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 #include "../../mount.h"
20 #include "../fdinfo.h"
22 #define FANOTIFY_DEFAULT_MAX_EVENTS 16384
23 #define FANOTIFY_DEFAULT_MAX_MARKS 8192
24 #define FANOTIFY_DEFAULT_MAX_LISTENERS 128
26 extern const struct fsnotify_ops fanotify_fsnotify_ops;
28 static struct kmem_cache *fanotify_mark_cache __read_mostly;
29 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
31 struct fanotify_response_event {
32 struct list_head list;
34 struct fsnotify_event *event;
38 * Get an fsnotify notification event if one exists and is small
39 * enough to fit in "count". Return an error pointer if the count
40 * is not large enough.
42 * Called with the group->notification_mutex held.
44 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
47 BUG_ON(!mutex_is_locked(&group->notification_mutex));
49 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
51 if (fsnotify_notify_queue_is_empty(group))
54 if (FAN_EVENT_METADATA_LEN > count)
55 return ERR_PTR(-EINVAL);
57 /* held the notification_mutex the whole time, so this is the
58 * same event we peeked above */
59 return fsnotify_remove_notify_event(group);
62 static int create_fd(struct fsnotify_group *group,
63 struct fsnotify_event *event,
67 struct file *new_file;
69 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
71 client_fd = get_unused_fd();
75 if (event->data_type != FSNOTIFY_EVENT_PATH) {
77 put_unused_fd(client_fd);
82 * we need a new file handle for the userspace program so it can read even if it was
83 * originally opened O_WRONLY.
85 /* it's possible this event was an overflow event. in that case dentry and mnt
86 * are NULL; That's fine, just don't call dentry open */
87 if (event->path.dentry && event->path.mnt)
88 new_file = dentry_open(&event->path,
89 group->fanotify_data.f_flags | FMODE_NONOTIFY,
92 new_file = ERR_PTR(-EOVERFLOW);
93 if (IS_ERR(new_file)) {
95 * we still send an event even if we can't open the file. this
96 * can happen when say tasks are gone and we try to open their
97 * /proc files or we try to open a WRONLY file like in sysfs
98 * we just send the errno to userspace since there isn't much
101 put_unused_fd(client_fd);
102 client_fd = PTR_ERR(new_file);
110 static int fill_event_metadata(struct fsnotify_group *group,
111 struct fanotify_event_metadata *metadata,
112 struct fsnotify_event *event,
117 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
118 group, metadata, event);
121 metadata->event_len = FAN_EVENT_METADATA_LEN;
122 metadata->metadata_len = FAN_EVENT_METADATA_LEN;
123 metadata->vers = FANOTIFY_METADATA_VERSION;
124 metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
125 metadata->pid = pid_vnr(event->tgid);
126 if (unlikely(event->mask & FAN_Q_OVERFLOW))
127 metadata->fd = FAN_NOFD;
129 metadata->fd = create_fd(group, event, file);
130 if (metadata->fd < 0)
137 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
138 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
141 struct fanotify_response_event *re, *return_re = NULL;
143 mutex_lock(&group->fanotify_data.access_mutex);
144 list_for_each_entry(re, &group->fanotify_data.access_list, list) {
148 list_del_init(&re->list);
152 mutex_unlock(&group->fanotify_data.access_mutex);
154 pr_debug("%s: found return_re=%p\n", __func__, return_re);
159 static int process_access_response(struct fsnotify_group *group,
160 struct fanotify_response *response_struct)
162 struct fanotify_response_event *re;
163 __s32 fd = response_struct->fd;
164 __u32 response = response_struct->response;
166 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
169 * make sure the response is valid, if invalid we do nothing and either
170 * userspace can send a valid response or we will clean it up after the
184 re = dequeue_re(group, fd);
188 re->event->response = response;
190 wake_up(&group->fanotify_data.access_waitq);
192 kmem_cache_free(fanotify_response_event_cache, re);
197 static int prepare_for_access_response(struct fsnotify_group *group,
198 struct fsnotify_event *event,
