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 /* this just sits here and wastes global memory. used to just pad userspace messages with zeros */
51 static struct inotify_event nul_inotify_event;
53 /* these are configurable via /proc/sys/fs/inotify/ */
54 static int inotify_max_user_instances __read_mostly;
55 static int inotify_max_queued_events __read_mostly;
56 int inotify_max_user_watches __read_mostly;
58 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
59 struct kmem_cache *event_priv_cachep __read_mostly;
62 * When inotify registers a new group it increments this and uses that
63 * value as an offset to set the fsnotify group "name" and priority.
65 static atomic_t inotify_grp_num;
69 #include <linux/sysctl.h>
73 ctl_table inotify_table[] = {
75 .ctl_name = INOTIFY_MAX_USER_INSTANCES,
76 .procname = "max_user_instances",
77 .data = &inotify_max_user_instances,
78 .maxlen = sizeof(int),
80 .proc_handler = &proc_dointvec_minmax,
81 .strategy = &sysctl_intvec,
85 .ctl_name = INOTIFY_MAX_USER_WATCHES,
86 .procname = "max_user_watches",
87 .data = &inotify_max_user_watches,
88 .maxlen = sizeof(int),
90 .proc_handler = &proc_dointvec_minmax,
91 .strategy = &sysctl_intvec,
95 .ctl_name = INOTIFY_MAX_QUEUED_EVENTS,
96 .procname = "max_queued_events",
97 .data = &inotify_max_queued_events,
98 .maxlen = sizeof(int),
100 .proc_handler = &proc_dointvec_minmax,
101 .strategy = &sysctl_intvec,
106 #endif /* CONFIG_SYSCTL */
108 static inline __u32 inotify_arg_to_mask(u32 arg)
112 /* everything should accept their own ignored and cares about children */
113 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
115 /* mask off the flags used to open the fd */
116 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
121 static inline u32 inotify_mask_to_arg(__u32 mask)
123 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
127 /* intofiy userspace file descriptor functions */
128 static unsigned int inotify_poll(struct file *file, poll_table *wait)
130 struct fsnotify_group *group = file->private_data;
133 poll_wait(file, &group->notification_waitq, wait);
134 mutex_lock(&group->notification_mutex);
135 if (!fsnotify_notify_queue_is_empty(group))
136 ret = POLLIN | POLLRDNORM;
137 mutex_unlock(&group->notification_mutex);
143 * Get an inotify_kernel_event if one exists and is small
144 * enough to fit in "count". Return an error pointer if
147 * Called with the group->notification_mutex held.
149 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
152 size_t event_size = sizeof(struct inotify_event);
153 struct fsnotify_event *event;
155 if (fsnotify_notify_queue_is_empty(group))
158 event = fsnotify_peek_notify_event(group);
160 event_size += roundup(event->name_len, event_size);
162 if (event_size > count)
163 return ERR_PTR(-EINVAL);
165 /* held the notification_mutex the whole time, so this is the
166 * same event we peeked above */
167 fsnotify_remove_notify_event(group);
173 * Copy an event to user space, returning how much we copied.
175 * We already checked that the event size is smaller than the
176 * buffer we had in "get_one_event()" above.
178 static ssize_t copy_event_to_user(struct fsnotify_group *group,
179 struct fsnotify_event *event,
182 struct inotify_event inotify_event;
183 struct fsnotify_event_private_data *fsn_priv;
184 struct inotify_event_private_data *priv;
185 size_t event_size = sizeof(struct inotify_event);
188 /* we get the inotify watch descriptor from the event private data */
189 spin_lock(&event->lock);
190 fsn_priv = fsnotify_remove_priv_from_event(group, event);
191 spin_unlock(&event->lock);
194 inotify_event.wd = -1;
196 priv = container_of(fsn_priv, struct inotify_event_private_data,
197 fsnotify_event_priv_data);
198 inotify_event.wd = priv->wd;
199 inotify_free_event_priv(fsn_priv);
202 /* round up event->name_len so it is a multiple of event_size */
203 name_len = roundup(event->name_len, event_size);
204 inotify_event.len = name_len;
206 inotify_event.mask = inotify_mask_to_arg(event->mask);
207 inotify_event.cookie = event->sync_cookie;
209 /* send the main event */
210 if (copy_to_user(buf, &inotify_event, event_size))
216 * fsnotify only stores the pathname, so here we have to send the pathname
217 * and then pad that pathname out to a multiple of sizeof(inotify_event)
218 * with zeros. I get my zeros from the nul_inotify_event.
