2 * fs/inotify_user.c - inotify support for userspace
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> /* fs_initcall */
30 #include <linux/inotify.h>
31 #include <linux/kernel.h> /* roundup() */
32 #include <linux/namei.h> /* LOOKUP_FOLLOW */
33 #include <linux/sched/signal.h>
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>
41 #include <linux/memcontrol.h>
44 #include "../fdinfo.h"
46 #include <asm/ioctls.h>
48 /* configurable via /proc/sys/fs/inotify/ */
49 static int inotify_max_queued_events __read_mostly;
51 struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
55 #include <linux/sysctl.h>
59 struct ctl_table inotify_table[] = {
61 .procname = "max_user_instances",
62 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
63 .maxlen = sizeof(int),
65 .proc_handler = proc_dointvec_minmax,
69 .procname = "max_user_watches",
70 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
71 .maxlen = sizeof(int),
73 .proc_handler = proc_dointvec_minmax,
77 .procname = "max_queued_events",
78 .data = &inotify_max_queued_events,
79 .maxlen = sizeof(int),
81 .proc_handler = proc_dointvec_minmax,
86 #endif /* CONFIG_SYSCTL */
88 static inline __u32 inotify_arg_to_mask(u32 arg)
93 * everything should accept their own ignored, cares about children,
94 * and should receive events when the inode is unmounted
96 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT);
98 /* mask off the flags used to open the fd */
99 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
104 static inline u32 inotify_mask_to_arg(__u32 mask)
106 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
110 /* intofiy userspace file descriptor functions */
111 static __poll_t inotify_poll(struct file *file, poll_table *wait)
113 struct fsnotify_group *group = file->private_data;
116 poll_wait(file, &group->notification_waitq, wait);
117 spin_lock(&group->notification_lock);
118 if (!fsnotify_notify_queue_is_empty(group))
119 ret = EPOLLIN | EPOLLRDNORM;
120 spin_unlock(&group->notification_lock);
125 static int round_event_name_len(struct fsnotify_event *fsn_event)
127 struct inotify_event_info *event;
129 event = INOTIFY_E(fsn_event);
130 if (!event->name_len)
132 return roundup(event->name_len + 1, sizeof(struct inotify_event));
136 * Get an inotify_kernel_event if one exists and is small
137 * enough to fit in "count". Return an error pointer if
140 * Called with the group->notification_lock held.
142 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
145 size_t event_size = sizeof(struct inotify_event);
146 struct fsnotify_event *event;
148 if (fsnotify_notify_queue_is_empty(group))
151 event = fsnotify_peek_first_event(group);
153 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
155 event_size += round_event_name_len(event);
156 if (event_size > count)
157 return ERR_PTR(-EINVAL);
159 /* held the notification_lock the whole time, so this is the
160 * same event we peeked above */
161 fsnotify_remove_first_event(group);
167 * Copy an event to user space, returning how much we copied.
169 * We already checked that the event size is smaller than the
170 * buffer we had in "get_one_event()" above.
172 static ssize_t copy_event_to_user(struct fsnotify_group *group,
173 struct fsnotify_event *fsn_event,
176 struct inotify_event inotify_event;
177 struct inotify_event_info *event;
178 size_t event_size = sizeof(struct inotify_event);
182 pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
184 event = INOTIFY_E(fsn_event);
185 name_len = event->name_len;
187 * round up name length so it is a multiple of event_size
188 * plus an extra byte for the terminating '\0'.
