blob: 2e4e2a57708cf95dcdf348235376ef901f4dc295 [file] [log] [blame]
Robert Love0eeca282005-07-12 17:06:03 -04001/*
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>
Robert Love0eeca282005-07-12 17:06:03 -040032#include <linux/init.h>
33#include <linux/list.h>
34#include <linux/writeback.h>
35#include <linux/inotify.h>
36
37#include <asm/ioctls.h>
38
39static atomic_t inotify_cookie;
40
41static kmem_cache_t *watch_cachep;
42static kmem_cache_t *event_cachep;
43
44static struct vfsmount *inotify_mnt;
45
Robert Love0399cb02005-07-13 12:38:18 -040046/* these are configurable via /proc/sys/fs/inotify/ */
47int inotify_max_user_instances;
Robert Love0eeca282005-07-12 17:06:03 -040048int inotify_max_user_watches;
49int inotify_max_queued_events;
50
51/*
52 * Lock ordering:
53 *
54 * dentry->d_lock (used to keep d_move() away from dentry->d_parent)
55 * iprune_sem (synchronize shrink_icache_memory())
56 * inode_lock (protects the super_block->s_inodes list)
57 * inode->inotify_sem (protects inode->inotify_watches and watches->i_list)
58 * inotify_dev->sem (protects inotify_device and watches->d_list)
59 */
60
61/*
62 * Lifetimes of the three main data structures--inotify_device, inode, and
63 * inotify_watch--are managed by reference count.
64 *
Robert Loveb6807162005-07-25 15:07:13 -040065 * inotify_device: Lifetime is from inotify_init() until release. Additional
66 * references can bump the count via get_inotify_dev() and drop the count via
Robert Love0eeca282005-07-12 17:06:03 -040067 * put_inotify_dev().
68 *
69 * inotify_watch: Lifetime is from create_watch() to destory_watch().
70 * Additional references can bump the count via get_inotify_watch() and drop
71 * the count via put_inotify_watch().
72 *
73 * inode: Pinned so long as the inode is associated with a watch, from
74 * create_watch() to put_inotify_watch().
75 */
76
77/*
Robert Loveb6807162005-07-25 15:07:13 -040078 * struct inotify_device - represents an inotify instance
Robert Love0eeca282005-07-12 17:06:03 -040079 *
80 * This structure is protected by the semaphore 'sem'.
81 */
82struct inotify_device {
83 wait_queue_head_t wq; /* wait queue for i/o */
84 struct idr idr; /* idr mapping wd -> watch */
85 struct semaphore sem; /* protects this bad boy */
86 struct list_head events; /* list of queued events */
87 struct list_head watches; /* list of watches */
88 atomic_t count; /* reference count */
89 struct user_struct *user; /* user who opened this dev */
90 unsigned int queue_size; /* size of the queue (bytes) */
91 unsigned int event_count; /* number of pending events */
92 unsigned int max_events; /* maximum number of events */
John McCutchanb9c55d22005-08-01 11:00:45 -040093 u32 last_wd; /* the last wd allocated */
Robert Love0eeca282005-07-12 17:06:03 -040094};
95
96/*
97 * struct inotify_kernel_event - An inotify event, originating from a watch and
98 * queued for user-space. A list of these is attached to each instance of the
99 * device. In read(), this list is walked and all events that can fit in the
100 * buffer are returned.
101 *
102 * Protected by dev->sem of the device in which we are queued.
103 */
104struct inotify_kernel_event {
105 struct inotify_event event; /* the user-space event */
106 struct list_head list; /* entry in inotify_device's list */
107 char *name; /* filename, if any */
108};
109
110/*
111 * struct inotify_watch - represents a watch request on a specific inode
112 *
113 * d_list is protected by dev->sem of the associated watch->dev.
114 * i_list and mask are protected by inode->inotify_sem of the associated inode.
115 * dev, inode, and wd are never written to once the watch is created.
