2 * fs/inotify_user.c - inotify support for userspace
5 * John McCutchan <ttb@tentacle.dhs.org>
6 * Robert Love <rml@novell.com>
8 * Copyright (C) 2005 John McCutchan
9 * Copyright 2006 Hewlett-Packard Development Company, L.P.
11 * Copyright (C) 2009 Eric Paris <Red Hat Inc>
12 * inotify was largely rewriten to make use of the fsnotify infrastructure
14 * This program is free software; you can redistribute it and/or modify it
15 * under the terms of the GNU General Public License as published by the
16 * Free Software Foundation; either version 2, or (at your option) any
19 * This program is distributed in the hope that it will be useful, but
20 * WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 * General Public License for more details.
25 #include <linux/file.h>
26 #include <linux/fs.h> /* struct inode */
27 #include <linux/fsnotify_backend.h>
28 #include <linux/idr.h>
29 #include <linux/init.h> /* module_init */
30 #include <linux/inotify.h>
31 #include <linux/kernel.h> /* roundup() */
32 #include <linux/namei.h> /* LOOKUP_FOLLOW */
33 #include <linux/sched.h> /* struct user */
34 #include <linux/slab.h> /* struct kmem_cache */
35 #include <linux/syscalls.h>
36 #include <linux/types.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/uaccess.h>
39 #include <linux/poll.h>
40 #include <linux/wait.h>
44 #include <asm/ioctls.h>
46 /* these are configurable via /proc/sys/fs/inotify/ */
47 static int inotify_max_user_instances __read_mostly
;
48 static int inotify_max_queued_events __read_mostly
;
49 static int inotify_max_user_watches __read_mostly
;
51 static struct kmem_cache
*inotify_inode_mark_cachep __read_mostly
;
52 struct kmem_cache
*event_priv_cachep __read_mostly
;
56 #include <linux/sysctl.h>
60 ctl_table inotify_table
[] = {
62 .procname
= "max_user_instances",
63 .data
= &inotify_max_user_instances
,
64 .maxlen
= sizeof(int),
66 .proc_handler
= proc_dointvec_minmax
,
70 .procname
= "max_user_watches",
71 .data
= &inotify_max_user_watches
,
72 .maxlen
= sizeof(int),
74 .proc_handler
= proc_dointvec_minmax
,
78 .procname
= "max_queued_events",
79 .data
= &inotify_max_queued_events
,
80 .maxlen
= sizeof(int),
82 .proc_handler
= proc_dointvec_minmax
,
87 #endif /* CONFIG_SYSCTL */
89 static inline __u32
inotify_arg_to_mask(u32 arg
)
94 * everything should accept their own ignored, cares about children,
95 * and should receive events when the inode is unmounted
97 mask
= (FS_IN_IGNORED
| FS_EVENT_ON_CHILD
| FS_UNMOUNT
);
99 /* mask off the flags used to open the fd */
100 mask
|= (arg
& (IN_ALL_EVENTS
| IN_ONESHOT
));
105 static inline u32
inotify_mask_to_arg(__u32 mask
)
107 return mask
& (IN_ALL_EVENTS
| IN_ISDIR
| IN_UNMOUNT
| IN_IGNORED
|
111 /* intofiy userspace file descriptor functions */
112 static unsigned int inotify_poll(struct file
*file
, poll_table
*wait
)
114 struct fsnotify_group
*group
= file
->private_data
;
117 poll_wait(file
, &group
->notification_waitq
, wait
);
118 mutex_lock(&group
->notification_mutex
);
119 if (!fsnotify_notify_queue_is_empty(group
))
120 ret
= POLLIN
| POLLRDNORM
;
121 mutex_unlock(&group
->notification_mutex
);
127 * Get an inotify_kernel_event if one exists and is small
128 * enough to fit in "count". Return an error pointer if
131 * Called with the group->notification_mutex held.
