1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * fs/inotify_user.c - inotify support for userspace
6 * John McCutchan <ttb@tentacle.dhs.org>
7 * Robert Love <rml@novell.com>
9 * Copyright (C) 2005 John McCutchan
10 * Copyright 2006 Hewlett-Packard Development Company, L.P.
12 * Copyright (C) 2009 Eric Paris <Red Hat Inc>
13 * inotify was largely rewriten to make use of the fsnotify infrastructure
16 #include <linux/file.h>
17 #include <linux/fs.h> /* struct inode */
18 #include <linux/fsnotify_backend.h>
19 #include <linux/idr.h>
20 #include <linux/init.h> /* fs_initcall */
21 #include <linux/inotify.h>
22 #include <linux/kernel.h> /* roundup() */
23 #include <linux/namei.h> /* LOOKUP_FOLLOW */
24 #include <linux/sched/signal.h>
25 #include <linux/slab.h> /* struct kmem_cache */
26 #include <linux/syscalls.h>
27 #include <linux/types.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/uaccess.h>
30 #include <linux/poll.h>
31 #include <linux/wait.h>
32 #include <linux/memcontrol.h>
33 #include <linux/security.h>
36 #include "../fdinfo.h"
38 #include <asm/ioctls.h>
41 * An inotify watch requires allocating an inotify_inode_mark structure as
42 * well as pinning the watched inode. Doubling the size of a VFS inode
43 * should be more than enough to cover the additional filesystem inode
46 #define INOTIFY_WATCH_COST (sizeof(struct inotify_inode_mark) + \
47 2 * sizeof(struct inode))
49 /* configurable via /proc/sys/fs/inotify/ */
50 static int inotify_max_queued_events __read_mostly;
52 struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
56 #include <linux/sysctl.h>
58 struct ctl_table inotify_table[] = {
60 .procname = "max_user_instances",
61 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
62 .maxlen = sizeof(int),
64 .proc_handler = proc_dointvec_minmax,
65 .extra1 = SYSCTL_ZERO,
68 .procname = "max_user_watches",
69 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
70 .maxlen = sizeof(int),
72 .proc_handler = proc_dointvec_minmax,
73 .extra1 = SYSCTL_ZERO,
76 .procname = "max_queued_events",
77 .data = &inotify_max_queued_events,
78 .maxlen = sizeof(int),
80 .proc_handler = proc_dointvec_minmax,
85 #endif /* CONFIG_SYSCTL */
87 static inline __u32 inotify_arg_to_mask(struct inode *inode, u32 arg)
92 * Everything should accept their own ignored and should receive events
93 * when the inode is unmounted. All directories care about children.
95 mask = (FS_IN_IGNORED | FS_UNMOUNT);
96 if (S_ISDIR(inode->i_mode))
97 mask |= FS_EVENT_ON_CHILD;
99 /* mask off the flags used to open the fd */
100 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
105 static inline u32 inotify_mask_to_arg(__u32 mask)
107 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
111 /* intofiy userspace file descriptor functions */
112 static __poll_t inotify_poll(struct file *file, poll_table *wait)
114 struct fsnotify_group *group = file->private_data;
117 poll_wait(file, &group->notification_waitq, wait);
118 spin_lock(&group->notification_lock);
119 if (!fsnotify_notify_queue_is_empty(group))
120 ret = EPOLLIN | EPOLLRDNORM;
121 spin_unlock(&group->notification_lock);
126 static int round_event_name_len(struct fsnotify_event *fsn_event)
128 struct inotify_event_info *event;
130 event = INOTIFY_E(fsn_event);
131 if (!event->name_len)
133 return roundup(event->name_len + 1, sizeof(struct inotify_event));
137 * Get an inotify_kernel_event if one exists and is small
138 * enough to fit in "count". Return an error pointer if
141 * Called with the group->notification_lock held.
143 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
146 size_t event_size = sizeof(struct inotify_event);
147 struct fsnotify_event *event;
149 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,
346 unsigned int flags, __u64 mask)
350 error = user_path_at(AT_FDCWD, dirname, flags, path);
353 /* you can only watch an inode if you have read permissions on it */
354 error = path_permission(path, MAY_READ);
359 error = security_path_notify(path, mask,
360 FSNOTIFY_OBJ_TYPE_INODE);
367 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
368 struct inotify_inode_mark *i_mark)
372 idr_preload(GFP_KERNEL);
375 ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
377 /* we added the mark to the idr, take a reference */
379 fsnotify_get_mark(&i_mark->fsn_mark);
382 spin_unlock(idr_lock);
384 return ret < 0 ? ret : 0;
387 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
390 struct idr *idr = &group->inotify_data.idr;
391 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
392 struct inotify_inode_mark *i_mark;
394 assert_spin_locked(idr_lock);
396 i_mark = idr_find(idr, wd);
398 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
400 fsnotify_get_mark(fsn_mark);
401 /* One ref for being in the idr, one ref we just took */
402 BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
408 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
411 struct inotify_inode_mark *i_mark;
412 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
415 i_mark = inotify_idr_find_locked(group, wd);
416 spin_unlock(idr_lock);
422 * Remove the mark from the idr (if present) and drop the reference
423 * on the mark because it was in the idr.
