xref: /linux/fs/notify/fanotify/fanotify_user.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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 #include <linux/sched/signal.h>
18 
19 #include <asm/ioctls.h>
20 
21 #include "../../mount.h"
22 #include "../fdinfo.h"
23 #include "fanotify.h"
24 
25 #define FANOTIFY_DEFAULT_MAX_EVENTS	16384
26 #define FANOTIFY_DEFAULT_MAX_MARKS	8192
27 #define FANOTIFY_DEFAULT_MAX_LISTENERS	128
28 
29 /*
30  * All flags that may be specified in parameter event_f_flags of fanotify_init.
31  *
32  * Internal and external open flags are stored together in field f_flags of
33  * struct file. Only external open flags shall be allowed in event_f_flags.
34  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
35  * excluded.
36  */
37 #define	FANOTIFY_INIT_ALL_EVENT_F_BITS				( \
38 		O_ACCMODE	| O_APPEND	| O_NONBLOCK	| \
39 		__O_SYNC	| O_DSYNC	| O_CLOEXEC     | \
40 		O_LARGEFILE	| O_NOATIME	)
41 
42 extern const struct fsnotify_ops fanotify_fsnotify_ops;
43 
44 static struct kmem_cache *fanotify_mark_cache __read_mostly;
45 struct kmem_cache *fanotify_event_cachep __read_mostly;
46 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
47 
48 /*
49  * Get an fsnotify notification event if one exists and is small
50  * enough to fit in "count". Return an error pointer if the count
51  * is not large enough.
52  *
53  * Called with the group->notification_lock held.
54  */
55 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
56 					    size_t count)
57 {
58 	assert_spin_locked(&group->notification_lock);
59 
60 	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
61 
62 	if (fsnotify_notify_queue_is_empty(group))
63 		return NULL;
64 
65 	if (FAN_EVENT_METADATA_LEN > count)
66 		return ERR_PTR(-EINVAL);
67 
68 	/* held the notification_lock the whole time, so this is the
69 	 * same event we peeked above */
70 	return fsnotify_remove_first_event(group);
71 }
72 
73 static int create_fd(struct fsnotify_group *group,
74 		     struct fanotify_event_info *event,
75 		     struct file **file)
76 {
77 	int client_fd;
78 	struct file *new_file;
79 
80 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
81 
82 	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
83 	if (client_fd < 0)
84 		return client_fd;
85 
86 	/*
87 	 * we need a new file handle for the userspace program so it can read even if it was
88 	 * originally opened O_WRONLY.
89 	 */
90 	/* it's possible this event was an overflow event.  in that case dentry and mnt
91 	 * are NULL;  That's fine, just don't call dentry open */
92 	if (event->path.dentry && event->path.mnt)
93 		new_file = dentry_open(&event->path,
94 				       group->fanotify_data.f_flags | FMODE_NONOTIFY,
95 				       current_cred());
96 	else
97 		new_file = ERR_PTR(-EOVERFLOW);
98 	if (IS_ERR(new_file)) {
99 		/*
100 		 * we still send an event even if we can't open the file.  this
101 		 * can happen when say tasks are gone and we try to open their
102 		 * /proc files or we try to open a WRONLY file like in sysfs
103 		 * we just send the errno to userspace since there isn't much
104 		 * else we can do.
