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 .procname = "max_user_instances",
73 .data = &inotify_max_user_instances,
74 .maxlen = sizeof(int),
76 .proc_handler = proc_dointvec_minmax,
80 .procname = "max_user_watches",
81 .data = &inotify_max_user_watches,
82 .maxlen = sizeof(int),
84 .proc_handler = proc_dointvec_minmax,
88 .procname = "max_queued_events",
89 .data = &inotify_max_queued_events,
90 .maxlen = sizeof(int),
92 .proc_handler = proc_dointvec_minmax,
97 #endif /* CONFIG_SYSCTL */
99 static inline __u32 inotify_arg_to_mask(u32 arg)
103 /* everything should accept their own ignored and cares about children */
104 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
106 /* mask off the flags used to open the fd */
107 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
112 static inline u32 inotify_mask_to_arg(__u32 mask)
114 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
118 /* intofiy userspace file descriptor functions */
119 static unsigned int inotify_poll(struct file *file, poll_table *wait)
121 struct fsnotify_group *group = file->private_data;
124 poll_wait(file, &group->notification_waitq, wait);
125 mutex_lock(&group->notification_mutex);
126 if (!fsnotify_notify_queue_is_empty(group))
127 ret = POLLIN | POLLRDNORM;
128 mutex_unlock(&group->notification_mutex);
134 * Get an inotify_kernel_event if one exists and is small
135 * enough to fit in "count". Return an error pointer if
138 * Called with the group->notification_mutex held.
140 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
143 size_t event_size = sizeof(struct inotify_event);
144 struct fsnotify_event *event;
146 if (fsnotify_notify_queue_is_empty(group))
149 event = fsnotify_peek_notify_event(group);
152 event_size += roundup(event->name_len + 1, event_size);
154 if (event_size > count)
155 return ERR_PTR(-EINVAL);
157 /* held the notification_mutex the whole time, so this is the
158 * same event we peeked above */
159 fsnotify_remove_notify_event(group);
165 * Copy an event to user space, returning how much we copied.
167 * We already checked that the event size is smaller than the
168 * buffer we had in "get_one_event()" above.
170 static ssize_t copy_event_to_user(struct fsnotify_group *group,
171 struct fsnotify_event *event,
174 struct inotify_event inotify_event;
175 struct fsnotify_event_private_data *fsn_priv;
176 struct inotify_event_private_data *priv;
177 size_t event_size = sizeof(struct inotify_event);
180 /* we get the inotify watch descriptor from the event private data */
181 spin_lock(&event->lock);
182 fsn_priv = fsnotify_remove_priv_from_event(group, event);
183 spin_unlock(&event->lock);
186 inotify_event.wd = -1;
188 priv = container_of(fsn_priv, struct inotify_event_private_data,
189 fsnotify_event_priv_data);
190 inotify_event.wd = priv->wd;
191 inotify_free_event_priv(fsn_priv);
195 * round up event->name_len so it is a multiple of event_size
196 * plus an extra byte for the terminating '\0'.
199 name_len = roundup(event->name_len + 1, event_size);
200 inotify_event.len = name_len;
202 inotify_event.mask = inotify_mask_to_arg(event->mask);
203 inotify_event.cookie = event->sync_cookie;
205 /* send the main event */
206 if (copy_to_user(buf, &inotify_event, event_size))
212 * fsnotify only stores the pathname, so here we have to send the pathname
213 * and then pad that pathname out to a multiple of sizeof(inotify_event)
214 * with zeros. I get my zeros from the nul_inotify_event.
217 unsigned int len_to_zero = name_len - event->name_len;
218 /* copy the path name */
219 if (copy_to_user(buf, event->file_name, event->name_len))
221 buf += event->name_len;
223 /* fill userspace with 0's */
224 if (clear_user(buf, len_to_zero))
227 event_size += name_len;
233 static ssize_t inotify_read(struct file *file, char __user *buf,
234 size_t count, loff_t *pos)
236 struct fsnotify_group *group;
237 struct fsnotify_event *kevent;
243 group = file->private_data;
246 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
248 mutex_lock(&group->notification_mutex);
249 kevent = get_one_event(group, count);
250 mutex_unlock(&group->notification_mutex);
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_fasync(int fd, struct file *file, int on)
286 struct fsnotify_group *group = file->private_data;
288 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
291 static int inotify_release(struct inode *ignored, struct file *file)
293 struct fsnotify_group *group = file->private_data;
294 struct user_struct *user = group->inotify_data.user;
296 fsnotify_clear_marks_by_group(group);
298 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
299 fsnotify_put_group(group);
301 atomic_dec(&user->inotify_devs);
306 static long inotify_ioctl(struct file *file, unsigned int cmd,
309 struct fsnotify_group *group;
310 struct fsnotify_event_holder *holder;
311 struct fsnotify_event *event;
316 group = file->private_data;
317 p = (void __user *) arg;
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,
348 * find_inode - resolve a user-given path to a specific inode
350 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
354 error = user_path_at(AT_FDCWD, dirname, flags, path);
357 /* you can only watch an inode if you have read permissions on it */
358 error = inode_permission(path->dentry->d_inode, MAY_READ);
365 * Remove the mark from the idr (if present) and drop the reference
366 * on the mark because it was in the idr.
