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/magic.h> /* superblock magic number */
33 #include <linux/mount.h> /* mntget */
34 #include <linux/namei.h> /* LOOKUP_FOLLOW */
35 #include <linux/path.h> /* struct path */
36 #include <linux/sched.h> /* struct user */
37 #include <linux/slab.h> /* struct kmem_cache */
38 #include <linux/syscalls.h>
39 #include <linux/types.h>
40 #include <linux/uaccess.h>
41 #include <linux/poll.h>
42 #include <linux/wait.h>
46 #include <asm/ioctls.h>
48 static struct vfsmount
*inotify_mnt __read_mostly
;
50 /* these are configurable via /proc/sys/fs/inotify/ */
51 static int inotify_max_user_instances __read_mostly
;
52 static int inotify_max_queued_events __read_mostly
;
53 int inotify_max_user_watches __read_mostly
;
55 static struct kmem_cache
*inotify_inode_mark_cachep __read_mostly
;
56 struct kmem_cache
*event_priv_cachep __read_mostly
;
59 * When inotify registers a new group it increments this and uses that
60 * value as an offset to set the fsnotify group "name" and priority.
62 static atomic_t inotify_grp_num
;
66 #include <linux/sysctl.h>
70 ctl_table inotify_table
[] = {
72 .ctl_name
= INOTIFY_MAX_USER_INSTANCES
,
73 .procname
= "max_user_instances",
74 .data
= &inotify_max_user_instances
,
75 .maxlen
= sizeof(int),
77 .proc_handler
= &proc_dointvec_minmax
,
78 .strategy
= &sysctl_intvec
,
82 .ctl_name
= INOTIFY_MAX_USER_WATCHES
,
83 .procname
= "max_user_watches",
84 .data
= &inotify_max_user_watches
,
85 .maxlen
= sizeof(int),
87 .proc_handler
= &proc_dointvec_minmax
,
88 .strategy
= &sysctl_intvec
,
92 .ctl_name
= INOTIFY_MAX_QUEUED_EVENTS
,
93 .procname
= "max_queued_events",
94 .data
= &inotify_max_queued_events
,
95 .maxlen
= sizeof(int),
97 .proc_handler
= &proc_dointvec_minmax
,
98 .strategy
= &sysctl_intvec
,
103 #endif /* CONFIG_SYSCTL */
105 static inline __u32
inotify_arg_to_mask(u32 arg
)
109 /* everything should accept their own ignored and cares about children */
110 mask
= (FS_IN_IGNORED
| FS_EVENT_ON_CHILD
);
112 /* mask off the flags used to open the fd */
113 mask
|= (arg
& (IN_ALL_EVENTS
| IN_ONESHOT
));
118 static inline u32
inotify_mask_to_arg(__u32 mask
)
120 return mask
& (IN_ALL_EVENTS
| IN_ISDIR
| IN_UNMOUNT
| IN_IGNORED
|
124 /* intofiy userspace file descriptor functions */
125 static unsigned int inotify_poll(struct file
*file
, poll_table
*wait
)
127 struct fsnotify_group
*group
= file
->private_data
;
130 poll_wait(file
, &group
->notification_waitq
, wait
);
131 mutex_lock(&group
->notification_mutex
);
132 if (!fsnotify_notify_queue_is_empty(group
))
133 ret
= POLLIN
| POLLRDNORM
;
134 mutex_unlock(&group
->notification_mutex
);
140 * Get an inotify_kernel_event if one exists and is small
141 * enough to fit in "count". Return an error pointer if
144 * Called with the group->notification_mutex held.
146 static struct fsnotify_event
*get_one_event(struct fsnotify_group
*group
,
149 size_t event_size
= sizeof(struct inotify_event
);
150 struct fsnotify_event
*event
;
152 if (fsnotify_notify_queue_is_empty(group
))
155 event
= fsnotify_peek_notify_event(group
);
158 event_size
+= roundup(event
->name_len
+ 1, event_size
);
160 if (event_size
> count
)
161 return ERR_PTR(-EINVAL
);
163 /* held the notification_mutex the whole time, so this is the
164 * same event we peeked above */
165 fsnotify_remove_notify_event(group
);
171 * Copy an event to user space, returning how much we copied.
173 * We already checked that the event size is smaller than the
174 * buffer we had in "get_one_event()" above.
