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);
157 event_size += roundup(event->name_len, event_size);
159 if (event_size > count)
160 return ERR_PTR(-EINVAL);
162 /* held the notification_mutex the whole time, so this is the
163 * same event we peeked above */
164 fsnotify_remove_notify_event(group);
170 * Copy an event to user space, returning how much we copied.
172 * We already checked that the event size is smaller than the
173 * buffer we had in "get_one_event()" above.
175 static ssize_t copy_event_to_user(struct fsnotify_group *group,
176 struct fsnotify_event *event,
179 struct inotify_event inotify_event;
180 struct fsnotify_event_private_data *fsn_priv;
181 struct inotify_event_private_data *priv;
182 size_t event_size = sizeof(struct inotify_event);
185 /* we get the inotify watch descriptor from the event private data */
186 spin_lock(&event->lock);
187 fsn_priv = fsnotify_remove_priv_from_event(group, event);
188 spin_unlock(&event->lock);
191 inotify_event.wd = -1;
193 priv = container_of(fsn_priv, struct inotify_event_private_data,
194 fsnotify_event_priv_data);
195 inotify_event.wd = priv->wd;
196 inotify_free_event_priv(fsn_priv);
200 * round up event->name_len so it is a multiple of event_size
201 * plus an extra byte for the terminating '\0'.
204 name_len = roundup(event->name_len + 1, event_size);
205 inotify_event.len = name_len;
207 inotify_event.mask = inotify_mask_to_arg(event->mask);
208 inotify_event.cookie = event->sync_cookie;
210 /* send the main event */
211 if (copy_to_user(buf, &inotify_event, event_size))
217 * fsnotify only stores the pathname, so here we have to send the pathname
218 * and then pad that pathname out to a multiple of sizeof(inotify_event)
219 * with zeros. I get my zeros from the nul_inotify_event.
222 unsigned int len_to_zero = name_len - event->name_len;
223 /* copy the path name */
224 if (copy_to_user(buf, event->file_name, event->name_len))
226 buf += event->name_len;
228 /* fill userspace with 0's */
229 if (clear_user(buf, len_to_zero))
232 event_size += name_len;
238 static ssize_t inotify_read(struct file *file, char __user *buf,
239 size_t count, loff_t *pos)
241 struct fsnotify_group *group;
242 struct fsnotify_event *kevent;
248 group = file->private_data;
251 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
253 mutex_lock(&group->notification_mutex);
254 kevent = get_one_event(group, count);
255 mutex_unlock(&group->notification_mutex);
258 ret = PTR_ERR(kevent);
261 ret = copy_event_to_user(group, kevent, buf);
262 fsnotify_put_event(kevent);
271 if (file->f_flags & O_NONBLOCK)
274 if (signal_pending(current))
283 finish_wait(&group->notification_waitq, &wait);
284 if (start != buf && ret != -EFAULT)
289 static int inotify_fasync(int fd, struct file *file, int on)
291 struct fsnotify_group *group = file->private_data;
293 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
296 static int inotify_release(struct inode *ignored, struct file *file)
298 struct fsnotify_group *group = file->private_data;
299 struct user_struct *user = group->inotify_data.user;
301 fsnotify_clear_marks_by_group(group);
303 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
304 fsnotify_put_group(group);
306 atomic_dec(&user->inotify_devs);
311 static long inotify_ioctl(struct file *file, unsigned int cmd,
314 struct fsnotify_group *group;
315 struct fsnotify_event_holder *holder;
316 struct fsnotify_event *event;
321 group = file->private_data;
322 p = (void __user *) arg;
326 mutex_lock(&group->notification_mutex);
327 list_for_each_entry(holder, &group->notification_list, event_list) {
328 event = holder->event;
329 send_len += sizeof(struct inotify_event);
330 send_len += roundup(event->name_len,
331 sizeof(struct inotify_event));
333 mutex_unlock(&group->notification_mutex);
334 ret = put_user(send_len, (int __user *) p);
341 static const struct file_operations inotify_fops = {
342 .poll = inotify_poll,
343 .read = inotify_read,
344 .fasync = inotify_fasync,
345 .release = inotify_release,
346 .unlocked_ioctl = inotify_ioctl,
347 .compat_ioctl = inotify_ioctl,
352 * find_inode - resolve a user-given path to a specific inode
354 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
358 error = user_path_at(AT_FDCWD, dirname, flags, path);
361 /* you can only watch an inode if you have read permissions on it */
362 error = inode_permission(path->dentry->d_inode, MAY_READ);
369 * Remove the mark from the idr (if present) and drop the reference
370 * on the mark because it was in the idr.
