xref: /openbmc/linux/fs/notify/fanotify/fanotify_user.c (revision f8a11425075ff11b4b5784f077cb84f3d2dfb3f0)
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 (!capable(CAP_SYS_ADMIN) &&
428 	    task_tgid(current) != event->pid)
429 		metadata.pid = 0;
430 
431 	if (path && path->mnt && path->dentry) {
432 		fd = create_fd(group, path, &f);
433 		if (fd < 0)
434 			return fd;
435 	}
436 	metadata.fd = fd;
437 
438 	ret = -EFAULT;
439 	/*
440 	 * Sanity check copy size in case get_one_event() and
441 	 * event_len sizes ever get out of sync.
442 	 */
443 	if (WARN_ON_ONCE(metadata.event_len > count))
444 		goto out_close_fd;
445 
446 	if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
447 		goto out_close_fd;
448 
449 	buf += FAN_EVENT_METADATA_LEN;
450 	count -= FAN_EVENT_METADATA_LEN;
451 
452 	if (fanotify_is_perm_event(event->mask))
453 		FANOTIFY_PERM(event)->fd = fd;
454 
455 	if (f)
456 		fd_install(fd, f);
457 
458 	/* Event info records order is: dir fid + name, child fid */
459 	if (fanotify_event_dir_fh_len(event)) {
460 		info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
461 					     FAN_EVENT_INFO_TYPE_DFID;
462 		ret = copy_info_to_user(fanotify_event_fsid(event),
463 					fanotify_info_dir_fh(info),
464 					info_type, fanotify_info_name(info),
465 					info->name_len, buf, count);
466 		if (ret < 0)
467 			return ret;
468 
469 		buf += ret;
470 		count -= ret;
471 	}
472 
473 	if (fanotify_event_object_fh_len(event)) {
474 		const char *dot = NULL;
475 		int dot_len = 0;
476 
477 		if (fid_mode == FAN_REPORT_FID || info_type) {
478 			/*
479 			 * With only group flag FAN_REPORT_FID only type FID is
480 			 * reported. Second info record type is always FID.
481 			 */
482 			info_type = FAN_EVENT_INFO_TYPE_FID;
483 		} else if ((fid_mode & FAN_REPORT_NAME) &&
484 			   (event->mask & FAN_ONDIR)) {
485 			/*
486 			 * With group flag FAN_REPORT_NAME, if name was not
487 			 * recorded in an event on a directory, report the
488 			 * name "." with info type DFID_NAME.
489 			 */
490 			info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
491 			dot = ".";
492 			dot_len = 1;
493 		} else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
494 			   (event->mask & FAN_ONDIR)) {
495 			/*
496 			 * With group flag FAN_REPORT_DIR_FID, a single info
497 			 * record has type DFID for directory entry modification
498 			 * event and for event on a directory.
499 			 */
500 			info_type = FAN_EVENT_INFO_TYPE_DFID;
501 		} else {
502 			/*
503 			 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
504 			 * a single info record has type FID for event on a
505 			 * non-directory, when there is no directory to report.
506 			 * For example, on FAN_DELETE_SELF event.
507 			 */
508 			info_type = FAN_EVENT_INFO_TYPE_FID;
509 		}
510 
511 		ret = copy_info_to_user(fanotify_event_fsid(event),
512 					fanotify_event_object_fh(event),
513 					info_type, dot, dot_len, buf, count);
514 		if (ret < 0)
515 			return ret;
516 
517 		buf += ret;
518 		count -= ret;
519 	}
520 
521 	return metadata.event_len;
522 
523 out_close_fd:
524 	if (fd != FAN_NOFD) {
525 		put_unused_fd(fd);
526 		fput(f);
527 	}
528 	return ret;
529 }
530 
531 /* intofiy userspace file descriptor functions */
532 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
533 {
534 	struct fsnotify_group *group = file->private_data;
535 	__poll_t ret = 0;
536 
537 	poll_wait(file, &group->notification_waitq, wait);
538 	spin_lock(&group->notification_lock);
539 	if (!fsnotify_notify_queue_is_empty(group))
540 		ret = EPOLLIN | EPOLLRDNORM;
541 	spin_unlock(&group->notification_lock);
542 
543 	return ret;
544 }
545 
546 static ssize_t fanotify_read(struct file *file, char __user *buf,
547 			     size_t count, loff_t *pos)
548 {
549 	struct fsnotify_group *group;
550 	struct fanotify_event *event;
551 	char __user *start;
552 	int ret;
553 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
554 
555 	start = buf;
556 	group = file->private_data;
557 
558 	pr_debug("%s: group=%p\n", __func__, group);
559 
560 	add_wait_queue(&group->notification_waitq, &wait);
561 	while (1) {
562 		/*
563 		 * User can supply arbitrarily large buffer. Avoid softlockups
564 		 * in case there are lots of available events.
