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_flags(group->fanotify_data.f_flags);
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->reserved = 0;
122 metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
123 metadata->pid = pid_vnr(event->tgid);
124 if (unlikely(event->mask & FAN_Q_OVERFLOW))
125 metadata->fd = FAN_NOFD;
127 metadata->fd = create_fd(group, event);
128 if (metadata->fd < 0)
135 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
136 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
139 struct fanotify_response_event *re, *return_re = NULL;
141 mutex_lock(&group->fanotify_data.access_mutex);
142 list_for_each_entry(re, &group->fanotify_data.access_list, list) {
146 list_del_init(&re->list);
150 mutex_unlock(&group->fanotify_data.access_mutex);
152 pr_debug("%s: found return_re=%p\n", __func__, return_re);
157 static int process_access_response(struct fsnotify_group *group,
158 struct fanotify_response *response_struct)
160 struct fanotify_response_event *re;
161 __s32 fd = response_struct->fd;
162 __u32 response = response_struct->response;
164 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
167 * make sure the response is valid, if invalid we do nothing and either
168 * userspace can send a valid response or we will clean it up after the
182 re = dequeue_re(group, fd);
186 re->event->response = response;
188 wake_up(&group->fanotify_data.access_waitq);
190 kmem_cache_free(fanotify_response_event_cache, re);
195 static int prepare_for_access_response(struct fsnotify_group *group,
196 struct fsnotify_event *event,
199 struct fanotify_response_event *re;
201 if (!(event->mask & FAN_ALL_PERM_EVENTS))
204 re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
211 mutex_lock(&group->fanotify_data.access_mutex);
213 if (atomic_read(&group->fanotify_data.bypass_perm)) {
214 mutex_unlock(&group->fanotify_data.access_mutex);
215 kmem_cache_free(fanotify_response_event_cache, re);
216 event->response = FAN_ALLOW;
220 list_add_tail(&re->list, &group->fanotify_data.access_list);
221 mutex_unlock(&group->fanotify_data.access_mutex);
226 static void remove_access_response(struct fsnotify_group *group,
227 struct fsnotify_event *event,
230 struct fanotify_response_event *re;
232 if (!(event->mask & FAN_ALL_PERM_EVENTS))
235 re = dequeue_re(group, fd);
239 BUG_ON(re->event != event);
241 kmem_cache_free(fanotify_response_event_cache, re);
246 static int prepare_for_access_response(struct fsnotify_group *group,
247 struct fsnotify_event *event,
253 static void remove_access_response(struct fsnotify_group *group,
254 struct fsnotify_event *event,
261 static ssize_t copy_event_to_user(struct fsnotify_group *group,
262 struct fsnotify_event *event,
265 struct fanotify_event_metadata fanotify_event_metadata;
268 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
270 ret = fill_event_metadata(group, &fanotify_event_metadata, event);
274 fd = fanotify_event_metadata.fd;
275 ret = prepare_for_access_response(group, event, fd);
280 if (copy_to_user(buf, &fanotify_event_metadata,
281 fanotify_event_metadata.event_len))
282 goto out_kill_access_response;
284 return fanotify_event_metadata.event_len;
286 out_kill_access_response:
287 remove_access_response(group, event, fd);
292 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
293 if (event->mask & FAN_ALL_PERM_EVENTS) {
294 event->response = FAN_DENY;
295 wake_up(&group->fanotify_data.access_waitq);
301 /* intofiy userspace file descriptor functions */
302 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
304 struct fsnotify_group *group = file->private_data;
307 poll_wait(file, &group->notification_waitq, wait);
308 mutex_lock(&group->notification_mutex);
309 if (!fsnotify_notify_queue_is_empty(group))
310 ret = POLLIN | POLLRDNORM;
311 mutex_unlock(&group->notification_mutex);
316 static ssize_t fanotify_read(struct file *file, char __user *buf,
317 size_t count, loff_t *pos)
319 struct fsnotify_group *group;
320 struct fsnotify_event *kevent;
326 group = file->private_data;
328 pr_debug("%s: group=%p\n", __func__, group);
331 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
333 mutex_lock(&group->notification_mutex);
334 kevent = get_one_event(group, count);
335 mutex_unlock(&group->notification_mutex);
338 ret = PTR_ERR(kevent);
341 ret = copy_event_to_user(group, kevent, buf);
342 fsnotify_put_event(kevent);
351 if (file->f_flags & O_NONBLOCK)
354 if (signal_pending(current))
363 finish_wait(&group->notification_waitq, &wait);
364 if (start != buf && ret != -EFAULT)
369 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
371 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
372 struct fanotify_response response = { .fd = -1, .response = -1 };
373 struct fsnotify_group *group;
376 group = file->private_data;
378 if (count > sizeof(response))
379 count = sizeof(response);
381 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
383 if (copy_from_user(&response, buf, count))
386 ret = process_access_response(group, &response);
396 static int fanotify_release(struct inode *ignored, struct file *file)
398 struct fsnotify_group *group = file->private_data;
