1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/fanotify.h>
3 #include <linux/fcntl.h>
4 #include <linux/fdtable.h>
5 #include <linux/file.h>
6 #include <linux/fs.h>
7 #include <linux/anon_inodes.h>
8 #include <linux/fsnotify_backend.h>
9 #include <linux/init.h>
10 #include <linux/mount.h>
11 #include <linux/namei.h>
12 #include <linux/poll.h>
13 #include <linux/security.h>
14 #include <linux/syscalls.h>
15 #include <linux/slab.h>
16 #include <linux/types.h>
17 #include <linux/uaccess.h>
18 #include <linux/compat.h>
19 #include <linux/sched/signal.h>
20 #include <linux/memcontrol.h>
21 #include <linux/statfs.h>
22 #include <linux/exportfs.h>
23 
24 #include <asm/ioctls.h>
25 
26 #include "../../mount.h"
27 #include "../fdinfo.h"
28 #include "fanotify.h"
29 
30 #define FANOTIFY_DEFAULT_MAX_EVENTS	16384
31 #define FANOTIFY_OLD_DEFAULT_MAX_MARKS	8192
32 #define FANOTIFY_DEFAULT_MAX_GROUPS	128
33 #define FANOTIFY_DEFAULT_FEE_POOL_SIZE	32
34 
35 /*
36  * Legacy fanotify marks limits (8192) is per group and we introduced a tunable
37  * limit of marks per user, similar to inotify.  Effectively, the legacy limit
38  * of fanotify marks per user is <max marks per group> * <max groups per user>.
39  * This default limit (1M) also happens to match the increased limit of inotify
40  * max_user_watches since v5.10.
41  */
42 #define FANOTIFY_DEFAULT_MAX_USER_MARKS	\
43 	(FANOTIFY_OLD_DEFAULT_MAX_MARKS * FANOTIFY_DEFAULT_MAX_GROUPS)
44 
45 /*
46  * Most of the memory cost of adding an inode mark is pinning the marked inode.
47  * The size of the filesystem inode struct is not uniform across filesystems,
48  * so double the size of a VFS inode is used as a conservative approximation.
49  */
50 #define INODE_MARK_COST	(2 * sizeof(struct inode))
51 
52 /* configurable via /proc/sys/fs/fanotify/ */
53 static int fanotify_max_queued_events __read_mostly;
54 
55 #ifdef CONFIG_SYSCTL
56 
57 #include <linux/sysctl.h>
58 
59 static long ft_zero = 0;
60 static long ft_int_max = INT_MAX;
61 
62 static struct ctl_table fanotify_table[] = {
63 	{
64 		.procname	= "max_user_groups",
65 		.data	= &init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS],
66 		.maxlen		= sizeof(long),
67 		.mode		= 0644,
68 		.proc_handler	= proc_doulongvec_minmax,
69 		.extra1		= &ft_zero,
70 		.extra2		= &ft_int_max,
71 	},
72 	{
73 		.procname	= "max_user_marks",
74 		.data	= &init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS],
75 		.maxlen		= sizeof(long),
76 		.mode		= 0644,
77 		.proc_handler	= proc_doulongvec_minmax,
78 		.extra1		= &ft_zero,
79 		.extra2		= &ft_int_max,
80 	},
81 	{
82 		.procname	= "max_queued_events",
83 		.data		= &fanotify_max_queued_events,
84 		.maxlen		= sizeof(int),
85 		.mode		= 0644,
86 		.proc_handler	= proc_dointvec_minmax,
87 		.extra1		= SYSCTL_ZERO
88 	},
89 	{ }
90 };
91 
92 static void __init fanotify_sysctls_init(void)
93 {
94 	register_sysctl("fs/fanotify", fanotify_table);
95 }
96 #else
97 #define fanotify_sysctls_init() do { } while (0)
98 #endif /* CONFIG_SYSCTL */
99 
100 /*
101  * All flags that may be specified in parameter event_f_flags of fanotify_init.
102  *
103  * Internal and external open flags are stored together in field f_flags of
104  * struct file. Only external open flags shall be allowed in event_f_flags.
105  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
106  * excluded.
107  */
108 #define	FANOTIFY_INIT_ALL_EVENT_F_BITS				( \
109 		O_ACCMODE	| O_APPEND	| O_NONBLOCK	| \
110 		__O_SYNC	| O_DSYNC	| O_CLOEXEC     | \
111 		O_LARGEFILE	| O_NOATIME	)
112 
113 extern const struct fsnotify_ops fanotify_fsnotify_ops;
114 
115 struct kmem_cache *fanotify_mark_cache __read_mostly;
116 struct kmem_cache *fanotify_fid_event_cachep __read_mostly;
117 struct kmem_cache *fanotify_path_event_cachep __read_mostly;
118 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
119 
120 #define FANOTIFY_EVENT_ALIGN 4
121 #define FANOTIFY_FID_INFO_HDR_LEN \
122 	(sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle))
123 #define FANOTIFY_PIDFD_INFO_HDR_LEN \
124 	sizeof(struct fanotify_event_info_pidfd)
125 #define FANOTIFY_ERROR_INFO_LEN \
126 	(sizeof(struct fanotify_event_info_error))
127 
128 static int fanotify_fid_info_len(int fh_len, int name_len)
129 {
130 	int info_len = fh_len;
131 
132 	if (name_len)
133 		info_len += name_len + 1;
134 
135 	return roundup(FANOTIFY_FID_INFO_HDR_LEN + info_len,
136 		       FANOTIFY_EVENT_ALIGN);
137 }
138 
139 /* FAN_RENAME may have one or two dir+name info records */
140 static int fanotify_dir_name_info_len(struct fanotify_event *event)
141 {
142 	struct fanotify_info *info = fanotify_event_info(event);
143 	int dir_fh_len = fanotify_event_dir_fh_len(event);
144 	int dir2_fh_len = fanotify_event_dir2_fh_len(event);
145 	int info_len = 0;
146 
147 	if (dir_fh_len)
148 		info_len += fanotify_fid_info_len(dir_fh_len,
149 						  info->name_len);
150 	if (dir2_fh_len)
151 		info_len += fanotify_fid_info_len(dir2_fh_len,
152 						  info->name2_len);
153 
154 	return info_len;
155 }
156 
157 static size_t fanotify_event_len(unsigned int info_mode,
158 				 struct fanotify_event *event)
159 {
160 	size_t event_len = FAN_EVENT_METADATA_LEN;
161 	int fh_len;
162 	int dot_len = 0;
163 
164 	if (!info_mode)
165 		return event_len;
166 
167 	if (fanotify_is_error_event(event->mask))
168 		event_len += FANOTIFY_ERROR_INFO_LEN;
169 
170 	if (fanotify_event_has_any_dir_fh(event)) {
171 		event_len += fanotify_dir_name_info_len(event);
172 	} else if ((info_mode & FAN_REPORT_NAME) &&
173 		   (event->mask & FAN_ONDIR)) {
174 		/*
175 		 * With group flag FAN_REPORT_NAME, if name was not recorded in
176 		 * event on a directory, we will report the name ".".
177 		 */
178 		dot_len = 1;
179 	}
180 
181 	if (info_mode & FAN_REPORT_PIDFD)
182 		event_len += FANOTIFY_PIDFD_INFO_HDR_LEN;
183 
184 	if (fanotify_event_has_object_fh(event)) {
185 		fh_len = fanotify_event_object_fh_len(event);
186 		event_len += fanotify_fid_info_len(fh_len, dot_len);
187 	}
188 
189 	return event_len;
190 }
191 
192 /*
193  * Remove an hashed event from merge hash table.
194  */
195 static void fanotify_unhash_event(struct fsnotify_group *group,
196 				  struct fanotify_event *event)
197 {
198 	assert_spin_locked(&group->notification_lock);
199 
200 	pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
201 		 group, event, fanotify_event_hash_bucket(group, event));
202 
203 	if (WARN_ON_ONCE(hlist_unhashed(&event->merge_list)))
204 		return;
205 
206 	hlist_del_init(&event->merge_list);
207 }
208 
209 /*
210  * Get an fanotify notification event if one exists and is small
211  * enough to fit in "count". Return an error pointer if the count
212  * is not large enough. When permission event is dequeued, its state is
213  * updated accordingly.
214  */
215 static struct fanotify_event *get_one_event(struct fsnotify_group *group,
216 					    size_t count)
217 {
218 	size_t event_size;
219 	struct fanotify_event *event = NULL;
220 	struct fsnotify_event *fsn_event;
221 	unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
222 
223 	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
224 
225 	spin_lock(&group->notification_lock);
226 	fsn_event = fsnotify_peek_first_event(group);
227 	if (!fsn_event)
228 		goto out;
229 
230 	event = FANOTIFY_E(fsn_event);
231 	event_size = fanotify_event_len(info_mode, event);
232 
233 	if (event_size > count) {
234 		event = ERR_PTR(-EINVAL);
235 		goto out;
236 	}
237 
238 	/*
239 	 * Held the notification_lock the whole time, so this is the
240 	 * same event we peeked above.