133 static struct fsnotify_event
*get_one_event(struct fsnotify_group
*group
,
136 size_t event_size
= sizeof(struct inotify_event
);
137 struct fsnotify_event
*event
;
139 if (fsnotify_notify_queue_is_empty(group
))
142 event
= fsnotify_peek_notify_event(group
);
145 event_size
+= roundup(event
->name_len
+ 1, event_size
);
147 if (event_size
> count
)
148 return ERR_PTR(-EINVAL
);
150 /* held the notification_mutex the whole time, so this is the
151 * same event we peeked above */
152 fsnotify_remove_notify_event(group
);
158 * Copy an event to user space, returning how much we copied.
160 * We already checked that the event size is smaller than the
161 * buffer we had in "get_one_event()" above.
163 static ssize_t
copy_event_to_user(struct fsnotify_group
*group
,
164 struct fsnotify_event
*event
,
167 struct inotify_event inotify_event
;
168 struct fsnotify_event_private_data
*fsn_priv
;
169 struct inotify_event_private_data
*priv
;
170 size_t event_size
= sizeof(struct inotify_event
);
173 /* we get the inotify watch descriptor from the event private data */
174 spin_lock(&event
->lock
);
175 fsn_priv
= fsnotify_remove_priv_from_event(group
, event
);
176 spin_unlock(&event
->lock
);
179 inotify_event
.wd
= -1;
181 priv
= container_of(fsn_priv
, struct inotify_event_private_data
,
182 fsnotify_event_priv_data
);
183 inotify_event
.wd
= priv
->wd
;
184 inotify_free_event_priv(fsn_priv
);
188 * round up event->name_len so it is a multiple of event_size
189 * plus an extra byte for the terminating '\0'.
192 name_len
= roundup(event
->name_len
+ 1, event_size
);
193 inotify_event
.len
= name_len
;
195 inotify_event
.mask
= inotify_mask_to_arg(event
->mask
);
196 inotify_event
.cookie
= event
->sync_cookie
;
198 /* send the main event */
199 if (copy_to_user(buf
, &inotify_event
, event_size
))
205 * fsnotify only stores the pathname, so here we have to send the pathname
206 * and then pad that pathname out to a multiple of sizeof(inotify_event)
207 * with zeros. I get my zeros from the nul_inotify_event.
210 unsigned int len_to_zero
= name_len
- event
->name_len
;
211 /* copy the path name */
212 if (copy_to_user(buf
, event
->file_name
, event
->name_len
))
214 buf
+= event
->name_len
;
216 /* fill userspace with 0's */
217 if (clear_user(buf
, len_to_zero
))
220 event_size
+= name_len
;
226 static ssize_t
inotify_read(struct file
*file
, char __user
*buf
,
227 size_t count
, loff_t
*pos
)
229 struct fsnotify_group
*group
;
230 struct fsnotify_event
*kevent
;
236 group
= file
->private_data
;
239 prepare_to_wait(&group
->notification_waitq
, &wait
, TASK_INTERRUPTIBLE
);
241 mutex_lock(&group
->notification_mutex
);
242 kevent
= get_one_event(group
, count
);
243 mutex_unlock(&group
->notification_mutex
);
246 ret
= PTR_ERR(kevent
);
249 ret
= copy_event_to_user(group
, kevent
, buf
);
250 fsnotify_put_event(kevent
);
259 if (file
->f_flags
& O_NONBLOCK
)
262 if (signal_pending(current
))
271 finish_wait(&group
->notification_waitq
, &wait
);
272 if (start
!= buf
&& ret
!= -EFAULT
)
277 static int inotify_fasync(int fd
, struct file
*file
, int on
)
279 struct fsnotify_group
*group
= file
->private_data
;
281 return fasync_helper(fd
, file
, on
, &group
->inotify_data
.fa
) >= 0 ? 0 : -EIO
;