105 		 */
106 		put_unused_fd(client_fd);
107 		client_fd = PTR_ERR(new_file);
108 	} else {
109 		*file = new_file;
110 	}
111 
112 	return client_fd;
113 }
114 
115 static int fill_event_metadata(struct fsnotify_group *group,
116 			       struct fanotify_event_metadata *metadata,
117 			       struct fsnotify_event *fsn_event,
118 			       struct file **file)
119 {
120 	int ret = 0;
121 	struct fanotify_event_info *event;
122 
123 	pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
124 		 group, metadata, fsn_event);
125 
126 	*file = NULL;
127 	event = container_of(fsn_event, struct fanotify_event_info, fse);
128 	metadata->event_len = FAN_EVENT_METADATA_LEN;
129 	metadata->metadata_len = FAN_EVENT_METADATA_LEN;
130 	metadata->vers = FANOTIFY_METADATA_VERSION;
131 	metadata->reserved = 0;
132 	metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
133 	metadata->pid = pid_vnr(event->tgid);
134 	if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
135 		metadata->fd = FAN_NOFD;
136 	else {
137 		metadata->fd = create_fd(group, event, file);
138 		if (metadata->fd < 0)
139 			ret = metadata->fd;
140 	}
141 
142 	return ret;
143 }
144 
145 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
146 static struct fanotify_perm_event_info *dequeue_event(
147 				struct fsnotify_group *group, int fd)
148 {
149 	struct fanotify_perm_event_info *event, *return_e = NULL;
150 
151 	spin_lock(&group->notification_lock);
152 	list_for_each_entry(event, &group->fanotify_data.access_list,
153 			    fae.fse.list) {
154 		if (event->fd != fd)
155 			continue;
156 
157 		list_del_init(&event->fae.fse.list);
158 		return_e = event;
159 		break;
160 	}
161 	spin_unlock(&group->notification_lock);
162 
163 	pr_debug("%s: found return_re=%p\n", __func__, return_e);
164 
165 	return return_e;
166 }
167 
168 static int process_access_response(struct fsnotify_group *group,
169 				   struct fanotify_response *response_struct)
170 {
171 	struct fanotify_perm_event_info *event;
172 	int fd = response_struct->fd;
173 	int response = response_struct->response;
174 
175 	pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
176 		 fd, response);
177 	/*
178 	 * make sure the response is valid, if invalid we do nothing and either
179 	 * userspace can send a valid response or we will clean it up after the
180 	 * timeout
181 	 */
182 	switch (response) {
183 	case FAN_ALLOW:
184 	case FAN_DENY:
185 		break;
186 	default:
187 		return -EINVAL;
188 	}
189 
190 	if (fd < 0)
191 		return -EINVAL;
192 
193 	event = dequeue_event(group, fd);
194 	if (!event)
195 		return -ENOENT;
196 
197 	event->response = response;
198 	wake_up(&group->fanotify_data.access_waitq);
199 
200 	return 0;
201 }
202 #endif
203 
204 static ssize_t copy_event_to_user(struct fsnotify_group *group,
205 				  struct fsnotify_event *event,
206 				  char __user *buf)
207 {
208 	struct fanotify_event_metadata fanotify_event_metadata;
209 	struct file *f;
210 	int fd, ret;
211 
212 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
213 
214 	ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
215 	if (ret < 0)
216 		return ret;
217 
218 	fd = fanotify_event_metadata.fd;
219 	ret = -EFAULT;
220 	if (copy_to_user(buf, &fanotify_event_metadata,
221 			 fanotify_event_metadata.event_len))
222 		goto out_close_fd;
223 
224 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
225 	if (event->mask & FAN_ALL_PERM_EVENTS)
226 		FANOTIFY_PE(event)->fd = fd;
227 #endif
228 
229 	if (fd != FAN_NOFD)
230 		fd_install(fd, f);
231 	return fanotify_event_metadata.event_len;
232 
233 out_close_fd:
234 	if (fd != FAN_NOFD) {
235 		put_unused_fd(fd);
236 		fput(f);
237 	}
238 	return ret;
239 }
240 
241 /* intofiy userspace file descriptor functions */
242 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
243 {
244 	struct fsnotify_group *group = file->private_data;
245 	int ret = 0;
246 
247 	poll_wait(file, &group->notification_waitq, wait);
248 	spin_lock(&group->notification_lock);
249 	if (!fsnotify_notify_queue_is_empty(group))
250 		ret = POLLIN | POLLRDNORM;
251 	spin_unlock(&group->notification_lock);
252 
253 	return ret;
254 }
255 
256 static ssize_t fanotify_read(struct file *file, char __user *buf,
257 			     size_t count, loff_t *pos)
258 {
259 	struct fsnotify_group *group;
260 	struct fsnotify_event *kevent;
261 	char __user *start;
262 	int ret;
263 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
264 
265 	start = buf;
266 	group = file->private_data;
267 
268 	pr_debug("%s: group=%p\n", __func__, group);
269 
270 	add_wait_queue(&group->notification_waitq, &wait);
271 	while (1) {
272 		spin_lock(&group->notification_lock);
273 		kevent = get_one_event(group, count);
274 		spin_unlock(&group->notification_lock);
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 
293 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
294 			continue;
295 		}
296 
297 		ret = copy_event_to_user(group, kevent, buf);
298 		/*
299 		 * Permission events get queued to wait for response.  Other
300 		 * events can be destroyed now.
