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 ssize_t
fill_event_metadata(struct fsnotify_group
*group
,
110 struct fanotify_event_metadata
*metadata
,
111 struct fsnotify_event
*event
)
113 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__
,
114 group
, metadata
, event
);
116 metadata
->event_len
= FAN_EVENT_METADATA_LEN
;
117 metadata
->vers
= FANOTIFY_METADATA_VERSION
;
118 metadata
->mask
= event
->mask
& FAN_ALL_OUTGOING_EVENTS
;
119 metadata
->pid
= pid_vnr(event
->tgid
);
120 metadata
->fd
= create_fd(group
, event
);
125 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
126 static struct fanotify_response_event
*dequeue_re(struct fsnotify_group
*group
,
129 struct fanotify_response_event
*re
, *return_re
= NULL
;
131 mutex_lock(&group
->fanotify_data
.access_mutex
);
132 list_for_each_entry(re
, &group
->fanotify_data
.access_list
, list
) {
136 list_del_init(&re
->list
);
140 mutex_unlock(&group
->fanotify_data
.access_mutex
);
142 pr_debug("%s: found return_re=%p\n", __func__
, return_re
);
147 static int process_access_response(struct fsnotify_group
*group
,
148 struct fanotify_response
*response_struct
)
150 struct fanotify_response_event
*re
;
151 __s32 fd
= response_struct
->fd
;
152 __u32 response
= response_struct
->response
;
154 pr_debug("%s: group=%p fd=%d response=%d\n", __func__
, group
,
157 * make sure the response is valid, if invalid we do nothing and either
158 * userspace can send a valid responce or we will clean it up after the
172 re
= dequeue_re(group
, fd
);
176 re
->event
->response
= response
;
178 wake_up(&group
->fanotify_data
.access_waitq
);
180 kmem_cache_free(fanotify_response_event_cache
, re
);
185 static int prepare_for_access_response(struct fsnotify_group
*group
,
186 struct fsnotify_event
*event
,
189 struct fanotify_response_event
*re
;
191 if (!(event
->mask
& FAN_ALL_PERM_EVENTS
))
194 re
= kmem_cache_alloc(fanotify_response_event_cache
, GFP_KERNEL
);
201 mutex_lock(&group
->fanotify_data
.access_mutex
);
203 if (group
->fanotify_data
.bypass_perm
) {
204 mutex_unlock(&group
->fanotify_data
.access_mutex
);
205 kmem_cache_free(fanotify_response_event_cache
, re
);
206 event
->response
= FAN_ALLOW
;
210 list_add_tail(&re
->list
, &group
->fanotify_data
.access_list
);
211 mutex_unlock(&group
->fanotify_data
.access_mutex
);
216 static void remove_access_response(struct fsnotify_group
*group
,
217 struct fsnotify_event
*event
,
220 struct fanotify_response_event
*re
;
222 if (!(event
->mask
& FAN_ALL_PERM_EVENTS
))
225 re
= dequeue_re(group
, fd
);
229 BUG_ON(re
->event
!= event
);
231 kmem_cache_free(fanotify_response_event_cache
, re
);
236 static int prepare_for_access_response(struct fsnotify_group
*group
,
237 struct fsnotify_event
*event
,
243 static void remove_access_response(struct fsnotify_group
*group
,
244 struct fsnotify_event
*event
,
251 static ssize_t
copy_event_to_user(struct fsnotify_group
*group
,
252 struct fsnotify_event
*event
,
255 struct fanotify_event_metadata fanotify_event_metadata
;
258 pr_debug("%s: group=%p event=%p\n", __func__
, group
, event
);
260 fd
= fill_event_metadata(group
, &fanotify_event_metadata
, event
);
264 ret
= prepare_for_access_response(group
, event
, fd
);
269 if (copy_to_user(buf
, &fanotify_event_metadata
, FAN_EVENT_METADATA_LEN
))
270 goto out_kill_access_response
;
272 return FAN_EVENT_METADATA_LEN
;
274 out_kill_access_response
:
275 remove_access_response(group
, event
, fd
);
