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_notify_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();
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(wait);
263 
264 	start = buf;
265 	group = file->private_data;
266 
267 	pr_debug("%s: group=%p\n", __func__, group);
268 
269 	while (1) {
270 		prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
271 
272 		mutex_lock(&group->notification_mutex);
273 		kevent = get_one_event(group, count);
274 		mutex_unlock(&group->notification_mutex);
275 
276 		if (IS_ERR(kevent)) {
277 			ret = PTR_ERR(kevent);
278 			break;
279 		}
280 
281 		if (!kevent) {
282 			ret = -EAGAIN;
283 			if (file->f_flags & O_NONBLOCK)
284 				break;
285 
286 			ret = -ERESTARTSYS;
287 			if (signal_pending(current))
288 				break;
289 
290 			if (start != buf)
291 				break;
292 			schedule();
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 
322 	finish_wait(&group->notification_waitq, &wait);
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 	spin_lock(&group->fanotify_data.access_lock);
363 
364 	atomic_inc(&group->fanotify_data.bypass_perm);
365 
366 	list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
367 				 fae.fse.list) {
368 		pr_debug("%s: found group=%p event=%p\n", __func__, group,
369 			 event);
370 
371 		list_del_init(&event->fae.fse.list);
372 		event->response = FAN_ALLOW;
373 	}
374 	spin_unlock(&group->fanotify_data.access_lock);
375 
376 	wake_up(&group->fanotify_data.access_waitq);
377 #endif
378 
379 	/* matches the fanotify_init->fsnotify_alloc_group */
380 	fsnotify_destroy_group(group);
381 
382 	return 0;
383 }
384 
385 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
386 {
387 	struct fsnotify_group *group;
388 	struct fsnotify_event *fsn_event;
389 	void __user *p;
390 	int ret = -ENOTTY;
391 	size_t send_len = 0;
392 
393 	group = file->private_data;
394 
395 	p = (void __user *) arg;
396 
397 	switch (cmd) {
398 	case FIONREAD:
399 		mutex_lock(&group->notification_mutex);
400 		list_for_each_entry(fsn_event, &group->notification_list, list)
401 			send_len += FAN_EVENT_METADATA_LEN;
402 		mutex_unlock(&group->notification_mutex);
403 		ret = put_user(send_len, (int __user *) p);
404 		break;
405 	}
406 
407 	return ret;
408 }
409 
410 static const struct file_operations fanotify_fops = {
411 	.show_fdinfo	= fanotify_show_fdinfo,
412 	.poll		= fanotify_poll,
413 	.read		= fanotify_read,
414 	.write		= fanotify_write,
415 	.fasync		= NULL,
416 	.release	= fanotify_release,
417 	.unlocked_ioctl	= fanotify_ioctl,
418 	.compat_ioctl	= fanotify_ioctl,
419 	.llseek		= noop_llseek,
420 };
421 
422 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
423 {
424 	kmem_cache_free(fanotify_mark_cache, fsn_mark);
425 }
426 
427 static int fanotify_find_path(int dfd, const char __user *filename,
428 			      struct path *path, unsigned int flags)
429 {
430 	int ret;
431 
432 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
433 		 dfd, filename, flags);
434 
435 	if (filename == NULL) {
436 		struct fd f = fdget(dfd);
437 
438 		ret = -EBADF;
439 		if (!f.file)
440 			goto out;
441 
442 		ret = -ENOTDIR;
443 		if ((flags & FAN_MARK_ONLYDIR) &&
444 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
445 			fdput(f);
446 			goto out;
447 		}
448 
449 		*path = f.file->f_path;
450 		path_get(path);
451 		fdput(f);
452 	} else {
453 		unsigned int lookup_flags = 0;
454 
455 		if (!(flags & FAN_MARK_DONT_FOLLOW))
456 			lookup_flags |= LOOKUP_FOLLOW;
457 		if (flags & FAN_MARK_ONLYDIR)
458 			lookup_flags |= LOOKUP_DIRECTORY;
459 
460 		ret = user_path_at(dfd, filename, lookup_flags, path);
461 		if (ret)
462 			goto out;
463 	}
464 
465 	/* you can only watch an inode if you have read permissions on it */
466 	ret = inode_permission(path->dentry->d_inode, MAY_READ);
467 	if (ret)
468 		path_put(path);
469 out:
470 	return ret;
471 }
472 
473 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
474 					    __u32 mask,
475 					    unsigned int flags,
476 					    int *destroy)
477 {
478 	__u32 oldmask;
479 
480 	spin_lock(&fsn_mark->lock);
481 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
482 		oldmask = fsn_mark->mask;
483 		fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
484 	} else {
485 		oldmask = fsn_mark->ignored_mask;
486 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
487 	}
488 	spin_unlock(&fsn_mark->lock);
489 
