aboutsummaryrefslogtreecommitdiffstats
path: root/fs/inotify.c
diff options
context:
space:
mode:
authorRobert Love <rml@novell.com>2005-07-12 17:06:03 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2005-07-12 23:38:38 -0400
commit0eeca28300df110bd6ed54b31193c83b87921443 (patch)
tree7db42d8a18d80eca538f5b7d25e0532b8fa38b85 /fs/inotify.c
parentbd4c625c061c2a38568d0add3478f59172455159 (diff)
[PATCH] inotify
inotify is intended to correct the deficiencies of dnotify, particularly its inability to scale and its terrible user interface: * dnotify requires the opening of one fd per each directory that you intend to watch. This quickly results in too many open files and pins removable media, preventing unmount. * dnotify is directory-based. You only learn about changes to directories. Sure, a change to a file in a directory affects the directory, but you are then forced to keep a cache of stat structures. * dnotify's interface to user-space is awful. Signals? inotify provides a more usable, simple, powerful solution to file change notification: * inotify's interface is a system call that returns a fd, not SIGIO. You get a single fd, which is select()-able. * inotify has an event that says "the filesystem that the item you were watching is on was unmounted." * inotify can watch directories or files. Inotify is currently used by Beagle (a desktop search infrastructure), Gamin (a FAM replacement), and other projects. See Documentation/filesystems/inotify.txt. Signed-off-by: Robert Love <rml@novell.com> Cc: John McCutchan <ttb@tentacle.dhs.org> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'fs/inotify.c')
-rw-r--r--fs/inotify.c999
1 files changed, 999 insertions, 0 deletions
diff --git a/fs/inotify.c b/fs/inotify.c
new file mode 100644
index 000000000000..e423bfe0c86f
--- /dev/null
+++ b/fs/inotify.c
@@ -0,0 +1,999 @@
1/*
2 * fs/inotify.c - inode-based file event notifications
3 *
4 * Authors:
5 * John McCutchan <ttb@tentacle.dhs.org>
6 * Robert Love <rml@novell.com>
7 *
8 * Copyright (C) 2005 John McCutchan
9 *
10 * This program is free software; you can redistribute it and/or modify it
11 * under the terms of the GNU General Public License as published by the
12 * Free Software Foundation; either version 2, or (at your option) any
13 * later version.
14 *
15 * This program is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 */
20
21#include <linux/module.h>
22#include <linux/kernel.h>
23#include <linux/sched.h>
24#include <linux/spinlock.h>
25#include <linux/idr.h>
26#include <linux/slab.h>
27#include <linux/fs.h>
28#include <linux/file.h>
29#include <linux/mount.h>
30#include <linux/namei.h>
31#include <linux/poll.h>
32#include <linux/device.h>
33#include <linux/miscdevice.h>
34#include <linux/init.h>
35#include <linux/list.h>
36#include <linux/writeback.h>
37#include <linux/inotify.h>
38
39#include <asm/ioctls.h>
40
41static atomic_t inotify_cookie;
42
43static kmem_cache_t *watch_cachep;
44static kmem_cache_t *event_cachep;
45
46static struct vfsmount *inotify_mnt;
47
48/* These are configurable via /proc/sys/inotify */
49int inotify_max_user_devices;
50int inotify_max_user_watches;
51int inotify_max_queued_events;
52
53/*
54 * Lock ordering:
55 *
56 * dentry->d_lock (used to keep d_move() away from dentry->d_parent)
57 * iprune_sem (synchronize shrink_icache_memory())
58 * inode_lock (protects the super_block->s_inodes list)
59 * inode->inotify_sem (protects inode->inotify_watches and watches->i_list)
60 * inotify_dev->sem (protects inotify_device and watches->d_list)
61 */
62
63/*
64 * Lifetimes of the three main data structures--inotify_device, inode, and
65 * inotify_watch--are managed by reference count.
66 *
67 * inotify_device: Lifetime is from open until release. Additional references
68 * can bump the count via get_inotify_dev() and drop the count via
69 * put_inotify_dev().
70 *
71 * inotify_watch: Lifetime is from create_watch() to destory_watch().
72 * Additional references can bump the count via get_inotify_watch() and drop
73 * the count via put_inotify_watch().