368 static void inotify_remove_from_idr(struct fsnotify_group *group,
369 struct inotify_inode_mark_entry *ientry)
372 struct fsnotify_mark_entry *entry;
373 struct inotify_inode_mark_entry *found_ientry;
376 spin_lock(&group->inotify_data.idr_lock);
377 idr = &group->inotify_data.idr;
383 entry = idr_find(&group->inotify_data.idr, wd);
384 if (unlikely(!entry))
387 found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
388 if (unlikely(found_ientry != ientry)) {
389 /* We found an entry in the idr with the right wd, but it's
390 * not the entry we were told to remove. eparis seriously
391 * fucked up somewhere. */
397 /* One ref for being in the idr, one ref held by the caller */
398 BUG_ON(atomic_read(&entry->refcnt) < 2);
403 /* removed from the idr, drop that ref */
404 fsnotify_put_mark(entry);
406 spin_unlock(&group->inotify_data.idr_lock);
410 * Send IN_IGNORED for this wd, remove this wd from the idr.
412 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
413 struct fsnotify_group *group)
415 struct inotify_inode_mark_entry *ientry;
416 struct fsnotify_event *ignored_event;
417 struct inotify_event_private_data *event_priv;
418 struct fsnotify_event_private_data *fsn_event_priv;
421 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
422 FSNOTIFY_EVENT_NONE, NULL, 0,
427 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
429 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
430 if (unlikely(!event_priv))
431 goto skip_send_ignore;
433 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
435 fsn_event_priv->group = group;
436 event_priv->wd = ientry->wd;
438 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
440 inotify_free_event_priv(fsn_event_priv);
444 /* matches the reference taken when the event was created */
445 fsnotify_put_event(ignored_event);
447 /* remove this entry from the idr */
448 inotify_remove_from_idr(group, ientry);
450 atomic_dec(&group->inotify_data.user->inotify_watches);
453 /* ding dong the mark is dead */
454 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
456 struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
458 kmem_cache_free(inotify_inode_mark_cachep, ientry);
461 static int inotify_update_existing_watch(struct fsnotify_group *group,
465 struct fsnotify_mark_entry *entry;
466 struct inotify_inode_mark_entry *ientry;
467 __u32 old_mask, new_mask;
469 int add = (arg & IN_MASK_ADD);
472 /* don't allow invalid bits: we don't want flags set */
473 mask = inotify_arg_to_mask(arg);
477 spin_lock(&inode->i_lock);
478 entry = fsnotify_find_mark_entry(group, inode);
479 spin_unlock(&inode->i_lock);
483 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
485 spin_lock(&entry->lock);
487 old_mask = entry->mask;
490 new_mask = entry->mask;
493 new_mask = entry->mask;
496 spin_unlock(&entry->lock);
498 if (old_mask != new_mask) {
499 /* more bits in old than in new? */
500 int dropped = (old_mask & ~new_mask);
501 /* more bits in this entry than the inode's mask? */
502 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
503 /* more bits in this entry than the group? */
504 int do_group = (new_mask & ~group->mask);
506 /* update the inode with this new entry */
507 if (dropped || do_inode)
508 fsnotify_recalc_inode_mask(inode);
510 /* update the group mask with the new mask */
511 if (dropped || do_group)
512 fsnotify_recalc_group_mask(group);
518 /* match the get from fsnotify_find_mark_entry() */
519 fsnotify_put_mark(entry);
524 static int inotify_new_watch(struct fsnotify_group *group,
528 struct inotify_inode_mark_entry *tmp_ientry;
532 /* don't allow invalid bits: we don't want flags set */
533 mask = inotify_arg_to_mask(arg);
537 tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
538 if (unlikely(!tmp_ientry))
541 fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
542 tmp_ientry->fsn_entry.mask = mask;
546 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
550 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
553 spin_lock(&group->inotify_data.idr_lock);
554 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
555 group->inotify_data.last_wd+1,
557 spin_unlock(&group->inotify_data.idr_lock);
559 /* idr was out of memory allocate and try again */
565 /* we put the mark on the idr, take a reference */
566 fsnotify_get_mark(&tmp_ientry->fsn_entry);
568 /* we are on the idr, now get on the inode */
569 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
571 /* we failed to get on the inode, get off the idr */
572 inotify_remove_from_idr(group, tmp_ientry);
576 /* update the idr hint, who cares about races, it's just a hint */
577 group->inotify_data.last_wd = tmp_ientry->wd;
579 /* increment the number of watches the user has */
580 atomic_inc(&group->inotify_data.user->inotify_watches);
582 /* return the watch descriptor for this new entry */
583 ret = tmp_ientry->wd;
585 /* match the ref from fsnotify_init_markentry() */
586 fsnotify_put_mark(&tmp_ientry->fsn_entry);
588 /* if this mark added a new event update the group mask */
589 if (mask & ~group->mask)
590 fsnotify_recalc_group_mask(group);
594 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
599 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
604 /* try to update and existing watch with the new arg */
605 ret = inotify_update_existing_watch(group, inode, arg);
606 /* no mark present, try to add a new one */
608 ret = inotify_new_watch(group, inode, arg);
610 * inotify_new_watch could race with another thread which did an
611 * inotify_new_watch between the update_existing and the add watch
612 * here, go back and try to update an existing mark again.