116 */
117struct inotify_watch {
118 struct list_head d_list; /* entry in inotify_device's list */
119 struct list_head i_list; /* entry in inode's list */
120 atomic_t count; /* reference count */
121 struct inotify_device *dev; /* associated device */
122 struct inode *inode; /* associated inode */
123 s32 wd; /* watch descriptor */
124 u32 mask; /* event mask for this watch */
125};
126
Robert Love0399cb02005-07-13 12:38:18 -0400127#ifdef CONFIG_SYSCTL
128
129#include <linux/sysctl.h>
130
131static int zero;
132
133ctl_table inotify_table[] = {
134 {
135 .ctl_name = INOTIFY_MAX_USER_INSTANCES,
136 .procname = "max_user_instances",
137 .data = &inotify_max_user_instances,
138 .maxlen = sizeof(int),
139 .mode = 0644,
140 .proc_handler = &proc_dointvec_minmax,
141 .strategy = &sysctl_intvec,
142 .extra1 = &zero,
143 },
144 {
145 .ctl_name = INOTIFY_MAX_USER_WATCHES,
146 .procname = "max_user_watches",
147 .data = &inotify_max_user_watches,
148 .maxlen = sizeof(int),
149 .mode = 0644,
150 .proc_handler = &proc_dointvec_minmax,
151 .strategy = &sysctl_intvec,
152 .extra1 = &zero,
153 },
154 {
155 .ctl_name = INOTIFY_MAX_QUEUED_EVENTS,
156 .procname = "max_queued_events",
157 .data = &inotify_max_queued_events,
158 .maxlen = sizeof(int),
159 .mode = 0644,
160 .proc_handler = &proc_dointvec_minmax,
161 .strategy = &sysctl_intvec,
162 .extra1 = &zero
163 },
164 { .ctl_name = 0 }
165};
166#endif /* CONFIG_SYSCTL */
167
Robert Love0eeca282005-07-12 17:06:03 -0400168static inline void get_inotify_dev(struct inotify_device *dev)
169{
170 atomic_inc(&dev->count);
171}
172
173static inline void put_inotify_dev(struct inotify_device *dev)
174{
175 if (atomic_dec_and_test(&dev->count)) {
176 atomic_dec(&dev->user->inotify_devs);
177 free_uid(dev->user);
178 kfree(dev);
179 }
180}
181
182static inline void get_inotify_watch(struct inotify_watch *watch)
183{
184 atomic_inc(&watch->count);
185}
186
187/*
188 * put_inotify_watch - decrements the ref count on a given watch. cleans up
189 * the watch and its references if the count reaches zero.
190 */
191static inline void put_inotify_watch(struct inotify_watch *watch)
192{
193 if (atomic_dec_and_test(&watch->count)) {
194 put_inotify_dev(watch->dev);
195 iput(watch->inode);
196 kmem_cache_free(watch_cachep, watch);
197 }
198}
199
200/*
201 * kernel_event - create a new kernel event with the given parameters
202 *
203 * This function can sleep.
204 */
205static struct inotify_kernel_event * kernel_event(s32 wd, u32 mask, u32 cookie,
206 const char *name)
207{
208 struct inotify_kernel_event *kevent;
209
210 kevent = kmem_cache_alloc(event_cachep, GFP_KERNEL);
211 if (unlikely(!kevent))
212 return NULL;
213
214 /* we hand this out to user-space, so zero it just in case */
215 memset(&kevent->event, 0, sizeof(struct inotify_event));
216
217 kevent->event.wd = wd;
218 kevent->event.mask = mask;
219 kevent->event.cookie = cookie;
220
221 INIT_LIST_HEAD(&kevent->list);
222
223 if (name) {
224 size_t len, rem, event_size = sizeof(struct inotify_event);
225
226 /*
227 * We need to pad the filename so as to properly align an
228 * array of inotify_event structures. Because the structure is
229 * small and the common case is a small filename, we just round
230 * up to the next multiple of the structure's sizeof. This is
231 * simple and safe for all architectures.
232 */
233 len = strlen(name) + 1;
234 rem = event_size - len;
235 if (len > event_size) {
236 rem = event_size - (len % event_size);
237 if (len % event_size == 0)
238 rem = 0;
239 }
240
241 kevent->name = kmalloc(len + rem, GFP_KERNEL);
242 if (unlikely(!kevent->name)) {
243 kmem_cache_free(event_cachep, kevent);
244 return NULL;
245 }
246 memcpy(kevent->name, name, len);
247 if (rem)
248 memset(kevent->name + len, 0, rem);
249 kevent->event.len = len + rem;
250 } else {
251 kevent->event.len = 0;
252 kevent->name = NULL;
253 }
254
255 return kevent;
256}
257
258/*
259 * inotify_dev_get_event - return the next event in the given dev's queue
260 *
261 * Caller must hold dev->sem.
262 */
263static inline struct inotify_kernel_event *
264inotify_dev_get_event(struct inotify_device *dev)
265{
266 return list_entry(dev->events.next, struct inotify_kernel_event, list);
267}
268
269/*
270 * inotify_dev_queue_event - add a new event to the given device
271 *
272 * Caller must hold dev->sem. Can sleep (calls kernel_event()).