241 	 */
242 	fsnotify_remove_first_event(group);
243 	if (fanotify_is_perm_event(event->mask))
244 		FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED;
245 	if (fanotify_is_hashed_event(event->mask))
246 		fanotify_unhash_event(group, event);
247 out:
248 	spin_unlock(&group->notification_lock);
249 	return event;
250 }
251 
252 static int create_fd(struct fsnotify_group *group, const struct path *path,
253 		     struct file **file)
254 {
255 	int client_fd;
256 	struct file *new_file;
257 
258 	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
259 	if (client_fd < 0)
260 		return client_fd;
261 
262 	/*
263 	 * we need a new file handle for the userspace program so it can read even if it was
264 	 * originally opened O_WRONLY.
265 	 */
266 	new_file = dentry_open(path,
267 			       group->fanotify_data.f_flags | __FMODE_NONOTIFY,
268 			       current_cred());
269 	if (IS_ERR(new_file)) {
270 		put_unused_fd(client_fd);
271 		client_fd = PTR_ERR(new_file);
272 	} else {
273 		*file = new_file;
274 	}
275 
276 	return client_fd;
277 }
278 
279 static int process_access_response_info(const char __user *info,
280 					size_t info_len,
281 				struct fanotify_response_info_audit_rule *friar)
282 {
283 	if (info_len != sizeof(*friar))
284 		return -EINVAL;
285 
286 	if (copy_from_user(friar, info, sizeof(*friar)))
287 		return -EFAULT;
288 
289 	if (friar->hdr.type != FAN_RESPONSE_INFO_AUDIT_RULE)
290 		return -EINVAL;
291 	if (friar->hdr.pad != 0)
292 		return -EINVAL;
293 	if (friar->hdr.len != sizeof(*friar))
294 		return -EINVAL;
295 
296 	return info_len;
297 }
298 
299 /*
300  * Finish processing of permission event by setting it to ANSWERED state and
301  * drop group->notification_lock.
302  */
303 static void finish_permission_event(struct fsnotify_group *group,
304 				    struct fanotify_perm_event *event, u32 response,
305 				    struct fanotify_response_info_audit_rule *friar)
306 				    __releases(&group->notification_lock)
307 {
308 	bool destroy = false;
309 
310 	assert_spin_locked(&group->notification_lock);
311 	event->response = response & ~FAN_INFO;
312 	if (response & FAN_INFO)
313 		memcpy(&event->audit_rule, friar, sizeof(*friar));
314 
315 	if (event->state == FAN_EVENT_CANCELED)
316 		destroy = true;
317 	else
318 		event->state = FAN_EVENT_ANSWERED;
319 	spin_unlock(&group->notification_lock);
320 	if (destroy)
321 		fsnotify_destroy_event(group, &event->fae.fse);
322 }
323 
324 static int process_access_response(struct fsnotify_group *group,
325 				   struct fanotify_response *response_struct,
326 				   const char __user *info,
327 				   size_t info_len)
328 {
329 	struct fanotify_perm_event *event;
330 	int fd = response_struct->fd;
331 	u32 response = response_struct->response;
332 	int ret = info_len;
333 	struct fanotify_response_info_audit_rule friar;
334 
335 	pr_debug("%s: group=%p fd=%d response=%u buf=%p size=%zu\n", __func__,
336 		 group, fd, response, info, info_len);
337 	/*
338 	 * make sure the response is valid, if invalid we do nothing and either
339 	 * userspace can send a valid response or we will clean it up after the
340 	 * timeout
341 	 */
342 	if (response & ~FANOTIFY_RESPONSE_VALID_MASK)
343 		return -EINVAL;
344 
345 	switch (response & FANOTIFY_RESPONSE_ACCESS) {
346 	case FAN_ALLOW:
347 	case FAN_DENY:
348 		break;
349 	default:
350 		return -EINVAL;
351 	}
352 
353 	if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
354 		return -EINVAL;
355 
356 	if (response & FAN_INFO) {
357 		ret = process_access_response_info(info, info_len, &friar);
358 		if (ret < 0)
359 			return ret;
360 		if (fd == FAN_NOFD)
361 			return ret;
362 	} else {
363 		ret = 0;
364 	}
365 
366 	if (fd < 0)
367 		return -EINVAL;
368 
369 	spin_lock(&group->notification_lock);
370 	list_for_each_entry(event, &group->fanotify_data.access_list,
371 			    fae.fse.list) {
372 		if (event->fd != fd)
373 			continue;
374 
375 		list_del_init(&event->fae.fse.list);
376 		finish_permission_event(group, event, response, &friar);
377 		wake_up(&group->fanotify_data.access_waitq);
378 		return ret;
379 	}
380 	spin_unlock(&group->notification_lock);
381 
382 	return -ENOENT;
383 }
384 
385 static size_t copy_error_info_to_user(struct fanotify_event *event,
386 				      char __user *buf, int count)
387 {
388 	struct fanotify_event_info_error info = { };
389 	struct fanotify_error_event *fee = FANOTIFY_EE(event);
390 
391 	info.hdr.info_type = FAN_EVENT_INFO_TYPE_ERROR;
392 	info.hdr.len = FANOTIFY_ERROR_INFO_LEN;
393 
394 	if (WARN_ON(count < info.hdr.len))
395 		return -EFAULT;
396 
397 	info.error = fee->error;
398 	info.error_count = fee->err_count;
399 
400 	if (copy_to_user(buf, &info, sizeof(info)))
401 		return -EFAULT;
402 
403 	return info.hdr.len;
404 }
405 
406 static int copy_fid_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
407 				 int info_type, const char *name,
408 				 size_t name_len,
409 				 char __user *buf, size_t count)
410 {
411 	struct fanotify_event_info_fid info = { };
412 	struct file_handle handle = { };
413 	unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
414 	size_t fh_len = fh ? fh->len : 0;
415 	size_t info_len = fanotify_fid_info_len(fh_len, name_len);
416 	size_t len = info_len;
417 
418 	pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
419 		 __func__, fh_len, name_len, info_len, count);
420 
421 	if (WARN_ON_ONCE(len < sizeof(info) || len > count))
422 		return -EFAULT;
423 
424 	/*
425 	 * Copy event info fid header followed by variable sized file handle
426 	 * and optionally followed by variable sized filename.
427 	 */
428 	switch (info_type) {
429 	case FAN_EVENT_INFO_TYPE_FID:
430 	case FAN_EVENT_INFO_TYPE_DFID:
431 		if (WARN_ON_ONCE(name_len))
432 			return -EFAULT;
433 		break;
434 	case FAN_EVENT_INFO_TYPE_DFID_NAME:
435 	case FAN_EVENT_INFO_TYPE_OLD_DFID_NAME:
436 	case FAN_EVENT_INFO_TYPE_NEW_DFID_NAME:
437 		if (WARN_ON_ONCE(!name || !name_len))
438 			return -EFAULT;
439 		break;
440 	default:
441 		return -EFAULT;
442 	}
443 
444 	info.hdr.info_type = info_type;
445 	info.hdr.len = len;
446 	info.fsid = *fsid;
447 	if (copy_to_user(buf, &info, sizeof(info)))
448 		return -EFAULT;
449 
450 	buf += sizeof(info);
451 	len -= sizeof(info);
452 	if (WARN_ON_ONCE(len < sizeof(handle)))
453 		return -EFAULT;
454 
455 	handle.handle_type = fh->type;
456 	handle.handle_bytes = fh_len;
457 
458 	/* Mangle handle_type for bad file_handle */
459 	if (!fh_len)
460 		handle.handle_type = FILEID_INVALID;
461 
462 	if (copy_to_user(buf, &handle, sizeof(handle)))
463 		return -EFAULT;
464 
465 	buf += sizeof(handle);
466 	len -= sizeof(handle);
467 	if (WARN_ON_ONCE(len < fh_len))
468 		return -EFAULT;
469 
470 	/*
471 	 * For an inline fh and inline file name, copy through stack to exclude
472 	 * the copy from usercopy hardening protections.
