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 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)
93 /* everything should accept their own ignored and cares about children */
94 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
96 /* mask off the flags used to open the fd */
97 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
102 static inline u32 inotify_mask_to_arg(__u32 mask)
104 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
108 /* intofiy userspace file descriptor functions */
109 static unsigned int inotify_poll(struct file *file, poll_table *wait)
111 struct fsnotify_group *group = file->private_data;
114 poll_wait(file, &group->notification_waitq, wait);
115 mutex_lock(&group->notification_mutex);
116 if (!fsnotify_notify_queue_is_empty(group))
117 ret = POLLIN | POLLRDNORM;
118 mutex_unlock(&group->notification_mutex);
124 * Get an inotify_kernel_event if one exists and is small
125 * enough to fit in "count". Return an error pointer if
128 * Called with the group->notification_mutex held.
130 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
133 size_t event_size = sizeof(struct inotify_event);
134 struct fsnotify_event *event;
136 if (fsnotify_notify_queue_is_empty(group))
139 event = fsnotify_peek_notify_event(group);
142 event_size += roundup(event->name_len + 1, event_size);
144 if (event_size > count)
145 return ERR_PTR(-EINVAL);
147 /* held the notification_mutex the whole time, so this is the
148 * same event we peeked above */
149 fsnotify_remove_notify_event(group);
155 * Copy an event to user space, returning how much we copied.
157 * We already checked that the event size is smaller than the
158 * buffer we had in "get_one_event()" above.
160 static ssize_t copy_event_to_user(struct fsnotify_group *group,
161 struct fsnotify_event *event,
164 struct inotify_event inotify_event;
165 struct fsnotify_event_private_data *fsn_priv;
166 struct inotify_event_private_data *priv;
167 size_t event_size = sizeof(struct inotify_event);
170 /* we get the inotify watch descriptor from the event private data */
171 spin_lock(&event->lock);
172 fsn_priv = fsnotify_remove_priv_from_event(group, event);
173 spin_unlock(&event->lock);
176 inotify_event.wd = -1;
178 priv = container_of(fsn_priv, struct inotify_event_private_data,
179 fsnotify_event_priv_data);
180 inotify_event.wd = priv->wd;
181 inotify_free_event_priv(fsn_priv);
185 * round up event->name_len so it is a multiple of event_size
186 * plus an extra byte for the terminating '\0'.
189 name_len = roundup(event->name_len + 1, event_size);
190 inotify_event.len = name_len;
192 inotify_event.mask = inotify_mask_to_arg(event->mask);
193 inotify_event.cookie = event->sync_cookie;
195 /* send the main event */
196 if (copy_to_user(buf, &inotify_event, event_size))
202 * fsnotify only stores the pathname, so here we have to send the pathname
203 * and then pad that pathname out to a multiple of sizeof(inotify_event)
204 * with zeros. I get my zeros from the nul_inotify_event.
