xref: /linux/fs/notify/fanotify/fanotify_user.c (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
1 #include <linux/fanotify.h>
2 #include <linux/fcntl.h>
3 #include <linux/file.h>
4 #include <linux/fs.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>
16 #include <linux/compat.h>
17 
18 #include <asm/ioctls.h>
19 
20 #include "../../mount.h"
21 #include "../fdinfo.h"
22 #include "fanotify.h"
23 
24 #define FANOTIFY_DEFAULT_MAX_EVENTS	16384
25 #define FANOTIFY_DEFAULT_MAX_MARKS	8192
26 #define FANOTIFY_DEFAULT_MAX_LISTENERS	128
27 
28 /*
29  * All flags that may be specified in parameter event_f_flags of fanotify_init.
30  *
31  * Internal and external open flags are stored together in field f_flags of
32  * struct file. Only external open flags shall be allowed in event_f_flags.
33  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
34  * excluded.
35  */
36 #define	FANOTIFY_INIT_ALL_EVENT_F_BITS				( \
37 		O_ACCMODE	| O_APPEND	| O_NONBLOCK	| \
38 		__O_SYNC	| O_DSYNC	| O_CLOEXEC     | \
39 		O_LARGEFILE	| O_NOATIME	)
40 
41 extern const struct fsnotify_ops fanotify_fsnotify_ops;
42 
43 static struct kmem_cache *fanotify_mark_cache __read_mostly;
44 struct kmem_cache *fanotify_event_cachep __read_mostly;
45 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
46 
47 /*
48  * Get an fsnotify notification event if one exists and is small
49  * enough to fit in "count". Return an error pointer if the count
50  * is not large enough.
51  *
52  * Called with the group->notification_mutex held.
53  */
54 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
55 					    size_t count)
56 {
57 	BUG_ON(!mutex_is_locked(&group->notification_mutex));
58 
59 	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
60 
61 	if (fsnotify_notify_queue_is_empty(group))
62 		return NULL;
63 
64 	if (FAN_EVENT_METADATA_LEN > count)
65 		return ERR_PTR(-EINVAL);
66 
67 	/* held the notification_mutex the whole time, so this is the
68 	 * same event we peeked above */
69 	return fsnotify_remove_first_event(group);
70 }
71 
72 static int create_fd(struct fsnotify_group *group,
73 		     struct fanotify_event_info *event,
74 		     struct file **file)
75 {
76 	int client_fd;
77 	struct file *new_file;
78 
79 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
80 
81 	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
82 	if (client_fd < 0)
83 		return client_fd;
84 
85 	/*
86 	 * we need a new file handle for the userspace program so it can read even if it was
87 	 * originally opened O_WRONLY.
88 	 */
89 	/* it's possible this event was an overflow event.  in that case dentry and mnt
90 	 * are NULL;  That's fine, just don't call dentry open */
91 	if (event->path.dentry && event->path.mnt)
92 		new_file = dentry_open(&event->path,
93 				       group->fanotify_data.f_flags | FMODE_NONOTIFY,
94 				       current_cred());
95 	else
96 		new_file = ERR_PTR(-EOVERFLOW);
97 	if (IS_ERR(new_file)) {
98 		/*
99 		 * we still send an event even if we can't open the file.  this
100 		 * can happen when say tasks are gone and we try to open their
101 		 * /proc files or we try to open a WRONLY file like in sysfs
102 		 * we just send the errno to userspace since there isn't much
103 		 * else we can do.