473 	 */
474 	fh_buf = fanotify_fh_buf(fh);
475 	if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
476 		memcpy(bounce, fh_buf, fh_len);
477 		fh_buf = bounce;
478 	}
479 	if (copy_to_user(buf, fh_buf, fh_len))
480 		return -EFAULT;
481 
482 	buf += fh_len;
483 	len -= fh_len;
484 
485 	if (name_len) {
486 		/* Copy the filename with terminating null */
487 		name_len++;
488 		if (WARN_ON_ONCE(len < name_len))
489 			return -EFAULT;
490 
491 		if (copy_to_user(buf, name, name_len))
492 			return -EFAULT;
493 
494 		buf += name_len;
495 		len -= name_len;
496 	}
497 
498 	/* Pad with 0's */
499 	WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
500 	if (len > 0 && clear_user(buf, len))
501 		return -EFAULT;
502 
503 	return info_len;
504 }
505 
506 static int copy_pidfd_info_to_user(int pidfd,
507 				   char __user *buf,
508 				   size_t count)
509 {
510 	struct fanotify_event_info_pidfd info = { };
511 	size_t info_len = FANOTIFY_PIDFD_INFO_HDR_LEN;
512 
513 	if (WARN_ON_ONCE(info_len > count))
514 		return -EFAULT;
515 
516 	info.hdr.info_type = FAN_EVENT_INFO_TYPE_PIDFD;
517 	info.hdr.len = info_len;
518 	info.pidfd = pidfd;
519 
520 	if (copy_to_user(buf, &info, info_len))
521 		return -EFAULT;
522 
523 	return info_len;
524 }
525 
526 static int copy_info_records_to_user(struct fanotify_event *event,
527 				     struct fanotify_info *info,
528 				     unsigned int info_mode, int pidfd,
529 				     char __user *buf, size_t count)
530 {
531 	int ret, total_bytes = 0, info_type = 0;
532 	unsigned int fid_mode = info_mode & FANOTIFY_FID_BITS;
533 	unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
534 
535 	/*
536 	 * Event info records order is as follows:
537 	 * 1. dir fid + name
538 	 * 2. (optional) new dir fid + new name
539 	 * 3. (optional) child fid
540 	 */
541 	if (fanotify_event_has_dir_fh(event)) {
542 		info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
543 					     FAN_EVENT_INFO_TYPE_DFID;
544 
545 		/* FAN_RENAME uses special info types */
546 		if (event->mask & FAN_RENAME)
547 			info_type = FAN_EVENT_INFO_TYPE_OLD_DFID_NAME;
548 
549 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
550 					    fanotify_info_dir_fh(info),
551 					    info_type,
552 					    fanotify_info_name(info),
553 					    info->name_len, buf, count);
554 		if (ret < 0)
555 			return ret;
556 
557 		buf += ret;
558 		count -= ret;
559 		total_bytes += ret;
560 	}
561 
562 	/* New dir fid+name may be reported in addition to old dir fid+name */
563 	if (fanotify_event_has_dir2_fh(event)) {
564 		info_type = FAN_EVENT_INFO_TYPE_NEW_DFID_NAME;
565 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
566 					    fanotify_info_dir2_fh(info),
567 					    info_type,
568 					    fanotify_info_name2(info),
569 					    info->name2_len, buf, count);
570 		if (ret < 0)
571 			return ret;
572 
573 		buf += ret;
574 		count -= ret;
575 		total_bytes += ret;
576 	}
577 
578 	if (fanotify_event_has_object_fh(event)) {
579 		const char *dot = NULL;
580 		int dot_len = 0;
581 
582 		if (fid_mode == FAN_REPORT_FID || info_type) {
583 			/*
584 			 * With only group flag FAN_REPORT_FID only type FID is
585 			 * reported. Second info record type is always FID.
586 			 */
587 			info_type = FAN_EVENT_INFO_TYPE_FID;
588 		} else if ((fid_mode & FAN_REPORT_NAME) &&
589 			   (event->mask & FAN_ONDIR)) {
590 			/*
591 			 * With group flag FAN_REPORT_NAME, if name was not
592 			 * recorded in an event on a directory, report the name
593 			 * "." with info type DFID_NAME.
594 			 */
595 			info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
596 			dot = ".";
597 			dot_len = 1;
598 		} else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
599 			   (event->mask & FAN_ONDIR)) {
600 			/*
601 			 * With group flag FAN_REPORT_DIR_FID, a single info
602 			 * record has type DFID for directory entry modification
603 			 * event and for event on a directory.
604 			 */
605 			info_type = FAN_EVENT_INFO_TYPE_DFID;
606 		} else {
607 			/*
608 			 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
609 			 * a single info record has type FID for event on a
610 			 * non-directory, when there is no directory to report.
611 			 * For example, on FAN_DELETE_SELF event.
612 			 */
613 			info_type = FAN_EVENT_INFO_TYPE_FID;
614 		}
615 
616 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
617 					    fanotify_event_object_fh(event),
618 					    info_type, dot, dot_len,
619 					    buf, count);
620 		if (ret < 0)
621 			return ret;
622 
623 		buf += ret;
624 		count -= ret;
625 		total_bytes += ret;
626 	}
627 
628 	if (pidfd_mode) {
629 		ret = copy_pidfd_info_to_user(pidfd, buf, count);
630 		if (ret < 0)
631 			return ret;
632 
633 		buf += ret;
634 		count -= ret;
635 		total_bytes += ret;
636 	}
637 
638 	if (fanotify_is_error_event(event->mask)) {
639 		ret = copy_error_info_to_user(event, buf, count);
640 		if (ret < 0)
641 			return ret;
642 		buf += ret;
643 		count -= ret;
644 		total_bytes += ret;
645 	}
646 
647 	return total_bytes;
648 }
649 
650 static ssize_t copy_event_to_user(struct fsnotify_group *group,
651 				  struct fanotify_event *event,
652 				  char __user *buf, size_t count)
653 {
654 	struct fanotify_event_metadata metadata;
655 	const struct path *path = fanotify_event_path(event);
656 	struct fanotify_info *info = fanotify_event_info(event);
657 	unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
658 	unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
659 	struct file *f = NULL, *pidfd_file = NULL;
660 	int ret, pidfd = -ESRCH, fd = -EBADF;
661 
662 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
663 
664 	metadata.event_len = fanotify_event_len(info_mode, event);
665 	metadata.metadata_len = FAN_EVENT_METADATA_LEN;
666 	metadata.vers = FANOTIFY_METADATA_VERSION;
667 	metadata.reserved = 0;
668 	metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
669 	metadata.pid = pid_vnr(event->pid);
670 	/*
671 	 * For an unprivileged listener, event->pid can be used to identify the
672 	 * events generated by the listener process itself, without disclosing
673 	 * the pids of other processes.
674 	 */
675 	if (FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
676 	    task_tgid(current) != event->pid)
677 		metadata.pid = 0;
678 
679 	/*
680 	 * For now, fid mode is required for an unprivileged listener and
681 	 * fid mode does not report fd in events.  Keep this check anyway
682 	 * for safety in case fid mode requirement is relaxed in the future
683 	 * to allow unprivileged listener to get events with no fd and no fid.
684 	 */
685 	if (!FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
686 	    path && path->mnt && path->dentry) {
687 		fd = create_fd(group, path, &f);
688 		/*
689 		 * Opening an fd from dentry can fail for several reasons.
690 		 * For example, when tasks are gone and we try to open their
691 		 * /proc files or we try to open a WRONLY file like in sysfs
692 		 * or when trying to open a file that was deleted on the
693 		 * remote network server.
694 		 *
695 		 * For a group with FAN_REPORT_FD_ERROR, we will send the
696 		 * event with the error instead of the open fd, otherwise
697 		 * Userspace may not get the error at all.
698 		 * In any case, userspace will not know which file failed to
699 		 * open, so add a debug print for further investigation.
700 		 */
701 		if (fd < 0) {
702 			pr_debug("fanotify: create_fd(%pd2) failed err=%d\n",
703 				 path->dentry, fd);
704 			if (!FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR)) {
705 				/*
706 				 * Historically, we've handled EOPENSTALE in a
707 				 * special way and silently dropped such
708 				 * events. Now we have to keep it to maintain
709 				 * backward compatibility...
710 				 */
711 				if (fd == -EOPENSTALE)
712 					fd = 0;
713 				return fd;
714 			}
715 		}
716 	}
717 	if (FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR))
718 		metadata.fd = fd;
719 	else
720 		metadata.fd = fd >= 0 ? fd : FAN_NOFD;
721 
722 	if (pidfd_mode) {
723 		/*
724 		 * Complain if the FAN_REPORT_PIDFD and FAN_REPORT_TID mutual
725 		 * exclusion is ever lifted. At the time of incoporating pidfd
726 		 * support within fanotify, the pidfd API only supported the
727 		 * creation of pidfds for thread-group leaders.
728 		 */
729 		WARN_ON_ONCE(FAN_GROUP_FLAG(group, FAN_REPORT_TID));
730 
731 		/*
732 		 * The PIDTYPE_TGID check for an event->pid is performed
733 		 * preemptively in an attempt to catch out cases where the event
734 		 * listener reads events after the event generating process has
735 		 * already terminated.  Depending on flag FAN_REPORT_FD_ERROR,
736 		 * report either -ESRCH or FAN_NOPIDFD to the event listener in
737 		 * those cases with all other pidfd creation errors reported as
738 		 * the error code itself or as FAN_EPIDFD.