301 		 */
302 		if (!(kevent->mask & FAN_ALL_PERM_EVENTS)) {
303 			fsnotify_destroy_event(group, kevent);
304 			if (ret < 0)
305 				break;
306 		} else {
307 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
308 			if (ret < 0) {
309 				FANOTIFY_PE(kevent)->response = FAN_DENY;
310 				wake_up(&group->fanotify_data.access_waitq);
311 				break;
312 			}
313 			spin_lock(&group->notification_lock);
314 			list_add_tail(&kevent->list,
315 				      &group->fanotify_data.access_list);
316 			spin_unlock(&group->notification_lock);
317 #endif
318 		}
319 		buf += ret;
320 		count -= ret;
321 	}
322 	remove_wait_queue(&group->notification_waitq, &wait);
323 
324 	if (start != buf && ret != -EFAULT)
325 		ret = buf - start;
326 	return ret;
327 }
328 
329 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
330 {
331 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
332 	struct fanotify_response response = { .fd = -1, .response = -1 };
333 	struct fsnotify_group *group;
334 	int ret;
335 
336 	group = file->private_data;
337 
338 	if (count > sizeof(response))
339 		count = sizeof(response);
340 
341 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
342 
343 	if (copy_from_user(&response, buf, count))
344 		return -EFAULT;
345 
346 	ret = process_access_response(group, &response);
347 	if (ret < 0)
348 		count = ret;
349 
350 	return count;
351 #else
352 	return -EINVAL;
353 #endif
354 }
355 
356 static int fanotify_release(struct inode *ignored, struct file *file)
357 {
358 	struct fsnotify_group *group = file->private_data;
359 
360 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
361 	struct fanotify_perm_event_info *event, *next;
362 	struct fsnotify_event *fsn_event;
363 
364 	/*
365 	 * Stop new events from arriving in the notification queue. since
366 	 * userspace cannot use fanotify fd anymore, no event can enter or
367 	 * leave access_list by now either.
368 	 */
369 	fsnotify_group_stop_queueing(group);
370 
371 	/*
372 	 * Process all permission events on access_list and notification queue
373 	 * and simulate reply from userspace.
374 	 */
375 	spin_lock(&group->notification_lock);
376 	list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
377 				 fae.fse.list) {
378 		pr_debug("%s: found group=%p event=%p\n", __func__, group,
379 			 event);
380 
381 		list_del_init(&event->fae.fse.list);
382 		event->response = FAN_ALLOW;
383 	}
384 
385 	/*
386 	 * Destroy all non-permission events. For permission events just
387 	 * dequeue them and set the response. They will be freed once the
388 	 * response is consumed and fanotify_get_response() returns.