74 *
75 * inode: Pinned so long as the inode is associated with a watch, from
76 * create_watch() to put_inotify_watch().
77 */
78
79/*
80 * struct inotify_device - represents an open instance of an inotify device
81 *
82 * This structure is protected by the semaphore 'sem'.
83 */
84struct inotify_device {
85 wait_queue_head_t wq; /* wait queue for i/o */
86 struct idr idr; /* idr mapping wd -> watch */
87 struct semaphore sem; /* protects this bad boy */
88 struct list_head events; /* list of queued events */
89 struct list_head watches; /* list of watches */
90 atomic_t count; /* reference count */
91 struct user_struct *user; /* user who opened this dev */
92 unsigned int queue_size; /* size of the queue (bytes) */
93 unsigned int event_count; /* number of pending events */
94 unsigned int max_events; /* maximum number of events */
95};
96
97/*
98 * struct inotify_kernel_event - An inotify event, originating from a watch and
99 * queued for user-space. A list of these is attached to each instance of the
100 * device. In read(), this list is walked and all events that can fit in the
101 * buffer are returned.
102 *
103 * Protected by dev->sem of the device in which we are queued.
104 */
105struct inotify_kernel_event {
106 struct inotify_event event; /* the user-space event */
107 struct list_head list; /* entry in inotify_device's list */
108 char *name; /* filename, if any */
109};
110
111/*
112 * struct inotify_watch - represents a watch request on a specific inode
113 *
114 * d_list is protected by dev->sem of the associated watch->dev.
115 * i_list and mask are protected by inode->inotify_sem of the associated inode.
116 * dev, inode, and wd are never written to once the watch is created.
117 */
118struct inotify_watch {
119 struct list_head d_list; /* entry in inotify_device's list */
120 struct list_head i_list; /* entry in inode's list */
121 atomic_t count; /* reference count */
122 struct inotify_device *dev; /* associated device */
123 struct inode *inode; /* associated inode */
124 s32 wd; /* watch descriptor */
125 u32 mask; /* event mask for this watch */
126};
127
128static inline void get_inotify_dev(struct inotify_device *dev)
129{
130 atomic_inc(&dev->count);
131}
132
133static inline void put_inotify_dev(struct inotify_device *dev)
134{
135 if (atomic_dec_and_test(&dev->count)) {
136 atomic_dec(&dev->user->inotify_devs);
137 free_uid(dev->user);
138 kfree(dev);
139 }
140}
141
142static inline void get_inotify_watch(struct inotify_watch *watch)
143{
144 atomic_inc(&watch->count);
145}
146
147/*
148 * put_inotify_watch - decrements the ref count on a given watch. cleans up
149 * the watch and its references if the count reaches zero.
150 */
151static inline void put_inotify_watch(struct inotify_watch *watch)
152{
153 if (atomic_dec_and_test(&watch->count)) {
154 put_inotify_dev(watch->dev);
155 iput(watch->inode);
156 kmem_cache_free(watch_cachep, watch);
157 }
158}
159
160/*
161 * kernel_event - create a new kernel event with the given parameters
162 *
163 * This function can sleep.
164 */
165static struct inotify_kernel_event * kernel_event(s32 wd, u32 mask, u32 cookie,
166 const char *name)
167{
168 struct inotify_kernel_event *kevent;
169
170 kevent = kmem_cache_alloc(event_cachep, GFP_KERNEL);
171 if (unlikely(!kevent))
172 return NULL;
173
174 /* we hand this out to user-space, so zero it just in case */
175 memset(&kevent->event, 0, sizeof(struct inotify_event));
176
177 kevent->event.wd = wd;
178 kevent->event.mask = mask;
179 kevent->event.cookie = cookie;
180
181 INIT_LIST_HEAD(&kevent->list);
182
183 if (name) {
184 size_t len, rem, event_size = sizeof(struct inotify_event);
185
186 /*
187 * We need to pad the filename so as to properly align an
188 * array of inotify_event structures. Because the structure is
189 * small and the common case is a small filename, we just round
190 * up to the next multiple of the structure's sizeof. This is
191 * simple and safe for all architectures.
