1 #include <linux/fanotify.h>
2 #include <linux/fcntl.h>
3 #include <linux/file.h>
5 #include <linux/anon_inodes.h>
6 #include <linux/fsnotify_backend.h>
7 #include <linux/init.h>
8 #include <linux/mount.h>
9 #include <linux/namei.h>
10 #include <linux/poll.h>
11 #include <linux/security.h>
12 #include <linux/syscalls.h>
13 #include <linux/slab.h>
14 #include <linux/types.h>
15 #include <linux/uaccess.h>
17 #include <asm/ioctls.h>
19 #define FANOTIFY_DEFAULT_MAX_EVENTS 16384
20 #define FANOTIFY_DEFAULT_MAX_MARKS 8192
22 extern const struct fsnotify_ops fanotify_fsnotify_ops;
24 static struct kmem_cache *fanotify_mark_cache __read_mostly;
25 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
27 struct fanotify_response_event {
28 struct list_head list;
30 struct fsnotify_event *event;
34 * Get an fsnotify notification event if one exists and is small
35 * enough to fit in "count". Return an error pointer if the count
36 * is not large enough.
38 * Called with the group->notification_mutex held.
40 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
43 BUG_ON(!mutex_is_locked(&group->notification_mutex));
45 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
47 if (fsnotify_notify_queue_is_empty(group))
50 if (FAN_EVENT_METADATA_LEN > count)
51 return ERR_PTR(-EINVAL);
53 /* held the notification_mutex the whole time, so this is the
54 * same event we peeked above */
55 return fsnotify_remove_notify_event(group);
58 static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
61 struct dentry *dentry;
63 struct file *new_file;
65 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
67 client_fd = get_unused_fd();
71 if (event->data_type != FSNOTIFY_EVENT_PATH) {
73 put_unused_fd(client_fd);
78 * we need a new file handle for the userspace program so it can read even if it was
79 * originally opened O_WRONLY.
81 dentry = dget(event->path.dentry);
82 mnt = mntget(event->path.mnt);
83 /* it's possible this event was an overflow event. in that case dentry and mnt
84 * are NULL; That's fine, just don't call dentry open */
86 new_file = dentry_open(dentry, mnt,
87 group->fanotify_data.f_flags | FMODE_NONOTIFY,
90 new_file = ERR_PTR(-EOVERFLOW);
91 if (IS_ERR(new_file)) {
93 * we still send an event even if we can't open the file. this
94 * can happen when say tasks are gone and we try to open their
95 * /proc files or we try to open a WRONLY file like in sysfs
96 * we just send the errno to userspace since there isn't much
99 put_unused_fd(client_fd);
100 client_fd = PTR_ERR(new_file);
102 fd_install(client_fd, new_file);
108 static ssize_t fill_event_metadata(struct fsnotify_group *group,
109 struct fanotify_event_metadata *metadata,
110 struct fsnotify_event *event)
112 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
113 group, metadata, event);
115 metadata->event_len = FAN_EVENT_METADATA_LEN;
116 metadata->vers = FANOTIFY_METADATA_VERSION;
117 metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
118 metadata->pid = pid_vnr(event->tgid);
119 metadata->fd = create_fd(group, event);
124 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
125 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
128 struct fanotify_response_event *re, *return_re = NULL;
130 mutex_lock(&group->fanotify_data.access_mutex);
131 list_for_each_entry(re, &group->fanotify_data.access_list, list) {
135 list_del_init(&re->list);
139 mutex_unlock(&group->fanotify_data.access_mutex);
141 pr_debug("%s: found return_re=%p\n", __func__, return_re);
146 static int process_access_response(struct fsnotify_group *group,
147 struct fanotify_response *response_struct)
149 struct fanotify_response_event *re;
150 __s32 fd = response_struct->fd;
151 __u32 response = response_struct->response;
153 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
156 * make sure the response is valid, if invalid we do nothing and either
157 * userspace can send a valid responce or we will clean it up after the
171 re = dequeue_re(group, fd);
175 re->event->response = response;
177 wake_up(&group->fanotify_data.access_waitq);
179 kmem_cache_free(fanotify_response_event_cache, re);
184 static int prepare_for_access_response(struct fsnotify_group *group,
185 struct fsnotify_event *event,