389 	 */
390 	while (!fsnotify_notify_queue_is_empty(group)) {
391 		fsn_event = fsnotify_remove_first_event(group);
392 		if (!(fsn_event->mask & FAN_ALL_PERM_EVENTS)) {
393 			spin_unlock(&group->notification_lock);
394 			fsnotify_destroy_event(group, fsn_event);
395 			spin_lock(&group->notification_lock);
396 		} else
397 			FANOTIFY_PE(fsn_event)->response = FAN_ALLOW;
398 	}
399 	spin_unlock(&group->notification_lock);
400 
401 	/* Response for all permission events it set, wakeup waiters */
402 	wake_up(&group->fanotify_data.access_waitq);
403 #endif
404 
405 	/* matches the fanotify_init->fsnotify_alloc_group */
406 	fsnotify_destroy_group(group);
407 
408 	return 0;
409 }
410 
411 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
412 {
413 	struct fsnotify_group *group;
414 	struct fsnotify_event *fsn_event;
415 	void __user *p;
416 	int ret = -ENOTTY;
417 	size_t send_len = 0;
418 
419 	group = file->private_data;
420 
421 	p = (void __user *) arg;
422 
423 	switch (cmd) {
424 	case FIONREAD:
425 		spin_lock(&group->notification_lock);
426 		list_for_each_entry(fsn_event, &group->notification_list, list)
427 			send_len += FAN_EVENT_METADATA_LEN;
428 		spin_unlock(&group->notification_lock);
429 		ret = put_user(send_len, (int __user *) p);
430 		break;
431 	}
432 
433 	return ret;
434 }
435 
436 static const struct file_operations fanotify_fops = {
437 	.show_fdinfo	= fanotify_show_fdinfo,
438 	.poll		= fanotify_poll,
439 	.read		= fanotify_read,
440 	.write		= fanotify_write,
441 	.fasync		= NULL,
442 	.release	= fanotify_release,
443 	.unlocked_ioctl	= fanotify_ioctl,
444 	.compat_ioctl	= fanotify_ioctl,
445 	.llseek		= noop_llseek,
446 };
447 
448 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
449 {
450 	kmem_cache_free(fanotify_mark_cache, fsn_mark);
451 }
452 
453 static int fanotify_find_path(int dfd, const char __user *filename,
454 			      struct path *path, unsigned int flags)
455 {
456 	int ret;
457 
458 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
459 		 dfd, filename, flags);
460 
461 	if (filename == NULL) {
462 		struct fd f = fdget(dfd);
463 
464 		ret = -EBADF;
465 		if (!f.file)
466 			goto out;
467 
468 		ret = -ENOTDIR;
469 		if ((flags & FAN_MARK_ONLYDIR) &&
470 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
471 			fdput(f);
472 			goto out;
473 		}
474 
475 		*path = f.file->f_path;
476 		path_get(path);
477 		fdput(f);
478 	} else {
479 		unsigned int lookup_flags = 0;
480 
481 		if (!(flags & FAN_MARK_DONT_FOLLOW))
482 			lookup_flags |= LOOKUP_FOLLOW;
483 		if (flags & FAN_MARK_ONLYDIR)
484 			lookup_flags |= LOOKUP_DIRECTORY;
485 
486 		ret = user_path_at(dfd, filename, lookup_flags, path);
487 		if (ret)
488 			goto out;
489 	}
490 
491 	/* you can only watch an inode if you have read permissions on it */
492 	ret = inode_permission(path->dentry->d_inode, MAY_READ);
493 	if (ret)
494 		path_put(path);
495 out:
496 	return ret;
497 }
498 
499 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
500 					    __u32 mask,
501 					    unsigned int flags,
502 					    int *destroy)
503 {
504 	__u32 oldmask = 0;
505 
506 	spin_lock(&fsn_mark->lock);
507 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
508 		__u32 tmask = fsn_mark->mask & ~mask;
509 
510 		if (flags & FAN_MARK_ONDIR)
511 			tmask &= ~FAN_ONDIR;
512 
513 		oldmask = fsn_mark->mask;
514 		fsnotify_set_mark_mask_locked(fsn_mark, tmask);
515 	} else {
516 		__u32 tmask = fsn_mark->ignored_mask & ~mask;
517 		if (flags & FAN_MARK_ONDIR)
518 			tmask &= ~FAN_ONDIR;
519 
520 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
521 	}
522 	*destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
523 	spin_unlock(&fsn_mark->lock);
524 
525 	return mask & oldmask;
526 }
527 
528 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
529 					 struct vfsmount *mnt, __u32 mask,
530 					 unsigned int flags)
531 {
532 	struct fsnotify_mark *fsn_mark = NULL;
533 	__u32 removed;
534 	int destroy_mark;
535 
536 	mutex_lock(&group->mark_mutex);
537 	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