372 static void inotify_remove_from_idr(struct fsnotify_group *group,
373 struct inotify_inode_mark_entry *ientry)
376 struct fsnotify_mark_entry *entry;
377 struct inotify_inode_mark_entry *found_ientry;
380 spin_lock(&group->inotify_data.idr_lock);
381 idr = &group->inotify_data.idr;
387 entry = idr_find(&group->inotify_data.idr, wd);
388 if (unlikely(!entry))
391 found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
392 if (unlikely(found_ientry != ientry)) {
393 /* We found an entry in the idr with the right wd, but it's
394 * not the entry we were told to remove. eparis seriously
395 * fucked up somewhere. */
401 /* One ref for being in the idr, one ref held by the caller */
402 BUG_ON(atomic_read(&entry->refcnt) < 2);
407 /* removed from the idr, drop that ref */
408 fsnotify_put_mark(entry);
410 spin_unlock(&group->inotify_data.idr_lock);
414 * Send IN_IGNORED for this wd, remove this wd from the idr.
416 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
417 struct fsnotify_group *group)
419 struct inotify_inode_mark_entry *ientry;
420 struct fsnotify_event *ignored_event;
421 struct inotify_event_private_data *event_priv;
422 struct fsnotify_event_private_data *fsn_event_priv;
425 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
426 FSNOTIFY_EVENT_NONE, NULL, 0,
431 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
433 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
434 if (unlikely(!event_priv))
435 goto skip_send_ignore;
437 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
439 fsn_event_priv->group = group;
440 event_priv->wd = ientry->wd;
442 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
444 inotify_free_event_priv(fsn_event_priv);
448 /* matches the reference taken when the event was created */
449 fsnotify_put_event(ignored_event);
451 /* remove this entry from the idr */
452 inotify_remove_from_idr(group, ientry);
454 atomic_dec(&group->inotify_data.user->inotify_watches);
457 /* ding dong the mark is dead */
458 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
460 struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
462 kmem_cache_free(inotify_inode_mark_cachep, ientry);
465 static int inotify_update_existing_watch(struct fsnotify_group *group,
469 struct fsnotify_mark_entry *entry;
470 struct inotify_inode_mark_entry *ientry;
471 __u32 old_mask, new_mask;
473 int add = (arg & IN_MASK_ADD);
476 /* don't allow invalid bits: we don't want flags set */
477 mask = inotify_arg_to_mask(arg);
481 spin_lock(&inode->i_lock);
482 entry = fsnotify_find_mark_entry(group, inode);
483 spin_unlock(&inode->i_lock);
487 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
489 spin_lock(&entry->lock);
491 old_mask = entry->mask;
494 new_mask = entry->mask;
497 new_mask = entry->mask;
500 spin_unlock(&entry->lock);
502 if (old_mask != new_mask) {
503 /* more bits in old than in new? */
504 int dropped = (old_mask & ~new_mask);
505 /* more bits in this entry than the inode's mask? */
506 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
507 /* more bits in this entry than the group? */
508 int do_group = (new_mask & ~group->mask);
510 /* update the inode with this new entry */
511 if (dropped || do_inode)
512 fsnotify_recalc_inode_mask(inode);
514 /* update the group mask with the new mask */
515 if (dropped || do_group)
516 fsnotify_recalc_group_mask(group);
522 /* match the get from fsnotify_find_mark_entry() */
523 fsnotify_put_mark(entry);
528 static int inotify_new_watch(struct fsnotify_group *group,
532 struct inotify_inode_mark_entry *tmp_ientry;
536 /* don't allow invalid bits: we don't want flags set */
537 mask = inotify_arg_to_mask(arg);
541 tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
542 if (unlikely(!tmp_ientry))
545 fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
546 tmp_ientry->fsn_entry.mask = mask;
550 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
554 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
557 spin_lock(&group->inotify_data.idr_lock);
558 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
559 group->inotify_data.last_wd,
561 spin_unlock(&group->inotify_data.idr_lock);
563 /* idr was out of memory allocate and try again */
569 /* we put the mark on the idr, take a reference */