565 		 */
566 		cond_resched();
567 		event = get_one_event(group, count);
568 		if (IS_ERR(event)) {
569 			ret = PTR_ERR(event);
570 			break;
571 		}
572 
573 		if (!event) {
574 			ret = -EAGAIN;
575 			if (file->f_flags & O_NONBLOCK)
576 				break;
577 
578 			ret = -ERESTARTSYS;
579 			if (signal_pending(current))
580 				break;
581 
582 			if (start != buf)
583 				break;
584 
585 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
586 			continue;
587 		}
588 
589 		ret = copy_event_to_user(group, event, buf, count);
590 		if (unlikely(ret == -EOPENSTALE)) {
591 			/*
592 			 * We cannot report events with stale fd so drop it.
593 			 * Setting ret to 0 will continue the event loop and
594 			 * do the right thing if there are no more events to
595 			 * read (i.e. return bytes read, -EAGAIN or wait).
596 			 */
597 			ret = 0;
598 		}
599 
600 		/*
601 		 * Permission events get queued to wait for response.  Other
602 		 * events can be destroyed now.
603 		 */
604 		if (!fanotify_is_perm_event(event->mask)) {
605 			fsnotify_destroy_event(group, &event->fse);
606 		} else {
607 			if (ret <= 0) {
608 				spin_lock(&group->notification_lock);
609 				finish_permission_event(group,
610 					FANOTIFY_PERM(event), FAN_DENY);
611 				wake_up(&group->fanotify_data.access_waitq);
612 			} else {
613 				spin_lock(&group->notification_lock);
614 				list_add_tail(&event->fse.list,
615 					&group->fanotify_data.access_list);
616 				spin_unlock(&group->notification_lock);
617 			}
618 		}
619 		if (ret < 0)
620 			break;
621 		buf += ret;
622 		count -= ret;
623 	}
624 	remove_wait_queue(&group->notification_waitq, &wait);
625 
626 	if (start != buf && ret != -EFAULT)
627 		ret = buf - start;
628 	return ret;
629 }
630 
631 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
632 {
633 	struct fanotify_response response = { .fd = -1, .response = -1 };
634 	struct fsnotify_group *group;
635 	int ret;
636 
637 	if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
638 		return -EINVAL;
639 
640 	group = file->private_data;
641 
642 	if (count < sizeof(response))
643 		return -EINVAL;
644 
645 	count = sizeof(response);
646 
647 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
648 
649 	if (copy_from_user(&response, buf, count))
650 		return -EFAULT;
651 
652 	ret = process_access_response(group, &response);
653 	if (ret < 0)
654 		count = ret;
655 
656 	return count;
657 }
658 
659 static int fanotify_release(struct inode *ignored, struct file *file)
660 {
661 	struct fsnotify_group *group = file->private_data;
662 	struct fsnotify_event *fsn_event;
663 
664 	/*
665 	 * Stop new events from arriving in the notification queue. since
666 	 * userspace cannot use fanotify fd anymore, no event can enter or
667 	 * leave access_list by now either.
668 	 */
669 	fsnotify_group_stop_queueing(group);
670 
671 	/*
672 	 * Process all permission events on access_list and notification queue
673 	 * and simulate reply from userspace.
674 	 */
675 	spin_lock(&group->notification_lock);
676 	while (!list_empty(&group->fanotify_data.access_list)) {
677 		struct fanotify_perm_event *event;
678 
679 		event = list_first_entry(&group->fanotify_data.access_list,
680 				struct fanotify_perm_event, fae.fse.list);
681 		list_del_init(&event->fae.fse.list);
682 		finish_permission_event(group, event, FAN_ALLOW);
683 		spin_lock(&group->notification_lock);
684 	}
685 
686 	/*
687 	 * Destroy all non-permission events. For permission events just
688 	 * dequeue them and set the response. They will be freed once the
689 	 * response is consumed and fanotify_get_response() returns.
