xref: /linux/fs/notify/fanotify/fanotify_user.c (revision eed4edda910fe34dfae8c6bfbcf57f4593a54295)
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 "../fsnotify.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 static void __init fanotify_sysctls_init(void)
92 {
93 	register_sysctl("fs/fanotify", fanotify_table);
94 }
95 #else
96 #define fanotify_sysctls_init() do { } while (0)
97 #endif /* CONFIG_SYSCTL */
98 
99 /*
100  * All flags that may be specified in parameter event_f_flags of fanotify_init.
101  *
102  * Internal and external open flags are stored together in field f_flags of
103  * struct file. Only external open flags shall be allowed in event_f_flags.
104  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
105  * excluded.
106  */
107 #define	FANOTIFY_INIT_ALL_EVENT_F_BITS				( \
108 		O_ACCMODE	| O_APPEND	| O_NONBLOCK	| \
109 		__O_SYNC	| O_DSYNC	| O_CLOEXEC     | \
110 		O_LARGEFILE	| O_NOATIME	)
111 
112 extern const struct fsnotify_ops fanotify_fsnotify_ops;
113 
114 struct kmem_cache *fanotify_mark_cache __ro_after_init;
115 struct kmem_cache *fanotify_fid_event_cachep __ro_after_init;
116 struct kmem_cache *fanotify_path_event_cachep __ro_after_init;
117 struct kmem_cache *fanotify_perm_event_cachep __ro_after_init;
118 
119 #define FANOTIFY_EVENT_ALIGN 4
120 #define FANOTIFY_FID_INFO_HDR_LEN \
121 	(sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle))
122 #define FANOTIFY_PIDFD_INFO_HDR_LEN \
123 	sizeof(struct fanotify_event_info_pidfd)
124 #define FANOTIFY_ERROR_INFO_LEN \
125 	(sizeof(struct fanotify_event_info_error))
126 
127 static int fanotify_fid_info_len(int fh_len, int name_len)
128 {
129 	int info_len = fh_len;
130 
131 	if (name_len)
132 		info_len += name_len + 1;
133 
134 	return roundup(FANOTIFY_FID_INFO_HDR_LEN + info_len,
135 		       FANOTIFY_EVENT_ALIGN);
136 }
137 
138 /* FAN_RENAME may have one or two dir+name info records */
139 static int fanotify_dir_name_info_len(struct fanotify_event *event)
140 {
141 	struct fanotify_info *info = fanotify_event_info(event);
142 	int dir_fh_len = fanotify_event_dir_fh_len(event);
143 	int dir2_fh_len = fanotify_event_dir2_fh_len(event);
144 	int info_len = 0;
145 
146 	if (dir_fh_len)
147 		info_len += fanotify_fid_info_len(dir_fh_len,
148 						  info->name_len);
149 	if (dir2_fh_len)
150 		info_len += fanotify_fid_info_len(dir2_fh_len,
151 						  info->name2_len);
152 
153 	return info_len;
154 }
155 
156 static size_t fanotify_event_len(unsigned int info_mode,
157 				 struct fanotify_event *event)
158 {
159 	size_t event_len = FAN_EVENT_METADATA_LEN;
160 	int fh_len;
161 	int dot_len = 0;
162 
163 	if (!info_mode)
164 		return event_len;
165 
166 	if (fanotify_is_error_event(event->mask))
167 		event_len += FANOTIFY_ERROR_INFO_LEN;
168 
169 	if (fanotify_event_has_any_dir_fh(event)) {
170 		event_len += fanotify_dir_name_info_len(event);
171 	} else if ((info_mode & FAN_REPORT_NAME) &&
172 		   (event->mask & FAN_ONDIR)) {
173 		/*
174 		 * With group flag FAN_REPORT_NAME, if name was not recorded in
175 		 * event on a directory, we will report the name ".".
176 		 */
177 		dot_len = 1;
178 	}
179 
180 	if (info_mode & FAN_REPORT_PIDFD)
181 		event_len += FANOTIFY_PIDFD_INFO_HDR_LEN;
182 
183 	if (fanotify_event_has_object_fh(event)) {
184 		fh_len = fanotify_event_object_fh_len(event);
185 		event_len += fanotify_fid_info_len(fh_len, dot_len);
186 	}
187 
188 	return event_len;
189 }
190 
191 /*
192  * Remove an hashed event from merge hash table.
193  */
194 static void fanotify_unhash_event(struct fsnotify_group *group,
195 				  struct fanotify_event *event)
196 {
197 	assert_spin_locked(&group->notification_lock);
198 
199 	pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
200 		 group, event, fanotify_event_hash_bucket(group, event));
201 
202 	if (WARN_ON_ONCE(hlist_unhashed(&event->merge_list)))
203 		return;
204 
205 	hlist_del_init(&event->merge_list);
206 }
207 
208 /*
209  * Get an fanotify notification event if one exists and is small
210  * enough to fit in "count". Return an error pointer if the count
211  * is not large enough. When permission event is dequeued, its state is
212  * updated accordingly.
213  */
214 static struct fanotify_event *get_one_event(struct fsnotify_group *group,
215 					    size_t count)
216 {
217 	size_t event_size;
218 	struct fanotify_event *event = NULL;
219 	struct fsnotify_event *fsn_event;
220 	unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
221 
222 	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
223 
224 	spin_lock(&group->notification_lock);
225 	fsn_event = fsnotify_peek_first_event(group);
226 	if (!fsn_event)
227 		goto out;
228 
229 	event = FANOTIFY_E(fsn_event);
230 	event_size = fanotify_event_len(info_mode, event);
231 
232 	if (event_size > count) {
233 		event = ERR_PTR(-EINVAL);
234 		goto out;
235 	}
236 
237 	/*
238 	 * Held the notification_lock the whole time, so this is the
239 	 * same event we peeked above.
240 	 */
241 	fsnotify_remove_first_event(group);
242 	if (fanotify_is_perm_event(event->mask))
243 		FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED;
244 	if (fanotify_is_hashed_event(event->mask))
245 		fanotify_unhash_event(group, event);
246 out:
247 	spin_unlock(&group->notification_lock);
248 	return event;
249 }
250 
251 static int create_fd(struct fsnotify_group *group, const struct path *path,
252 		     struct file **file)
253 {
254 	int client_fd;
255 	struct file *new_file;
256 
257 	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
258 	if (client_fd < 0)
259 		return client_fd;
260 
261 	/*
262 	 * we need a new file handle for the userspace program so it can read even if it was
263 	 * originally opened O_WRONLY.
264 	 */
265 	new_file = dentry_open(path,
266 			       group->fanotify_data.f_flags | __FMODE_NONOTIFY,
267 			       current_cred());
268 	if (IS_ERR(new_file)) {
269 		/*
270 		 * we still send an event even if we can't open the file.  this
271 		 * can happen when say tasks are gone and we try to open their
272 		 * /proc files or we try to open a WRONLY file like in sysfs
273 		 * we just send the errno to userspace since there isn't much
274 		 * else we can do.
275 		 */
276 		put_unused_fd(client_fd);
277 		client_fd = PTR_ERR(new_file);
278 	} else {
279 		*file = new_file;
280 	}
281 
282 	return client_fd;
283 }
284 
285 static int process_access_response_info(const char __user *info,
286 					size_t info_len,
287 				struct fanotify_response_info_audit_rule *friar)
288 {
289 	if (info_len != sizeof(*friar))
290 		return -EINVAL;
291 
292 	if (copy_from_user(friar, info, sizeof(*friar)))
293 		return -EFAULT;
294 
295 	if (friar->hdr.type != FAN_RESPONSE_INFO_AUDIT_RULE)
296 		return -EINVAL;
297 	if (friar->hdr.pad != 0)
298 		return -EINVAL;
299 	if (friar->hdr.len != sizeof(*friar))
300 		return -EINVAL;
301 
302 	return info_len;
303 }
304 
305 /*
306  * Finish processing of permission event by setting it to ANSWERED state and
307  * drop group->notification_lock.
