keeping shrink_icache_memory() at bay.
357 * We temporarily drop inode_lock, however, and CAN block.
359 void inotify_unmount_inodes(struct list_head *list)
361 struct inode *inode, *next_i, *need_iput = NULL;
363 list_for_each_entry_safe(inode, next_i, list, i_sb_list) {
364 struct inotify_watch *watch, *next_w;
365 struct inode *need_iput_tmp;
366 struct list_head *watches;
369 * If i_count is zero, the inode cannot have any watches and
370 * doing an __iget/iput with MS_ACTIVE clear would actually
371 * evict all inodes with zero i_count from icache which is
372 * unnecessarily violent and may in fact be illegal to do.
374 if (!atomic_read(&inode->i_count))
375 continue;
378 * We cannot __iget() an inode in state I_CLEAR, I_FREEING, or
379 * I_WILL_FREE which is fine because by that point the inode
380 * cannot have any associated watches.
382 if (inode->i_state & (I_CLEAR | I_FREEING | I_WILL_FREE))
383 continue;
385 need_iput_tmp = need_iput;
386 need_iput = NULL;
387 /* In case inotify_remove_watch_locked() drops a reference. */
388 if (inode != need_iput_tmp)
389 __iget(inode);
390 else
391 need_iput_tmp = NULL;
392 /* In case the dropping of a reference would nuke next_i. */
393 if ((&next_i->i_sb_list != list) &&
394 atomic_read(&next_i->i_count) &&
395 !(next_i->i_state & (I_CLEAR | I_FREEING |
396 I_WILL_FREE))) {
397 __iget(next_i);
398 need_iput = next_i;
402 * We can safely drop inode_lock here because we hold
403 * references on both inode and next_i. Also no new inodes
404 * will be added since the umount has begun. Finally,
405 * iprune_mutex keeps shrink_icache_memory() away.
407 spin_unlock(&inode_lock);
409 if (need_iput_tmp)
410 iput(need_iput_tmp);
412 /* for each watch, send IN_UNMOUNT and then remove it */
413 mutex_lock(&inode->inotify_mutex);
414 watches = &inode->inotify_watches;
415 list_for_each_entry_safe(watch, next_w, watches, i_list) {
416 struct inotify_handle *ih= watch->ih;
417 mutex_lock(&ih->mutex);
418 ih->in_ops->handle_event(watch, watch->wd, IN_UNMOUNT, 0,
419 NULL, NULL);
420 inotify_remove_watch_locked(ih, watch);
421 mutex_unlock(&ih->mutex);
423 mutex_unlock(&inode->inotify_mutex);
424 iput(inode);
426 spin_lock(&inode_lock);
429 EXPORT_SYMBOL_GPL(inotify_unmount_inodes);
432 * inotify_inode_is_dead - an inode has been deleted, cleanup any watches
433 * @inode: inode that is about to be removed
435 void inotify_inode_is_dead(struct inode *inode)
437 struct inotify_watch *watch, *next;
439 mutex_lock(&inode->inotify_mutex);
440 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
441 struct inotify_handle *ih = watch->ih;
442 mutex_lock(&ih->mutex);
443 inotify_remove_watch_locked(ih, watch);
444 mutex_unlock(&ih->mutex);
446 mutex_unlock(&inode->inotify_mutex);
448 EXPORT_SYMBOL_GPL(inotify_inode_is_dead);
450 /* Kernel Consumer API */
453 * inotify_init - allocate and initialize an inotify instance
454 * @ops: caller's inotify operations
456 struct inotify_handle *inotify_init(const struct inotify_operations *ops)
458 struct inotify_handle *ih;
460 ih = kmalloc(sizeof(struct inotify_handle), GFP_KERNEL);
461 if (unlikely(!ih))
462 return ERR_PTR(-ENOMEM);
464 idr_init(&ih->idr);
465 INIT_LIST_HEAD(&ih->watches);
466 mutex_init(&ih->mutex);
467 ih->last_wd = 0;
468 ih->in_ops = ops;
469 atomic_set(&ih->count, 0);
470 get_inotify_handle(ih);
472 return ih;
474 EXPORT_SYMBOL_GPL(inotify_init);
477 * inotify_init_watch - initialize an inotify watch
478 * @watch: watch to initialize
480 void inotify_init_watch(struct inotify_watch *watch)
482 INIT_LIST_HEAD(&watch->h_list);
483 INIT_LIST_HEAD(&watch->i_list);
484 atomic_set(&watch->count, 0);
485 get_inotify_watch(watch); /* initial get */
487 EXPORT_SYMBOL_GPL(inotify_init_watch);
490 * inotify_destroy - clean up and destroy an inotify instance
491 * @ih: inotify handle
493 void inotify_destroy(struct inotify_handle *ih)
496 * Destroy all of the watches for this handle. Unfortunately, not very
497 * pretty. We cannot do a simple iteration over the list, because we
498 * do not know the inode until we iterate to the watch. But we need to
499 * hold inode->inotify_mutex before ih->mutex. The following works.