690 	 */
691 	while ((fsn_event = fsnotify_remove_first_event(group))) {
692 		struct fanotify_event *event = FANOTIFY_E(fsn_event);
693 
694 		if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
695 			spin_unlock(&group->notification_lock);
696 			fsnotify_destroy_event(group, fsn_event);
697 		} else {
698 			finish_permission_event(group, FANOTIFY_PERM(event),
699 						FAN_ALLOW);
700 		}
701 		spin_lock(&group->notification_lock);
702 	}
703 	spin_unlock(&group->notification_lock);
704 
705 	/* Response for all permission events it set, wakeup waiters */
706 	wake_up(&group->fanotify_data.access_waitq);
707 
708 	/* matches the fanotify_init->fsnotify_alloc_group */
709 	fsnotify_destroy_group(group);
710 
711 	return 0;
712 }
713 
714 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
715 {
716 	struct fsnotify_group *group;
717 	struct fsnotify_event *fsn_event;
718 	void __user *p;
719 	int ret = -ENOTTY;
720 	size_t send_len = 0;
721 
722 	group = file->private_data;
723 
724 	p = (void __user *) arg;
725 
726 	switch (cmd) {
727 	case FIONREAD:
728 		spin_lock(&group->notification_lock);
729 		list_for_each_entry(fsn_event, &group->notification_list, list)
730 			send_len += FAN_EVENT_METADATA_LEN;
731 		spin_unlock(&group->notification_lock);
732 		ret = put_user(send_len, (int __user *) p);
733 		break;
734 	}
735 
736 	return ret;
737 }
738 
739 static const struct file_operations fanotify_fops = {
740 	.show_fdinfo	= fanotify_show_fdinfo,
741 	.poll		= fanotify_poll,
742 	.read		= fanotify_read,
743 	.write		= fanotify_write,
744 	.fasync		= NULL,
745 	.release	= fanotify_release,
746 	.unlocked_ioctl	= fanotify_ioctl,
747 	.compat_ioctl	= compat_ptr_ioctl,
748 	.llseek		= noop_llseek,
749 };
750 
751 static int fanotify_find_path(int dfd, const char __user *filename,
752 			      struct path *path, unsigned int flags, __u64 mask,
753 			      unsigned int obj_type)
754 {
755 	int ret;
756 
757 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
758 		 dfd, filename, flags);
759 
760 	if (filename == NULL) {
761 		struct fd f = fdget(dfd);
762 
763 		ret = -EBADF;
764 		if (!f.file)
765 			goto out;
766 
767 		ret = -ENOTDIR;
768 		if ((flags & FAN_MARK_ONLYDIR) &&
769 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
770 			fdput(f);
771 			goto out;
772 		}
773 
774 		*path = f.file->f_path;
775 		path_get(path);
776 		fdput(f);
777 	} else {
778 		unsigned int lookup_flags = 0;
779 
780 		if (!(flags & FAN_MARK_DONT_FOLLOW))
781 			lookup_flags |= LOOKUP_FOLLOW;
782 		if (flags & FAN_MARK_ONLYDIR)
783 			lookup_flags |= LOOKUP_DIRECTORY;
784 
785 		ret = user_path_at(dfd, filename, lookup_flags, path);
786 		if (ret)
787 			goto out;
788 	}
789 
790 	/* you can only watch an inode if you have read permissions on it */
791 	ret = path_permission(path, MAY_READ);
792 	if (ret) {
793 		path_put(path);
794 		goto out;
795 	}
796 
797 	ret = security_path_notify(path, mask, obj_type);
798 	if (ret)
799 		path_put(path);
800 
801 out:
802 	return ret;
803 }
804 
805 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
806 					    __u32 mask, unsigned int flags,
807 					    __u32 umask, int *destroy)
808 {
809 	__u32 oldmask = 0;
810 
811 	/* umask bits cannot be removed by user */
812 	mask &= ~umask;
813 	spin_lock(&fsn_mark->lock);
814 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
815 		oldmask = fsn_mark->mask;
816 		fsn_mark->mask &= ~mask;
817 	} else {
818 		fsn_mark->ignored_mask &= ~mask;
819 	}
820 	/*
821 	 * We need to keep the mark around even if remaining mask cannot
822 	 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
823 	 * changes to the mask.
824 	 * Destroy mark when only umask bits remain.