308  */
309 static void finish_permission_event(struct fsnotify_group *group,
310 				    struct fanotify_perm_event *event, u32 response,
311 				    struct fanotify_response_info_audit_rule *friar)
312 				    __releases(&group->notification_lock)
313 {
314 	bool destroy = false;
315 
316 	assert_spin_locked(&group->notification_lock);
317 	event->response = response & ~FAN_INFO;
318 	if (response & FAN_INFO)
319 		memcpy(&event->audit_rule, friar, sizeof(*friar));
320 
321 	if (event->state == FAN_EVENT_CANCELED)
322 		destroy = true;
323 	else
324 		event->state = FAN_EVENT_ANSWERED;
325 	spin_unlock(&group->notification_lock);
326 	if (destroy)
327 		fsnotify_destroy_event(group, &event->fae.fse);
328 }
329 
330 static int process_access_response(struct fsnotify_group *group,
331 				   struct fanotify_response *response_struct,
332 				   const char __user *info,
333 				   size_t info_len)
334 {
335 	struct fanotify_perm_event *event;
336 	int fd = response_struct->fd;
337 	u32 response = response_struct->response;
338 	int ret = info_len;
339 	struct fanotify_response_info_audit_rule friar;
340 
341 	pr_debug("%s: group=%p fd=%d response=%u buf=%p size=%zu\n", __func__,
342 		 group, fd, response, info, info_len);
343 	/*
344 	 * make sure the response is valid, if invalid we do nothing and either
345 	 * userspace can send a valid response or we will clean it up after the
346 	 * timeout
347 	 */
348 	if (response & ~FANOTIFY_RESPONSE_VALID_MASK)
349 		return -EINVAL;
350 
351 	switch (response & FANOTIFY_RESPONSE_ACCESS) {
352 	case FAN_ALLOW:
353 	case FAN_DENY:
354 		break;
355 	default:
356 		return -EINVAL;
357 	}
358 
359 	if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
360 		return -EINVAL;
361 
362 	if (response & FAN_INFO) {
363 		ret = process_access_response_info(info, info_len, &friar);
364 		if (ret < 0)
365 			return ret;
366 		if (fd == FAN_NOFD)
367 			return ret;
368 	} else {
369 		ret = 0;
370 	}
371 
372 	if (fd < 0)
373 		return -EINVAL;
374 
375 	spin_lock(&group->notification_lock);
376 	list_for_each_entry(event, &group->fanotify_data.access_list,
377 			    fae.fse.list) {
378 		if (event->fd != fd)
379 			continue;
380 
381 		list_del_init(&event->fae.fse.list);
382 		finish_permission_event(group, event, response, &friar);
383 		wake_up(&group->fanotify_data.access_waitq);
384 		return ret;
385 	}
386 	spin_unlock(&group->notification_lock);
387 
388 	return -ENOENT;
389 }
390 
391 static size_t copy_error_info_to_user(struct fanotify_event *event,
392 				      char __user *buf, int count)
393 {
394 	struct fanotify_event_info_error info = { };
395 	struct fanotify_error_event *fee = FANOTIFY_EE(event);
396 
397 	info.hdr.info_type = FAN_EVENT_INFO_TYPE_ERROR;
398 	info.hdr.len = FANOTIFY_ERROR_INFO_LEN;
399 
400 	if (WARN_ON(count < info.hdr.len))
401 		return -EFAULT;
402 
403 	info.error = fee->error;
404 	info.error_count = fee->err_count;
405 
406 	if (copy_to_user(buf, &info, sizeof(info)))
407 		return -EFAULT;
408 
409 	return info.hdr.len;
410 }
411 
412 static int copy_fid_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
413 				 int info_type, const char *name,
414 				 size_t name_len,
415 				 char __user *buf, size_t count)
416 {
417 	struct fanotify_event_info_fid info = { };
418 	struct file_handle handle = { };
419 	unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
420 	size_t fh_len = fh ? fh->len : 0;
421 	size_t info_len = fanotify_fid_info_len(fh_len, name_len);
422 	size_t len = info_len;
423 
424 	pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
425 		 __func__, fh_len, name_len, info_len, count);
426 
427 	if (WARN_ON_ONCE(len < sizeof(info) || len > count))
428 		return -EFAULT;
429 
430 	/*
431 	 * Copy event info fid header followed by variable sized file handle
432 	 * and optionally followed by variable sized filename.
433 	 */
434 	switch (info_type) {
435 	case FAN_EVENT_INFO_TYPE_FID:
436 	case FAN_EVENT_INFO_TYPE_DFID:
437 		if (WARN_ON_ONCE(name_len))
438 			return -EFAULT;
439 		break;
440 	case FAN_EVENT_INFO_TYPE_DFID_NAME:
441 	case FAN_EVENT_INFO_TYPE_OLD_DFID_NAME:
442 	case FAN_EVENT_INFO_TYPE_NEW_DFID_NAME:
443 		if (WARN_ON_ONCE(!name || !name_len))
444 			return -EFAULT;
445 		break;
446 	default:
447 		return -EFAULT;
448 	}
449 
450 	info.hdr.info_type = info_type;
451 	info.hdr.len = len;
452 	info.fsid = *fsid;
453 	if (copy_to_user(buf, &info, sizeof(info)))
454 		return -EFAULT;
455 
456 	buf += sizeof(info);
457 	len -= sizeof(info);
458 	if (WARN_ON_ONCE(len < sizeof(handle)))
459 		return -EFAULT;
460 
461 	handle.handle_type = fh->type;
462 	handle.handle_bytes = fh_len;
463 
464 	/* Mangle handle_type for bad file_handle */
465 	if (!fh_len)
466 		handle.handle_type = FILEID_INVALID;
467 
468 	if (copy_to_user(buf, &handle, sizeof(handle)))
469 		return -EFAULT;
470 
471 	buf += sizeof(handle);
472 	len -= sizeof(handle);
473 	if (WARN_ON_ONCE(len < fh_len))
474 		return -EFAULT;
475 
476 	/*
477 	 * For an inline fh and inline file name, copy through stack to exclude
478 	 * the copy from usercopy hardening protections.
479 	 */
480 	fh_buf = fanotify_fh_buf(fh);
481 	if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
482 		memcpy(bounce, fh_buf, fh_len);
483 		fh_buf = bounce;
484 	}
485 	if (copy_to_user(buf, fh_buf, fh_len))
486 		return -EFAULT;
487 
488 	buf += fh_len;
489 	len -= fh_len;
490 
491 	if (name_len) {
492 		/* Copy the filename with terminating null */
493 		name_len++;
494 		if (WARN_ON_ONCE(len < name_len))
495 			return -EFAULT;
496 
497 		if (copy_to_user(buf, name, name_len))
498 			return -EFAULT;
499 
500 		buf += name_len;
501 		len -= name_len;
502 	}
503 
504 	/* Pad with 0's */
505 	WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
506 	if (len > 0 && clear_user(buf, len))
507 		return -EFAULT;
508 
509 	return info_len;
510 }
511 
512 static int copy_pidfd_info_to_user(int pidfd,
513 				   char __user *buf,
514 				   size_t count)
515 {
516 	struct fanotify_event_info_pidfd info = { };
517 	size_t info_len = FANOTIFY_PIDFD_INFO_HDR_LEN;
518 
519 	if (WARN_ON_ONCE(info_len > count))
520 		return -EFAULT;
521 
522 	info.hdr.info_type = FAN_EVENT_INFO_TYPE_PIDFD;
523 	info.hdr.len = info_len;
524 	info.pidfd = pidfd;
525 
526 	if (copy_to_user(buf, &info, info_len))
527 		return -EFAULT;
528 
529 	return info_len;
530 }
531 
532 static int copy_info_records_to_user(struct fanotify_event *event,
533 				     struct fanotify_info *info,
534 				     unsigned int info_mode, int pidfd,
535 				     char __user *buf, size_t count)
536 {
537 	int ret, total_bytes = 0, info_type = 0;
538 	unsigned int fid_mode = info_mode & FANOTIFY_FID_BITS;
539 	unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
540 
541 	/*
542 	 * Event info records order is as follows:
543 	 * 1. dir fid + name
544 	 * 2. (optional) new dir fid + new name
545 	 * 3. (optional) child fid
546 	 */
547 	if (fanotify_event_has_dir_fh(event)) {
548 		info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
549 					     FAN_EVENT_INFO_TYPE_DFID;
550 
551 		/* FAN_RENAME uses special info types */
552 		if (event->mask & FAN_RENAME)
553 			info_type = FAN_EVENT_INFO_TYPE_OLD_DFID_NAME;
554 
555 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
556 					    fanotify_info_dir_fh(info),
557 					    info_type,
558 					    fanotify_info_name(info),
559 					    info->name_len, buf, count);
560 		if (ret < 0)
561 			return ret;
562 
563 		buf += ret;
564 		count -= ret;
565 		total_bytes += ret;
566 	}
567 
568 	/* New dir fid+name may be reported in addition to old dir fid+name */
569 	if (fanotify_event_has_dir2_fh(event)) {
570 		info_type = FAN_EVENT_INFO_TYPE_NEW_DFID_NAME;
571 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
572 					    fanotify_info_dir2_fh(info),
573 					    info_type,
574 					    fanotify_info_name2(info),
575 					    info->name2_len, buf, count);
576 		if (ret < 0)
577 			return ret;
578 
579 		buf += ret;
580 		count -= ret;
581 		total_bytes += ret;
582 	}
583 
584 	if (fanotify_event_has_object_fh(event)) {
585 		const char *dot = NULL;
586 		int dot_len = 0;
587 
588 		if (fid_mode == FAN_REPORT_FID || info_type) {
589 			/*
590 			 * With only group flag FAN_REPORT_FID only type FID is
591 			 * reported. Second info record type is always FID.
