xref: /linux/fs/notify/fanotify/fanotify_user.c (revision bb5b94f5bbe75470912b70fb08880fc5273aa62d)
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_OLD_DEFAULT_MAX_MARKS	8192
31 #define FANOTIFY_DEFAULT_MAX_GROUPS	128
32 
33 /*
34  * Legacy fanotify marks limits (8192) is per group and we introduced a tunable
35  * limit of marks per user, similar to inotify.  Effectively, the legacy limit
36  * of fanotify marks per user is <max marks per group> * <max groups per user>.
37  * This default limit (1M) also happens to match the increased limit of inotify
38  * max_user_watches since v5.10.
39  */
40 #define FANOTIFY_DEFAULT_MAX_USER_MARKS	\
41 	(FANOTIFY_OLD_DEFAULT_MAX_MARKS * FANOTIFY_DEFAULT_MAX_GROUPS)
42 
43 /*
44  * Most of the memory cost of adding an inode mark is pinning the marked inode.
45  * The size of the filesystem inode struct is not uniform across filesystems,
46  * so double the size of a VFS inode is used as a conservative approximation.
47  */
48 #define INODE_MARK_COST	(2 * sizeof(struct inode))
49 
50 /* configurable via /proc/sys/fs/fanotify/ */
51 static int fanotify_max_queued_events __read_mostly;
52 
53 #ifdef CONFIG_SYSCTL
54 
55 #include <linux/sysctl.h>
56 
57 struct ctl_table fanotify_table[] = {
58 	{
59 		.procname	= "max_user_groups",
60 		.data	= &init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS],
61 		.maxlen		= sizeof(int),
62 		.mode		= 0644,
63 		.proc_handler	= proc_dointvec_minmax,
64 		.extra1		= SYSCTL_ZERO,
65 	},
66 	{
67 		.procname	= "max_user_marks",
68 		.data	= &init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS],
69 		.maxlen		= sizeof(int),
70 		.mode		= 0644,
71 		.proc_handler	= proc_dointvec_minmax,
72 		.extra1		= SYSCTL_ZERO,
73 	},
74 	{
75 		.procname	= "max_queued_events",
76 		.data		= &fanotify_max_queued_events,
77 		.maxlen		= sizeof(int),
78 		.mode		= 0644,
79 		.proc_handler	= proc_dointvec_minmax,
80 		.extra1		= SYSCTL_ZERO
81 	},
82 	{ }
83 };
84 #endif /* CONFIG_SYSCTL */
85 
86 /*
87  * All flags that may be specified in parameter event_f_flags of fanotify_init.
88  *
89  * Internal and external open flags are stored together in field f_flags of
90  * struct file. Only external open flags shall be allowed in event_f_flags.
91  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
92  * excluded.
93  */
94 #define	FANOTIFY_INIT_ALL_EVENT_F_BITS				( \
95 		O_ACCMODE	| O_APPEND	| O_NONBLOCK	| \
96 		__O_SYNC	| O_DSYNC	| O_CLOEXEC     | \
97 		O_LARGEFILE	| O_NOATIME	)
98 
99 extern const struct fsnotify_ops fanotify_fsnotify_ops;
100 
101 struct kmem_cache *fanotify_mark_cache __read_mostly;
102 struct kmem_cache *fanotify_fid_event_cachep __read_mostly;
103 struct kmem_cache *fanotify_path_event_cachep __read_mostly;
104 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
105 
106 #define FANOTIFY_EVENT_ALIGN 4
107 #define FANOTIFY_INFO_HDR_LEN \
108 	(sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle))
109 
110 static int fanotify_fid_info_len(int fh_len, int name_len)
111 {
112 	int info_len = fh_len;
113 
114 	if (name_len)
115 		info_len += name_len + 1;
116 
117 	return roundup(FANOTIFY_INFO_HDR_LEN + info_len, FANOTIFY_EVENT_ALIGN);
118 }
119 
120 static int fanotify_event_info_len(unsigned int fid_mode,
121 				   struct fanotify_event *event)
122 {
123 	struct fanotify_info *info = fanotify_event_info(event);
124 	int dir_fh_len = fanotify_event_dir_fh_len(event);
125 	int fh_len = fanotify_event_object_fh_len(event);
126 	int info_len = 0;
127 	int dot_len = 0;
128 
129 	if (dir_fh_len) {
130 		info_len += fanotify_fid_info_len(dir_fh_len, info->name_len);
131 	} else if ((fid_mode & FAN_REPORT_NAME) && (event->mask & FAN_ONDIR)) {
132 		/*
133 		 * With group flag FAN_REPORT_NAME, if name was not recorded in
134 		 * event on a directory, we will report the name ".".
135 		 */
136 		dot_len = 1;
137 	}
138 
139 	if (fh_len)
140 		info_len += fanotify_fid_info_len(fh_len, dot_len);
141 
142 	return info_len;
143 }
144 
145 /*
146  * Remove an hashed event from merge hash table.
147  */
148 static void fanotify_unhash_event(struct fsnotify_group *group,
149 				  struct fanotify_event *event)
150 {
151 	assert_spin_locked(&group->notification_lock);
152 
153 	pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
154 		 group, event, fanotify_event_hash_bucket(group, event));
155 
156 	if (WARN_ON_ONCE(hlist_unhashed(&event->merge_list)))
157 		return;
158 
159 	hlist_del_init(&event->merge_list);
160 }
161 
162 /*
163  * Get an fanotify notification event if one exists and is small
164  * enough to fit in "count". Return an error pointer if the count
165  * is not large enough. When permission event is dequeued, its state is
166  * updated accordingly.
167  */
168 static struct fanotify_event *get_one_event(struct fsnotify_group *group,
169 					    size_t count)
170 {
171 	size_t event_size = FAN_EVENT_METADATA_LEN;
172 	struct fanotify_event *event = NULL;
173 	struct fsnotify_event *fsn_event;
174 	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
175 
176 	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
177 
178 	spin_lock(&group->notification_lock);
179 	fsn_event = fsnotify_peek_first_event(group);
180 	if (!fsn_event)
181 		goto out;
182 
183 	event = FANOTIFY_E(fsn_event);
184 	if (fid_mode)
185 		event_size += fanotify_event_info_len(fid_mode, event);
186 
187 	if (event_size > count) {
188 		event = ERR_PTR(-EINVAL);
189 		goto out;
190 	}
191 
192 	/*
193 	 * Held the notification_lock the whole time, so this is the
194 	 * same event we peeked above.