490 	*destroy = !(oldmask & ~mask);
491 
492 	return mask & oldmask;
493 }
494 
495 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
496 					 struct vfsmount *mnt, __u32 mask,
497 					 unsigned int flags)
498 {
499 	struct fsnotify_mark *fsn_mark = NULL;
500 	__u32 removed;
501 	int destroy_mark;
502 
503 	mutex_lock(&group->mark_mutex);
504 	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
505 	if (!fsn_mark) {
506 		mutex_unlock(&group->mark_mutex);
507 		return -ENOENT;
508 	}
509 
510 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
511 						 &destroy_mark);
512 	if (destroy_mark)
513 		fsnotify_destroy_mark_locked(fsn_mark, group);
514 	mutex_unlock(&group->mark_mutex);
515 
516 	fsnotify_put_mark(fsn_mark);
517 	if (removed & real_mount(mnt)->mnt_fsnotify_mask)
518 		fsnotify_recalc_vfsmount_mask(mnt);
519 
520 	return 0;
521 }
522 
523 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
524 				      struct inode *inode, __u32 mask,
525 				      unsigned int flags)
526 {
527 	struct fsnotify_mark *fsn_mark = NULL;
528 	__u32 removed;
529 	int destroy_mark;
530 
531 	mutex_lock(&group->mark_mutex);
532 	fsn_mark = fsnotify_find_inode_mark(group, inode);
533 	if (!fsn_mark) {
534 		mutex_unlock(&group->mark_mutex);
535 		return -ENOENT;
536 	}
537 
538 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
539 						 &destroy_mark);
540 	if (destroy_mark)
541 		fsnotify_destroy_mark_locked(fsn_mark, group);
542 	mutex_unlock(&group->mark_mutex);
543 
544 	/* matches the fsnotify_find_inode_mark() */
545 	fsnotify_put_mark(fsn_mark);
546 	if (removed & inode->i_fsnotify_mask)
547 		fsnotify_recalc_inode_mask(inode);
548 
549 	return 0;
550 }
551 
552 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
553 				       __u32 mask,
554 				       unsigned int flags)
555 {
556 	__u32 oldmask = -1;
557 
558 	spin_lock(&fsn_mark->lock);
559 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
560 		oldmask = fsn_mark->mask;
561 		fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
562 	} else {
563 		__u32 tmask = fsn_mark->ignored_mask | mask;
564 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
565 		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
566 			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
567 	}
568 
569 	if (!(flags & FAN_MARK_ONDIR)) {
570 		__u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
571 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
572 	}
573 
574 	spin_unlock(&fsn_mark->lock);
575 
576 	return mask & ~oldmask;
577 }
578 
579 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
580 						   struct inode *inode,
581 						   struct vfsmount *mnt)
582 {
583 	struct fsnotify_mark *mark;
584 	int ret;
585 
586 	if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
587 		return ERR_PTR(-ENOSPC);
588 
589 	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
590 	if (!mark)
591 		return ERR_PTR(-ENOMEM);
592 
593 	fsnotify_init_mark(mark, fanotify_free_mark);
594 	ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
595 	if (ret) {
596 		fsnotify_put_mark(mark);
597 		return ERR_PTR(ret);
598 	}
599 
600 	return mark;
601 }
602 
603 
604 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
605 				      struct vfsmount *mnt, __u32 mask,
606 				      unsigned int flags)
607 {
608 	struct fsnotify_mark *fsn_mark;
609 	__u32 added;
610 
611 	mutex_lock(&group->mark_mutex);
612 	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
613 	if (!fsn_mark) {
614 		fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
615 		if (IS_ERR(fsn_mark)) {
616 			mutex_unlock(&group->mark_mutex);
617 			return PTR_ERR(fsn_mark);
618 		}
619 	}
620 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
621 	mutex_unlock(&group->mark_mutex);
622 
623 	if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
624 		fsnotify_recalc_vfsmount_mask(mnt);
625 
626 	fsnotify_put_mark(fsn_mark);
627 	return 0;
628 }
629 
630 static int fanotify_add_inode_mark(struct fsnotify_group *group,
631 				   struct inode *inode, __u32 mask,
632 				   unsigned int flags)
633 {
634 	struct fsnotify_mark *fsn_mark;
635 	__u32 added;
636 
637 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
638 
639 	/*
640 	 * If some other task has this inode open for write we should not add
641 	 * an ignored mark, unless that ignored mark is supposed to survive
642 	 * modification changes anyway.