273 */
274static void inotify_dev_queue_event(struct inotify_device *dev,
275 struct inotify_watch *watch, u32 mask,
276 u32 cookie, const char *name)
277{
278 struct inotify_kernel_event *kevent, *last;
279
280 /* coalescing: drop this event if it is a dupe of the previous */
281 last = inotify_dev_get_event(dev);
282 if (last && last->event.mask == mask && last->event.wd == watch->wd &&
283 last->event.cookie == cookie) {
284 const char *lastname = last->name;
285
286 if (!name && !lastname)
287 return;
288 if (name && lastname && !strcmp(lastname, name))
289 return;
290 }
291
292 /* the queue overflowed and we already sent the Q_OVERFLOW event */
293 if (unlikely(dev->event_count > dev->max_events))
294 return;
295
296 /* if the queue overflows, we need to notify user space */
297 if (unlikely(dev->event_count == dev->max_events))
298 kevent = kernel_event(-1, IN_Q_OVERFLOW, cookie, NULL);
299 else
300 kevent = kernel_event(watch->wd, mask, cookie, name);
301
302 if (unlikely(!kevent))
303 return;
304
305 /* queue the event and wake up anyone waiting */
306 dev->event_count++;
307 dev->queue_size += sizeof(struct inotify_event) + kevent->event.len;
308 list_add_tail(&kevent->list, &dev->events);
309 wake_up_interruptible(&dev->wq);
310}
311
312/*
313 * remove_kevent - cleans up and ultimately frees the given kevent
314 *
315 * Caller must hold dev->sem.
316 */
317static void remove_kevent(struct inotify_device *dev,
318 struct inotify_kernel_event *kevent)
319{
320 list_del(&kevent->list);
321
322 dev->event_count--;
323 dev->queue_size -= sizeof(struct inotify_event) + kevent->event.len;
324
325 kfree(kevent->name);
326 kmem_cache_free(event_cachep, kevent);
327}
328
329/*
330 * inotify_dev_event_dequeue - destroy an event on the given device
331 *
332 * Caller must hold dev->sem.
333 */
334static void inotify_dev_event_dequeue(struct inotify_device *dev)
335{
336 if (!list_empty(&dev->events)) {
337 struct inotify_kernel_event *kevent;
338 kevent = inotify_dev_get_event(dev);
339 remove_kevent(dev, kevent);
340 }
341}
342
343/*
344 * inotify_dev_get_wd - returns the next WD for use by the given dev
345 *
346 * Callers must hold dev->sem. This function can sleep.
347 */
348static int inotify_dev_get_wd(struct inotify_device *dev,
349 struct inotify_watch *watch)
350{
351 int ret;
352
353 do {
354 if (unlikely(!idr_pre_get(&dev->idr, GFP_KERNEL)))
355 return -ENOSPC;
John McCutchan7c657f22005-08-26 14:02:04 -0400356 ret = idr_get_new_above(&dev->idr, watch, dev->last_wd+1, &watch->wd);
Robert Love0eeca282005-07-12 17:06:03 -0400357 } while (ret == -EAGAIN);
358
359 return ret;
360}
361
362/*
363 * find_inode - resolve a user-given path to a specific inode and return a nd
364 */
365static int find_inode(const char __user *dirname, struct nameidata *nd)
366{
367 int error;
368
369 error = __user_walk(dirname, LOOKUP_FOLLOW, nd);
370 if (error)
371 return error;
372 /* you can only watch an inode if you have read permissions on it */
373 error = permission(nd->dentry->d_inode, MAY_READ, NULL);
374 if (error)
Robert Loveb6807162005-07-25 15:07:13 -0400375 path_release(nd);
Robert Love0eeca282005-07-12 17:06:03 -0400376 return error;
377}
378
379/*
380 * create_watch - creates a watch on the given device.
381 *
382 * Callers must hold dev->sem. Calls inotify_dev_get_wd() so may sleep.
383 * Both 'dev' and 'inode' (by way of nameidata) need to be pinned.
384 */
385static struct inotify_watch *create_watch(struct inotify_device *dev,
386 u32 mask, struct inode *inode)
387{
388 struct inotify_watch *watch;
389 int ret;
390
Robert Loveb6807162005-07-25 15:07:13 -0400391 if (atomic_read(&dev->user->inotify_watches) >=
392 inotify_max_user_watches)
Robert Love0eeca282005-07-12 17:06:03 -0400393 return ERR_PTR(-ENOSPC);
394
395 watch = kmem_cache_alloc(watch_cachep, GFP_KERNEL);
396 if (unlikely(!watch))
397 return ERR_PTR(-ENOMEM);
398
399 ret = inotify_dev_get_wd(dev, watch);
400 if (unlikely(ret)) {
401 kmem_cache_free(watch_cachep, watch);
402 return ERR_PTR(ret);
403 }
404
Robert Love0bf955c2005-08-15 12:27:54 -0400405 dev->last_wd = watch->wd;
Robert Love0eeca282005-07-12 17:06:03 -0400406 watch->mask = mask;
407 atomic_set(&watch->count, 0);
408 INIT_LIST_HEAD(&watch->d_list);
409 INIT_LIST_HEAD(&watch->i_list);
410
411 /* save a reference to device and bump the count to make it official */
412 get_inotify_dev(dev);
413 watch->dev = dev;
414
415 /*
416 * Save a reference to the inode and bump the ref count to make it
417 * official. We hold a reference to nameidata, which makes this safe.