825 	 */
826 	*destroy = !((fsn_mark->mask | fsn_mark->ignored_mask) & ~umask);
827 	spin_unlock(&fsn_mark->lock);
828 
829 	return mask & oldmask;
830 }
831 
832 static int fanotify_remove_mark(struct fsnotify_group *group,
833 				fsnotify_connp_t *connp, __u32 mask,
834 				unsigned int flags, __u32 umask)
835 {
836 	struct fsnotify_mark *fsn_mark = NULL;
837 	__u32 removed;
838 	int destroy_mark;
839 
840 	mutex_lock(&group->mark_mutex);
841 	fsn_mark = fsnotify_find_mark(connp, group);
842 	if (!fsn_mark) {
843 		mutex_unlock(&group->mark_mutex);
844 		return -ENOENT;
845 	}
846 
847 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
848 						 umask, &destroy_mark);
849 	if (removed & fsnotify_conn_mask(fsn_mark->connector))
850 		fsnotify_recalc_mask(fsn_mark->connector);
851 	if (destroy_mark)
852 		fsnotify_detach_mark(fsn_mark);
853 	mutex_unlock(&group->mark_mutex);
854 	if (destroy_mark)
855 		fsnotify_free_mark(fsn_mark);
856 
857 	/* matches the fsnotify_find_mark() */
858 	fsnotify_put_mark(fsn_mark);
859 	return 0;
860 }
861 
862 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
863 					 struct vfsmount *mnt, __u32 mask,
864 					 unsigned int flags, __u32 umask)
865 {
866 	return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
867 				    mask, flags, umask);
868 }
869 
870 static int fanotify_remove_sb_mark(struct fsnotify_group *group,
871 				   struct super_block *sb, __u32 mask,
872 				   unsigned int flags, __u32 umask)
873 {
874 	return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
875 				    flags, umask);
876 }
877 
878 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
879 				      struct inode *inode, __u32 mask,
880 				      unsigned int flags, __u32 umask)
881 {
882 	return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
883 				    flags, umask);
884 }
885 
886 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
887 				       __u32 mask,
888 				       unsigned int flags)
889 {
890 	__u32 oldmask = -1;
891 
892 	spin_lock(&fsn_mark->lock);
893 	if (!(flags & FAN_MARK_IGNORED_MASK)) {
894 		oldmask = fsn_mark->mask;
895 		fsn_mark->mask |= mask;
896 	} else {
897 		fsn_mark->ignored_mask |= mask;
898 		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
899 			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
900 	}
901 	spin_unlock(&fsn_mark->lock);
902 
903 	return mask & ~oldmask;
904 }
905 
906 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
907 						   fsnotify_connp_t *connp,
908 						   unsigned int type,
909 						   __kernel_fsid_t *fsid)
910 {
911 	struct ucounts *ucounts = group->fanotify_data.ucounts;
912 	struct fsnotify_mark *mark;
913 	int ret;
914 
915 	/*
916 	 * Enforce per user marks limits per user in all containing user ns.
917 	 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
918 	 * in the limited groups account.
919 	 */
920 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
921 	    !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
922 		return ERR_PTR(-ENOSPC);
923 
924 	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
925 	if (!mark) {
926 		ret = -ENOMEM;
927 		goto out_dec_ucounts;
928 	}
929 
930 	fsnotify_init_mark(mark, group);
931 	ret = fsnotify_add_mark_locked(mark, connp, type, 0, fsid);
932 	if (ret) {
933 		fsnotify_put_mark(mark);
934 		goto out_dec_ucounts;
935 	}
936 
937 	return mark;
938 
939 out_dec_ucounts:
940 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
941 		dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
942 	return ERR_PTR(ret);
943 }
944 
945 
946 static int fanotify_add_mark(struct fsnotify_group *group,
947 			     fsnotify_connp_t *connp, unsigned int type,
948 			     __u32 mask, unsigned int flags,
949 			     __kernel_fsid_t *fsid)
950 {
951 	struct fsnotify_mark *fsn_mark;
952 	__u32 added;
953 
954 	mutex_lock(&group->mark_mutex);
955 	fsn_mark = fsnotify_find_mark(connp, group);
956 	if (!fsn_mark) {
957 		fsn_mark = fanotify_add_new_mark(group, connp, type, fsid);
958 		if (IS_ERR(fsn_mark)) {
959 			mutex_unlock(&group->mark_mutex);
960 			return PTR_ERR(fsn_mark);
961 		}
962 	}
963 	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
964 	if (added & ~fsnotify_conn_mask(fsn_mark->connector))
965 		fsnotify_recalc_mask(fsn_mark->connector);
966 	mutex_unlock(&group->mark_mutex);
967 
968 	fsnotify_put_mark(fsn_mark);
969 	return 0;
970 }
971 
972 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
973 				      struct vfsmount *mnt, __u32 mask,
974 				      unsigned int flags, __kernel_fsid_t *fsid)
975 {
976 	return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
977 				 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
978 }
979 
980 static int fanotify_add_sb_mark(struct fsnotify_group *group,
981 				struct super_block *sb, __u32 mask,
982 				unsigned int flags, __kernel_fsid_t *fsid)
983 {
984 	return fanotify_add_mark(group, &sb->s_fsnotify_marks,
985 				 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
986 }
987 
988 static int fanotify_add_inode_mark(struct fsnotify_group *group,
989 				   struct inode *inode, __u32 mask,
990 				   unsigned int flags, __kernel_fsid_t *fsid)
991 {
992 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
993 
994 	/*
995 	 * If some other task has this inode open for write we should not add
996 	 * an ignored mark, unless that ignored mark is supposed to survive
997 	 * modification changes anyway.