643 	 */
644 	if ((flags & FAN_MARK_IGNORED_MASK) &&
645 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
646 	    (atomic_read(&inode->i_writecount) > 0))
647 		return 0;
648 
649 	mutex_lock(&group->mark_mutex);
650 	fsn_mark = fsnotify_find_inode_mark(group, inode);
651 	if (!fsn_mark) {
652 		fsn_mark = fanotify_add_new_mark(group, inode, NULL);
653 		if (IS_ERR(fsn_mark)) {
654 			mutex_unlock(&group->mark_mutex);
655 			return PTR_ERR(fsn_mark);
656 		}
657 	}
658 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
659 	mutex_unlock(&group->mark_mutex);
660 
661 	if (added & ~inode->i_fsnotify_mask)
662 		fsnotify_recalc_inode_mask(inode);
663 
664 	fsnotify_put_mark(fsn_mark);
665 	return 0;
666 }
667 
668 /* fanotify syscalls */
669 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
670 {
671 	struct fsnotify_group *group;
672 	int f_flags, fd;
673 	struct user_struct *user;
674 	struct fanotify_event_info *oevent;
675 
676 	pr_debug("%s: flags=%d event_f_flags=%d\n",
677 		__func__, flags, event_f_flags);
678 
679 	if (!capable(CAP_SYS_ADMIN))
680 		return -EPERM;
681 
682 	if (flags & ~FAN_ALL_INIT_FLAGS)
683 		return -EINVAL;
684 
685 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
686 		return -EINVAL;
687 
688 	switch (event_f_flags & O_ACCMODE) {
689 	case O_RDONLY:
690 	case O_RDWR:
691 	case O_WRONLY:
692 		break;
693 	default:
694 		return -EINVAL;
695 	}
696 
697 	user = get_current_user();
698 	if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
699 		free_uid(user);
700 		return -EMFILE;
701 	}
702 
703 	f_flags = O_RDWR | FMODE_NONOTIFY;
704 	if (flags & FAN_CLOEXEC)
705 		f_flags |= O_CLOEXEC;
706 	if (flags & FAN_NONBLOCK)
707 		f_flags |= O_NONBLOCK;
708 
709 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
710 	group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
711 	if (IS_ERR(group)) {
712 		free_uid(user);
713 		return PTR_ERR(group);
714 	}
715 
716 	group->fanotify_data.user = user;
717 	atomic_inc(&user->fanotify_listeners);
718 
719 	oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
720 	if (unlikely(!oevent)) {
721 		fd = -ENOMEM;
722 		goto out_destroy_group;
723 	}
724 	group->overflow_event = &oevent->fse;
725 
726 	if (force_o_largefile())
727 		event_f_flags |= O_LARGEFILE;
728 	group->fanotify_data.f_flags = event_f_flags;
729 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
730 	spin_lock_init(&group->fanotify_data.access_lock);
731 	init_waitqueue_head(&group->fanotify_data.access_waitq);
732 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
733 	atomic_set(&group->fanotify_data.bypass_perm, 0);
734 #endif
735 	switch (flags & FAN_ALL_CLASS_BITS) {
736 	case FAN_CLASS_NOTIF:
737 		group->priority = FS_PRIO_0;
738 		break;
739 	case FAN_CLASS_CONTENT:
740 		group->priority = FS_PRIO_1;
741 		break;
742 	case FAN_CLASS_PRE_CONTENT:
743 		group->priority = FS_PRIO_2;
744 		break;
745 	default:
746 		fd = -EINVAL;
747 		goto out_destroy_group;
748 	}
749 
750 	if (flags & FAN_UNLIMITED_QUEUE) {
751 		fd = -EPERM;
752 		if (!capable(CAP_SYS_ADMIN))
753 			goto out_destroy_group;
754 		group->max_events = UINT_MAX;
755 	} else {
756 		group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
757 	}
758 
759 	if (flags & FAN_UNLIMITED_MARKS) {
760 		fd = -EPERM;
761 		if (!capable(CAP_SYS_ADMIN))
762 			goto out_destroy_group;
763 		group->fanotify_data.max_marks = UINT_MAX;
764 	} else {
765 		group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
766 	}
767 
768 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
769 	if (fd < 0)
770 		goto out_destroy_group;
771 
772 	return fd;
773 
774 out_destroy_group:
775 	fsnotify_destroy_group(group);
776 	return fd;
777 }
778 
779 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
780 			      __u64, mask, int, dfd,
781 			      const char  __user *, pathname)
782 {