190 pad_name_len = round_event_name_len(fsn_event);
191 inotify_event.len = pad_name_len;
192 inotify_event.mask = inotify_mask_to_arg(event->mask);
193 inotify_event.wd = event->wd;
194 inotify_event.cookie = event->sync_cookie;
196 /* send the main event */
197 if (copy_to_user(buf, &inotify_event, event_size))
203 * fsnotify only stores the pathname, so here we have to send the pathname
204 * and then pad that pathname out to a multiple of sizeof(inotify_event)
208 /* copy the path name */
209 if (copy_to_user(buf, event->name, name_len))
213 /* fill userspace with 0's */
214 if (clear_user(buf, pad_name_len - name_len))
216 event_size += pad_name_len;
222 static ssize_t inotify_read(struct file *file, char __user *buf,
223 size_t count, loff_t *pos)
225 struct fsnotify_group *group;
226 struct fsnotify_event *kevent;
229 DEFINE_WAIT_FUNC(wait, woken_wake_function);
232 group = file->private_data;
234 add_wait_queue(&group->notification_waitq, &wait);
236 spin_lock(&group->notification_lock);
237 kevent = get_one_event(group, count);
238 spin_unlock(&group->notification_lock);
240 pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
243 ret = PTR_ERR(kevent);
246 ret = copy_event_to_user(group, kevent, buf);
247 fsnotify_destroy_event(group, kevent);
256 if (file->f_flags & O_NONBLOCK)
259 if (signal_pending(current))
265 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
267 remove_wait_queue(&group->notification_waitq, &wait);
269 if (start != buf && ret != -EFAULT)
274 static int inotify_release(struct inode *ignored, struct file *file)
276 struct fsnotify_group *group = file->private_data;
278 pr_debug("%s: group=%p\n", __func__, group);
280 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
281 fsnotify_destroy_group(group);
286 static long inotify_ioctl(struct file *file, unsigned int cmd,
289 struct fsnotify_group *group;
290 struct fsnotify_event *fsn_event;
295 group = file->private_data;
296 p = (void __user *) arg;
298 pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
302 spin_lock(&group->notification_lock);
303 list_for_each_entry(fsn_event, &group->notification_list,
305 send_len += sizeof(struct inotify_event);
306 send_len += round_event_name_len(fsn_event);
308 spin_unlock(&group->notification_lock);
309 ret = put_user(send_len, (int __user *) p);
311 #ifdef CONFIG_CHECKPOINT_RESTORE
312 case INOTIFY_IOC_SETNEXTWD:
314 if (arg >= 1 && arg <= INT_MAX) {
315 struct inotify_group_private_data *data;
317 data = &group->inotify_data;
318 spin_lock(&data->idr_lock);
319 idr_set_cursor(&data->idr, (unsigned int)arg);
320 spin_unlock(&data->idr_lock);
324 #endif /* CONFIG_CHECKPOINT_RESTORE */
330 static const struct file_operations inotify_fops = {
331 .show_fdinfo = inotify_show_fdinfo,
332 .poll = inotify_poll,
333 .read = inotify_read,
334 .fasync = fsnotify_fasync,
335 .release = inotify_release,
336 .unlocked_ioctl = inotify_ioctl,
337 .compat_ioctl = inotify_ioctl,
338 .llseek = noop_llseek,
343 * find_inode - resolve a user-given path to a specific inode
345 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
349 error = user_path_at(AT_FDCWD, dirname, flags, path);
352 /* you can only watch an inode if you have read permissions on it */
353 error = inode_permission(path->dentry->d_inode, MAY_READ);
359 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
360 struct inotify_inode_mark *i_mark)
364 idr_preload(GFP_KERNEL);
367 ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
369 /* we added the mark to the idr, take a reference */
371 fsnotify_get_mark(&i_mark->fsn_mark);
374 spin_unlock(idr_lock);
376 return ret < 0 ? ret : 0;
379 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
382 struct idr *idr = &group->inotify_data.idr;
383 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
384 struct inotify_inode_mark *i_mark;
386 assert_spin_locked(idr_lock);
388 i_mark = idr_find(idr, wd);
390 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
392 fsnotify_get_mark(fsn_mark);
393 /* One ref for being in the idr, one ref we just took */
394 BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
400 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
403 struct inotify_inode_mark *i_mark;
404 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
407 i_mark = inotify_idr_find_locked(group, wd);
408 spin_unlock(idr_lock);
414 * Remove the mark from the idr (if present) and drop the reference
415 * on the mark because it was in the idr.
417 static void inotify_remove_from_idr(struct fsnotify_group *group,
418 struct inotify_inode_mark *i_mark)
420 struct idr *idr = &group->inotify_data.idr;
421 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
422 struct inotify_inode_mark *found_i_mark = NULL;
429 * does this i_mark think it is in the idr? we shouldn't get called
433 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
434 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
438 /* Lets look in the idr to see if we find it */
439 found_i_mark = inotify_idr_find_locked(group, wd);
440 if (unlikely(!found_i_mark)) {
441 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
442 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
447 * We found an mark in the idr at the right wd, but it's
448 * not the mark we were told to remove. eparis seriously
449 * fucked up somewhere.