592 			 */
593 			info_type = FAN_EVENT_INFO_TYPE_FID;
594 		} else if ((fid_mode & FAN_REPORT_NAME) &&
595 			   (event->mask & FAN_ONDIR)) {
596 			/*
597 			 * With group flag FAN_REPORT_NAME, if name was not
598 			 * recorded in an event on a directory, report the name
599 			 * "." with info type DFID_NAME.
600 			 */
601 			info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
602 			dot = ".";
603 			dot_len = 1;
604 		} else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
605 			   (event->mask & FAN_ONDIR)) {
606 			/*
607 			 * With group flag FAN_REPORT_DIR_FID, a single info
608 			 * record has type DFID for directory entry modification
609 			 * event and for event on a directory.
610 			 */
611 			info_type = FAN_EVENT_INFO_TYPE_DFID;
612 		} else {
613 			/*
614 			 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
615 			 * a single info record has type FID for event on a
616 			 * non-directory, when there is no directory to report.
617 			 * For example, on FAN_DELETE_SELF event.
618 			 */
619 			info_type = FAN_EVENT_INFO_TYPE_FID;
620 		}
621 
622 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
623 					    fanotify_event_object_fh(event),
624 					    info_type, dot, dot_len,
625 					    buf, count);
626 		if (ret < 0)
627 			return ret;
628 
629 		buf += ret;
630 		count -= ret;
631 		total_bytes += ret;
632 	}
633 
634 	if (pidfd_mode) {
635 		ret = copy_pidfd_info_to_user(pidfd, buf, count);
636 		if (ret < 0)
637 			return ret;
638 
639 		buf += ret;
640 		count -= ret;
641 		total_bytes += ret;
642 	}
643 
644 	if (fanotify_is_error_event(event->mask)) {
645 		ret = copy_error_info_to_user(event, buf, count);
646 		if (ret < 0)
647 			return ret;
648 		buf += ret;
649 		count -= ret;
650 		total_bytes += ret;
651 	}
652 
653 	return total_bytes;
654 }
655 
656 static ssize_t copy_event_to_user(struct fsnotify_group *group,
657 				  struct fanotify_event *event,
658 				  char __user *buf, size_t count)
659 {
660 	struct fanotify_event_metadata metadata;
661 	const struct path *path = fanotify_event_path(event);
662 	struct fanotify_info *info = fanotify_event_info(event);
663 	unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
664 	unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
665 	struct file *f = NULL, *pidfd_file = NULL;
666 	int ret, pidfd = FAN_NOPIDFD, fd = FAN_NOFD;
667 
668 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
669 
670 	metadata.event_len = fanotify_event_len(info_mode, event);
671 	metadata.metadata_len = FAN_EVENT_METADATA_LEN;
672 	metadata.vers = FANOTIFY_METADATA_VERSION;
673 	metadata.reserved = 0;
674 	metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
675 	metadata.pid = pid_vnr(event->pid);
676 	/*
677 	 * For an unprivileged listener, event->pid can be used to identify the
678 	 * events generated by the listener process itself, without disclosing
679 	 * the pids of other processes.
680 	 */
681 	if (FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
682 	    task_tgid(current) != event->pid)
683 		metadata.pid = 0;
684 
685 	/*
686 	 * For now, fid mode is required for an unprivileged listener and
687 	 * fid mode does not report fd in events.  Keep this check anyway
688 	 * for safety in case fid mode requirement is relaxed in the future
689 	 * to allow unprivileged listener to get events with no fd and no fid.
690 	 */
691 	if (!FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
692 	    path && path->mnt && path->dentry) {
693 		fd = create_fd(group, path, &f);
694 		if (fd < 0)
695 			return fd;
696 	}
697 	metadata.fd = fd;
698 
699 	if (pidfd_mode) {
700 		/*
701 		 * Complain if the FAN_REPORT_PIDFD and FAN_REPORT_TID mutual
702 		 * exclusion is ever lifted. At the time of incoporating pidfd
703 		 * support within fanotify, the pidfd API only supported the
704 		 * creation of pidfds for thread-group leaders.
705 		 */
706 		WARN_ON_ONCE(FAN_GROUP_FLAG(group, FAN_REPORT_TID));
707 
708 		/*
709 		 * The PIDTYPE_TGID check for an event->pid is performed
710 		 * preemptively in an attempt to catch out cases where the event
711 		 * listener reads events after the event generating process has
712 		 * already terminated. Report FAN_NOPIDFD to the event listener
713 		 * in those cases, with all other pidfd creation errors being
714 		 * reported as FAN_EPIDFD.
715 		 */
716 		if (metadata.pid == 0 ||
717 		    !pid_has_task(event->pid, PIDTYPE_TGID)) {
718 			pidfd = FAN_NOPIDFD;
719 		} else {
720 			pidfd = pidfd_prepare(event->pid, 0, &pidfd_file);
721 			if (pidfd < 0)
722 				pidfd = FAN_EPIDFD;
723 		}
724 	}
725 
726 	ret = -EFAULT;
727 	/*
728 	 * Sanity check copy size in case get_one_event() and
729 	 * event_len sizes ever get out of sync.
730 	 */
731 	if (WARN_ON_ONCE(metadata.event_len > count))
732 		goto out_close_fd;
733 
734 	if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
735 		goto out_close_fd;
736 
737 	buf += FAN_EVENT_METADATA_LEN;
738 	count -= FAN_EVENT_METADATA_LEN;
739 
740 	if (fanotify_is_perm_event(event->mask))
741 		FANOTIFY_PERM(event)->fd = fd;
742 
743 	if (info_mode) {
744 		ret = copy_info_records_to_user(event, info, info_mode, pidfd,
745 						buf, count);
746 		if (ret < 0)
747 			goto out_close_fd;
748 	}
749 
750 	if (f)
751 		fd_install(fd, f);
752 
753 	if (pidfd_file)
754 		fd_install(pidfd, pidfd_file);
755 
756 	return metadata.event_len;
757 
758 out_close_fd:
759 	if (fd != FAN_NOFD) {
760 		put_unused_fd(fd);
761 		fput(f);
762 	}
763 
764 	if (pidfd >= 0) {
765 		put_unused_fd(pidfd);
766 		fput(pidfd_file);
767 	}
768 
769 	return ret;
770 }
771 
772 /* intofiy userspace file descriptor functions */
773 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
774 {
775 	struct fsnotify_group *group = file->private_data;
776 	__poll_t ret = 0;
777 
778 	poll_wait(file, &group->notification_waitq, wait);
779 	spin_lock(&group->notification_lock);
780 	if (!fsnotify_notify_queue_is_empty(group))
781 		ret = EPOLLIN | EPOLLRDNORM;
782 	spin_unlock(&group->notification_lock);
783 
784 	return ret;
785 }
786 
787 static ssize_t fanotify_read(struct file *file, char __user *buf,
788 			     size_t count, loff_t *pos)
789 {
790 	struct fsnotify_group *group;
791 	struct fanotify_event *event;
792 	char __user *start;
793 	int ret;
794 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
795 
796 	start = buf;
797 	group = file->private_data;
798 
799 	pr_debug("%s: group=%p\n", __func__, group);
800 
801 	add_wait_queue(&group->notification_waitq, &wait);
802 	while (1) {
803 		/*
804 		 * User can supply arbitrarily large buffer. Avoid softlockups
805 		 * in case there are lots of available events.
806 		 */
807 		cond_resched();
808 		event = get_one_event(group, count);
809 		if (IS_ERR(event)) {
810 			ret = PTR_ERR(event);
811 			break;
812 		}
813 
814 		if (!event) {
815 			ret = -EAGAIN;
816 			if (file->f_flags & O_NONBLOCK)
817 				break;
818 
819 			ret = -ERESTARTSYS;
820 			if (signal_pending(current))
821 				break;
822 
823 			if (start != buf)
824 				break;
825 
826 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
827 			continue;
828 		}
829 
830 		ret = copy_event_to_user(group, event, buf, count);
831 		if (unlikely(ret == -EOPENSTALE)) {
832 			/*
833 			 * We cannot report events with stale fd so drop it.
834 			 * Setting ret to 0 will continue the event loop and
835 			 * do the right thing if there are no more events to
836 			 * read (i.e. return bytes read, -EAGAIN or wait).
837 			 */
838 			ret = 0;
839 		}
840 
841 		/*
842 		 * Permission events get queued to wait for response.  Other
843 		 * events can be destroyed now.