221 unsigned int len_to_zero = name_len - event->name_len;
222 /* copy the path name */
223 if (copy_to_user(buf, event->file_name, event->name_len))
225 buf += event->name_len;
227 /* fill userspace with 0's from nul_inotify_event */
228 if (copy_to_user(buf, &nul_inotify_event, len_to_zero))
231 event_size += name_len;
237 static ssize_t inotify_read(struct file *file, char __user *buf,
238 size_t count, loff_t *pos)
240 struct fsnotify_group *group;
241 struct fsnotify_event *kevent;
247 group = file->private_data;
250 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
252 mutex_lock(&group->notification_mutex);
253 kevent = get_one_event(group, count);
254 mutex_unlock(&group->notification_mutex);
257 ret = PTR_ERR(kevent);
260 ret = copy_event_to_user(group, kevent, buf);
261 fsnotify_put_event(kevent);
270 if (file->f_flags & O_NONBLOCK)
273 if (signal_pending(current))
282 finish_wait(&group->notification_waitq, &wait);
283 if (start != buf && ret != -EFAULT)
288 static int inotify_fasync(int fd, struct file *file, int on)
290 struct fsnotify_group *group = file->private_data;
292 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
295 static int inotify_release(struct inode *ignored, struct file *file)
297 struct fsnotify_group *group = file->private_data;
298 struct user_struct *user = group->inotify_data.user;
300 fsnotify_clear_marks_by_group(group);
302 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
303 fsnotify_put_group(group);
305 atomic_dec(&user->inotify_devs);
310 static long inotify_ioctl(struct file *file, unsigned int cmd,
313 struct fsnotify_group *group;
314 struct fsnotify_event_holder *holder;
315 struct fsnotify_event *event;
320 group = file->private_data;
321 p = (void __user *) arg;
325 mutex_lock(&group->notification_mutex);
326 list_for_each_entry(holder, &group->notification_list, event_list) {
327 event = holder->event;
328 send_len += sizeof(struct inotify_event);
329 send_len += roundup(event->name_len,
330 sizeof(struct inotify_event));
332 mutex_unlock(&group->notification_mutex);
333 ret = put_user(send_len, (int __user *) p);
340 static const struct file_operations inotify_fops = {
341 .poll = inotify_poll,
342 .read = inotify_read,
343 .fasync = inotify_fasync,
344 .release = inotify_release,
345 .unlocked_ioctl = inotify_ioctl,
346 .compat_ioctl = inotify_ioctl,
351 * find_inode - resolve a user-given path to a specific inode
353 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
357 error = user_path_at(AT_FDCWD, dirname, flags, path);
360 /* you can only watch an inode if you have read permissions on it */
361 error = inode_permission(path->dentry->d_inode, MAY_READ);
367 static void inotify_remove_from_idr(struct fsnotify_group *group,
368 struct inotify_inode_mark_entry *ientry)
372 spin_lock(&group->inotify_data.idr_lock);
373 idr = &group->inotify_data.idr;
374 idr_remove(idr, ientry->wd);
375 spin_unlock(&group->inotify_data.idr_lock);
379 * Send IN_IGNORED for this wd, remove this wd from the idr, and drop the
380 * internal reference help on the mark because it is in the idr.