418 */
419 watch->inode = igrab(inode);
420
421 /* bump our own count, corresponding to our entry in dev->watches */
422 get_inotify_watch(watch);
423
424 atomic_inc(&dev->user->inotify_watches);
425
426 return watch;
427}
428
429/*
430 * inotify_find_dev - find the watch associated with the given inode and dev
431 *
432 * Callers must hold inode->inotify_sem.
433 */
434static struct inotify_watch *inode_find_dev(struct inode *inode,
435 struct inotify_device *dev)
436{
437 struct inotify_watch *watch;
438
439 list_for_each_entry(watch, &inode->inotify_watches, i_list) {
440 if (watch->dev == dev)
441 return watch;
442 }
443
444 return NULL;
445}
446
447/*
448 * remove_watch_no_event - remove_watch() without the IN_IGNORED event.
449 */
450static void remove_watch_no_event(struct inotify_watch *watch,
451 struct inotify_device *dev)
452{
453 list_del(&watch->i_list);
454 list_del(&watch->d_list);
455
456 atomic_dec(&dev->user->inotify_watches);
457 idr_remove(&dev->idr, watch->wd);
458 put_inotify_watch(watch);
459}
460
461/*
462 * remove_watch - Remove a watch from both the device and the inode. Sends
463 * the IN_IGNORED event to the given device signifying that the inode is no
464 * longer watched.
465 *
466 * Callers must hold both inode->inotify_sem and dev->sem. We drop a
467 * reference to the inode before returning.
468 *
469 * The inode is not iput() so as to remain atomic. If the inode needs to be
470 * iput(), the call returns one. Otherwise, it returns zero.
471 */
472static void remove_watch(struct inotify_watch *watch,struct inotify_device *dev)
473{
474 inotify_dev_queue_event(dev, watch, IN_IGNORED, 0, NULL);
475 remove_watch_no_event(watch, dev);
476}
477
478/*
479 * inotify_inode_watched - returns nonzero if there are watches on this inode
480 * and zero otherwise. We call this lockless, we do not care if we race.
481 */
482static inline int inotify_inode_watched(struct inode *inode)
483{
484 return !list_empty(&inode->inotify_watches);
485}
486
487/* Kernel API */
488
489/**
490 * inotify_inode_queue_event - queue an event to all watches on this inode
491 * @inode: inode event is originating from
492 * @mask: event mask describing this event
493 * @cookie: cookie for synchronization, or zero
494 * @name: filename, if any
495 */
496void inotify_inode_queue_event(struct inode *inode, u32 mask, u32 cookie,
497 const char *name)
498{
499 struct inotify_watch *watch, *next;
500
501 if (!inotify_inode_watched(inode))
502 return;
503
504 down(&inode->inotify_sem);
505 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
506 u32 watch_mask = watch->mask;
507 if (watch_mask & mask) {
508 struct inotify_device *dev = watch->dev;
509 get_inotify_watch(watch);
510 down(&dev->sem);
511 inotify_dev_queue_event(dev, watch, mask, cookie, name);
512 if (watch_mask & IN_ONESHOT)
513 remove_watch_no_event(watch, dev);
514 up(&dev->sem);
515 put_inotify_watch(watch);
516 }
517 }
518 up(&inode->inotify_sem);
519}
520EXPORT_SYMBOL_GPL(inotify_inode_queue_event);
521
522/**
523 * inotify_dentry_parent_queue_event - queue an event to a dentry's parent
524 * @dentry: the dentry in question, we queue against this dentry's parent
525 * @mask: event mask describing this event
526 * @cookie: cookie for synchronization, or zero
527 * @name: filename, if any
528 */
529void inotify_dentry_parent_queue_event(struct dentry *dentry, u32 mask,
530 u32 cookie, const char *name)
531{
532 struct dentry *parent;
533 struct inode *inode;
534
535 spin_lock(&dentry->d_lock);
536 parent = dentry->d_parent;
537 inode = parent->d_inode;
538
539 if (inotify_inode_watched(inode)) {
540 dget(parent);
541 spin_unlock(&dentry->d_lock);
542 inotify_inode_queue_event(inode, mask, cookie, name);
543 dput(parent);
544 } else
545 spin_unlock(&dentry->d_lock);
546}
547EXPORT_SYMBOL_GPL(inotify_dentry_parent_queue_event);
548
549/**
550 * inotify_get_cookie - return a unique cookie for use in synchronizing events.
551 */
552u32 inotify_get_cookie(void)
553{
554 return atomic_inc_return(&inotify_cookie);
555}
556EXPORT_SYMBOL_GPL(inotify_get_cookie);
557
558/**
559 * inotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
560 * @list: list of inodes being unmounted (sb->s_inodes)
561 *
562 * Called with inode_lock held, protecting the unmounting super block's list
563 * of inodes, and with iprune_sem held, keeping shrink_icache_memory() at bay.
564 * We temporarily drop inode_lock, however, and CAN block.