176 static ssize_t
copy_event_to_user(struct fsnotify_group
*group
,
177 struct fsnotify_event
*event
,
180 struct inotify_event inotify_event
;
181 struct fsnotify_event_private_data
*fsn_priv
;
182 struct inotify_event_private_data
*priv
;
183 size_t event_size
= sizeof(struct inotify_event
);
186 /* we get the inotify watch descriptor from the event private data */
187 spin_lock(&event
->lock
);
188 fsn_priv
= fsnotify_remove_priv_from_event(group
, event
);
189 spin_unlock(&event
->lock
);
192 inotify_event
.wd
= -1;
194 priv
= container_of(fsn_priv
, struct inotify_event_private_data
,
195 fsnotify_event_priv_data
);
196 inotify_event
.wd
= priv
->wd
;
197 inotify_free_event_priv(fsn_priv
);
201 * round up event->name_len so it is a multiple of event_size
202 * plus an extra byte for the terminating '\0'.
205 name_len
= roundup(event
->name_len
+ 1, event_size
);
206 inotify_event
.len
= name_len
;
208 inotify_event
.mask
= inotify_mask_to_arg(event
->mask
);
209 inotify_event
.cookie
= event
->sync_cookie
;
211 /* send the main event */
212 if (copy_to_user(buf
, &inotify_event
, event_size
))
218 * fsnotify only stores the pathname, so here we have to send the pathname
219 * and then pad that pathname out to a multiple of sizeof(inotify_event)
220 * with zeros. I get my zeros from the nul_inotify_event.
223 unsigned int len_to_zero
= name_len
- event
->name_len
;
224 /* copy the path name */
225 if (copy_to_user(buf
, event
->file_name
, event
->name_len
))
227 buf
+= event
->name_len
;
229 /* fill userspace with 0's */
230 if (clear_user(buf
, len_to_zero
))
233 event_size
+= name_len
;
239 static ssize_t
inotify_read(struct file
*file
, char __user
*buf
,
240 size_t count
, loff_t
*pos
)
242 struct fsnotify_group
*group
;
243 struct fsnotify_event
*kevent
;
249 group
= file
->private_data
;
252 prepare_to_wait(&group
->notification_waitq
, &wait
, TASK_INTERRUPTIBLE
);
254 mutex_lock(&group
->notification_mutex
);
255 kevent
= get_one_event(group
, count
);
256 mutex_unlock(&group
->notification_mutex
);
259 ret
= PTR_ERR(kevent
);
262 ret
= copy_event_to_user(group
, kevent
, buf
);
263 fsnotify_put_event(kevent
);
272 if (file
->f_flags
& O_NONBLOCK
)
275 if (signal_pending(current
))
284 finish_wait(&group
->notification_waitq
, &wait
);
285 if (start
!= buf
&& ret
!= -EFAULT
)
290 static int inotify_fasync(int fd
, struct file
*file
, int on
)
292 struct fsnotify_group
*group
= file
->private_data
;
294 return fasync_helper(fd
, file
, on
, &group
->inotify_data
.fa
) >= 0 ? 0 : -EIO
;
297 static int inotify_release(struct inode
*ignored
, struct file
*file
)
299 struct fsnotify_group
*group
= file
->private_data
;
300 struct user_struct
*user
= group
->inotify_data
.user
;
302 fsnotify_clear_marks_by_group(group
);
304 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
305 fsnotify_put_group(group
);
307 atomic_dec(&user
->inotify_devs
);
312 static long inotify_ioctl(struct file
*file
, unsigned int cmd
,
315 struct fsnotify_group
*group
;
316 struct fsnotify_event_holder
*holder
;
317 struct fsnotify_event
*event
;
322 group
= file
->private_data
;
323 p
= (void __user
*) arg
;
327 mutex_lock(&group
->notification_mutex
);
328 list_for_each_entry(holder
, &group
->notification_list
, event_list
) {
329 event
= holder
->event
;
330 send_len
+= sizeof(struct inotify_event
);
332 send_len
+= roundup(event
->name_len
+ 1,
333 sizeof(struct inotify_event
));
335 mutex_unlock(&group
->notification_mutex
);
336 ret
= put_user(send_len
, (int __user
*) p
);
343 static const struct file_operations inotify_fops
= {
344 .poll
= inotify_poll
,
345 .read
= inotify_read
,
346 .fasync
= inotify_fasync
,
347 .release
= inotify_release
,
348 .unlocked_ioctl
= inotify_ioctl
,
349 .compat_ioctl
= inotify_ioctl
,
354 * find_inode - resolve a user-given path to a specific inode
356 static int inotify_find_inode(const char __user
*dirname
, struct path
*path
, unsigned flags
)
360 error
= user_path_at(AT_FDCWD
, dirname
, flags
, path
);
363 /* you can only watch an inode if you have read permissions on it */
364 error
= inode_permission(path
->dentry
->d_inode
, MAY_READ
);
371 * Remove the mark from the idr (if present) and drop the reference
372 * on the mark because it was in the idr.