998 	 */
999 	if ((flags & FAN_MARK_IGNORED_MASK) &&
1000 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1001 	    inode_is_open_for_write(inode))
1002 		return 0;
1003 
1004 	return fanotify_add_mark(group, &inode->i_fsnotify_marks,
1005 				 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
1006 }
1007 
1008 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1009 {
1010 	struct fanotify_event *oevent;
1011 
1012 	oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1013 	if (!oevent)
1014 		return NULL;
1015 
1016 	fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1017 	oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1018 
1019 	return &oevent->fse;
1020 }
1021 
1022 static struct hlist_head *fanotify_alloc_merge_hash(void)
1023 {
1024 	struct hlist_head *hash;
1025 
1026 	hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1027 		       GFP_KERNEL_ACCOUNT);
1028 	if (!hash)
1029 		return NULL;
1030 
1031 	__hash_init(hash, FANOTIFY_HTABLE_SIZE);
1032 
1033 	return hash;
1034 }
1035 
1036 /* fanotify syscalls */
1037 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1038 {
1039 	struct fsnotify_group *group;
1040 	int f_flags, fd;
1041 	unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1042 	unsigned int class = flags & FANOTIFY_CLASS_BITS;
1043 
1044 	pr_debug("%s: flags=%x event_f_flags=%x\n",
1045 		 __func__, flags, event_f_flags);
1046 
1047 	if (!capable(CAP_SYS_ADMIN)) {
1048 		/*
1049 		 * An unprivileged user can setup an fanotify group with
1050 		 * limited functionality - an unprivileged group is limited to
1051 		 * notification events with file handles and it cannot use
1052 		 * unlimited queue/marks.
1053 		 */
1054 		if ((flags & FANOTIFY_ADMIN_INIT_FLAGS) || !fid_mode)
1055 			return -EPERM;
1056 	}
1057 
1058 #ifdef CONFIG_AUDITSYSCALL
1059 	if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1060 #else
1061 	if (flags & ~FANOTIFY_INIT_FLAGS)
1062 #endif
1063 		return -EINVAL;
1064 
1065 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1066 		return -EINVAL;
1067 
1068 	switch (event_f_flags & O_ACCMODE) {
1069 	case O_RDONLY:
1070 	case O_RDWR:
1071 	case O_WRONLY:
1072 		break;
1073 	default:
1074 		return -EINVAL;
1075 	}
1076 
1077 	if (fid_mode && class != FAN_CLASS_NOTIF)
1078 		return -EINVAL;
1079 
1080 	/*
1081 	 * Child name is reported with parent fid so requires dir fid.
1082 	 * We can report both child fid and dir fid with or without name.