195 	 */
196 	fsnotify_remove_first_event(group);
197 	if (fanotify_is_perm_event(event->mask))
198 		FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED;
199 	if (fanotify_is_hashed_event(event->mask))
200 		fanotify_unhash_event(group, event);
201 out:
202 	spin_unlock(&group->notification_lock);
203 	return event;
204 }
205 
206 static int create_fd(struct fsnotify_group *group, struct path *path,
207 		     struct file **file)
208 {
209 	int client_fd;
210 	struct file *new_file;
211 
212 	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
213 	if (client_fd < 0)
214 		return client_fd;
215 
216 	/*
217 	 * we need a new file handle for the userspace program so it can read even if it was
218 	 * originally opened O_WRONLY.
219 	 */
220 	new_file = dentry_open(path,
221 			       group->fanotify_data.f_flags | FMODE_NONOTIFY,
222 			       current_cred());
223 	if (IS_ERR(new_file)) {
224 		/*
225 		 * we still send an event even if we can't open the file.  this
226 		 * can happen when say tasks are gone and we try to open their
227 		 * /proc files or we try to open a WRONLY file like in sysfs
228 		 * we just send the errno to userspace since there isn't much
229 		 * else we can do.
230 		 */
231 		put_unused_fd(client_fd);
232 		client_fd = PTR_ERR(new_file);
233 	} else {
234 		*file = new_file;
235 	}
236 
237 	return client_fd;
238 }
239 
240 /*
241  * Finish processing of permission event by setting it to ANSWERED state and
242  * drop group->notification_lock.
243  */
244 static void finish_permission_event(struct fsnotify_group *group,
245 				    struct fanotify_perm_event *event,
246 				    unsigned int response)
247 				    __releases(&group->notification_lock)
248 {
249 	bool destroy = false;
250 
251 	assert_spin_locked(&group->notification_lock);
252 	event->response = response;
253 	if (event->state == FAN_EVENT_CANCELED)
254 		destroy = true;
255 	else
256 		event->state = FAN_EVENT_ANSWERED;
257 	spin_unlock(&group->notification_lock);
258 	if (destroy)
259 		fsnotify_destroy_event(group, &event->fae.fse);
260 }
261 
262 static int process_access_response(struct fsnotify_group *group,
263 				   struct fanotify_response *response_struct)
264 {
265 	struct fanotify_perm_event *event;
266 	int fd = response_struct->fd;
267 	int response = response_struct->response;
268 
269 	pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
270 		 fd, response);
271 	/*
272 	 * make sure the response is valid, if invalid we do nothing and either
273 	 * userspace can send a valid response or we will clean it up after the
274 	 * timeout
275 	 */
276 	switch (response & ~FAN_AUDIT) {
277 	case FAN_ALLOW:
278 	case FAN_DENY:
279 		break;
280 	default:
281 		return -EINVAL;
282 	}
283 
284 	if (fd < 0)
285 		return -EINVAL;
286 
287 	if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
288 		return -EINVAL;
289 
290 	spin_lock(&group->notification_lock);
291 	list_for_each_entry(event, &group->fanotify_data.access_list,
292 			    fae.fse.list) {
293 		if (event->fd != fd)
294 			continue;
295 
296 		list_del_init(&event->fae.fse.list);
297 		finish_permission_event(group, event, response);
298 		wake_up(&group->fanotify_data.access_waitq);
299 		return 0;
300 	}
301 	spin_unlock(&group->notification_lock);
302 
303 	return -ENOENT;
304 }
305 
306 static int copy_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
307 			     int info_type, const char *name, size_t name_len,
308 			     char __user *buf, size_t count)
309 {
310 	struct fanotify_event_info_fid info = { };
311 	struct file_handle handle = { };
312 	unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
313 	size_t fh_len = fh ? fh->len : 0;
314 	size_t info_len = fanotify_fid_info_len(fh_len, name_len);
315 	size_t len = info_len;
316 
317 	pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
318 		 __func__, fh_len, name_len, info_len, count);
319 
320 	if (!fh_len)
321 		return 0;
322 
323 	if (WARN_ON_ONCE(len < sizeof(info) || len > count))
324 		return -EFAULT;
325 
326 	/*
327 	 * Copy event info fid header followed by variable sized file handle
328 	 * and optionally followed by variable sized filename.
329 	 */
330 	switch (info_type) {
331 	case FAN_EVENT_INFO_TYPE_FID:
332 	case FAN_EVENT_INFO_TYPE_DFID:
333 		if (WARN_ON_ONCE(name_len))
334 			return -EFAULT;
335 		break;
336 	case FAN_EVENT_INFO_TYPE_DFID_NAME:
337 		if (WARN_ON_ONCE(!name || !name_len))
338 			return -EFAULT;
339 		break;
340 	default:
341 		return -EFAULT;
342 	}
343 
344 	info.hdr.info_type = info_type;
345 	info.hdr.len = len;
346 	info.fsid = *fsid;
347 	if (copy_to_user(buf, &info, sizeof(info)))
348 		return -EFAULT;
349 
350 	buf += sizeof(info);
351 	len -= sizeof(info);
352 	if (WARN_ON_ONCE(len < sizeof(handle)))
353 		return -EFAULT;
354 
355 	handle.handle_type = fh->type;
356 	handle.handle_bytes = fh_len;
357 	if (copy_to_user(buf, &handle, sizeof(handle)))
358 		return -EFAULT;
359 
360 	buf += sizeof(handle);
361 	len -= sizeof(handle);
362 	if (WARN_ON_ONCE(len < fh_len))
363 		return -EFAULT;
364 
365 	/*
366 	 * For an inline fh and inline file name, copy through stack to exclude
367 	 * the copy from usercopy hardening protections.