538 	if (!fsn_mark) {
539 		mutex_unlock(&group->mark_mutex);
540 		return -ENOENT;
541 	}
542 
543 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
544 						 &destroy_mark);
545 	if (destroy_mark)
546 		fsnotify_detach_mark(fsn_mark);
547 	mutex_unlock(&group->mark_mutex);
548 	if (destroy_mark)
549 		fsnotify_free_mark(fsn_mark);
550 
551 	fsnotify_put_mark(fsn_mark);
552 	if (removed & real_mount(mnt)->mnt_fsnotify_mask)
553 		fsnotify_recalc_vfsmount_mask(mnt);
554 
555 	return 0;
556 }
557 
558 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
559 				      struct inode *inode, __u32 mask,
560 				      unsigned int flags)
561 {
562 	struct fsnotify_mark *fsn_mark = NULL;
563 	__u32 removed;
564 	int destroy_mark;
565 
566 	mutex_lock(&group->mark_mutex);
567 	fsn_mark = fsnotify_find_inode_mark(group, inode);
568 	if (!fsn_mark) {
569 		mutex_unlock(&group->mark_mutex);
570 		return -ENOENT;
571 	}
572 
573 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
574 						 &destroy_mark);
575 	if (destroy_mark)
576 		fsnotify_detach_mark(fsn_mark);
577 	mutex_unlock(&group->mark_mutex);
578 	if (destroy_mark)
579 		fsnotify_free_mark(fsn_mark);
580 
581 	/* matches the fsnotify_find_inode_mark() */
582 	fsnotify_put_mark(fsn_mark);
583 	if (removed & inode->i_fsnotify_mask)
584 		fsnotify_recalc_inode_mask(inode);
585 
586 	return 0;
587 }
588 
589 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
590 				       __u32 mask,
591 				       unsigned int flags)
592 {
593 	__u32 oldmask = -1;
594 
595 	spin_lock(&fsn_mark->lock);
596 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
597 		__u32 tmask = fsn_mark->mask | mask;
598 
599 		if (flags & FAN_MARK_ONDIR)
600 			tmask |= FAN_ONDIR;
601 
602 		oldmask = fsn_mark->mask;
603 		fsnotify_set_mark_mask_locked(fsn_mark, tmask);
604 	} else {
605 		__u32 tmask = fsn_mark->ignored_mask | mask;
606 		if (flags & FAN_MARK_ONDIR)
607 			tmask |= FAN_ONDIR;
608 
609 		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
610 		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
611 			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
612 	}
613 	spin_unlock(&fsn_mark->lock);
614 
615 	return mask & ~oldmask;
616 }
617 
618 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
619 						   struct inode *inode,
620 						   struct vfsmount *mnt)
621 {
622 	struct fsnotify_mark *mark;
623 	int ret;
624 
625 	if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
626 		return ERR_PTR(-ENOSPC);
627 
628 	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
629 	if (!mark)
630 		return ERR_PTR(-ENOMEM);
631 
632 	fsnotify_init_mark(mark, fanotify_free_mark);
633 	ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
634 	if (ret) {
635 		fsnotify_put_mark(mark);
636 		return ERR_PTR(ret);
637 	}
638 
639 	return mark;
640 }
641 
642 
643 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
644 				      struct vfsmount *mnt, __u32 mask,
645 				      unsigned int flags)
646 {
647 	struct fsnotify_mark *fsn_mark;
648 	__u32 added;
649 
650 	mutex_lock(&group->mark_mutex);
651 	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
652 	if (!fsn_mark) {
653 		fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
654 		if (IS_ERR(fsn_mark)) {
655 			mutex_unlock(&group->mark_mutex);
656 			return PTR_ERR(fsn_mark);
657 		}
658 	}
659 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
660 	mutex_unlock(&group->mark_mutex);
661 
662 	if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
663 		fsnotify_recalc_vfsmount_mask(mnt);
664 
665 	fsnotify_put_mark(fsn_mark);
666 	return 0;
667 }
668 
669 static int fanotify_add_inode_mark(struct fsnotify_group *group,
670 				   struct inode *inode, __u32 mask,
671 				   unsigned int flags)
672 {
673 	struct fsnotify_mark *fsn_mark;
674 	__u32 added;
675 
676 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
677 
678 	/*
679 	 * If some other task has this inode open for write we should not add
680 	 * an ignored mark, unless that ignored mark is supposed to survive
681 	 * modification changes anyway.