104 		 */
105 		put_unused_fd(client_fd);
106 		client_fd = PTR_ERR(new_file);
107 	} else {
108 		*file = new_file;
109 	}
110 
111 	return client_fd;
112 }
113 
114 static int fill_event_metadata(struct fsnotify_group *group,
115 			       struct fanotify_event_metadata *metadata,
116 			       struct fsnotify_event *fsn_event,
117 			       struct file **file)
118 {
119 	int ret = 0;
120 	struct fanotify_event_info *event;
121 
122 	pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
123 		 group, metadata, fsn_event);
124 
125 	*file = NULL;
126 	event = container_of(fsn_event, struct fanotify_event_info, fse);
127 	metadata->event_len = FAN_EVENT_METADATA_LEN;
128 	metadata->metadata_len = FAN_EVENT_METADATA_LEN;
129 	metadata->vers = FANOTIFY_METADATA_VERSION;
130 	metadata->reserved = 0;
131 	metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
132 	metadata->pid = pid_vnr(event->tgid);
133 	if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
134 		metadata->fd = FAN_NOFD;
135 	else {
136 		metadata->fd = create_fd(group, event, file);
137 		if (metadata->fd < 0)
138 			ret = metadata->fd;
139 	}
140 
141 	return ret;
142 }
143 
144 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
145 static struct fanotify_perm_event_info *dequeue_event(
146 				struct fsnotify_group *group, int fd)
147 {
148 	struct fanotify_perm_event_info *event, *return_e = NULL;
149 
150 	spin_lock(&group->fanotify_data.access_lock);
151 	list_for_each_entry(event, &group->fanotify_data.access_list,
152 			    fae.fse.list) {
153 		if (event->fd != fd)
154 			continue;
155 
156 		list_del_init(&event->fae.fse.list);
157 		return_e = event;
158 		break;
159 	}
160 	spin_unlock(&group->fanotify_data.access_lock);
161 
162 	pr_debug("%s: found return_re=%p\n", __func__, return_e);
163 
164 	return return_e;
165 }
166 
167 static int process_access_response(struct fsnotify_group *group,
168 				   struct fanotify_response *response_struct)
169 {
170 	struct fanotify_perm_event_info *event;
171 	int fd = response_struct->fd;
172 	int response = response_struct->response;
173 
174 	pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
175 		 fd, response);
176 	/*
177 	 * make sure the response is valid, if invalid we do nothing and either
178 	 * userspace can send a valid response or we will clean it up after the
179 	 * timeout
180 	 */
181 	switch (response) {
182 	case FAN_ALLOW:
183 	case FAN_DENY:
184 		break;
185 	default:
186 		return -EINVAL;
187 	}
188 
189 	if (fd < 0)
190 		return -EINVAL;
191 
192 	event = dequeue_event(group, fd);
193 	if (!event)
194 		return -ENOENT;
195 
196 	event->response = response;
197 	wake_up(&group->fanotify_data.access_waitq);
198 
199 	return 0;
200 }
201 #endif
202 
203 static ssize_t copy_event_to_user(struct fsnotify_group *group,
204 				  struct fsnotify_event *event,
205 				  char __user *buf)
206 {
207 	struct fanotify_event_metadata fanotify_event_metadata;
208 	struct file *f;
209 	int fd, ret;
210 
211 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
212 
213 	ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
214 	if (ret < 0)
215 		return ret;
216 
217 	fd = fanotify_event_metadata.fd;
218 	ret = -EFAULT;
219 	if (copy_to_user(buf, &fanotify_event_metadata,
220 			 fanotify_event_metadata.event_len))
221 		goto out_close_fd;
222 
223 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
224 	if (event->mask & FAN_ALL_PERM_EVENTS)
225 		FANOTIFY_PE(event)->fd = fd;
226 #endif
227 
228 	if (fd != FAN_NOFD)
229 		fd_install(fd, f);
230 	return fanotify_event_metadata.event_len;
231 
232 out_close_fd:
233 	if (fd != FAN_NOFD) {
234 		put_unused_fd(fd);
235 		fput(f);
236 	}
237 	return ret;
238 }
239 
240 /* intofiy userspace file descriptor functions */
241 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
242 {
243 	struct fsnotify_group *group = file->private_data;
244 	int ret = 0;
245 
246 	poll_wait(file, &group->notification_waitq, wait);
247 	mutex_lock(&group->notification_mutex);
248 	if (!fsnotify_notify_queue_is_empty(group))
249 		ret = POLLIN | POLLRDNORM;
250 	mutex_unlock(&group->notification_mutex);
251 
252 	return ret;
253 }
254 
255 static ssize_t fanotify_read(struct file *file, char __user *buf,
256 			     size_t count, loff_t *pos)
257 {
258 	struct fsnotify_group *group;
259 	struct fsnotify_event *kevent;
260 	char __user *start;
261 	int ret;
262 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
263 
264 	start = buf;
265 	group = file->private_data;
266 
267 	pr_debug("%s: group=%p\n", __func__, group);
268 
269 	add_wait_queue(&group->notification_waitq, &wait);
270 	while (1) {
271 		mutex_lock(&group->notification_mutex);
272 		kevent = get_one_event(group, count);
273 		mutex_unlock(&group->notification_mutex);
274 
275 		if (IS_ERR(kevent)) {
276 			ret = PTR_ERR(kevent);
277 			break;
278 		}
279 
280 		if (!kevent) {
281 			ret = -EAGAIN;
282 			if (file->f_flags & O_NONBLOCK)
283 				break;
284 
285 			ret = -ERESTARTSYS;
286 			if (signal_pending(current))
287 				break;
288 
289 			if (start != buf)
290 				break;
291 
292 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
293 			continue;
294 		}
295 
296 		ret = copy_event_to_user(group, kevent, buf);
297 		/*
298 		 * Permission events get queued to wait for response.  Other
299 		 * events can be destroyed now.