565 */
566void inotify_unmount_inodes(struct list_head *list)
567{
568 struct inode *inode, *next_i, *need_iput = NULL;
569
570 list_for_each_entry_safe(inode, next_i, list, i_sb_list) {
571 struct inotify_watch *watch, *next_w;
572 struct inode *need_iput_tmp;
573 struct list_head *watches;
574
575 /*
576 * If i_count is zero, the inode cannot have any watches and
577 * doing an __iget/iput with MS_ACTIVE clear would actually
578 * evict all inodes with zero i_count from icache which is
579 * unnecessarily violent and may in fact be illegal to do.
580 */
581 if (!atomic_read(&inode->i_count))
582 continue;
583
584 /*
585 * We cannot __iget() an inode in state I_CLEAR, I_FREEING, or
586 * I_WILL_FREE which is fine because by that point the inode
587 * cannot have any associated watches.
588 */
589 if (inode->i_state & (I_CLEAR | I_FREEING | I_WILL_FREE))
590 continue;
591
592 need_iput_tmp = need_iput;
593 need_iput = NULL;
594 /* In case the remove_watch() drops a reference. */
595 if (inode != need_iput_tmp)
596 __iget(inode);
597 else
598 need_iput_tmp = NULL;
599 /* In case the dropping of a reference would nuke next_i. */
600 if ((&next_i->i_sb_list != list) &&
601 atomic_read(&next_i->i_count) &&
602 !(next_i->i_state & (I_CLEAR | I_FREEING |
603 I_WILL_FREE))) {
604 __iget(next_i);
605 need_iput = next_i;
606 }
607
608 /*
609 * We can safely drop inode_lock here because we hold
610 * references on both inode and next_i. Also no new inodes
611 * will be added since the umount has begun. Finally,
612 * iprune_sem keeps shrink_icache_memory() away.
613 */
614 spin_unlock(&inode_lock);
615
616 if (need_iput_tmp)
617 iput(need_iput_tmp);
618
619 /* for each watch, send IN_UNMOUNT and then remove it */
620 down(&inode->inotify_sem);
621 watches = &inode->inotify_watches;
622 list_for_each_entry_safe(watch, next_w, watches, i_list) {
623 struct inotify_device *dev = watch->dev;
624 down(&dev->sem);
625 inotify_dev_queue_event(dev, watch, IN_UNMOUNT,0,NULL);
626 remove_watch(watch, dev);
627 up(&dev->sem);
628 }
629 up(&inode->inotify_sem);
630 iput(inode);
631
632 spin_lock(&inode_lock);
633 }
634}
635EXPORT_SYMBOL_GPL(inotify_unmount_inodes);
636
637/**
638 * inotify_inode_is_dead - an inode has been deleted, cleanup any watches
639 * @inode: inode that is about to be removed
640 */
641void inotify_inode_is_dead(struct inode *inode)
642{
643 struct inotify_watch *watch, *next;
644
645 down(&inode->inotify_sem);
646 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
647 struct inotify_device *dev = watch->dev;
648 down(&dev->sem);
649 remove_watch(watch, dev);
650 up(&dev->sem);
651 }
652 up(&inode->inotify_sem);
653}
654EXPORT_SYMBOL_GPL(inotify_inode_is_dead);
655
656/* Device Interface */
657
658static unsigned int inotify_poll(struct file *file, poll_table *wait)
659{
660 struct inotify_device *dev = file->private_data;
661 int ret = 0;
662
663 poll_wait(file, &dev->wq, wait);
664 down(&dev->sem);
665 if (!list_empty(&dev->events))
666 ret = POLLIN | POLLRDNORM;
667 up(&dev->sem);
668
669 return ret;
670}
671
672static ssize_t inotify_read(struct file *file, char __user *buf,
673 size_t count, loff_t *pos)
674{
675 size_t event_size = sizeof (struct inotify_event);
676 struct inotify_device *dev;
677 char __user *start;
678 int ret;
679 DEFINE_WAIT(wait);
680
681 start = buf;
682 dev = file->private_data;
683
684 while (1) {
685 int events;
686
687 prepare_to_wait(&dev->wq, &wait, TASK_INTERRUPTIBLE);
688
689 down(&dev->sem);
690 events = !list_empty(&dev->events);
691 up(&dev->sem);
692 if (events) {
693 ret = 0;
694 break;
695 }
696
697 if (file->f_flags & O_NONBLOCK) {
698 ret = -EAGAIN;
699 break;
700 }
701
702 if (signal_pending(current)) {
703 ret = -EINTR;
704 break;
705 }
706
707 schedule();
708 }
709
710 finish_wait(&dev->wq, &wait);
711 if (ret)
712 return ret;
713
714 down(&dev->sem);
715 while (1) {
716 struct inotify_kernel_event *kevent;
717
718 ret = buf - start;
719 if (list_empty(&dev->events))
720 break;
721
722 kevent = inotify_dev_get_event(dev);
723 if (event_size + kevent->event.len > count)
724 break;
725
726 if (copy_to_user(buf, &kevent->event, event_size)) {
727 ret = -EFAULT;
728 break;
729 }
730 buf += event_size;
731 count -= event_size;
732
733 if (kevent->name) {
734 if (copy_to_user(buf, kevent->name, kevent->event.len)){
735 ret = -EFAULT;
736 break;
737 }
738 buf += kevent->event.len;
739 count -= kevent->event.len;
740 }
741
742 remove_kevent(dev, kevent);
743 }
744 up(&dev->sem);
745
746 return ret;
747}
748
749static int inotify_release(struct inode *ignored, struct file *file)
750{
751 struct inotify_device *dev = file->private_data;
752
753 /*
754 * Destroy all of the watches on this device. Unfortunately, not very
755 * pretty. We cannot do a simple iteration over the list, because we
756 * do not know the inode until we iterate to the watch. But we need to
757 * hold inode->inotify_sem before dev->sem. The following works.