374 static void inotify_remove_from_idr(struct fsnotify_group
*group
,
375 struct inotify_inode_mark_entry
*ientry
)
378 struct fsnotify_mark_entry
*entry
;
379 struct inotify_inode_mark_entry
*found_ientry
;
382 spin_lock(&group
->inotify_data
.idr_lock
);
383 idr
= &group
->inotify_data
.idr
;
389 entry
= idr_find(&group
->inotify_data
.idr
, wd
);
390 if (unlikely(!entry
))
393 found_ientry
= container_of(entry
, struct inotify_inode_mark_entry
, fsn_entry
);
394 if (unlikely(found_ientry
!= ientry
)) {
395 /* We found an entry in the idr with the right wd, but it's
396 * not the entry we were told to remove. eparis seriously
397 * fucked up somewhere. */
403 /* One ref for being in the idr, one ref held by the caller */
404 BUG_ON(atomic_read(&entry
->refcnt
) < 2);
409 /* removed from the idr, drop that ref */
410 fsnotify_put_mark(entry
);
412 spin_unlock(&group
->inotify_data
.idr_lock
);
416 * Send IN_IGNORED for this wd, remove this wd from the idr.
418 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry
*entry
,
419 struct fsnotify_group
*group
)
421 struct inotify_inode_mark_entry
*ientry
;
422 struct fsnotify_event
*ignored_event
;
423 struct inotify_event_private_data
*event_priv
;
424 struct fsnotify_event_private_data
*fsn_event_priv
;
427 ignored_event
= fsnotify_create_event(NULL
, FS_IN_IGNORED
, NULL
,
428 FSNOTIFY_EVENT_NONE
, NULL
, 0,
433 ientry
= container_of(entry
, struct inotify_inode_mark_entry
, fsn_entry
);
435 event_priv
= kmem_cache_alloc(event_priv_cachep
, GFP_NOFS
);
436 if (unlikely(!event_priv
))
437 goto skip_send_ignore
;
439 fsn_event_priv
= &event_priv
->fsnotify_event_priv_data
;
441 fsn_event_priv
->group
= group
;
442 event_priv
->wd
= ientry
->wd
;
444 ret
= fsnotify_add_notify_event(group
, ignored_event
, fsn_event_priv
);
446 inotify_free_event_priv(fsn_event_priv
);
450 /* matches the reference taken when the event was created */
451 fsnotify_put_event(ignored_event
);
453 /* remove this entry from the idr */
454 inotify_remove_from_idr(group
, ientry
);
456 atomic_dec(&group
->inotify_data
.user
->inotify_watches
);
459 /* ding dong the mark is dead */
460 static void inotify_free_mark(struct fsnotify_mark_entry
*entry
)
462 struct inotify_inode_mark_entry
*ientry
= (struct inotify_inode_mark_entry
*)entry
;
464 kmem_cache_free(inotify_inode_mark_cachep
, ientry
);
467 static int inotify_update_existing_watch(struct fsnotify_group
*group
,
471 struct fsnotify_mark_entry
*entry
;
472 struct inotify_inode_mark_entry
*ientry
;
473 __u32 old_mask
, new_mask
;
475 int add
= (arg
& IN_MASK_ADD
);
478 /* don't allow invalid bits: we don't want flags set */
479 mask
= inotify_arg_to_mask(arg
);
483 spin_lock(&inode
->i_lock
);
484 entry
= fsnotify_find_mark_entry(group
, inode
);
485 spin_unlock(&inode
->i_lock
);
489 ientry
= container_of(entry
, struct inotify_inode_mark_entry
, fsn_entry
);
491 spin_lock(&entry
->lock
);
493 old_mask
= entry
->mask
;
496 new_mask
= entry
->mask
;
499 new_mask
= entry
->mask
;
502 spin_unlock(&entry