368 	 */
369 	fh_buf = fanotify_fh_buf(fh);
370 	if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
371 		memcpy(bounce, fh_buf, fh_len);
372 		fh_buf = bounce;
373 	}
374 	if (copy_to_user(buf, fh_buf, fh_len))
375 		return -EFAULT;
376 
377 	buf += fh_len;
378 	len -= fh_len;
379 
380 	if (name_len) {
381 		/* Copy the filename with terminating null */
382 		name_len++;
383 		if (WARN_ON_ONCE(len < name_len))
384 			return -EFAULT;
385 
386 		if (copy_to_user(buf, name, name_len))
387 			return -EFAULT;
388 
389 		buf += name_len;
390 		len -= name_len;
391 	}
392 
393 	/* Pad with 0's */
394 	WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
395 	if (len > 0 && clear_user(buf, len))
396 		return -EFAULT;
397 
398 	return info_len;
399 }
400 
401 static ssize_t copy_event_to_user(struct fsnotify_group *group,
402 				  struct fanotify_event *event,
403 				  char __user *buf, size_t count)
404 {
405 	struct fanotify_event_metadata metadata;
406 	struct path *path = fanotify_event_path(event);
407 	struct fanotify_info *info = fanotify_event_info(event);
408 	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
409 	struct file *f = NULL;
410 	int ret, fd = FAN_NOFD;
411 	int info_type = 0;
412 
413 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
414 
415 	metadata.event_len = FAN_EVENT_METADATA_LEN +
416 				fanotify_event_info_len(fid_mode, event);
417 	metadata.metadata_len = FAN_EVENT_METADATA_LEN;
418 	metadata.vers = FANOTIFY_METADATA_VERSION;
419 	metadata.reserved = 0;
420 	metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
421 	metadata.pid = pid_vnr(event->pid);
422 	/*
423 	 * For an unprivileged listener, event->pid can be used to identify the
424 	 * events generated by the listener process itself, without disclosing
425 	 * the pids of other processes.
426 	 */
427 	if (FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
428 	    task_tgid(current) != event->pid)
429 		metadata.pid = 0;
430 
431 	/*
432 	 * For now, fid mode is required for an unprivileged listener and
433 	 * fid mode does not report fd in events.  Keep this check anyway
434 	 * for safety in case fid mode requirement is relaxed in the future
435 	 * to allow unprivileged listener to get events with no fd and no fid.
436 	 */
437 	if (!FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
438 	    path && path->mnt && path->dentry) {
439 		fd = create_fd(group, path, &f);
440 		if (fd < 0)
441 			return fd;
442 	}
443 	metadata.fd = fd;
444 
445 	ret = -EFAULT;
446 	/*
447 	 * Sanity check copy size in case get_one_event() and
448 	 * event_len sizes ever get out of sync.
449 	 */
450 	if (WARN_ON_ONCE(metadata.event_len > count))
451 		goto out_close_fd;
452 
453 	if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
454 		goto out_close_fd;
455 
456 	buf += FAN_EVENT_METADATA_LEN;
457 	count -= FAN_EVENT_METADATA_LEN;
458 
459 	if (fanotify_is_perm_event(event->mask))
460 		FANOTIFY_PERM(event)->fd = fd;
461 
462 	if (f)
463 		fd_install(fd, f);
464 
465 	/* Event info records order is: dir fid + name, child fid */
466 	if (fanotify_event_dir_fh_len(event)) {
467 		info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
468 					     FAN_EVENT_INFO_TYPE_DFID;
469 		ret = copy_info_to_user(fanotify_event_fsid(event),
470 					fanotify_info_dir_fh(info),
471 					info_type, fanotify_info_name(info),
472 					info->name_len, buf, count);
473 		if (ret < 0)
474 			goto out_close_fd;
475 
476 		buf += ret;
477 		count -= ret;
478 	}
479 
480 	if (fanotify_event_object_fh_len(event)) {
481 		const char *dot = NULL;
482 		int dot_len = 0;
483 
484 		if (fid_mode == FAN_REPORT_FID || info_type) {
485 			/*
486 			 * With only group flag FAN_REPORT_FID only type FID is
487 			 * reported. Second info record type is always FID.
488 			 */
489 			info_type = FAN_EVENT_INFO_TYPE_FID;
490 		} else if ((fid_mode & FAN_REPORT_NAME) &&
491 			   (event->mask & FAN_ONDIR)) {
492 			/*
493 			 * With group flag FAN_REPORT_NAME, if name was not
494 			 * recorded in an event on a directory, report the
495 			 * name "." with info type DFID_NAME.
496 			 */
497 			info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
498 			dot = ".";
499 			dot_len = 1;
500 		} else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
501 			   (event->mask & FAN_ONDIR)) {
502 			/*
503 			 * With group flag FAN_REPORT_DIR_FID, a single info
504 			 * record has type DFID for directory entry modification
505 			 * event and for event on a directory.
506 			 */
507 			info_type = FAN_EVENT_INFO_TYPE_DFID;
508 		} else {
509 			/*
510 			 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
511 			 * a single info record has type FID for event on a
512 			 * non-directory, when there is no directory to report.
513 			 * For example, on FAN_DELETE_SELF event.
514 			 */
515 			info_type = FAN_EVENT_INFO_TYPE_FID;
516 		}
517 
518 		ret = copy_info_to_user(fanotify_event_fsid(event),
519 					fanotify_event_object_fh(event),
520 					info_type, dot, dot_len, buf, count);
521 		if (ret < 0)
522 			goto out_close_fd;
523 
524 		buf += ret;
525 		count -= ret;
526 	}
527 
528 	return metadata.event_len;
529 
530 out_close_fd:
531 	if (fd != FAN_NOFD) {
532 		put_unused_fd(fd);
533 		fput(f);
534 	}
535 	return ret;
536 }
537 
538 /* intofiy userspace file descriptor functions */
539 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
540 {
541 	struct fsnotify_group *group = file->private_data;
542 	__poll_t ret = 0;
543 
544 	poll_wait(file, &group->notification_waitq, wait);
545 	spin_lock(&group->notification_lock);
546 	if (!fsnotify_notify_queue_is_empty(group))
547 		ret = EPOLLIN | EPOLLRDNORM;
548 	spin_unlock(&group->notification_lock);
549 
550 	return ret;
551 }
552 
553 static ssize_t fanotify_read(struct file *file, char __user *buf,
554 			     size_t count, loff_t *pos)
555 {
556 	struct fsnotify_group *group;
557 	struct fanotify_event *event;
558 	char __user *start;
559 	int ret;
560 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
561 
562 	start = buf;
563 	group = file->private_data;
564 
565 	pr_debug("%s: group=%p\n", __func__, group);
566 
567 	add_wait_queue(&group->notification_waitq, &wait);
568 	while (1) {
569 		/*
570 		 * User can supply arbitrarily large buffer. Avoid softlockups
571 		 * in case there are lots of available events.