844 		 */
845 		if (!fanotify_is_perm_event(event->mask)) {
846 			fsnotify_destroy_event(group, &event->fse);
847 		} else {
848 			if (ret <= 0) {
849 				spin_lock(&group->notification_lock);
850 				finish_permission_event(group,
851 					FANOTIFY_PERM(event), FAN_DENY, NULL);
852 				wake_up(&group->fanotify_data.access_waitq);
853 			} else {
854 				spin_lock(&group->notification_lock);
855 				list_add_tail(&event->fse.list,
856 					&group->fanotify_data.access_list);
857 				spin_unlock(&group->notification_lock);
858 			}
859 		}
860 		if (ret < 0)
861 			break;
862 		buf += ret;
863 		count -= ret;
864 	}
865 	remove_wait_queue(&group->notification_waitq, &wait);
866 
867 	if (start != buf && ret != -EFAULT)
868 		ret = buf - start;
869 	return ret;
870 }
871 
872 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
873 {
874 	struct fanotify_response response;
875 	struct fsnotify_group *group;
876 	int ret;
877 	const char __user *info_buf = buf + sizeof(struct fanotify_response);
878 	size_t info_len;
879 
880 	if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
881 		return -EINVAL;
882 
883 	group = file->private_data;
884 
885 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
886 
887 	if (count < sizeof(response))
888 		return -EINVAL;
889 
890 	if (copy_from_user(&response, buf, sizeof(response)))
891 		return -EFAULT;
892 
893 	info_len = count - sizeof(response);
894 
895 	ret = process_access_response(group, &response, info_buf, info_len);
896 	if (ret < 0)
897 		count = ret;
898 	else
899 		count = sizeof(response) + ret;
900 
901 	return count;
902 }
903 
904 static int fanotify_release(struct inode *ignored, struct file *file)
905 {
906 	struct fsnotify_group *group = file->private_data;
907 	struct fsnotify_event *fsn_event;
908 
909 	/*
910 	 * Stop new events from arriving in the notification queue. since
911 	 * userspace cannot use fanotify fd anymore, no event can enter or
912 	 * leave access_list by now either.
913 	 */
914 	fsnotify_group_stop_queueing(group);
915 
916 	/*
917 	 * Process all permission events on access_list and notification queue
918 	 * and simulate reply from userspace.
919 	 */
920 	spin_lock(&group->notification_lock);
921 	while (!list_empty(&group->fanotify_data.access_list)) {
922 		struct fanotify_perm_event *event;
923 
924 		event = list_first_entry(&group->fanotify_data.access_list,
925 				struct fanotify_perm_event, fae.fse.list);
926 		list_del_init(&event->fae.fse.list);
927 		finish_permission_event(group, event, FAN_ALLOW, NULL);
928 		spin_lock(&group->notification_lock);
929 	}
930 
931 	/*
932 	 * Destroy all non-permission events. For permission events just
933 	 * dequeue them and set the response. They will be freed once the
934 	 * response is consumed and fanotify_get_response() returns.
935 	 */
936 	while ((fsn_event = fsnotify_remove_first_event(group))) {
937 		struct fanotify_event *event = FANOTIFY_E(fsn_event);
938 
939 		if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
940 			spin_unlock(&group->notification_lock);
941 			fsnotify_destroy_event(group, fsn_event);
942 		} else {
943 			finish_permission_event(group, FANOTIFY_PERM(event),
944 						FAN_ALLOW, NULL);
945 		}
946 		spin_lock(&group->notification_lock);
947 	}
948 	spin_unlock(&group->notification_lock);
949 
950 	/* Response for all permission events it set, wakeup waiters */
951 	wake_up(&group->fanotify_data.access_waitq);
952 
953 	/* matches the fanotify_init->fsnotify_alloc_group */
954 	fsnotify_destroy_group(group);
955 
956 	return 0;
957 }
958 
959 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
960 {
961 	struct fsnotify_group *group;
962 	struct fsnotify_event *fsn_event;
963 	void __user *p;
964 	int ret = -ENOTTY;
965 	size_t send_len = 0;
966 
967 	group = file->private_data;
968 
969 	p = (void __user *) arg;
970 
971 	switch (cmd) {
972 	case FIONREAD:
973 		spin_lock(&group->notification_lock);
974 		list_for_each_entry(fsn_event, &group->notification_list, list)
975 			send_len += FAN_EVENT_METADATA_LEN;
976 		spin_unlock(&group->notification_lock);
977 		ret = put_user(send_len, (int __user *) p);
978 		break;
979 	}
980 
981 	return ret;
982 }
983 
984 static const struct file_operations fanotify_fops = {
985 	.show_fdinfo	= fanotify_show_fdinfo,
986 	.poll		= fanotify_poll,
987 	.read		= fanotify_read,
988 	.write		= fanotify_write,
989 	.fasync		= NULL,
990 	.release	= fanotify_release,
991 	.unlocked_ioctl	= fanotify_ioctl,
992 	.compat_ioctl	= compat_ptr_ioctl,
993 	.llseek		= noop_llseek,
994 };
995 
996 static int fanotify_find_path(int dfd, const char __user *filename,
997 			      struct path *path, unsigned int flags, __u64 mask,
998 			      unsigned int obj_type)
999 {
1000 	int ret;
1001 
1002 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
1003 		 dfd, filename, flags);
1004 
1005 	if (filename == NULL) {
1006 		struct fd f = fdget(dfd);
1007 
1008 		ret = -EBADF;
1009 		if (!f.file)
1010 			goto out;
1011 
1012 		ret = -ENOTDIR;
1013 		if ((flags & FAN_MARK_ONLYDIR) &&
1014 		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
1015 			fdput(f);
1016 			goto out;
1017 		}
1018 
1019 		*path = f.file->f_path;
1020 		path_get(path);
1021 		fdput(f);
1022 	} else {
1023 		unsigned int lookup_flags = 0;
1024 
1025 		if (!(flags & FAN_MARK_DONT_FOLLOW))
1026 			lookup_flags |= LOOKUP_FOLLOW;
1027 		if (flags & FAN_MARK_ONLYDIR)
1028 			lookup_flags |= LOOKUP_DIRECTORY;
1029 
1030 		ret = user_path_at(dfd, filename, lookup_flags, path);
1031 		if (ret)
1032 			goto out;
1033 	}
1034 
1035 	/* you can only watch an inode if you have read permissions on it */
1036 	ret = path_permission(path, MAY_READ);
1037 	if (ret) {
1038 		path_put(path);
1039 		goto out;
1040 	}
1041 
1042 	ret = security_path_notify(path, mask, obj_type);
1043 	if (ret)
1044 		path_put(path);
1045 
1046 out:
1047 	return ret;
1048 }
1049 
1050 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
1051 					    __u32 mask, unsigned int flags,
1052 					    __u32 umask, int *destroy)
1053 {
1054 	__u32 oldmask, newmask;
1055 
1056 	/* umask bits cannot be removed by user */
1057 	mask &= ~umask;
1058 	spin_lock(&fsn_mark->lock);
1059 	oldmask = fsnotify_calc_mask(fsn_mark);
1060 	if (!(flags & FANOTIFY_MARK_IGNORE_BITS)) {
1061 		fsn_mark->mask &= ~mask;
1062 	} else {
1063 		fsn_mark->ignore_mask &= ~mask;
1064 	}
1065 	newmask = fsnotify_calc_mask(fsn_mark);
1066 	/*
1067 	 * We need to keep the mark around even if remaining mask cannot
1068 	 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
1069 	 * changes to the mask.
1070 	 * Destroy mark when only umask bits remain.
1071 	 */
1072 	*destroy = !((fsn_mark->mask | fsn_mark->ignore_mask) & ~umask);
1073 	spin_unlock(&fsn_mark->lock);
1074 
1075 	return oldmask & ~newmask;
1076 }
1077 
1078 static int fanotify_remove_mark(struct fsnotify_group *group,
1079 				fsnotify_connp_t *connp, __u32 mask,
1080 				unsigned int flags, __u32 umask)
1081 {
1082 	struct fsnotify_mark *fsn_mark = NULL;
1083 	__u32 removed;
1084 	int destroy_mark;
1085 
1086 	fsnotify_group_lock(group);
1087 	fsn_mark = fsnotify_find_mark(connp, group);
1088 	if (!fsn_mark) {
1089 		fsnotify_group_unlock(group);
1090 		return -ENOENT;
1091 	}
1092 
1093 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
1094 						 umask, &destroy_mark);
1095 	if (removed & fsnotify_conn_mask(fsn_mark->connector))
1096 		fsnotify_recalc_mask(fsn_mark->connector);
1097 	if (destroy_mark)
1098 		fsnotify_detach_mark(fsn_mark);
1099 	fsnotify_group_unlock(group);
1100 	if (destroy_mark)
1101 		fsnotify_free_mark(fsn_mark);
1102 
1103 	/* matches the fsnotify_find_mark() */
1104 	fsnotify_put_mark(fsn_mark);
1105 	return 0;
1106 }
1107 
1108 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
1109 					 struct vfsmount *mnt, __u32 mask,
1110 					 unsigned int flags, __u32 umask)
1111 {
1112 	return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
1113 				    mask, flags, umask);
1114 }
1115 
1116 static int fanotify_remove_sb_mark(struct fsnotify_group *group,
1117 				   struct super_block *sb, __u32 mask,
1118 				   unsigned int flags, __u32 umask)
1119 {
1120 	return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
1121 				    flags, umask);
1122 }
1123 
1124 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
1125 				      struct inode *inode, __u32 mask,
1126 				      unsigned int flags, __u32 umask)
1127 {
1128 	return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
1129 				    flags, umask);
1130 }
1131 
1132 static bool fanotify_mark_update_flags(struct fsnotify_mark *fsn_mark,
1133 				       unsigned int fan_flags)
1134 {
1135 	bool want_iref = !(fan_flags & FAN_MARK_EVICTABLE);
1136 	unsigned int ignore = fan_flags & FANOTIFY_MARK_IGNORE_BITS;
1137 	bool recalc = false;
1138 
1139 	/*
1140 	 * When using FAN_MARK_IGNORE for the first time, mark starts using
1141 	 * independent event flags in ignore mask.  After that, trying to
1142 	 * update the ignore mask with the old FAN_MARK_IGNORED_MASK API
1143 	 * will result in EEXIST error.