192 */
193 len = strlen(name) + 1;
194 rem = event_size - len;
195 if (len > event_size) {
196 rem = event_size - (len % event_size);
197 if (len % event_size == 0)
198 rem = 0;
199 }
200
201 kevent->name = kmalloc(len + rem, GFP_KERNEL);
202 if (unlikely(!kevent->name)) {
203 kmem_cache_free(event_cachep, kevent);
204 return NULL;
205 }
206 memcpy(kevent->name, name, len);
207 if (rem)
208 memset(kevent->name + len, 0, rem);
209 kevent->event.len = len + rem;
210 } else {
211 kevent->event.len = 0;
212 kevent->name = NULL;
213 }
214
215 return kevent;
216}
217
218/*
219 * inotify_dev_get_event - return the next event in the given dev's queue
220 *
221 * Caller must hold dev->sem.
222 */
223static inline struct inotify_kernel_event *
224inotify_dev_get_event(struct inotify_device *dev)
225{
226 return list_entry(dev->events.next, struct inotify_kernel_event, list);
227}
228
229/*
230 * inotify_dev_queue_event - add a new event to the given device
231 *
232 * Caller must hold dev->sem. Can sleep (calls kernel_event()).
233 */
234static void inotify_dev_queue_event(struct inotify_device *dev,
235 struct inotify_watch *watch, u32 mask,
236 u32 cookie, const char *name)
237{
238 struct inotify_kernel_event *kevent, *last;
239
240 /* coalescing: drop this event if it is a dupe of the previous */
241 last = inotify_dev_get_event(dev);
242 if (last && last->event.mask == mask && last->event.wd == watch->wd &&
243 last->event.cookie == cookie) {
244 const char *lastname = last->name;
245
246 if (!name && !lastname)
247 return;
248 if (name && lastname && !strcmp(lastname, name))
249 return;
250 }
251
252 /* the queue overflowed and we already sent the Q_OVERFLOW event */
253 if (unlikely(dev->event_count > dev->max_events))
254 return;
255
256 /* if the queue overflows, we need to notify user space */
257 if (unlikely(dev->event_count == dev->max_events))
258 kevent = kernel_event(-1, IN_Q_OVERFLOW, cookie, NULL);
259 else
260 kevent = kernel_event(watch->wd, mask, cookie, name);
261
262 if (unlikely(!kevent))
263 return;
264
265 /* queue the event and wake up anyone waiting */
266 dev->event_count++;
267 dev->queue_size += sizeof(struct inotify_event) + kevent->event.len;
268 list_add_tail(&kevent->list, &dev->events);
269 wake_up_interruptible(&dev->wq);
270}
271
272/*
273 * remove_kevent - cleans up and ultimately frees the given kevent
274 *
275 * Caller must hold dev->sem.
276 */
277static void remove_kevent(struct inotify_device *dev,
278 struct inotify_kernel_event *kevent)
279{
280 list_del(&kevent->list);
281
282 dev->event_count--;
283 dev->queue_size -= sizeof(struct inotify_event) + kevent->event.len;
284
285 kfree(kevent->name);
286 kmem_cache_free(event_cachep, kevent);
287}
288
289/*
290 * inotify_dev_event_dequeue - destroy an event on the given device
291 *
292 * Caller must hold dev->sem.
293 */
294static void inotify_dev_event_dequeue(struct inotify_device *dev)
295{
296 if (!list_empty(&dev->events)) {
297 struct inotify_kernel_event *kevent;
298 kevent = inotify_dev_get_event(dev);
299 remove_kevent(dev, kevent);
300 }
301}
302
303/*
304 * inotify_dev_get_wd - returns the next WD for use by the given dev
305 *
306 * Callers must hold dev->sem. This function can sleep.
307 */
308static int inotify_dev_get_wd(struct inotify_device *dev,
309 struct inotify_watch *watch)
310{
311 int ret;
312
313 do {
314 if (unlikely(!idr_pre_get(&dev->idr, GFP_KERNEL)))
315 return -ENOSPC;
316 ret = idr_get_new(&dev->idr, watch, &watch->wd);
317 } while (ret == -EAGAIN);
318
319 return ret;
320}
321
322/*
323 * find_inode - resolve a user-given path to a specific inode and return a nd
324 */
325static int find_inode(const char __user *dirname, struct nameidata *nd)
326{
327 int error;
328
329 error = __user_walk(dirname, LOOKUP_FOLLOW, nd);
330 if (error)
331 return error;
332 /* you can only watch an inode if you have read permissions on it */
333 error = permission(nd->dentry->d_inode, MAY_READ, NULL);
334 if (error)
335 path_release (nd);
336 return error;
337}
338
339/*
340 * create_watch - creates a watch on the given device.
