1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/fanotify.h>
3 #include <linux/fcntl.h>
4 #include <linux/file.h>
5 #include <linux/fs.h>
6 #include <linux/anon_inodes.h>
7 #include <linux/fsnotify_backend.h>
8 #include <linux/init.h>
9 #include <linux/mount.h>
10 #include <linux/namei.h>
11 #include <linux/poll.h>
12 #include <linux/security.h>
13 #include <linux/syscalls.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <linux/uaccess.h>
17 #include <linux/compat.h>
18 #include <linux/sched/signal.h>
19 #include <linux/memcontrol.h>
20 #include <linux/statfs.h>
21 #include <linux/exportfs.h>
22 
23 #include <asm/ioctls.h>
24 
25 #include "../../mount.h"
26 #include "../fdinfo.h"
27 #include "fanotify.h"
28 
29 #define FANOTIFY_DEFAULT_MAX_EVENTS	16384
30 #define FANOTIFY_DEFAULT_MAX_MARKS	8192
31 #define FANOTIFY_DEFAULT_MAX_LISTENERS	128
32 
33 /*
34  * All flags that may be specified in parameter event_f_flags of fanotify_init.
35  *
36  * Internal and external open flags are stored together in field f_flags of
37  * struct file. Only external open flags shall be allowed in event_f_flags.
38  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
39  * excluded.
40  */
41 #define	FANOTIFY_INIT_ALL_EVENT_F_BITS				( \
42 		O_ACCMODE	| O_APPEND	| O_NONBLOCK	| \
43 		__O_SYNC	| O_DSYNC	| O_CLOEXEC     | \
44 		O_LARGEFILE	| O_NOATIME	)
45 
46 extern const struct fsnotify_ops fanotify_fsnotify_ops;
47 
48 struct kmem_cache *fanotify_mark_cache __read_mostly;
49 struct kmem_cache *fanotify_event_cachep __read_mostly;
50 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
51 
52 #define FANOTIFY_EVENT_ALIGN 4
53 
54 static int fanotify_event_info_len(struct fanotify_event *event)
55 {
56 	if (!fanotify_event_has_fid(event))
57 		return 0;
58 
59 	return roundup(sizeof(struct fanotify_event_info_fid) +
60 		       sizeof(struct file_handle) + event->fh_len,
61 		       FANOTIFY_EVENT_ALIGN);
62 }
63 
64 /*
65  * Get an fsnotify notification event if one exists and is small
66  * enough to fit in "count". Return an error pointer if the count
67  * is not large enough. When permission event is dequeued, its state is
68  * updated accordingly.
69  */
70 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
71 					    size_t count)
72 {
73 	size_t event_size = FAN_EVENT_METADATA_LEN;
74 	struct fsnotify_event *fsn_event = NULL;
75 
76 	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
77 
78 	spin_lock(&group->notification_lock);
79 	if (fsnotify_notify_queue_is_empty(group))
80 		goto out;
81 
82 	if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
83 		event_size += fanotify_event_info_len(
84 			FANOTIFY_E(fsnotify_peek_first_event(group)));
85 	}
86 
87 	if (event_size > count) {
88 		fsn_event = ERR_PTR(-EINVAL);
89 		goto out;
90 	}
91 	fsn_event = fsnotify_remove_first_event(group);
92 	if (fanotify_is_perm_event(FANOTIFY_E(fsn_event)->mask))
93 		FANOTIFY_PE(fsn_event)->state = FAN_EVENT_REPORTED;
94 out:
95 	spin_unlock(&group->notification_lock);
96 	return fsn_event;
97 }
98 
99 static int create_fd(struct fsnotify_group *group,
100 		     struct fanotify_event *event,
101 		     struct file **file)
102 {
103 	int client_fd;
104 	struct file *new_file;
105 
106 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
107 
108 	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
109 	if (client_fd < 0)
110 		return client_fd;
111 
112 	/*
113 	 * we need a new file handle for the userspace program so it can read even if it was
114 	 * originally opened O_WRONLY.
115 	 */
116 	/* it's possible this event was an overflow event.  in that case dentry and mnt
117 	 * are NULL;  That's fine, just don't call dentry open */
118 	if (event->path.dentry && event->path.mnt)
119 		new_file = dentry_open(&event->path,
120 				       group->fanotify_data.f_flags | FMODE_NONOTIFY,
121 				       current_cred());
122 	else
123 		new_file = ERR_PTR(-EOVERFLOW);
124 	if (IS_ERR(new_file)) {
125 		/*
126 		 * we still send an event even if we can't open the file.  this
127 		 * can happen when say tasks are gone and we try to open their
128 		 * /proc files or we try to open a WRONLY file like in sysfs
129 		 * we just send the errno to userspace since there isn't much
130 		 * else we can do.
