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>
43 #include "../fdinfo.h"
45 #include <asm/ioctls.h>
47 /* these are configurable via /proc/sys/fs/inotify/ */
48 static int inotify_max_user_instances __read_mostly;
49 static int inotify_max_queued_events __read_mostly;
50 static int inotify_max_user_watches __read_mostly;
52 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
53 struct kmem_cache *event_priv_cachep __read_mostly;
57 #include <linux/sysctl.h>
61 ctl_table inotify_table[] = {
63 .procname = "max_user_instances",
64 .data = &inotify_max_user_instances,
65 .maxlen = sizeof(int),
67 .proc_handler = proc_dointvec_minmax,
71 .procname = "max_user_watches",
72 .data = &inotify_max_user_watches,
73 .maxlen = sizeof(int),
75 .proc_handler = proc_dointvec_minmax,
79 .procname = "max_queued_events",
80 .data = &inotify_max_queued_events,
81 .maxlen = sizeof(int),
83 .proc_handler = proc_dointvec_minmax,
88 #endif /* CONFIG_SYSCTL */
90 static inline __u32 inotify_arg_to_mask(u32 arg)
95 * everything should accept their own ignored, cares about children,
96 * and should receive events when the inode is unmounted
98 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT);
100 /* mask off the flags used to open the fd */
101 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
106 static inline u32 inotify_mask_to_arg(__u32 mask)
108 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
112 /* intofiy userspace file descriptor functions */
113 static unsigned int inotify_poll(struct file *file, poll_table *wait)
115 struct fsnotify_group *group = file->private_data;
118 poll_wait(file, &group->notification_waitq, wait);
119 mutex_lock(&group->notification_mutex);
120 if (!fsnotify_notify_queue_is_empty(group))
121 ret = POLLIN | POLLRDNORM;
122 mutex_unlock(&group->notification_mutex);
128 * Get an inotify_kernel_event if one exists and is small
129 * enough to fit in "count". Return an error pointer if
132 * Called with the group->notification_mutex held.
134 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
137 size_t event_size = sizeof(struct inotify_event);
138 struct fsnotify_event *event;
140 if (fsnotify_notify_queue_is_empty(group))
143 event = fsnotify_peek_notify_event(group);
145 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
148 event_size += roundup(event->name_len + 1, event_size);
150 if (event_size > count)
151 return ERR_PTR(-EINVAL);
153 /* held the notification_mutex the whole time, so this is the
154 * same event we peeked above */
155 fsnotify_remove_notify_event(group);
161 * Copy an event to user space, returning how much we copied.
163 * We already checked that the event size is smaller than the
164 * buffer we had in "get_one_event()" above.
166 static ssize_t copy_event_to_user(struct fsnotify_group *group,
167 struct fsnotify_event *event,
170 struct inotify_event inotify_event;
171 struct fsnotify_event_private_data *fsn_priv;
172 struct inotify_event_private_data *priv;
173 size_t event_size = sizeof(struct inotify_event);
176 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
178 /* we get the inotify watch descriptor from the event private data */
179 spin_lock(&event->lock);
180 fsn_priv = fsnotify_remove_priv_from_event(group, event);
181 spin_unlock(&event->lock);
184 inotify_event.wd = -1;
186 priv = container_of(fsn_priv, struct inotify_event_private_data,
187 fsnotify_event_priv_data);
188 inotify_event.wd = priv->wd;
189 inotify_free_event_priv(fsn_priv);
193 * round up event->name_len so it is a multiple of event_size
194 * plus an extra byte for the terminating '\0'.
197 name_len = roundup(event->name_len + 1, event_size);
198 inotify_event.len = name_len;
200 inotify_event.mask = inotify_mask_to_arg(event->mask);
201 inotify_event.cookie = event->sync_cookie;
203 /* send the main event */
204 if (copy_to_user(buf, &inotify_event, event_size))
210 * fsnotify only stores the pathname, so here we have to send the pathname
211 * and then pad that pathname out to a multiple of sizeof(inotify_event)
212 * with zeros. I get my zeros from the nul_inotify_event.