201 struct fanotify_response_event *re;
203 if (!(event->mask & FAN_ALL_PERM_EVENTS))
206 re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
213 mutex_lock(&group->fanotify_data.access_mutex);
215 if (atomic_read(&group->fanotify_data.bypass_perm)) {
216 mutex_unlock(&group->fanotify_data.access_mutex);
217 kmem_cache_free(fanotify_response_event_cache, re);
218 event->response = FAN_ALLOW;
222 list_add_tail(&re->list, &group->fanotify_data.access_list);
223 mutex_unlock(&group->fanotify_data.access_mutex);
229 static int prepare_for_access_response(struct fsnotify_group *group,
230 struct fsnotify_event *event,
238 static ssize_t copy_event_to_user(struct fsnotify_group *group,
239 struct fsnotify_event *event,
242 struct fanotify_event_metadata fanotify_event_metadata;
246 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
248 ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
252 fd = fanotify_event_metadata.fd;
254 if (copy_to_user(buf, &fanotify_event_metadata,
255 fanotify_event_metadata.event_len))
258 ret = prepare_for_access_response(group, event, fd);
264 return fanotify_event_metadata.event_len;
267 if (fd != FAN_NOFD) {
272 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
273 if (event->mask & FAN_ALL_PERM_EVENTS) {
274 event->response = FAN_DENY;
275 wake_up(&group->fanotify_data.access_waitq);
281 /* intofiy userspace file descriptor functions */
282 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
284 struct fsnotify_group *group = file->private_data;
287 poll_wait(file, &group->notification_waitq, wait);
288 mutex_lock(&group->notification_mutex);
289 if (!fsnotify_notify_queue_is_empty(group))
290 ret = POLLIN | POLLRDNORM;
291 mutex_unlock(&group->notification_mutex);
296 static ssize_t fanotify_read(struct file *file, char __user *buf,
297 size_t count, loff_t *pos)
299 struct fsnotify_group *group;
300 struct fsnotify_event *kevent;
306 group = file->private_data;
308 pr_debug("%s: group=%p\n", __func__, group);
311 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
313 mutex_lock(&group->notification_mutex);
314 kevent = get_one_event(group, count);
315 mutex_unlock(&group->notification_mutex);
318 ret = PTR_ERR(kevent);
321 ret = copy_event_to_user(group, kevent, buf);
322 fsnotify_put_event(kevent);
331 if (file->f_flags & O_NONBLOCK)
334 if (signal_pending(current))
343 finish_wait(&group->notification_waitq, &wait);
344 if (start != buf && ret != -EFAULT)
349 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
351 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
352 struct fanotify_response response = { .fd = -1, .response = -1 };
353 struct fsnotify_group *group;
356 group = file->private_data;
358 if (count > sizeof(response))
359 count = sizeof(response);
361 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
363 if (copy_from_user(&response, buf, count))
366 ret = process_access_response(group, &response);
376 static int fanotify_release(struct inode *ignored, struct file *file)
378 struct fsnotify_group *group = file->private_data;
380 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
381 struct fanotify_response_event *re, *lre;
383 mutex_lock(&group->fanotify_data.access_mutex);
385 atomic_inc(&group->fanotify_data.bypass_perm);
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);
401 if (file->f_flags & FASYNC)
402 fsnotify_fasync(-1, file, 0);
404 /* matches the fanotify_init->fsnotify_alloc_group */
405 fsnotify_destroy_group(group);
410 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
412 struct fsnotify_group *group;
413 struct fsnotify_event_holder *holder;
418 group = file->private_data;
420 p = (void __user *) arg;
424 mutex_lock(&group->notification_mutex);
425 list_for_each_entry(holder, &group->notification_list, event_list)
426 send_len += FAN_EVENT_METADATA_LEN;
427 mutex_unlock(&group->notification_mutex);
428 ret = put_user(send_len, (int __user *) p);
435 static const struct file_operations fanotify_fops = {
436 .show_fdinfo = fanotify_show_fdinfo,
