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 | IN_EXCL_UNLINK));
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);
144 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
147 event_size += roundup(event->name_len + 1, event_size);
149 if (event_size > count)
150 return ERR_PTR(-EINVAL);
152 /* held the notification_mutex the whole time, so this is the
153 * same event we peeked above */
154 fsnotify_remove_notify_event(group);
160 * Copy an event to user space, returning how much we copied.
162 * We already checked that the event size is smaller than the
163 * buffer we had in "get_one_event()" above.
165 static ssize_t copy_event_to_user(struct fsnotify_group *group,
166 struct fsnotify_event *event,
169 struct inotify_event inotify_event;
170 struct fsnotify_event_private_data *fsn_priv;
171 struct inotify_event_private_data *priv;
172 size_t event_size = sizeof(struct inotify_event);
175 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
177 /* we get the inotify watch descriptor from the event private data */
178 spin_lock(&event->lock);
179 fsn_priv = fsnotify_remove_priv_from_event(group, event);
180 spin_unlock(&event->lock);
183 inotify_event.wd = -1;
185 priv = container_of(fsn_priv, struct inotify_event_private_data,
186 fsnotify_event_priv_data);
187 inotify_event.wd = priv->wd;
188 inotify_free_event_priv(fsn_priv);
192 * round up event->name_len so it is a multiple of event_size
193 * plus an extra byte for the terminating '\0'.
196 name_len = roundup(event->name_len + 1, event_size);
197 inotify_event.len = name_len;
199 inotify_event.mask = inotify_mask_to_arg(event->mask);
200 inotify_event.cookie = event->sync_cookie;
202 /* send the main event */
203 if (copy_to_user(buf, &inotify_event, event_size))
209 * fsnotify only stores the pathname, so here we have to send the pathname
210 * and then pad that pathname out to a multiple of sizeof(inotify_event)
211 * with zeros. I get my zeros from the nul_inotify_event.
214 unsigned int len_to_zero = name_len - event->name_len;
215 /* copy the path name */
216 if (copy_to_user(buf, event->file_name, event->name_len))
218 buf += event->name_len;
220 /* fill userspace with 0's */
221 if (clear_user(buf, len_to_zero))
224 event_size += name_len;
230 static ssize_t inotify_read(struct file *file, char __user *buf,
231 size_t count, loff_t *pos)
233 struct fsnotify_group *group;
234 struct fsnotify_event *kevent;
240 group = file->private_data;
243 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
245 mutex_lock(&group->notification_mutex);
246 kevent = get_one_event(group, count);
247 mutex_unlock(&group->notification_mutex);
249 pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
252 ret = PTR_ERR(kevent);
255 ret = copy_event_to_user(group, kevent, buf);
256 fsnotify_put_event(kevent);
265 if (file->f_flags & O_NONBLOCK)
268 if (signal_pending(current))
277 finish_wait(&group->notification_waitq, &wait);
278 if (start != buf && ret != -EFAULT)
283 static int inotify_fasync(int fd, struct file *file, int on)
285 struct fsnotify_group *group = file->private_data;
287 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
290 static int inotify_release(struct inode *ignored, struct file *file)
292 struct fsnotify_group *group = file->private_data;
294 pr_debug("%s: group=%p\n", __func__, group);
296 fsnotify_clear_marks_by_group(group);
298 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
299 fsnotify_put_group(group);
304 static long inotify_ioctl(struct file *file, unsigned int cmd,
307 struct fsnotify_group *group;
308 struct fsnotify_event_holder *holder;
309 struct fsnotify_event *event;
314 group = file->private_data;
315 p = (void __user *) arg;
317 pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
321 mutex_lock(&group->notification_mutex);
322 list_for_each_entry(holder, &group->notification_list, event_list) {
323 event = holder->event;
324 send_len += sizeof(struct inotify_event);
326 send_len += roundup(event->name_len + 1,
327 sizeof(struct inotify_event));
329 mutex_unlock(&group->notification_mutex);
330 ret = put_user(send_len, (int __user *) p);
337 static const struct file_operations inotify_fops = {
338 .poll = inotify_poll,
339 .read = inotify_read,
340 .fasync = inotify_fasync,
341 .release = inotify_release,
342 .unlocked_ioctl = inotify_ioctl,
343 .compat_ioctl = inotify_ioctl,
344 .llseek = noop_llseek,
349 * find_inode - resolve a user-given path to a specific inode
351 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
355 error = user_path_at(AT_FDCWD, dirname, flags, path);
358 /* you can only watch an inode if you have read permissions on it */
359 error = inode_permission(path->dentry->d_inode, MAY_READ);
365 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
367 struct inotify_inode_mark *i_mark)
372 if (unlikely(!idr_pre_get(idr, GFP_KERNEL)))
376 ret = idr_get_new_above(idr, i_mark, *last_wd + 1,
378 /* we added the mark to the idr, take a reference */
380 *last_wd = i_mark->wd;
381 fsnotify_get_mark(&i_mark->fsn_mark);
383 spin_unlock(idr_lock);
384 } while (ret == -EAGAIN);
389 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
392 struct idr *idr = &group->inotify_data.idr;
393 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
394 struct inotify_inode_mark *i_mark;
396 assert_spin_locked(idr_lock);
398 i_mark = idr_find(idr, wd);
400 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
402 fsnotify_get_mark(fsn_mark);
403 /* One ref for being in the idr, one ref we just took */
404 BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
410 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
413 struct inotify_inode_mark *i_mark;
414 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
417 i_mark = inotify_idr_find_locked(group, wd);
418 spin_unlock(idr_lock);
423 static void do_inotify_remove_from_idr(struct fsnotify_group *group,
424 struct inotify_inode_mark *i_mark)
426 struct idr *idr = &group->inotify_data.idr;
427 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
430 assert_spin_locked(idr_lock);
434 /* removed from the idr, drop that ref */
435 fsnotify_put_mark(&i_mark->fsn_mark);
439 * Remove the mark from the idr (if present) and drop the reference
440 * on the mark because it was in the idr.