425 static void inotify_remove_from_idr(struct fsnotify_group *group,
426 struct inotify_inode_mark *i_mark)
428 struct idr *idr = &group->inotify_data.idr;
429 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
430 struct inotify_inode_mark *found_i_mark = NULL;
437 * does this i_mark think it is in the idr? we shouldn't get called
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);
446 /* Lets look in the idr to see if we find it */
447 found_i_mark = inotify_idr_find_locked(group, wd);
448 if (unlikely(!found_i_mark)) {
449 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
450 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
455 * We found an mark in the idr at the right wd, but it's
456 * not the mark we were told to remove. eparis seriously
457 * fucked up somewhere.
459 if (unlikely(found_i_mark != i_mark)) {
460 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
461 "found_i_mark=%p found_i_mark->wd=%d "
462 "found_i_mark->group=%p\n", __func__, i_mark,
463 i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
464 found_i_mark->wd, found_i_mark->fsn_mark.group);
469 * One ref for being in the idr
470 * one ref grabbed by inotify_idr_find
472 if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
473 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
474 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
475 /* we can't really recover with bad ref cnting.. */
480 /* Removed from the idr, drop that ref. */
481 fsnotify_put_mark(&i_mark->fsn_mark);
484 spin_unlock(idr_lock);
485 /* match the ref taken by inotify_idr_find_locked() */
487 fsnotify_put_mark(&found_i_mark->fsn_mark);
491 * Send IN_IGNORED for this wd, remove this wd from the idr.
493 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
494 struct fsnotify_group *group)
496 struct inotify_inode_mark *i_mark;
498 /* Queue ignore event for the watch */
499 inotify_handle_inode_event(fsn_mark, FS_IN_IGNORED, NULL, NULL, NULL,
502 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
503 /* remove this mark from the idr */
504 inotify_remove_from_idr(group, i_mark);
506 dec_inotify_watches(group->inotify_data.ucounts);
509 static int inotify_update_existing_watch(struct fsnotify_group *group,
513 struct fsnotify_mark *fsn_mark;
514 struct inotify_inode_mark *i_mark;
515 __u32 old_mask, new_mask;
517 int add = (arg & IN_MASK_ADD);
518 int create = (arg & IN_MASK_CREATE);
521 mask = inotify_arg_to_mask(inode, arg);
523 fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
531 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
533 spin_lock(&fsn_mark->lock);
534 old_mask = fsn_mark->mask;
536 fsn_mark->mask |= mask;
538 fsn_mark->mask = mask;
539 new_mask = fsn_mark->mask;
540 spin_unlock(&fsn_mark->lock);
542 if (old_mask != new_mask) {
543 /* more bits in old than in new? */
544 int dropped = (old_mask & ~new_mask);
545 /* more bits in this fsn_mark than the inode's mask? */
546 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
548 /* update the inode with this new fsn_mark */
549 if (dropped || do_inode)
550 fsnotify_recalc_mask(inode->i_fsnotify_marks);
558 /* match the get from fsnotify_find_mark() */
559 fsnotify_put_mark(fsn_mark);
564 static int inotify_new_watch(struct fsnotify_group *group,
568 struct inotify_inode_mark *tmp_i_mark;
571 struct idr *idr = &group->inotify_data.idr;
572 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
574 mask = inotify_arg_to_mask(inode, arg);
576 tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
577 if (unlikely(!tmp_i_mark))
580 fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
581 tmp_i_mark->fsn_mark.mask = mask;
584 ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
588 /* increment the number of watches the user has */
589 if (!inc_inotify_watches(group->inotify_data.ucounts)) {
590 inotify_remove_from_idr(group, tmp_i_mark);
595 /* we are on the idr, now get on the inode */
596 ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
598 /* we failed to get on the inode, get off the idr */
599 inotify_remove_from_idr(group, tmp_i_mark);
604 /* return the watch descriptor for this new mark */
605 ret = tmp_i_mark->wd;
608 /* match the ref from fsnotify_init_mark() */
609 fsnotify_put_mark(&tmp_i_mark->fsn_mark);
614 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
618 mutex_lock(&group->mark_mutex);
619 /* try to update and existing watch with the new arg */
620 ret = inotify_update_existing_watch(group, inode, arg);
621 /* no mark present, try to add a new one */
623 ret = inotify_new_watch(group, inode, arg);
624 mutex_unlock(&group->mark_mutex);
629 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
631 struct fsnotify_group *group;
632 struct inotify_event_info *oevent;
634 group = fsnotify_alloc_user_group(&inotify_fsnotify_ops);
638 oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL_ACCOUNT);
639 if (unlikely(!oevent)) {
640 fsnotify_destroy_group(group);