400 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
401 struct fanotify_response_event *re, *lre;
403 mutex_lock(&group->fanotify_data.access_mutex);
405 atomic_inc(&group->fanotify_data.bypass_perm);
407 list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
408 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
411 list_del_init(&re->list);
412 re->event->response = FAN_ALLOW;
414 kmem_cache_free(fanotify_response_event_cache, re);
416 mutex_unlock(&group->fanotify_data.access_mutex);
418 wake_up(&group->fanotify_data.access_waitq);
420 /* matches the fanotify_init->fsnotify_alloc_group */
421 fsnotify_put_group(group);
426 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
428 struct fsnotify_group *group;
429 struct fsnotify_event_holder *holder;
434 group = file->private_data;
436 p = (void __user *) arg;
440 mutex_lock(&group->notification_mutex);
441 list_for_each_entry(holder, &group->notification_list, event_list)
442 send_len += FAN_EVENT_METADATA_LEN;
443 mutex_unlock(&group->notification_mutex);
444 ret = put_user(send_len, (int __user *) p);
451 static const struct file_operations fanotify_fops = {
452 .poll = fanotify_poll,
453 .read = fanotify_read,
454 .write = fanotify_write,
456 .release = fanotify_release,
457 .unlocked_ioctl = fanotify_ioctl,
458 .compat_ioctl = fanotify_ioctl,
459 .llseek = noop_llseek,
462 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
464 kmem_cache_free(fanotify_mark_cache, fsn_mark);
467 static int fanotify_find_path(int dfd, const char __user *filename,
468 struct path *path, unsigned int flags)
472 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
473 dfd, filename, flags);
475 if (filename == NULL) {
480 file = fget_light(dfd, &fput_needed);
485 if ((flags & FAN_MARK_ONLYDIR) &&
486 !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
487 fput_light(file, fput_needed);
491 *path = file->f_path;
493 fput_light(file, fput_needed);
495 unsigned int lookup_flags = 0;
497 if (!(flags & FAN_MARK_DONT_FOLLOW))
498 lookup_flags |= LOOKUP_FOLLOW;
499 if (flags & FAN_MARK_ONLYDIR)
500 lookup_flags |= LOOKUP_DIRECTORY;
502 ret = user_path_at(dfd, filename, lookup_flags, path);
507 /* you can only watch an inode if you have read permissions on it */
508 ret = inode_permission(path->dentry->d_inode, MAY_READ);
515 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
521 spin_lock(&fsn_mark->lock);
522 if (!(flags & FAN_MARK_IGNORED_MASK)) {
523 oldmask = fsn_mark->mask;
524 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
526 oldmask = fsn_mark->ignored_mask;
527 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
529 spin_unlock(&fsn_mark->lock);
531 if (!(oldmask & ~mask))
532 fsnotify_destroy_mark(fsn_mark);
534 return mask & oldmask;
537 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
538 struct vfsmount *mnt, __u32 mask,
541 struct fsnotify_mark *fsn_mark = NULL;
544 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
548 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
549 fsnotify_put_mark(fsn_mark);
550 if (removed & mnt->mnt_fsnotify_mask)
551 fsnotify_recalc_vfsmount_mask(mnt);
556 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
557 struct inode *inode, __u32 mask,
560 struct fsnotify_mark *fsn_mark = NULL;
563 fsn_mark = fsnotify_find_inode_mark(group, inode);
567 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
568 /* matches the fsnotify_find_inode_mark() */
569 fsnotify_put_mark(fsn_mark);
570 if (removed & inode->i_fsnotify_mask)
571 fsnotify_recalc_inode_mask(inode);
576 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
582 spin_lock(&fsn_mark->lock);
583 if (!(flags & FAN_MARK_IGNORED_MASK)) {
584 oldmask = fsn_mark->mask;
585 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
587 __u32 tmask = fsn_mark->ignored_mask | mask;
588 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
589 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
590 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
593 if (!(flags & FAN_MARK_ONDIR)) {
594 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
595 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
598 spin_unlock(&fsn_mark->lock);
600 return mask & ~oldmask;
603 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
604 struct vfsmount *mnt, __u32 mask,
607 struct fsnotify_mark *fsn_mark;
611 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
613 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
616 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
620 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
621 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
625 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
627 if (added & ~mnt->mnt_fsnotify_mask)
628 fsnotify_recalc_vfsmount_mask(mnt);
630 fsnotify_put_mark(fsn_mark);
634 static int fanotify_add_inode_mark(struct fsnotify_group *group,
635 struct inode *inode, __u32 mask,
638 struct fsnotify_mark *fsn_mark;
642 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
645 * If some other task has this inode open for write we should not add
646 * an ignored mark, unless that ignored mark is supposed to survive
647 * modification changes anyway.