215 unsigned int len_to_zero = name_len - event->name_len;
216 /* copy the path name */
217 if (copy_to_user(buf, event->file_name, event->name_len))
219 buf += event->name_len;
221 /* fill userspace with 0's */
222 if (clear_user(buf, len_to_zero))
225 event_size += name_len;
231 static ssize_t inotify_read(struct file *file, char __user *buf,
232 size_t count, loff_t *pos)
234 struct fsnotify_group *group;
235 struct fsnotify_event *kevent;
241 group = file->private_data;
244 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
246 mutex_lock(&group->notification_mutex);
247 kevent = get_one_event(group, count);
248 mutex_unlock(&group->notification_mutex);
250 pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
253 ret = PTR_ERR(kevent);
256 ret = copy_event_to_user(group, kevent, buf);
257 fsnotify_put_event(kevent);
266 if (file->f_flags & O_NONBLOCK)
269 if (signal_pending(current))
278 finish_wait(&group->notification_waitq, &wait);
279 if (start != buf && ret != -EFAULT)
284 static int inotify_release(struct inode *ignored, struct file *file)
286 struct fsnotify_group *group = file->private_data;
288 pr_debug("%s: group=%p\n", __func__, group);
290 if (file->f_flags & FASYNC)
291 fsnotify_fasync(-1, file, 0);
293 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
294 fsnotify_destroy_group(group);
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;
312 pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
316 mutex_lock(&group->notification_mutex);
317 list_for_each_entry(holder, &group->notification_list, event_list) {
318 event = holder->event;
319 send_len += sizeof(struct inotify_event);
321 send_len += roundup(event->name_len + 1,
322 sizeof(struct inotify_event));
324 mutex_unlock(&group->notification_mutex);
325 ret = put_user(send_len, (int __user *) p);
332 static const struct file_operations inotify_fops = {
333 .show_fdinfo = inotify_show_fdinfo,
334 .poll = inotify_poll,
335 .read = inotify_read,
336 .fasync = fsnotify_fasync,
337 .release = inotify_release,
338 .unlocked_ioctl = inotify_ioctl,
339 .compat_ioctl = inotify_ioctl,
340 .llseek = noop_llseek,
345 * find_inode - resolve a user-given path to a specific inode
347 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
351 error = user_path_at(AT_FDCWD, dirname, flags, path);
354 /* you can only watch an inode if you have read permissions on it */
355 error = inode_permission(path->dentry->d_inode, MAY_READ);
361 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
363 struct inotify_inode_mark *i_mark)
367 idr_preload(GFP_KERNEL);
370 ret = idr_alloc(idr, i_mark, *last_wd + 1, 0, GFP_NOWAIT);
372 /* we added the mark to the idr, take a reference */
374 *last_wd = i_mark->wd;
375 fsnotify_get_mark(&i_mark->fsn_mark);
378 spin_unlock(idr_lock);
380 return ret < 0 ? ret : 0;
383 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
386 struct idr *idr = &group->inotify_data.idr;
387 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
388 struct inotify_inode_mark *i_mark;
390 assert_spin_locked(idr_lock);
392 i_mark = idr_find(idr, wd);
394 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
396 fsnotify_get_mark(fsn_mark);
397 /* One ref for being in the idr, one ref we just took */
398 BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
404 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
407 struct inotify_inode_mark *i_mark;
408 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
411 i_mark = inotify_idr_find_locked(group, wd);
412 spin_unlock(idr_lock);
417 static void do_inotify_remove_from_idr(struct fsnotify_group *group,
418 struct inotify_inode_mark *i_mark)
420 struct idr *idr = &group->inotify_data.idr;
421 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
424 assert_spin_locked(idr_lock);
428 /* removed from the idr, drop that ref */
429 fsnotify_put_mark(&i_mark->fsn_mark);
433 * Remove the mark from the idr (if present) and drop the reference
434 * on the mark because it was in the idr.