1144 	 */
1145 	if (ignore == FAN_MARK_IGNORE)
1146 		fsn_mark->flags |= FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS;
1147 
1148 	/*
1149 	 * Setting FAN_MARK_IGNORED_SURV_MODIFY for the first time may lead to
1150 	 * the removal of the FS_MODIFY bit in calculated mask if it was set
1151 	 * because of an ignore mask that is now going to survive FS_MODIFY.
1152 	 */
1153 	if (ignore && (fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1154 	    !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) {
1155 		fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
1156 		if (!(fsn_mark->mask & FS_MODIFY))
1157 			recalc = true;
1158 	}
1159 
1160 	if (fsn_mark->connector->type != FSNOTIFY_OBJ_TYPE_INODE ||
1161 	    want_iref == !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1162 		return recalc;
1163 
1164 	/*
1165 	 * NO_IREF may be removed from a mark, but not added.
1166 	 * When removed, fsnotify_recalc_mask() will take the inode ref.
1167 	 */
1168 	WARN_ON_ONCE(!want_iref);
1169 	fsn_mark->flags &= ~FSNOTIFY_MARK_FLAG_NO_IREF;
1170 
1171 	return true;
1172 }
1173 
1174 static bool fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
1175 				      __u32 mask, unsigned int fan_flags)
1176 {
1177 	bool recalc;
1178 
1179 	spin_lock(&fsn_mark->lock);
1180 	if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS))
1181 		fsn_mark->mask |= mask;
1182 	else
1183 		fsn_mark->ignore_mask |= mask;
1184 
1185 	recalc = fsnotify_calc_mask(fsn_mark) &
1186 		~fsnotify_conn_mask(fsn_mark->connector);
1187 
1188 	recalc |= fanotify_mark_update_flags(fsn_mark, fan_flags);
1189 	spin_unlock(&fsn_mark->lock);
1190 
1191 	return recalc;
1192 }
1193 
1194 struct fan_fsid {
1195 	struct super_block *sb;
1196 	__kernel_fsid_t id;
1197 	bool weak;
1198 };
1199 
1200 static int fanotify_set_mark_fsid(struct fsnotify_group *group,
1201 				  struct fsnotify_mark *mark,
1202 				  struct fan_fsid *fsid)
1203 {
1204 	struct fsnotify_mark_connector *conn;
1205 	struct fsnotify_mark *old;
1206 	struct super_block *old_sb = NULL;
1207 
1208 	FANOTIFY_MARK(mark)->fsid = fsid->id;
1209 	mark->flags |= FSNOTIFY_MARK_FLAG_HAS_FSID;
1210 	if (fsid->weak)
1211 		mark->flags |= FSNOTIFY_MARK_FLAG_WEAK_FSID;
1212 
1213 	/* First mark added will determine if group is single or multi fsid */
1214 	if (list_empty(&group->marks_list))
1215 		return 0;
1216 
1217 	/* Find sb of an existing mark */
1218 	list_for_each_entry(old, &group->marks_list, g_list) {
1219 		conn = READ_ONCE(old->connector);
1220 		if (!conn)
1221 			continue;
1222 		old_sb = fsnotify_connector_sb(conn);
1223 		if (old_sb)
1224 			break;
1225 	}
1226 
1227 	/* Only detached marks left? */
1228 	if (!old_sb)
1229 		return 0;
1230 
1231 	/* Do not allow mixing of marks with weak and strong fsid */
1232 	if ((mark->flags ^ old->flags) & FSNOTIFY_MARK_FLAG_WEAK_FSID)
1233 		return -EXDEV;
1234 
1235 	/* Allow mixing of marks with strong fsid from different fs */
1236 	if (!fsid->weak)
1237 		return 0;
1238 
1239 	/* Do not allow mixing marks with weak fsid from different fs */
1240 	if (old_sb != fsid->sb)
1241 		return -EXDEV;
1242 
1243 	/* Do not allow mixing marks from different btrfs sub-volumes */
1244 	if (!fanotify_fsid_equal(&FANOTIFY_MARK(old)->fsid,
1245 				 &FANOTIFY_MARK(mark)->fsid))
1246 		return -EXDEV;
1247 
1248 	return 0;
1249 }
1250 
1251 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
1252 						   fsnotify_connp_t *connp,
1253 						   unsigned int obj_type,
1254 						   unsigned int fan_flags,
1255 						   struct fan_fsid *fsid)
1256 {
1257 	struct ucounts *ucounts = group->fanotify_data.ucounts;
1258 	struct fanotify_mark *fan_mark;
1259 	struct fsnotify_mark *mark;
1260 	int ret;
1261 
1262 	/*
1263 	 * Enforce per user marks limits per user in all containing user ns.
1264 	 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
1265 	 * in the limited groups account.
1266 	 */
1267 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
1268 	    !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
1269 		return ERR_PTR(-ENOSPC);
1270 
1271 	fan_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
1272 	if (!fan_mark) {
1273 		ret = -ENOMEM;
1274 		goto out_dec_ucounts;
1275 	}
1276 
1277 	mark = &fan_mark->fsn_mark;
1278 	fsnotify_init_mark(mark, group);
1279 	if (fan_flags & FAN_MARK_EVICTABLE)
1280 		mark->flags |= FSNOTIFY_MARK_FLAG_NO_IREF;
1281 
1282 	/* Cache fsid of filesystem containing the marked object */
1283 	if (fsid) {
1284 		ret = fanotify_set_mark_fsid(group, mark, fsid);
1285 		if (ret)
1286 			goto out_put_mark;
1287 	} else {
1288 		fan_mark->fsid.val[0] = fan_mark->fsid.val[1] = 0;
1289 	}
1290 
1291 	ret = fsnotify_add_mark_locked(mark, connp, obj_type, 0);
1292 	if (ret)
1293 		goto out_put_mark;
1294 
1295 	return mark;
1296 
1297 out_put_mark:
1298 	fsnotify_put_mark(mark);
1299 out_dec_ucounts:
1300 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1301 		dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
1302 	return ERR_PTR(ret);
1303 }
1304 
1305 static int fanotify_group_init_error_pool(struct fsnotify_group *group)
1306 {
1307 	if (mempool_initialized(&group->fanotify_data.error_events_pool))
1308 		return 0;
1309 
1310 	return mempool_init_kmalloc_pool(&group->fanotify_data.error_events_pool,
1311 					 FANOTIFY_DEFAULT_FEE_POOL_SIZE,
1312 					 sizeof(struct fanotify_error_event));
1313 }
1314 
1315 static int fanotify_may_update_existing_mark(struct fsnotify_mark *fsn_mark,
1316 					      unsigned int fan_flags)
1317 {
1318 	/*
1319 	 * Non evictable mark cannot be downgraded to evictable mark.
1320 	 */
1321 	if (fan_flags & FAN_MARK_EVICTABLE &&
1322 	    !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1323 		return -EEXIST;
1324 
1325 	/*
1326 	 * New ignore mask semantics cannot be downgraded to old semantics.
1327 	 */
1328 	if (fan_flags & FAN_MARK_IGNORED_MASK &&
1329 	    fsn_mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS)
1330 		return -EEXIST;
1331 
1332 	/*
1333 	 * An ignore mask that survives modify could never be downgraded to not
1334 	 * survive modify.  With new FAN_MARK_IGNORE semantics we make that rule
1335 	 * explicit and return an error when trying to update the ignore mask
1336 	 * without the original FAN_MARK_IGNORED_SURV_MODIFY value.