300 		 */
301 		if (!(kevent->mask & FAN_ALL_PERM_EVENTS)) {
302 			fsnotify_destroy_event(group, kevent);
303 			if (ret < 0)
304 				break;
305 		} else {
306 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
307 			if (ret < 0) {
308 				FANOTIFY_PE(kevent)->response = FAN_DENY;
309 				wake_up(&group->fanotify_data.access_waitq);
310 				break;
311 			}
312 			spin_lock(&group->fanotify_data.access_lock);
313 			list_add_tail(&kevent->list,
314 				      &group->fanotify_data.access_list);
315 			spin_unlock(&group->fanotify_data.access_lock);
316 #endif
317 		}
318 		buf += ret;
319 		count -= ret;
320 	}
321 	remove_wait_queue(&group->notification_waitq, &wait);
322 
323 	if (start != buf && ret != -EFAULT)
324 		ret = buf - start;
325 	return ret;
326 }
327 
328 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
329 {
330 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
331 	struct fanotify_response response = { .fd = -1, .response = -1 };
332 	struct fsnotify_group *group;
333 	int ret;
334 
335 	group = file->private_data;
336 
337 	if (count > sizeof(response))
338 		count = sizeof(response);
339 
340 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
341 
342 	if (copy_from_user(&response, buf, count))
343 		return -EFAULT;
344 
345 	ret = process_access_response(group, &response);
346 	if (ret < 0)
347 		count = ret;
348 
349 	return count;
350 #else
351 	return -EINVAL;
352 #endif
353 }
354 
355 static int fanotify_release(struct inode *ignored, struct file *file)
356 {
357 	struct fsnotify_group *group = file->private_data;
358 
359 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
360 	struct fanotify_perm_event_info *event, *next;
361 
362 	/*
363 	 * There may be still new events arriving in the notification queue
364 	 * but since userspace cannot use fanotify fd anymore, no event can
365 	 * enter or leave access_list by now.
366 	 */
367 	spin_lock(&group->fanotify_data.access_lock);
368 
369 	atomic_inc(&group->fanotify_data.bypass_perm);
370 
371 	list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
372 				 fae.fse.list) {
373 		pr_debug("%s: found group=%p event=%p\n", __func__, group,
374 			 event);
375 
376 		list_del_init(&event->fae.fse.list);
377 		event->response = FAN_ALLOW;
378 	}
379 	spin_unlock(&group->fanotify_data.access_lock);
380 
381 	/*
382 	 * Since bypass_perm is set, newly queued events will not wait for
383 	 * access response. Wake up the already sleeping ones now.
384 	 * synchronize_srcu() in fsnotify_destroy_group() will wait for all
385 	 * processes sleeping in fanotify_handle_event() waiting for access
386 	 * response and thus also for all permission events to be freed.