501 while (1) {
502 struct inotify_watch *watch;
503 struct list_head *watches;
504 struct inode *inode;
506 mutex_lock(&ih->mutex);
507 watches = &ih->watches;
508 if (list_empty(watches)) {
509 mutex_unlock(&ih->mutex);
510 break;
512 watch = list_first_entry(watches, struct inotify_watch, h_list);
513 get_inotify_watch(watch);
514 mutex_unlock(&ih->mutex);
516 inode = watch->inode;
517 mutex_lock(&inode->inotify_mutex);
518 mutex_lock(&ih->mutex);
520 /* make sure we didn't race with another list removal */
521 if (likely(idr_find(&ih->idr, watch->wd))) {
522 remove_watch_no_event(watch, ih);
523 put_inotify_watch(watch);
526 mutex_unlock(&ih->mutex);
527 mutex_unlock(&inode->inotify_mutex);
528 put_inotify_watch(watch);
531 /* free this handle: the put matching the get in inotify_init() */
532 put_inotify_handle(ih);
534 EXPORT_SYMBOL_GPL(inotify_destroy);
537 * inotify_find_watch - find an existing watch for an (ih,inode) pair
538 * @ih: inotify handle
539 * @inode: inode to watch
540 * @watchp: pointer to existing inotify_watch
542 * Caller must pin given inode (via nameidata).
544 s32 inotify_find_watch(struct inotify_handle *ih, struct inode *inode,
545 struct inotify_watch **watchp)
547 struct inotify_watch *old;
548 int ret = -ENOENT;
550 mutex_lock(&inode->inotify_mutex);
551 mutex_lock(&ih->mutex);
553 old = inode_find_handle(inode, ih);
554 if (unlikely(old)) {
555 get_inotify_watch(old); /* caller must put watch */
556 *watchp = old;
557 ret = old->wd;
560 mutex_unlock(&ih->mutex);
561 mutex_unlock(&inode->inotify_mutex);
563 return ret;
565 EXPORT_SYMBOL_GPL(inotify_find_watch);
568 * inotify_find_update_watch - find and update the mask of an existing watch
569 * @ih: inotify handle
570 * @inode: inode's watch to update
571 * @mask: mask of events to watch
573 * Caller must pin given inode (via nameidata).
575 s32 inotify_find_update_watch(struct inotify_handle *ih, struct inode *inode,
576 u32 mask)
578 struct inotify_watch *old;
579 int mask_add = 0;
580 int ret;
582 if (mask & IN_MASK_ADD)
583 mask_add = 1;
585 /* don't allow invalid bits: we don't want flags set */
586 mask &= IN_ALL_EVENTS | IN_ONESHOT;
587 if (unlikely(!mask))
588 return -EINVAL;
590 mutex_lock(&inode->inotify_mutex);
591 mutex_lock(&ih->mutex);
594 * Handle the case of re-adding a watch on an (inode,ih) pair that we
595 * are already watching. We just update the mask and return its wd.
