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
21 #define FANOTIFY_DEFAULT_MAX_LISTENERS 128
23 extern const struct fsnotify_ops fanotify_fsnotify_ops;
25 static struct kmem_cache *fanotify_mark_cache __read_mostly;
26 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
28 struct fanotify_response_event {
29 struct list_head list;
31 struct fsnotify_event *event;
35 * Get an fsnotify notification event if one exists and is small
36 * enough to fit in "count". Return an error pointer if the count
37 * is not large enough.
39 * Called with the group->notification_mutex held.
41 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
44 BUG_ON(!mutex_is_locked(&group->notification_mutex));
46 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
48 if (fsnotify_notify_queue_is_empty(group))
51 if (FAN_EVENT_METADATA_LEN > count)
52 return ERR_PTR(-EINVAL);
54 /* held the notification_mutex the whole time, so this is the
55 * same event we peeked above */
56 return fsnotify_remove_notify_event(group);
59 static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
62 struct dentry *dentry;
64 struct file *new_file;
66 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
68 client_fd = get_unused_fd();
72 if (event->data_type != FSNOTIFY_EVENT_PATH) {
74 put_unused_fd(client_fd);
79 * we need a new file handle for the userspace program so it can read even if it was
80 * originally opened O_WRONLY.
82 dentry = dget(event->path.dentry);
83 mnt = mntget(event->path.mnt);
84 /* it's possible this event was an overflow event. in that case dentry and mnt
85 * are NULL; That's fine, just don't call dentry open */
87 new_file = dentry_open(dentry, mnt,
88 group->fanotify_data.f_flags | FMODE_NONOTIFY,
91 new_file = ERR_PTR(-EOVERFLOW);
92 if (IS_ERR(new_file)) {
94 * we still send an event even if we can't open the file. this
95 * can happen when say tasks are gone and we try to open their
96 * /proc files or we try to open a WRONLY file like in sysfs
97 * we just send the errno to userspace since there isn't much
100 put_unused_fd(client_fd);
101 client_fd = PTR_ERR(new_file);
103 fd_install(client_fd, new_file);
109 static int fill_event_metadata(struct fsnotify_group *group,
110 struct fanotify_event_metadata *metadata,
111 struct fsnotify_event *event)
115 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
116 group, metadata, event);
118 metadata->event_len = FAN_EVENT_METADATA_LEN;
119 metadata->metadata_len = FAN_EVENT_METADATA_LEN;
120 metadata->vers = FANOTIFY_METADATA_VERSION;
121 metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
122 metadata->pid = pid_vnr(event->tgid);
123 if (unlikely(event->mask & FAN_Q_OVERFLOW))
124 metadata->fd = FAN_NOFD;
126 metadata->fd = create_fd(group, event);
127 if (metadata->fd < 0)
134 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
135 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
138 struct fanotify_response_event *re, *return_re = NULL;
140 mutex_lock(&group->fanotify_data.access_mutex);
141 list_for_each_entry(re, &group->fanotify_data.access_list, list) {
145 list_del_init(&re->list);
149 mutex_unlock(&group->fanotify_data.access_mutex);
151 pr_debug("%s: found return_re=%p\n", __func__, return_re);
156 static int process_access_response(struct fsnotify_group *group,
157 struct fanotify_response *response_struct)
159 struct fanotify_response_event *re;
160 __s32 fd = response_struct->fd;
161 __u32 response = response_struct->response;
163 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
166 * make sure the response is valid, if invalid we do nothing and either
167 * userspace can send a valid responce or we will clean it up after the
181 re = dequeue_re(group, fd);
185 re->event->response = response;
187 wake_up(&group->fanotify_data.access_waitq);
189 kmem_cache_free(fanotify_response_event_cache, re);
194 static int prepare_for_access_response(struct fsnotify_group *group,
195 struct fsnotify_event *event,
198 struct fanotify_response_event *re;
200 if (!(event->mask & FAN_ALL_PERM_EVENTS))
203 re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
210 mutex_lock(&group->fanotify_data.access_mutex);
212 if (atomic_read(&group->fanotify_data.bypass_perm)) {