387 	 */
388 	wake_up(&group->fanotify_data.access_waitq);
389 #endif
390 
391 	/* matches the fanotify_init->fsnotify_alloc_group */
392 	fsnotify_destroy_group(group);
393 
394 	return 0;
395 }
396 
397 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
398 {
399 	struct fsnotify_group *group;
400 	struct fsnotify_event *fsn_event;
401 	void __user *p;
402 	int ret = -ENOTTY;
403 	size_t send_len = 0;
404 
405 	group = file->private_data;
406 
407 	p = (void __user *) arg;
408 
409 	switch (cmd) {
410 	case FIONREAD:
411 		mutex_lock(&group->notification_mutex);
412 		list_for_each_entry(fsn_event, &group->notification_list, list)
413 			send_len += FAN_EVENT_METADATA_LEN;
414 		mutex_unlock(&group->notification_mutex);
415 		ret = put_user(send_len, (int __user *) p);
416 		break;
417 	}
418 
419 	return ret;
420 }
421 
422 static const struct file_operations fanotify_fops = {
423 	.show_fdinfo	= fanotify_show_fdinfo,
424 	.poll		= fanotify_poll,
425 	.read		= fanotify_read,
426 	.write		= fanotify_write,
427 	.fasync		= NULL,
428 	.release	= fanotify_release,
429 	.unlocked_ioctl	= fanotify_ioctl,
430 	.compat_ioctl	= fanotify_ioctl,
431 	.llseek		= noop_llseek,
432 };
433 
434 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
435 {
436 	kmem_cache_free(fanotify_mark_cache, fsn_mark);
437 }
438 
439 static int fanotify_find_path(int dfd, const char __user *filename,
440 			      struct path *path, unsigned int flags)
441 {
442 	int ret;
443 
444 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
445 		 dfd, filename, flags);
446 
447 	if (filename == NULL) {
448 		struct fd f = fdget(dfd);
449 
450 		ret = -EBADF;
451 		if (!f.file)
452 			goto out;
453 
454 		ret = -ENOTDIR;
455 		if ((flags & FAN_MARK_ONLYDIR) &&
456 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
457 			fdput(f);
458 			goto out;
459 		}
460 
461 		*path = f.file->f_path;
462 		path_get(path);
463 		fdput(f);
464 	} else {
465 		unsigned int lookup_flags = 0;
466 
467 		if (!(flags & FAN_MARK_DONT_FOLLOW))
468 			lookup_flags |= LOOKUP_FOLLOW;
469 		if (flags & FAN_MARK_ONLYDIR)
470 			lookup_flags |= LOOKUP_DIRECTORY;
471 
472 		ret = user_path_at(dfd, filename, lookup_flags, path);
473 		if (ret)
474 			goto out;
475 	}
476 
477 	/* you can only watch an inode if you have read permissions on it */
478 	ret = inode_permission(path->dentry->d_inode, MAY_READ);
479 	if (ret)
480 		path_put(path);
481 out:
482 	return ret;
483 }
484 
485 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
486 					    __u32 mask,
487 					    unsigned int flags,
488 					    int *destroy)
489 {
490 	__u32 oldmask;
491 
492 	spin_lock(&fsn_mark->lock);
493 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
494 		oldmask = fsn_mark->mask;
495 		fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
496 	} else {
497 		oldmask = fsn_mark->ignored_mask;
498 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
499 	}
500 	spin_unlock(&fsn_mark->lock);
501 
502 	*destroy = !(oldmask & ~mask);
503 
504 	return mask & oldmask;
505 }
506 
507 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
508 					 struct vfsmount *mnt, __u32 mask,
509 					 unsigned int flags)
510 {
511 	struct fsnotify_mark *fsn_mark = NULL;
512 	__u32 removed;
513 	int destroy_mark;
514 
515 	mutex_lock(&group->mark_mutex);
516 	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
517 	if (!fsn_mark) {
518 		mutex_unlock(&group->mark_mutex);
519 		return -ENOENT;
520 	}
521 
522 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
523 						 &destroy_mark);
524 	if (destroy_mark)
525 		fsnotify_destroy_mark_locked(fsn_mark, group);
526 	mutex_unlock(&group->mark_mutex);
527 