572 		 */
573 		cond_resched();
574 		event = get_one_event(group, count);
575 		if (IS_ERR(event)) {
576 			ret = PTR_ERR(event);
577 			break;
578 		}
579 
580 		if (!event) {
581 			ret = -EAGAIN;
582 			if (file->f_flags & O_NONBLOCK)
583 				break;
584 
585 			ret = -ERESTARTSYS;
586 			if (signal_pending(current))
587 				break;
588 
589 			if (start != buf)
590 				break;
591 
592 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
593 			continue;
594 		}
595 
596 		ret = copy_event_to_user(group, event, buf, count);
597 		if (unlikely(ret == -EOPENSTALE)) {
598 			/*
599 			 * We cannot report events with stale fd so drop it.
600 			 * Setting ret to 0 will continue the event loop and
601 			 * do the right thing if there are no more events to
602 			 * read (i.e. return bytes read, -EAGAIN or wait).
603 			 */
604 			ret = 0;
605 		}
606 
607 		/*
608 		 * Permission events get queued to wait for response.  Other
609 		 * events can be destroyed now.
610 		 */
611 		if (!fanotify_is_perm_event(event->mask)) {
612 			fsnotify_destroy_event(group, &event->fse);
613 		} else {
614 			if (ret <= 0) {
615 				spin_lock(&group->notification_lock);
616 				finish_permission_event(group,
617 					FANOTIFY_PERM(event), FAN_DENY);
618 				wake_up(&group->fanotify_data.access_waitq);
619 			} else {
620 				spin_lock(&group->notification_lock);
621 				list_add_tail(&event->fse.list,
622 					&group->fanotify_data.access_list);
623 				spin_unlock(&group->notification_lock);
624 			}
625 		}
626 		if (ret < 0)
627 			break;
628 		buf += ret;
629 		count -= ret;
630 	}
631 	remove_wait_queue(&group->notification_waitq, &wait);
632 
633 	if (start != buf && ret != -EFAULT)
634 		ret = buf - start;
635 	return ret;
636 }
637 
638 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
639 {
640 	struct fanotify_response response = { .fd = -1, .response = -1 };
641 	struct fsnotify_group *group;
642 	int ret;
643 
644 	if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
645 		return -EINVAL;
646 
647 	group = file->private_data;
648 
649 	if (count < sizeof(response))
650 		return -EINVAL;
651 
652 	count = sizeof(response);
653 
654 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
655 
656 	if (copy_from_user(&response, buf, count))
657 		return -EFAULT;
658 
659 	ret = process_access_response(group, &response);
660 	if (ret < 0)
661 		count = ret;
662 
663 	return count;
664 }
665 
666 static int fanotify_release(struct inode *ignored, struct file *file)
667 {
668 	struct fsnotify_group *group = file->private_data;
669 	struct fsnotify_event *fsn_event;
670 
671 	/*
672 	 * Stop new events from arriving in the notification queue. since
673 	 * userspace cannot use fanotify fd anymore, no event can enter or
674 	 * leave access_list by now either.
675 	 */
676 	fsnotify_group_stop_queueing(group);
677 
678 	/*
679 	 * Process all permission events on access_list and notification queue
680 	 * and simulate reply from userspace.
681 	 */
682 	spin_lock(&group->notification_lock);
683 	while (!list_empty(&group->fanotify_data.access_list)) {
684 		struct fanotify_perm_event *event;
685 
686 		event = list_first_entry(&group->fanotify_data.access_list,
687 				struct fanotify_perm_event, fae.fse.list);
688 		list_del_init(&event->fae.fse.list);
689 		finish_permission_event(group, event, FAN_ALLOW);
690 		spin_lock(&group->notification_lock);
691 	}
692 
693 	/*
694 	 * Destroy all non-permission events. For permission events just
695 	 * dequeue them and set the response. They will be freed once the
696 	 * response is consumed and fanotify_get_response() returns.
697 	 */
698 	while ((fsn_event = fsnotify_remove_first_event(group))) {
699 		struct fanotify_event *event = FANOTIFY_E(fsn_event);
700 
701 		if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
702 			spin_unlock(&group->notification_lock);
703 			fsnotify_destroy_event(group, fsn_event);
704 		} else {
705 			finish_permission_event(group, FANOTIFY_PERM(event),
706 						FAN_ALLOW);
707 		}
708 		spin_lock(&group->notification_lock);
709 	}
710 	spin_unlock(&group->notification_lock);
711 
712 	/* Response for all permission events it set, wakeup waiters */
713 	wake_up(&group->fanotify_data.access_waitq);
714 
715 	/* matches the fanotify_init->fsnotify_alloc_group */
716 	fsnotify_destroy_group(group);
717 
718 	return 0;
719 }
720 
721 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
722 {
723 	struct fsnotify_group *group;
724 	struct fsnotify_event *fsn_event;
725 	void __user *p;
726 	int ret = -ENOTTY;
727 	size_t send_len = 0;
728 
729 	group = file->private_data;
730 
731 	p = (void __user *) arg;
732 
733 	switch (cmd) {
734 	case FIONREAD:
735 		spin_lock(&group->notification_lock);
736 		list_for_each_entry(fsn_event, &group->notification_list, list)
737 			send_len += FAN_EVENT_METADATA_LEN;
738 		spin_unlock(&group->notification_lock);
739 		ret = put_user(send_len, (int __user *) p);
740 		break;
741 	}
742 
743 	return ret;
744 }
745 
746 static const struct file_operations fanotify_fops = {
747 	.show_fdinfo	= fanotify_show_fdinfo,
748 	.poll		= fanotify_poll,
749 	.read		= fanotify_read,
750 	.write		= fanotify_write,
751 	.fasync		= NULL,
752 	.release	= fanotify_release,
753 	.unlocked_ioctl	= fanotify_ioctl,
754 	.compat_ioctl	= compat_ptr_ioctl,
755 	.llseek		= noop_llseek,
756 };
757 
758 static int fanotify_find_path(int dfd, const char __user *filename,
759 			      struct path *path, unsigned int flags, __u64 mask,
760 			      unsigned int obj_type)
761 {
762 	int ret;
763 
764 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
765 		 dfd, filename, flags);
766 
767 	if (filename == NULL) {
768 		struct fd f = fdget(dfd);
769 
770 		ret = -EBADF;
771 		if (!f.file)
772 			goto out;
773 
774 		ret = -ENOTDIR;
775 		if ((flags & FAN_MARK_ONLYDIR) &&
776 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
777 			fdput(f);
778 			goto out;
779 		}
780 
781 		*path = f.file->f_path;
782 		path_get(path);
783 		fdput(f);
784 	} else {
785 		unsigned int lookup_flags = 0;
786 
787 		if (!(flags & FAN_MARK_DONT_FOLLOW))
788 			lookup_flags |= LOOKUP_FOLLOW;
789 		if (flags & FAN_MARK_ONLYDIR)
790 			lookup_flags |= LOOKUP_DIRECTORY;
791 
792 		ret = user_path_at(dfd, filename, lookup_flags, path);
793 		if (ret)
794 			goto out;
795 	}
796 
797 	/* you can only watch an inode if you have read permissions on it */