442 static void inotify_remove_from_idr(struct fsnotify_group *group,
443 struct inotify_inode_mark *i_mark)
445 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
446 struct inotify_inode_mark *found_i_mark = NULL;
453 * does this i_mark think it is in the idr? we shouldn't get called
457 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
458 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
459 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
463 /* Lets look in the idr to see if we find it */
464 found_i_mark = inotify_idr_find_locked(group, wd);
465 if (unlikely(!found_i_mark)) {
466 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
467 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
468 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
473 * We found an mark in the idr at the right wd, but it's
474 * not the mark we were told to remove. eparis seriously
475 * fucked up somewhere.
477 if (unlikely(found_i_mark != i_mark)) {
478 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
479 "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
480 "found_i_mark->group=%p found_i_mark->inode=%p\n",
481 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
482 i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd,
483 found_i_mark->fsn_mark.group,
484 found_i_mark->fsn_mark.i.inode);
489 * One ref for being in the idr
490 * one ref held by the caller trying to kill us
491 * one ref grabbed by inotify_idr_find
493 if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
494 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
495 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
496 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
497 /* we can't really recover with bad ref cnting.. */
501 do_inotify_remove_from_idr(group, i_mark);
503 /* match the ref taken by inotify_idr_find_locked() */
505 fsnotify_put_mark(&found_i_mark->fsn_mark);
507 spin_unlock(idr_lock);
511 * Send IN_IGNORED for this wd, remove this wd from the idr.
513 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
514 struct fsnotify_group *group)
516 struct inotify_inode_mark *i_mark;
517 struct fsnotify_event *ignored_event, *notify_event;
518 struct inotify_event_private_data *event_priv;
519 struct fsnotify_event_private_data *fsn_event_priv;
522 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
523 FSNOTIFY_EVENT_NONE, NULL, 0,
528 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
530 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
531 if (unlikely(!event_priv))
532 goto skip_send_ignore;
534 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
536 fsn_event_priv->group = group;
537 event_priv->wd = i_mark->wd;
539 notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL);
541 if (IS_ERR(notify_event))
542 ret = PTR_ERR(notify_event);
544 fsnotify_put_event(notify_event);
545 inotify_free_event_priv(fsn_event_priv);
550 /* matches the reference taken when the event was created */
551 fsnotify_put_event(ignored_event);
553 /* remove this mark from the idr */
554 inotify_remove_from_idr(group, i_mark);
556 atomic_dec(&group->inotify_data.user->inotify_watches);
559 /* ding dong the mark is dead */
560 static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
562 struct inotify_inode_mark *i_mark;
564 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
566 kmem_cache_free(inotify_inode_mark_cachep, i_mark);
569 static int inotify_update_existing_watch(struct fsnotify_group *group,
573 struct fsnotify_mark *fsn_mark;
574 struct inotify_inode_mark *i_mark;
575 __u32 old_mask, new_mask;
577 int add = (arg & IN_MASK_ADD);
580 /* don't allow invalid bits: we don't want flags set */
581 mask = inotify_arg_to_mask(arg);
582 if (unlikely(!(mask & IN_ALL_EVENTS)))
585 fsn_mark = fsnotify_find_inode_mark(group, inode);
589 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
591 spin_lock(&fsn_mark->lock);
593 old_mask = fsn_mark->mask;
595 fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask));
597 fsnotify_set_mark_mask_locked(fsn_mark, mask);
598 new_mask = fsn_mark->mask;
600 spin_unlock(&fsn_mark->lock);
602 if (old_mask != new_mask) {
603 /* more bits in old than in new? */
604 int dropped = (old_mask & ~new_mask);
605 /* more bits in this fsn_mark than the inode's mask? */
606 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
608 /* update the inode with this new fsn_mark */
609 if (dropped || do_inode)
610 fsnotify_recalc_inode_mask(inode);
617 /* match the get from fsnotify_find_mark() */
618 fsnotify_put_mark(fsn_mark);
623 static int inotify_new_watch(struct fsnotify_group *group,
627 struct inotify_inode_mark *tmp_i_mark;
630 struct idr *idr = &group->inotify_data.idr;
631 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
633 /* don't allow invalid bits: we don't want flags set */
634 mask = inotify_arg_to_mask(arg);
635 if (unlikely(!(mask & IN_ALL_EVENTS)))
638 tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
639 if (unlikely(!tmp_i_mark))
642 fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark);
643 tmp_i_mark->fsn_mark.mask = mask;