758 */
759 while (1) {
760 struct inotify_watch *watch;
761 struct list_head *watches;
762 struct inode *inode;
763
764 down(&dev->sem);
765 watches = &dev->watches;
766 if (list_empty(watches)) {
767 up(&dev->sem);
768 break;
769 }
770 watch = list_entry(watches->next, struct inotify_watch, d_list);
771 get_inotify_watch(watch);
772 up(&dev->sem);
773
774 inode = watch->inode;
775 down(&inode->inotify_sem);
776 down(&dev->sem);
777 remove_watch_no_event(watch, dev);
778 up(&dev->sem);
779 up(&inode->inotify_sem);
780 put_inotify_watch(watch);
781 }
782
783 /* destroy all of the events on this device */
784 down(&dev->sem);
785 while (!list_empty(&dev->events))
786 inotify_dev_event_dequeue(dev);
787 up(&dev->sem);
788
Robert Loveb6807162005-07-25 15:07:13 -0400789 /* free this device: the put matching the get in inotify_init() */
Robert Love0eeca282005-07-12 17:06:03 -0400790 put_inotify_dev(dev);
791
792 return 0;
793}
794
795/*
Robert Loveb6807162005-07-25 15:07:13 -0400796 * inotify_ignore - remove a given wd from this inotify instance.
Robert Love0eeca282005-07-12 17:06:03 -0400797 *
798 * Can sleep.
799 */
800static int inotify_ignore(struct inotify_device *dev, s32 wd)
801{
802 struct inotify_watch *watch;
803 struct inode *inode;
804
805 down(&dev->sem);
806 watch = idr_find(&dev->idr, wd);
807 if (unlikely(!watch)) {
808 up(&dev->sem);
809 return -EINVAL;
810 }
811 get_inotify_watch(watch);
812 inode = watch->inode;
813 up(&dev->sem);
814
815 down(&inode->inotify_sem);
816 down(&dev->sem);
817
818 /* make sure that we did not race */
819 watch = idr_find(&dev->idr, wd);
820 if (likely(watch))
821 remove_watch(watch, dev);
822
823 up(&dev->sem);
824 up(&inode->inotify_sem);
825 put_inotify_watch(watch);
826
827 return 0;
828}
829
830static long inotify_ioctl(struct file *file, unsigned int cmd,
831 unsigned long arg)
832{
833 struct inotify_device *dev;
834 void __user *p;
835 int ret = -ENOTTY;
836
837 dev = file->private_data;
838 p = (void __user *) arg;
839
840 switch (cmd) {
841 case FIONREAD:
842 ret = put_user(dev->queue_size, (int __user *) p);
843 break;
844 }
845
846 return ret;
847}
848
849static struct file_operations inotify_fops = {
850 .poll = inotify_poll,
851 .read = inotify_read,
852 .release = inotify_release,
853 .unlocked_ioctl = inotify_ioctl,
854 .compat_ioctl = inotify_ioctl,
855};
856
857asmlinkage long sys_inotify_init(void)
858{
859 struct inotify_device *dev;
860 struct user_struct *user;
Robert Loveb6807162005-07-25 15:07:13 -0400861 struct file *filp;
862 int fd, ret;
Robert Love0eeca282005-07-12 17:06:03 -0400863
864 fd = get_unused_fd();
Robert Loveb6807162005-07-25 15:07:13 -0400865 if (fd < 0)
866 return fd;
Robert Love0eeca282005-07-12 17:06:03 -0400867
868 filp = get_empty_filp();
869 if (!filp) {
Robert Love0eeca282005-07-12 17:06:03 -0400870 ret = -ENFILE;
Robert Love5eb22cb2005-07-25 15:12:19 -0400871 goto out_put_fd;
Robert Love0eeca282005-07-12 17:06:03 -0400872 }
Robert Love0eeca282005-07-12 17:06:03 -0400873
874 user = get_uid(current->user);
Robert Loveb6807162005-07-25 15:07:13 -0400875 if (unlikely(atomic_read(&user->inotify_devs) >=
876 inotify_max_user_instances)) {
Robert Love0eeca282005-07-12 17:06:03 -0400877 ret = -EMFILE;
Robert Love5eb22cb2005-07-25 15:12:19 -0400878 goto out_free_uid;