798 	ret = path_permission(path, MAY_READ);
799 	if (ret) {
800 		path_put(path);
801 		goto out;
802 	}
803 
804 	ret = security_path_notify(path, mask, obj_type);
805 	if (ret)
806 		path_put(path);
807 
808 out:
809 	return ret;
810 }
811 
812 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
813 					    __u32 mask, unsigned int flags,
814 					    __u32 umask, int *destroy)
815 {
816 	__u32 oldmask = 0;
817 
818 	/* umask bits cannot be removed by user */
819 	mask &= ~umask;
820 	spin_lock(&fsn_mark->lock);
821 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
822 		oldmask = fsn_mark->mask;
823 		fsn_mark->mask &= ~mask;
824 	} else {
825 		fsn_mark->ignored_mask &= ~mask;
826 	}
827 	/*
828 	 * We need to keep the mark around even if remaining mask cannot
829 	 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
830 	 * changes to the mask.
831 	 * Destroy mark when only umask bits remain.
832 	 */
833 	*destroy = !((fsn_mark->mask | fsn_mark->ignored_mask) & ~umask);
834 	spin_unlock(&fsn_mark->lock);
835 
836 	return mask & oldmask;
837 }
838 
839 static int fanotify_remove_mark(struct fsnotify_group *group,
840 				fsnotify_connp_t *connp, __u32 mask,
841 				unsigned int flags, __u32 umask)
842 {
843 	struct fsnotify_mark *fsn_mark = NULL;
844 	__u32 removed;
845 	int destroy_mark;
846 
847 	mutex_lock(&group->mark_mutex);
848 	fsn_mark = fsnotify_find_mark(connp, group);
849 	if (!fsn_mark) {
850 		mutex_unlock(&group->mark_mutex);
851 		return -ENOENT;
852 	}
853 
854 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
855 						 umask, &destroy_mark);
856 	if (removed & fsnotify_conn_mask(fsn_mark->connector))
857 		fsnotify_recalc_mask(fsn_mark->connector);
858 	if (destroy_mark)
859 		fsnotify_detach_mark(fsn_mark);
860 	mutex_unlock(&group->mark_mutex);
861 	if (destroy_mark)
862 		fsnotify_free_mark(fsn_mark);
863 
864 	/* matches the fsnotify_find_mark() */
865 	fsnotify_put_mark(fsn_mark);
866 	return 0;
867 }
868 
869 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
870 					 struct vfsmount *mnt, __u32 mask,
871 					 unsigned int flags, __u32 umask)
872 {
873 	return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
874 				    mask, flags, umask);
875 }
876 
877 static int fanotify_remove_sb_mark(struct fsnotify_group *group,
878 				   struct super_block *sb, __u32 mask,
879 				   unsigned int flags, __u32 umask)
880 {
881 	return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
882 				    flags, umask);
883 }
884 
885 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
886 				      struct inode *inode, __u32 mask,
887 				      unsigned int flags, __u32 umask)
888 {
889 	return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
890 				    flags, umask);
891 }
892 
893 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
894 				       __u32 mask,
895 				       unsigned int flags)
896 {
897 	__u32 oldmask = -1;
898 
899 	spin_lock(&fsn_mark->lock);
900 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
901 		oldmask = fsn_mark->mask;
902 		fsn_mark->mask |= mask;
903 	} else {
904 		fsn_mark->ignored_mask |= mask;
905 		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
906 			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
907 	}
908 	spin_unlock(&fsn_mark->lock);
909 
910 	return mask & ~oldmask;
911 }
912 
913 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
914 						   fsnotify_connp_t *connp,
915 						   unsigned int type,
916 						   __kernel_fsid_t *fsid)
917 {
918 	struct ucounts *ucounts = group->fanotify_data.ucounts;
919 	struct fsnotify_mark *mark;
920 	int ret;
921 
922 	/*
923 	 * Enforce per user marks limits per user in all containing user ns.
924 	 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
925 	 * in the limited groups account.
926 	 */
927 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
928 	    !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
929 		return ERR_PTR(-ENOSPC);
930 
931 	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
932 	if (!mark) {
933 		ret = -ENOMEM;
934 		goto out_dec_ucounts;
935 	}
936 
937 	fsnotify_init_mark(mark, group);
938 	ret = fsnotify_add_mark_locked(mark, connp, type, 0, fsid);
939 	if (ret) {
940 		fsnotify_put_mark(mark);
941 		goto out_dec_ucounts;
942 	}
943 
944 	return mark;
945 
946 out_dec_ucounts:
947 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
948 		dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
949 	return ERR_PTR(ret);
950 }
951 
952 
953 static int fanotify_add_mark(struct fsnotify_group *group,
954 			     fsnotify_connp_t *connp, unsigned int type,
955 			     __u32 mask, unsigned int flags,
956 			     __kernel_fsid_t *fsid)
957 {
958 	struct fsnotify_mark *fsn_mark;
959 	__u32 added;
960 
961 	mutex_lock(&group->mark_mutex);
962 	fsn_mark = fsnotify_find_mark(connp, group);
963 	if (!fsn_mark) {
964 		fsn_mark = fanotify_add_new_mark(group, connp, type, fsid);
965 		if (IS_ERR(fsn_mark)) {
966 			mutex_unlock(&group->mark_mutex);
967 			return PTR_ERR(fsn_mark);
968 		}
969 	}
970 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
971 	if (added & ~fsnotify_conn_mask(fsn_mark->connector))
972 		fsnotify_recalc_mask(fsn_mark->connector);
973 	mutex_unlock(&group->mark_mutex);
974 
975 	fsnotify_put_mark(fsn_mark);
976 	return 0;
977 }
978 
979 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
980 				      struct vfsmount *mnt, __u32 mask,
981 				      unsigned int flags, __kernel_fsid_t *fsid)
982 {
983 	return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
984 				 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
985 }
986 
987 static int fanotify_add_sb_mark(struct fsnotify_group *group,
988 				struct super_block *sb, __u32 mask,
989 				unsigned int flags, __kernel_fsid_t *fsid)
990 {
991 	return fanotify_add_mark(group, &sb->s_fsnotify_marks,
992 				 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
993 }
994 
995 static int fanotify_add_inode_mark(struct fsnotify_group *group,
996 				   struct inode *inode, __u32 mask,
997 				   unsigned int flags, __kernel_fsid_t *fsid)
998 {
999 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
1000 
1001 	/*
1002 	 * If some other task has this inode open for write we should not add
1003 	 * an ignored mark, unless that ignored mark is supposed to survive
1004 	 * modification changes anyway.