341 *
342 * Callers must hold dev->sem. Calls inotify_dev_get_wd() so may sleep.
343 * Both 'dev' and 'inode' (by way of nameidata) need to be pinned.
344 */
345static struct inotify_watch *create_watch(struct inotify_device *dev,
346 u32 mask, struct inode *inode)
347{
348 struct inotify_watch *watch;
349 int ret;
350
351 if (atomic_read(&dev->user->inotify_watches) >= inotify_max_user_watches)
352 return ERR_PTR(-ENOSPC);
353
354 watch = kmem_cache_alloc(watch_cachep, GFP_KERNEL);
355 if (unlikely(!watch))
356 return ERR_PTR(-ENOMEM);
357
358 ret = inotify_dev_get_wd(dev, watch);
359 if (unlikely(ret)) {
360 kmem_cache_free(watch_cachep, watch);
361 return ERR_PTR(ret);
362 }
363
364 watch->mask = mask;
365 atomic_set(&watch->count, 0);
366 INIT_LIST_HEAD(&watch->d_list);
367 INIT_LIST_HEAD(&watch->i_list);
368
369 /* save a reference to device and bump the count to make it official */
370 get_inotify_dev(dev);
371 watch->dev = dev;
372
373 /*
374 * Save a reference to the inode and bump the ref count to make it
375 * official. We hold a reference to nameidata, which makes this safe.
376 */
377 watch->inode = igrab(inode);
378
379 /* bump our own count, corresponding to our entry in dev->watches */
380 get_inotify_watch(watch);
381
382 atomic_inc(&dev->user->inotify_watches);
383
384 return watch;
385}
386
387/*
388 * inotify_find_dev - find the watch associated with the given inode and dev
389 *
390 * Callers must hold inode->inotify_sem.
391 */
392static struct inotify_watch *inode_find_dev(struct inode *inode,
393 struct inotify_device *dev)
394{
395 struct inotify_watch *watch;
396
397 list_for_each_entry(watch, &inode->inotify_watches, i_list) {
398 if (watch->dev == dev)
399 return watch;
400 }
401
402 return NULL;
403}
404
405/*
406 * remove_watch_no_event - remove_watch() without the IN_IGNORED event.
407 */
408static void remove_watch_no_event(struct inotify_watch *watch,
409 struct inotify_device *dev)
410{
411 list_del(&watch->i_list);
412 list_del(&watch->d_list);
413
414 atomic_dec(&dev->user->inotify_watches);
415 idr_remove(&dev->idr, watch->wd);
416 put_inotify_watch(watch);
417}
418
419/*
420 * remove_watch - Remove a watch from both the device and the inode. Sends
421 * the IN_IGNORED event to the given device signifying that the inode is no
422 * longer watched.
423 *
424 * Callers must hold both inode->inotify_sem and dev->sem. We drop a
425 * reference to the inode before returning.
426 *
427 * The inode is not iput() so as to remain atomic. If the inode needs to be
428 * iput(), the call returns one. Otherwise, it returns zero.
429 */
430static void remove_watch(struct inotify_watch *watch,struct inotify_device *dev)
431{
432 inotify_dev_queue_event(dev, watch, IN_IGNORED, 0, NULL);
433 remove_watch_no_event(watch, dev);
434}
435
436/*
437 * inotify_inode_watched - returns nonzero if there are watches on this inode
438 * and zero otherwise. We call this lockless, we do not care if we race.