382 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
383 struct fsnotify_group *group)
385 struct inotify_inode_mark_entry *ientry;
386 struct fsnotify_event *ignored_event;
387 struct inotify_event_private_data *event_priv;
388 struct fsnotify_event_private_data *fsn_event_priv;
391 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
392 FSNOTIFY_EVENT_NONE, NULL, 0,
397 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
399 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
400 if (unlikely(!event_priv))
401 goto skip_send_ignore;
403 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
405 fsn_event_priv->group = group;
406 event_priv->wd = ientry->wd;
408 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
410 inotify_free_event_priv(fsn_event_priv);
414 /* matches the reference taken when the event was created */
415 fsnotify_put_event(ignored_event);
417 /* remove this entry from the idr */
418 inotify_remove_from_idr(group, ientry);
420 /* removed from idr, drop that reference */
421 fsnotify_put_mark(entry);
423 atomic_dec(&group->inotify_data.user->inotify_watches);
426 /* ding dong the mark is dead */
427 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
429 struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
431 kmem_cache_free(inotify_inode_mark_cachep, ientry);
434 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
436 struct fsnotify_mark_entry *entry = NULL;
437 struct inotify_inode_mark_entry *ientry;
438 struct inotify_inode_mark_entry *tmp_ientry;
440 int add = (arg & IN_MASK_ADD);
442 __u32 old_mask, new_mask;
444 /* don't allow invalid bits: we don't want flags set */
445 mask = inotify_arg_to_mask(arg);
449 tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
450 if (unlikely(!tmp_ientry))
452 /* we set the mask at the end after attaching it */
453 fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
457 spin_lock(&inode->i_lock);
458 entry = fsnotify_find_mark_entry(group, inode);
459 spin_unlock(&inode->i_lock);
461 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
464 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
468 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
471 spin_lock(&group->inotify_data.idr_lock);
472 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
473 group->inotify_data.last_wd,
475 spin_unlock(&group->inotify_data.idr_lock);
482 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
484 inotify_remove_from_idr(group, tmp_ientry);
490 /* tmp_ientry has been added to the inode, so we are all set up.
491 * now we just need to make sure tmp_ientry doesn't get freed and
492 * we need to set up entry and ientry so the generic code can
495 entry = &ientry->fsn_entry;
498 atomic_inc(&group->inotify_data.user->inotify_watches);
500 /* update the idr hint */
501 group->inotify_data.last_wd = ientry->wd;
503 /* we put the mark on the idr, take a reference */
504 fsnotify_get_mark(entry);
509 spin_lock(&entry->lock);
511 old_mask = entry->mask;
514 new_mask = entry->mask;
517 new_mask = entry->mask;
520 spin_unlock(&entry->lock);
522 if (old_mask != new_mask) {
523 /* more bits in old than in new? */
524 int dropped = (old_mask & ~new_mask);
525 /* more bits in this entry than the inode's mask? */
526 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
527 /* more bits in this entry than the group? */
528 int do_group = (new_mask & ~group->mask);
530 /* update the inode with this new entry */
531 if (dropped || do_inode)
532 fsnotify_recalc_inode_mask(inode);
534 /* update the group mask with the new mask */
535 if (dropped || do_group)
536 fsnotify_recalc_group_mask(group);
539 /* this either matches fsnotify_find_mark_entry, or init_mark_entry
540 * depending on which path we took... */
541 fsnotify_put_mark(entry);
544 /* could be an error, could be that we found an existing mark */
546 /* on the idr but didn't make it on the inode */
547 if (tmp_ientry->wd != -1)
548 inotify_remove_from_idr(group, tmp_ientry);
549 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
555 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
