• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
fanotify_sysctls_init(void)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 
fanotify_fid_info_len(int fh_len,int name_len)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 */
fanotify_dir_name_info_len(struct fanotify_event * event)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 
fanotify_event_len(unsigned int info_mode,struct fanotify_event * event)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  */
fanotify_unhash_event(struct fsnotify_group * group,struct fanotify_event * event)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  */
get_one_event(struct fsnotify_group * group,size_t count)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 
create_fd(struct fsnotify_group * group,const struct path * path,struct file ** file)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 		put_unused_fd(client_fd);
270 		client_fd = PTR_ERR(new_file);
271 	} else {
272 		*file = new_file;
273 	}
274 
275 	return client_fd;
276 }
277 
process_access_response_info(const char __user * info,size_t info_len,struct fanotify_response_info_audit_rule * friar)278 static int process_access_response_info(const char __user *info,
279 					size_t info_len,
280 				struct fanotify_response_info_audit_rule *friar)
281 {
282 	if (info_len != sizeof(*friar))
283 		return -EINVAL;
284 
285 	if (copy_from_user(friar, info, sizeof(*friar)))
286 		return -EFAULT;
287 
288 	if (friar->hdr.type != FAN_RESPONSE_INFO_AUDIT_RULE)
289 		return -EINVAL;
290 	if (friar->hdr.pad != 0)
291 		return -EINVAL;
292 	if (friar->hdr.len != sizeof(*friar))
293 		return -EINVAL;
294 
295 	return info_len;
296 }
297 
298 /*
299  * Finish processing of permission event by setting it to ANSWERED state and
300  * drop group->notification_lock.
301  */
finish_permission_event(struct fsnotify_group * group,struct fanotify_perm_event * event,u32 response,struct fanotify_response_info_audit_rule * friar)302 static void finish_permission_event(struct fsnotify_group *group,
303 				    struct fanotify_perm_event *event, u32 response,
304 				    struct fanotify_response_info_audit_rule *friar)
305 				    __releases(&group->notification_lock)
306 {
307 	bool destroy = false;
308 
309 	assert_spin_locked(&group->notification_lock);
310 	event->response = response & ~FAN_INFO;
311 	if (response & FAN_INFO)
312 		memcpy(&event->audit_rule, friar, sizeof(*friar));
313 
314 	if (event->state == FAN_EVENT_CANCELED)
315 		destroy = true;
316 	else
317 		event->state = FAN_EVENT_ANSWERED;
318 	spin_unlock(&group->notification_lock);
319 	if (destroy)
320 		fsnotify_destroy_event(group, &event->fae.fse);
321 }
322 
process_access_response(struct fsnotify_group * group,struct fanotify_response * response_struct,const char __user * info,size_t info_len)323 static int process_access_response(struct fsnotify_group *group,
324 				   struct fanotify_response *response_struct,
325 				   const char __user *info,
326 				   size_t info_len)
327 {
328 	struct fanotify_perm_event *event;
329 	int fd = response_struct->fd;
330 	u32 response = response_struct->response;
331 	int ret = info_len;
332 	struct fanotify_response_info_audit_rule friar;
333 
334 	pr_debug("%s: group=%p fd=%d response=%u buf=%p size=%zu\n", __func__,
335 		 group, fd, response, info, info_len);
336 	/*
337 	 * make sure the response is valid, if invalid we do nothing and either
338 	 * userspace can send a valid response or we will clean it up after the
339 	 * timeout
340 	 */
341 	if (response & ~FANOTIFY_RESPONSE_VALID_MASK)
342 		return -EINVAL;
343 
344 	switch (response & FANOTIFY_RESPONSE_ACCESS) {
345 	case FAN_ALLOW:
346 	case FAN_DENY:
347 		break;
348 	default:
349 		return -EINVAL;
350 	}
351 
352 	if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
353 		return -EINVAL;
354 
355 	if (response & FAN_INFO) {
356 		ret = process_access_response_info(info, info_len, &friar);
357 		if (ret < 0)
358 			return ret;
359 		if (fd == FAN_NOFD)
360 			return ret;
361 	} else {
362 		ret = 0;
363 	}
364 
365 	if (fd < 0)
366 		return -EINVAL;
367 
368 	spin_lock(&group->notification_lock);