439 */
440static inline int inotify_inode_watched(struct inode *inode)
441{
442 return !list_empty(&inode->inotify_watches);
443}
444
445/* Kernel API */
446
447/**
448 * inotify_inode_queue_event - queue an event to all watches on this inode
449 * @inode: inode event is originating from
450 * @mask: event mask describing this event
451 * @cookie: cookie for synchronization, or zero
452 * @name: filename, if any
453 */
454void inotify_inode_queue_event(struct inode *inode, u32 mask, u32 cookie,
455 const char *name)
456{
457 struct inotify_watch *watch, *next;
458
459 if (!inotify_inode_watched(inode))
460 return;
461
462 down(&inode->inotify_sem);
463 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
464 u32 watch_mask = watch->mask;
465 if (watch_mask & mask) {
466 struct inotify_device *dev = watch->dev;
467 get_inotify_watch(watch);
468 down(&dev->sem);
469 inotify_dev_queue_event(dev, watch, mask, cookie, name);
470 if (watch_mask & IN_ONESHOT)
471 remove_watch_no_event(watch, dev);
472 up(&dev->sem);
473 put_inotify_watch(watch);
474 }
475 }
476 up(&inode->inotify_sem);
477}
478EXPORT_SYMBOL_GPL(inotify_inode_queue_event);
479
480/**
481 * inotify_dentry_parent_queue_event - queue an event to a dentry's parent
482 * @dentry: the dentry in question, we queue against this dentry's parent
483 * @mask: event mask describing this event
484 * @cookie: cookie for synchronization, or zero
485 * @name: filename, if any
486 */
487void inotify_dentry_parent_queue_event(struct dentry *dentry, u32 mask,
488 u32 cookie, const char *name)
489{
490 struct dentry *parent;
491 struct inode *inode;
492
493 spin_lock(&dentry->d_lock);
494 parent = dentry->d_parent;
495 inode = parent->d_inode;
496
497 if (inotify_inode_watched(inode)) {
498 dget(parent);
499 spin_unlock(&dentry->d_lock);
500 inotify_inode_queue_event(inode, mask, cookie, name);
501 dput(parent);
502 } else
503 spin_unlock(&dentry->d_lock);
504}
505EXPORT_SYMBOL_GPL(inotify_dentry_parent_queue_event);
506
507/**
508 * inotify_get_cookie - return a unique cookie for use in synchronizing events.
509 */
510u32 inotify_get_cookie(void)
511{
512 return atomic_inc_return(&inotify_cookie);
513}
514EXPORT_SYMBOL_GPL(inotify_get_cookie);
515
516/**
517 * inotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
518 * @list: list of inodes being unmounted (sb->s_inodes)
519 *
520 * Called with inode_lock held, protecting the unmounting super block's list
521 * of inodes, and with iprune_sem held, keeping shrink_icache_memory() at bay.
522 * We temporarily drop inode_lock, however, and CAN block.
523 */
524void inotify_unmount_inodes(struct list_head *list)
525{
526 struct inode *inode, *next_i, *need_iput = NULL;
527
528 list_for_each_entry_safe(inode, next_i, list, i_sb_list) {
529 struct inotify_watch *watch, *next_w;
530 struct inode *need_iput_tmp;
531 struct list_head *watches;
532
533 /*
534 * If i_count is zero, the inode cannot have any watches and
535 * doing an __iget/iput with MS_ACTIVE clear would actually
536 * evict all inodes with zero i_count from icache which is
537 * unnecessarily violent and may in fact be illegal to do.
538 */
539 if (!atomic_read(&inode->i_count))
540 continue;
541
542 /*
543 * We cannot __iget() an inode in state I_CLEAR, I_FREEING, or
544 * I_WILL_FREE which is fine because by that point the inode
545 * cannot have any associated watches.
546 */
547 if (inode->i_state & (I_CLEAR | I_FREEING | I_WILL_FREE))
548 continue;
549
550 need_iput_tmp = need_iput;
551 need_iput = NULL;
552 /* In case the remove_watch() drops a reference. */
553 if (inode != need_iput_tmp)
554 __iget(inode);
555 else
556 need_iput_tmp = NULL;
557 /* In case the dropping of a reference would nuke next_i. */
558 if ((&next_i->i_sb_list != list) &&
559 atomic_read(&next_i->i_count) &&
560 !(next_i->i_state & (I_CLEAR | I_FREEING |
561 I_WILL_FREE))) {
562 __iget(next_i);
563 need_iput = next_i;
564 }
565
566 /*
567 * We can safely drop inode_lock here because we hold
568 * references on both inode and next_i. Also no new inodes
569 * will be added since the umount has begun. Finally,
570 * iprune_sem keeps shrink_icache_memory() away.