131 		 */
132 		put_unused_fd(client_fd);
133 		client_fd = PTR_ERR(new_file);
134 	} else {
135 		*file = new_file;
136 	}
137 
138 	return client_fd;
139 }
140 
141 /*
142  * Finish processing of permission event by setting it to ANSWERED state and
143  * drop group->notification_lock.
144  */
145 static void finish_permission_event(struct fsnotify_group *group,
146 				    struct fanotify_perm_event *event,
147 				    unsigned int response)
148 				    __releases(&group->notification_lock)
149 {
150 	bool destroy = false;
151 
152 	assert_spin_locked(&group->notification_lock);
153 	event->response = response;
154 	if (event->state == FAN_EVENT_CANCELED)
155 		destroy = true;
156 	else
157 		event->state = FAN_EVENT_ANSWERED;
158 	spin_unlock(&group->notification_lock);
159 	if (destroy)
160 		fsnotify_destroy_event(group, &event->fae.fse);
161 }
162 
163 static int process_access_response(struct fsnotify_group *group,
164 				   struct fanotify_response *response_struct)
165 {
166 	struct fanotify_perm_event *event;
167 	int fd = response_struct->fd;
168 	int response = response_struct->response;
169 
170 	pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
171 		 fd, response);
172 	/*
173 	 * make sure the response is valid, if invalid we do nothing and either
174 	 * userspace can send a valid response or we will clean it up after the
175 	 * timeout
176 	 */
177 	switch (response & ~FAN_AUDIT) {
178 	case FAN_ALLOW:
179 	case FAN_DENY:
180 		break;
181 	default:
182 		return -EINVAL;
183 	}
184 
185 	if (fd < 0)
186 		return -EINVAL;
187 
188 	if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
189 		return -EINVAL;
190 
191 	spin_lock(&group->notification_lock);
192 	list_for_each_entry(event, &group->fanotify_data.access_list,
193 			    fae.fse.list) {
194 		if (event->fd != fd)
195 			continue;
196 
197 		list_del_init(&event->fae.fse.list);
198 		finish_permission_event(group, event, response);
199 		wake_up(&group->fanotify_data.access_waitq);
200 		return 0;
201 	}
202 	spin_unlock(&group->notification_lock);
203 
204 	return -ENOENT;
205 }
206 
207 static int copy_fid_to_user(struct fanotify_event *event, char __user *buf)
208 {
209 	struct fanotify_event_info_fid info = { };
210 	struct file_handle handle = { };
211 	unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh;
212 	size_t fh_len = event->fh_len;
213 	size_t len = fanotify_event_info_len(event);
214 
215 	if (!len)
216 		return 0;
217 
218 	if (WARN_ON_ONCE(len < sizeof(info) + sizeof(handle) + fh_len))
219 		return -EFAULT;
220 
221 	/* Copy event info fid header followed by vaiable sized file handle */
222 	info.hdr.info_type = FAN_EVENT_INFO_TYPE_FID;
223 	info.hdr.len = len;
224 	info.fsid = event->fid.fsid;
225 	if (copy_to_user(buf, &info, sizeof(info)))
226 		return -EFAULT;
227 
228 	buf += sizeof(info);
229 	len -= sizeof(info);
230 	handle.handle_type = event->fh_type;
231 	handle.handle_bytes = fh_len;
232 	if (copy_to_user(buf, &handle, sizeof(handle)))
233 		return -EFAULT;
234 
235 	buf += sizeof(handle);
236 	len -= sizeof(handle);
237 	/*
238 	 * For an inline fh, copy through stack to exclude the copy from
239 	 * usercopy hardening protections.
240 	 */
241 	fh = fanotify_event_fh(event);
242 	if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
243 		memcpy(bounce, fh, fh_len);
244 		fh = bounce;
245 	}
246 	if (copy_to_user(buf, fh, fh_len))
247 		return -EFAULT;
248 
249 	/* Pad with 0's */
250 	buf += fh_len;
251 	len -= fh_len;
252 	WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
253 	if (len > 0 && clear_user(buf, len))
254 		return -EFAULT;
255 
256 	return 0;
257 }
258 
259 static ssize_t copy_event_to_user(struct fsnotify_group *group,
260 				  struct fsnotify_event *fsn_event,
261 				  char __user *buf, size_t count)
262 {
263 	struct fanotify_event_metadata metadata;
264 	struct fanotify_event *event;
265 	struct file *f = NULL;
266 	int ret, fd = FAN_NOFD;
267 
268 	pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
269 
270 	event = container_of(fsn_event, struct fanotify_event, fse);
271 	metadata.event_len = FAN_EVENT_METADATA_LEN;
272 	metadata.metadata_len = FAN_EVENT_METADATA_LEN;
273 	metadata.vers = FANOTIFY_METADATA_VERSION;
274 	metadata.reserved = 0;
275 	metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
276 	metadata.pid = pid_vnr(event->pid);
277 
278 	if (fanotify_event_has_path(event)) {
279 		fd = create_fd(group, event, &f);
280 		if (fd < 0)
281 			return fd;
282 	} else if (fanotify_event_has_fid(event)) {
283 		metadata.event_len += fanotify_event_info_len(event);
284 	}
285 	metadata.fd = fd;
286 
287 	ret = -EFAULT;
288 	/*
289 	 * Sanity check copy size in case get_one_event() and
290 	 * fill_event_metadata() event_len sizes ever get out of sync.