597 old = inode_find_handle(inode, ih);
598 if (unlikely(!old)) {
599 ret = -ENOENT;
600 goto out;
603 if (mask_add)
604 old->mask |= mask;
605 else
606 old->mask = mask;
607 ret = old->wd;
608 out:
609 mutex_unlock(&ih->mutex);
610 mutex_unlock(&inode->inotify_mutex);
611 return ret;
613 EXPORT_SYMBOL_GPL(inotify_find_update_watch);
616 * inotify_add_watch - add a watch to an inotify instance
617 * @ih: inotify handle
618 * @watch: caller allocated watch structure
619 * @inode: inode to watch
620 * @mask: mask of events to watch
622 * Caller must pin given inode (via nameidata).
623 * Caller must ensure it only calls inotify_add_watch() once per watch.
624 * Calls inotify_handle_get_wd() so may sleep.
626 s32 inotify_add_watch(struct inotify_handle *ih, struct inotify_watch *watch,
627 struct inode *inode, u32 mask)
629 int ret = 0;
631 /* don't allow invalid bits: we don't want flags set */
632 mask &= IN_ALL_EVENTS | IN_ONESHOT;
633 if (unlikely(!mask))
634 return -EINVAL;
635 watch->mask = mask;
637 mutex_lock(&inode->inotify_mutex);
638 mutex_lock(&ih->mutex);
640 /* Initialize a new watch */
641 ret = inotify_handle_get_wd(ih, watch);
642 if (unlikely(ret))
643 goto out;
644 ret = watch->wd;
646 /* save a reference to handle and bump the count to make it official */
647 get_inotify_handle(ih);
648 watch->ih = ih;
651 * Save a reference to the inode and bump the ref count to make it
652 * official. We hold a reference to nameidata, which makes this safe.
654 watch->inode = igrab(inode);
656 if (!inotify_inode_watched(inode))
657 set_dentry_child_flags(inode, 1);
659 /* Add the watch to the handle's and the inode's list */
660 list_add(&watch->h_list, &ih->watches);
661 list_add(&watch->i_list, &inode->inotify_watches);
662 out:
663 mutex_unlock(&ih->mutex);
664 mutex_unlock(&inode->inotify_mutex);
665 return ret;
667 EXPORT_SYMBOL_GPL(inotify_add_watch);
670 * inotify_rm_wd - remove a watch from an inotify instance
671 * @ih: inotify handle
672 * @wd: watch descriptor to remove
674 * Can sleep.
676 int inotify_rm_wd(struct inotify_handle *ih, u32 wd)
678 struct inotify_watch *watch;
679 struct inode *inode;
681 mutex_lock(&ih->mutex);
682 watch = idr_find(&ih->idr, wd);
683 if (unlikely(!watch)) {
684 mutex_unlock(&ih->mutex);
685 return -EINVAL;
687 get_inotify_watch(watch);
688 inode = watch->inode;
689 mutex_unlock(&ih->mutex);
691 mutex_lock(&inode->inotify_mutex);
692 mutex_lock(&ih->mutex);
694 /* make sure that we did not race */
695 if (likely(idr_find(&ih->idr, wd) == watch))
696 inotify_remove_watch_locked(ih, watch);
698 mutex_unlock(&ih->mutex);
699 mutex_unlock(&inode->inotify_mutex);
700 put_inotify_watch(watch);
702 return 0;
704 EXPORT_SYMBOL_GPL(inotify_rm_wd);
707 * inotify_rm_watch - remove a watch from an inotify instance
708 * @ih: inotify handle
709 * @watch: watch to remove
711 * Can sleep.
713 int inotify_rm_watch(struct inotify_handle *ih,
714 struct inotify_watch *watch)
716 return inotify_rm_wd(ih, watch->wd);
718 EXPORT_SYMBOL_GPL(inotify_rm_watch);
721 * inotify_setup - core initialization function
723 static int __init inotify_setup(void)
725 atomic_set(&inotify_cookie, 0);
727 return 0;
730 module_init(inotify_setup);