570 fsnotify_get_mark(&tmp_ientry->fsn_entry);
572 /* we are on the idr, now get on the inode */
573 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
575 /* we failed to get on the inode, get off the idr */
576 inotify_remove_from_idr(group, tmp_ientry);
580 /* update the idr hint, who cares about races, it's just a hint */
581 group->inotify_data.last_wd = tmp_ientry->wd;
583 /* increment the number of watches the user has */
584 atomic_inc(&group->inotify_data.user->inotify_watches);
586 /* return the watch descriptor for this new entry */
587 ret = tmp_ientry->wd;
589 /* match the ref from fsnotify_init_markentry() */
590 fsnotify_put_mark(&tmp_ientry->fsn_entry);
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 = 1;
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;
651 /* Check the IN_* constants for consistency. */
652 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
653 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
655 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
658 fd = get_unused_fd_flags(flags & O_CLOEXEC);
662 filp = get_empty_filp();
668 user = get_current_user();
669 if (unlikely(atomic_read(&user->inotify_devs) >=
670 inotify_max_user_instances)) {
675 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
676 group = inotify_new_group(user, inotify_max_queued_events);
678 ret = PTR_ERR(group);
682 filp->f_op = &inotify_fops;
683 filp->f_path.mnt = mntget(inotify_mnt);
684 filp->f_path.dentry = dget(inotify_mnt->mnt_root);
685 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
686 filp->f_mode = FMODE_READ;
687 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
688 filp->private_data = group;
690 atomic_inc(&user->inotify_devs);
692 fd_install(fd, filp);
704 SYSCALL_DEFINE0(inotify_init)
706 return sys_inotify_init1(0);
709 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
712 struct fsnotify_group *group;
716 int ret, fput_needed;
719 filp = fget_light(fd, &fput_needed);
723 /* verify that this is indeed an inotify instance */
724 if (unlikely(filp->f_op != &inotify_fops)) {
729 if (!(mask & IN_DONT_FOLLOW))
730 flags |= LOOKUP_FOLLOW;
731 if (mask & IN_ONLYDIR)
732 flags |= LOOKUP_DIRECTORY;
734 ret = inotify_find_inode(pathname, &path, flags);
738 /* inode held in place by reference to path; group by fget on fd */
739 inode = path.dentry->d_inode;
740 group = filp->private_data;
742 /* create/update an inode mark */
743 ret = inotify_update_watch(group, inode, mask);
745 goto path_put_and_out;
750 fput_light(filp, fput_needed);
754 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
756 struct fsnotify_group *group;
757 struct fsnotify_mark_entry *entry;
759 int ret = 0, fput_needed;
761 filp = fget_light(fd, &fput_needed);
765 /* verify that this is indeed an inotify instance */
766 if (unlikely(filp->f_op != &inotify_fops)) {
771 group = filp->private_data;
773 spin_lock(&group->inotify_data.idr_lock);
774 entry = idr_find(&group->inotify_data.idr, wd);
775 if (unlikely(!entry)) {
776 spin_unlock(&group->inotify_data.idr_lock);
780 fsnotify_get_mark(entry);
781 spin_unlock(&group->inotify_data.idr_lock);
783 fsnotify_destroy_mark_by_entry(entry);
784 fsnotify_put_mark(entry);
787 fput_light(filp, fput_needed);
792 inotify_get_sb(struct file_system_type *fs_type, int flags,
793 const char *dev_name, void *data, struct vfsmount *mnt)
795 return get_sb_pseudo(fs_type, "inotify", NULL,
796 INOTIFYFS_SUPER_MAGIC, mnt);
799 static struct file_system_type inotify_fs_type = {
801 .get_sb = inotify_get_sb,
802 .kill_sb = kill_anon_super,
806 * inotify_user_setup - Our initialization function. Note that we cannnot return
807 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
808 * must result in panic().
810 static int __init inotify_user_setup(void)
814 ret = register_filesystem(&inotify_fs_type);
816 panic("inotify: register_filesystem returned %d!\n", ret);
818 inotify_mnt = kern_mount(&inotify_fs_type);
819 if (IS_ERR(inotify_mnt))
820 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
822 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
823 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
825 inotify_max_queued_events = 16384;
826 inotify_max_user_instances = 128;
827 inotify_max_user_watches = 8192;
831 module_init(inotify_user_setup);