437 .poll = fanotify_poll,
438 .read = fanotify_read,
439 .write = fanotify_write,
441 .release = fanotify_release,
442 .unlocked_ioctl = fanotify_ioctl,
443 .compat_ioctl = fanotify_ioctl,
444 .llseek = noop_llseek,
447 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
449 kmem_cache_free(fanotify_mark_cache, fsn_mark);
452 static int fanotify_find_path(int dfd, const char __user *filename,
453 struct path *path, unsigned int flags)
457 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
458 dfd, filename, flags);
460 if (filename == NULL) {
461 struct fd f = fdget(dfd);
468 if ((flags & FAN_MARK_ONLYDIR) &&
469 !(S_ISDIR(file_inode(f.file)->i_mode))) {
474 *path = f.file->f_path;
478 unsigned int lookup_flags = 0;
480 if (!(flags & FAN_MARK_DONT_FOLLOW))
481 lookup_flags |= LOOKUP_FOLLOW;
482 if (flags & FAN_MARK_ONLYDIR)
483 lookup_flags |= LOOKUP_DIRECTORY;
485 ret = user_path_at(dfd, filename, lookup_flags, path);
490 /* you can only watch an inode if you have read permissions on it */
491 ret = inode_permission(path->dentry->d_inode, MAY_READ);
498 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
505 spin_lock(&fsn_mark->lock);
506 if (!(flags & FAN_MARK_IGNORED_MASK)) {
507 oldmask = fsn_mark->mask;
508 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
510 oldmask = fsn_mark->ignored_mask;
511 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
513 spin_unlock(&fsn_mark->lock);
515 *destroy = !(oldmask & ~mask);
517 return mask & oldmask;
520 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
521 struct vfsmount *mnt, __u32 mask,
524 struct fsnotify_mark *fsn_mark = NULL;
528 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
532 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
535 fsnotify_destroy_mark(fsn_mark, group);
537 fsnotify_put_mark(fsn_mark);
538 if (removed & real_mount(mnt)->mnt_fsnotify_mask)
539 fsnotify_recalc_vfsmount_mask(mnt);
544 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
545 struct inode *inode, __u32 mask,
548 struct fsnotify_mark *fsn_mark = NULL;
552 fsn_mark = fsnotify_find_inode_mark(group, inode);
556 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
559 fsnotify_destroy_mark(fsn_mark, group);
560 /* matches the fsnotify_find_inode_mark() */
561 fsnotify_put_mark(fsn_mark);
562 if (removed & inode->i_fsnotify_mask)
563 fsnotify_recalc_inode_mask(inode);
568 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
574 spin_lock(&fsn_mark->lock);
575 if (!(flags & FAN_MARK_IGNORED_MASK)) {
576 oldmask = fsn_mark->mask;
577 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
579 __u32 tmask = fsn_mark->ignored_mask | mask;
580 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
581 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
582 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
585 if (!(flags & FAN_MARK_ONDIR)) {
586 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
587 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
590 spin_unlock(&fsn_mark->lock);
592 return mask & ~oldmask;
595 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
596 struct vfsmount *mnt, __u32 mask,
599 struct fsnotify_mark *fsn_mark;
603 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
605 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
608 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
612 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
613 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
617 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
619 if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
620 fsnotify_recalc_vfsmount_mask(mnt);
622 fsnotify_put_mark(fsn_mark);
626 static int fanotify_add_inode_mark(struct fsnotify_group *group,
627 struct inode *inode, __u32 mask,
630 struct fsnotify_mark *fsn_mark;
634 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
637 * If some other task has this inode open for write we should not add
638 * an ignored mark, unless that ignored mark is supposed to survive
639 * modification changes anyway.