291 	 */
292 	if (WARN_ON_ONCE(metadata.event_len > count))
293 		goto out_close_fd;
294 
295 	if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
296 		goto out_close_fd;
297 
298 	if (fanotify_is_perm_event(event->mask))
299 		FANOTIFY_PE(fsn_event)->fd = fd;
300 
301 	if (fanotify_event_has_path(event)) {
302 		fd_install(fd, f);
303 	} else if (fanotify_event_has_fid(event)) {
304 		ret = copy_fid_to_user(event, buf + FAN_EVENT_METADATA_LEN);
305 		if (ret < 0)
306 			return ret;
307 	}
308 
309 	return metadata.event_len;
310 
311 out_close_fd:
312 	if (fd != FAN_NOFD) {
313 		put_unused_fd(fd);
314 		fput(f);
315 	}
316 	return ret;
317 }
318 
319 /* intofiy userspace file descriptor functions */
320 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
321 {
322 	struct fsnotify_group *group = file->private_data;
323 	__poll_t ret = 0;
324 
325 	poll_wait(file, &group->notification_waitq, wait);
326 	spin_lock(&group->notification_lock);
327 	if (!fsnotify_notify_queue_is_empty(group))
328 		ret = EPOLLIN | EPOLLRDNORM;
329 	spin_unlock(&group->notification_lock);
330 
331 	return ret;
332 }
333 
334 static ssize_t fanotify_read(struct file *file, char __user *buf,
335 			     size_t count, loff_t *pos)
336 {
337 	struct fsnotify_group *group;
338 	struct fsnotify_event *kevent;
339 	char __user *start;
340 	int ret;
341 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
342 
343 	start = buf;
344 	group = file->private_data;
345 
346 	pr_debug("%s: group=%p\n", __func__, group);
347 
348 	add_wait_queue(&group->notification_waitq, &wait);
349 	while (1) {
350 		kevent = get_one_event(group, count);
351 		if (IS_ERR(kevent)) {
352 			ret = PTR_ERR(kevent);
353 			break;
354 		}
355 
356 		if (!kevent) {
357 			ret = -EAGAIN;
358 			if (file->f_flags & O_NONBLOCK)
359 				break;
360 
361 			ret = -ERESTARTSYS;
362 			if (signal_pending(current))
363 				break;
364 
365 			if (start != buf)
366 				break;
367 
368 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
369 			continue;
370 		}
371 
372 		ret = copy_event_to_user(group, kevent, buf, count);
373 		if (unlikely(ret == -EOPENSTALE)) {
374 			/*
375 			 * We cannot report events with stale fd so drop it.
376 			 * Setting ret to 0 will continue the event loop and
377 			 * do the right thing if there are no more events to
378 			 * read (i.e. return bytes read, -EAGAIN or wait).
379 			 */
380 			ret = 0;
381 		}
382 
383 		/*
384 		 * Permission events get queued to wait for response.  Other
385 		 * events can be destroyed now.
386 		 */
387 		if (!fanotify_is_perm_event(FANOTIFY_E(kevent)->mask)) {
388 			fsnotify_destroy_event(group, kevent);
389 		} else {
390 			if (ret <= 0) {
391 				spin_lock(&group->notification_lock);
392 				finish_permission_event(group,
393 					FANOTIFY_PE(kevent), FAN_DENY);
394 				wake_up(&group->fanotify_data.access_waitq);
395 			} else {
396 				spin_lock(&group->notification_lock);
397 				list_add_tail(&kevent->list,
398 					&group->fanotify_data.access_list);
399 				spin_unlock(&group->notification_lock);
400 			}
401 		}
402 		if (ret < 0)
403 			break;
404 		buf += ret;
405 		count -= ret;
406 	}
407 	remove_wait_queue(&group->notification_waitq, &wait);
408 
409 	if (start != buf && ret != -EFAULT)
410 		ret = buf - start;
411 	return ret;
412 }
413 
414 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
415 {
416 	struct fanotify_response response = { .fd = -1, .response = -1 };
417 	struct fsnotify_group *group;
418 	int ret;
419 
420 	if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
421 		return -EINVAL;
422 
423 	group = file->private_data;
424 
425 	if (count > sizeof(response))
426 		count = sizeof(response);
427 
428 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
429 
430 	if (copy_from_user(&response, buf, count))
431 		return -EFAULT;
432 
433 	ret = process_access_response(group, &response);
434 	if (ret < 0)
435 		count = ret;
436 
437 	return count;
438 }
439 
440 static int fanotify_release(struct inode *ignored, struct file *file)
441 {
442 	struct fsnotify_group *group = file->private_data;
443 	struct fanotify_perm_event *event;
444 	struct fsnotify_event *fsn_event;
445 
446 	/*
447 	 * Stop new events from arriving in the notification queue. since
448 	 * userspace cannot use fanotify fd anymore, no event can enter or
449 	 * leave access_list by now either.