213 mutex_unlock(&group->fanotify_data.access_mutex);
214 kmem_cache_free(fanotify_response_event_cache, re);
215 event->response = FAN_ALLOW;
219 list_add_tail(&re->list, &group->fanotify_data.access_list);
220 mutex_unlock(&group->fanotify_data.access_mutex);
225 static void remove_access_response(struct fsnotify_group *group,
226 struct fsnotify_event *event,
229 struct fanotify_response_event *re;
231 if (!(event->mask & FAN_ALL_PERM_EVENTS))
234 re = dequeue_re(group, fd);
238 BUG_ON(re->event != event);
240 kmem_cache_free(fanotify_response_event_cache, re);
245 static int prepare_for_access_response(struct fsnotify_group *group,
246 struct fsnotify_event *event,
252 static void remove_access_response(struct fsnotify_group *group,
253 struct fsnotify_event *event,
260 static ssize_t copy_event_to_user(struct fsnotify_group *group,
261 struct fsnotify_event *event,
264 struct fanotify_event_metadata fanotify_event_metadata;
267 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
269 ret = fill_event_metadata(group, &fanotify_event_metadata, event);
273 fd = fanotify_event_metadata.fd;
274 ret = prepare_for_access_response(group, event, fd);
279 if (copy_to_user(buf, &fanotify_event_metadata,
280 fanotify_event_metadata.event_len))
281 goto out_kill_access_response;
283 return fanotify_event_metadata.event_len;
285 out_kill_access_response:
286 remove_access_response(group, event, fd);
291 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
292 if (event->mask & FAN_ALL_PERM_EVENTS) {
293 event->response = FAN_DENY;
294 wake_up(&group->fanotify_data.access_waitq);
300 /* intofiy userspace file descriptor functions */
301 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
303 struct fsnotify_group *group = file->private_data;
306 poll_wait(file, &group->notification_waitq, wait);
307 mutex_lock(&group->notification_mutex);
308 if (!fsnotify_notify_queue_is_empty(group))
309 ret = POLLIN | POLLRDNORM;
310 mutex_unlock(&group->notification_mutex);
315 static ssize_t fanotify_read(struct file *file, char __user *buf,
316 size_t count, loff_t *pos)
318 struct fsnotify_group *group;
319 struct fsnotify_event *kevent;
325 group = file->private_data;
327 pr_debug("%s: group=%p\n", __func__, group);
330 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
332 mutex_lock(&group->notification_mutex);
333 kevent = get_one_event(group, count);
334 mutex_unlock(&group->notification_mutex);
337 ret = PTR_ERR(kevent);
340 ret = copy_event_to_user(group, kevent, buf);
341 fsnotify_put_event(kevent);
350 if (file->f_flags & O_NONBLOCK)
353 if (signal_pending(current))
362 finish_wait(&group->notification_waitq, &wait);
363 if (start != buf && ret != -EFAULT)
368 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
370 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
371 struct fanotify_response response = { .fd = -1, .response = -1 };
372 struct fsnotify_group *group;
375 group = file->private_data;
377 if (count > sizeof(response))
378 count = sizeof(response);
380 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
382 if (copy_from_user(&response, buf, count))
385 ret = process_access_response(group, &response);
395 static int fanotify_release(struct inode *ignored, struct file *file)
397 struct fsnotify_group *group = file->private_data;
399 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
400 struct fanotify_response_event *re, *lre;
402 mutex_lock(&group->fanotify_data.access_mutex);
404 atomic_inc(&group->fanotify_data.bypass_perm);
406 list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
407 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
410 list_del_init(&re->list);
411 re->event->response = FAN_ALLOW;
413 kmem_cache_free(fanotify_response_event_cache, re);
415 mutex_unlock(&group->fanotify_data.access_mutex);
417 wake_up(&group->fanotify_data.access_waitq);
419 /* matches the fanotify_init->fsnotify_alloc_group */
420 fsnotify_put_group(group);
425 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
427 struct fsnotify_group *group;
428 struct fsnotify_event_holder *holder;
433 group = file->private_data;
435 p = (void __user *) arg;
439 mutex_lock(&group->notification_mutex);
440 list_for_each_entry(holder, &group->notification_list, event_list)