739 		 */
740 		if (metadata.pid && pid_has_task(event->pid, PIDTYPE_TGID))
741 			pidfd = pidfd_prepare(event->pid, 0, &pidfd_file);
742 
743 		if (!FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR) && pidfd < 0)
744 			pidfd = pidfd == -ESRCH ? FAN_NOPIDFD : FAN_EPIDFD;
745 	}
746 
747 	ret = -EFAULT;
748 	/*
749 	 * Sanity check copy size in case get_one_event() and
750 	 * event_len sizes ever get out of sync.
751 	 */
752 	if (WARN_ON_ONCE(metadata.event_len > count))
753 		goto out_close_fd;
754 
755 	if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
756 		goto out_close_fd;
757 
758 	buf += FAN_EVENT_METADATA_LEN;
759 	count -= FAN_EVENT_METADATA_LEN;
760 
761 	if (info_mode) {
762 		ret = copy_info_records_to_user(event, info, info_mode, pidfd,
763 						buf, count);
764 		if (ret < 0)
765 			goto out_close_fd;
766 	}
767 
768 	if (f)
769 		fd_install(fd, f);
770 
771 	if (pidfd_file)
772 		fd_install(pidfd, pidfd_file);
773 
774 	if (fanotify_is_perm_event(event->mask))
775 		FANOTIFY_PERM(event)->fd = fd;
776 
777 	return metadata.event_len;
778 
779 out_close_fd:
780 	if (f) {
781 		put_unused_fd(fd);
782 		fput(f);
783 	}
784 
785 	if (pidfd_file) {
786 		put_unused_fd(pidfd);
787 		fput(pidfd_file);
788 	}
789 
790 	return ret;
791 }
792 
793 /* intofiy userspace file descriptor functions */
794 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
795 {
796 	struct fsnotify_group *group = file->private_data;
797 	__poll_t ret = 0;
798 
799 	poll_wait(file, &group->notification_waitq, wait);
800 	spin_lock(&group->notification_lock);
801 	if (!fsnotify_notify_queue_is_empty(group))
802 		ret = EPOLLIN | EPOLLRDNORM;
803 	spin_unlock(&group->notification_lock);
804 
805 	return ret;
806 }
807 
808 static ssize_t fanotify_read(struct file *file, char __user *buf,
809 			     size_t count, loff_t *pos)
810 {
811 	struct fsnotify_group *group;
812 	struct fanotify_event *event;
813 	char __user *start;
814 	int ret;
815 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
816 
817 	start = buf;
818 	group = file->private_data;
819 
820 	pr_debug("%s: group=%p\n", __func__, group);
821 
822 	add_wait_queue(&group->notification_waitq, &wait);
823 	while (1) {
824 		/*
825 		 * User can supply arbitrarily large buffer. Avoid softlockups
826 		 * in case there are lots of available events.
827 		 */
828 		cond_resched();
829 		event = get_one_event(group, count);
830 		if (IS_ERR(event)) {
831 			ret = PTR_ERR(event);
832 			break;
833 		}
834 
835 		if (!event) {
836 			ret = -EAGAIN;
837 			if (file->f_flags & O_NONBLOCK)
838 				break;
839 
840 			ret = -ERESTARTSYS;
841 			if (signal_pending(current))
842 				break;
843 
844 			if (start != buf)
845 				break;
846 
847 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
848 			continue;
849 		}
850 
851 		ret = copy_event_to_user(group, event, buf, count);
852 
853 		/*
854 		 * Permission events get queued to wait for response.  Other
855 		 * events can be destroyed now.
856 		 */
857 		if (!fanotify_is_perm_event(event->mask)) {
858 			fsnotify_destroy_event(group, &event->fse);
859 		} else {
860 			if (ret <= 0 || FANOTIFY_PERM(event)->fd < 0) {
861 				spin_lock(&group->notification_lock);
862 				finish_permission_event(group,
863 					FANOTIFY_PERM(event), FAN_DENY, NULL);
864 				wake_up(&group->fanotify_data.access_waitq);
865 			} else {
866 				spin_lock(&group->notification_lock);
867 				list_add_tail(&event->fse.list,
868 					&group->fanotify_data.access_list);
869 				spin_unlock(&group->notification_lock);
870 			}
871 		}
872 		if (ret < 0)
873 			break;
874 		buf += ret;
875 		count -= ret;
876 	}
877 	remove_wait_queue(&group->notification_waitq, &wait);
878 
879 	if (start != buf && ret != -EFAULT)
880 		ret = buf - start;
881 	return ret;
882 }
883 
884 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
885 {
886 	struct fanotify_response response;
887 	struct fsnotify_group *group;
888 	int ret;
889 	const char __user *info_buf = buf + sizeof(struct fanotify_response);
890 	size_t info_len;
891 
892 	if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
893 		return -EINVAL;
894 
895 	group = file->private_data;
896 
897 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
898 
899 	if (count < sizeof(response))
900 		return -EINVAL;
901 
902 	if (copy_from_user(&response, buf, sizeof(response)))
903 		return -EFAULT;
904 
905 	info_len = count - sizeof(response);
906 
907 	ret = process_access_response(group, &response, info_buf, info_len);
908 	if (ret < 0)
909 		count = ret;
910 	else
911 		count = sizeof(response) + ret;
912 
913 	return count;
914 }
915 
916 static int fanotify_release(struct inode *ignored, struct file *file)
917 {
918 	struct fsnotify_group *group = file->private_data;
919 	struct fsnotify_event *fsn_event;
920 
921 	/*
922 	 * Stop new events from arriving in the notification queue. since
923 	 * userspace cannot use fanotify fd anymore, no event can enter or
924 	 * leave access_list by now either.
925 	 */
926 	fsnotify_group_stop_queueing(group);
927 
928 	/*
929 	 * Process all permission events on access_list and notification queue
930 	 * and simulate reply from userspace.
931 	 */
932 	spin_lock(&group->notification_lock);
933 	while (!list_empty(&group->fanotify_data.access_list)) {
934 		struct fanotify_perm_event *event;
935 
936 		event = list_first_entry(&group->fanotify_data.access_list,
937 				struct fanotify_perm_event, fae.fse.list);
938 		list_del_init(&event->fae.fse.list);
939 		finish_permission_event(group, event, FAN_ALLOW, NULL);
940 		spin_lock(&group->notification_lock);
941 	}
942 
943 	/*
944 	 * Destroy all non-permission events. For permission events just
945 	 * dequeue them and set the response. They will be freed once the
946 	 * response is consumed and fanotify_get_response() returns.
947 	 */
948 	while ((fsn_event = fsnotify_remove_first_event(group))) {
949 		struct fanotify_event *event = FANOTIFY_E(fsn_event);
950 
951 		if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
952 			spin_unlock(&group->notification_lock);
953 			fsnotify_destroy_event(group, fsn_event);
954 		} else {
955 			finish_permission_event(group, FANOTIFY_PERM(event),
956 						FAN_ALLOW, NULL);
957 		}
958 		spin_lock(&group->notification_lock);
959 	}
960 	spin_unlock(&group->notification_lock);
961 
962 	/* Response for all permission events it set, wakeup waiters */
963 	wake_up(&group->fanotify_data.access_waitq);
964 
965 	/* matches the fanotify_init->fsnotify_alloc_group */
966 	fsnotify_destroy_group(group);
967 
968 	return 0;
969 }
970 
971 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
972 {
973 	struct fsnotify_group *group;
974 	struct fsnotify_event *fsn_event;
975 	void __user *p;
976 	int ret = -ENOTTY;
977 	size_t send_len = 0;
978 
979 	group = file->private_data;
980 
981 	p = (void __user *) arg;
982 
983 	switch (cmd) {
984 	case FIONREAD:
985 		spin_lock(&group->notification_lock);
986 		list_for_each_entry(fsn_event, &group->notification_list, list)
987 			send_len += FAN_EVENT_METADATA_LEN;
988 		spin_unlock(&group->notification_lock);
989 		ret = put_user(send_len, (int __user *) p);
990 		break;
991 	}
992 
993 	return ret;
994 }
995 
996 static const struct file_operations fanotify_fops = {
997 	.show_fdinfo	= fanotify_show_fdinfo,
998 	.poll		= fanotify_poll,
999 	.read		= fanotify_read,
1000 	.write		= fanotify_write,
1001 	.fasync		= NULL,
1002 	.release	= fanotify_release,
1003 	.unlocked_ioctl	= fanotify_ioctl,
1004 	.compat_ioctl	= compat_ptr_ioctl,
1005 	.llseek		= noop_llseek,
1006 };
1007 
1008 static int fanotify_find_path(int dfd, const char __user *filename,
1009 			      struct path *path, unsigned int flags, __u64 mask,
1010 			      unsigned int obj_type)
1011 {
1012 	int ret;
1013 
1014 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
1015 		 dfd, filename, flags);
1016 
1017 	if (filename == NULL) {
1018 		struct fd f = fdget(dfd);
1019 
1020 		ret = -EBADF;
1021 		if (!f.file)
1022 			goto out;
1023 
1024 		ret = -ENOTDIR;
1025 		if ((flags & FAN_MARK_ONLYDIR) &&
1026 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
1027 			fdput(f);
1028 			goto out;
1029 		}
1030 
1031 		*path = f.file->f_path;
1032 		path_get(path);
1033 		fdput(f);
1034 	} else {
1035 		unsigned int lookup_flags = 0;
1036 
1037 		if (!(flags & FAN_MARK_DONT_FOLLOW))
1038 			lookup_flags |= LOOKUP_FOLLOW;
1039 		if (flags & FAN_MARK_ONLYDIR)
1040 			lookup_flags |= LOOKUP_DIRECTORY;
1041 
1042 		ret = user_path_at(dfd, filename, lookup_flags, path);
1043 		if (ret)
1044 			goto out;
1045 	}
1046 
1047 	/* you can only watch an inode if you have read permissions on it */
1048 	ret = path_permission(path, MAY_READ);
1049 	if (ret) {
1050 		path_put(path);
1051 		goto out;
1052 	}
1053 
1054 	ret = security_path_notify(path, mask, obj_type);
1055 	if (ret)
1056 		path_put(path);
1057 
1058 out:
1059 	return ret;
1060 }
1061 
1062 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
1063 					    __u32 mask, unsigned int flags,
1064 					    __u32 umask, int *destroy)
1065 {
1066 	__u32 oldmask, newmask;
1067 
1068 	/* umask bits cannot be removed by user */
1069 	mask &= ~umask;
1070 	spin_lock(&fsn_mark->lock);
1071 	oldmask = fsnotify_calc_mask(fsn_mark);
1072 	if (!(flags & FANOTIFY_MARK_IGNORE_BITS)) {
1073 		fsn_mark->mask &= ~mask;
1074 	} else {
1075 		fsn_mark->ignore_mask &= ~mask;
1076 	}
1077 	newmask = fsnotify_calc_mask(fsn_mark);
1078 	/*
1079 	 * We need to keep the mark around even if remaining mask cannot
1080 	 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
1081 	 * changes to the mask.