188 struct fanotify_response_event *re;
190 if (!(event->mask & FAN_ALL_PERM_EVENTS))
193 re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
200 mutex_lock(&group->fanotify_data.access_mutex);
202 if (group->fanotify_data.bypass_perm) {
203 mutex_unlock(&group->fanotify_data.access_mutex);
204 kmem_cache_free(fanotify_response_event_cache, re);
205 event->response = FAN_ALLOW;
209 list_add_tail(&re->list, &group->fanotify_data.access_list);
210 mutex_unlock(&group->fanotify_data.access_mutex);
215 static void remove_access_response(struct fsnotify_group *group,
216 struct fsnotify_event *event,
219 struct fanotify_response_event *re;
221 if (!(event->mask & FAN_ALL_PERM_EVENTS))
224 re = dequeue_re(group, fd);
228 BUG_ON(re->event != event);
230 kmem_cache_free(fanotify_response_event_cache, re);
235 static int prepare_for_access_response(struct fsnotify_group *group,
236 struct fsnotify_event *event,
242 static void remove_access_response(struct fsnotify_group *group,
243 struct fsnotify_event *event,
250 static ssize_t copy_event_to_user(struct fsnotify_group *group,
251 struct fsnotify_event *event,
254 struct fanotify_event_metadata fanotify_event_metadata;
257 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
259 fd = fill_event_metadata(group, &fanotify_event_metadata, event);
263 ret = prepare_for_access_response(group, event, fd);
268 if (copy_to_user(buf, &fanotify_event_metadata, FAN_EVENT_METADATA_LEN))
269 goto out_kill_access_response;
271 return FAN_EVENT_METADATA_LEN;
273 out_kill_access_response:
274 remove_access_response(group, event, fd);
280 /* intofiy userspace file descriptor functions */
281 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
283 struct fsnotify_group *group = file->private_data;
286 poll_wait(file, &group->notification_waitq, wait);
287 mutex_lock(&group->notification_mutex);
288 if (!fsnotify_notify_queue_is_empty(group))
289 ret = POLLIN | POLLRDNORM;
290 mutex_unlock(&group->notification_mutex);
295 static ssize_t fanotify_read(struct file *file, char __user *buf,
296 size_t count, loff_t *pos)
298 struct fsnotify_group *group;
299 struct fsnotify_event *kevent;
305 group = file->private_data;
307 pr_debug("%s: group=%p\n", __func__, group);
310 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
312 mutex_lock(&group->notification_mutex);
313 kevent = get_one_event(group, count);
314 mutex_unlock(&group->notification_mutex);
317 ret = PTR_ERR(kevent);
320 ret = copy_event_to_user(group, kevent, buf);
321 fsnotify_put_event(kevent);
330 if (file->f_flags & O_NONBLOCK)
333 if (signal_pending(current))
342 finish_wait(&group->notification_waitq, &wait);
343 if (start != buf && ret != -EFAULT)
348 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
350 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
351 struct fanotify_response response = { .fd = -1, .response = -1 };
352 struct fsnotify_group *group;
355 group = file->private_data;
357 if (count > sizeof(response))
358 count = sizeof(response);
360 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
362 if (copy_from_user(&response, buf, count))
365 ret = process_access_response(group, &response);
375 static int fanotify_release(struct inode *ignored, struct file *file)
377 struct fsnotify_group *group = file->private_data;
378 struct fanotify_response_event *re, *lre;
380 pr_debug("%s: file=%p group=%p\n", __func__, file, group);
382 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
383 mutex_lock(&group->fanotify_data.access_mutex);
385 group->fanotify_data.bypass_perm = true;
387 list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
388 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
391 list_del_init(&re->list);
392 re->event->response = FAN_ALLOW;
394 kmem_cache_free(fanotify_response_event_cache, re);
396 mutex_unlock(&group->fanotify_data.access_mutex);
398 wake_up(&group->fanotify_data.access_waitq);
400 /* matches the fanotify_init->fsnotify_alloc_group */
401 fsnotify_put_group(group);
406 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
408 struct fsnotify_group *group;
409 struct fsnotify_event_holder *holder;
414 group = file->private_data;
416 p = (void __user *) arg;