571 */
572 spin_unlock(&inode_lock);
573
574 if (need_iput_tmp)
575 iput(need_iput_tmp);
576
577 /* for each watch, send IN_UNMOUNT and then remove it */
578 down(&inode->inotify_sem);
579 watches = &inode->inotify_watches;
580 list_for_each_entry_safe(watch, next_w, watches, i_list) {
581 struct inotify_device *dev = watch->dev;
582 down(&dev->sem);
583 inotify_dev_queue_event(dev, watch, IN_UNMOUNT,0,NULL);
584 remove_watch(watch, dev);
585 up(&dev->sem);
586 }
587 up(&inode->inotify_sem);
588 iput(inode);
589
590 spin_lock(&inode_lock);
591 }
592}
593EXPORT_SYMBOL_GPL(inotify_unmount_inodes);
594
595/**
596 * inotify_inode_is_dead - an inode has been deleted, cleanup any watches
597 * @inode: inode that is about to be removed
598 */
599void inotify_inode_is_dead(struct inode *inode)
600{
601 struct inotify_watch *watch, *next;
602
603 down(&inode->inotify_sem);
604 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
605 struct inotify_device *dev = watch->dev;
606 down(&dev->sem);
607 remove_watch(watch, dev);
608 up(&dev->sem);
609 }
610 up(&inode->inotify_sem);
611}
612EXPORT_SYMBOL_GPL(inotify_inode_is_dead);
613
614/* Device Interface */
615
616static unsigned int inotify_poll(struct file *file, poll_table *wait)
617{
618 struct inotify_device *dev = file->private_data;
619 int ret = 0;
620
621 poll_wait(file, &dev->wq, wait);
622 down(&dev->sem);
623 if (!list_empty(&dev->events))
624 ret = POLLIN | POLLRDNORM;
625 up(&dev->sem);
626
627 return ret;
628}
629
630static ssize_t inotify_read(struct file *file, char __user *buf,
631 size_t count, loff_t *pos)
632{
633 size_t event_size = sizeof (struct inotify_event);
634 struct inotify_device *dev;
635 char __user *start;
636 int ret;
637 DEFINE_WAIT(wait);
638
639 start = buf;
640 dev = file->private_data;
641
642 while (1) {
643 int events;
644
645 prepare_to_wait(&dev->wq, &wait, TASK_INTERRUPTIBLE);
646
647 down(&dev->sem);
648 events = !list_empty(&dev->events);
649 up(&dev->sem);
650 if (events) {
651 ret = 0;
652 break;
653 }
654
655 if (file->f_flags & O_NONBLOCK) {
656 ret = -EAGAIN;
657 break;
658 }
659
660 if (signal_pending(current)) {
661 ret = -EINTR;
662 break;
663 }
664
665 schedule();
666 }
667
668 finish_wait(&dev->wq, &wait);
669 if (ret)
670 return ret;
671
672 down(&dev->sem);
673 while (1) {
674 struct inotify_kernel_event *kevent;
675
676 ret = buf - start;
677 if (list_empty(&dev->events))
678 break;
679
680 kevent = inotify_dev_get_event(dev);
681 if (event_size + kevent->event.len > count)
682 break;
683
684 if (copy_to_user(buf, &kevent->event, event_size)) {
685 ret = -EFAULT;
686 break;
687 }
688 buf += event_size;
689 count -= event_size;
690
691 if (kevent->name) {
692 if (copy_to_user(buf, kevent->name, kevent->event.len)){
693 ret = -EFAULT;
694 break;
695 }
696 buf += kevent->event.len;
697 count -= kevent->event.len;
698 }
699
700 remove_kevent(dev, kevent);
701 }
702 up(&dev->sem);
703
704 return ret;
705}
706
707static int inotify_release(struct inode *ignored, struct file *file)
708{
709 struct inotify_device *dev = file->private_data;
710
711 /*
712 * Destroy all of the watches on this device. Unfortunately, not very
713 * pretty. We cannot do a simple iteration over the list, because we
714 * do not know the inode until we iterate to the watch. But we need to
715 * hold inode->inotify_sem before dev->sem. The following works.