436 static void inotify_remove_from_idr(struct fsnotify_group *group,
437 struct inotify_inode_mark *i_mark)
439 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
440 struct inotify_inode_mark *found_i_mark = NULL;
447 * does this i_mark think it is in the idr? we shouldn't get called
451 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
452 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
453 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
457 /* Lets look in the idr to see if we find it */
458 found_i_mark = inotify_idr_find_locked(group, wd);
459 if (unlikely(!found_i_mark)) {
460 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
461 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
462 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
467 * We found an mark in the idr at the right wd, but it's
468 * not the mark we were told to remove. eparis seriously
469 * fucked up somewhere.
471 if (unlikely(found_i_mark != i_mark)) {
472 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
473 "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
474 "found_i_mark->group=%p found_i_mark->inode=%p\n",
475 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
476 i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd,
477 found_i_mark->fsn_mark.group,
478 found_i_mark->fsn_mark.i.inode);
483 * One ref for being in the idr
484 * one ref held by the caller trying to kill us
485 * one ref grabbed by inotify_idr_find
487 if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
488 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
489 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
490 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
491 /* we can't really recover with bad ref cnting.. */
495 do_inotify_remove_from_idr(group, i_mark);
497 /* match the ref taken by inotify_idr_find_locked() */
499 fsnotify_put_mark(&found_i_mark->fsn_mark);
501 spin_unlock(idr_lock);
505 * Send IN_IGNORED for this wd, remove this wd from the idr.
507 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
508 struct fsnotify_group *group)
510 struct inotify_inode_mark *i_mark;
511 struct fsnotify_event *ignored_event, *notify_event;
512 struct inotify_event_private_data *event_priv;
513 struct fsnotify_event_private_data *fsn_event_priv;
516 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
518 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
519 FSNOTIFY_EVENT_NONE, NULL, 0,
522 goto skip_send_ignore;
524 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
525 if (unlikely(!event_priv))
526 goto skip_send_ignore;
528 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
530 fsnotify_get_group(group);
531 fsn_event_priv->group = group;
532 event_priv->wd = i_mark->wd;
534 notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL);
536 if (IS_ERR(notify_event))
537 ret = PTR_ERR(notify_event);
539 fsnotify_put_event(notify_event);
540 inotify_free_event_priv(fsn_event_priv);
544 /* matches the reference taken when the event was created */
546 fsnotify_put_event(ignored_event);
548 /* remove this mark from the idr */
549 inotify_remove_from_idr(group, i_mark);
551 atomic_dec(&group->inotify_data.user->inotify_watches);
554 /* ding dong the mark is dead */
555 static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
557 struct inotify_inode_mark *i_mark;
559 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
561 kmem_cache_free(inotify_inode_mark_cachep, i_mark);
564 static int inotify_update_existing_watch(struct fsnotify_group *group,
568 struct fsnotify_mark *fsn_mark;
569 struct inotify_inode_mark *i_mark;
570 __u32 old_mask, new_mask;
572 int add = (arg & IN_MASK_ADD);
575 /* don't allow invalid bits: we don't want flags set */
576 mask = inotify_arg_to_mask(arg);
578 fsn_mark = fsnotify_find_inode_mark(group, inode);
582 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
584 spin_lock(&fsn_mark->lock);
586 old_mask = fsn_mark->mask;
588 fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask));
590 fsnotify_set_mark_mask_locked(fsn_mark, mask);
591 new_mask = fsn_mark->mask;
593 spin_unlock(&fsn_mark->lock);
595 if (old_mask != new_mask) {
596 /* more bits in old than in new? */
597 int dropped = (old_mask & ~new_mask);
598 /* more bits in this fsn_mark than the inode's mask? */
599 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
601 /* update the inode with this new fsn_mark */
602 if (dropped || do_inode)
603 fsnotify_recalc_inode_mask(inode);
610 /* match the get from fsnotify_find_mark() */
611 fsnotify_put_mark(fsn_mark);
616 static int inotify_new_watch(struct fsnotify_group *group,
620 struct inotify_inode_mark *tmp_i_mark;
623 struct idr *idr = &group->inotify_data.idr;
624 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
626 /* don't allow invalid bits: we don't want flags set */
627 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, NULL, 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;
661 /* match the ref from fsnotify_init_mark() */
662 fsnotify_put_mark(&tmp_i_mark->fsn_mark);
667 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
672 /* try to update and existing watch with the new arg */
673 ret = inotify_update_existing_watch(group, inode, arg);
674 /* no mark present, try to add a new one */
676 ret = inotify_new_watch(group, inode, arg);
678 * inotify_new_watch could race with another thread which did an
679 * inotify_new_watch between the update_existing and the add watch
680 * here, go back and try to update an existing mark again.