528 	fsnotify_put_mark(fsn_mark);
529 	if (removed & real_mount(mnt)->mnt_fsnotify_mask)
530 		fsnotify_recalc_vfsmount_mask(mnt);
531 
532 	return 0;
533 }
534 
535 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
536 				      struct inode *inode, __u32 mask,
537 				      unsigned int flags)
538 {
539 	struct fsnotify_mark *fsn_mark = NULL;
540 	__u32 removed;
541 	int destroy_mark;
542 
543 	mutex_lock(&group->mark_mutex);
544 	fsn_mark = fsnotify_find_inode_mark(group, inode);
545 	if (!fsn_mark) {
546 		mutex_unlock(&group->mark_mutex);
547 		return -ENOENT;
548 	}
549 
550 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
551 						 &destroy_mark);
552 	if (destroy_mark)
553 		fsnotify_destroy_mark_locked(fsn_mark, group);
554 	mutex_unlock(&group->mark_mutex);
555 
556 	/* matches the fsnotify_find_inode_mark() */
557 	fsnotify_put_mark(fsn_mark);
558 	if (removed & inode->i_fsnotify_mask)
559 		fsnotify_recalc_inode_mask(inode);
560 
561 	return 0;
562 }
563 
564 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
565 				       __u32 mask,
566 				       unsigned int flags)
567 {
568 	__u32 oldmask = -1;
569 
570 	spin_lock(&fsn_mark->lock);
571 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
572 		oldmask = fsn_mark->mask;
573 		fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
574 	} else {
575 		__u32 tmask = fsn_mark->ignored_mask | mask;
576 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
577 		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
578 			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
579 	}
580 
581 	if (!(flags & FAN_MARK_ONDIR)) {
582 		__u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
583 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
584 	}
585 
586 	spin_unlock(&fsn_mark->lock);
587 
588 	return mask & ~oldmask;
589 }
590 
591 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
592 						   struct inode *inode,
593 						   struct vfsmount *mnt)
594 {
595 	struct fsnotify_mark *mark;
596 	int ret;
597 
598 	if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
599 		return ERR_PTR(-ENOSPC);
600 
601 	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
602 	if (!mark)
603 		return ERR_PTR(-ENOMEM);
604 
605 	fsnotify_init_mark(mark, fanotify_free_mark);
606 	ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
607 	if (ret) {
608 		fsnotify_put_mark(mark);
609 		return ERR_PTR(ret);
610 	}
611 
612 	return mark;
613 }
614 
615 
616 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
617 				      struct vfsmount *mnt, __u32 mask,
618 				      unsigned int flags)
619 {
620 	struct fsnotify_mark *fsn_mark;
621 	__u32 added;
622 
623 	mutex_lock(&group->mark_mutex);
624 	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
625 	if (!fsn_mark) {
626 		fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
627 		if (IS_ERR(fsn_mark)) {
628 			mutex_unlock(&group->mark_mutex);
629 			return PTR_ERR(fsn_mark);
630 		}
631 	}
632 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
633 	mutex_unlock(&group->mark_mutex);
634 
635 	if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
636 		fsnotify_recalc_vfsmount_mask(mnt);
637 
638 	fsnotify_put_mark(fsn_mark);
639 	return 0;
640 }
641 
642 static int fanotify_add_inode_mark(struct fsnotify_group *group,
643 				   struct inode *inode, __u32 mask,
644 				   unsigned int flags)
645 {
646 	struct fsnotify_mark *fsn_mark;
647 	__u32 added;
648 
649 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
650 
651 	/*
652 	 * If some other task has this inode open for write we should not add
653 	 * an ignored mark, unless that ignored mark is supposed to survive
654 	 * modification changes anyway.