284 static int inotify_release(struct inode
*ignored
, struct file
*file
)
286 struct fsnotify_group
*group
= file
->private_data
;
287 struct user_struct
*user
= group
->inotify_data
.user
;
289 fsnotify_clear_marks_by_group(group
);
291 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
292 fsnotify_put_group(group
);
294 atomic_dec(&user
->inotify_devs
);
299 static long inotify_ioctl(struct file
*file
, unsigned int cmd
,
302 struct fsnotify_group
*group
;
303 struct fsnotify_event_holder
*holder
;
304 struct fsnotify_event
*event
;
309 group
= file
->private_data
;
310 p
= (void __user
*) arg
;
314 mutex_lock(&group
->notification_mutex
);
315 list_for_each_entry(holder
, &group
->notification_list
, event_list
) {
316 event
= holder
->event
;
317 send_len
+= sizeof(struct inotify_event
);
319 send_len
+= roundup(event
->name_len
+ 1,
320 sizeof(struct inotify_event
));
322 mutex_unlock(&group
->notification_mutex
);
323 ret
= put_user(send_len
, (int __user
*) p
);
330 static const struct file_operations inotify_fops
= {
331 .poll
= inotify_poll
,
332 .read
= inotify_read
,
333 .fasync
= inotify_fasync
,
334 .release
= inotify_release
,
335 .unlocked_ioctl
= inotify_ioctl
,
336 .compat_ioctl
= inotify_ioctl
,
341 * find_inode - resolve a user-given path to a specific inode
343 static int inotify_find_inode(const char __user
*dirname
, struct path
*path
, unsigned flags
)
347 error
= user_path_at(AT_FDCWD
, dirname
, flags
, path
);
350 /* you can only watch an inode if you have read permissions on it */
351 error
= inode_permission(path
->dentry
->d_inode
, MAY_READ
);
357 static int inotify_add_to_idr(struct idr
*idr
, spinlock_t
*idr_lock
,
359 struct inotify_inode_mark
*i_mark
)
364 if (unlikely(!idr_pre_get(idr
, GFP_KERNEL
)))
368 ret
= idr_get_new_above(idr
, i_mark
, *last_wd
+ 1,
370 /* we added the mark to the idr, take a reference */
372 *last_wd
= i_mark
->wd
;
373 fsnotify_get_mark(&i_mark
->fsn_mark
);
375 spin_unlock(idr_lock
);
376 } while (ret
== -EAGAIN
);
381 static struct inotify_inode_mark
*inotify_idr_find_locked(struct fsnotify_group
*group
,
384 struct idr
*idr
= &group
->inotify_data
.idr
;
385 spinlock_t
*idr_lock
= &group
->inotify_data
.idr_lock
;
386 struct inotify_inode_mark
*i_mark
;
388 assert_spin_locked(idr_lock
);
390 i_mark
= idr_find(idr
, wd
);
392 struct fsnotify_mark
*fsn_mark
= &i_mark
->fsn_mark
;
394 fsnotify_get_mark(fsn_mark
);
395 /* One ref for being in the idr, one ref we just took */
396 BUG_ON(atomic_read(&fsn_mark
->refcnt
) < 2);
402 static struct inotify_inode_mark
*inotify_idr_find(struct fsnotify_group
*group
,
405 struct inotify_inode_mark
*i_mark
;
406 spinlock_t
*idr_lock
= &group
->inotify_data
.idr_lock
;
409 i_mark
= inotify_idr_find_locked(group
, wd
);
410 spin_unlock(idr_lock
);
415 static void do_inotify_remove_from_idr(struct fsnotify_group
*group
,
416 struct inotify_inode_mark
*i_mark
)
418 struct idr
*idr
= &group
->inotify_data
.idr
;
419 spinlock_t
*idr_lock
= &group
->inotify_data
.idr_lock
;
422 assert_spin_locked(idr_lock
);
426 /* removed from the idr, drop that ref */
427 fsnotify_put_mark(&i_mark
->fsn_mark
);
431 * Remove the mark from the idr (if present) and drop the reference
432 * on the mark because it was in the idr.