207 unsigned int len_to_zero = name_len - event->name_len;
208 /* copy the path name */
209 if (copy_to_user(buf, event->file_name, event->name_len))
211 buf += event->name_len;
213 /* fill userspace with 0's */
214 if (clear_user(buf, len_to_zero))
217 event_size += name_len;
223 static ssize_t inotify_read(struct file *file, char __user *buf,
224 size_t count, loff_t *pos)
226 struct fsnotify_group *group;
227 struct fsnotify_event *kevent;
233 group = file->private_data;
236 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
238 mutex_lock(&group->notification_mutex);
239 kevent = get_one_event(group, count);
240 mutex_unlock(&group->notification_mutex);
243 ret = PTR_ERR(kevent);
246 ret = copy_event_to_user(group, kevent, buf);
247 fsnotify_put_event(kevent);
256 if (file->f_flags & O_NONBLOCK)
259 if (signal_pending(current))
268 finish_wait(&group->notification_waitq, &wait);
269 if (start != buf && ret != -EFAULT)
274 static int inotify_fasync(int fd, struct file *file, int on)
276 struct fsnotify_group *group = file->private_data;
278 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
281 static int inotify_release(struct inode *ignored, struct file *file)
283 struct fsnotify_group *group = file->private_data;
284 struct user_struct *user = group->inotify_data.user;
286 fsnotify_clear_marks_by_group(group);
288 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
289 fsnotify_put_group(group);
291 atomic_dec(&user->inotify_devs);
296 static long inotify_ioctl(struct file *file, unsigned int cmd,
299 struct fsnotify_group *group;
300 struct fsnotify_event_holder *holder;
301 struct fsnotify_event *event;
306 group = file->private_data;
307 p = (void __user *) arg;
311 mutex_lock(&group->notification_mutex);
312 list_for_each_entry(holder, &group->notification_list, event_list) {
313 event = holder->event;
314 send_len += sizeof(struct inotify_event);
316 send_len += roundup(event->name_len + 1,
317 sizeof(struct inotify_event));
319 mutex_unlock(&group->notification_mutex);
320 ret = put_user(send_len, (int __user *) p);
327 static const struct file_operations inotify_fops = {
328 .poll = inotify_poll,
329 .read = inotify_read,
330 .fasync = inotify_fasync,
331 .release = inotify_release,
332 .unlocked_ioctl = inotify_ioctl,
333 .compat_ioctl = inotify_ioctl,
338 * find_inode - resolve a user-given path to a specific inode
340 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
344 error = user_path_at(AT_FDCWD, dirname, flags, path);
347 /* you can only watch an inode if you have read permissions on it */
348 error = inode_permission(path->dentry->d_inode, MAY_READ);
354 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
356 struct inotify_inode_mark *i_mark)
361 if (unlikely(!idr_pre_get(idr, GFP_KERNEL)))
365 ret = idr_get_new_above(idr, i_mark, *last_wd + 1,
367 /* we added the mark to the idr, take a reference */
369 *last_wd = i_mark->wd;
370 fsnotify_get_mark(&i_mark->fsn_mark);
372 spin_unlock(idr_lock);
373 } while (ret == -EAGAIN);
378 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
381 struct idr *idr = &group->inotify_data.idr;
382 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
383 struct inotify_inode_mark *i_mark;
385 assert_spin_locked(idr_lock);
387 i_mark = idr_find(idr, wd);
389 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
391 fsnotify_get_mark(fsn_mark);
392 /* One ref for being in the idr, one ref we just took */
393 BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
399 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
402 struct inotify_inode_mark *i_mark;
403 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
406 i_mark = inotify_idr_find_locked(group, wd);
407 spin_unlock(idr_lock);
412 static void do_inotify_remove_from_idr(struct fsnotify_group *group,
413 struct inotify_inode_mark *i_mark)
415 struct idr *idr = &group->inotify_data.idr;
416 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
419 assert_spin_locked(idr_lock);
423 /* removed from the idr, drop that ref */
424 fsnotify_put_mark(&i_mark->fsn_mark);
428 * Remove the mark from the idr (if present) and drop the reference
429 * on the mark because it was in the idr.
431 static void inotify_remove_from_idr(struct fsnotify_group *group,
432 struct inotify_inode_mark *i_mark)
434 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
435 struct inotify_inode_mark *found_i_mark = NULL;
442 * does this i_mark think it is in the idr? we shouldn't get called
446 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
447 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
448 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
452 /* Lets look in the idr to see if we find it */
453 found_i_mark = inotify_idr_find_locked(group, wd);
454 if (unlikely(!found_i_mark)) {
455 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
456 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
457 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
462 * We found an mark in the idr at the right wd, but it's
463 * not the mark we were told to remove. eparis seriously
464 * fucked up somewhere.