->lock
);
504 if (old_mask
!= new_mask
) {
505 /* more bits in old than in new? */
506 int dropped
= (old_mask
& ~new_mask
);
507 /* more bits in this entry than the inode's mask? */
508 int do_inode
= (new_mask
& ~inode
->i_fsnotify_mask
);
509 /* more bits in this entry than the group? */
510 int do_group
= (new_mask
& ~group
->mask
);
512 /* update the inode with this new entry */
513 if (dropped
|| do_inode
)
514 fsnotify_recalc_inode_mask(inode
);
516 /* update the group mask with the new mask */
517 if (dropped
|| do_group
)
518 fsnotify_recalc_group_mask(group
);
524 /* match the get from fsnotify_find_mark_entry() */
525 fsnotify_put_mark(entry
);
530 static int inotify_new_watch(struct fsnotify_group
*group
,
534 struct inotify_inode_mark_entry
*tmp_ientry
;
538 /* don't allow invalid bits: we don't want flags set */
539 mask
= inotify_arg_to_mask(arg
);
543 tmp_ientry
= kmem_cache_alloc(inotify_inode_mark_cachep
, GFP_KERNEL
);
544 if (unlikely(!tmp_ientry
))
547 fsnotify_init_mark(&tmp_ientry
->fsn_entry
, inotify_free_mark
);
548 tmp_ientry
->fsn_entry
.mask
= mask
;
552 if (atomic_read(&group
->inotify_data
.user
->inotify_watches
) >= inotify_max_user_watches
)
556 if (unlikely(!idr_pre_get(&group
->inotify_data
.idr
, GFP_KERNEL
)))
559 spin_lock(&group
->inotify_data
.idr_lock
);
560 ret
= idr_get_new_above(&group
->inotify_data
.idr
, &tmp_ientry
->fsn_entry
,
561 group
->inotify_data
.last_wd
,
563 spin_unlock(&group
->inotify_data
.idr_lock
);
565 /* idr was out of memory allocate and try again */
571 /* we put the mark on the idr, take a reference */
572 fsnotify_get_mark(&tmp_ientry
->fsn_entry
);
574 /* we are on the idr, now get on the inode */
575 ret
= fsnotify_add_mark(&tmp_ientry
->fsn_entry
, group
, inode
);
577 /* we failed to get on the inode, get off the idr */
578 inotify_remove_from_idr(group
, tmp_ientry
);
582 /* update the idr hint, who cares about races, it's just a hint */
583 group
->inotify_data
.last_wd
= tmp_ientry
->wd
;
585 /* increment the number of watches the user has */
586 atomic_inc(&group
->inotify_data
.user
->inotify_watches
);
588 /* return the watch descriptor for this new entry */
589 ret
= tmp_ientry
->wd
;
591 /* match the ref from fsnotify_init_markentry() */
592 fsnotify_put_mark(&tmp_ientry
->fsn_entry
);
594 /* if this mark added a new event update the group mask */
595 if (mask
& ~group
->mask
)
596 fsnotify_recalc_group_mask(group
);
600 kmem_cache_free(inotify_inode_mark_cachep
, tmp_ientry
);
605 static int inotify_update_watch(struct fsnotify_group
*group
, struct inode
*inode
, u32 arg
)
610 /* try to update and existing watch with the new arg */
611 ret
= inotify_update_existing_watch(group
, inode
, arg
);
612 /* no mark present, try to add a new one */
614 ret
= inotify_new_watch(group
, inode
, arg
);
616 * inotify_new_watch could race with another thread which did an
617 * inotify_new_watch between the update_existing and the add watch
618 * here, go back and try to update an existing mark again.
