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 * This program is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by the
13 * Free Software Foundation; either version 2, or (at your option) any
16 * This program is distributed in the hope that it will be useful, but
17 * WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/namei.h>
29 #include <linux/poll.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/inotify.h>
33 #include <linux/syscalls.h>
34 #include <linux/magic.h>
36 #include <asm/ioctls.h>
38 static struct kmem_cache *watch_cachep __read_mostly;
39 static struct kmem_cache *event_cachep __read_mostly;
41 static struct vfsmount *inotify_mnt __read_mostly;
43 /* these are configurable via /proc/sys/fs/inotify/ */
44 static int inotify_max_user_instances __read_mostly;
45 static int inotify_max_user_watches __read_mostly;
46 static int inotify_max_queued_events __read_mostly;
51 * inotify_dev->up_mutex (ensures we don't re-add the same watch)
52 * inode->inotify_mutex (protects inode's watch list)
53 * inotify_handle->mutex (protects inotify_handle's watch list)
54 * inotify_dev->ev_mutex (protects device's event queue)
58 * Lifetimes of the main data structures:
60 * inotify_device: Lifetime is managed by reference count, from
61 * sys_inotify_init() until release. Additional references can bump the count
62 * via get_inotify_dev() and drop the count via put_inotify_dev().
64 * inotify_user_watch: Lifetime is from create_watch() to the receipt of an
65 * IN_IGNORED event from inotify, or when using IN_ONESHOT, to receipt of the
66 * first event, or to inotify_destroy().
70 * struct inotify_device - represents an inotify instance
72 * This structure is protected by the mutex 'mutex'.
74 struct inotify_device {
75 wait_queue_head_t wq; /* wait queue for i/o */
76 struct mutex ev_mutex; /* protects event queue */
77 struct mutex up_mutex; /* synchronizes watch updates */
78 struct list_head events; /* list of queued events */
79 struct user_struct *user; /* user who opened this dev */
80 struct inotify_handle *ih; /* inotify handle */
81 struct fasync_struct *fa; /* async notification */
82 atomic_t count; /* reference count */
83 unsigned int queue_size; /* size of the queue (bytes) */
84 unsigned int event_count; /* number of pending events */
85 unsigned int max_events; /* maximum number of events */
89 * struct inotify_kernel_event - An inotify event, originating from a watch and
90 * queued for user-space. A list of these is attached to each instance of the
91 * device. In read(), this list is walked and all events that can fit in the
92 * buffer are returned.
94 * Protected by dev->ev_mutex of the device in which we are queued.
96 struct inotify_kernel_event {
97 struct inotify_event event; /* the user-space event */
98 struct list_head list; /* entry in inotify_device's list */
99 char *name; /* filename, if any */
103 * struct inotify_user_watch - our version of an inotify_watch, we add
104 * a reference to the associated inotify_device.
106 struct inotify_user_watch {
107 struct inotify_device *dev; /* associated device */
108 struct inotify_watch wdata; /* inotify watch data */
113 #include <linux/sysctl.h>
117 ctl_table inotify_table[] = {
119 .ctl_name = INOTIFY_MAX_USER_INSTANCES,
120 .procname = "max_user_instances",
121 .data = &inotify_max_user_instances,
122 .maxlen = sizeof(int),
124 .proc_handler = &proc_dointvec_minmax,
125 .strategy = &sysctl_intvec,
129 .ctl_name = INOTIFY_MAX_USER_WATCHES,
130 .procname = "max_user_watches",
131 .data = &inotify_max_user_watches,
132 .maxlen = sizeof(int),
134 .proc_handler = &proc_dointvec_minmax,
135 .strategy = &sysctl_intvec,
139 .ctl_name = INOTIFY_MAX_QUEUED_EVENTS,
140 .procname = "max_queued_events",
141 .data = &inotify_max_queued_events,
142 .maxlen = sizeof(int),
144 .proc_handler = &proc_dointvec_minmax,
145 .strategy = &sysctl_intvec,
150 #endif /* CONFIG_SYSCTL */
152 static inline void get_inotify_dev(struct inotify_device *dev)
154 atomic_inc(&dev->count);
157 static inline void put_inotify_dev(struct inotify_device *dev)
159 if (atomic_dec_and_test(&dev->count)) {
160 atomic_dec(&dev->user->inotify_devs);
167 * free_inotify_user_watch - cleans up the watch and its references
169 static void free_inotify_user_watch(struct inotify_watch *w)
171 struct inotify_user_watch *watch;
172 struct inotify_device *dev;
174 watch = container_of(w, struct inotify_user_watch, wdata);
177 atomic_dec(&dev->user->inotify_watches);
178 put_inotify_dev(dev);
179 kmem_cache_free(watch_cachep, watch);
183 * kernel_event - create a new kernel event with the given parameters
185 * This function can sleep.