466 if (unlikely(found_i_mark != i_mark)) {
467 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
468 "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
469 "found_i_mark->group=%p found_i_mark->inode=%p\n",
470 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
471 i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd,
472 found_i_mark->fsn_mark.group,
473 found_i_mark->fsn_mark.i.inode);
478 * One ref for being in the idr
479 * one ref held by the caller trying to kill us
480 * one ref grabbed by inotify_idr_find
482 if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
483 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
484 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
485 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
486 /* we can't really recover with bad ref cnting.. */
490 do_inotify_remove_from_idr(group, i_mark);
492 /* match the ref taken by inotify_idr_find_locked() */
494 fsnotify_put_mark(&found_i_mark->fsn_mark);
496 spin_unlock(idr_lock);
500 * Send IN_IGNORED for this wd, remove this wd from the idr.
502 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
503 struct fsnotify_group *group)
505 struct inotify_inode_mark *i_mark;
506 struct fsnotify_event *ignored_event;
507 struct inotify_event_private_data *event_priv;
508 struct fsnotify_event_private_data *fsn_event_priv;
511 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
512 FSNOTIFY_EVENT_NONE, NULL, 0,
517 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
519 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
520 if (unlikely(!event_priv))
521 goto skip_send_ignore;
523 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
525 fsn_event_priv->group = group;
526 event_priv->wd = i_mark->wd;
528 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL);
530 inotify_free_event_priv(fsn_event_priv);
534 /* matches the reference taken when the event was created */
535 fsnotify_put_event(ignored_event);
537 /* remove this mark from the idr */
538 inotify_remove_from_idr(group, i_mark);
540 atomic_dec(&group->inotify_data.user->inotify_watches);
543 /* ding dong the mark is dead */
544 static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
546 struct inotify_inode_mark *i_mark;
548 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
550 kmem_cache_free(inotify_inode_mark_cachep, i_mark);
553 static int inotify_update_existing_watch(struct fsnotify_group *group,
557 struct fsnotify_mark *fsn_mark;
558 struct inotify_inode_mark *i_mark;
559 __u32 old_mask, new_mask;
561 int add = (arg & IN_MASK_ADD);
564 /* don't allow invalid bits: we don't want flags set */
565 mask = inotify_arg_to_mask(arg);
569 fsn_mark = fsnotify_find_mark(group, inode);
573 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
575 spin_lock(&fsn_mark->lock);
577 old_mask = fsn_mark->mask;
579 fsn_mark->mask |= mask;
580 new_mask = fsn_mark->mask;
582 fsn_mark->mask = mask;
583 new_mask = fsn_mark->mask;
586 spin_unlock(&fsn_mark->lock);
588 if (old_mask != new_mask) {
589 /* more bits in old than in new? */
590 int dropped = (old_mask & ~new_mask);
591 /* more bits in this fsn_mark than the inode's mask? */
592 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
593 /* more bits in this fsn_mark than the group? */
594 int do_group = (new_mask & ~group->mask);
596 /* update the inode with this new fsn_mark */
597 if (dropped || do_inode)
598 fsnotify_recalc_inode_mask(inode);
600 /* update the group mask with the new mask */
601 if (dropped || do_group)
602 fsnotify_recalc_group_mask(group);
608 /* match the get from fsnotify_find_mark() */
609 fsnotify_put_mark(fsn_mark);
614 static int inotify_new_watch(struct fsnotify_group *group,
618 struct inotify_inode_mark *tmp_i_mark;
621 struct idr *idr = &group->inotify_data.idr;
622 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
624 /* don't allow invalid bits: we don't want flags set */
625 mask = inotify_arg_to_mask(arg);
629 tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
630 if (unlikely(!tmp_i_mark))
633 fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark);
634 tmp_i_mark->fsn_mark.mask = mask;
638 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
641 ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd,
646 /* we are on the idr, now get on the inode */
647 ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, 0);
649 /* we failed to get on the inode, get off the idr */
650 inotify_remove_from_idr(group, tmp_i_mark);
654 /* increment the number of watches the user has */
655 atomic_inc(&group->inotify_data.user->inotify_watches);
657 /* return the watch descriptor for this new mark */
658 ret = tmp_i_mark->wd;
660 /* if this mark added a new event update the group mask */
661 if (mask & ~group->mask)
662 fsnotify_recalc_group_mask(group);
665 /* match the ref from fsnotify_init_mark() */
666 fsnotify_put_mark(&tmp_i_mark->fsn_mark);
671 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
676 /* try to update and existing watch with the new arg */
677 ret = inotify_update_existing_watch(group, inode, arg);
678 /* no mark present, try to add a new one */
680 ret = inotify_new_watch(group, inode, arg);
682 * inotify_new_watch could race with another thread which did an
683 * inotify_new_watch between the update_existing and the add watch
684 * here, go back and try to update an existing mark again.