682 	 */
683 	if ((flags & FAN_MARK_IGNORED_MASK) &&
684 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
685 	    (atomic_read(&inode->i_writecount) > 0))
686 		return 0;
687 
688 	mutex_lock(&group->mark_mutex);
689 	fsn_mark = fsnotify_find_inode_mark(group, inode);
690 	if (!fsn_mark) {
691 		fsn_mark = fanotify_add_new_mark(group, inode, NULL);
692 		if (IS_ERR(fsn_mark)) {
693 			mutex_unlock(&group->mark_mutex);
694 			return PTR_ERR(fsn_mark);
695 		}
696 	}
697 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
698 	mutex_unlock(&group->mark_mutex);
699 
700 	if (added & ~inode->i_fsnotify_mask)
701 		fsnotify_recalc_inode_mask(inode);
702 
703 	fsnotify_put_mark(fsn_mark);
704 	return 0;
705 }
706 
707 /* fanotify syscalls */
708 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
709 {
710 	struct fsnotify_group *group;
711 	int f_flags, fd;
712 	struct user_struct *user;
713 	struct fanotify_event_info *oevent;
714 
715 	pr_debug("%s: flags=%d event_f_flags=%d\n",
716 		__func__, flags, event_f_flags);
717 
718 	if (!capable(CAP_SYS_ADMIN))
719 		return -EPERM;
720 
721 	if (flags & ~FAN_ALL_INIT_FLAGS)
722 		return -EINVAL;
723 
724 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
725 		return -EINVAL;
726 
727 	switch (event_f_flags & O_ACCMODE) {
728 	case O_RDONLY:
729 	case O_RDWR:
730 	case O_WRONLY:
731 		break;
732 	default:
733 		return -EINVAL;
734 	}
735 
736 	user = get_current_user();
737 	if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
738 		free_uid(user);
739 		return -EMFILE;
740 	}
741 
742 	f_flags = O_RDWR | FMODE_NONOTIFY;
743 	if (flags & FAN_CLOEXEC)
744 		f_flags |= O_CLOEXEC;
745 	if (flags & FAN_NONBLOCK)
746 		f_flags |= O_NONBLOCK;
747 
748 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
749 	group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
750 	if (IS_ERR(group)) {
751 		free_uid(user);
752 		return PTR_ERR(group);
753 	}
754 
755 	group->fanotify_data.user = user;
756 	atomic_inc(&user->fanotify_listeners);
757 
758 	oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
759 	if (unlikely(!oevent)) {
760 		fd = -ENOMEM;
761 		goto out_destroy_group;
762 	}
763 	group->overflow_event = &oevent->fse;
764 
765 	if (force_o_largefile())
766 		event_f_flags |= O_LARGEFILE;
767 	group->fanotify_data.f_flags = event_f_flags;
768 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
769 	init_waitqueue_head(&group->fanotify_data.access_waitq);
770 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
771 #endif
772 	switch (flags & FAN_ALL_CLASS_BITS) {
773 	case FAN_CLASS_NOTIF:
774 		group->priority = FS_PRIO_0;
775 		break;
776 	case FAN_CLASS_CONTENT:
777 		group->priority = FS_PRIO_1;
778 		break;
779 	case FAN_CLASS_PRE_CONTENT:
780 		group->priority = FS_PRIO_2;
781 		break;
782 	default:
783 		fd = -EINVAL;
784 		goto out_destroy_group;
785 	}
786 
787 	if (flags & FAN_UNLIMITED_QUEUE) {
788 		fd = -EPERM;
789 		if (!capable(CAP_SYS_ADMIN))
790 			goto out_destroy_group;
791 		group->max_events = UINT_MAX;
792 	} else {
793 		group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
794 	}
795 
796 	if (flags & FAN_UNLIMITED_MARKS) {
797 		fd = -EPERM;
798 		if (!capable(CAP_SYS_ADMIN))
799 			goto out_destroy_group;
800 		group->fanotify_data.max_marks = UINT_MAX;
801 	} else {
802 		group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
803 	}
804 
805 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
806 	if (fd < 0)
807 		goto out_destroy_group;
808 
809 	return fd;
810 
811 out_destroy_group:
812 	fsnotify_destroy_group(group);
813 	return fd;
814 }
815 
816 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
817 			      __u64, mask, int, dfd,
818 			      const char  __user *, pathname)
819 {
820 	struct inode *inode = NULL;
821 	struct vfsmount *mnt = NULL;
