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 static long it_zero = 0;
59 static long it_int_max = INT_MAX;
61 static struct ctl_table inotify_table[] = {
63 .procname = "max_user_instances",
64 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
65 .maxlen = sizeof(long),
67 .proc_handler = proc_doulongvec_minmax,
69 .extra2 = &it_int_max,
72 .procname = "max_user_watches",
73 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
74 .maxlen = sizeof(long),
76 .proc_handler = proc_doulongvec_minmax,
78 .extra2 = &it_int_max,
81 .procname = "max_queued_events",
82 .data = &inotify_max_queued_events,
83 .maxlen = sizeof(int),
85 .proc_handler = proc_dointvec_minmax,
91 static void __init inotify_sysctls_init(void)
93 register_sysctl("fs/inotify", inotify_table);
97 #define inotify_sysctls_init() do { } while (0)
98 #endif /* CONFIG_SYSCTL */
100 static inline __u32 inotify_arg_to_mask(struct inode *inode, u32 arg)
105 * Everything should receive events when the inode is unmounted.
106 * All directories care about children.
109 if (S_ISDIR(inode->i_mode))
110 mask |= FS_EVENT_ON_CHILD;
112 /* mask off the flags used to open the fd */
113 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
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 __poll_t 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 spin_lock(&group->notification_lock);
132 if (!fsnotify_notify_queue_is_empty(group))
133 ret = EPOLLIN | EPOLLRDNORM;
134 spin_unlock(&group->notification_lock);
139 static int round_event_name_len(struct fsnotify_event *fsn_event)
141 struct inotify_event_info *event;
143 event = INOTIFY_E(fsn_event);
144 if (!event->name_len)
146 return roundup(event->name_len + 1, sizeof(struct inotify_event));
150 * Get an inotify_kernel_event if one exists and is small
151 * enough to fit in "count". Return an error pointer if
154 * Called with the group->notification_lock held.
156 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
159 size_t event_size = sizeof(struct inotify_event);
160 struct fsnotify_event *event;
162 event = fsnotify_peek_first_event(group);
166 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
168 event_size += round_event_name_len(event);
169 if (event_size > count)
170 return ERR_PTR(-EINVAL);
172 /* held the notification_lock the whole time, so this is the
173 * same event we peeked above */
174 fsnotify_remove_first_event(group);
180 * Copy an event to user space, returning how much we copied.
182 * We already checked that the event size is smaller than the
183 * buffer we had in "get_one_event()" above.
185 static ssize_t copy_event_to_user(struct fsnotify_group *group,
186 struct fsnotify_event *fsn_event,
189 struct inotify_event inotify_event;
190 struct inotify_event_info *event;
191 size_t event_size = sizeof(struct inotify_event);
195 pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
197 event = INOTIFY_E(fsn_event);
198 name_len = event->name_len;
200 * round up name length so it is a multiple of event_size
201 * plus an extra byte for the terminating '\0'.
