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 extern const struct fsnotify_ops fanotify_fsnotify_ops;
21 static struct kmem_cache *fanotify_mark_cache __read_mostly;
22 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
24 struct fanotify_response_event {
25 struct list_head list;
27 struct fsnotify_event *event;
31 * Get an fsnotify notification event if one exists and is small
32 * enough to fit in "count". Return an error pointer if the count
33 * is not large enough.
35 * Called with the group->notification_mutex held.
37 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
40 BUG_ON(!mutex_is_locked(&group->notification_mutex));
42 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
44 if (fsnotify_notify_queue_is_empty(group))
47 if (FAN_EVENT_METADATA_LEN > count)
48 return ERR_PTR(-EINVAL);
50 /* held the notification_mutex the whole time, so this is the
51 * same event we peeked above */
52 return fsnotify_remove_notify_event(group);
55 static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
58 struct dentry *dentry;
60 struct file *new_file;
62 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
64 client_fd = get_unused_fd();
68 if (event->data_type != FSNOTIFY_EVENT_PATH) {
70 put_unused_fd(client_fd);
75 * we need a new file handle for the userspace program so it can read even if it was
76 * originally opened O_WRONLY.
78 dentry = dget(event->path.dentry);
79 mnt = mntget(event->path.mnt);
80 /* it's possible this event was an overflow event. in that case dentry and mnt
81 * are NULL; That's fine, just don't call dentry open */
83 new_file = dentry_open(dentry, mnt,
84 group->fanotify_data.f_flags | FMODE_NONOTIFY,
87 new_file = ERR_PTR(-EOVERFLOW);
88 if (IS_ERR(new_file)) {
90 * we still send an event even if we can't open the file. this
91 * can happen when say tasks are gone and we try to open their
92 * /proc files or we try to open a WRONLY file like in sysfs
93 * we just send the errno to userspace since there isn't much
96 put_unused_fd(client_fd);
97 client_fd = PTR_ERR(new_file);
99 fd_install(client_fd, new_file);
105 static ssize_t fill_event_metadata(struct fsnotify_group *group,
106 struct fanotify_event_metadata *metadata,
107 struct fsnotify_event *event)
109 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
110 group, metadata, event);
112 metadata->event_len = FAN_EVENT_METADATA_LEN;
113 metadata->vers = FANOTIFY_METADATA_VERSION;
114 metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
115 metadata->pid = pid_vnr(event->tgid);
116 metadata->fd = create_fd(group, event);
121 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
122 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
125 struct fanotify_response_event *re, *return_re = NULL;
127 mutex_lock(&group->fanotify_data.access_mutex);
128 list_for_each_entry(re, &group->fanotify_data.access_list, list) {
132 list_del_init(&re->list);
136 mutex_unlock(&group->fanotify_data.access_mutex);
138 pr_debug("%s: found return_re=%p\n", __func__, return_re);
143 static int process_access_response(struct fsnotify_group *group,
144 struct fanotify_response *response_struct)
146 struct fanotify_response_event *re;
147 __s32 fd = response_struct->fd;
148 __u32 response = response_struct->response;
150 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
153 * make sure the response is valid, if invalid we do nothing and either
154 * userspace can send a valid responce or we will clean it up after the
168 re = dequeue_re(group, fd);
172 re->event->response = response;
174 wake_up(&group->fanotify_data.access_waitq);
176 kmem_cache_free(fanotify_response_event_cache, re);
181 static int prepare_for_access_response(struct fsnotify_group *group,
182 struct fsnotify_event *event,
185 struct fanotify_response_event *re;
187 if (!(event->mask & FAN_ALL_PERM_EVENTS))
190 re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
197 mutex_lock(&group->fanotify_data.access_mutex);
199 if (group->fanotify_data.bypass_perm) {
200 mutex_unlock(&group->fanotify_data.access_mutex);
201 kmem_cache_free(fanotify_response_event_cache, re);
202 event->response = FAN_ALLOW;
206 list_add_tail(&re->list, &group->fanotify_data.access_list);
207 mutex_unlock(&group->fanotify_data.access_mutex);
212 static void remove_access_response(struct fsnotify_group *group,
213 struct fsnotify_event *event,
216 struct fanotify_response_event *re;
218 if (!(event->mask & FAN_ALL_PERM_EVENTS))
221 re = dequeue_re(group, fd);
225 BUG_ON(re->event != event);
227 kmem_cache_free(fanotify_response_event_cache, re);
232 static int prepare_for_access_response(struct fsnotify_group *group,
233 struct fsnotify_event *event,
239 static void remove_access_response(struct fsnotify_group *group,
240 struct fsnotify_event *event,
247 static ssize_t copy_event_to_user(struct fsnotify_group *group,
248 struct fsnotify_event *event,
251 struct fanotify_event_metadata fanotify_event_metadata;
254 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
256 fd = fill_event_metadata(group, &fanotify_event_metadata, event);
260 ret = prepare_for_access_response(group, event, fd);
265 if (copy_to_user(buf, &fanotify_event_metadata, FAN_EVENT_METADATA_LEN))
266 goto out_kill_access_response;