441 send_len += FAN_EVENT_METADATA_LEN;
442 mutex_unlock(&group->notification_mutex);
443 ret = put_user(send_len, (int __user *) p);
450 static const struct file_operations fanotify_fops = {
451 .poll = fanotify_poll,
452 .read = fanotify_read,
453 .write = fanotify_write,
455 .release = fanotify_release,
456 .unlocked_ioctl = fanotify_ioctl,
457 .compat_ioctl = fanotify_ioctl,
458 .llseek = noop_llseek,
461 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
463 kmem_cache_free(fanotify_mark_cache, fsn_mark);
466 static int fanotify_find_path(int dfd, const char __user *filename,
467 struct path *path, unsigned int flags)
471 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
472 dfd, filename, flags);
474 if (filename == NULL) {
479 file = fget_light(dfd, &fput_needed);
484 if ((flags & FAN_MARK_ONLYDIR) &&
485 !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
486 fput_light(file, fput_needed);
490 *path = file->f_path;
492 fput_light(file, fput_needed);
494 unsigned int lookup_flags = 0;
496 if (!(flags & FAN_MARK_DONT_FOLLOW))
497 lookup_flags |= LOOKUP_FOLLOW;
498 if (flags & FAN_MARK_ONLYDIR)
499 lookup_flags |= LOOKUP_DIRECTORY;
501 ret = user_path_at(dfd, filename, lookup_flags, path);
506 /* you can only watch an inode if you have read permissions on it */
507 ret = inode_permission(path->dentry->d_inode, MAY_READ);
514 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
520 spin_lock(&fsn_mark->lock);
521 if (!(flags & FAN_MARK_IGNORED_MASK)) {
522 oldmask = fsn_mark->mask;
523 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
525 oldmask = fsn_mark->ignored_mask;
526 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
528 spin_unlock(&fsn_mark->lock);
530 if (!(oldmask & ~mask))
531 fsnotify_destroy_mark(fsn_mark);
533 return mask & oldmask;
536 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
537 struct vfsmount *mnt, __u32 mask,
540 struct fsnotify_mark *fsn_mark = NULL;
543 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
547 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
548 fsnotify_put_mark(fsn_mark);
549 if (removed & mnt->mnt_fsnotify_mask)
550 fsnotify_recalc_vfsmount_mask(mnt);
555 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
556 struct inode *inode, __u32 mask,
559 struct fsnotify_mark *fsn_mark = NULL;
562 fsn_mark = fsnotify_find_inode_mark(group, inode);
566 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
567 /* matches the fsnotify_find_inode_mark() */
568 fsnotify_put_mark(fsn_mark);
569 if (removed & inode->i_fsnotify_mask)
570 fsnotify_recalc_inode_mask(inode);
575 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
581 spin_lock(&fsn_mark->lock);
582 if (!(flags & FAN_MARK_IGNORED_MASK)) {
583 oldmask = fsn_mark->mask;
584 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
586 __u32 tmask = fsn_mark->ignored_mask | mask;
587 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
588 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
589 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
592 if (!(flags & FAN_MARK_ONDIR)) {
593 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
594 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
597 spin_unlock(&fsn_mark->lock);
599 return mask & ~oldmask;
602 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
603 struct vfsmount *mnt, __u32 mask,
606 struct fsnotify_mark *fsn_mark;
610 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
612 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
615 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
619 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
620 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
624 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
626 if (added & ~mnt->mnt_fsnotify_mask)
627 fsnotify_recalc_vfsmount_mask(mnt);
629 fsnotify_put_mark(fsn_mark);
633 static int fanotify_add_inode_mark(struct fsnotify_group *group,
634 struct inode *inode, __u32 mask,
637 struct fsnotify_mark *fsn_mark;
641 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
644 * If some other task has this inode open for write we should not add
645 * an ignored mark, unless that ignored mark is supposed to survive
646 * modification changes anyway.