187 static struct inotify_kernel_event * kernel_event(s32 wd, u32 mask, u32 cookie,
190 struct inotify_kernel_event *kevent;
192 kevent = kmem_cache_alloc(event_cachep, GFP_NOFS);
193 if (unlikely(!kevent))
196 /* we hand this out to user-space, so zero it just in case */
197 memset(&kevent->event, 0, sizeof(struct inotify_event));
199 kevent->event.wd = wd;
200 kevent->event.mask = mask;
201 kevent->event.cookie = cookie;
203 INIT_LIST_HEAD(&kevent->list);
206 size_t len, rem, event_size = sizeof(struct inotify_event);
209 * We need to pad the filename so as to properly align an
210 * array of inotify_event structures. Because the structure is
211 * small and the common case is a small filename, we just round
212 * up to the next multiple of the structure's sizeof. This is
213 * simple and safe for all architectures.
215 len = strlen(name) + 1;
216 rem = event_size - len;
217 if (len > event_size) {
218 rem = event_size - (len % event_size);
219 if (len % event_size == 0)
223 kevent->name = kmalloc(len + rem, GFP_KERNEL);
224 if (unlikely(!kevent->name)) {
225 kmem_cache_free(event_cachep, kevent);
228 memcpy(kevent->name, name, len);
230 memset(kevent->name + len, 0, rem);
231 kevent->event.len = len + rem;
233 kevent->event.len = 0;
241 * inotify_dev_get_event - return the next event in the given dev's queue
243 * Caller must hold dev->ev_mutex.
245 static inline struct inotify_kernel_event *
246 inotify_dev_get_event(struct inotify_device *dev)
248 return list_entry(dev->events.next, struct inotify_kernel_event, list);
252 * inotify_dev_get_last_event - return the last event in the given dev's queue
254 * Caller must hold dev->ev_mutex.
256 static inline struct inotify_kernel_event *
257 inotify_dev_get_last_event(struct inotify_device *dev)
259 if (list_empty(&dev->events))
261 return list_entry(dev->events.prev, struct inotify_kernel_event, list);
265 * inotify_dev_queue_event - event handler registered with core inotify, adds
266 * a new event to the given device
268 * Can sleep (calls kernel_event()).
270 static void inotify_dev_queue_event(struct inotify_watch *w, u32 wd, u32 mask,
271 u32 cookie, const char *name,
272 struct inode *ignored)
274 struct inotify_user_watch *watch;
275 struct inotify_device *dev;
276 struct inotify_kernel_event *kevent, *last;
278 watch = container_of(w, struct inotify_user_watch, wdata);
281 mutex_lock(&dev->ev_mutex);
283 /* we can safely put the watch as we don't reference it while
284 * generating the event
286 if (mask & IN_IGNORED || w->mask & IN_ONESHOT)
287 put_inotify_watch(w); /* final put */
289 /* coalescing: drop this event if it is a dupe of the previous */
290 last = inotify_dev_get_last_event(dev);
291 if (last && last->event.mask == mask && last->event.wd == wd &&
292 last->event.cookie == cookie) {
293 const char *lastname = last->name;
295 if (!name && !lastname)
297 if (name && lastname && !strcmp(lastname, name))
301 /* the queue overflowed and we already sent the Q_OVERFLOW event */
302 if (unlikely(dev->event_count > dev->max_events))
305 /* if the queue overflows, we need to notify user space */
306 if (unlikely(dev->event_count == dev->max_events))
307 kevent = kernel_event(-1, IN_Q_OVERFLOW, cookie, NULL);
309 kevent = kernel_event(wd, mask, cookie, name);
311 if (unlikely(!kevent))
314 /* queue the event and wake up anyone waiting */
316 dev->queue_size += sizeof(struct inotify_event) + kevent->event.len;
317 list_add_tail(&kevent->list, &dev->events);
318 wake_up_interruptible(&dev->wq);
319 kill_fasync(&dev->fa, SIGIO, POLL_IN);
322 mutex_unlock(&dev->ev_mutex);
326 * remove_kevent - cleans up the given kevent
328 * Caller must hold dev->ev_mutex.
330 static void remove_kevent(struct inotify_device *dev,
331 struct inotify_kernel_event *kevent)
333 list_del(&kevent->list);
336 dev->queue_size -= sizeof(struct inotify_event) + kevent->event.len;
340 * free_kevent - frees the given kevent.
342 static void free_kevent(struct inotify_kernel_event *kevent)
345 kmem_cache_free(event_cachep, kevent);
349 * inotify_dev_event_dequeue - destroy an event on the given device
351 * Caller must hold dev->ev_mutex.