620 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
622 struct fsnotify_group *group;
623 unsigned int grp_num;
625 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
626 grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
627 group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
631 group->max_events = max_events;
633 spin_lock_init(&group->inotify_data.idr_lock);
634 idr_init(&group->inotify_data.idr);
635 group->inotify_data.last_wd = 0;
636 group->inotify_data.user = user;
637 group->inotify_data.fa = NULL;
643 /* inotify syscalls */
644 SYSCALL_DEFINE1(inotify_init1, int, flags)
646 struct fsnotify_group *group;
647 struct user_struct *user;
652 /* Check the IN_* constants for consistency. */
653 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
654 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
656 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
659 fd = get_unused_fd_flags(flags & O_CLOEXEC);
663 user = get_current_user();
664 if (unlikely(atomic_read(&user->inotify_devs) >=
665 inotify_max_user_instances)) {
670 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
671 group = inotify_new_group(user, inotify_max_queued_events);
673 ret = PTR_ERR(group);
677 atomic_inc(&user->inotify_devs);
679 path.mnt = inotify_mnt;
680 path.dentry = inotify_mnt->mnt_root;
682 filp = alloc_file(&path, FMODE_READ, &inotify_fops);
686 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
687 filp->private_data = group;
689 fd_install(fd, filp);
696 atomic_dec(&user->inotify_devs);
703 SYSCALL_DEFINE0(inotify_init)
705 return sys_inotify_init1(0);
708 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
711 struct fsnotify_group *group;
715 int ret, fput_needed;
718 filp = fget_light(fd, &fput_needed);
722 /* verify that this is indeed an inotify instance */
723 if (unlikely(filp->f_op != &inotify_fops)) {
728 if (!(mask & IN_DONT_FOLLOW))
729 flags |= LOOKUP_FOLLOW;
730 if (mask & IN_ONLYDIR)
731 flags |= LOOKUP_DIRECTORY;
733 ret = inotify_find_inode(pathname, &path, flags);
737 /* inode held in place by reference to path; group by fget on fd */
738 inode = path.dentry->d_inode;
739 group = filp->private_data;
741 /* create/update an inode mark */
742 ret = inotify_update_watch(group, inode, mask);
745 fput_light(filp, fput_needed);
749 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
751 struct fsnotify_group *group;
752 struct fsnotify_mark_entry *entry;
754 int ret = 0, fput_needed;
756 filp = fget_light(fd, &fput_needed);
760 /* verify that this is indeed an inotify instance */
761 if (unlikely(filp->f_op != &inotify_fops)) {
766 group = filp->private_data;
768 spin_lock(&group->inotify_data.idr_lock);
769 entry = idr_find(&group->inotify_data.idr, wd);
770 if (unlikely(!entry)) {
771 spin_unlock(&group->inotify_data.idr_lock);
775 fsnotify_get_mark(entry);
776 spin_unlock(&group->inotify_data.idr_lock);
778 fsnotify_destroy_mark_by_entry(entry);
779 fsnotify_put_mark(entry);
782 fput_light(filp, fput_needed);
787 inotify_get_sb(struct file_system_type *fs_type, int flags,
788 const char *dev_name, void *data, struct vfsmount *mnt)
790 return get_sb_pseudo(fs_type, "inotify", NULL,
791 INOTIFYFS_SUPER_MAGIC, mnt);
794 static struct file_system_type inotify_fs_type = {
796 .get_sb = inotify_get_sb,
797 .kill_sb = kill_anon_super,
801 * inotify_user_setup - Our initialization function. Note that we cannnot return
802 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
803 * must result in panic().
805 static int __init inotify_user_setup(void)
809 ret = register_filesystem(&inotify_fs_type);
811 panic("inotify: register_filesystem returned %d!\n", ret);
813 inotify_mnt = kern_mount(&inotify_fs_type);
814 if (IS_ERR(inotify_mnt))
815 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
817 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
818 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
820 inotify_max_queued_events = 16384;
821 inotify_max_user_instances = 128;
822 inotify_max_user_watches = 8192;
826 module_init(inotify_user_setup);