1337 	 */
1338 	if (fan_flags & FAN_MARK_IGNORE &&
1339 	    !(fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1340 	    fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)
1341 		return -EEXIST;
1342 
1343 	return 0;
1344 }
1345 
1346 static int fanotify_add_mark(struct fsnotify_group *group,
1347 			     fsnotify_connp_t *connp, unsigned int obj_type,
1348 			     __u32 mask, unsigned int fan_flags,
1349 			     struct fan_fsid *fsid)
1350 {
1351 	struct fsnotify_mark *fsn_mark;
1352 	bool recalc;
1353 	int ret = 0;
1354 
1355 	fsnotify_group_lock(group);
1356 	fsn_mark = fsnotify_find_mark(connp, group);
1357 	if (!fsn_mark) {
1358 		fsn_mark = fanotify_add_new_mark(group, connp, obj_type,
1359 						 fan_flags, fsid);
1360 		if (IS_ERR(fsn_mark)) {
1361 			fsnotify_group_unlock(group);
1362 			return PTR_ERR(fsn_mark);
1363 		}
1364 	}
1365 
1366 	/*
1367 	 * Check if requested mark flags conflict with an existing mark flags.
1368 	 */
1369 	ret = fanotify_may_update_existing_mark(fsn_mark, fan_flags);
1370 	if (ret)
1371 		goto out;
1372 
1373 	/*
1374 	 * Error events are pre-allocated per group, only if strictly
1375 	 * needed (i.e. FAN_FS_ERROR was requested).
1376 	 */
1377 	if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS) &&
1378 	    (mask & FAN_FS_ERROR)) {
1379 		ret = fanotify_group_init_error_pool(group);
1380 		if (ret)
1381 			goto out;
1382 	}
1383 
1384 	recalc = fanotify_mark_add_to_mask(fsn_mark, mask, fan_flags);
1385 	if (recalc)
1386 		fsnotify_recalc_mask(fsn_mark->connector);
1387 
1388 out:
1389 	fsnotify_group_unlock(group);
1390 
1391 	fsnotify_put_mark(fsn_mark);
1392 	return ret;
1393 }
1394 
1395 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
1396 				      struct vfsmount *mnt, __u32 mask,
1397 				      unsigned int flags, struct fan_fsid *fsid)
1398 {
1399 	return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
1400 				 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
1401 }
1402 
1403 static int fanotify_add_sb_mark(struct fsnotify_group *group,
1404 				struct super_block *sb, __u32 mask,
1405 				unsigned int flags, struct fan_fsid *fsid)
1406 {
1407 	return fanotify_add_mark(group, &sb->s_fsnotify_marks,
1408 				 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
1409 }
1410 
1411 static int fanotify_add_inode_mark(struct fsnotify_group *group,
1412 				   struct inode *inode, __u32 mask,
1413 				   unsigned int flags, struct fan_fsid *fsid)
1414 {
1415 	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
1416 
1417 	/*
1418 	 * If some other task has this inode open for write we should not add
1419 	 * an ignore mask, unless that ignore mask is supposed to survive
1420 	 * modification changes anyway.
1421 	 */
1422 	if ((flags & FANOTIFY_MARK_IGNORE_BITS) &&
1423 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1424 	    inode_is_open_for_write(inode))
1425 		return 0;
1426 
1427 	return fanotify_add_mark(group, &inode->i_fsnotify_marks,
1428 				 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
1429 }
1430 
1431 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1432 {
1433 	struct fanotify_event *oevent;
1434 
1435 	oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1436 	if (!oevent)
1437 		return NULL;
1438 
1439 	fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1440 	oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1441 
1442 	return &oevent->fse;
1443 }
1444 
1445 static struct hlist_head *fanotify_alloc_merge_hash(void)
1446 {
1447 	struct hlist_head *hash;
1448 
1449 	hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1450 		       GFP_KERNEL_ACCOUNT);
1451 	if (!hash)
1452 		return NULL;
1453 
1454 	__hash_init(hash, FANOTIFY_HTABLE_SIZE);
1455 
1456 	return hash;
1457 }
1458 
1459 /* fanotify syscalls */
1460 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1461 {
1462 	struct fsnotify_group *group;
1463 	int f_flags, fd;
1464 	unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1465 	unsigned int class = flags & FANOTIFY_CLASS_BITS;
1466 	unsigned int internal_flags = 0;
1467 
1468 	pr_debug("%s: flags=%x event_f_flags=%x\n",
1469 		 __func__, flags, event_f_flags);
1470 
1471 	if (!capable(CAP_SYS_ADMIN)) {
1472 		/*
1473 		 * An unprivileged user can setup an fanotify group with
1474 		 * limited functionality - an unprivileged group is limited to
1475 		 * notification events with file handles and it cannot use
1476 		 * unlimited queue/marks.
1477 		 */
1478 		if ((flags & FANOTIFY_ADMIN_INIT_FLAGS) || !fid_mode)
1479 			return -EPERM;
1480 
1481 		/*
1482 		 * Setting the internal flag FANOTIFY_UNPRIV on the group
1483 		 * prevents setting mount/filesystem marks on this group and
1484 		 * prevents reporting pid and open fd in events.
1485 		 */
1486 		internal_flags |= FANOTIFY_UNPRIV;
1487 	}
1488 
1489 #ifdef CONFIG_AUDITSYSCALL
1490 	if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1491 #else
1492 	if (flags & ~FANOTIFY_INIT_FLAGS)
1493 #endif
1494 		return -EINVAL;
1495 
1496 	/*
1497 	 * A pidfd can only be returned for a thread-group leader; thus
1498 	 * FAN_REPORT_PIDFD and FAN_REPORT_TID need to remain mutually
1499 	 * exclusive.
1500 	 */
1501 	if ((flags & FAN_REPORT_PIDFD) && (flags & FAN_REPORT_TID))
1502 		return -EINVAL;
1503 
1504 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1505 		return -EINVAL;
1506 
1507 	switch (event_f_flags & O_ACCMODE) {
1508 	case O_RDONLY:
1509 	case O_RDWR:
1510 	case O_WRONLY:
1511 		break;
1512 	default:
1513 		return -EINVAL;
1514 	}
1515 
1516 	if (fid_mode && class != FAN_CLASS_NOTIF)
1517 		return -EINVAL;
1518 
1519 	/*
1520 	 * Child name is reported with parent fid so requires dir fid.
1521 	 * We can report both child fid and dir fid with or without name.
1522 	 */
1523 	if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1524 		return -EINVAL;
1525 
1526 	/*
1527 	 * FAN_REPORT_TARGET_FID requires FAN_REPORT_NAME and FAN_REPORT_FID
1528 	 * and is used as an indication to report both dir and child fid on all
1529 	 * dirent events.
1530 	 */
1531 	if ((fid_mode & FAN_REPORT_TARGET_FID) &&
1532 	    (!(fid_mode & FAN_REPORT_NAME) || !(fid_mode & FAN_REPORT_FID)))
1533 		return -EINVAL;
1534 
1535 	f_flags = O_RDWR | __FMODE_NONOTIFY;
1536 	if (flags & FAN_CLOEXEC)
1537 		f_flags |= O_CLOEXEC;
1538 	if (flags & FAN_NONBLOCK)
1539 		f_flags |= O_NONBLOCK;
1540 
1541 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
1542 	group = fsnotify_alloc_group(&fanotify_fsnotify_ops,
1543 				     FSNOTIFY_GROUP_USER | FSNOTIFY_GROUP_NOFS);
1544 	if (IS_ERR(group)) {
1545 		return PTR_ERR(group);
1546 	}
1547 
1548 	/* Enforce groups limits per user in all containing user ns */
1549 	group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1550 						  current_euid(),
1551 						  UCOUNT_FANOTIFY_GROUPS);
1552 	if (!group->fanotify_data.ucounts) {
1553 		fd = -EMFILE;
1554 		goto out_destroy_group;
1555 	}
1556 
1557 	group->fanotify_data.flags = flags | internal_flags;
1558 	group->memcg = get_mem_cgroup_from_mm(current->mm);
1559 
1560 	group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1561 	if (!group->fanotify_data.merge_hash) {
1562 		fd = -ENOMEM;
1563 		goto out_destroy_group;
1564 	}
1565 
1566 	group->overflow_event = fanotify_alloc_overflow_event();
1567 	if (unlikely(!group->overflow_event)) {
1568 		fd = -ENOMEM;
1569 		goto out_destroy_group;
1570 	}
1571 
1572 	if (force_o_largefile())
1573 		event_f_flags |= O_LARGEFILE;
1574 	group->fanotify_data.f_flags = event_f_flags;
1575 	init_waitqueue_head(&group->fanotify_data.access_waitq);
1576 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
1577 	switch (class) {
1578 	case FAN_CLASS_NOTIF:
1579 		group->priority = FS_PRIO_0;
1580 		break;
1581 	case FAN_CLASS_CONTENT:
1582 		group->priority = FS_PRIO_1;
1583 		break;
1584 	case FAN_CLASS_PRE_CONTENT:
1585 		group->priority = FS_PRIO_2;
1586 		break;
1587 	default:
1588 		fd = -EINVAL;
1589 		goto out_destroy_group;
1590 	}
1591 
1592 	if (flags & FAN_UNLIMITED_QUEUE) {
1593 		fd = -EPERM;
1594 		if (!capable(CAP_SYS_ADMIN))
1595 			goto out_destroy_group;
1596 		group->max_events = UINT_MAX;
1597 	} else {
1598 		group->max_events = fanotify_max_queued_events;
1599 	}
1600 
1601 	if (flags & FAN_UNLIMITED_MARKS) {
1602 		fd = -EPERM;
1603 		if (!capable(CAP_SYS_ADMIN))
1604 			goto out_destroy_group;
1605 	}
1606 
1607 	if (flags & FAN_ENABLE_AUDIT) {
1608 		fd = -EPERM;
1609 		if (!capable(CAP_AUDIT_WRITE))
1610 			goto out_destroy_group;
1611 	}
1612 
1613 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
1614 	if (fd < 0)
1615 		goto out_destroy_group;
1616 
1617 	return fd;
1618 
1619 out_destroy_group:
1620 	fsnotify_destroy_group(group);
1621 	return fd;
1622 }
1623 
1624 static int fanotify_test_fsid(struct dentry *dentry, unsigned int flags,
1625 			      struct fan_fsid *fsid)
1626 {
1627 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1628 	__kernel_fsid_t root_fsid;
1629 	int err;
1630 
1631 	/*
1632 	 * Make sure dentry is not of a filesystem with zero fsid (e.g. fuse).
