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