450 	 */
451 	fsnotify_group_stop_queueing(group);
452 
453 	/*
454 	 * Process all permission events on access_list and notification queue
455 	 * and simulate reply from userspace.
456 	 */
457 	spin_lock(&group->notification_lock);
458 	while (!list_empty(&group->fanotify_data.access_list)) {
459 		event = list_first_entry(&group->fanotify_data.access_list,
460 				struct fanotify_perm_event, fae.fse.list);
461 		list_del_init(&event->fae.fse.list);
462 		finish_permission_event(group, event, FAN_ALLOW);
463 		spin_lock(&group->notification_lock);
464 	}
465 
466 	/*
467 	 * Destroy all non-permission events. For permission events just
468 	 * dequeue them and set the response. They will be freed once the
469 	 * response is consumed and fanotify_get_response() returns.
470 	 */
471 	while (!fsnotify_notify_queue_is_empty(group)) {
472 		fsn_event = fsnotify_remove_first_event(group);
473 		if (!(FANOTIFY_E(fsn_event)->mask & FANOTIFY_PERM_EVENTS)) {
474 			spin_unlock(&group->notification_lock);
475 			fsnotify_destroy_event(group, fsn_event);
476 		} else {
477 			finish_permission_event(group, FANOTIFY_PE(fsn_event),
478 						FAN_ALLOW);
479 		}
480 		spin_lock(&group->notification_lock);
481 	}
482 	spin_unlock(&group->notification_lock);
483 
484 	/* Response for all permission events it set, wakeup waiters */
485 	wake_up(&group->fanotify_data.access_waitq);
486 
487 	/* matches the fanotify_init->fsnotify_alloc_group */
488 	fsnotify_destroy_group(group);
489 
490 	return 0;
491 }
492 
493 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
494 {
495 	struct fsnotify_group *group;
496 	struct fsnotify_event *fsn_event;
497 	void __user *p;
498 	int ret = -ENOTTY;
499 	size_t send_len = 0;
500 
501 	group = file->private_data;
502 
503 	p = (void __user *) arg;
504 
505 	switch (cmd) {
506 	case FIONREAD:
507 		spin_lock(&group->notification_lock);
508 		list_for_each_entry(fsn_event, &group->notification_list, list)
509 			send_len += FAN_EVENT_METADATA_LEN;
510 		spin_unlock(&group->notification_lock);
511 		ret = put_user(send_len, (int __user *) p);
512 		break;
513 	}
514 
515 	return ret;
516 }
517 
518 static const struct file_operations fanotify_fops = {
519 	.show_fdinfo	= fanotify_show_fdinfo,
520 	.poll		= fanotify_poll,
521 	.read		= fanotify_read,
522 	.write		= fanotify_write,
523 	.fasync		= NULL,
524 	.release	= fanotify_release,
525 	.unlocked_ioctl	= fanotify_ioctl,
526 	.compat_ioctl	= compat_ptr_ioctl,
527 	.llseek		= noop_llseek,
528 };
529 
530 static int fanotify_find_path(int dfd, const char __user *filename,
531 			      struct path *path, unsigned int flags, __u64 mask,
532 			      unsigned int obj_type)
533 {
534 	int ret;
535 
536 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
537 		 dfd, filename, flags);
538 
539 	if (filename == NULL) {
540 		struct fd f = fdget(dfd);
541 
542 		ret = -EBADF;
543 		if (!f.file)
544 			goto out;
545 
546 		ret = -ENOTDIR;
547 		if ((flags & FAN_MARK_ONLYDIR) &&
548 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
549 			fdput(f);
550 			goto out;
551 		}
552 
553 		*path = f.file->f_path;
554 		path_get(path);
555 		fdput(f);
556 	} else {
557 		unsigned int lookup_flags = 0;
558 
559 		if (!(flags & FAN_MARK_DONT_FOLLOW))
560 			lookup_flags |= LOOKUP_FOLLOW;
561 		if (flags & FAN_MARK_ONLYDIR)
562 			lookup_flags |= LOOKUP_DIRECTORY;
563 
564 		ret = user_path_at(dfd, filename, lookup_flags, path);
565 		if (ret)
566 			goto out;
567 	}
568 
569 	/* you can only watch an inode if you have read permissions on it */
570 	ret = inode_permission(path->dentry->d_inode, MAY_READ);
571 	if (ret) {
572 		path_put(path);
573 		goto out;
574 	}
575 
576 	ret = security_path_notify(path, mask, obj_type);
577 	if (ret)
578 		path_put(path);
579 
580 out:
581 	return ret;
582 }
583 
584 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
585 					    __u32 mask,
586 					    unsigned int flags,
587 					    int *destroy)
588 {
589 	__u32 oldmask = 0;
590 
591 	spin_lock(&fsn_mark->lock);
592 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
593 		oldmask = fsn_mark->mask;
594 		fsn_mark->mask &= ~mask;
595 	} else {
596 		fsn_mark->ignored_mask &= ~mask;
597 	}
598 	*destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
599 	spin_unlock(&fsn_mark->lock);
600 
601 	return mask & oldmask;
602 }
603 
604 static int fanotify_remove_mark(struct fsnotify_group *group,
605 				fsnotify_connp_t *connp, __u32 mask,
606 				unsigned int flags)
607 {