434 static void inotify_remove_from_idr(struct fsnotify_group
*group
,
435 struct inotify_inode_mark
*i_mark
)
437 spinlock_t
*idr_lock
= &group
->inotify_data
.idr_lock
;
438 struct inotify_inode_mark
*found_i_mark
= NULL
;
445 * does this i_mark think it is in the idr? we shouldn't get called
449 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
450 " i_mark->inode=%p\n", __func__
, i_mark
, i_mark
->wd
,
451 i_mark
->fsn_mark
.group
, i_mark
->fsn_mark
.i
.inode
);
455 /* Lets look in the idr to see if we find it */
456 found_i_mark
= inotify_idr_find_locked(group
, wd
);
457 if (unlikely(!found_i_mark
)) {
458 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
459 " i_mark->inode=%p\n", __func__
, i_mark
, i_mark
->wd
,
460 i_mark
->fsn_mark
.group
, i_mark
->fsn_mark
.i
.inode
);
465 * We found an mark in the idr at the right wd, but it's
466 * not the mark we were told to remove. eparis seriously
467 * fucked up somewhere.
469 if (unlikely(found_i_mark
!= i_mark
)) {
470 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
471 "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
472 "found_i_mark->group=%p found_i_mark->inode=%p\n",
473 __func__
, i_mark
, i_mark
->wd
, i_mark
->fsn_mark
.group
,
474 i_mark
->fsn_mark
.i
.inode
, found_i_mark
, found_i_mark
->wd
,
475 found_i_mark
->fsn_mark
.group
,
476 found_i_mark
->fsn_mark
.i
.inode
);
481 * One ref for being in the idr
482 * one ref held by the caller trying to kill us
483 * one ref grabbed by inotify_idr_find
485 if (unlikely(atomic_read(&i_mark
->fsn_mark
.refcnt
) < 3)) {
486 printk(KERN_ERR
"%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
487 " i_mark->inode=%p\n", __func__
, i_mark
, i_mark
->wd
,
488 i_mark
->fsn_mark
.group
, i_mark
->fsn_mark
.i
.inode
);
489 /* we can't really recover with bad ref cnting.. */
493 do_inotify_remove_from_idr(group
, i_mark
);
495 /* match the ref taken by inotify_idr_find_locked() */
497 fsnotify_put_mark(&found_i_mark
->fsn_mark
);
499 spin_unlock(idr_lock
);
503 * Send IN_IGNORED for this wd, remove this wd from the idr.
505 void inotify_ignored_and_remove_idr(struct fsnotify_mark
*fsn_mark
,
506 struct fsnotify_group
*group
)
508 struct inotify_inode_mark
*i_mark
;
509 struct fsnotify_event
*ignored_event
;
510 struct inotify_event_private_data
*event_priv
;
511 struct fsnotify_event_private_data
*fsn_event_priv
;
514 ignored_event
= fsnotify_create_event(NULL
, FS_IN_IGNORED
, NULL
,
515 FSNOTIFY_EVENT_NONE
, NULL
, 0,
520 i_mark
= container_of(fsn_mark
, struct inotify_inode_mark
, fsn_mark
);
522 event_priv
= kmem_cache_alloc(event_priv_cachep
, GFP_NOFS
);
523 if (unlikely(!event_priv
))
524 goto skip_send_ignore
;
526 fsn_event_priv
= &event_priv
->fsnotify_event_priv_data
;
528 fsn_event_priv
->group
= group
;
529 event_priv
->wd
= i_mark
->wd
;
531 ret
= fsnotify_add_notify_event(group
, ignored_event
, fsn_event_priv
, NULL
, NULL
);
533 inotify_free_event_priv(fsn_event_priv
);
537 /* matches the reference taken when the event was created */
538 fsnotify_put_event(ignored_event
);
540 /* remove this mark from the idr */
541 inotify_remove_from_idr(group
, i_mark
);
543 atomic_dec(&group
->inotify_data
.user
->inotify_watches
);
546 /* ding dong the mark is dead */
547 static void inotify_free_mark(struct fsnotify_mark
*fsn_mark
)
549 struct inotify_inode_mark
*i_mark
;
551 i_mark
= container_of(fsn_mark