648 if ((flags & FAN_MARK_IGNORED_MASK) &&
649 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
650 (atomic_read(&inode->i_writecount) > 0))
653 fsn_mark = fsnotify_find_inode_mark(group, inode);
655 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
658 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
662 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
663 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
667 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
669 if (added & ~inode->i_fsnotify_mask)
670 fsnotify_recalc_inode_mask(inode);
672 fsnotify_put_mark(fsn_mark);
676 /* fanotify syscalls */
677 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
679 struct fsnotify_group *group;
681 struct user_struct *user;
683 pr_debug("%s: flags=%d event_f_flags=%d\n",
684 __func__, flags, event_f_flags);
686 if (!capable(CAP_SYS_ADMIN))
689 if (flags & ~FAN_ALL_INIT_FLAGS)
692 user = get_current_user();
693 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
698 f_flags = O_RDWR | FMODE_NONOTIFY;
699 if (flags & FAN_CLOEXEC)
700 f_flags |= O_CLOEXEC;
701 if (flags & FAN_NONBLOCK)
702 f_flags |= O_NONBLOCK;
704 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
705 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
708 return PTR_ERR(group);
711 group->fanotify_data.user = user;
712 atomic_inc(&user->fanotify_listeners);
714 group->fanotify_data.f_flags = event_f_flags;
715 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
716 mutex_init(&group->fanotify_data.access_mutex);
717 init_waitqueue_head(&group->fanotify_data.access_waitq);
718 INIT_LIST_HEAD(&group->fanotify_data.access_list);
719 atomic_set(&group->fanotify_data.bypass_perm, 0);
721 switch (flags & FAN_ALL_CLASS_BITS) {
722 case FAN_CLASS_NOTIF:
723 group->priority = FS_PRIO_0;
725 case FAN_CLASS_CONTENT:
726 group->priority = FS_PRIO_1;
728 case FAN_CLASS_PRE_CONTENT:
729 group->priority = FS_PRIO_2;
736 if (flags & FAN_UNLIMITED_QUEUE) {
738 if (!capable(CAP_SYS_ADMIN))
740 group->max_events = UINT_MAX;
742 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
745 if (flags & FAN_UNLIMITED_MARKS) {
747 if (!capable(CAP_SYS_ADMIN))
749 group->fanotify_data.max_marks = UINT_MAX;
751 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
754 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
761 fsnotify_put_group(group);
765 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
767 const char __user * pathname)
769 struct inode *inode = NULL;
770 struct vfsmount *mnt = NULL;
771 struct fsnotify_group *group;
774 int ret, fput_needed;
776 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
777 __func__, fanotify_fd, flags, dfd, pathname, mask);
779 /* we only use the lower 32 bits as of right now. */
780 if (mask & ((__u64)0xffffffff << 32))
783 if (flags & ~FAN_ALL_MARK_FLAGS)
785 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
786 case FAN_MARK_ADD: /* fallthrough */
787 case FAN_MARK_REMOVE:
796 if (mask & FAN_ONDIR) {
797 flags |= FAN_MARK_ONDIR;
801 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
802 if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
804 if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
808 filp = fget_light(fanotify_fd, &fput_needed);
812 /* verify that this is indeed an fanotify instance */
814 if (unlikely(filp->f_op != &fanotify_fops))
816 group = filp->private_data;
819 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
820 * allowed to set permissions events.
823 if (mask & FAN_ALL_PERM_EVENTS &&
824 group->priority == FS_PRIO_0)
827 ret = fanotify_find_path(dfd, pathname, &path, flags);
831 /* inode held in place by reference to path; group by fget on fd */
832 if (!(flags & FAN_MARK_MOUNT))
833 inode = path.dentry->d_inode;
837 /* create/update an inode mark */
838 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
840 if (flags & FAN_MARK_MOUNT)
841 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
843 ret = fanotify_add_inode_mark(group, inode, mask, flags);
845 case FAN_MARK_REMOVE:
846 if (flags & FAN_MARK_MOUNT)
847 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
849 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
852 if (flags & FAN_MARK_MOUNT)
853 fsnotify_clear_vfsmount_marks_by_group(group);
855 fsnotify_clear_inode_marks_by_group(group);
863 fput_light(filp, fput_needed);
867 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
868 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
869 long dfd, long pathname)
871 return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
873 (const char __user *) pathname);
875 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
879 * fanotify_user_setup - Our initialization function. Note that we cannnot return
880 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
881 * must result in panic().
883 static int __init fanotify_user_setup(void)
885 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
886 fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
891 device_initcall(fanotify_user_setup);