649 if ((flags & FAN_MARK_IGNORED_MASK) &&
650 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
651 (atomic_read(&inode->i_writecount) > 0))
654 fsn_mark = fsnotify_find_inode_mark(group, inode);
656 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
659 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
663 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
664 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
668 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
670 if (added & ~inode->i_fsnotify_mask)
671 fsnotify_recalc_inode_mask(inode);
673 fsnotify_put_mark(fsn_mark);
677 /* fanotify syscalls */
678 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
680 struct fsnotify_group *group;
682 struct user_struct *user;
684 pr_debug("%s: flags=%d event_f_flags=%d\n",
685 __func__, flags, event_f_flags);
687 if (!capable(CAP_SYS_ADMIN))
690 if (flags & ~FAN_ALL_INIT_FLAGS)
693 user = get_current_user();
694 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
699 f_flags = O_RDWR | FMODE_NONOTIFY;
700 if (flags & FAN_CLOEXEC)
701 f_flags |= O_CLOEXEC;
702 if (flags & FAN_NONBLOCK)
703 f_flags |= O_NONBLOCK;
705 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
706 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
709 return PTR_ERR(group);
712 group->fanotify_data.user = user;
713 atomic_inc(&user->fanotify_listeners);
715 group->fanotify_data.f_flags = event_f_flags;
716 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
717 mutex_init(&group->fanotify_data.access_mutex);
718 init_waitqueue_head(&group->fanotify_data.access_waitq);
719 INIT_LIST_HEAD(&group->fanotify_data.access_list);
720 atomic_set(&group->fanotify_data.bypass_perm, 0);
722 switch (flags & FAN_ALL_CLASS_BITS) {
723 case FAN_CLASS_NOTIF:
724 group->priority = FS_PRIO_0;
726 case FAN_CLASS_CONTENT:
727 group->priority = FS_PRIO_1;
729 case FAN_CLASS_PRE_CONTENT:
730 group->priority = FS_PRIO_2;
737 if (flags & FAN_UNLIMITED_QUEUE) {
739 if (!capable(CAP_SYS_ADMIN))
741 group->max_events = UINT_MAX;
743 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
746 if (flags & FAN_UNLIMITED_MARKS) {
748 if (!capable(CAP_SYS_ADMIN))
750 group->fanotify_data.max_marks = UINT_MAX;
752 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
755 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
762 fsnotify_put_group(group);
766 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
768 const char __user * pathname)
770 struct inode *inode = NULL;
771 struct vfsmount *mnt = NULL;
772 struct fsnotify_group *group;
775 int ret, fput_needed;
777 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
778 __func__, fanotify_fd, flags, dfd, pathname, mask);
780 /* we only use the lower 32 bits as of right now. */
781 if (mask & ((__u64)0xffffffff << 32))
784 if (flags & ~FAN_ALL_MARK_FLAGS)
786 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
787 case FAN_MARK_ADD: /* fallthrough */
788 case FAN_MARK_REMOVE:
797 if (mask & FAN_ONDIR) {
798 flags |= FAN_MARK_ONDIR;
802 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
803 if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
805 if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
809 filp = fget_light(fanotify_fd, &fput_needed);
813 /* verify that this is indeed an fanotify instance */
815 if (unlikely(filp->f_op != &fanotify_fops))
817 group = filp->private_data;
820 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
821 * allowed to set permissions events.
824 if (mask & FAN_ALL_PERM_EVENTS &&
825 group->priority == FS_PRIO_0)
828 ret = fanotify_find_path(dfd, pathname, &path, flags);
832 /* inode held in place by reference to path; group by fget on fd */
833 if (!(flags & FAN_MARK_MOUNT))
834 inode = path.dentry->d_inode;
838 /* create/update an inode mark */
839 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
841 if (flags & FAN_MARK_MOUNT)
842 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
844 ret = fanotify_add_inode_mark(group, inode, mask, flags);
846 case FAN_MARK_REMOVE:
847 if (flags & FAN_MARK_MOUNT)
848 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
850 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
853 if (flags & FAN_MARK_MOUNT)
854 fsnotify_clear_vfsmount_marks_by_group(group);
856 fsnotify_clear_inode_marks_by_group(group);
864 fput_light(filp, fput_needed);
868 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
869 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
870 long dfd, long pathname)
872 return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
874 (const char __user *) pathname);
876 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
880 * fanotify_user_setup - Our initialization function. Note that we cannot return
881 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
882 * must result in panic().
884 static int __init fanotify_user_setup(void)
886 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
887 fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
892 device_initcall(fanotify_user_setup);