716 */
717 while (1) {
718 struct inotify_watch *watch;
719 struct list_head *watches;
720 struct inode *inode;
721
722 down(&dev->sem);
723 watches = &dev->watches;
724 if (list_empty(watches)) {
725 up(&dev->sem);
726 break;
727 }
728 watch = list_entry(watches->next, struct inotify_watch, d_list);
729 get_inotify_watch(watch);
730 up(&dev->sem);
731
732 inode = watch->inode;
733 down(&inode->inotify_sem);
734 down(&dev->sem);
735 remove_watch_no_event(watch, dev);
736 up(&dev->sem);
737 up(&inode->inotify_sem);
738 put_inotify_watch(watch);
739 }
740
741 /* destroy all of the events on this device */
742 down(&dev->sem);
743 while (!list_empty(&dev->events))
744 inotify_dev_event_dequeue(dev);
745 up(&dev->sem);
746
747 /* free this device: the put matching the get in inotify_open() */
748 put_inotify_dev(dev);
749
750 return 0;
751}
752
753/*
754 * inotify_ignore - handle the INOTIFY_IGNORE ioctl, asking that a given wd be
755 * removed from the device.
756 *
757 * Can sleep.
758 */
759static int inotify_ignore(struct inotify_device *dev, s32 wd)
760{
761 struct inotify_watch *watch;
762 struct inode *inode;
763
764 down(&dev->sem);
765 watch = idr_find(&dev->idr, wd);
766 if (unlikely(!watch)) {
767 up(&dev->sem);
768 return -EINVAL;
769 }
770 get_inotify_watch(watch);
771 inode = watch->inode;
772 up(&dev->sem);
773
774 down(&inode->inotify_sem);
775 down(&dev->sem);
776
777 /* make sure that we did not race */
778 watch = idr_find(&dev->idr, wd);
779 if (likely(watch))
780 remove_watch(watch, dev);
781
782 up(&dev->sem);
783 up(&inode->inotify_sem);
784 put_inotify_watch(watch);
785
786 return 0;
787}
788
789static long inotify_ioctl(struct file *file, unsigned int cmd,
790 unsigned long arg)
791{
792 struct inotify_device *dev;
793 void __user *p;
794 int ret = -ENOTTY;
795
796 dev = file->private_data;
797 p = (void __user *) arg;
798
799 switch (cmd) {
800 case FIONREAD:
801 ret = put_user(dev->queue_size, (int __user *) p);
802 break;
803 }
804
805 return ret;
806}
807
808static struct file_operations inotify_fops = {
809 .poll = inotify_poll,
810 .read = inotify_read,
811 .release = inotify_release,
812 .unlocked_ioctl = inotify_ioctl,
813 .compat_ioctl = inotify_ioctl,
814};
815
816asmlinkage long sys_inotify_init(void)
817{
818 struct inotify_device *dev;
819 struct user_struct *user;
820 int ret = -ENOTTY;
821 int fd;
822 struct file *filp;
823
824 fd = get_unused_fd();
825 if (fd < 0) {
826 ret = fd;
827 goto out;
828 }
829
830 filp = get_empty_filp();
831 if (!filp) {
832 put_unused_fd(fd);
833 ret = -ENFILE;
834 goto out;
835 }
836 filp->f_op = &inotify_fops;
837 filp->f_vfsmnt = mntget(inotify_mnt);
838 filp->f_dentry = dget(inotify_mnt->mnt_root);
839 filp->f_mapping = filp->f_dentry->d_inode->i_mapping;
840 filp->f_mode = FMODE_READ;
841 filp->f_flags = O_RDONLY;
842
843 user = get_uid(current->user);
844
845 if (unlikely(atomic_read(&user->inotify_devs) >= inotify_max_user_devices)) {
846 ret = -EMFILE;
847 goto out_err;
848 }
849
850 dev = kmalloc(sizeof(struct inotify_device), GFP_KERNEL);
851 if (unlikely(!dev)) {
852 ret = -ENOMEM;
853 goto out_err;
854 }
855
856 idr_init(&dev->idr);
857 INIT_LIST_HEAD(&dev->events);
858 INIT_LIST_HEAD(&dev->watches);
859 init_waitqueue_head(&dev->wq);
860 sema_init(&dev->sem, 1);
861 dev->event_count = 0;