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 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 __u32 tmask
= fsn_mark
->ignored_mask
| mask
;
568 fsnotify_set_mark_ignored_mask_locked(fsn_mark
, tmask
);
569 if (flags
& FAN_MARK_IGNORED_SURV_MODIFY
)
570 fsn_mark
->flags
|= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY
;
573 if (!(flags
& FAN_MARK_ONDIR
)) {
574 __u32 tmask
= fsn_mark
->ignored_mask
| FAN_ONDIR
;
575 fsnotify_set_mark_ignored_mask_locked(fsn_mark
, tmask
);
578 spin_unlock(&fsn_mark
->lock
);
580 return mask
& ~oldmask
;
583 static int fanotify_add_vfsmount_mark(struct fsnotify_group
*group
,
584 struct vfsmount
*mnt
, __u32 mask
,
587 struct fsnotify_mark
*fsn_mark
;
590 fsn_mark
= fsnotify_find_vfsmount_mark(group
, mnt
);
594 if (atomic_read(&group
->num_marks
) > group
->fanotify_data
.max_marks
)
597 fsn_mark
= kmem_cache_alloc(fanotify_mark_cache
, GFP_KERNEL
);
601 fsnotify_init_mark(fsn_mark
, fanotify_free_mark
);
602 ret
= fsnotify_add_mark(fsn_mark
, group
, NULL
, mnt
, 0);
604 fanotify_free_mark(fsn_mark
);
608 added
= fanotify_mark_add_to_mask(fsn_mark
, mask
, flags
);
609 fsnotify_put_mark(fsn_mark
);
610 if (added
& ~mnt
->mnt_fsnotify_mask
)
611 fsnotify_recalc_vfsmount_mask(mnt
);
616 static int fanotify_add_inode_mark(struct fsnotify_group
*group
,
617 struct inode
*inode
, __u32 mask
,
620 struct fsnotify_mark
*fsn_mark
;
623 pr_debug("%s: group=%p inode=%p\n", __func__
, group
, inode
);
626 * If some other task has this inode open for write we should not add
627 * an ignored mark, unless that ignored mark is supposed to survive
628 * modification changes anyway.
630 if ((flags
& FAN_MARK_IGNORED_MASK
) &&
631 !(flags
& FAN_MARK_IGNORED_SURV_MODIFY
) &&
632 (atomic_read(&inode
->i_writecount
) > 0))
635 fsn_mark
= fsnotify_find_inode_mark(group
, inode
);
639 if (atomic_read(&group
->num_marks
) > group
->fanotify_data
.max_marks
)
642 fsn_mark
= kmem_cache_alloc(fanotify_mark_cache
, GFP_KERNEL
);
646 fsnotify_init_mark(fsn_mark
, fanotify_free_mark
);
647 ret
= fsnotify_add_mark(fsn_mark
, group
, inode
, NULL
, 0);
649 fanotify_free_mark(fsn_mark
);
653 added
= fanotify_mark_add_to_mask(fsn_mark
, mask
, flags
);
654 fsnotify_put_mark(fsn_mark
);
655 if (added
& ~inode
->i_fsnotify_mask
)
656 fsnotify_recalc_inode_mask(inode
);
660 /* fanotify syscalls */
661 SYSCALL_DEFINE2(fanotify_init
, unsigned int, flags
, unsigned int, event_f_flags
)
663 struct fsnotify_group
*group
;
665 struct user_struct
*user
;
667 pr_debug("%s: flags=%d event_f_flags=%d\n",
668 __func__
, flags
, event_f_flags
);
670 if (!capable(CAP_SYS_ADMIN
))
673 if (flags
& ~FAN_ALL_INIT_FLAGS
)
676 user
= get_current_user();
677 if (atomic_read(&user
->fanotify_listeners
) > FANOTIFY_DEFAULT_MAX_LISTENERS
) {
682 f_flags
= O_RDWR
| FMODE_NONOTIFY
;
683 if (flags
& FAN_CLOEXEC
)
684 f_flags
|= O_CLOEXEC
;
685 if (flags
& FAN_NONBLOCK
)
686 f_flags
|= O_NONBLOCK
;
688 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
689 group
= fsnotify_alloc_group(&fanotify_fsnotify_ops
);
691 return PTR_ERR(group
);
693 group
->fanotify_data
.user
= user
;
694 atomic_inc(&user
->fanotify_listeners
);
696 group
->fanotify_data
.f_flags
= event_f_flags
;
697 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
698 mutex_init(&group
->fanotify_data
.access_mutex
);
699 init_waitqueue_head(&group
->fanotify_data
.access_waitq
);
700 INIT_LIST_HEAD(&group
->fanotify_data
.access_list
);
702 switch (flags
& FAN_ALL_CLASS_BITS