608 	struct fsnotify_mark *fsn_mark = NULL;
609 	__u32 removed;
610 	int destroy_mark;
611 
612 	mutex_lock(&group->mark_mutex);
613 	fsn_mark = fsnotify_find_mark(connp, group);
614 	if (!fsn_mark) {
615 		mutex_unlock(&group->mark_mutex);
616 		return -ENOENT;
617 	}
618 
619 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
620 						 &destroy_mark);
621 	if (removed & fsnotify_conn_mask(fsn_mark->connector))
622 		fsnotify_recalc_mask(fsn_mark->connector);
623 	if (destroy_mark)
624 		fsnotify_detach_mark(fsn_mark);
625 	mutex_unlock(&group->mark_mutex);
626 	if (destroy_mark)
627 		fsnotify_free_mark(fsn_mark);
628 
629 	/* matches the fsnotify_find_mark() */
630 	fsnotify_put_mark(fsn_mark);
631 	return 0;
632 }
633 
634 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
635 					 struct vfsmount *mnt, __u32 mask,
636 					 unsigned int flags)
637 {
638 	return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
639 				    mask, flags);
640 }
641 
642 static int fanotify_remove_sb_mark(struct fsnotify_group *group,
643 				      struct super_block *sb, __u32 mask,
644 				      unsigned int flags)
645 {
646 	return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask, flags);
647 }
648 
649 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
650 				      struct inode *inode, __u32 mask,
651 				      unsigned int flags)
652 {
653 	return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
654 				    flags);
655 }
656 
657 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
658 				       __u32 mask,
659 				       unsigned int flags)
660 {
661 	__u32 oldmask = -1;
662 
663 	spin_lock(&fsn_mark->lock);
664 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
665 		oldmask = fsn_mark->mask;
666 		fsn_mark->mask |= mask;
667 	} else {
668 		fsn_mark->ignored_mask |= mask;
669 		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
670 			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
671 	}
672 	spin_unlock(&fsn_mark->lock);
673 
674 	return mask & ~oldmask;
675 }
676 
677 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
678 						   fsnotify_connp_t *connp,
679 						   unsigned int type,
680 						   __kernel_fsid_t *fsid)
681 {
682 	struct fsnotify_mark *mark;
683 	int ret;
684 
685 	if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
686 		return ERR_PTR(-ENOSPC);
687 
688 	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
689 	if (!mark)
690 		return ERR_PTR(-ENOMEM);
691 
692 	fsnotify_init_mark(mark, group);
693 	ret = fsnotify_add_mark_locked(mark, connp, type, 0, fsid);
694 	if (ret) {
695 		fsnotify_put_mark(mark);
696 		return ERR_PTR(ret);
697 	}
698 
699 	return mark;
700 }
701 
702 
703 static int fanotify_add_mark(struct fsnotify_group *group,
704 			     fsnotify_connp_t *connp, unsigned int type,
705 			     __u32 mask, unsigned int flags,
706 			     __kernel_fsid_t *fsid)
707 {
708 	struct fsnotify_mark *fsn_mark;
709 	__u32 added;
710 
711 	mutex_lock(&group->mark_mutex);
712 	fsn_mark = fsnotify_find_mark(connp, group);
713 	if (!fsn_mark) {
714 		fsn_mark = fanotify_add_new_mark(group, connp, type, fsid);
715 		if (IS_ERR(fsn_mark)) {
716 			mutex_unlock(&group->mark_mutex);
717 			return PTR_ERR(fsn_mark);
718 		}
719 	}
720 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
721 	if (added & ~fsnotify_conn_mask(fsn_mark->connector))
722 		fsnotify_recalc_mask(fsn_mark->connector);
723 	mutex_unlock(&group->mark_mutex);
724 
725 	fsnotify_put_mark(fsn_mark);
726 	return 0;
727 }
728 
729 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
730 				      struct vfsmount *mnt, __u32 mask,
731 				      unsigned int flags, __kernel_fsid_t *fsid)
732 {
733 	return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
734 				 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
735 }
736 
737 static int fanotify_add_sb_mark(struct fsnotify_group *group,
738 				struct super_block *sb, __u32 mask,
739 				unsigned int flags, __kernel_fsid_t *fsid)
740 {
741 	return fanotify_add_mark(group, &sb->s_fsnotify_marks,
742 				 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
743 }
744 
745 static int fanotify_add_inode_mark(struct fsnotify_group *group,
746 				   struct inode *inode, __u32 mask,
747 				   unsigned int flags, __kernel_fsid_t *fsid)
748 {
749 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
750 
751 	/*
752 	 * If some other task has this inode open for write we should not add
753 	 * an ignored mark, unless that ignored mark is supposed to survive
754 	 * modification changes anyway.