822 	struct fsnotify_group *group;
823 	struct fd f;
824 	struct path path;
825 	int ret;
826 
827 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
828 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
829 
830 	/* we only use the lower 32 bits as of right now. */
831 	if (mask & ((__u64)0xffffffff << 32))
832 		return -EINVAL;
833 
834 	if (flags & ~FAN_ALL_MARK_FLAGS)
835 		return -EINVAL;
836 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
837 	case FAN_MARK_ADD:		/* fallthrough */
838 	case FAN_MARK_REMOVE:
839 		if (!mask)
840 			return -EINVAL;
841 		break;
842 	case FAN_MARK_FLUSH:
843 		if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
844 			return -EINVAL;
845 		break;
846 	default:
847 		return -EINVAL;
848 	}
849 
850 	if (mask & FAN_ONDIR) {
851 		flags |= FAN_MARK_ONDIR;
852 		mask &= ~FAN_ONDIR;
853 	}
854 
855 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
856 	if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
857 #else
858 	if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
859 #endif
860 		return -EINVAL;
861 
862 	f = fdget(fanotify_fd);
863 	if (unlikely(!f.file))
864 		return -EBADF;
865 
866 	/* verify that this is indeed an fanotify instance */
867 	ret = -EINVAL;
868 	if (unlikely(f.file->f_op != &fanotify_fops))
869 		goto fput_and_out;
870 	group = f.file->private_data;
871 
872 	/*
873 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
874 	 * allowed to set permissions events.
875 	 */
876 	ret = -EINVAL;
877 	if (mask & FAN_ALL_PERM_EVENTS &&
878 	    group->priority == FS_PRIO_0)
879 		goto fput_and_out;
880 
881 	if (flags & FAN_MARK_FLUSH) {
882 		ret = 0;
883 		if (flags & FAN_MARK_MOUNT)
884 			fsnotify_clear_vfsmount_marks_by_group(group);
885 		else
886 			fsnotify_clear_inode_marks_by_group(group);
887 		goto fput_and_out;
888 	}
889 
890 	ret = fanotify_find_path(dfd, pathname, &path, flags);
891 	if (ret)
892 		goto fput_and_out;
893 
894 	/* inode held in place by reference to path; group by fget on fd */
895 	if (!(flags & FAN_MARK_MOUNT))
896 		inode = path.dentry->d_inode;
897 	else
898 		mnt = path.mnt;
899 
900 	/* create/update an inode mark */
901 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
902 	case FAN_MARK_ADD:
903 		if (flags & FAN_MARK_MOUNT)
904 			ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
905 		else
906 			ret = fanotify_add_inode_mark(group, inode, mask, flags);
907 		break;
908 	case FAN_MARK_REMOVE:
909 		if (flags & FAN_MARK_MOUNT)
910 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
911 		else
912 			ret = fanotify_remove_inode_mark(group, inode, mask, flags);
913 		break;
914 	default:
915 		ret = -EINVAL;
916 	}
917 
918 	path_put(&path);
919 fput_and_out:
920 	fdput(f);
921 	return ret;
922 }
923 
924 #ifdef CONFIG_COMPAT
925 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
926 				int, fanotify_fd, unsigned int, flags,
927 				__u32, mask0, __u32, mask1, int, dfd,
928 				const char  __user *, pathname)
929 {
930 	return sys_fanotify_mark(fanotify_fd, flags,
931 #ifdef __BIG_ENDIAN
932 				((__u64)mask0 << 32) | mask1,
933 #else
934 				((__u64)mask1 << 32) | mask0,
935 #endif
936 				 dfd, pathname);
937 }
938 #endif
939 
940 /*
941  * fanotify_user_setup - Our initialization function.  Note that we cannot return
942  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
943  * must result in panic().
944  */
945 static int __init fanotify_user_setup(void)
946 {
947 	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
948 	fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
949 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
950 	fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
951 						SLAB_PANIC);
952 #endif
953 
954 	return 0;
955 }
956 device_initcall(fanotify_user_setup);
957