) {
703 case FAN_CLASS_NOTIF
:
704 group
->priority
= FS_PRIO_0
;
706 case FAN_CLASS_CONTENT
:
707 group
->priority
= FS_PRIO_1
;
709 case FAN_CLASS_PRE_CONTENT
:
710 group
->priority
= FS_PRIO_2
;
717 if (flags
& FAN_UNLIMITED_QUEUE
) {
719 if (!capable(CAP_SYS_ADMIN
))
721 group
->max_events
= UINT_MAX
;
723 group
->max_events
= FANOTIFY_DEFAULT_MAX_EVENTS
;
726 if (flags
& FAN_UNLIMITED_MARKS
) {
728 if (!capable(CAP_SYS_ADMIN
))
730 group
->fanotify_data
.max_marks
= UINT_MAX
;
732 group
->fanotify_data
.max_marks
= FANOTIFY_DEFAULT_MAX_MARKS
;
735 fd
= anon_inode_getfd("[fanotify]", &fanotify_fops
, group
, f_flags
);
742 fsnotify_put_group(group
);
746 SYSCALL_DEFINE(fanotify_mark
)(int fanotify_fd
, unsigned int flags
,
748 const char __user
* pathname
)
750 struct inode
*inode
= NULL
;
751 struct vfsmount
*mnt
= NULL
;
752 struct fsnotify_group
*group
;
755 int ret
, fput_needed
;
757 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
758 __func__
, fanotify_fd
, flags
, dfd
, pathname
, mask
);
760 /* we only use the lower 32 bits as of right now. */
761 if (mask
& ((__u64
)0xffffffff << 32))
764 if (flags
& ~FAN_ALL_MARK_FLAGS
)
766 switch (flags
& (FAN_MARK_ADD
| FAN_MARK_REMOVE
| FAN_MARK_FLUSH
)) {
768 case FAN_MARK_REMOVE
:
775 if (mask
& FAN_ONDIR
) {
776 flags
|= FAN_MARK_ONDIR
;
780 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
781 if (mask
& ~(FAN_ALL_EVENTS
| FAN_ALL_PERM_EVENTS
| FAN_EVENT_ON_CHILD
))
783 if (mask
& ~(FAN_ALL_EVENTS
| FAN_EVENT_ON_CHILD
))
787 filp
= fget_light(fanotify_fd
, &fput_needed
);
791 /* verify that this is indeed an fanotify instance */
793 if (unlikely(filp
->f_op
!= &fanotify_fops
))
795 group
= filp
->private_data
;
798 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
799 * allowed to set permissions events.
802 if (mask
& FAN_ALL_PERM_EVENTS
&&
803 group
->priority
== FS_PRIO_0
)
806 ret
= fanotify_find_path(dfd
, pathname
, &path
, flags
);
810 /* inode held in place by reference to path; group by fget on fd */
811 if (!(flags
& FAN_MARK_MOUNT
))
812 inode
= path
.dentry
->d_inode
;
816 /* create/update an inode mark */
817 switch (flags
& (FAN_MARK_ADD
| FAN_MARK_REMOVE
| FAN_MARK_FLUSH
)) {
819 if (flags
& FAN_MARK_MOUNT
)
820 ret
= fanotify_add_vfsmount_mark(group
, mnt
, mask
, flags
);
822 ret
= fanotify_add_inode_mark(group
, inode
, mask
, flags
);
824 case FAN_MARK_REMOVE
:
825 if (flags
& FAN_MARK_MOUNT
)
826 ret
= fanotify_remove_vfsmount_mark(group
, mnt
, mask
, flags
);
828 ret
= fanotify_remove_inode_mark(group
, inode
, mask
, flags
);
831 if (flags
& FAN_MARK_MOUNT
)
832 fsnotify_clear_vfsmount_marks_by_group(group
);
834 fsnotify_clear_inode_marks_by_group(group
);
842 fput_light(filp
, fput_needed
);
846 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
847 asmlinkage
long SyS_fanotify_mark(long fanotify_fd
, long flags
, __u64 mask
,
848 long dfd
, long pathname
)
850 return SYSC_fanotify_mark((int) fanotify_fd
, (unsigned int) flags
,
852 (const char __user
*) pathname
);
854 SYSCALL_ALIAS(sys_fanotify_mark
, SyS_fanotify_mark
);
858 * fanotify_user_setup - Our initialization function. Note that we cannnot return
859 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
860 * must result in panic().
862 static int __init
fanotify_user_setup(void)
864 fanotify_mark_cache
= KMEM_CACHE(fsnotify_mark
, SLAB_PANIC
);
865 fanotify_response_event_cache
= KMEM_CACHE(fanotify_response_event
,
870 device_initcall(fanotify_user_setup
);