755 	 */
756 	if ((flags & FAN_MARK_IGNORED_MASK) &&
757 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
758 	    inode_is_open_for_write(inode))
759 		return 0;
760 
761 	return fanotify_add_mark(group, &inode->i_fsnotify_marks,
762 				 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
763 }
764 
765 /* fanotify syscalls */
766 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
767 {
768 	struct fsnotify_group *group;
769 	int f_flags, fd;
770 	struct user_struct *user;
771 	struct fanotify_event *oevent;
772 
773 	pr_debug("%s: flags=%x event_f_flags=%x\n",
774 		 __func__, flags, event_f_flags);
775 
776 	if (!capable(CAP_SYS_ADMIN))
777 		return -EPERM;
778 
779 #ifdef CONFIG_AUDITSYSCALL
780 	if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
781 #else
782 	if (flags & ~FANOTIFY_INIT_FLAGS)
783 #endif
784 		return -EINVAL;
785 
786 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
787 		return -EINVAL;
788 
789 	switch (event_f_flags & O_ACCMODE) {
790 	case O_RDONLY:
791 	case O_RDWR:
792 	case O_WRONLY:
793 		break;
794 	default:
795 		return -EINVAL;
796 	}
797 
798 	if ((flags & FAN_REPORT_FID) &&
799 	    (flags & FANOTIFY_CLASS_BITS) != FAN_CLASS_NOTIF)
800 		return -EINVAL;
801 
802 	user = get_current_user();
803 	if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
804 		free_uid(user);
805 		return -EMFILE;
806 	}
807 
808 	f_flags = O_RDWR | FMODE_NONOTIFY;
809 	if (flags & FAN_CLOEXEC)
810 		f_flags |= O_CLOEXEC;
811 	if (flags & FAN_NONBLOCK)
812 		f_flags |= O_NONBLOCK;
813 
814 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
815 	group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
816 	if (IS_ERR(group)) {
817 		free_uid(user);
818 		return PTR_ERR(group);
819 	}
820 
821 	group->fanotify_data.user = user;
822 	group->fanotify_data.flags = flags;
823 	atomic_inc(&user->fanotify_listeners);
824 	group->memcg = get_mem_cgroup_from_mm(current->mm);
825 
826 	oevent = fanotify_alloc_event(group, NULL, FS_Q_OVERFLOW, NULL,
827 				      FSNOTIFY_EVENT_NONE, NULL);
828 	if (unlikely(!oevent)) {
829 		fd = -ENOMEM;
830 		goto out_destroy_group;
831 	}
832 	group->overflow_event = &oevent->fse;
833 
834 	if (force_o_largefile())
835 		event_f_flags |= O_LARGEFILE;
836 	group->fanotify_data.f_flags = event_f_flags;
837 	init_waitqueue_head(&group->fanotify_data.access_waitq);
838 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
839 	switch (flags & FANOTIFY_CLASS_BITS) {
840 	case FAN_CLASS_NOTIF:
841 		group->priority = FS_PRIO_0;
842 		break;
843 	case FAN_CLASS_CONTENT:
844 		group->priority = FS_PRIO_1;
845 		break;
846 	case FAN_CLASS_PRE_CONTENT:
847 		group->priority = FS_PRIO_2;
848 		break;
849 	default:
850 		fd = -EINVAL;
851 		goto out_destroy_group;
852 	}
853 
854 	if (flags & FAN_UNLIMITED_QUEUE) {
855 		fd = -EPERM;
856 		if (!capable(CAP_SYS_ADMIN))
857 			goto out_destroy_group;
858 		group->max_events = UINT_MAX;
859 	} else {
860 		group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
861 	}
862 
863 	if (flags & FAN_UNLIMITED_MARKS) {
864 		fd = -EPERM;
865 		if (!capable(CAP_SYS_ADMIN))
866 			goto out_destroy_group;
867 		group->fanotify_data.max_marks = UINT_MAX;
868 	} else {
869 		group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
870 	}
871 
872 	if (flags & FAN_ENABLE_AUDIT) {
873 		fd = -EPERM;
874 		if (!capable(CAP_AUDIT_WRITE))
875 			goto out_destroy_group;
876 	}
877 
878 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
879 	if (fd < 0)
880 		goto out_destroy_group;
881 
882 	return fd;
883 
884 out_destroy_group:
885 	fsnotify_destroy_group(group);
886 	return fd;
887 }
888 
889 /* Check if filesystem can encode a unique fid */
890 static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
891 {
892 	__kernel_fsid_t root_fsid;
893 	int err;
894 
895 	/*
896 	 * Make sure path is not in filesystem with zero fsid (e.g. tmpfs).