626 static struct fsnotify_group
*inotify_new_group(struct user_struct
*user
, unsigned int max_events
)
628 struct fsnotify_group
*group
;
629 unsigned int grp_num
;
631 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
632 grp_num
= (INOTIFY_GROUP_NUM
- atomic_inc_return(&inotify_grp_num
));
633 group
= fsnotify_obtain_group(grp_num
, 0, &inotify_fsnotify_ops
);
637 group
->max_events
= max_events
;
639 spin_lock_init(&group
->inotify_data
.idr_lock
);
640 idr_init(&group
->inotify_data
.idr
);
641 group
->inotify_data
.last_wd
= 1;
642 group
->inotify_data
.user
= user
;
643 group
->inotify_data
.fa
= NULL
;
649 /* inotify syscalls */
650 SYSCALL_DEFINE1(inotify_init1
, int, flags
)
652 struct fsnotify_group
*group
;
653 struct user_struct
*user
;
657 /* Check the IN_* constants for consistency. */
658 BUILD_BUG_ON(IN_CLOEXEC
!= O_CLOEXEC
);
659 BUILD_BUG_ON(IN_NONBLOCK
!= O_NONBLOCK
);
661 if (flags
& ~(IN_CLOEXEC
| IN_NONBLOCK
))
664 fd
= get_unused_fd_flags(flags
& O_CLOEXEC
);
668 filp
= get_empty_filp();
674 user
= get_current_user();
675 if (unlikely(atomic_read(&user
->inotify_devs
) >=
676 inotify_max_user_instances
)) {
681 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
682 group
= inotify_new_group(user
, inotify_max_queued_events
);
684 ret
= PTR_ERR(group
);
688 filp
->f_op
= &inotify_fops
;
689 filp
->f_path
.mnt
= mntget(inotify_mnt
);
690 filp
->f_path
.dentry
= dget(inotify_mnt
->mnt_root
);
691 filp
->f_mapping
= filp
->f_path
.dentry
->d_inode
->i_mapping
;
692 filp
->f_mode
= FMODE_READ
;
693 filp
->f_flags
= O_RDONLY
| (flags
& O_NONBLOCK
);
694 filp
->private_data
= group
;
696 atomic_inc(&user
->inotify_devs
);
698 fd_install(fd
, filp
);
710 SYSCALL_DEFINE0(inotify_init
)
712 return sys_inotify_init1(0);
715 SYSCALL_DEFINE3(inotify_add_watch
, int, fd
, const char __user
*, pathname
,
718 struct fsnotify_group
*group
;
722 int ret
, fput_needed
;
725 filp
= fget_light(fd
, &fput_needed
);
729 /* verify that this is indeed an inotify instance */
730 if (unlikely(filp
->f_op
!= &inotify_fops
)) {
735 if (!(mask
& IN_DONT_FOLLOW
))
736 flags
|= LOOKUP_FOLLOW
;
737 if (mask
& IN_ONLYDIR
)
738 flags
|= LOOKUP_DIRECTORY
;
740 ret
= inotify_find_inode(pathname
, &path
, flags
);
744 /* inode held in place by reference to path; group by fget on fd */
745 inode
= path
.dentry
->d_inode
;
746 group
= filp
->private_data
;
748 /* create/update an inode mark */
749 ret
= inotify_update_watch(group
, inode
, mask
);
751 goto path_put_and_out
;
756 fput_light(filp
, fput_needed
);
760 SYSCALL_DEFINE2(inotify_rm_watch
, int, fd
, __s32
, wd
)
762 struct fsnotify_group
*group
;
763 struct fsnotify_mark_entry
*entry
;
765 int ret
= 0, fput_needed
;
767 filp
= fget_light(fd
, &fput_needed
);
771 /* verify that this is indeed an inotify instance */
772 if (unlikely(filp
->f_op
!= &inotify_fops
)) {
777 group
= filp
->private_data
;
779 spin_lock(&group
->inotify_data
.idr_lock
);
780 entry
= idr_find(&group
->inotify_data
.idr
, wd
);
781 if (unlikely(!entry
)) {
782 spin_unlock(&group
->inotify_data
.idr_lock
);
786 fsnotify_get_mark(entry
);
787 spin_unlock(&group
->inotify_data
.idr_lock
);
789 fsnotify_destroy_mark_by_entry(entry
);
790 fsnotify_put_mark(entry
);
793 fput_light(filp
, fput_needed
);
798 inotify_get_sb(struct file_system_type
*fs_type
, int flags
,
799 const char *dev_name
, void *data
, struct vfsmount
*mnt
)
801 return get_sb_pseudo(fs_type
, "inotify", NULL
,
802 INOTIFYFS_SUPER_MAGIC
, mnt
);
805 static struct file_system_type inotify_fs_type
= {
807 .get_sb
= inotify_get_sb
,
808 .kill_sb
= kill_anon_super
,
812 * inotify_user_setup - Our initialization function. Note that we cannnot return
813 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
814 * must result in panic().
816 static int __init
inotify_user_setup(void)
820 ret
= register_filesystem(&inotify_fs_type
);
822 panic("inotify: register_filesystem returned %d!\n", ret
);
824 inotify_mnt
= kern_mount(&inotify_fs_type
);
825 if (IS_ERR(inotify_mnt
))
826 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt
));
828 inotify_inode_mark_cachep
= KMEM_CACHE(inotify_inode_mark_entry
, SLAB_PANIC
);
829 event_priv_cachep
= KMEM_CACHE(inotify_event_private_data
, SLAB_PANIC
);
831 inotify_max_queued_events
= 16384;
832 inotify_max_user_instances
= 128;
833 inotify_max_user_watches
= 8192;
837 module_init(inotify_user_setup
);