641 if ((flags & FAN_MARK_IGNORED_MASK) &&
642 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
643 (atomic_read(&inode->i_writecount) > 0))
646 fsn_mark = fsnotify_find_inode_mark(group, inode);
648 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
651 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
655 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
656 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
660 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
662 if (added & ~inode->i_fsnotify_mask)
663 fsnotify_recalc_inode_mask(inode);
665 fsnotify_put_mark(fsn_mark);
669 /* fanotify syscalls */
670 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
672 struct fsnotify_group *group;
674 struct user_struct *user;
676 pr_debug("%s: flags=%d event_f_flags=%d\n",
677 __func__, flags, event_f_flags);
679 if (!capable(CAP_SYS_ADMIN))
682 if (flags & ~FAN_ALL_INIT_FLAGS)
685 user = get_current_user();
686 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
691 f_flags = O_RDWR | FMODE_NONOTIFY;
692 if (flags & FAN_CLOEXEC)
693 f_flags |= O_CLOEXEC;
694 if (flags & FAN_NONBLOCK)
695 f_flags |= O_NONBLOCK;
697 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
698 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
701 return PTR_ERR(group);
704 group->fanotify_data.user = user;
705 atomic_inc(&user->fanotify_listeners);
707 group->fanotify_data.f_flags = event_f_flags;
708 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
709 mutex_init(&group->fanotify_data.access_mutex);
710 init_waitqueue_head(&group->fanotify_data.access_waitq);
711 INIT_LIST_HEAD(&group->fanotify_data.access_list);
712 atomic_set(&group->fanotify_data.bypass_perm, 0);
714 switch (flags & FAN_ALL_CLASS_BITS) {
715 case FAN_CLASS_NOTIF:
716 group->priority = FS_PRIO_0;
718 case FAN_CLASS_CONTENT:
719 group->priority = FS_PRIO_1;
721 case FAN_CLASS_PRE_CONTENT:
722 group->priority = FS_PRIO_2;
726 goto out_destroy_group;
729 if (flags & FAN_UNLIMITED_QUEUE) {
731 if (!capable(CAP_SYS_ADMIN))
732 goto out_destroy_group;
733 group->max_events = UINT_MAX;
735 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
738 if (flags & FAN_UNLIMITED_MARKS) {
740 if (!capable(CAP_SYS_ADMIN))
741 goto out_destroy_group;
742 group->fanotify_data.max_marks = UINT_MAX;
744 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
747 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
749 goto out_destroy_group;
754 fsnotify_destroy_group(group);
758 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
760 const char __user * pathname)
762 struct inode *inode = NULL;
763 struct vfsmount *mnt = NULL;
764 struct fsnotify_group *group;
769 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
770 __func__, fanotify_fd, flags, dfd, pathname, mask);
772 /* we only use the lower 32 bits as of right now. */
773 if (mask & ((__u64)0xffffffff << 32))
776 if (flags & ~FAN_ALL_MARK_FLAGS)
778 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
779 case FAN_MARK_ADD: /* fallthrough */
780 case FAN_MARK_REMOVE:
789 if (mask & FAN_ONDIR) {
790 flags |= FAN_MARK_ONDIR;
794 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
795 if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
797 if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
801 f = fdget(fanotify_fd);
802 if (unlikely(!f.file))
805 /* verify that this is indeed an fanotify instance */
807 if (unlikely(f.file->f_op != &fanotify_fops))
809 group = f.file->private_data;
812 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
813 * allowed to set permissions events.
816 if (mask & FAN_ALL_PERM_EVENTS &&
817 group->priority == FS_PRIO_0)
820 ret = fanotify_find_path(dfd, pathname, &path, flags);
824 /* inode held in place by reference to path; group by fget on fd */
825 if (!(flags & FAN_MARK_MOUNT))
826 inode = path.dentry->d_inode;
830 /* create/update an inode mark */
831 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
833 if (flags & FAN_MARK_MOUNT)
834 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
836 ret = fanotify_add_inode_mark(group, inode, mask, flags);
838 case FAN_MARK_REMOVE:
839 if (flags & FAN_MARK_MOUNT)
840 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
842 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
845 if (flags & FAN_MARK_MOUNT)
846 fsnotify_clear_vfsmount_marks_by_group(group);
848 fsnotify_clear_inode_marks_by_group(group);
860 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
861 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
862 long dfd, long pathname)
864 return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
866 (const char __user *) pathname);
868 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
872 * fanotify_user_setup - Our initialization function. Note that we cannot return
873 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
874 * must result in panic().
876 static int __init fanotify_user_setup(void)
878 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
879 fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
884 device_initcall(fanotify_user_setup);