420 mutex_lock(&group->notification_mutex);
421 list_for_each_entry(holder, &group->notification_list, event_list)
422 send_len += FAN_EVENT_METADATA_LEN;
423 mutex_unlock(&group->notification_mutex);
424 ret = put_user(send_len, (int __user *) p);
431 static const struct file_operations fanotify_fops = {
432 .poll = fanotify_poll,
433 .read = fanotify_read,
434 .write = fanotify_write,
436 .release = fanotify_release,
437 .unlocked_ioctl = fanotify_ioctl,
438 .compat_ioctl = fanotify_ioctl,
439 .llseek = noop_llseek,
442 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
444 kmem_cache_free(fanotify_mark_cache, fsn_mark);
447 static int fanotify_find_path(int dfd, const char __user *filename,
448 struct path *path, unsigned int flags)
452 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
453 dfd, filename, flags);
455 if (filename == NULL) {
460 file = fget_light(dfd, &fput_needed);
465 if ((flags & FAN_MARK_ONLYDIR) &&
466 !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
467 fput_light(file, fput_needed);
471 *path = file->f_path;
473 fput_light(file, fput_needed);
475 unsigned int lookup_flags = 0;
477 if (!(flags & FAN_MARK_DONT_FOLLOW))
478 lookup_flags |= LOOKUP_FOLLOW;
479 if (flags & FAN_MARK_ONLYDIR)
480 lookup_flags |= LOOKUP_DIRECTORY;
482 ret = user_path_at(dfd, filename, lookup_flags, path);
487 /* you can only watch an inode if you have read permissions on it */
488 ret = inode_permission(path->dentry->d_inode, MAY_READ);
495 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
501 spin_lock(&fsn_mark->lock);
502 if (!(flags & FAN_MARK_IGNORED_MASK)) {
503 oldmask = fsn_mark->mask;
504 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
506 oldmask = fsn_mark->ignored_mask;
507 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
509 spin_unlock(&fsn_mark->lock);
511 if (!(oldmask & ~mask))
512 fsnotify_destroy_mark(fsn_mark);
514 return mask & oldmask;
517 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
518 struct vfsmount *mnt, __u32 mask,
521 struct fsnotify_mark *fsn_mark = NULL;
524 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
528 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
529 fsnotify_put_mark(fsn_mark);
530 if (removed & mnt->mnt_fsnotify_mask)
531 fsnotify_recalc_vfsmount_mask(mnt);
536 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
537 struct inode *inode, __u32 mask,
540 struct fsnotify_mark *fsn_mark = NULL;
543 fsn_mark = fsnotify_find_inode_mark(group, inode);
547 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
548 /* matches the fsnotify_find_inode_mark() */
549 fsnotify_put_mark(fsn_mark);
550 if (removed & inode->i_fsnotify_mask)
551 fsnotify_recalc_inode_mask(inode);
556 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
562 spin_lock(&fsn_mark->lock);
563 if (!(flags & FAN_MARK_IGNORED_MASK)) {
564 oldmask = fsn_mark->mask;
565 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
567 oldmask = fsn_mark->ignored_mask;
568 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask | mask));
569 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
570 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
572 spin_unlock(&fsn_mark->lock);
574 return mask & ~oldmask;
577 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
578 struct vfsmount *mnt, __u32 mask,
581 struct fsnotify_mark *fsn_mark;
584 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
588 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
591 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
595 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
596 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
598 fanotify_free_mark(fsn_mark);
602 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
603 fsnotify_put_mark(fsn_mark);
604 if (added & ~mnt->mnt_fsnotify_mask)
605 fsnotify_recalc_vfsmount_mask(mnt);
610 static int fanotify_add_inode_mark(struct fsnotify_group *group,
611 struct inode *inode, __u32 mask,
614 struct fsnotify_mark *fsn_mark;
617 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
620 * If some other task has this inode open for write we should not add
621 * an ignored mark, unless that ignored mark is supposed to survive
622 * modification changes anyway.