1082 	 * Destroy mark when only umask bits remain.
1083 	 */
1084 	*destroy = !((fsn_mark->mask | fsn_mark->ignore_mask) & ~umask);
1085 	spin_unlock(&fsn_mark->lock);
1086 
1087 	return oldmask & ~newmask;
1088 }
1089 
1090 static int fanotify_remove_mark(struct fsnotify_group *group,
1091 				fsnotify_connp_t *connp, __u32 mask,
1092 				unsigned int flags, __u32 umask)
1093 {
1094 	struct fsnotify_mark *fsn_mark = NULL;
1095 	__u32 removed;
1096 	int destroy_mark;
1097 
1098 	fsnotify_group_lock(group);
1099 	fsn_mark = fsnotify_find_mark(connp, group);
1100 	if (!fsn_mark) {
1101 		fsnotify_group_unlock(group);
1102 		return -ENOENT;
1103 	}
1104 
1105 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
1106 						 umask, &destroy_mark);
1107 	if (removed & fsnotify_conn_mask(fsn_mark->connector))
1108 		fsnotify_recalc_mask(fsn_mark->connector);
1109 	if (destroy_mark)
1110 		fsnotify_detach_mark(fsn_mark);
1111 	fsnotify_group_unlock(group);
1112 	if (destroy_mark)
1113 		fsnotify_free_mark(fsn_mark);
1114 
1115 	/* matches the fsnotify_find_mark() */
1116 	fsnotify_put_mark(fsn_mark);
1117 	return 0;
1118 }
1119 
1120 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
1121 					 struct vfsmount *mnt, __u32 mask,
1122 					 unsigned int flags, __u32 umask)
1123 {
1124 	return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
1125 				    mask, flags, umask);
1126 }
1127 
1128 static int fanotify_remove_sb_mark(struct fsnotify_group *group,
1129 				   struct super_block *sb, __u32 mask,
1130 				   unsigned int flags, __u32 umask)
1131 {
1132 	return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
1133 				    flags, umask);
1134 }
1135 
1136 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
1137 				      struct inode *inode, __u32 mask,
1138 				      unsigned int flags, __u32 umask)
1139 {
1140 	return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
1141 				    flags, umask);
1142 }
1143 
1144 static bool fanotify_mark_update_flags(struct fsnotify_mark *fsn_mark,
1145 				       unsigned int fan_flags)
1146 {
1147 	bool want_iref = !(fan_flags & FAN_MARK_EVICTABLE);
1148 	unsigned int ignore = fan_flags & FANOTIFY_MARK_IGNORE_BITS;
1149 	bool recalc = false;
1150 
1151 	/*
1152 	 * When using FAN_MARK_IGNORE for the first time, mark starts using
1153 	 * independent event flags in ignore mask.  After that, trying to
1154 	 * update the ignore mask with the old FAN_MARK_IGNORED_MASK API
1155 	 * will result in EEXIST error.
1156 	 */
1157 	if (ignore == FAN_MARK_IGNORE)
1158 		fsn_mark->flags |= FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS;
1159 
1160 	/*
1161 	 * Setting FAN_MARK_IGNORED_SURV_MODIFY for the first time may lead to
1162 	 * the removal of the FS_MODIFY bit in calculated mask if it was set
1163 	 * because of an ignore mask that is now going to survive FS_MODIFY.
1164 	 */
1165 	if (ignore && (fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1166 	    !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) {
1167 		fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
1168 		if (!(fsn_mark->mask & FS_MODIFY))
1169 			recalc = true;
1170 	}
1171 
1172 	if (fsn_mark->connector->type != FSNOTIFY_OBJ_TYPE_INODE ||
1173 	    want_iref == !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1174 		return recalc;
1175 
1176 	/*
1177 	 * NO_IREF may be removed from a mark, but not added.
1178 	 * When removed, fsnotify_recalc_mask() will take the inode ref.
1179 	 */
1180 	WARN_ON_ONCE(!want_iref);
1181 	fsn_mark->flags &= ~FSNOTIFY_MARK_FLAG_NO_IREF;
1182 
1183 	return true;
1184 }
1185 
1186 static bool fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
1187 				      __u32 mask, unsigned int fan_flags)
1188 {
1189 	bool recalc;
1190 
1191 	spin_lock(&fsn_mark->lock);
1192 	if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS))
1193 		fsn_mark->mask |= mask;
1194 	else
1195 		fsn_mark->ignore_mask |= mask;
1196 
1197 	recalc = fsnotify_calc_mask(fsn_mark) &
1198 		~fsnotify_conn_mask(fsn_mark->connector);
1199 
1200 	recalc |= fanotify_mark_update_flags(fsn_mark, fan_flags);
1201 	spin_unlock(&fsn_mark->lock);
1202 
1203 	return recalc;
1204 }
1205 
1206 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
1207 						   fsnotify_connp_t *connp,
1208 						   unsigned int obj_type,
1209 						   unsigned int fan_flags,
1210 						   __kernel_fsid_t *fsid)
1211 {
1212 	struct ucounts *ucounts = group->fanotify_data.ucounts;
1213 	struct fsnotify_mark *mark;
1214 	int ret;
1215 
1216 	/*
1217 	 * Enforce per user marks limits per user in all containing user ns.
1218 	 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
1219 	 * in the limited groups account.
1220 	 */
1221 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
1222 	    !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
1223 		return ERR_PTR(-ENOSPC);
1224 
1225 	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
1226 	if (!mark) {
1227 		ret = -ENOMEM;
1228 		goto out_dec_ucounts;
1229 	}
1230 
1231 	fsnotify_init_mark(mark, group);
1232 	if (fan_flags & FAN_MARK_EVICTABLE)
1233 		mark->flags |= FSNOTIFY_MARK_FLAG_NO_IREF;
1234 
1235 	ret = fsnotify_add_mark_locked(mark, connp, obj_type, 0, fsid);
1236 	if (ret) {
1237 		fsnotify_put_mark(mark);
1238 		goto out_dec_ucounts;
1239 	}
1240 
1241 	return mark;
1242 
1243 out_dec_ucounts:
1244 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1245 		dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
1246 	return ERR_PTR(ret);
1247 }
1248 
1249 static int fanotify_group_init_error_pool(struct fsnotify_group *group)
1250 {
1251 	if (mempool_initialized(&group->fanotify_data.error_events_pool))
1252 		return 0;
1253 
1254 	return mempool_init_kmalloc_pool(&group->fanotify_data.error_events_pool,
1255 					 FANOTIFY_DEFAULT_FEE_POOL_SIZE,
1256 					 sizeof(struct fanotify_error_event));
1257 }
1258 
1259 static int fanotify_may_update_existing_mark(struct fsnotify_mark *fsn_mark,
1260 					      unsigned int fan_flags)
1261 {
1262 	/*
1263 	 * Non evictable mark cannot be downgraded to evictable mark.