688 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
690 struct fsnotify_group *group;
692 group = fsnotify_alloc_group(&inotify_fsnotify_ops);
696 group->max_events = max_events;
698 spin_lock_init(&group->inotify_data.idr_lock);
699 idr_init(&group->inotify_data.idr);
700 group->inotify_data.last_wd = 0;
701 group->inotify_data.user = get_current_user();
703 if (atomic_inc_return(&group->inotify_data.user->inotify_devs) >
704 inotify_max_user_instances) {
705 fsnotify_destroy_group(group);
706 return ERR_PTR(-EMFILE);
713 /* inotify syscalls */
714 SYSCALL_DEFINE1(inotify_init1, int, flags)
716 struct fsnotify_group *group;
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 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
727 group = inotify_new_group(inotify_max_queued_events);
729 return PTR_ERR(group);
731 ret = anon_inode_getfd("inotify", &inotify_fops, group,
734 fsnotify_destroy_group(group);
739 SYSCALL_DEFINE0(inotify_init)
741 return sys_inotify_init1(0);
744 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
747 struct fsnotify_group *group;
755 if (unlikely(!f.file))
758 /* verify that this is indeed an inotify instance */
759 if (unlikely(f.file->f_op != &inotify_fops)) {
764 if (!(mask & IN_DONT_FOLLOW))
765 flags |= LOOKUP_FOLLOW;
766 if (mask & IN_ONLYDIR)
767 flags |= LOOKUP_DIRECTORY;
769 ret = inotify_find_inode(pathname, &path, flags);
773 /* inode held in place by reference to path; group by fget on fd */
774 inode = path.dentry->d_inode;
775 group = f.file->private_data;
777 /* create/update an inode mark */
778 ret = inotify_update_watch(group, inode, mask);
785 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
787 struct fsnotify_group *group;
788 struct inotify_inode_mark *i_mark;
793 if (unlikely(!f.file))
796 /* verify that this is indeed an inotify instance */
798 if (unlikely(f.file->f_op != &inotify_fops))
801 group = f.file->private_data;
804 i_mark = inotify_idr_find(group, wd);
805 if (unlikely(!i_mark))
810 fsnotify_destroy_mark(&i_mark->fsn_mark, group);
812 /* match ref taken by inotify_idr_find */
813 fsnotify_put_mark(&i_mark->fsn_mark);
821 * inotify_user_setup - Our initialization function. Note that we cannot return
822 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
823 * must result in panic().
825 static int __init inotify_user_setup(void)
827 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
828 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
829 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
830 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
831 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
832 BUILD_BUG_ON(IN_OPEN != FS_OPEN);
833 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
834 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
835 BUILD_BUG_ON(IN_CREATE != FS_CREATE);
836 BUILD_BUG_ON(IN_DELETE != FS_DELETE);
837 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
838 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
839 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
840 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
841 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
842 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
843 BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
844 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
846 BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
848 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
849 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
851 inotify_max_queued_events = 16384;
852 inotify_max_user_instances = 128;
853 inotify_max_user_watches = 8192;
857 module_init(inotify_user_setup);