268 return FAN_EVENT_METADATA_LEN;
270 out_kill_access_response:
271 remove_access_response(group, event, fd);
277 /* intofiy userspace file descriptor functions */
278 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
280 struct fsnotify_group *group = file->private_data;
283 poll_wait(file, &group->notification_waitq, wait);
284 mutex_lock(&group->notification_mutex);
285 if (!fsnotify_notify_queue_is_empty(group))
286 ret = POLLIN | POLLRDNORM;
287 mutex_unlock(&group->notification_mutex);
292 static ssize_t fanotify_read(struct file *file, char __user *buf,
293 size_t count, loff_t *pos)
295 struct fsnotify_group *group;
296 struct fsnotify_event *kevent;
302 group = file->private_data;
304 pr_debug("%s: group=%p\n", __func__, group);
307 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
309 mutex_lock(&group->notification_mutex);
310 kevent = get_one_event(group, count);
311 mutex_unlock(&group->notification_mutex);
314 ret = PTR_ERR(kevent);
317 ret = copy_event_to_user(group, kevent, buf);
318 fsnotify_put_event(kevent);
327 if (file->f_flags & O_NONBLOCK)
330 if (signal_pending(current))
339 finish_wait(&group->notification_waitq, &wait);
340 if (start != buf && ret != -EFAULT)
345 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
347 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
348 struct fanotify_response response = { .fd = -1, .response = -1 };
349 struct fsnotify_group *group;
352 group = file->private_data;
354 if (count > sizeof(response))
355 count = sizeof(response);
357 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
359 if (copy_from_user(&response, buf, count))
362 ret = process_access_response(group, &response);
372 static int fanotify_release(struct inode *ignored, struct file *file)
374 struct fsnotify_group *group = file->private_data;
375 struct fanotify_response_event *re, *lre;
377 pr_debug("%s: file=%p group=%p\n", __func__, file, group);
379 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
380 mutex_lock(&group->fanotify_data.access_mutex);
382 group->fanotify_data.bypass_perm = true;
384 list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
385 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
388 list_del_init(&re->list);
389 re->event->response = FAN_ALLOW;
391 kmem_cache_free(fanotify_response_event_cache, re);
393 mutex_unlock(&group->fanotify_data.access_mutex);
395 wake_up(&group->fanotify_data.access_waitq);
397 /* matches the fanotify_init->fsnotify_alloc_group */
398 fsnotify_put_group(group);
403 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
405 struct fsnotify_group *group;
406 struct fsnotify_event_holder *holder;
411 group = file->private_data;
413 p = (void __user *) arg;
417 mutex_lock(&group->notification_mutex);
418 list_for_each_entry(holder, &group->notification_list, event_list)
419 send_len += FAN_EVENT_METADATA_LEN;
420 mutex_unlock(&group->notification_mutex);
421 ret = put_user(send_len, (int __user *) p);
428 static const struct file_operations fanotify_fops = {
429 .poll = fanotify_poll,
430 .read = fanotify_read,
431 .write = fanotify_write,
433 .release = fanotify_release,
434 .unlocked_ioctl = fanotify_ioctl,
435 .compat_ioctl = fanotify_ioctl,
436 .llseek = noop_llseek,
439 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
441 kmem_cache_free(fanotify_mark_cache, fsn_mark);
444 static int fanotify_find_path(int dfd, const char __user *filename,
445 struct path *path, unsigned int flags)
449 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
450 dfd, filename, flags);
452 if (filename == NULL) {
457 file = fget_light(dfd, &fput_needed);
462 if ((flags & FAN_MARK_ONLYDIR) &&
463 !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
464 fput_light(file, fput_needed);
468 *path = file->f_path;
470 fput_light(file, fput_needed);
472 unsigned int lookup_flags = 0;
474 if (!(flags & FAN_MARK_DONT_FOLLOW))
475 lookup_flags |= LOOKUP_FOLLOW;
476 if (flags & FAN_MARK_ONLYDIR)
477 lookup_flags |= LOOKUP_DIRECTORY;
479 ret = user_path_at(dfd, filename, lookup_flags, path);
484 /* you can only watch an inode if you have read permissions on it */
485 ret = inode_permission(path->dentry->d_inode, MAY_READ);
492 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
498 spin_lock(&fsn_mark->lock);
499 if (!(flags & FAN_MARK_IGNORED_MASK)) {
500 oldmask = fsn_mark->mask;
501 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
503 oldmask = fsn_mark->ignored_mask;
504 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
506 spin_unlock(&fsn_mark->lock);
508 if (!(oldmask & ~mask))
509 fsnotify_destroy_mark(fsn_mark);
511 return mask & oldmask;
514 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
515 struct vfsmount *mnt, __u32 mask,
518 struct fsnotify_mark *fsn_mark = NULL;
521 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
525 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
526 fsnotify_put_mark(fsn_mark);
527 if (removed & mnt->mnt_fsnotify_mask)
528 fsnotify_recalc_vfsmount_mask(mnt);
533 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
534 struct inode *inode, __u32 mask,
537 struct fsnotify_mark *fsn_mark = NULL;
540 fsn_mark = fsnotify_find_inode_mark(group, inode);
544 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