557 struct fsnotify_group *group;
558 unsigned int grp_num;
560 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
561 grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
562 group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
566 group->max_events = max_events;
568 spin_lock_init(&group->inotify_data.idr_lock);
569 idr_init(&group->inotify_data.idr);
570 group->inotify_data.last_wd = 1;
571 group->inotify_data.user = user;
572 group->inotify_data.fa = NULL;
578 /* inotify syscalls */
579 SYSCALL_DEFINE1(inotify_init1, int, flags)
581 struct fsnotify_group *group;
582 struct user_struct *user;
586 /* Check the IN_* constants for consistency. */
587 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
588 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
590 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
593 fd = get_unused_fd_flags(flags & O_CLOEXEC);
597 filp = get_empty_filp();
603 user = get_current_user();
604 if (unlikely(atomic_read(&user->inotify_devs) >=
605 inotify_max_user_instances)) {
610 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
611 group = inotify_new_group(user, inotify_max_queued_events);
613 ret = PTR_ERR(group);
617 filp->f_op = &inotify_fops;
618 filp->f_path.mnt = mntget(inotify_mnt);
619 filp->f_path.dentry = dget(inotify_mnt->mnt_root);
620 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
621 filp->f_mode = FMODE_READ;
622 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
623 filp->private_data = group;
625 atomic_inc(&user->inotify_devs);
627 fd_install(fd, filp);
639 SYSCALL_DEFINE0(inotify_init)
641 return sys_inotify_init1(0);
644 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
647 struct fsnotify_group *group;
651 int ret, fput_needed;
654 filp = fget_light(fd, &fput_needed);
658 /* verify that this is indeed an inotify instance */
659 if (unlikely(filp->f_op != &inotify_fops)) {
664 if (!(mask & IN_DONT_FOLLOW))
665 flags |= LOOKUP_FOLLOW;
666 if (mask & IN_ONLYDIR)
667 flags |= LOOKUP_DIRECTORY;
669 ret = inotify_find_inode(pathname, &path, flags);
673 /* inode held in place by reference to path; group by fget on fd */
674 inode = path.dentry->d_inode;
675 group = filp->private_data;
677 /* create/update an inode mark */
678 ret = inotify_update_watch(group, inode, mask);
680 goto path_put_and_out;
685 fput_light(filp, fput_needed);
689 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
691 struct fsnotify_group *group;
692 struct fsnotify_mark_entry *entry;
694 int ret = 0, fput_needed;
696 filp = fget_light(fd, &fput_needed);
700 /* verify that this is indeed an inotify instance */
701 if (unlikely(filp->f_op != &inotify_fops)) {
706 group = filp->private_data;
708 spin_lock(&group->inotify_data.idr_lock);
709 entry = idr_find(&group->inotify_data.idr, wd);
710 if (unlikely(!entry)) {
711 spin_unlock(&group->inotify_data.idr_lock);
715 fsnotify_get_mark(entry);
716 spin_unlock(&group->inotify_data.idr_lock);
718 fsnotify_destroy_mark_by_entry(entry);
719 fsnotify_put_mark(entry);
722 fput_light(filp, fput_needed);
727 inotify_get_sb(struct file_system_type *fs_type, int flags,
728 const char *dev_name, void *data, struct vfsmount *mnt)
730 return get_sb_pseudo(fs_type, "inotify", NULL,
731 INOTIFYFS_SUPER_MAGIC, mnt);
734 static struct file_system_type inotify_fs_type = {
736 .get_sb = inotify_get_sb,
737 .kill_sb = kill_anon_super,
741 * inotify_user_setup - Our initialization function. Note that we cannnot return
742 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
743 * must result in panic().
745 static int __init inotify_user_setup(void)
749 ret = register_filesystem(&inotify_fs_type);
751 panic("inotify: register_filesystem returned %d!\n", ret);
753 inotify_mnt = kern_mount(&inotify_fs_type);
754 if (IS_ERR(inotify_mnt))
755 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
757 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
758 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
760 inotify_max_queued_events = 16384;
761 inotify_max_user_instances = 128;
762 inotify_max_user_watches = 8192;
766 module_init(inotify_user_setup);