203 pad_name_len = round_event_name_len(fsn_event);
204 inotify_event.len = pad_name_len;
205 inotify_event.mask = inotify_mask_to_arg(event->mask);
206 inotify_event.wd = event->wd;
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)
221 /* copy the path name */
222 if (copy_to_user(buf, event->name, name_len))
226 /* fill userspace with 0's */
227 if (clear_user(buf, pad_name_len - name_len))
229 event_size += pad_name_len;
235 static ssize_t inotify_read(struct file *file, char __user *buf,
236 size_t count, loff_t *pos)
238 struct fsnotify_group *group;
239 struct fsnotify_event *kevent;
242 DEFINE_WAIT_FUNC(wait, woken_wake_function);
245 group = file->private_data;
247 add_wait_queue(&group->notification_waitq, &wait);
249 spin_lock(&group->notification_lock);
250 kevent = get_one_event(group, count);
251 spin_unlock(&group->notification_lock);
253 pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
256 ret = PTR_ERR(kevent);
259 ret = copy_event_to_user(group, kevent, buf);
260 fsnotify_destroy_event(group, kevent);
269 if (file->f_flags & O_NONBLOCK)
272 if (signal_pending(current))
278 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
280 remove_wait_queue(&group->notification_waitq, &wait);
282 if (start != buf && ret != -EFAULT)
287 static int inotify_release(struct inode *ignored, struct file *file)
289 struct fsnotify_group *group = file->private_data;
291 pr_debug("%s: group=%p\n", __func__, group);
293 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
294 fsnotify_destroy_group(group);
299 static long inotify_ioctl(struct file *file, unsigned int cmd,
302 struct fsnotify_group *group;
303 struct fsnotify_event *fsn_event;
308 group = file->private_data;
309 p = (void __user *) arg;
311 pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
315 spin_lock(&group->notification_lock);
316 list_for_each_entry(fsn_event, &group->notification_list,
318 send_len += sizeof(struct inotify_event);
319 send_len += round_event_name_len(fsn_event);
321 spin_unlock(&group->notification_lock);
322 ret = put_user(send_len, (int __user *) p);
324 #ifdef CONFIG_CHECKPOINT_RESTORE
325 case INOTIFY_IOC_SETNEXTWD:
327 if (arg >= 1 && arg <= INT_MAX) {
328 struct inotify_group_private_data *data;
330 data = &group->inotify_data;
331 spin_lock(&data->idr_lock);
332 idr_set_cursor(&data->idr, (unsigned int)arg);
333 spin_unlock(&data->idr_lock);
337 #endif /* CONFIG_CHECKPOINT_RESTORE */
343 static const struct file_operations inotify_fops = {
344 .show_fdinfo = inotify_show_fdinfo,
345 .poll = inotify_poll,
346 .read = inotify_read,
347 .fasync = fsnotify_fasync,
348 .release = inotify_release,
349 .unlocked_ioctl = inotify_ioctl,
350 .compat_ioctl = inotify_ioctl,
351 .llseek = noop_llseek,
356 * find_inode - resolve a user-given path to a specific inode
358 static int inotify_find_inode(const char __user *dirname, struct path *path,
359 unsigned int flags, __u64 mask)
363 error = user_path_at(AT_FDCWD, dirname, flags, path);
366 /* you can only watch an inode if you have read permissions on it */
367 error = path_permission(path, MAY_READ);
372 error = security_path_notify(path, mask,
373 FSNOTIFY_OBJ_TYPE_INODE);
380 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
381 struct inotify_inode_mark *i_mark)
385 idr_preload(GFP_KERNEL);
388 ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
390 /* we added the mark to the idr, take a reference */
392 fsnotify_get_mark(&i_mark->fsn_mark);
395 spin_unlock(idr_lock);
397 return ret < 0 ? ret : 0;
400 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
403 struct idr *idr = &group->inotify_data.idr;
404 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
405 struct inotify_inode_mark *i_mark;
407 assert_spin_locked(idr_lock);
409 i_mark = idr_find(idr, wd);
411 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
413 fsnotify_get_mark(fsn_mark);
414 /* One ref for being in the idr, one ref we just took */
415 BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
421 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
424 struct inotify_inode_mark *i_mark;
425 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
428 i_mark = inotify_idr_find_locked(group, wd);
429 spin_unlock(idr_lock);
435 * Remove the mark from the idr (if present) and drop the reference
436 * on the mark because it was in the idr.
438 static void inotify_remove_from_idr(struct fsnotify_group *group,
439 struct inotify_inode_mark *i_mark)
441 struct idr *idr = &group->inotify_data.idr;
442 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
443 struct inotify_inode_mark *found_i_mark = NULL;
450 * does this i_mark think it is in the idr? we shouldn't get called
454 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
455 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
459 /* Lets look in the idr to see if we find it */
460 found_i_mark = inotify_idr_find_locked(group, wd);
461 if (unlikely(!found_i_mark)) {
462 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
463 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
468 * We found an mark in the idr at the right wd, but it's
469 * not the mark we were told to remove. eparis seriously
470 * fucked up somewhere.