862 dev->queue_size = 0;
863 dev->max_events = inotify_max_queued_events;
864 dev->user = user;
865 atomic_set(&dev->count, 0);
866
867 get_inotify_dev(dev);
868 atomic_inc(&user->inotify_devs);
869
870 filp->private_data = dev;
871 fd_install (fd, filp);
872 return fd;
873out_err:
874 put_unused_fd (fd);
875 put_filp (filp);
876 free_uid(user);
877out:
878 return ret;
879}
880
881asmlinkage long sys_inotify_add_watch(int fd, const char *path, u32 mask)
882{
883 struct inotify_watch *watch, *old;
884 struct inode *inode;
885 struct inotify_device *dev;
886 struct nameidata nd;
887 struct file *filp;
888 int ret;
889
890 filp = fget(fd);
891 if (!filp)
892 return -EBADF;
893
894 dev = filp->private_data;
895
896 ret = find_inode ((const char __user*)path, &nd);
897 if (ret)
898 goto fput_and_out;
899
900 /* Held in place by reference in nd */
901 inode = nd.dentry->d_inode;
902
903 down(&inode->inotify_sem);
904 down(&dev->sem);
905
906 /* don't let user-space set invalid bits: we don't want flags set */
907 mask &= IN_ALL_EVENTS;
908 if (!mask) {
909 ret = -EINVAL;
910 goto out;
911 }
912
913 /*
914 * Handle the case of re-adding a watch on an (inode,dev) pair that we
915 * are already watching. We just update the mask and return its wd.
916 */
917 old = inode_find_dev(inode, dev);
918 if (unlikely(old)) {
919 old->mask = mask;
920 ret = old->wd;
921 goto out;
922 }
923
924 watch = create_watch(dev, mask, inode);
925 if (unlikely(IS_ERR(watch))) {
926 ret = PTR_ERR(watch);
927 goto out;
928 }
929
930 /* Add the watch to the device's and the inode's list */
931 list_add(&watch->d_list, &dev->watches);
932 list_add(&watch->i_list, &inode->inotify_watches);
933 ret = watch->wd;
934out:
935 path_release (&nd);
936 up(&dev->sem);
937 up(&inode->inotify_sem);
938fput_and_out:
939 fput(filp);
940 return ret;
941}
942
943asmlinkage long sys_inotify_rm_watch(int fd, u32 wd)
944{
945 struct file *filp;
946 struct inotify_device *dev;
947 int ret;
948
949 filp = fget(fd);
950 if (!filp)
951 return -EBADF;
952 dev = filp->private_data;
953 ret = inotify_ignore (dev, wd);
954 fput(filp);
955 return ret;
956}
957
958static struct super_block *
959inotify_get_sb(struct file_system_type *fs_type, int flags,
960 const char *dev_name, void *data)
961{
962 return get_sb_pseudo(fs_type, "inotify", NULL, 0xBAD1DEA);
963}
964
965static struct file_system_type inotify_fs_type = {
966 .name = "inotifyfs",
967 .get_sb = inotify_get_sb,
968 .kill_sb = kill_anon_super,
969};
970
971/*
972 * inotify_init - Our initialization function. Note that we cannnot return
973 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
974 * must result in panic().
975 */
976static int __init inotify_init(void)
977{
978 register_filesystem(&inotify_fs_type);
979 inotify_mnt = kern_mount(&inotify_fs_type);
980
981 inotify_max_queued_events = 8192;
982 inotify_max_user_devices = 128;
983 inotify_max_user_watches = 8192;
984
985 atomic_set(&inotify_cookie, 0);
986
987 watch_cachep = kmem_cache_create("inotify_watch_cache",
988 sizeof(struct inotify_watch),
989 0, SLAB_PANIC, NULL, NULL);
990 event_cachep = kmem_cache_create("inotify_event_cache",
991 sizeof(struct inotify_kernel_event),
992 0, SLAB_PANIC, NULL, NULL);
993
994 printk(KERN_INFO "inotify syscall\n");
995
996 return 0;
997}
998
999module_init(inotify_init);