655 	 */
656 	if ((flags & FAN_MARK_IGNORED_MASK) &&
657 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
658 	    (atomic_read(&inode->i_writecount) > 0))
659 		return 0;
660 
661 	mutex_lock(&group->mark_mutex);
662 	fsn_mark = fsnotify_find_inode_mark(group, inode);
663 	if (!fsn_mark) {
664 		fsn_mark = fanotify_add_new_mark(group, inode, NULL);
665 		if (IS_ERR(fsn_mark)) {
666 			mutex_unlock(&group->mark_mutex);
667 			return PTR_ERR(fsn_mark);
668 		}
669 	}
670 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
671 	mutex_unlock(&group->mark_mutex);
672 
673 	if (added & ~inode->i_fsnotify_mask)
674 		fsnotify_recalc_inode_mask(inode);
675 
676 	fsnotify_put_mark(fsn_mark);
677 	return 0;
678 }
679 
680 /* fanotify syscalls */
681 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
682 {
683 	struct fsnotify_group *group;
684 	int f_flags, fd;
685 	struct user_struct *user;
686 	struct fanotify_event_info *oevent;
687 
688 	pr_debug("%s: flags=%d event_f_flags=%d\n",
689 		__func__, flags, event_f_flags);
690 
691 	if (!capable(CAP_SYS_ADMIN))
692 		return -EPERM;
693 
694 	if (flags & ~FAN_ALL_INIT_FLAGS)
695 		return -EINVAL;
696 
697 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
698 		return -EINVAL;
699 
700 	switch (event_f_flags & O_ACCMODE) {
701 	case O_RDONLY:
702 	case O_RDWR:
703 	case O_WRONLY:
704 		break;
705 	default:
706 		return -EINVAL;
707 	}
708 
709 	user = get_current_user();
710 	if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
711 		free_uid(user);
712 		return -EMFILE;
713 	}
714 
715 	f_flags = O_RDWR | FMODE_NONOTIFY;
716 	if (flags & FAN_CLOEXEC)
717 		f_flags |= O_CLOEXEC;
718 	if (flags & FAN_NONBLOCK)
719 		f_flags |= O_NONBLOCK;
720 
721 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
722 	group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
723 	if (IS_ERR(group)) {
724 		free_uid(user);
725 		return PTR_ERR(group);
726 	}
727 
728 	group->fanotify_data.user = user;
729 	atomic_inc(&user->fanotify_listeners);
730 
731 	oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
732 	if (unlikely(!oevent)) {
733 		fd = -ENOMEM;
734 		goto out_destroy_group;
735 	}
736 	group->overflow_event = &oevent->fse;
737 
738 	if (force_o_largefile())
739 		event_f_flags |= O_LARGEFILE;
740 	group->fanotify_data.f_flags = event_f_flags;
741 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
742 	spin_lock_init(&group->fanotify_data.access_lock);
743 	init_waitqueue_head(&group->fanotify_data.access_waitq);
744 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
745 	atomic_set(&group->fanotify_data.bypass_perm, 0);
746 #endif
747 	switch (flags & FAN_ALL_CLASS_BITS) {
748 	case FAN_CLASS_NOTIF:
749 		group->priority = FS_PRIO_0;
750 		break;
751 	case FAN_CLASS_CONTENT:
752 		group->priority = FS_PRIO_1;
753 		break;
754 	case FAN_CLASS_PRE_CONTENT:
755 		group->priority = FS_PRIO_2;
756 		break;
757 	default:
758 		fd = -EINVAL;
759 		goto out_destroy_group;
760 	}
761 
762 	if (flags & FAN_UNLIMITED_QUEUE) {
763 		fd = -EPERM;
764 		if (!capable(CAP_SYS_ADMIN))
765 			goto out_destroy_group;
766 		group->max_events = UINT_MAX;
767 	} else {
768 		group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
769 	}
770 
771 	if (flags & FAN_UNLIMITED_MARKS) {
772 		fd = -EPERM;
773 		if (!capable(CAP_SYS_ADMIN))
774 			goto out_destroy_group;
775 		group->fanotify_data.max_marks = UINT_MAX;
776 	} else {
777 		group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
778 	}
779 
780 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
781 	if (fd < 0)
782 		goto out_destroy_group;
783 
784 	return fd;
785 
786 out_destroy_group:
787 	fsnotify_destroy_group(group);
788 	return fd;
789 }
790 
791 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
792 			      __u64, mask, int, dfd,
793 			      const char  __user *, pathname)
794 {