1633 	 */
1634 	err = vfs_get_fsid(dentry, &fsid->id);
1635 	if (err)
1636 		return err;
1637 
1638 	fsid->sb = dentry->d_sb;
1639 	if (!fsid->id.val[0] && !fsid->id.val[1]) {
1640 		err = -ENODEV;
1641 		goto weak;
1642 	}
1643 
1644 	/*
1645 	 * Make sure dentry is not of a filesystem subvolume (e.g. btrfs)
1646 	 * which uses a different fsid than sb root.
1647 	 */
1648 	err = vfs_get_fsid(dentry->d_sb->s_root, &root_fsid);
1649 	if (err)
1650 		return err;
1651 
1652 	if (!fanotify_fsid_equal(&root_fsid, &fsid->id)) {
1653 		err = -EXDEV;
1654 		goto weak;
1655 	}
1656 
1657 	fsid->weak = false;
1658 	return 0;
1659 
1660 weak:
1661 	/* Allow weak fsid when marking inodes */
1662 	fsid->weak = true;
1663 	return (mark_type == FAN_MARK_INODE) ? 0 : err;
1664 }
1665 
1666 /* Check if filesystem can encode a unique fid */
1667 static int fanotify_test_fid(struct dentry *dentry, unsigned int flags)
1668 {
1669 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1670 	const struct export_operations *nop = dentry->d_sb->s_export_op;
1671 
1672 	/*
1673 	 * We need to make sure that the filesystem supports encoding of
1674 	 * file handles so user can use name_to_handle_at() to compare fids
1675 	 * reported with events to the file handle of watched objects.
1676 	 */
1677 	if (!exportfs_can_encode_fid(nop))
1678 		return -EOPNOTSUPP;
1679 
1680 	/*
1681 	 * For sb/mount mark, we also need to make sure that the filesystem
1682 	 * supports decoding file handles, so user has a way to map back the
1683 	 * reported fids to filesystem objects.
1684 	 */
1685 	if (mark_type != FAN_MARK_INODE && !exportfs_can_decode_fh(nop))
1686 		return -EOPNOTSUPP;
1687 
1688 	return 0;
1689 }
1690 
1691 static int fanotify_events_supported(struct fsnotify_group *group,
1692 				     const struct path *path, __u64 mask,
1693 				     unsigned int flags)
1694 {
1695 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1696 	/* Strict validation of events in non-dir inode mask with v5.17+ APIs */
1697 	bool strict_dir_events = FAN_GROUP_FLAG(group, FAN_REPORT_TARGET_FID) ||
1698 				 (mask & FAN_RENAME) ||
1699 				 (flags & FAN_MARK_IGNORE);
1700 
1701 	/*
1702 	 * Some filesystems such as 'proc' acquire unusual locks when opening
1703 	 * files. For them fanotify permission events have high chances of
1704 	 * deadlocking the system - open done when reporting fanotify event
1705 	 * blocks on this "unusual" lock while another process holding the lock
1706 	 * waits for fanotify permission event to be answered. Just disallow
1707 	 * permission events for such filesystems.
1708 	 */
1709 	if (mask & FANOTIFY_PERM_EVENTS &&
1710 	    path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1711 		return -EINVAL;
1712 
1713 	/*
1714 	 * mount and sb marks are not allowed on kernel internal pseudo fs,
1715 	 * like pipe_mnt, because that would subscribe to events on all the
1716 	 * anonynous pipes in the system.
1717 	 *
1718 	 * SB_NOUSER covers all of the internal pseudo fs whose objects are not
1719 	 * exposed to user's mount namespace, but there are other SB_KERNMOUNT
1720 	 * fs, like nsfs, debugfs, for which the value of allowing sb and mount
1721 	 * mark is questionable. For now we leave them alone.
1722 	 */
1723 	if (mark_type != FAN_MARK_INODE &&
1724 	    path->mnt->mnt_sb->s_flags & SB_NOUSER)
1725 		return -EINVAL;
1726 
1727 	/*
1728 	 * We shouldn't have allowed setting dirent events and the directory
1729 	 * flags FAN_ONDIR and FAN_EVENT_ON_CHILD in mask of non-dir inode,
1730 	 * but because we always allowed it, error only when using new APIs.
1731 	 */
1732 	if (strict_dir_events && mark_type == FAN_MARK_INODE &&
1733 	    !d_is_dir(path->dentry) && (mask & FANOTIFY_DIRONLY_EVENT_BITS))
1734 		return -ENOTDIR;
1735 
1736 	return 0;
1737 }
1738 
1739 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1740 			    int dfd, const char  __user *pathname)
1741 {
1742 	struct inode *inode = NULL;
1743 	struct vfsmount *mnt = NULL;
1744 	struct fsnotify_group *group;
1745 	struct fd f;
1746 	struct path path;
1747 	struct fan_fsid __fsid, *fsid = NULL;
1748 	u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1749 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1750 	unsigned int mark_cmd = flags & FANOTIFY_MARK_CMD_BITS;
1751 	unsigned int ignore = flags & FANOTIFY_MARK_IGNORE_BITS;
1752 	unsigned int obj_type, fid_mode;
1753 	u32 umask = 0;
1754 	int ret;
1755 
1756 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1757 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
1758 
1759 	/* we only use the lower 32 bits as of right now. */
1760 	if (upper_32_bits(mask))
1761 		return -EINVAL;
1762 
1763 	if (flags & ~FANOTIFY_MARK_FLAGS)
1764 		return -EINVAL;
1765 
1766 	switch (mark_type) {
1767 	case FAN_MARK_INODE:
1768 		obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1769 		break;
1770 	case FAN_MARK_MOUNT:
1771 		obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1772 		break;
1773 	case FAN_MARK_FILESYSTEM:
1774 		obj_type = FSNOTIFY_OBJ_TYPE_SB;
1775 		break;
1776 	default:
1777 		return -EINVAL;
1778 	}
1779 
1780 	switch (mark_cmd) {
1781 	case FAN_MARK_ADD:
1782 	case FAN_MARK_REMOVE:
1783 		if (!mask)
1784 			return -EINVAL;
1785 		break;
1786 	case FAN_MARK_FLUSH:
1787 		if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1788 			return -EINVAL;
1789 		break;
1790 	default:
1791 		return -EINVAL;
1792 	}
1793 
1794 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1795 		valid_mask |= FANOTIFY_PERM_EVENTS;
1796 
1797 	if (mask & ~valid_mask)
1798 		return -EINVAL;
1799 
1800 
1801 	/* We don't allow FAN_MARK_IGNORE & FAN_MARK_IGNORED_MASK together */
1802 	if (ignore == (FAN_MARK_IGNORE | FAN_MARK_IGNORED_MASK))
1803 		return -EINVAL;
1804 
1805 	/*
1806 	 * Event flags (FAN_ONDIR, FAN_EVENT_ON_CHILD) have no effect with
1807 	 * FAN_MARK_IGNORED_MASK.