1083 	 */
1084 	if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1085 		return -EINVAL;
1086 
1087 	f_flags = O_RDWR | FMODE_NONOTIFY;
1088 	if (flags & FAN_CLOEXEC)
1089 		f_flags |= O_CLOEXEC;
1090 	if (flags & FAN_NONBLOCK)
1091 		f_flags |= O_NONBLOCK;
1092 
1093 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
1094 	group = fsnotify_alloc_user_group(&fanotify_fsnotify_ops);
1095 	if (IS_ERR(group)) {
1096 		return PTR_ERR(group);
1097 	}
1098 
1099 	/* Enforce groups limits per user in all containing user ns */
1100 	group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1101 						  current_euid(),
1102 						  UCOUNT_FANOTIFY_GROUPS);
1103 	if (!group->fanotify_data.ucounts) {
1104 		fd = -EMFILE;
1105 		goto out_destroy_group;
1106 	}
1107 
1108 	group->fanotify_data.flags = flags;
1109 	group->memcg = get_mem_cgroup_from_mm(current->mm);
1110 
1111 	group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1112 	if (!group->fanotify_data.merge_hash) {
1113 		fd = -ENOMEM;
1114 		goto out_destroy_group;
1115 	}
1116 
1117 	group->overflow_event = fanotify_alloc_overflow_event();
1118 	if (unlikely(!group->overflow_event)) {
1119 		fd = -ENOMEM;
1120 		goto out_destroy_group;
1121 	}
1122 
1123 	if (force_o_largefile())
1124 		event_f_flags |= O_LARGEFILE;
1125 	group->fanotify_data.f_flags = event_f_flags;
1126 	init_waitqueue_head(&group->fanotify_data.access_waitq);
1127 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
1128 	switch (class) {
1129 	case FAN_CLASS_NOTIF:
1130 		group->priority = FS_PRIO_0;
1131 		break;
1132 	case FAN_CLASS_CONTENT:
1133 		group->priority = FS_PRIO_1;
1134 		break;
1135 	case FAN_CLASS_PRE_CONTENT:
1136 		group->priority = FS_PRIO_2;
1137 		break;
1138 	default:
1139 		fd = -EINVAL;
1140 		goto out_destroy_group;
1141 	}
1142 
1143 	if (flags & FAN_UNLIMITED_QUEUE) {
1144 		fd = -EPERM;
1145 		if (!capable(CAP_SYS_ADMIN))
1146 			goto out_destroy_group;
1147 		group->max_events = UINT_MAX;
1148 	} else {
1149 		group->max_events = fanotify_max_queued_events;
1150 	}
1151 
1152 	if (flags & FAN_UNLIMITED_MARKS) {
1153 		fd = -EPERM;
1154 		if (!capable(CAP_SYS_ADMIN))
1155 			goto out_destroy_group;
1156 	}
1157 
1158 	if (flags & FAN_ENABLE_AUDIT) {
1159 		fd = -EPERM;
1160 		if (!capable(CAP_AUDIT_WRITE))
1161 			goto out_destroy_group;
1162 	}
1163 
1164 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
1165 	if (fd < 0)
1166 		goto out_destroy_group;
1167 
1168 	return fd;
1169 
1170 out_destroy_group:
1171 	fsnotify_destroy_group(group);
1172 	return fd;
1173 }
1174 
1175 /* Check if filesystem can encode a unique fid */
1176 static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
1177 {
1178 	__kernel_fsid_t root_fsid;
1179 	int err;
1180 
1181 	/*
1182 	 * Make sure path is not in filesystem with zero fsid (e.g. tmpfs).
1183 	 */
1184 	err = vfs_get_fsid(path->dentry, fsid);
1185 	if (err)
1186 		return err;
1187 
1188 	if (!fsid->val[0] && !fsid->val[1])
1189 		return -ENODEV;
1190 
1191 	/*
1192 	 * Make sure path is not inside a filesystem subvolume (e.g. btrfs)
1193 	 * which uses a different fsid than sb root.
1194 	 */
1195 	err = vfs_get_fsid(path->dentry->d_sb->s_root, &root_fsid);
1196 	if (err)
1197 		return err;
1198 
1199 	if (root_fsid.val[0] != fsid->val[0] ||
1200 	    root_fsid.val[1] != fsid->val[1])
1201 		return -EXDEV;
1202 
1203 	/*
1204 	 * We need to make sure that the file system supports at least
1205 	 * encoding a file handle so user can use name_to_handle_at() to
1206 	 * compare fid returned with event to the file handle of watched
1207 	 * objects. However, name_to_handle_at() requires that the
1208 	 * filesystem also supports decoding file handles.
1209 	 */
1210 	if (!path->dentry->d_sb->s_export_op ||
1211 	    !path->dentry->d_sb->s_export_op->fh_to_dentry)
1212 		return -EOPNOTSUPP;
1213 
1214 	return 0;
1215 }
1216 
1217 static int fanotify_events_supported(struct path *path, __u64 mask)
1218 {
1219 	/*
1220 	 * Some filesystems such as 'proc' acquire unusual locks when opening
1221 	 * files. For them fanotify permission events have high chances of
1222 	 * deadlocking the system - open done when reporting fanotify event
1223 	 * blocks on this "unusual" lock while another process holding the lock
1224 	 * waits for fanotify permission event to be answered. Just disallow
1225 	 * permission events for such filesystems.