795 	struct inode *inode = NULL;
796 	struct vfsmount *mnt = NULL;
797 	struct fsnotify_group *group;
798 	struct fd f;
799 	struct path path;
800 	int ret;
801 
802 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
803 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
804 
805 	/* we only use the lower 32 bits as of right now. */
806 	if (mask & ((__u64)0xffffffff << 32))
807 		return -EINVAL;
808 
809 	if (flags & ~FAN_ALL_MARK_FLAGS)
810 		return -EINVAL;
811 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
812 	case FAN_MARK_ADD:		/* fallthrough */
813 	case FAN_MARK_REMOVE:
814 		if (!mask)
815 			return -EINVAL;
816 		break;
817 	case FAN_MARK_FLUSH:
818 		if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
819 			return -EINVAL;
820 		break;
821 	default:
822 		return -EINVAL;
823 	}
824 
825 	if (mask & FAN_ONDIR) {
826 		flags |= FAN_MARK_ONDIR;
827 		mask &= ~FAN_ONDIR;
828 	}
829 
830 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
831 	if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
832 #else
833 	if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
834 #endif
835 		return -EINVAL;
836 
837 	f = fdget(fanotify_fd);
838 	if (unlikely(!f.file))
839 		return -EBADF;
840 
841 	/* verify that this is indeed an fanotify instance */
842 	ret = -EINVAL;
843 	if (unlikely(f.file->f_op != &fanotify_fops))
844 		goto fput_and_out;
845 	group = f.file->private_data;
846 
847 	/*
848 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
849 	 * allowed to set permissions events.
850 	 */
851 	ret = -EINVAL;
852 	if (mask & FAN_ALL_PERM_EVENTS &&
853 	    group->priority == FS_PRIO_0)
854 		goto fput_and_out;
855 
856 	if (flags & FAN_MARK_FLUSH) {
857 		ret = 0;
858 		if (flags & FAN_MARK_MOUNT)
859 			fsnotify_clear_vfsmount_marks_by_group(group);
860 		else
861 			fsnotify_clear_inode_marks_by_group(group);
862 		goto fput_and_out;
863 	}
864 
865 	ret = fanotify_find_path(dfd, pathname, &path, flags);
866 	if (ret)
867 		goto fput_and_out;
868 
869 	/* inode held in place by reference to path; group by fget on fd */
870 	if (!(flags & FAN_MARK_MOUNT))
871 		inode = path.dentry->d_inode;
872 	else
873 		mnt = path.mnt;
874 
875 	/* create/update an inode mark */
876 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
877 	case FAN_MARK_ADD:
878 		if (flags & FAN_MARK_MOUNT)
879 			ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
880 		else
881 			ret = fanotify_add_inode_mark(group, inode, mask, flags);
882 		break;
883 	case FAN_MARK_REMOVE:
884 		if (flags & FAN_MARK_MOUNT)
885 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
886 		else
887 			ret = fanotify_remove_inode_mark(group, inode, mask, flags);
888 		break;
889 	default:
890 		ret = -EINVAL;
891 	}
892 
893 	path_put(&path);
894 fput_and_out:
895 	fdput(f);
896 	return ret;
897 }
898 
899 #ifdef CONFIG_COMPAT
900 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
901 				int, fanotify_fd, unsigned int, flags,
902 				__u32, mask0, __u32, mask1, int, dfd,
903 				const char  __user *, pathname)
904 {
905 	return sys_fanotify_mark(fanotify_fd, flags,
906 #ifdef __BIG_ENDIAN
907 				((__u64)mask0 << 32) | mask1,
908 #else
909 				((__u64)mask1 << 32) | mask0,
910 #endif
911 				 dfd, pathname);
912 }
913 #endif
914 
915 /*
916  * fanotify_user_setup - Our initialization function.  Note that we cannot return
917  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
918  * must result in panic().
919  */
920 static int __init fanotify_user_setup(void)
921 {
922 	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
923 	fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
924 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
925 	fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
926 						SLAB_PANIC);
927 #endif
928 
929 	return 0;
930 }
931 device_initcall(fanotify_user_setup);
932