897 	 */
898 	err = vfs_get_fsid(path->dentry, fsid);
899 	if (err)
900 		return err;
901 
902 	if (!fsid->val[0] && !fsid->val[1])
903 		return -ENODEV;
904 
905 	/*
906 	 * Make sure path is not inside a filesystem subvolume (e.g. btrfs)
907 	 * which uses a different fsid than sb root.
908 	 */
909 	err = vfs_get_fsid(path->dentry->d_sb->s_root, &root_fsid);
910 	if (err)
911 		return err;
912 
913 	if (root_fsid.val[0] != fsid->val[0] ||
914 	    root_fsid.val[1] != fsid->val[1])
915 		return -EXDEV;
916 
917 	/*
918 	 * We need to make sure that the file system supports at least
919 	 * encoding a file handle so user can use name_to_handle_at() to
920 	 * compare fid returned with event to the file handle of watched
921 	 * objects. However, name_to_handle_at() requires that the
922 	 * filesystem also supports decoding file handles.
923 	 */
924 	if (!path->dentry->d_sb->s_export_op ||
925 	    !path->dentry->d_sb->s_export_op->fh_to_dentry)
926 		return -EOPNOTSUPP;
927 
928 	return 0;
929 }
930 
931 static int fanotify_events_supported(struct path *path, __u64 mask)
932 {
933 	/*
934 	 * Some filesystems such as 'proc' acquire unusual locks when opening
935 	 * files. For them fanotify permission events have high chances of
936 	 * deadlocking the system - open done when reporting fanotify event
937 	 * blocks on this "unusual" lock while another process holding the lock
938 	 * waits for fanotify permission event to be answered. Just disallow
939 	 * permission events for such filesystems.
940 	 */
941 	if (mask & FANOTIFY_PERM_EVENTS &&
942 	    path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
943 		return -EINVAL;
944 	return 0;
945 }
946 
947 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
948 			    int dfd, const char  __user *pathname)
949 {
950 	struct inode *inode = NULL;
951 	struct vfsmount *mnt = NULL;
952 	struct fsnotify_group *group;
953 	struct fd f;
954 	struct path path;
955 	__kernel_fsid_t __fsid, *fsid = NULL;
956 	u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
957 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
958 	unsigned int obj_type;
959 	int ret;
960 
961 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
962 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
963 
964 	/* we only use the lower 32 bits as of right now. */
965 	if (mask & ((__u64)0xffffffff << 32))
966 		return -EINVAL;
967 
968 	if (flags & ~FANOTIFY_MARK_FLAGS)
969 		return -EINVAL;
970 
971 	switch (mark_type) {
972 	case FAN_MARK_INODE:
973 		obj_type = FSNOTIFY_OBJ_TYPE_INODE;
974 		break;
975 	case FAN_MARK_MOUNT:
976 		obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
977 		break;
978 	case FAN_MARK_FILESYSTEM:
979 		obj_type = FSNOTIFY_OBJ_TYPE_SB;
980 		break;
981 	default:
982 		return -EINVAL;
983 	}
984 
985 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
986 	case FAN_MARK_ADD:		/* fallthrough */
987 	case FAN_MARK_REMOVE:
988 		if (!mask)
989 			return -EINVAL;
990 		break;
991 	case FAN_MARK_FLUSH:
992 		if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
993 			return -EINVAL;
994 		break;
995 	default:
996 		return -EINVAL;
997 	}
998 
999 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1000 		valid_mask |= FANOTIFY_PERM_EVENTS;
1001 
1002 	if (mask & ~valid_mask)
1003 		return -EINVAL;
1004 
1005 	f = fdget(fanotify_fd);
1006 	if (unlikely(!f.file))
1007 		return -EBADF;
1008 
1009 	/* verify that this is indeed an fanotify instance */
1010 	ret = -EINVAL;
1011 	if (unlikely(f.file->f_op != &fanotify_fops))
1012 		goto fput_and_out;
1013 	group = f.file->private_data;
1014 
1015 	/*
1016 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
1017 	 * allowed to set permissions events.