647 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
650 ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd,
655 /* we are on the idr, now get on the inode */
656 ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0);
658 /* we failed to get on the inode, get off the idr */
659 inotify_remove_from_idr(group, tmp_i_mark);
663 /* increment the number of watches the user has */
664 atomic_inc(&group->inotify_data.user->inotify_watches);
666 /* return the watch descriptor for this new mark */
667 ret = tmp_i_mark->wd;
670 /* match the ref from fsnotify_init_mark() */
671 fsnotify_put_mark(&tmp_i_mark->fsn_mark);
676 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
681 /* try to update and existing watch with the new arg */
682 ret = inotify_update_existing_watch(group, inode, arg);
683 /* no mark present, try to add a new one */
685 ret = inotify_new_watch(group, inode, arg);
687 * inotify_new_watch could race with another thread which did an
688 * inotify_new_watch between the update_existing and the add watch
689 * here, go back and try to update an existing mark again.
697 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
699 struct fsnotify_group *group;
701 group = fsnotify_alloc_group(&inotify_fsnotify_ops);
705 group->max_events = max_events;
707 spin_lock_init(&group->inotify_data.idr_lock);
708 idr_init(&group->inotify_data.idr);
709 group->inotify_data.last_wd = 0;
710 group->inotify_data.fa = NULL;
711 group->inotify_data.user = get_current_user();
713 if (atomic_inc_return(&group->inotify_data.user->inotify_devs) >
714 inotify_max_user_instances) {
715 fsnotify_put_group(group);
716 return ERR_PTR(-EMFILE);
723 /* inotify syscalls */
724 SYSCALL_DEFINE1(inotify_init1, int, flags)
726 struct fsnotify_group *group;
729 /* Check the IN_* constants for consistency. */
730 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
731 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
733 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
736 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
737 group = inotify_new_group(inotify_max_queued_events);
739 return PTR_ERR(group);
741 ret = anon_inode_getfd("inotify", &inotify_fops, group,
744 fsnotify_put_group(group);
749 SYSCALL_DEFINE0(inotify_init)
751 return sys_inotify_init1(0);
754 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
757 struct fsnotify_group *group;
761 int ret, fput_needed;
764 filp = fget_light(fd, &fput_needed);
768 /* verify that this is indeed an inotify instance */
769 if (unlikely(filp->f_op != &inotify_fops)) {
774 if (!(mask & IN_DONT_FOLLOW))
775 flags |= LOOKUP_FOLLOW;
776 if (mask & IN_ONLYDIR)
777 flags |= LOOKUP_DIRECTORY;
779 ret = inotify_find_inode(pathname, &path, flags);
783 /* inode held in place by reference to path; group by fget on fd */
784 inode = path.dentry->d_inode;
785 group = filp->private_data;
787 /* create/update an inode mark */
788 ret = inotify_update_watch(group, inode, mask);
791 fput_light(filp, fput_needed);
795 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
797 struct fsnotify_group *group;
798 struct inotify_inode_mark *i_mark;
800 int ret = 0, fput_needed;
802 filp = fget_light(fd, &fput_needed);
806 /* verify that this is indeed an inotify instance */
808 if (unlikely(filp->f_op != &inotify_fops))
811 group = filp->private_data;
814 i_mark = inotify_idr_find(group, wd);
815 if (unlikely(!i_mark))
820 fsnotify_destroy_mark(&i_mark->fsn_mark);
822 /* match ref taken by inotify_idr_find */
823 fsnotify_put_mark(&i_mark->fsn_mark);
826 fput_light(filp, fput_needed);
831 * inotify_user_setup - Our initialization function. Note that we cannot return
832 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
833 * must result in panic().
835 static int __init inotify_user_setup(void)
837 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
838 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
839 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
840 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
841 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
842 BUILD_BUG_ON(IN_OPEN != FS_OPEN);
843 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
844 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
845 BUILD_BUG_ON(IN_CREATE != FS_CREATE);
846 BUILD_BUG_ON(IN_DELETE != FS_DELETE);
847 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
848 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
849 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
850 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
851 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
852 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
853 BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
854 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
856 BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
858 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
859 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
861 inotify_max_queued_events = 16384;
862 inotify_max_user_instances = 128;
863 inotify_max_user_watches = 8192;
867 module_init(inotify_user_setup);