472 if (unlikely(found_i_mark != i_mark)) {
473 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
474 "found_i_mark=%p found_i_mark->wd=%d "
475 "found_i_mark->group=%p\n", __func__, i_mark,
476 i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
477 found_i_mark->wd, found_i_mark->fsn_mark.group);
482 * One ref for being in the idr
483 * one ref grabbed by inotify_idr_find
485 if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
486 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
487 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
488 /* we can't really recover with bad ref cnting.. */
493 /* Removed from the idr, drop that ref. */
494 fsnotify_put_mark(&i_mark->fsn_mark);
497 spin_unlock(idr_lock);
498 /* match the ref taken by inotify_idr_find_locked() */
500 fsnotify_put_mark(&found_i_mark->fsn_mark);
504 * Send IN_IGNORED for this wd, remove this wd from the idr.
506 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
507 struct fsnotify_group *group)
509 struct inotify_inode_mark *i_mark;
511 /* Queue ignore event for the watch */
512 inotify_handle_inode_event(fsn_mark, FS_IN_IGNORED, NULL, NULL, NULL,
515 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
516 /* remove this mark from the idr */
517 inotify_remove_from_idr(group, i_mark);
519 dec_inotify_watches(group->inotify_data.ucounts);
522 static int inotify_update_existing_watch(struct fsnotify_group *group,
526 struct fsnotify_mark *fsn_mark;
527 struct inotify_inode_mark *i_mark;
528 __u32 old_mask, new_mask;
530 int add = (arg & IN_MASK_ADD);
531 int create = (arg & IN_MASK_CREATE);
534 mask = inotify_arg_to_mask(inode, arg);
536 fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
544 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
546 spin_lock(&fsn_mark->lock);
547 old_mask = fsn_mark->mask;
549 fsn_mark->mask |= mask;
551 fsn_mark->mask = mask;
552 new_mask = fsn_mark->mask;
553 spin_unlock(&fsn_mark->lock);
555 if (old_mask != new_mask) {
556 /* more bits in old than in new? */
557 int dropped = (old_mask & ~new_mask);
558 /* more bits in this fsn_mark than the inode's mask? */
559 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
561 /* update the inode with this new fsn_mark */
562 if (dropped || do_inode)
563 fsnotify_recalc_mask(inode->i_fsnotify_marks);
571 /* match the get from fsnotify_find_mark() */
572 fsnotify_put_mark(fsn_mark);
577 static int inotify_new_watch(struct fsnotify_group *group,
581 struct inotify_inode_mark *tmp_i_mark;
584 struct idr *idr = &group->inotify_data.idr;
585 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
587 mask = inotify_arg_to_mask(inode, arg);
589 tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
590 if (unlikely(!tmp_i_mark))
593 fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
594 tmp_i_mark->fsn_mark.mask = mask;
597 ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
601 /* increment the number of watches the user has */
602 if (!inc_inotify_watches(group->inotify_data.ucounts)) {
603 inotify_remove_from_idr(group, tmp_i_mark);
608 /* we are on the idr, now get on the inode */
609 ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
611 /* we failed to get on the inode, get off the idr */
612 inotify_remove_from_idr(group, tmp_i_mark);
617 /* return the watch descriptor for this new mark */
618 ret = tmp_i_mark->wd;
621 /* match the ref from fsnotify_init_mark() */
622 fsnotify_put_mark(&tmp_i_mark->fsn_mark);
627 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
631 mutex_lock(&group->mark_mutex);
632 /* try to update and existing watch with the new arg */
633 ret = inotify_update_existing_watch(group, inode, arg);
634 /* no mark present, try to add a new one */
636 ret = inotify_new_watch(group, inode, arg);
637 mutex_unlock(&group->mark_mutex);
642 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
644 struct fsnotify_group *group;
645 struct inotify_event_info *oevent;
647 group = fsnotify_alloc_user_group(&inotify_fsnotify_ops);
651 oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL_ACCOUNT);
652 if (unlikely(!oevent)) {
653 fsnotify_destroy_group(group);
654 return ERR_PTR(-ENOMEM);
656 group->overflow_event = &oevent->fse;
657 fsnotify_init_event(group->overflow_event);
658 oevent->mask = FS_Q_OVERFLOW;
660 oevent->sync_cookie = 0;
661 oevent->name_len = 0;
663 group->max_events = max_events;
664 group->memcg = get_mem_cgroup_from_mm(current->mm);
666 spin_lock_init(&group->inotify_data.idr_lock);
667 idr_init(&group->inotify_data.idr);