353 static void inotify_dev_event_dequeue(struct inotify_device *dev)
355 if (!list_empty(&dev->events)) {
356 struct inotify_kernel_event *kevent;
357 kevent = inotify_dev_get_event(dev);
358 remove_kevent(dev, kevent);
364 * find_inode - resolve a user-given path to a specific inode
366 static int find_inode(const char __user *dirname, struct path *path,
371 error = user_path_at(AT_FDCWD, dirname, flags, path);
374 /* you can only watch an inode if you have read permissions on it */
375 error = inode_permission(path->dentry->d_inode, MAY_READ);
382 * create_watch - creates a watch on the given device.
384 * Callers must hold dev->up_mutex.
386 static int create_watch(struct inotify_device *dev, struct inode *inode,
389 struct inotify_user_watch *watch;
392 if (atomic_read(&dev->user->inotify_watches) >=
393 inotify_max_user_watches)
396 watch = kmem_cache_alloc(watch_cachep, GFP_KERNEL);
397 if (unlikely(!watch))
400 /* save a reference to device and bump the count to make it official */
401 get_inotify_dev(dev);
404 atomic_inc(&dev->user->inotify_watches);
406 inotify_init_watch(&watch->wdata);
407 ret = inotify_add_watch(dev->ih, &watch->wdata, inode, mask);
409 free_inotify_user_watch(&watch->wdata);
414 /* Device Interface */
416 static unsigned int inotify_poll(struct file *file, poll_table *wait)
418 struct inotify_device *dev = file->private_data;
421 poll_wait(file, &dev->wq, wait);
422 mutex_lock(&dev->ev_mutex);
423 if (!list_empty(&dev->events))
424 ret = POLLIN | POLLRDNORM;
425 mutex_unlock(&dev->ev_mutex);
431 * Get an inotify_kernel_event if one exists and is small
432 * enough to fit in "count". Return an error pointer if
435 * Called with the device ev_mutex held.
437 static struct inotify_kernel_event *get_one_event(struct inotify_device *dev,
440 size_t event_size = sizeof(struct inotify_event);
441 struct inotify_kernel_event *kevent;
443 if (list_empty(&dev->events))
446 kevent = inotify_dev_get_event(dev);
448 event_size += kevent->event.len;
450 if (event_size > count)
451 return ERR_PTR(-EINVAL);
453 remove_kevent(dev, kevent);
458 * Copy an event to user space, returning how much we copied.
460 * We already checked that the event size is smaller than the
461 * buffer we had in "get_one_event()" above.
463 static ssize_t copy_event_to_user(struct inotify_kernel_event *kevent,
466 size_t event_size = sizeof(struct inotify_event);
468 if (copy_to_user(buf, &kevent->event, event_size))
474 if (copy_to_user(buf, kevent->name, kevent->event.len))
477 event_size += kevent->event.len;
482 static ssize_t inotify_read(struct file *file, char __user *buf,
483 size_t count, loff_t *pos)
485 struct inotify_device *dev;
491 dev = file->private_data;
494 struct inotify_kernel_event *kevent;
496 prepare_to_wait(&dev->wq, &wait, TASK_INTERRUPTIBLE);
498 mutex_lock(&dev->ev_mutex);
499 kevent = get_one_event(dev, count);
500 mutex_unlock(&dev->ev_mutex);
503 ret = PTR_ERR(kevent);
506 ret = copy_event_to_user(kevent, buf);
516 if (file->f_flags & O_NONBLOCK)
519 if (signal_pending(current))
528 finish_wait(&dev->wq, &wait);
529 if (start != buf && ret != -EFAULT)
534 static int inotify_fasync(int fd, struct file *file, int on)
536 struct inotify_device *dev = file->private_data;
538 return fasync_helper(fd, file, on, &dev->fa) >= 0 ? 0 : -EIO;
541 static int inotify_release(struct inode *ignored, struct file *file)
543 struct inotify_device *dev = file->private_data;
545 inotify_destroy(dev->ih);
547 /* destroy all of the events on this device */
548 mutex_lock(&dev->ev_mutex);
549 while (!list_empty(&dev->events))
550 inotify_dev_event_dequeue(dev);
551 mutex_unlock(&dev->ev_mutex);
553 /* free this device: the put matching the get in inotify_init() */
554 put_inotify_dev(dev);
559 static long inotify_ioctl(struct file *file, unsigned int cmd,
562 struct inotify_device *dev;
566 dev = file->private_data;
567 p = (void __user *) arg;
571 ret = put_user(dev->queue_size, (int __user *) p);
578 static const struct file_operations inotify_fops = {
579 .poll = inotify_poll,
580 .read = inotify_read,