1264 	 */
1265 	if (fan_flags & FAN_MARK_EVICTABLE &&
1266 	    !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1267 		return -EEXIST;
1268 
1269 	/*
1270 	 * New ignore mask semantics cannot be downgraded to old semantics.
1271 	 */
1272 	if (fan_flags & FAN_MARK_IGNORED_MASK &&
1273 	    fsn_mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS)
1274 		return -EEXIST;
1275 
1276 	/*
1277 	 * An ignore mask that survives modify could never be downgraded to not
1278 	 * survive modify.  With new FAN_MARK_IGNORE semantics we make that rule
1279 	 * explicit and return an error when trying to update the ignore mask
1280 	 * without the original FAN_MARK_IGNORED_SURV_MODIFY value.
1281 	 */
1282 	if (fan_flags & FAN_MARK_IGNORE &&
1283 	    !(fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1284 	    fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)
1285 		return -EEXIST;
1286 
1287 	return 0;
1288 }
1289 
1290 static int fanotify_add_mark(struct fsnotify_group *group,
1291 			     fsnotify_connp_t *connp, unsigned int obj_type,
1292 			     __u32 mask, unsigned int fan_flags,
1293 			     __kernel_fsid_t *fsid)
1294 {
1295 	struct fsnotify_mark *fsn_mark;
1296 	bool recalc;
1297 	int ret = 0;
1298 
1299 	fsnotify_group_lock(group);
1300 	fsn_mark = fsnotify_find_mark(connp, group);
1301 	if (!fsn_mark) {
1302 		fsn_mark = fanotify_add_new_mark(group, connp, obj_type,
1303 						 fan_flags, fsid);
1304 		if (IS_ERR(fsn_mark)) {
1305 			fsnotify_group_unlock(group);
1306 			return PTR_ERR(fsn_mark);
1307 		}
1308 	}
1309 
1310 	/*
1311 	 * Check if requested mark flags conflict with an existing mark flags.
1312 	 */
1313 	ret = fanotify_may_update_existing_mark(fsn_mark, fan_flags);
1314 	if (ret)
1315 		goto out;
1316 
1317 	/*
1318 	 * Error events are pre-allocated per group, only if strictly
1319 	 * needed (i.e. FAN_FS_ERROR was requested).
1320 	 */
1321 	if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS) &&
1322 	    (mask & FAN_FS_ERROR)) {
1323 		ret = fanotify_group_init_error_pool(group);
1324 		if (ret)
1325 			goto out;
1326 	}
1327 
1328 	recalc = fanotify_mark_add_to_mask(fsn_mark, mask, fan_flags);
1329 	if (recalc)
1330 		fsnotify_recalc_mask(fsn_mark->connector);
1331 
1332 out:
1333 	fsnotify_group_unlock(group);
1334 
1335 	fsnotify_put_mark(fsn_mark);
1336 	return ret;
1337 }
1338 
1339 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
1340 				      struct vfsmount *mnt, __u32 mask,
1341 				      unsigned int flags, __kernel_fsid_t *fsid)
1342 {
1343 	return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
1344 				 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
1345 }
1346 
1347 static int fanotify_add_sb_mark(struct fsnotify_group *group,
1348 				struct super_block *sb, __u32 mask,
1349 				unsigned int flags, __kernel_fsid_t *fsid)
1350 {
1351 	return fanotify_add_mark(group, &sb->s_fsnotify_marks,
1352 				 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
1353 }
1354 
1355 static int fanotify_add_inode_mark(struct fsnotify_group *group,
1356 				   struct inode *inode, __u32 mask,
1357 				   unsigned int flags, __kernel_fsid_t *fsid)
1358 {
1359 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
1360 
1361 	/*
1362 	 * If some other task has this inode open for write we should not add
1363 	 * an ignore mask, unless that ignore mask is supposed to survive
1364 	 * modification changes anyway.
1365 	 */
1366 	if ((flags & FANOTIFY_MARK_IGNORE_BITS) &&
1367 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1368 	    inode_is_open_for_write(inode))
1369 		return 0;
1370 
1371 	return fanotify_add_mark(group, &inode->i_fsnotify_marks,
1372 				 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
1373 }
1374 
1375 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1376 {
1377 	struct fanotify_event *oevent;
1378 
1379 	oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1380 	if (!oevent)
1381 		return NULL;
1382 
1383 	fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1384 	oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1385 
1386 	return &oevent->fse;
1387 }
1388 
1389 static struct hlist_head *fanotify_alloc_merge_hash(void)
1390 {
1391 	struct hlist_head *hash;
1392 
1393 	hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1394 		       GFP_KERNEL_ACCOUNT);
1395 	if (!hash)
1396 		return NULL;
1397 
1398 	__hash_init(hash, FANOTIFY_HTABLE_SIZE);
1399 
1400 	return hash;
1401 }
1402 
1403 /* fanotify syscalls */
1404 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1405 {
1406 	struct fsnotify_group *group;
1407 	int f_flags, fd;
1408 	unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1409 	unsigned int class = flags & FANOTIFY_CLASS_BITS;
1410 	unsigned int internal_flags = 0;
1411 
1412 	pr_debug("%s: flags=%x event_f_flags=%x\n",
1413 		 __func__, flags, event_f_flags);
1414 
1415 	if (!capable(CAP_SYS_ADMIN)) {
1416 		/*
1417 		 * An unprivileged user can setup an fanotify group with
1418 		 * limited functionality - an unprivileged group is limited to
1419 		 * notification events with file handles and it cannot use
1420 		 * unlimited queue/marks.
1421 		 */
1422 		if ((flags & FANOTIFY_ADMIN_INIT_FLAGS) || !fid_mode)
1423 			return -EPERM;
1424 
1425 		/*
1426 		 * Setting the internal flag FANOTIFY_UNPRIV on the group
1427 		 * prevents setting mount/filesystem marks on this group and
1428 		 * prevents reporting pid and open fd in events.
1429 		 */
1430 		internal_flags |= FANOTIFY_UNPRIV;
1431 	}
1432 
1433 #ifdef CONFIG_AUDITSYSCALL
1434 	if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1435 #else
1436 	if (flags & ~FANOTIFY_INIT_FLAGS)
1437 #endif
1438 		return -EINVAL;
1439 
1440 	/*
1441 	 * A pidfd can only be returned for a thread-group leader; thus
1442 	 * FAN_REPORT_PIDFD and FAN_REPORT_TID need to remain mutually
1443 	 * exclusive.
1444 	 */
1445 	if ((flags & FAN_REPORT_PIDFD) && (flags & FAN_REPORT_TID))
1446 		return -EINVAL;
1447 
1448 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1449 		return -EINVAL;
1450 
1451 	switch (event_f_flags & O_ACCMODE) {
1452 	case O_RDONLY:
1453 	case O_RDWR:
1454 	case O_WRONLY:
1455 		break;
1456 	default:
1457 		return -EINVAL;
1458 	}
1459 
1460 	if (fid_mode && class != FAN_CLASS_NOTIF)
1461 		return -EINVAL;
1462 
1463 	/*
1464 	 * Child name is reported with parent fid so requires dir fid.
1465 	 * We can report both child fid and dir fid with or without name.
1466 	 */
1467 	if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1468 		return -EINVAL;
1469 
1470 	/*
1471 	 * FAN_REPORT_TARGET_FID requires FAN_REPORT_NAME and FAN_REPORT_FID
1472 	 * and is used as an indication to report both dir and child fid on all
1473 	 * dirent events.