668 group->inotify_data.ucounts = inc_ucount(current_user_ns(),
670 UCOUNT_INOTIFY_INSTANCES);
672 if (!group->inotify_data.ucounts) {
673 fsnotify_destroy_group(group);
674 return ERR_PTR(-EMFILE);
681 /* inotify syscalls */
682 static int do_inotify_init(int flags)
684 struct fsnotify_group *group;
687 /* Check the IN_* constants for consistency. */
688 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
689 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
691 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
694 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
695 group = inotify_new_group(inotify_max_queued_events);
697 return PTR_ERR(group);
699 ret = anon_inode_getfd("inotify", &inotify_fops, group,
702 fsnotify_destroy_group(group);
707 SYSCALL_DEFINE1(inotify_init1, int, flags)
709 return do_inotify_init(flags);
712 SYSCALL_DEFINE0(inotify_init)
714 return do_inotify_init(0);
717 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
720 struct fsnotify_group *group;
728 * We share a lot of code with fs/dnotify. We also share
729 * the bit layout between inotify's IN_* and the fsnotify
730 * FS_*. This check ensures that only the inotify IN_*
731 * bits get passed in and set in watches/events.
733 if (unlikely(mask & ~ALL_INOTIFY_BITS))
736 * Require at least one valid bit set in the mask.
737 * Without _something_ set, we would have no events to
740 if (unlikely(!(mask & ALL_INOTIFY_BITS)))
744 if (unlikely(!f.file))
747 /* IN_MASK_ADD and IN_MASK_CREATE don't make sense together */
748 if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
753 /* verify that this is indeed an inotify instance */
754 if (unlikely(f.file->f_op != &inotify_fops)) {
759 if (!(mask & IN_DONT_FOLLOW))
760 flags |= LOOKUP_FOLLOW;
761 if (mask & IN_ONLYDIR)
762 flags |= LOOKUP_DIRECTORY;
764 ret = inotify_find_inode(pathname, &path, flags,
765 (mask & IN_ALL_EVENTS));
769 /* inode held in place by reference to path; group by fget on fd */
770 inode = path.dentry->d_inode;
771 group = f.file->private_data;
773 /* create/update an inode mark */
774 ret = inotify_update_watch(group, inode, mask);
781 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
783 struct fsnotify_group *group;
784 struct inotify_inode_mark *i_mark;
789 if (unlikely(!f.file))
792 /* verify that this is indeed an inotify instance */
793 if (unlikely(f.file->f_op != &inotify_fops))
796 group = f.file->private_data;
798 i_mark = inotify_idr_find(group, wd);
799 if (unlikely(!i_mark))
804 fsnotify_destroy_mark(&i_mark->fsn_mark, group);
806 /* match ref taken by inotify_idr_find */
807 fsnotify_put_mark(&i_mark->fsn_mark);
815 * inotify_user_setup - Our initialization function. Note that we cannot return
816 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
817 * must result in panic().
819 static int __init inotify_user_setup(void)
821 unsigned long watches_max;
826 * Allow up to 1% of addressable memory to be allocated for inotify
827 * watches (per user) limited to the range [8192, 1048576].
829 watches_max = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
831 watches_max = clamp(watches_max, 8192UL, 1048576UL);
833 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
834 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
835 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
836 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
837 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
838 BUILD_BUG_ON(IN_OPEN != FS_OPEN);
839 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
840 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
841 BUILD_BUG_ON(IN_CREATE != FS_CREATE);
842 BUILD_BUG_ON(IN_DELETE != FS_DELETE);
843 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
844 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
845 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
846 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
847 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
848 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
849 BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
850 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
852 BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
854 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
855 SLAB_PANIC|SLAB_ACCOUNT);
857 inotify_max_queued_events = 16384;
858 init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
859 init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = watches_max;
860 inotify_sysctls_init();
864 fs_initcall(inotify_user_setup);