1808 	 */
1809 	if (ignore == FAN_MARK_IGNORED_MASK) {
1810 		mask &= ~FANOTIFY_EVENT_FLAGS;
1811 		umask = FANOTIFY_EVENT_FLAGS;
1812 	}
1813 
1814 	f = fdget(fanotify_fd);
1815 	if (unlikely(!f.file))
1816 		return -EBADF;
1817 
1818 	/* verify that this is indeed an fanotify instance */
1819 	ret = -EINVAL;
1820 	if (unlikely(f.file->f_op != &fanotify_fops))
1821 		goto fput_and_out;
1822 	group = f.file->private_data;
1823 
1824 	/*
1825 	 * An unprivileged user is not allowed to setup mount nor filesystem
1826 	 * marks.  This also includes setting up such marks by a group that
1827 	 * was initialized by an unprivileged user.
1828 	 */
1829 	ret = -EPERM;
1830 	if ((!capable(CAP_SYS_ADMIN) ||
1831 	     FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV)) &&
1832 	    mark_type != FAN_MARK_INODE)
1833 		goto fput_and_out;
1834 
1835 	/*
1836 	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
1837 	 * allowed to set permissions events.
1838 	 */
1839 	ret = -EINVAL;
1840 	if (mask & FANOTIFY_PERM_EVENTS &&
1841 	    group->priority == FS_PRIO_0)
1842 		goto fput_and_out;
1843 
1844 	if (mask & FAN_FS_ERROR &&
1845 	    mark_type != FAN_MARK_FILESYSTEM)
1846 		goto fput_and_out;
1847 
1848 	/*
1849 	 * Evictable is only relevant for inode marks, because only inode object
1850 	 * can be evicted on memory pressure.
1851 	 */
1852 	if (flags & FAN_MARK_EVICTABLE &&
1853 	     mark_type != FAN_MARK_INODE)
1854 		goto fput_and_out;
1855 
1856 	/*
1857 	 * Events that do not carry enough information to report
1858 	 * event->fd require a group that supports reporting fid.  Those
1859 	 * events are not supported on a mount mark, because they do not
1860 	 * carry enough information (i.e. path) to be filtered by mount
1861 	 * point.
1862 	 */
1863 	fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1864 	if (mask & ~(FANOTIFY_FD_EVENTS|FANOTIFY_EVENT_FLAGS) &&
1865 	    (!fid_mode || mark_type == FAN_MARK_MOUNT))
1866 		goto fput_and_out;
1867 
1868 	/*
1869 	 * FAN_RENAME uses special info type records to report the old and
1870 	 * new parent+name.  Reporting only old and new parent id is less
1871 	 * useful and was not implemented.
1872 	 */
1873 	if (mask & FAN_RENAME && !(fid_mode & FAN_REPORT_NAME))
1874 		goto fput_and_out;
1875 
1876 	if (mark_cmd == FAN_MARK_FLUSH) {
1877 		ret = 0;
1878 		if (mark_type == FAN_MARK_MOUNT)
1879 			fsnotify_clear_vfsmount_marks_by_group(group);
1880 		else if (mark_type == FAN_MARK_FILESYSTEM)
1881 			fsnotify_clear_sb_marks_by_group(group);
1882 		else
1883 			fsnotify_clear_inode_marks_by_group(group);
1884 		goto fput_and_out;
1885 	}
1886 
1887 	ret = fanotify_find_path(dfd, pathname, &path, flags,
1888 			(mask & ALL_FSNOTIFY_EVENTS), obj_type);
1889 	if (ret)
1890 		goto fput_and_out;
1891 
1892 	if (mark_cmd == FAN_MARK_ADD) {
1893 		ret = fanotify_events_supported(group, &path, mask, flags);
1894 		if (ret)
1895 			goto path_put_and_out;
1896 	}
1897 
1898 	if (fid_mode) {
1899 		ret = fanotify_test_fsid(path.dentry, flags, &__fsid);
1900 		if (ret)
1901 			goto path_put_and_out;
1902 
1903 		ret = fanotify_test_fid(path.dentry, flags);
1904 		if (ret)
1905 			goto path_put_and_out;
1906 
1907 		fsid = &__fsid;
1908 	}
1909 
1910 	/* inode held in place by reference to path; group by fget on fd */
1911 	if (mark_type == FAN_MARK_INODE)
1912 		inode = path.dentry->d_inode;
1913 	else
1914 		mnt = path.mnt;
1915 
1916 	ret = mnt ? -EINVAL : -EISDIR;
1917 	/* FAN_MARK_IGNORE requires SURV_MODIFY for sb/mount/dir marks */
1918 	if (mark_cmd == FAN_MARK_ADD && ignore == FAN_MARK_IGNORE &&
1919 	    (mnt || S_ISDIR(inode->i_mode)) &&
1920 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY))
1921 		goto path_put_and_out;
1922 
1923 	/* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1924 	if (mnt || !S_ISDIR(inode->i_mode)) {
1925 		mask &= ~FAN_EVENT_ON_CHILD;
1926 		umask = FAN_EVENT_ON_CHILD;
1927 		/*
1928 		 * If group needs to report parent fid, register for getting
1929 		 * events with parent/name info for non-directory.
1930 		 */
1931 		if ((fid_mode & FAN_REPORT_DIR_FID) &&
1932 		    (flags & FAN_MARK_ADD) && !ignore)
1933 			mask |= FAN_EVENT_ON_CHILD;
1934 	}
1935 
1936 	/* create/update an inode mark */
1937 	switch (mark_cmd) {
1938 	case FAN_MARK_ADD:
1939 		if (mark_type == FAN_MARK_MOUNT)
1940 			ret = fanotify_add_vfsmount_mark(group, mnt, mask,
1941 							 flags, fsid);
1942 		else if (mark_type == FAN_MARK_FILESYSTEM)
1943 			ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
1944 						   flags, fsid);
1945 		else
1946 			ret = fanotify_add_inode_mark(group, inode, mask,
1947 						      flags, fsid);
1948 		break;
1949 	case FAN_MARK_REMOVE:
1950 		if (mark_type == FAN_MARK_MOUNT)
1951 			ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
1952 							    flags, umask);
1953 		else if (mark_type == FAN_MARK_FILESYSTEM)
1954 			ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
1955 						      flags, umask);
1956 		else
1957 			ret = fanotify_remove_inode_mark(group, inode, mask,
1958 							 flags, umask);
1959 		break;
1960 	default:
1961 		ret = -EINVAL;
1962 	}
1963 
1964 path_put_and_out:
1965 	path_put(&path);
1966 fput_and_out:
1967 	fdput(f);
1968 	return ret;
1969 }
1970 
1971 #ifndef CONFIG_ARCH_SPLIT_ARG64
1972 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1973 			      __u64, mask, int, dfd,
1974 			      const char  __user *, pathname)
1975 {
1976 	return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1977 }
1978 #endif
1979 
1980 #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT)
1981 SYSCALL32_DEFINE6(fanotify_mark,
1982 				int, fanotify_fd, unsigned int, flags,
1983 				SC_ARG64(mask), int, dfd,
1984 				const char  __user *, pathname)
1985 {
1986 	return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
1987 				dfd, pathname);
1988 }
1989 #endif
1990 
1991 /*
1992  * fanotify_user_setup - Our initialization function.  Note that we cannot return
1993  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
1994  * must result in panic().
1995  */
1996 static int __init fanotify_user_setup(void)
1997 {
1998 	struct sysinfo si;
1999 	int max_marks;
2000 
2001 	si_meminfo(&si);
2002 	/*
2003 	 * Allow up to 1% of addressable memory to be accounted for per user
2004 	 * marks limited to the range [8192, 1048576]. mount and sb marks are
2005 	 * a lot cheaper than inode marks, but there is no reason for a user
2006 	 * to have many of those, so calculate by the cost of inode marks.
2007 	 */
2008 	max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
2009 		    INODE_MARK_COST;
2010 	max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
2011 				     FANOTIFY_DEFAULT_MAX_USER_MARKS);
2012 
2013 	BUILD_BUG_ON(FANOTIFY_INIT_FLAGS & FANOTIFY_INTERNAL_GROUP_FLAGS);
2014 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 12);
2015 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 11);
2016 
2017 	fanotify_mark_cache = KMEM_CACHE(fanotify_mark,
2018 					 SLAB_PANIC|SLAB_ACCOUNT);
2019 	fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
2020 					       SLAB_PANIC);
2021 	fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
2022 						SLAB_PANIC);
2023 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
2024 		fanotify_perm_event_cachep =
2025 			KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
2026 	}
2027 
2028 	fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
2029 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
2030 					FANOTIFY_DEFAULT_MAX_GROUPS;
2031 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
2032 	fanotify_sysctls_init();
2033 
2034 	return 0;
2035 }
2036 device_initcall(fanotify_user_setup);
2037