451 if (unlikely(found_i_mark != i_mark)) {
452 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
453 "found_i_mark=%p found_i_mark->wd=%d "
454 "found_i_mark->group=%p\n", __func__, i_mark,
455 i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
456 found_i_mark->wd, found_i_mark->fsn_mark.group);
461 * One ref for being in the idr
462 * one ref grabbed by inotify_idr_find
464 if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
465 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
466 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
467 /* we can't really recover with bad ref cnting.. */
472 /* Removed from the idr, drop that ref. */
473 fsnotify_put_mark(&i_mark->fsn_mark);
476 spin_unlock(idr_lock);
477 /* match the ref taken by inotify_idr_find_locked() */
479 fsnotify_put_mark(&found_i_mark->fsn_mark);
483 * Send IN_IGNORED for this wd, remove this wd from the idr.
485 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
486 struct fsnotify_group *group)
488 struct inotify_inode_mark *i_mark;
489 struct fsnotify_iter_info iter_info = { };
491 fsnotify_iter_set_report_type_mark(&iter_info, FSNOTIFY_OBJ_TYPE_INODE,
494 /* Queue ignore event for the watch */
495 inotify_handle_event(group, NULL, FS_IN_IGNORED, NULL,
496 FSNOTIFY_EVENT_NONE, NULL, 0, &iter_info);
498 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
499 /* remove this mark from the idr */
500 inotify_remove_from_idr(group, i_mark);
502 dec_inotify_watches(group->inotify_data.ucounts);
505 static int inotify_update_existing_watch(struct fsnotify_group *group,
509 struct fsnotify_mark *fsn_mark;
510 struct inotify_inode_mark *i_mark;
511 __u32 old_mask, new_mask;
513 int add = (arg & IN_MASK_ADD);
514 int create = (arg & IN_MASK_CREATE);
517 mask = inotify_arg_to_mask(arg);
519 fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
525 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
527 spin_lock(&fsn_mark->lock);
528 old_mask = fsn_mark->mask;
530 fsn_mark->mask |= mask;
532 fsn_mark->mask = mask;
533 new_mask = fsn_mark->mask;
534 spin_unlock(&fsn_mark->lock);
536 if (old_mask != new_mask) {
537 /* more bits in old than in new? */
538 int dropped = (old_mask & ~new_mask);
539 /* more bits in this fsn_mark than the inode's mask? */
540 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
542 /* update the inode with this new fsn_mark */
543 if (dropped || do_inode)
544 fsnotify_recalc_mask(inode->i_fsnotify_marks);
551 /* match the get from fsnotify_find_mark() */
552 fsnotify_put_mark(fsn_mark);
557 static int inotify_new_watch(struct fsnotify_group *group,
561 struct inotify_inode_mark *tmp_i_mark;
564 struct idr *idr = &group->inotify_data.idr;
565 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
567 mask = inotify_arg_to_mask(arg);
569 tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
570 if (unlikely(!tmp_i_mark))
573 fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
574 tmp_i_mark->fsn_mark.mask = mask;
577 ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
581 /* increment the number of watches the user has */
582 if (!inc_inotify_watches(group->inotify_data.ucounts)) {
583 inotify_remove_from_idr(group, tmp_i_mark);
588 /* we are on the idr, now get on the inode */
589 ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
591 /* we failed to get on the inode, get off the idr */
592 inotify_remove_from_idr(group, tmp_i_mark);
597 /* return the watch descriptor for this new mark */
598 ret = tmp_i_mark->wd;
601 /* match the ref from fsnotify_init_mark() */
602 fsnotify_put_mark(&tmp_i_mark->fsn_mark);
607 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
611 mutex_lock(&group->mark_mutex);
612 /* try to update and existing watch with the new arg */
613 ret = inotify_update_existing_watch(group, inode, arg);
614 /* no mark present, try to add a new one */
616 ret = inotify_new_watch(group, inode, arg);
617 mutex_unlock(&group->mark_mutex);
622 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
624 struct fsnotify_group *group;
625 struct inotify_event_info *oevent;
627 group = fsnotify_alloc_group(&inotify_fsnotify_ops);
631 oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL);
632 if (unlikely(!oevent)) {
633 fsnotify_destroy_group(group);
634 return ERR_PTR(-ENOMEM);
636 group->overflow_event = &oevent->fse;
637 fsnotify_init_event(group->overflow_event, NULL);
638 oevent->mask = FS_Q_OVERFLOW;