1005 	 */
1006 	if ((flags & FAN_MARK_IGNORED_MASK) &&
1007 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1008 	    inode_is_open_for_write(inode))
1009 		return 0;
1010 
1011 	return fanotify_add_mark(group, &inode->i_fsnotify_marks,
1012 				 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
1013 }
1014 
1015 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1016 {
1017 	struct fanotify_event *oevent;
1018 
1019 	oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1020 	if (!oevent)
1021 		return NULL;
1022 
1023 	fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1024 	oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1025 
1026 	return &oevent->fse;
1027 }
1028 
1029 static struct hlist_head *fanotify_alloc_merge_hash(void)
1030 {
1031 	struct hlist_head *hash;
1032 
1033 	hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1034 		       GFP_KERNEL_ACCOUNT);
1035 	if (!hash)
1036 		return NULL;
1037 
1038 	__hash_init(hash, FANOTIFY_HTABLE_SIZE);
1039 
1040 	return hash;
1041 }
1042 
1043 /* fanotify syscalls */
1044 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1045 {
1046 	struct fsnotify_group *group;
1047 	int f_flags, fd;
1048 	unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1049 	unsigned int class = flags & FANOTIFY_CLASS_BITS;
1050 	unsigned int internal_flags = 0;
1051 
1052 	pr_debug("%s: flags=%x event_f_flags=%x\n",
1053 		 __func__, flags, event_f_flags);
1054 
1055 	if (!capable(CAP_SYS_ADMIN)) {
1056 		/*
1057 		 * An unprivileged user can setup an fanotify group with
1058 		 * limited functionality - an unprivileged group is limited to
1059 		 * notification events with file handles and it cannot use
1060 		 * unlimited queue/marks.
1061 		 */
1062 		if ((flags & FANOTIFY_ADMIN_INIT_FLAGS) || !fid_mode)
1063 			return -EPERM;
1064 
1065 		/*
1066 		 * Setting the internal flag FANOTIFY_UNPRIV on the group
1067 		 * prevents setting mount/filesystem marks on this group and
1068 		 * prevents reporting pid and open fd in events.
1069 		 */
1070 		internal_flags |= FANOTIFY_UNPRIV;
1071 	}
1072 
1073 #ifdef CONFIG_AUDITSYSCALL
1074 	if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1075 #else
1076 	if (flags & ~FANOTIFY_INIT_FLAGS)
1077 #endif
1078 		return -EINVAL;
1079 
1080 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1081 		return -EINVAL;
1082 
1083 	switch (event_f_flags & O_ACCMODE) {
1084 	case O_RDONLY:
1085 	case O_RDWR:
1086 	case O_WRONLY:
1087 		break;
1088 	default:
1089 		return -EINVAL;
1090 	}
1091 
1092 	if (fid_mode && class != FAN_CLASS_NOTIF)
1093 		return -EINVAL;
1094 
1095 	/*
1096 	 * Child name is reported with parent fid so requires dir fid.
1097 	 * We can report both child fid and dir fid with or without name.
1098 	 */
1099 	if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1100 		return -EINVAL;
1101 
1102 	f_flags = O_RDWR | FMODE_NONOTIFY;
1103 	if (flags & FAN_CLOEXEC)
1104 		f_flags |= O_CLOEXEC;
1105 	if (flags & FAN_NONBLOCK)
1106 		f_flags |= O_NONBLOCK;
1107 
1108 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
1109 	group = fsnotify_alloc_user_group(&fanotify_fsnotify_ops);
1110 	if (IS_ERR(group)) {
1111 		return PTR_ERR(group);
1112 	}
1113 
1114 	/* Enforce groups limits per user in all containing user ns */
1115 	group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1116 						  current_euid(),
1117 						  UCOUNT_FANOTIFY_GROUPS);
1118 	if (!group->fanotify_data.ucounts) {
1119 		fd = -EMFILE;
1120 		goto out_destroy_group;
1121 	}
1122 
1123 	group->fanotify_data.flags = flags | internal_flags;
1124 	group->memcg = get_mem_cgroup_from_mm(current->mm);
1125 
1126 	group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1127 	if (!group->fanotify_data.merge_hash) {
1128 		fd = -ENOMEM;
1129 		goto out_destroy_group;
1130 	}
1131 
1132 	group->overflow_event = fanotify_alloc_overflow_event();
1133 	if (unlikely(!group->overflow_event)) {
1134 		fd = -ENOMEM;
1135 		goto out_destroy_group;
1136 	}
1137 
1138 	if (force_o_largefile())
1139 		event_f_flags |= O_LARGEFILE;
1140 	group->fanotify_data.f_flags = event_f_flags;
1141 	init_waitqueue_head(&group->fanotify_data.access_waitq);
1142 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
1143 	switch (class) {
1144 	case FAN_CLASS_NOTIF:
1145 		group->priority = FS_PRIO_0;
1146 		break;
1147 	case FAN_CLASS_CONTENT:
1148 		group->priority = FS_PRIO_1;
1149 		break;
1150 	case FAN_CLASS_PRE_CONTENT:
1151 		group->priority = FS_PRIO_2;
1152 		break;
1153 	default:
1154 		fd = -EINVAL;
1155 		goto out_destroy_group;
1156 	}
1157 
1158 	if (flags & FAN_UNLIMITED_QUEUE) {
1159 		fd = -EPERM;
1160 		if (!capable(CAP_SYS_ADMIN))
1161 			goto out_destroy_group;
1162 		group->max_events = UINT_MAX;
1163 	} else {
1164 		group->max_events = fanotify_max_queued_events;
1165 	}
1166 
1167 	if (flags & FAN_UNLIMITED_MARKS) {
1168 		fd = -EPERM;
1169 		if (!capable(CAP_SYS_ADMIN))
1170 			goto out_destroy_group;
1171 	}
1172 
1173 	if (flags & FAN_ENABLE_AUDIT) {
1174 		fd = -EPERM;
1175 		if (!capable(CAP_AUDIT_WRITE))
1176 			goto out_destroy_group;
1177 	}
1178 
1179 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
1180 	if (fd < 0)
1181 		goto out_destroy_group;
1182 
1183 	return fd;
1184 
1185 out_destroy_group:
1186 	fsnotify_destroy_group(group);
1187 	return fd;
1188 }
1189 
1190 /* Check if filesystem can encode a unique fid */
1191 static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
1192 {
1193 	__kernel_fsid_t root_fsid;
1194 	int err;
1195 
1196 	/*
1197 	 * Make sure path is not in filesystem with zero fsid (e.g. tmpfs).