641 return ERR_PTR(-ENOMEM);
643 group->overflow_event = &oevent->fse;
644 fsnotify_init_event(group->overflow_event);
645 oevent->mask = FS_Q_OVERFLOW;
647 oevent->sync_cookie = 0;
648 oevent->name_len = 0;
650 group->max_events = max_events;
651 group->memcg = get_mem_cgroup_from_mm(current->mm);
653 spin_lock_init(&group->inotify_data.idr_lock);
654 idr_init(&group->inotify_data.idr);
655 group->inotify_data.ucounts = inc_ucount(current_user_ns(),
657 UCOUNT_INOTIFY_INSTANCES);
659 if (!group->inotify_data.ucounts) {
660 fsnotify_destroy_group(group);
661 return ERR_PTR(-EMFILE);
668 /* inotify syscalls */
669 static int do_inotify_init(int flags)
671 struct fsnotify_group *group;
674 /* Check the IN_* constants for consistency. */
675 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
676 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
678 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
681 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
682 group = inotify_new_group(inotify_max_queued_events);
684 return PTR_ERR(group);
686 ret = anon_inode_getfd("inotify", &inotify_fops, group,
689 fsnotify_destroy_group(group);
694 SYSCALL_DEFINE1(inotify_init1, int, flags)
696 return do_inotify_init(flags);
699 SYSCALL_DEFINE0(inotify_init)
701 return do_inotify_init(0);
704 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
707 struct fsnotify_group *group;
715 * We share a lot of code with fs/dnotify. We also share
716 * the bit layout between inotify's IN_* and the fsnotify
717 * FS_*. This check ensures that only the inotify IN_*
718 * bits get passed in and set in watches/events.
720 if (unlikely(mask & ~ALL_INOTIFY_BITS))
723 * Require at least one valid bit set in the mask.
724 * Without _something_ set, we would have no events to
727 if (unlikely(!(mask & ALL_INOTIFY_BITS)))
731 if (unlikely(!f.file))
734 /* IN_MASK_ADD and IN_MASK_CREATE don't make sense together */
735 if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
740 /* verify that this is indeed an inotify instance */
741 if (unlikely(f.file->f_op != &inotify_fops)) {
746 if (!(mask & IN_DONT_FOLLOW))
747 flags |= LOOKUP_FOLLOW;
748 if (mask & IN_ONLYDIR)
749 flags |= LOOKUP_DIRECTORY;
751 ret = inotify_find_inode(pathname, &path, flags,
752 (mask & IN_ALL_EVENTS));
756 /* inode held in place by reference to path; group by fget on fd */
757 inode = path.dentry->d_inode;
758 group = f.file->private_data;
760 /* create/update an inode mark */
761 ret = inotify_update_watch(group, inode, mask);
768 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
770 struct fsnotify_group *group;
771 struct inotify_inode_mark *i_mark;
776 if (unlikely(!f.file))
779 /* verify that this is indeed an inotify instance */
780 if (unlikely(f.file->f_op != &inotify_fops))
783 group = f.file->private_data;
785 i_mark = inotify_idr_find(group, wd);
786 if (unlikely(!i_mark))
791 fsnotify_destroy_mark(&i_mark->fsn_mark, group);
793 /* match ref taken by inotify_idr_find */
794 fsnotify_put_mark(&i_mark->fsn_mark);
802 * inotify_user_setup - Our initialization function. Note that we cannot return
803 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
804 * must result in panic().
806 static int __init inotify_user_setup(void)
808 unsigned long watches_max;
813 * Allow up to 1% of addressable memory to be allocated for inotify
814 * watches (per user) limited to the range [8192, 1048576].
816 watches_max = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
818 watches_max = clamp(watches_max, 8192UL, 1048576UL);
820 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
821 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
822 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
823 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
824 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
825 BUILD_BUG_ON(IN_OPEN != FS_OPEN);
826 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
827 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
828 BUILD_BUG_ON(IN_CREATE != FS_CREATE);
829 BUILD_BUG_ON(IN_DELETE != FS_DELETE);
830 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
831 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
832 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
833 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
834 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
835 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
836 BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
837 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
839 BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
841 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
842 SLAB_PANIC|SLAB_ACCOUNT);
844 inotify_max_queued_events = 16384;
845 init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
846 init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = watches_max;
850 fs_initcall(inotify_user_setup);