1474 	 */
1475 	if ((fid_mode & FAN_REPORT_TARGET_FID) &&
1476 	    (!(fid_mode & FAN_REPORT_NAME) || !(fid_mode & FAN_REPORT_FID)))
1477 		return -EINVAL;
1478 
1479 	f_flags = O_RDWR | __FMODE_NONOTIFY;
1480 	if (flags & FAN_CLOEXEC)
1481 		f_flags |= O_CLOEXEC;
1482 	if (flags & FAN_NONBLOCK)
1483 		f_flags |= O_NONBLOCK;
1484 
1485 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
1486 	group = fsnotify_alloc_group(&fanotify_fsnotify_ops,
1487 				     FSNOTIFY_GROUP_USER | FSNOTIFY_GROUP_NOFS);
1488 	if (IS_ERR(group)) {
1489 		return PTR_ERR(group);
1490 	}
1491 
1492 	/* Enforce groups limits per user in all containing user ns */
1493 	group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1494 						  current_euid(),
1495 						  UCOUNT_FANOTIFY_GROUPS);
1496 	if (!group->fanotify_data.ucounts) {
1497 		fd = -EMFILE;
1498 		goto out_destroy_group;
1499 	}
1500 
1501 	group->fanotify_data.flags = flags | internal_flags;
1502 	group->memcg = get_mem_cgroup_from_mm(current->mm);
1503 
1504 	group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1505 	if (!group->fanotify_data.merge_hash) {
1506 		fd = -ENOMEM;
1507 		goto out_destroy_group;
1508 	}
1509 
1510 	group->overflow_event = fanotify_alloc_overflow_event();
1511 	if (unlikely(!group->overflow_event)) {
1512 		fd = -ENOMEM;
1513 		goto out_destroy_group;
1514 	}
1515 
1516 	if (force_o_largefile())
1517 		event_f_flags |= O_LARGEFILE;
1518 	group->fanotify_data.f_flags = event_f_flags;
1519 	init_waitqueue_head(&group->fanotify_data.access_waitq);
1520 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
1521 	switch (class) {
1522 	case FAN_CLASS_NOTIF:
1523 		group->priority = FS_PRIO_0;
1524 		break;
1525 	case FAN_CLASS_CONTENT:
1526 		group->priority = FS_PRIO_1;
1527 		break;
1528 	case FAN_CLASS_PRE_CONTENT:
1529 		group->priority = FS_PRIO_2;
1530 		break;
1531 	default:
1532 		fd = -EINVAL;
1533 		goto out_destroy_group;
1534 	}
1535 
1536 	if (flags & FAN_UNLIMITED_QUEUE) {
1537 		fd = -EPERM;
1538 		if (!capable(CAP_SYS_ADMIN))
1539 			goto out_destroy_group;
1540 		group->max_events = UINT_MAX;
1541 	} else {
1542 		group->max_events = fanotify_max_queued_events;
1543 	}
1544 
1545 	if (flags & FAN_UNLIMITED_MARKS) {
1546 		fd = -EPERM;
1547 		if (!capable(CAP_SYS_ADMIN))
1548 			goto out_destroy_group;
1549 	}
1550 
1551 	if (flags & FAN_ENABLE_AUDIT) {
1552 		fd = -EPERM;
1553 		if (!capable(CAP_AUDIT_WRITE))
1554 			goto out_destroy_group;
1555 	}
1556 
1557 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
1558 	if (fd < 0)
1559 		goto out_destroy_group;
1560 
1561 	return fd;
1562 
1563 out_destroy_group:
1564 	fsnotify_destroy_group(group);
1565 	return fd;
1566 }
1567 
1568 static int fanotify_test_fsid(struct dentry *dentry, __kernel_fsid_t *fsid)
1569 {
1570 	__kernel_fsid_t root_fsid;
1571 	int err;
1572 
1573 	/*
1574 	 * Make sure dentry is not of a filesystem with zero fsid (e.g. fuse).
1575 	 */
1576 	err = vfs_get_fsid(dentry, fsid);
1577 	if (err)
1578 		return err;
1579 
1580 	if (!fsid->val[0] && !fsid->val[1])
1581 		return -ENODEV;
1582 
1583 	/*
1584 	 * Make sure dentry is not of a filesystem subvolume (e.g. btrfs)
1585 	 * which uses a different fsid than sb root.
1586 	 */
1587 	err = vfs_get_fsid(dentry->d_sb->s_root, &root_fsid);
1588 	if (err)
1589 		return err;
1590 
1591 	if (root_fsid.val[0] != fsid->val[0] ||
1592 	    root_fsid.val[1] != fsid->val[1])
1593 		return -EXDEV;
1594 
1595 	return 0;
1596 }
1597 
1598 /* Check if filesystem can encode a unique fid */
1599 static int fanotify_test_fid(struct dentry *dentry, unsigned int flags)
1600 {
1601 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1602 	const struct export_operations *nop = dentry->d_sb->s_export_op;
1603 
1604 	/*
1605 	 * We need to make sure that the filesystem supports encoding of
1606 	 * file handles so user can use name_to_handle_at() to compare fids
1607 	 * reported with events to the file handle of watched objects.
1608 	 */
1609 	if (!nop)
1610 		return -EOPNOTSUPP;
1611 
1612 	/*
1613 	 * For sb/mount mark, we also need to make sure that the filesystem
1614 	 * supports decoding file handles, so user has a way to map back the
1615 	 * reported fids to filesystem objects.
1616 	 */
1617 	if (mark_type != FAN_MARK_INODE && !nop->fh_to_dentry)
1618 		return -EOPNOTSUPP;
1619 
1620 	return 0;
1621 }
1622 
1623 static int fanotify_events_supported(struct fsnotify_group *group,
1624 				     const struct path *path, __u64 mask,
1625 				     unsigned int flags)
1626 {
1627 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1628 	/* Strict validation of events in non-dir inode mask with v5.17+ APIs */
1629 	bool strict_dir_events = FAN_GROUP_FLAG(group, FAN_REPORT_TARGET_FID) ||
1630 				 (mask & FAN_RENAME) ||
1631 				 (flags & FAN_MARK_IGNORE);
1632 
1633 	/*
1634 	 * Some filesystems such as 'proc' acquire unusual locks when opening
1635 	 * files. For them fanotify permission events have high chances of
1636 	 * deadlocking the system - open done when reporting fanotify event
1637 	 * blocks on this "unusual" lock while another process holding the lock
1638 	 * waits for fanotify permission event to be answered. Just disallow
1639 	 * permission events for such filesystems.
1640 	 */
1641 	if (mask & FANOTIFY_PERM_EVENTS &&
1642 	    path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1643 		return -EINVAL;
1644 
1645 	/*
1646 	 * mount and sb marks are not allowed on kernel internal pseudo fs,
1647 	 * like pipe_mnt, because that would subscribe to events on all the
1648 	 * anonynous pipes in the system.
1649 	 *
1650 	 * SB_NOUSER covers all of the internal pseudo fs whose objects are not
1651 	 * exposed to user's mount namespace, but there are other SB_KERNMOUNT
1652 	 * fs, like nsfs, debugfs, for which the value of allowing sb and mount
1653 	 * mark is questionable. For now we leave them alone.
1654 	 */
1655 	if (mark_type != FAN_MARK_INODE &&
1656 	    path->mnt->mnt_sb->s_flags & SB_NOUSER)
1657 		return -EINVAL;
1658 
1659 	/*
1660 	 * We shouldn't have allowed setting dirent events and the directory
1661 	 * flags FAN_ONDIR and FAN_EVENT_ON_CHILD in mask of non-dir inode,
1662 	 * but because we always allowed it, error only when using new APIs.
1663 	 */
1664 	if (strict_dir_events && mark_type == FAN_MARK_INODE &&
1665 	    !d_is_dir(path->dentry) && (mask & FANOTIFY_DIRONLY_EVENT_BITS))
1666 		return -ENOTDIR;
1667 
1668 	return 0;
1669 }
1670 
1671 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1672 			    int dfd, const char  __user *pathname)
1673 {
1674 	struct inode *inode = NULL;
1675 	struct vfsmount *mnt = NULL;
1676 	struct fsnotify_group *group;
1677 	struct fd f;
1678 	struct path path;
1679 	__kernel_fsid_t __fsid, *fsid = NULL;
1680 	u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1681 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1682 	unsigned int mark_cmd = flags & FANOTIFY_MARK_CMD_BITS;
1683 	unsigned int ignore = flags & FANOTIFY_MARK_IGNORE_BITS;
1684 	unsigned int obj_type, fid_mode;
1685 	u32 umask = 0;
1686 	int ret;
1687 
1688 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1689 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
1690 
1691 	/* we only use the lower 32 bits as of right now. */
1692 	if (upper_32_bits(mask))
1693 		return -EINVAL;
1694 
1695 	if (flags & ~FANOTIFY_MARK_FLAGS)
1696 		return -EINVAL;
1697 
1698 	switch (mark_type) {
1699 	case FAN_MARK_INODE:
1700 		obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1701 		break;
1702 	case FAN_MARK_MOUNT:
1703 		obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1704 		break;
1705 	case FAN_MARK_FILESYSTEM:
1706 		obj_type = FSNOTIFY_OBJ_TYPE_SB;
1707 		break;
1708 	default:
1709 		return -EINVAL;
1710 	}
1711 
1712 	switch (mark_cmd) {
1713 	case FAN_MARK_ADD:
1714 	case FAN_MARK_REMOVE:
1715 		if (!mask)
1716 			return -EINVAL;
1717 		break;
1718 	case FAN_MARK_FLUSH:
1719 		if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1720 			return -EINVAL;
1721 		break;
1722 	default:
1723 		return -EINVAL;
1724 	}
1725 
1726 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1727 		valid_mask |= FANOTIFY_PERM_EVENTS;
1728 
1729 	if (mask & ~valid_mask)
1730 		return -EINVAL;
1731 
1732 
1733 	/* We don't allow FAN_MARK_IGNORE & FAN_MARK_IGNORED_MASK together */
1734 	if (ignore == (FAN_MARK_IGNORE | FAN_MARK_IGNORED_MASK))
1735 		return -EINVAL;
1736 
1737 	/*
1738 	 * Event flags (FAN_ONDIR, FAN_EVENT_ON_CHILD) have no effect with
1739 	 * FAN_MARK_IGNORED_MASK.