783 	struct inode *inode = NULL;
784 	struct vfsmount *mnt = NULL;
785 	struct fsnotify_group *group;
786 	struct fd f;
787 	struct path path;
788 	int ret;
789 
790 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
791 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
792 
793 	/* we only use the lower 32 bits as of right now. */
794 	if (mask & ((__u64)0xffffffff << 32))
795 		return -EINVAL;
796 
797 	if (flags & ~FAN_ALL_MARK_FLAGS)
798 		return -EINVAL;
799 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
800 	case FAN_MARK_ADD:		/* fallthrough */
801 	case FAN_MARK_REMOVE:
802 		if (!mask)
803 			return -EINVAL;
804 		break;
805 	case FAN_MARK_FLUSH:
806 		if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
807 			return -EINVAL;
808 		break;
809 	default:
810 		return -EINVAL;
811 	}
812 
813 	if (mask & FAN_ONDIR) {
814 		flags |= FAN_MARK_ONDIR;
815 		mask &= ~FAN_ONDIR;
816 	}
817 
818 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
819 	if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
820 #else
821 	if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
822 #endif
823 		return -EINVAL;
824 
825 	f = fdget(fanotify_fd);
826 	if (unlikely(!f.file))
827 		return -EBADF;
828 
829 	/* verify that this is indeed an fanotify instance */
830 	ret = -EINVAL;
831 	if (unlikely(f.file->f_op != &fanotify_fops))
832 		goto fput_and_out;
833 	group = f.file->private_data;
834 
835 	/*
836 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
837 	 * allowed to set permissions events.
838 	 */
839 	ret = -EINVAL;
840 	if (mask & FAN_ALL_PERM_EVENTS &&
841 	    group->priority == FS_PRIO_0)
842 		goto fput_and_out;
843 
844 	if (flags & FAN_MARK_FLUSH) {
845 		ret = 0;
846 		if (flags & FAN_MARK_MOUNT)
847 			fsnotify_clear_vfsmount_marks_by_group(group);
848 		else
849 			fsnotify_clear_inode_marks_by_group(group);
850 		goto fput_and_out;
851 	}
852 
853 	ret = fanotify_find_path(dfd, pathname, &path, flags);
854 	if (ret)
855 		goto fput_and_out;
856 
857 	/* inode held in place by reference to path; group by fget on fd */
858 	if (!(flags & FAN_MARK_MOUNT))
859 		inode = path.dentry->d_inode;
860 	else
861 		mnt = path.mnt;
862 
863 	/* create/update an inode mark */
864 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
865 	case FAN_MARK_ADD:
866 		if (flags & FAN_MARK_MOUNT)
867 			ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
868 		else
869 			ret = fanotify_add_inode_mark(group, inode, mask, flags);
870 		break;
871 	case FAN_MARK_REMOVE:
872 		if (flags & FAN_MARK_MOUNT)
873 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
874 		else
875 			ret = fanotify_remove_inode_mark(group, inode, mask, flags);
876 		break;
877 	default:
878 		ret = -EINVAL;
879 	}
880 
881 	path_put(&path);
882 fput_and_out:
883 	fdput(f);
884 	return ret;
885 }
886 
887 #ifdef CONFIG_COMPAT
888 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
889 				int, fanotify_fd, unsigned int, flags,
890 				__u32, mask0, __u32, mask1, int, dfd,
891 				const char  __user *, pathname)
892 {
893 	return sys_fanotify_mark(fanotify_fd, flags,
894 #ifdef __BIG_ENDIAN
895 				((__u64)mask0 << 32) | mask1,
896 #else
897 				((__u64)mask1 << 32) | mask0,
898 #endif
899 				 dfd, pathname);
900 }
901 #endif
902 
903 /*
904  * fanotify_user_setup - Our initialization function.  Note that we cannot return
905  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
906  * must result in panic().
907  */
908 static int __init fanotify_user_setup(void)
909 {
910 	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
911 	fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
912 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
913 	fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
914 						SLAB_PANIC);
915 #endif
916 
917 	return 0;
918 }
919 device_initcall(fanotify_user_setup);
920