Robert Love0eeca282005-07-12 17:06:03 -0400879 }
880
881 dev = kmalloc(sizeof(struct inotify_device), GFP_KERNEL);
882 if (unlikely(!dev)) {
883 ret = -ENOMEM;
Robert Love5eb22cb2005-07-25 15:12:19 -0400884 goto out_free_uid;
Robert Love0eeca282005-07-12 17:06:03 -0400885 }
886
Robert Loveb6807162005-07-25 15:07:13 -0400887 filp->f_op = &inotify_fops;
888 filp->f_vfsmnt = mntget(inotify_mnt);
889 filp->f_dentry = dget(inotify_mnt->mnt_root);
890 filp->f_mapping = filp->f_dentry->d_inode->i_mapping;
891 filp->f_mode = FMODE_READ;
892 filp->f_flags = O_RDONLY;
893 filp->private_data = dev;
894
Robert Love0eeca282005-07-12 17:06:03 -0400895 idr_init(&dev->idr);
896 INIT_LIST_HEAD(&dev->events);
897 INIT_LIST_HEAD(&dev->watches);
898 init_waitqueue_head(&dev->wq);
899 sema_init(&dev->sem, 1);
900 dev->event_count = 0;
901 dev->queue_size = 0;
902 dev->max_events = inotify_max_queued_events;
903 dev->user = user;
John McCutchanb9c55d22005-08-01 11:00:45 -0400904 dev->last_wd = 0;
Robert Love0eeca282005-07-12 17:06:03 -0400905 atomic_set(&dev->count, 0);
906
907 get_inotify_dev(dev);
908 atomic_inc(&user->inotify_devs);
Robert Loveb6807162005-07-25 15:07:13 -0400909 fd_install(fd, filp);
Robert Love0eeca282005-07-12 17:06:03 -0400910
Robert Love0eeca282005-07-12 17:06:03 -0400911 return fd;
Robert Love5eb22cb2005-07-25 15:12:19 -0400912out_free_uid:
Robert Love0eeca282005-07-12 17:06:03 -0400913 free_uid(user);
Robert Love5eb22cb2005-07-25 15:12:19 -0400914 put_filp(filp);
915out_put_fd:
916 put_unused_fd(fd);
Robert Love0eeca282005-07-12 17:06:03 -0400917 return ret;
918}
919
Robert Loveb6807162005-07-25 15:07:13 -0400920asmlinkage long sys_inotify_add_watch(int fd, const char __user *path, u32 mask)
Robert Love0eeca282005-07-12 17:06:03 -0400921{
922 struct inotify_watch *watch, *old;
923 struct inode *inode;
924 struct inotify_device *dev;
925 struct nameidata nd;
926 struct file *filp;
Robert Love33ea2f52005-07-25 15:08:37 -0400927 int ret, fput_needed;
Robert Love0eeca282005-07-12 17:06:03 -0400928
Robert Love33ea2f52005-07-25 15:08:37 -0400929 filp = fget_light(fd, &fput_needed);
930 if (unlikely(!filp))
Robert Love0eeca282005-07-12 17:06:03 -0400931 return -EBADF;
932
Robert Love783bc292005-07-25 15:10:08 -0400933 /* verify that this is indeed an inotify instance */
934 if (unlikely(filp->f_op != &inotify_fops)) {
935 ret = -EINVAL;
936 goto fput_and_out;
937 }
938
Robert Loveb6807162005-07-25 15:07:13 -0400939 ret = find_inode(path, &nd);
940 if (unlikely(ret))
Robert Love0eeca282005-07-12 17:06:03 -0400941 goto fput_and_out;
942
Robert Loveb6807162005-07-25 15:07:13 -0400943 /* inode held in place by reference to nd; dev by fget on fd */
Robert Love0eeca282005-07-12 17:06:03 -0400944 inode = nd.dentry->d_inode;
Robert Loveb6807162005-07-25 15:07:13 -0400945 dev = filp->private_data;
Robert Love0eeca282005-07-12 17:06:03 -0400946
947 down(&inode->inotify_sem);
948 down(&dev->sem);
949
950 /* don't let user-space set invalid bits: we don't want flags set */
951 mask &= IN_ALL_EVENTS;
Robert Loveb6807162005-07-25 15:07:13 -0400952 if (unlikely(!mask)) {
Robert Love0eeca282005-07-12 17:06:03 -0400953 ret = -EINVAL;
954 goto out;
955 }
956
957 /*
958 * Handle the case of re-adding a watch on an (inode,dev) pair that we
959 * are already watching. We just update the mask and return its wd.