1018 	 */
1019 	ret = -EINVAL;
1020 	if (mask & FANOTIFY_PERM_EVENTS &&
1021 	    group->priority == FS_PRIO_0)
1022 		goto fput_and_out;
1023 
1024 	/*
1025 	 * Events with data type inode do not carry enough information to report
1026 	 * event->fd, so we do not allow setting a mask for inode events unless
1027 	 * group supports reporting fid.
1028 	 * inode events are not supported on a mount mark, because they do not
1029 	 * carry enough information (i.e. path) to be filtered by mount point.
1030 	 */
1031 	if (mask & FANOTIFY_INODE_EVENTS &&
1032 	    (!FAN_GROUP_FLAG(group, FAN_REPORT_FID) ||
1033 	     mark_type == FAN_MARK_MOUNT))
1034 		goto fput_and_out;
1035 
1036 	if (flags & FAN_MARK_FLUSH) {
1037 		ret = 0;
1038 		if (mark_type == FAN_MARK_MOUNT)
1039 			fsnotify_clear_vfsmount_marks_by_group(group);
1040 		else if (mark_type == FAN_MARK_FILESYSTEM)
1041 			fsnotify_clear_sb_marks_by_group(group);
1042 		else
1043 			fsnotify_clear_inode_marks_by_group(group);
1044 		goto fput_and_out;
1045 	}
1046 
1047 	ret = fanotify_find_path(dfd, pathname, &path, flags,
1048 			(mask & ALL_FSNOTIFY_EVENTS), obj_type);
1049 	if (ret)
1050 		goto fput_and_out;
1051 
1052 	if (flags & FAN_MARK_ADD) {
1053 		ret = fanotify_events_supported(&path, mask);
1054 		if (ret)
1055 			goto path_put_and_out;
1056 	}
1057 
1058 	if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
1059 		ret = fanotify_test_fid(&path, &__fsid);
1060 		if (ret)
1061 			goto path_put_and_out;
1062 
1063 		fsid = &__fsid;
1064 	}
1065 
1066 	/* inode held in place by reference to path; group by fget on fd */
1067 	if (mark_type == FAN_MARK_INODE)
1068 		inode = path.dentry->d_inode;
1069 	else
1070 		mnt = path.mnt;
1071 
1072 	/* create/update an inode mark */
1073 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
1074 	case FAN_MARK_ADD:
1075 		if (mark_type == FAN_MARK_MOUNT)
1076 			ret = fanotify_add_vfsmount_mark(group, mnt, mask,
1077 							 flags, fsid);
1078 		else if (mark_type == FAN_MARK_FILESYSTEM)
1079 			ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
1080 						   flags, fsid);
1081 		else
1082 			ret = fanotify_add_inode_mark(group, inode, mask,
1083 						      flags, fsid);
1084 		break;
1085 	case FAN_MARK_REMOVE:
1086 		if (mark_type == FAN_MARK_MOUNT)
1087 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
1088 							    flags);
1089 		else if (mark_type == FAN_MARK_FILESYSTEM)
1090 			ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
1091 						      flags);
1092 		else
1093 			ret = fanotify_remove_inode_mark(group, inode, mask,
1094 							 flags);
1095 		break;
1096 	default:
1097 		ret = -EINVAL;
1098 	}
1099 
1100 path_put_and_out:
1101 	path_put(&path);
1102 fput_and_out:
1103 	fdput(f);
1104 	return ret;
1105 }
1106 
1107 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1108 			      __u64, mask, int, dfd,
1109 			      const char  __user *, pathname)
1110 {
1111 	return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1112 }
1113 
1114 #ifdef CONFIG_COMPAT
1115 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
1116 				int, fanotify_fd, unsigned int, flags,
1117 				__u32, mask0, __u32, mask1, int, dfd,
1118 				const char  __user *, pathname)
1119 {
1120 	return do_fanotify_mark(fanotify_fd, flags,
1121 #ifdef __BIG_ENDIAN
1122 				((__u64)mask0 << 32) | mask1,
1123 #else
1124 				((__u64)mask1 << 32) | mask0,
1125 #endif
1126 				 dfd, pathname);
1127 }
1128 #endif
1129 
1130 /*
1131  * fanotify_user_setup - Our initialization function.  Note that we cannot return
1132  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
1133  * must result in panic().
1134  */
1135 static int __init fanotify_user_setup(void)
1136 {
1137 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 8);
1138 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9);
1139 
1140 	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark,
1141 					 SLAB_PANIC|SLAB_ACCOUNT);
1142 	fanotify_event_cachep = KMEM_CACHE(fanotify_event, SLAB_PANIC);
1143 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
1144 		fanotify_perm_event_cachep =
1145 			KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
1146 	}
1147 
1148 	return 0;
1149 }
1150 device_initcall(fanotify_user_setup);
1151