1198 	 */
1199 	err = vfs_get_fsid(path->dentry, fsid);
1200 	if (err)
1201 		return err;
1202 
1203 	if (!fsid->val[0] && !fsid->val[1])
1204 		return -ENODEV;
1205 
1206 	/*
1207 	 * Make sure path is not inside a filesystem subvolume (e.g. btrfs)
1208 	 * which uses a different fsid than sb root.
1209 	 */
1210 	err = vfs_get_fsid(path->dentry->d_sb->s_root, &root_fsid);
1211 	if (err)
1212 		return err;
1213 
1214 	if (root_fsid.val[0] != fsid->val[0] ||
1215 	    root_fsid.val[1] != fsid->val[1])
1216 		return -EXDEV;
1217 
1218 	/*
1219 	 * We need to make sure that the file system supports at least
1220 	 * encoding a file handle so user can use name_to_handle_at() to
1221 	 * compare fid returned with event to the file handle of watched
1222 	 * objects. However, name_to_handle_at() requires that the
1223 	 * filesystem also supports decoding file handles.
1224 	 */
1225 	if (!path->dentry->d_sb->s_export_op ||
1226 	    !path->dentry->d_sb->s_export_op->fh_to_dentry)
1227 		return -EOPNOTSUPP;
1228 
1229 	return 0;
1230 }
1231 
1232 static int fanotify_events_supported(struct path *path, __u64 mask)
1233 {
1234 	/*
1235 	 * Some filesystems such as 'proc' acquire unusual locks when opening
1236 	 * files. For them fanotify permission events have high chances of
1237 	 * deadlocking the system - open done when reporting fanotify event
1238 	 * blocks on this "unusual" lock while another process holding the lock
1239 	 * waits for fanotify permission event to be answered. Just disallow
1240 	 * permission events for such filesystems.
1241 	 */
1242 	if (mask & FANOTIFY_PERM_EVENTS &&
1243 	    path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1244 		return -EINVAL;
1245 	return 0;
1246 }
1247 
1248 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1249 			    int dfd, const char  __user *pathname)
1250 {
1251 	struct inode *inode = NULL;
1252 	struct vfsmount *mnt = NULL;
1253 	struct fsnotify_group *group;
1254 	struct fd f;
1255 	struct path path;
1256 	__kernel_fsid_t __fsid, *fsid = NULL;
1257 	u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1258 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1259 	bool ignored = flags & FAN_MARK_IGNORED_MASK;
1260 	unsigned int obj_type, fid_mode;
1261 	u32 umask = 0;
1262 	int ret;
1263 
1264 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1265 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
1266 
1267 	/* we only use the lower 32 bits as of right now. */
1268 	if (upper_32_bits(mask))
1269 		return -EINVAL;
1270 
1271 	if (flags & ~FANOTIFY_MARK_FLAGS)
1272 		return -EINVAL;
1273 
1274 	switch (mark_type) {
1275 	case FAN_MARK_INODE:
1276 		obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1277 		break;
1278 	case FAN_MARK_MOUNT:
1279 		obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1280 		break;
1281 	case FAN_MARK_FILESYSTEM:
1282 		obj_type = FSNOTIFY_OBJ_TYPE_SB;
1283 		break;
1284 	default:
1285 		return -EINVAL;
1286 	}
1287 
1288 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
1289 	case FAN_MARK_ADD:
1290 	case FAN_MARK_REMOVE:
1291 		if (!mask)
1292 			return -EINVAL;
1293 		break;
1294 	case FAN_MARK_FLUSH:
1295 		if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1296 			return -EINVAL;
1297 		break;
1298 	default:
1299 		return -EINVAL;
1300 	}
1301 
1302 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1303 		valid_mask |= FANOTIFY_PERM_EVENTS;
1304 
1305 	if (mask & ~valid_mask)
1306 		return -EINVAL;
1307 
1308 	/* Event flags (ONDIR, ON_CHILD) are meaningless in ignored mask */
1309 	if (ignored)
1310 		mask &= ~FANOTIFY_EVENT_FLAGS;
1311 
1312 	f = fdget(fanotify_fd);
1313 	if (unlikely(!f.file))
1314 		return -EBADF;
1315 
1316 	/* verify that this is indeed an fanotify instance */
1317 	ret = -EINVAL;
1318 	if (unlikely(f.file->f_op != &fanotify_fops))
1319 		goto fput_and_out;
1320 	group = f.file->private_data;
1321 
1322 	/*
1323 	 * An unprivileged user is not allowed to setup mount nor filesystem
1324 	 * marks.  This also includes setting up such marks by a group that
1325 	 * was initialized by an unprivileged user.