1740 	 */
1741 	if (ignore == FAN_MARK_IGNORED_MASK) {
1742 		mask &= ~FANOTIFY_EVENT_FLAGS;
1743 		umask = FANOTIFY_EVENT_FLAGS;
1744 	}
1745 
1746 	f = fdget(fanotify_fd);
1747 	if (unlikely(!f.file))
1748 		return -EBADF;
1749 
1750 	/* verify that this is indeed an fanotify instance */
1751 	ret = -EINVAL;
1752 	if (unlikely(f.file->f_op != &fanotify_fops))
1753 		goto fput_and_out;
1754 	group = f.file->private_data;
1755 
1756 	/*
1757 	 * An unprivileged user is not allowed to setup mount nor filesystem
1758 	 * marks.  This also includes setting up such marks by a group that
1759 	 * was initialized by an unprivileged user.
1760 	 */
1761 	ret = -EPERM;
1762 	if ((!capable(CAP_SYS_ADMIN) ||
1763 	     FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV)) &&
1764 	    mark_type != FAN_MARK_INODE)
1765 		goto fput_and_out;
1766 
1767 	/*
1768 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
1769 	 * allowed to set permissions events.
1770 	 */
1771 	ret = -EINVAL;
1772 	if (mask & FANOTIFY_PERM_EVENTS &&
1773 	    group->priority == FS_PRIO_0)
1774 		goto fput_and_out;
1775 
1776 	if (mask & FAN_FS_ERROR &&
1777 	    mark_type != FAN_MARK_FILESYSTEM)
1778 		goto fput_and_out;
1779 
1780 	/*
1781 	 * Evictable is only relevant for inode marks, because only inode object
1782 	 * can be evicted on memory pressure.
1783 	 */
1784 	if (flags & FAN_MARK_EVICTABLE &&
1785 	     mark_type != FAN_MARK_INODE)
1786 		goto fput_and_out;
1787 
1788 	/*
1789 	 * Events that do not carry enough information to report
1790 	 * event->fd require a group that supports reporting fid.  Those
1791 	 * events are not supported on a mount mark, because they do not
1792 	 * carry enough information (i.e. path) to be filtered by mount
1793 	 * point.
1794 	 */
1795 	fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1796 	if (mask & ~(FANOTIFY_FD_EVENTS|FANOTIFY_EVENT_FLAGS) &&
1797 	    (!fid_mode || mark_type == FAN_MARK_MOUNT))
1798 		goto fput_and_out;
1799 
1800 	/*
1801 	 * FAN_RENAME uses special info type records to report the old and
1802 	 * new parent+name.  Reporting only old and new parent id is less
1803 	 * useful and was not implemented.
1804 	 */
1805 	if (mask & FAN_RENAME && !(fid_mode & FAN_REPORT_NAME))
1806 		goto fput_and_out;
1807 
1808 	if (mark_cmd == FAN_MARK_FLUSH) {
1809 		ret = 0;
1810 		if (mark_type == FAN_MARK_MOUNT)
1811 			fsnotify_clear_vfsmount_marks_by_group(group);
1812 		else if (mark_type == FAN_MARK_FILESYSTEM)
1813 			fsnotify_clear_sb_marks_by_group(group);
1814 		else
1815 			fsnotify_clear_inode_marks_by_group(group);
1816 		goto fput_and_out;
1817 	}
1818 
1819 	ret = fanotify_find_path(dfd, pathname, &path, flags,
1820 			(mask & ALL_FSNOTIFY_EVENTS), obj_type);
1821 	if (ret)
1822 		goto fput_and_out;
1823 
1824 	if (mark_cmd == FAN_MARK_ADD) {
1825 		ret = fanotify_events_supported(group, &path, mask, flags);
1826 		if (ret)
1827 			goto path_put_and_out;
1828 	}
1829 
1830 	if (fid_mode) {
1831 		ret = fanotify_test_fsid(path.dentry, &__fsid);
1832 		if (ret)
1833 			goto path_put_and_out;
1834 
1835 		ret = fanotify_test_fid(path.dentry, flags);
1836 		if (ret)
1837 			goto path_put_and_out;
1838 
1839 		fsid = &__fsid;
1840 	}
1841 
1842 	/* inode held in place by reference to path; group by fget on fd */
1843 	if (mark_type == FAN_MARK_INODE)
1844 		inode = path.dentry->d_inode;
1845 	else
1846 		mnt = path.mnt;
1847 
1848 	ret = mnt ? -EINVAL : -EISDIR;
1849 	/* FAN_MARK_IGNORE requires SURV_MODIFY for sb/mount/dir marks */
1850 	if (mark_cmd == FAN_MARK_ADD && ignore == FAN_MARK_IGNORE &&
1851 	    (mnt || S_ISDIR(inode->i_mode)) &&
1852 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY))
1853 		goto path_put_and_out;
1854 
1855 	/* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1856 	if (mnt || !S_ISDIR(inode->i_mode)) {
1857 		mask &= ~FAN_EVENT_ON_CHILD;
1858 		umask = FAN_EVENT_ON_CHILD;
1859 		/*
1860 		 * If group needs to report parent fid, register for getting
1861 		 * events with parent/name info for non-directory.
1862 		 */
1863 		if ((fid_mode & FAN_REPORT_DIR_FID) &&
1864 		    (flags & FAN_MARK_ADD) && !ignore)
1865 			mask |= FAN_EVENT_ON_CHILD;
1866 	}
1867 
1868 	/* create/update an inode mark */
1869 	switch (mark_cmd) {
1870 	case FAN_MARK_ADD:
1871 		if (mark_type == FAN_MARK_MOUNT)
1872 			ret = fanotify_add_vfsmount_mark(group, mnt, mask,
1873 							 flags, fsid);
1874 		else if (mark_type == FAN_MARK_FILESYSTEM)
1875 			ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
1876 						   flags, fsid);
1877 		else
1878 			ret = fanotify_add_inode_mark(group, inode, mask,
1879 						      flags, fsid);
1880 		break;
1881 	case FAN_MARK_REMOVE:
1882 		if (mark_type == FAN_MARK_MOUNT)
1883 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
1884 							    flags, umask);
1885 		else if (mark_type == FAN_MARK_FILESYSTEM)
1886 			ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
1887 						      flags, umask);
1888 		else
1889 			ret = fanotify_remove_inode_mark(group, inode, mask,
1890 							 flags, umask);
1891 		break;
1892 	default:
1893 		ret = -EINVAL;
1894 	}
1895 
1896 path_put_and_out:
1897 	path_put(&path);
1898 fput_and_out:
1899 	fdput(f);
1900 	return ret;
1901 }
1902 
1903 #ifndef CONFIG_ARCH_SPLIT_ARG64
1904 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1905 			      __u64, mask, int, dfd,
1906 			      const char  __user *, pathname)
1907 {
1908 	return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1909 }
1910 #endif
1911 
1912 #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT)
1913 SYSCALL32_DEFINE6(fanotify_mark,
1914 				int, fanotify_fd, unsigned int, flags,
1915 				SC_ARG64(mask), int, dfd,
1916 				const char  __user *, pathname)
1917 {
1918 	return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
1919 				dfd, pathname);
1920 }
1921 #endif
1922 
1923 /*
1924  * fanotify_user_setup - Our initialization function.  Note that we cannot return
1925  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
1926  * must result in panic().
1927  */
1928 static int __init fanotify_user_setup(void)
1929 {
1930 	struct sysinfo si;
1931 	int max_marks;
1932 
1933 	si_meminfo(&si);
1934 	/*
1935 	 * Allow up to 1% of addressable memory to be accounted for per user
1936 	 * marks limited to the range [8192, 1048576]. mount and sb marks are
1937 	 * a lot cheaper than inode marks, but there is no reason for a user
1938 	 * to have many of those, so calculate by the cost of inode marks.
1939 	 */
1940 	max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
1941 		    INODE_MARK_COST;
1942 	max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
1943 				     FANOTIFY_DEFAULT_MAX_USER_MARKS);
1944 
1945 	BUILD_BUG_ON(FANOTIFY_INIT_FLAGS & FANOTIFY_INTERNAL_GROUP_FLAGS);
1946 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 13);
1947 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 11);
1948 
1949 	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark,
1950 					 SLAB_PANIC|SLAB_ACCOUNT);
1951 	fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
1952 					       SLAB_PANIC);
1953 	fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
1954 						SLAB_PANIC);
1955 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
1956 		fanotify_perm_event_cachep =
1957 			KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
1958 	}
1959 
1960 	fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
1961 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
1962 					FANOTIFY_DEFAULT_MAX_GROUPS;
1963 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
1964 	fanotify_sysctls_init();
1965 
1966 	return 0;
1967 }
1968 device_initcall(fanotify_user_setup);
1969