, struct inotify_inode_mark
, fsn_mark
);
553 kmem_cache_free(inotify_inode_mark_cachep
, i_mark
);
556 static int inotify_update_existing_watch(struct fsnotify_group
*group
,
560 struct fsnotify_mark
*fsn_mark
;
561 struct inotify_inode_mark
*i_mark
;
562 __u32 old_mask
, new_mask
;
564 int add
= (arg
& IN_MASK_ADD
);
567 /* don't allow invalid bits: we don't want flags set */
568 mask
= inotify_arg_to_mask(arg
);
572 fsn_mark
= fsnotify_find_inode_mark(group
, inode
);
576 i_mark
= container_of(fsn_mark
, struct inotify_inode_mark
, fsn_mark
);
578 spin_lock(&fsn_mark
->lock
);
580 old_mask
= fsn_mark
->mask
;
582 fsnotify_set_mark_mask_locked(fsn_mark
, (fsn_mark
->mask
| mask
));
584 fsnotify_set_mark_mask_locked(fsn_mark
, mask
);
585 new_mask
= fsn_mark
->mask
;
587 spin_unlock(&fsn_mark
->lock
);
589 if (old_mask
!= new_mask
) {
590 /* more bits in old than in new? */
591 int dropped
= (old_mask
& ~new_mask
);
592 /* more bits in this fsn_mark than the inode's mask? */
593 int do_inode
= (new_mask
& ~inode
->i_fsnotify_mask
);
594 /* more bits in this fsn_mark than the group? */
595 int do_group
= (new_mask
& ~group
->mask
);
597 /* update the inode with this new fsn_mark */
598 if (dropped
|| do_inode
)
599 fsnotify_recalc_inode_mask(inode
);
601 /* update the group mask with the new mask */
602 if (dropped
|| do_group
)
603 fsnotify_recalc_group_mask(group
);
609 /* match the get from fsnotify_find_mark() */
610 fsnotify_put_mark(fsn_mark
);
615 static int inotify_new_watch(struct fsnotify_group
*group
,
619 struct inotify_inode_mark
*tmp_i_mark
;
622 struct idr
*idr
= &group
->inotify_data
.idr
;
623 spinlock_t
*idr_lock
= &group
->inotify_data
.idr_lock
;
625 /* don't allow invalid bits: we don't want flags set */
626 mask
= inotify_arg_to_mask(arg
);
630 tmp_i_mark
= kmem_cache_alloc(inotify_inode_mark_cachep
, GFP_KERNEL
);
631 if (unlikely(!tmp_i_mark
))
634 fsnotify_init_mark(&tmp_i_mark
->fsn_mark
, inotify_free_mark
);
635 tmp_i_mark
->fsn_mark
.mask
= mask
;
639 if (atomic_read(&group
->inotify_data
.user
->inotify_watches
) >= inotify_max_user_watches
)
642 ret
= inotify_add_to_idr(idr
, idr_lock
, &group
->inotify_data
.last_wd
,
647 /* we are on the idr, now get on the inode */
648 ret
= fsnotify_add_mark(&tmp_i_mark
->fsn_mark
, group
, inode
, NULL
, 0);
650 /* we failed to get on the inode, get off the idr */
651 inotify_remove_from_idr(group
, tmp_i_mark
);
655 /* increment the number of watches the user has */
656 atomic_inc(&group
->inotify_data
.user
->inotify_watches
);
658 /* return the watch descriptor for this new mark */
659 ret
= tmp_i_mark
->wd
;
661 /* if this mark added a new event update the group mask */
662 if (mask
& ~group
->mask
)
663 fsnotify_recalc_group_mask(group
);
666 /* match the ref from fsnotify_init_mark() */
667 fsnotify_put_mark(&tmp_i_mark
->fsn_mark
);
672 static int inotify_update_watch(struct fsnotify_group
*group
, struct inode
*inode
, u32 arg
)
677 /* try to update and existing watch with the new arg */
678 ret
= inotify_update_existing_watch(group
, inode
, arg
);
679 /* no mark present, try to add a new one */
681 ret
= inotify_new_watch(group
, inode
, arg
);
683 * inotify_new_watch could race with another thread which did an
684 * inotify_new_watch between the update_existing and the add watch
685 * here, go back and try to update an existing mark again.