692 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
694 struct fsnotify_group *group;
696 group = fsnotify_alloc_group(&inotify_fsnotify_ops);
700 group->max_events = max_events;
702 spin_lock_init(&group->inotify_data.idr_lock);
703 idr_init(&group->inotify_data.idr);
704 group->inotify_data.last_wd = 0;
705 group->inotify_data.user = user;
706 group->inotify_data.fa = NULL;
712 /* inotify syscalls */
713 SYSCALL_DEFINE1(inotify_init1, int, flags)
715 struct fsnotify_group *group;
716 struct user_struct *user;
719 /* Check the IN_* constants for consistency. */
720 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
721 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
723 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
726 user = get_current_user();
727 if (unlikely(atomic_read(&user->inotify_devs) >=
728 inotify_max_user_instances)) {
733 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
734 group = inotify_new_group(user, inotify_max_queued_events);
736 ret = PTR_ERR(group);
740 atomic_inc(&user->inotify_devs);
742 ret = anon_inode_getfd("inotify", &inotify_fops, group,
747 atomic_dec(&user->inotify_devs);
753 SYSCALL_DEFINE0(inotify_init)
755 return sys_inotify_init1(0);
758 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
761 struct fsnotify_group *group;
765 int ret, fput_needed;
768 filp = fget_light(fd, &fput_needed);
772 /* verify that this is indeed an inotify instance */
773 if (unlikely(filp->f_op != &inotify_fops)) {
778 if (!(mask & IN_DONT_FOLLOW))
779 flags |= LOOKUP_FOLLOW;
780 if (mask & IN_ONLYDIR)
781 flags |= LOOKUP_DIRECTORY;
783 ret = inotify_find_inode(pathname, &path, flags);
787 /* inode held in place by reference to path; group by fget on fd */
788 inode = path.dentry->d_inode;
789 group = filp->private_data;
791 /* create/update an inode mark */
792 ret = inotify_update_watch(group, inode, mask);
795 fput_light(filp, fput_needed);
799 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
801 struct fsnotify_group *group;
802 struct inotify_inode_mark *i_mark;
804 int ret = 0, fput_needed;
806 filp = fget_light(fd, &fput_needed);
810 /* verify that this is indeed an inotify instance */
812 if (unlikely(filp->f_op != &inotify_fops))
815 group = filp->private_data;
818 i_mark = inotify_idr_find(group, wd);
819 if (unlikely(!i_mark))
824 fsnotify_destroy_mark(&i_mark->fsn_mark);
826 /* match ref taken by inotify_idr_find */
827 fsnotify_put_mark(&i_mark->fsn_mark);
830 fput_light(filp, fput_needed);
835 * inotify_user_setup - Our initialization function. Note that we cannnot return
836 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
837 * must result in panic().
839 static int __init inotify_user_setup(void)
841 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
842 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
844 inotify_max_queued_events = 16384;
845 inotify_max_user_instances = 128;
846 inotify_max_user_watches = 8192;
850 module_init(inotify_user_setup);