624 if ((flags & FAN_MARK_IGNORED_MASK) &&
625 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
626 (atomic_read(&inode->i_writecount) > 0))
629 fsn_mark = fsnotify_find_inode_mark(group, inode);
633 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
636 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
640 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
641 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
643 fanotify_free_mark(fsn_mark);
647 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
648 fsnotify_put_mark(fsn_mark);
649 if (added & ~inode->i_fsnotify_mask)
650 fsnotify_recalc_inode_mask(inode);
654 /* fanotify syscalls */
655 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
657 struct fsnotify_group *group;
660 pr_debug("%s: flags=%d event_f_flags=%d\n",
661 __func__, flags, event_f_flags);
663 if (!capable(CAP_SYS_ADMIN))
666 if (flags & ~FAN_ALL_INIT_FLAGS)
669 f_flags = O_RDWR | FMODE_NONOTIFY;
670 if (flags & FAN_CLOEXEC)
671 f_flags |= O_CLOEXEC;
672 if (flags & FAN_NONBLOCK)
673 f_flags |= O_NONBLOCK;
675 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
676 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
678 return PTR_ERR(group);
680 group->fanotify_data.f_flags = event_f_flags;
681 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
682 mutex_init(&group->fanotify_data.access_mutex);
683 init_waitqueue_head(&group->fanotify_data.access_waitq);
684 INIT_LIST_HEAD(&group->fanotify_data.access_list);
686 switch (flags & FAN_ALL_CLASS_BITS) {
687 case FAN_CLASS_NOTIF:
688 group->priority = FS_PRIO_0;
690 case FAN_CLASS_CONTENT:
691 group->priority = FS_PRIO_1;
693 case FAN_CLASS_PRE_CONTENT:
694 group->priority = FS_PRIO_2;
701 if (flags & FAN_UNLIMITED_QUEUE) {
703 if (!capable(CAP_SYS_ADMIN))
705 group->max_events = UINT_MAX;
707 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
710 if (flags & FAN_UNLIMITED_MARKS) {
712 if (!capable(CAP_SYS_ADMIN))
714 group->fanotify_data.max_marks = UINT_MAX;
716 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
719 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
726 fsnotify_put_group(group);
730 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
732 const char __user * pathname)
734 struct inode *inode = NULL;
735 struct vfsmount *mnt = NULL;
736 struct fsnotify_group *group;
739 int ret, fput_needed;
741 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
742 __func__, fanotify_fd, flags, dfd, pathname, mask);
744 /* we only use the lower 32 bits as of right now. */
745 if (mask & ((__u64)0xffffffff << 32))
748 if (flags & ~FAN_ALL_MARK_FLAGS)
750 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
752 case FAN_MARK_REMOVE:
758 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
759 if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
761 if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
765 filp = fget_light(fanotify_fd, &fput_needed);
769 /* verify that this is indeed an fanotify instance */
771 if (unlikely(filp->f_op != &fanotify_fops))
773 group = filp->private_data;
776 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
777 * allowed to set permissions events.
780 if (mask & FAN_ALL_PERM_EVENTS &&
781 group->priority == FS_PRIO_0)
784 ret = fanotify_find_path(dfd, pathname, &path, flags);
788 /* inode held in place by reference to path; group by fget on fd */
789 if (!(flags & FAN_MARK_MOUNT))
790 inode = path.dentry->d_inode;
794 /* create/update an inode mark */
795 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
797 if (flags & FAN_MARK_MOUNT)
798 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
800 ret = fanotify_add_inode_mark(group, inode, mask, flags);
802 case FAN_MARK_REMOVE:
803 if (flags & FAN_MARK_MOUNT)
804 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
806 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
809 if (flags & FAN_MARK_MOUNT)
810 fsnotify_clear_vfsmount_marks_by_group(group);
812 fsnotify_clear_inode_marks_by_group(group);
820 fput_light(filp, fput_needed);
824 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
825 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
826 long dfd, long pathname)
828 return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
830 (const char __user *) pathname);
832 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
836 * fanotify_user_setup - Our initialization function. Note that we cannnot return
837 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
838 * must result in panic().
840 static int __init fanotify_user_setup(void)
842 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
843 fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
848 device_initcall(fanotify_user_setup);