693 static struct fsnotify_group
*inotify_new_group(struct user_struct
*user
, unsigned int max_events
)
695 struct fsnotify_group
*group
;
697 group
= fsnotify_alloc_group(&inotify_fsnotify_ops
);
701 group
->max_events
= max_events
;
703 spin_lock_init(&group
->inotify_data
.idr_lock
);
704 idr_init(&group
->inotify_data
.idr
);
705 group
->inotify_data
.last_wd
= 0;
706 group
->inotify_data
.user
= user
;
707 group
->inotify_data
.fa
= NULL
;
713 /* inotify syscalls */
714 SYSCALL_DEFINE1(inotify_init1
, int, flags
)
716 struct fsnotify_group
*group
;
717 struct user_struct
*user
;
720 /* Check the IN_* constants for consistency. */
721 BUILD_BUG_ON(IN_CLOEXEC
!= O_CLOEXEC
);
722 BUILD_BUG_ON(IN_NONBLOCK
!= O_NONBLOCK
);
724 if (flags
& ~(IN_CLOEXEC
| IN_NONBLOCK
))
727 user
= get_current_user();
728 if (unlikely(atomic_read(&user
->inotify_devs
) >=
729 inotify_max_user_instances
)) {
734 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
735 group
= inotify_new_group(user
, inotify_max_queued_events
);
737 ret
= PTR_ERR(group
);
741 atomic_inc(&user
->inotify_devs
);
743 ret
= anon_inode_getfd("inotify", &inotify_fops
, group
,
748 atomic_dec(&user
->inotify_devs
);
754 SYSCALL_DEFINE0(inotify_init
)
756 return sys_inotify_init1(0);
759 SYSCALL_DEFINE3(inotify_add_watch
, int, fd
, const char __user
*, pathname
,
762 struct fsnotify_group
*group
;
766 int ret
, fput_needed
;
769 filp
= fget_light(fd
, &fput_needed
);
773 /* verify that this is indeed an inotify instance */
774 if (unlikely(filp
->f_op
!= &inotify_fops
)) {
779 if (!(mask
& IN_DONT_FOLLOW
))
780 flags
|= LOOKUP_FOLLOW
;
781 if (mask
& IN_ONLYDIR
)
782 flags
|= LOOKUP_DIRECTORY
;
784 ret
= inotify_find_inode(pathname
, &path
, flags
);
788 /* inode held in place by reference to path; group by fget on fd */
789 inode
= path
.dentry
->d_inode
;
790 group
= filp
->private_data
;
792 /* create/update an inode mark */
793 ret
= inotify_update_watch(group
, inode
, mask
);
796 fput_light(filp
, fput_needed
);
800 SYSCALL_DEFINE2(inotify_rm_watch
, int, fd
, __s32
, wd
)
802 struct fsnotify_group
*group
;
803 struct inotify_inode_mark
*i_mark
;
805 int ret
= 0, fput_needed
;
807 filp
= fget_light(fd
, &fput_needed
);
811 /* verify that this is indeed an inotify instance */
813 if (unlikely(filp
->f_op
!= &inotify_fops
))
816 group
= filp
->private_data
;
819 i_mark
= inotify_idr_find(group
, wd
);
820 if (unlikely(!i_mark
))
825 fsnotify_destroy_mark(&i_mark
->fsn_mark
);
827 /* match ref taken by inotify_idr_find */
828 fsnotify_put_mark(&i_mark
->fsn_mark
);
831 fput_light(filp
, fput_needed
);
836 * inotify_user_setup - Our initialization function. Note that we cannnot return
837 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
838 * must result in panic().
840 static int __init
inotify_user_setup(void)
842 inotify_inode_mark_cachep
= KMEM_CACHE(inotify_inode_mark
, SLAB_PANIC
);
843 event_priv_cachep
= KMEM_CACHE(inotify_event_private_data
, SLAB_PANIC
);
845 inotify_max_queued_events
= 16384;
846 inotify_max_user_instances
= 128;
847 inotify_max_user_watches
= 8192;
851 module_init(inotify_user_setup
);