1226 	 */
1227 	if (mask & FANOTIFY_PERM_EVENTS &&
1228 	    path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1229 		return -EINVAL;
1230 	return 0;
1231 }
1232 
1233 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1234 			    int dfd, const char  __user *pathname)
1235 {
1236 	struct inode *inode = NULL;
1237 	struct vfsmount *mnt = NULL;
1238 	struct fsnotify_group *group;
1239 	struct fd f;
1240 	struct path path;
1241 	__kernel_fsid_t __fsid, *fsid = NULL;
1242 	u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1243 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1244 	bool ignored = flags & FAN_MARK_IGNORED_MASK;
1245 	unsigned int obj_type, fid_mode;
1246 	u32 umask = 0;
1247 	int ret;
1248 
1249 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1250 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
1251 
1252 	/* we only use the lower 32 bits as of right now. */
1253 	if (upper_32_bits(mask))
1254 		return -EINVAL;
1255 
1256 	if (flags & ~FANOTIFY_MARK_FLAGS)
1257 		return -EINVAL;
1258 
1259 	switch (mark_type) {
1260 	case FAN_MARK_INODE:
1261 		obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1262 		break;
1263 	case FAN_MARK_MOUNT:
1264 		obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1265 		break;
1266 	case FAN_MARK_FILESYSTEM:
1267 		obj_type = FSNOTIFY_OBJ_TYPE_SB;
1268 		break;
1269 	default:
1270 		return -EINVAL;
1271 	}
1272 
1273 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
1274 	case FAN_MARK_ADD:
1275 	case FAN_MARK_REMOVE:
1276 		if (!mask)
1277 			return -EINVAL;
1278 		break;
1279 	case FAN_MARK_FLUSH:
1280 		if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1281 			return -EINVAL;
1282 		break;
1283 	default:
1284 		return -EINVAL;
1285 	}
1286 
1287 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1288 		valid_mask |= FANOTIFY_PERM_EVENTS;
1289 
1290 	if (mask & ~valid_mask)
1291 		return -EINVAL;
1292 
1293 	/* Event flags (ONDIR, ON_CHILD) are meaningless in ignored mask */
1294 	if (ignored)
1295 		mask &= ~FANOTIFY_EVENT_FLAGS;
1296 
1297 	f = fdget(fanotify_fd);
1298 	if (unlikely(!f.file))
1299 		return -EBADF;
1300 
1301 	/* verify that this is indeed an fanotify instance */
1302 	ret = -EINVAL;
1303 	if (unlikely(f.file->f_op != &fanotify_fops))
1304 		goto fput_and_out;
1305 	group = f.file->private_data;
1306 
1307 	/*
1308 	 * An unprivileged user is not allowed to watch a mount point nor
1309 	 * a filesystem.
1310 	 */
1311 	ret = -EPERM;
1312 	if (!capable(CAP_SYS_ADMIN) &&
1313 	    mark_type != FAN_MARK_INODE)
1314 		goto fput_and_out;
1315 
1316 	/*
1317 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
1318 	 * allowed to set permissions events.
1319 	 */
1320 	ret = -EINVAL;
1321 	if (mask & FANOTIFY_PERM_EVENTS &&
1322 	    group->priority == FS_PRIO_0)
1323 		goto fput_and_out;
1324 
1325 	/*
1326 	 * Events with data type inode do not carry enough information to report
1327 	 * event->fd, so we do not allow setting a mask for inode events unless
1328 	 * group supports reporting fid.
1329 	 * inode events are not supported on a mount mark, because they do not
1330 	 * carry enough information (i.e. path) to be filtered by mount point.