545 /* matches the fsnotify_find_inode_mark() */
546 fsnotify_put_mark(fsn_mark);
547 if (removed & inode->i_fsnotify_mask)
548 fsnotify_recalc_inode_mask(inode);
553 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
559 spin_lock(&fsn_mark->lock);
560 if (!(flags & FAN_MARK_IGNORED_MASK)) {
561 oldmask = fsn_mark->mask;
562 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
564 oldmask = fsn_mark->ignored_mask;
565 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask | mask));
566 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
567 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
569 spin_unlock(&fsn_mark->lock);
571 return mask & ~oldmask;
574 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
575 struct vfsmount *mnt, __u32 mask,
578 struct fsnotify_mark *fsn_mark;
581 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
585 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
589 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
590 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
592 fanotify_free_mark(fsn_mark);
596 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
597 fsnotify_put_mark(fsn_mark);
598 if (added & ~mnt->mnt_fsnotify_mask)
599 fsnotify_recalc_vfsmount_mask(mnt);
604 static int fanotify_add_inode_mark(struct fsnotify_group *group,
605 struct inode *inode, __u32 mask,
608 struct fsnotify_mark *fsn_mark;
611 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
613 fsn_mark = fsnotify_find_inode_mark(group, inode);
617 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
621 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
622 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
624 fanotify_free_mark(fsn_mark);
628 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
629 fsnotify_put_mark(fsn_mark);
630 if (added & ~inode->i_fsnotify_mask)
631 fsnotify_recalc_inode_mask(inode);
635 /* fanotify syscalls */
636 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
638 struct fsnotify_group *group;
641 pr_debug("%s: flags=%d event_f_flags=%d\n",
642 __func__, flags, event_f_flags);
644 if (!capable(CAP_SYS_ADMIN))
647 if (flags & ~FAN_ALL_INIT_FLAGS)
650 f_flags = O_RDWR | FMODE_NONOTIFY;
651 if (flags & FAN_CLOEXEC)
652 f_flags |= O_CLOEXEC;
653 if (flags & FAN_NONBLOCK)
654 f_flags |= O_NONBLOCK;
656 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
657 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
659 return PTR_ERR(group);
661 group->fanotify_data.f_flags = event_f_flags;
662 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
663 mutex_init(&group->fanotify_data.access_mutex);
664 init_waitqueue_head(&group->fanotify_data.access_waitq);
665 INIT_LIST_HEAD(&group->fanotify_data.access_list);
668 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
675 fsnotify_put_group(group);
679 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
681 const char __user * pathname)
683 struct inode *inode = NULL;
684 struct vfsmount *mnt = NULL;
685 struct fsnotify_group *group;
688 int ret, fput_needed;
690 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
691 __func__, fanotify_fd, flags, dfd, pathname, mask);
693 /* we only use the lower 32 bits as of right now. */
694 if (mask & ((__u64)0xffffffff << 32))
697 if (flags & ~FAN_ALL_MARK_FLAGS)
699 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
701 case FAN_MARK_REMOVE:
707 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
708 if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
710 if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
714 filp = fget_light(fanotify_fd, &fput_needed);
718 /* verify that this is indeed an fanotify instance */
720 if (unlikely(filp->f_op != &fanotify_fops))
723 ret = fanotify_find_path(dfd, pathname, &path, flags);
727 /* inode held in place by reference to path; group by fget on fd */
728 if (!(flags & FAN_MARK_MOUNT))
729 inode = path.dentry->d_inode;
732 group = filp->private_data;
734 /* create/update an inode mark */
735 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
737 if (flags & FAN_MARK_MOUNT)
738 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
740 ret = fanotify_add_inode_mark(group, inode, mask, flags);
742 case FAN_MARK_REMOVE:
743 if (flags & FAN_MARK_MOUNT)
744 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
746 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
749 if (flags & FAN_MARK_MOUNT)
750 fsnotify_clear_vfsmount_marks_by_group(group);
752 fsnotify_clear_inode_marks_by_group(group);
760 fput_light(filp, fput_needed);
764 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
765 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
766 long dfd, long pathname)
768 return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
770 (const char __user *) pathname);
772 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
776 * fanotify_user_setup - Our initialization function. Note that we cannnot return
777 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
778 * must result in panic().
780 static int __init fanotify_user_setup(void)
782 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
783 fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
788 device_initcall(fanotify_user_setup);