1326 	 */
1327 	ret = -EPERM;
1328 	if ((!capable(CAP_SYS_ADMIN) ||
1329 	     FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV)) &&
1330 	    mark_type != FAN_MARK_INODE)
1331 		goto fput_and_out;
1332 
1333 	/*
1334 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
1335 	 * allowed to set permissions events.
1336 	 */
1337 	ret = -EINVAL;
1338 	if (mask & FANOTIFY_PERM_EVENTS &&
1339 	    group->priority == FS_PRIO_0)
1340 		goto fput_and_out;
1341 
1342 	/*
1343 	 * Events with data type inode do not carry enough information to report
1344 	 * event->fd, so we do not allow setting a mask for inode events unless
1345 	 * group supports reporting fid.
1346 	 * inode events are not supported on a mount mark, because they do not
1347 	 * carry enough information (i.e. path) to be filtered by mount point.
1348 	 */
1349 	fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1350 	if (mask & FANOTIFY_INODE_EVENTS &&
1351 	    (!fid_mode || mark_type == FAN_MARK_MOUNT))
1352 		goto fput_and_out;
1353 
1354 	if (flags & FAN_MARK_FLUSH) {
1355 		ret = 0;
1356 		if (mark_type == FAN_MARK_MOUNT)
1357 			fsnotify_clear_vfsmount_marks_by_group(group);
1358 		else if (mark_type == FAN_MARK_FILESYSTEM)
1359 			fsnotify_clear_sb_marks_by_group(group);
1360 		else
1361 			fsnotify_clear_inode_marks_by_group(group);
1362 		goto fput_and_out;
1363 	}
1364 
1365 	ret = fanotify_find_path(dfd, pathname, &path, flags,
1366 			(mask & ALL_FSNOTIFY_EVENTS), obj_type);
1367 	if (ret)
1368 		goto fput_and_out;
1369 
1370 	if (flags & FAN_MARK_ADD) {
1371 		ret = fanotify_events_supported(&path, mask);
1372 		if (ret)
1373 			goto path_put_and_out;
1374 	}
1375 
1376 	if (fid_mode) {
1377 		ret = fanotify_test_fid(&path, &__fsid);
1378 		if (ret)
1379 			goto path_put_and_out;
1380 
1381 		fsid = &__fsid;
1382 	}
1383 
1384 	/* inode held in place by reference to path; group by fget on fd */
1385 	if (mark_type == FAN_MARK_INODE)
1386 		inode = path.dentry->d_inode;
1387 	else
1388 		mnt = path.mnt;
1389 
1390 	/* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1391 	if (mnt || !S_ISDIR(inode->i_mode)) {
1392 		mask &= ~FAN_EVENT_ON_CHILD;
1393 		umask = FAN_EVENT_ON_CHILD;
1394 		/*
1395 		 * If group needs to report parent fid, register for getting
1396 		 * events with parent/name info for non-directory.
1397 		 */
1398 		if ((fid_mode & FAN_REPORT_DIR_FID) &&
1399 		    (flags & FAN_MARK_ADD) && !ignored)
1400 			mask |= FAN_EVENT_ON_CHILD;
1401 	}
1402 
1403 	/* create/update an inode mark */
1404 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
1405 	case FAN_MARK_ADD:
1406 		if (mark_type == FAN_MARK_MOUNT)
1407 			ret = fanotify_add_vfsmount_mark(group, mnt, mask,
1408 							 flags, fsid);
1409 		else if (mark_type == FAN_MARK_FILESYSTEM)
1410 			ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
1411 						   flags, fsid);
1412 		else
1413 			ret = fanotify_add_inode_mark(group, inode, mask,
1414 						      flags, fsid);
1415 		break;
1416 	case FAN_MARK_REMOVE:
1417 		if (mark_type == FAN_MARK_MOUNT)
1418 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
1419 							    flags, umask);
1420 		else if (mark_type == FAN_MARK_FILESYSTEM)
1421 			ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
1422 						      flags, umask);
1423 		else
1424 			ret = fanotify_remove_inode_mark(group, inode, mask,
1425 							 flags, umask);
1426 		break;
1427 	default:
1428 		ret = -EINVAL;
1429 	}
1430 
1431 path_put_and_out:
1432 	path_put(&path);
1433 fput_and_out:
1434 	fdput(f);
1435 	return ret;
1436 }
1437 
1438 #ifndef CONFIG_ARCH_SPLIT_ARG64
1439 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1440 			      __u64, mask, int, dfd,
1441 			      const char  __user *, pathname)
1442 {
1443 	return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1444 }
1445 #endif
1446 
1447 #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT)
1448 SYSCALL32_DEFINE6(fanotify_mark,
1449 				int, fanotify_fd, unsigned int, flags,
1450 				SC_ARG64(mask), int, dfd,
1451 				const char  __user *, pathname)
1452 {
1453 	return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
1454 				dfd, pathname);
1455 }
1456 #endif
1457 
1458 /*
1459  * fanotify_user_setup - Our initialization function.  Note that we cannot return
1460  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
1461  * must result in panic().
1462  */
1463 static int __init fanotify_user_setup(void)
1464 {
1465 	struct sysinfo si;
1466 	int max_marks;
1467 
1468 	si_meminfo(&si);
1469 	/*
1470 	 * Allow up to 1% of addressable memory to be accounted for per user
1471 	 * marks limited to the range [8192, 1048576]. mount and sb marks are
1472 	 * a lot cheaper than inode marks, but there is no reason for a user
1473 	 * to have many of those, so calculate by the cost of inode marks.
1474 	 */
1475 	max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
1476 		    INODE_MARK_COST;
1477 	max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
1478 				     FANOTIFY_DEFAULT_MAX_USER_MARKS);
1479 
1480 	BUILD_BUG_ON(FANOTIFY_INIT_FLAGS & FANOTIFY_INTERNAL_GROUP_FLAGS);
1481 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 10);
1482 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9);
1483 
1484 	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark,
1485 					 SLAB_PANIC|SLAB_ACCOUNT);
1486 	fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
1487 					       SLAB_PANIC);
1488 	fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
1489 						SLAB_PANIC);
1490 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
1491 		fanotify_perm_event_cachep =
1492 			KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
1493 	}
1494 
1495 	fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
1496 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
1497 					FANOTIFY_DEFAULT_MAX_GROUPS;
1498 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
1499 
1500 	return 0;
1501 }
1502 device_initcall(fanotify_user_setup);
1503