581 .fasync = inotify_fasync,
582 .release = inotify_release,
583 .unlocked_ioctl = inotify_ioctl,
584 .compat_ioctl = inotify_ioctl,
587 static const struct inotify_operations inotify_user_ops = {
588 .handle_event = inotify_dev_queue_event,
589 .destroy_watch = free_inotify_user_watch,
592 SYSCALL_DEFINE1(inotify_init1, int, flags)
594 struct inotify_device *dev;
595 struct inotify_handle *ih;
596 struct user_struct *user;
600 /* Check the IN_* constants for consistency. */
601 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
602 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
604 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
607 fd = get_unused_fd_flags(flags & O_CLOEXEC);
611 filp = get_empty_filp();
617 user = get_current_user();
618 if (unlikely(atomic_read(&user->inotify_devs) >=
619 inotify_max_user_instances)) {
624 dev = kmalloc(sizeof(struct inotify_device), GFP_KERNEL);
625 if (unlikely(!dev)) {
630 ih = inotify_init(&inotify_user_ops);
638 filp->f_op = &inotify_fops;
639 filp->f_path.mnt = mntget(inotify_mnt);
640 filp->f_path.dentry = dget(inotify_mnt->mnt_root);
641 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
642 filp->f_mode = FMODE_READ;
643 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
644 filp->private_data = dev;
646 INIT_LIST_HEAD(&dev->events);
647 init_waitqueue_head(&dev->wq);
648 mutex_init(&dev->ev_mutex);
649 mutex_init(&dev->up_mutex);
650 dev->event_count = 0;
652 dev->max_events = inotify_max_queued_events;
654 atomic_set(&dev->count, 0);
656 get_inotify_dev(dev);
657 atomic_inc(&user->inotify_devs);
658 fd_install(fd, filp);
671 SYSCALL_DEFINE0(inotify_init)
673 return sys_inotify_init1(0);
676 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
680 struct inotify_device *dev;
683 int ret, fput_needed;
686 filp = fget_light(fd, &fput_needed);
690 /* verify that this is indeed an inotify instance */
691 if (unlikely(filp->f_op != &inotify_fops)) {
696 if (!(mask & IN_DONT_FOLLOW))
697 flags |= LOOKUP_FOLLOW;
698 if (mask & IN_ONLYDIR)
699 flags |= LOOKUP_DIRECTORY;
701 ret = find_inode(pathname, &path, flags);
705 /* inode held in place by reference to path; dev by fget on fd */
706 inode = path.dentry->d_inode;
707 dev = filp->private_data;
709 mutex_lock(&dev->up_mutex);
710 ret = inotify_find_update_watch(dev->ih, inode, mask);
712 ret = create_watch(dev, inode, mask);
713 mutex_unlock(&dev->up_mutex);
717 fput_light(filp, fput_needed);
721 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
724 struct inotify_device *dev;
725 int ret, fput_needed;
727 filp = fget_light(fd, &fput_needed);
731 /* verify that this is indeed an inotify instance */
732 if (unlikely(filp->f_op != &inotify_fops)) {
737 dev = filp->private_data;
739 /* we free our watch data when we get IN_IGNORED */
740 ret = inotify_rm_wd(dev->ih, wd);
743 fput_light(filp, fput_needed);
748 inotify_get_sb(struct file_system_type *fs_type, int flags,
749 const char *dev_name, void *data, struct vfsmount *mnt)
751 return get_sb_pseudo(fs_type, "inotify", NULL,
752 INOTIFYFS_SUPER_MAGIC, mnt);
755 static struct file_system_type inotify_fs_type = {
757 .get_sb = inotify_get_sb,
758 .kill_sb = kill_anon_super,
762 * inotify_user_setup - Our initialization function. Note that we cannnot return
763 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
764 * must result in panic().
766 static int __init inotify_user_setup(void)
770 ret = register_filesystem(&inotify_fs_type);
772 panic("inotify: register_filesystem returned %d!\n", ret);
774 inotify_mnt = kern_mount(&inotify_fs_type);
775 if (IS_ERR(inotify_mnt))
776 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
778 inotify_max_queued_events = 16384;
779 inotify_max_user_instances = 128;
780 inotify_max_user_watches = 8192;
782 watch_cachep = kmem_cache_create("inotify_watch_cache",
783 sizeof(struct inotify_user_watch),
784 0, SLAB_PANIC, NULL);
785 event_cachep = kmem_cache_create("inotify_event_cache",
786 sizeof(struct inotify_kernel_event),
787 0, SLAB_PANIC, NULL);
792 module_init(inotify_user_setup);