640 oevent->sync_cookie = 0;
641 oevent->name_len = 0;
643 group->max_events = max_events;
644 group->memcg = get_mem_cgroup_from_mm(current->mm);
646 spin_lock_init(&group->inotify_data.idr_lock);
647 idr_init(&group->inotify_data.idr);
648 group->inotify_data.ucounts = inc_ucount(current_user_ns(),
650 UCOUNT_INOTIFY_INSTANCES);
652 if (!group->inotify_data.ucounts) {
653 fsnotify_destroy_group(group);
654 return ERR_PTR(-EMFILE);
661 /* inotify syscalls */
662 static int do_inotify_init(int flags)
664 struct fsnotify_group *group;
667 /* Check the IN_* constants for consistency. */
668 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
669 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
671 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
674 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
675 group = inotify_new_group(inotify_max_queued_events);
677 return PTR_ERR(group);
679 ret = anon_inode_getfd("inotify", &inotify_fops, group,
682 fsnotify_destroy_group(group);
687 SYSCALL_DEFINE1(inotify_init1, int, flags)
689 return do_inotify_init(flags);
692 SYSCALL_DEFINE0(inotify_init)
694 return do_inotify_init(0);
697 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
700 struct fsnotify_group *group;
708 * We share a lot of code with fs/dnotify. We also share
709 * the bit layout between inotify's IN_* and the fsnotify
710 * FS_*. This check ensures that only the inotify IN_*
711 * bits get passed in and set in watches/events.
713 if (unlikely(mask & ~ALL_INOTIFY_BITS))
716 * Require at least one valid bit set in the mask.
717 * Without _something_ set, we would have no events to
720 if (unlikely(!(mask & ALL_INOTIFY_BITS)))
724 if (unlikely(!f.file))
727 /* IN_MASK_ADD and IN_MASK_CREATE don't make sense together */
728 if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
733 /* verify that this is indeed an inotify instance */
734 if (unlikely(f.file->f_op != &inotify_fops)) {
739 if (!(mask & IN_DONT_FOLLOW))
740 flags |= LOOKUP_FOLLOW;
741 if (mask & IN_ONLYDIR)
742 flags |= LOOKUP_DIRECTORY;
744 ret = inotify_find_inode(pathname, &path, flags);
748 /* inode held in place by reference to path; group by fget on fd */
749 inode = path.dentry->d_inode;
750 group = f.file->private_data;
752 /* create/update an inode mark */
753 ret = inotify_update_watch(group, inode, mask);
760 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
762 struct fsnotify_group *group;
763 struct inotify_inode_mark *i_mark;
768 if (unlikely(!f.file))
771 /* verify that this is indeed an inotify instance */
773 if (unlikely(f.file->f_op != &inotify_fops))
776 group = f.file->private_data;
779 i_mark = inotify_idr_find(group, wd);
780 if (unlikely(!i_mark))
785 fsnotify_destroy_mark(&i_mark->fsn_mark, group);
787 /* match ref taken by inotify_idr_find */
788 fsnotify_put_mark(&i_mark->fsn_mark);
796 * inotify_user_setup - Our initialization function. Note that we cannot return
797 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
798 * must result in panic().
800 static int __init inotify_user_setup(void)
802 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
803 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
804 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
805 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
806 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
807 BUILD_BUG_ON(IN_OPEN != FS_OPEN);
808 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
809 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
810 BUILD_BUG_ON(IN_CREATE != FS_CREATE);
811 BUILD_BUG_ON(IN_DELETE != FS_DELETE);
812 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
813 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
814 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
815 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
816 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
817 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
818 BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
819 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
821 BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
823 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
824 SLAB_PANIC|SLAB_ACCOUNT);
826 inotify_max_queued_events = 16384;
827 init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
828 init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = 8192;
832 fs_initcall(inotify_user_setup);