960 */
961 old = inode_find_dev(inode, dev);
962 if (unlikely(old)) {
963 old->mask = mask;
964 ret = old->wd;
965 goto out;
966 }
967
968 watch = create_watch(dev, mask, inode);
969 if (unlikely(IS_ERR(watch))) {
970 ret = PTR_ERR(watch);
971 goto out;
972 }
973
974 /* Add the watch to the device's and the inode's list */
975 list_add(&watch->d_list, &dev->watches);
976 list_add(&watch->i_list, &inode->inotify_watches);
977 ret = watch->wd;
978out:
Robert Love0eeca282005-07-12 17:06:03 -0400979 up(&dev->sem);
980 up(&inode->inotify_sem);
Robert Love5eb22cb2005-07-25 15:12:19 -0400981 path_release(&nd);
Robert Love0eeca282005-07-12 17:06:03 -0400982fput_and_out:
Robert Love33ea2f52005-07-25 15:08:37 -0400983 fput_light(filp, fput_needed);
Robert Love0eeca282005-07-12 17:06:03 -0400984 return ret;
985}
986
987asmlinkage long sys_inotify_rm_watch(int fd, u32 wd)
988{
989 struct file *filp;
990 struct inotify_device *dev;
Robert Love33ea2f52005-07-25 15:08:37 -0400991 int ret, fput_needed;
Robert Love0eeca282005-07-12 17:06:03 -0400992
Robert Love33ea2f52005-07-25 15:08:37 -0400993 filp = fget_light(fd, &fput_needed);
994 if (unlikely(!filp))
Robert Love0eeca282005-07-12 17:06:03 -0400995 return -EBADF;
Robert Love783bc292005-07-25 15:10:08 -0400996
997 /* verify that this is indeed an inotify instance */
998 if (unlikely(filp->f_op != &inotify_fops)) {
999 ret = -EINVAL;
1000 goto out;
1001 }
1002
Robert Love0eeca282005-07-12 17:06:03 -04001003 dev = filp->private_data;
Robert Love9a556e82005-07-13 13:49:23 -04001004 ret = inotify_ignore(dev, wd);
Robert Love9a556e82005-07-13 13:49:23 -04001005
Robert Love783bc292005-07-25 15:10:08 -04001006out:
1007 fput_light(filp, fput_needed);
Robert Love0eeca282005-07-12 17:06:03 -04001008 return ret;
1009}
1010
1011static struct super_block *
1012inotify_get_sb(struct file_system_type *fs_type, int flags,
1013 const char *dev_name, void *data)
1014{
1015 return get_sb_pseudo(fs_type, "inotify", NULL, 0xBAD1DEA);
1016}
1017
1018static struct file_system_type inotify_fs_type = {
1019 .name = "inotifyfs",
1020 .get_sb = inotify_get_sb,
1021 .kill_sb = kill_anon_super,
1022};
1023
1024/*
Robert Loveb6807162005-07-25 15:07:13 -04001025 * inotify_setup - Our initialization function. Note that we cannnot return
Robert Love0eeca282005-07-12 17:06:03 -04001026 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
1027 * must result in panic().
1028 */
Robert Loveb6807162005-07-25 15:07:13 -04001029static int __init inotify_setup(void)
Robert Love0eeca282005-07-12 17:06:03 -04001030{
Robert Lovee5ca8442005-07-25 15:17:34 -04001031 int ret;
1032
1033 ret = register_filesystem(&inotify_fs_type);
1034 if (unlikely(ret))
1035 panic("inotify: register_filesystem returned %d!\n", ret);
1036
Robert Love0eeca282005-07-12 17:06:03 -04001037 inotify_mnt = kern_mount(&inotify_fs_type);
Andrew Morton89373de2005-07-26 14:08:38 -07001038 if (IS_ERR(inotify_mnt))
Robert Lovee5ca8442005-07-25 15:17:34 -04001039 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
Robert Love0eeca282005-07-12 17:06:03 -04001040
Robert Love1b2ccf02005-07-25 15:13:43 -04001041 inotify_max_queued_events = 16384;
1042 inotify_max_user_instances = 128;
Robert Love0eeca282005-07-12 17:06:03 -04001043 inotify_max_user_watches = 8192;
1044
1045 atomic_set(&inotify_cookie, 0);
1046
1047 watch_cachep = kmem_cache_create("inotify_watch_cache",
1048 sizeof(struct inotify_watch),
1049 0, SLAB_PANIC, NULL, NULL);
1050 event_cachep = kmem_cache_create("inotify_event_cache",
1051 sizeof(struct inotify_kernel_event),
1052 0, SLAB_PANIC, NULL, NULL);
1053
Robert Love0eeca282005-07-12 17:06:03 -04001054 return 0;
1055}
1056
Robert Loveb6807162005-07-25 15:07:13 -04001057module_init(inotify_setup);