1331 	 */
1332 	fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1333 	if (mask & FANOTIFY_INODE_EVENTS &&
1334 	    (!fid_mode || mark_type == FAN_MARK_MOUNT))
1335 		goto fput_and_out;
1336 
1337 	if (flags & FAN_MARK_FLUSH) {
1338 		ret = 0;
1339 		if (mark_type == FAN_MARK_MOUNT)
1340 			fsnotify_clear_vfsmount_marks_by_group(group);
1341 		else if (mark_type == FAN_MARK_FILESYSTEM)
1342 			fsnotify_clear_sb_marks_by_group(group);
1343 		else
1344 			fsnotify_clear_inode_marks_by_group(group);
1345 		goto fput_and_out;
1346 	}
1347 
1348 	ret = fanotify_find_path(dfd, pathname, &path, flags,
1349 			(mask & ALL_FSNOTIFY_EVENTS), obj_type);
1350 	if (ret)
1351 		goto fput_and_out;
1352 
1353 	if (flags & FAN_MARK_ADD) {
1354 		ret = fanotify_events_supported(&path, mask);
1355 		if (ret)
1356 			goto path_put_and_out;
1357 	}
1358 
1359 	if (fid_mode) {
1360 		ret = fanotify_test_fid(&path, &__fsid);
1361 		if (ret)
1362 			goto path_put_and_out;
1363 
1364 		fsid = &__fsid;
1365 	}
1366 
1367 	/* inode held in place by reference to path; group by fget on fd */
1368 	if (mark_type == FAN_MARK_INODE)
1369 		inode = path.dentry->d_inode;
1370 	else
1371 		mnt = path.mnt;
1372 
1373 	/* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1374 	if (mnt || !S_ISDIR(inode->i_mode)) {
1375 		mask &= ~FAN_EVENT_ON_CHILD;
1376 		umask = FAN_EVENT_ON_CHILD;
1377 		/*
1378 		 * If group needs to report parent fid, register for getting
1379 		 * events with parent/name info for non-directory.
1380 		 */
1381 		if ((fid_mode & FAN_REPORT_DIR_FID) &&
1382 		    (flags & FAN_MARK_ADD) && !ignored)
1383 			mask |= FAN_EVENT_ON_CHILD;
1384 	}
1385 
1386 	/* create/update an inode mark */
1387 	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
1388 	case FAN_MARK_ADD:
1389 		if (mark_type == FAN_MARK_MOUNT)
1390 			ret = fanotify_add_vfsmount_mark(group, mnt, mask,
1391 							 flags, fsid);
1392 		else if (mark_type == FAN_MARK_FILESYSTEM)
1393 			ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
1394 						   flags, fsid);
1395 		else
1396 			ret = fanotify_add_inode_mark(group, inode, mask,
1397 						      flags, fsid);
1398 		break;
1399 	case FAN_MARK_REMOVE:
1400 		if (mark_type == FAN_MARK_MOUNT)
1401 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
1402 							    flags, umask);
1403 		else if (mark_type == FAN_MARK_FILESYSTEM)
1404 			ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
1405 						      flags, umask);
1406 		else
1407 			ret = fanotify_remove_inode_mark(group, inode, mask,
1408 							 flags, umask);
1409 		break;
1410 	default:
1411 		ret = -EINVAL;
1412 	}
1413 
1414 path_put_and_out:
1415 	path_put(&path);
1416 fput_and_out:
1417 	fdput(f);
1418 	return ret;
1419 }
1420 
1421 #ifndef CONFIG_ARCH_SPLIT_ARG64
1422 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1423 			      __u64, mask, int, dfd,
1424 			      const char  __user *, pathname)
1425 {
1426 	return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1427 }
1428 #endif
1429 
1430 #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT)
1431 SYSCALL32_DEFINE6(fanotify_mark,
1432 				int, fanotify_fd, unsigned int, flags,
1433 				SC_ARG64(mask), int, dfd,
1434 				const char  __user *, pathname)
1435 {
1436 	return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
1437 				dfd, pathname);
1438 }
1439 #endif
1440 
1441 /*
1442  * fanotify_user_setup - Our initialization function.  Note that we cannot return
1443  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
1444  * must result in panic().
1445  */
1446 static int __init fanotify_user_setup(void)
1447 {
1448 	struct sysinfo si;
1449 	int max_marks;
1450 
1451 	si_meminfo(&si);
1452 	/*
1453 	 * Allow up to 1% of addressable memory to be accounted for per user
1454 	 * marks limited to the range [8192, 1048576]. mount and sb marks are
1455 	 * a lot cheaper than inode marks, but there is no reason for a user
1456 	 * to have many of those, so calculate by the cost of inode marks.
1457 	 */
1458 	max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
1459 		    INODE_MARK_COST;
1460 	max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
1461 				     FANOTIFY_DEFAULT_MAX_USER_MARKS);
1462 
1463 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 10);
1464 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9);
1465 
1466 	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark,
1467 					 SLAB_PANIC|SLAB_ACCOUNT);
1468 	fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
1469 					       SLAB_PANIC);
1470 	fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
1471 						SLAB_PANIC);
1472 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
1473 		fanotify_perm_event_cachep =
1474 			KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
1475 	}
1476 
1477 	fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
1478 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
1479 					FANOTIFY_DEFAULT_MAX_GROUPS;
1480 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
1481 
1482 	return 0;
1483 }
1484 device_initcall(fanotify_user_setup);
1485