369 	list_for_each_entry(event, &group->fanotify_data.access_list,
370 			    fae.fse.list) {
371 		if (event->fd != fd)
372 			continue;
373 
374 		list_del_init(&event->fae.fse.list);
375 		finish_permission_event(group, event, response, &friar);
376 		wake_up(&group->fanotify_data.access_waitq);
377 		return ret;
378 	}
379 	spin_unlock(&group->notification_lock);
380 
381 	return -ENOENT;
382 }
383 
copy_error_info_to_user(struct fanotify_event * event,char __user * buf,int count)384 static size_t copy_error_info_to_user(struct fanotify_event *event,
385 				      char __user *buf, int count)
386 {
387 	struct fanotify_event_info_error info = { };
388 	struct fanotify_error_event *fee = FANOTIFY_EE(event);
389 
390 	info.hdr.info_type = FAN_EVENT_INFO_TYPE_ERROR;
391 	info.hdr.len = FANOTIFY_ERROR_INFO_LEN;
392 
393 	if (WARN_ON(count < info.hdr.len))
394 		return -EFAULT;
395 
396 	info.error = fee->error;
397 	info.error_count = fee->err_count;
398 
399 	if (copy_to_user(buf, &info, sizeof(info)))
400 		return -EFAULT;
401 
402 	return info.hdr.len;
403 }
404 
copy_fid_info_to_user(__kernel_fsid_t * fsid,struct fanotify_fh * fh,int info_type,const char * name,size_t name_len,char __user * buf,size_t count)405 static int copy_fid_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
406 				 int info_type, const char *name,
407 				 size_t name_len,
408 				 char __user *buf, size_t count)
409 {
410 	struct fanotify_event_info_fid info = { };
411 	struct file_handle handle = { };
412 	unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
413 	size_t fh_len = fh ? fh->len : 0;
414 	size_t info_len = fanotify_fid_info_len(fh_len, name_len);
415 	size_t len = info_len;
416 
417 	pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
418 		 __func__, fh_len, name_len, info_len, count);
419 
420 	if (WARN_ON_ONCE(len < sizeof(info) || len > count))
421 		return -EFAULT;
422 
423 	/*
424 	 * Copy event info fid header followed by variable sized file handle
425 	 * and optionally followed by variable sized filename.
426 	 */
427 	switch (info_type) {
428 	case FAN_EVENT_INFO_TYPE_FID:
429 	case FAN_EVENT_INFO_TYPE_DFID:
430 		if (WARN_ON_ONCE(name_len))
431 			return -EFAULT;
432 		break;
433 	case FAN_EVENT_INFO_TYPE_DFID_NAME:
434 	case FAN_EVENT_INFO_TYPE_OLD_DFID_NAME:
435 	case FAN_EVENT_INFO_TYPE_NEW_DFID_NAME:
436 		if (WARN_ON_ONCE(!name || !name_len))
437 			return -EFAULT;
438 		break;
439 	default:
440 		return -EFAULT;
441 	}
442 
443 	info.hdr.info_type = info_type;
444 	info.hdr.len = len;
445 	info.fsid = *fsid;
446 	if (copy_to_user(buf, &info, sizeof(info)))
447 		return -EFAULT;
448 
449 	buf += sizeof(info);
450 	len -= sizeof(info);
451 	if (WARN_ON_ONCE(len < sizeof(handle)))
452 		return -EFAULT;
453 
454 	handle.handle_type = fh->type;
455 	handle.handle_bytes = fh_len;
456 
457 	/* Mangle handle_type for bad file_handle */
458 	if (!fh_len)
459 		handle.handle_type = FILEID_INVALID;
460 
461 	if (copy_to_user(buf, &handle, sizeof(handle)))
462 		return -EFAULT;
463 
464 	buf += sizeof(handle);
465 	len -= sizeof(handle);
466 	if (WARN_ON_ONCE(len < fh_len))
467 		return -EFAULT;
468 
469 	/*
470 	 * For an inline fh and inline file name, copy through stack to exclude
471 	 * the copy from usercopy hardening protections.
472 	 */
473 	fh_buf = fanotify_fh_buf(fh);
474 	if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
475 		memcpy(bounce, fh_buf, fh_len);
476 		fh_buf = bounce;
477 	}
478 	if (copy_to_user(buf, fh_buf, fh_len))
479 		return -EFAULT;
480 
481 	buf += fh_len;
482 	len -= fh_len;
483 
484 	if (name_len) {
485 		/* Copy the filename with terminating null */
486 		name_len++;
487 		if (WARN_ON_ONCE(len < name_len))
488 			return -EFAULT;
489 
490 		if (copy_to_user(buf, name, name_len))
491 			return -EFAULT;
492 
493 		buf += name_len;
494 		len -= name_len;
495 	}
496 
497 	/* Pad with 0's */
498 	WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
499 	if (len > 0 && clear_user(buf, len))
500 		return -EFAULT;
501 
502 	return info_len;
503 }
504 
copy_pidfd_info_to_user(int pidfd,char __user * buf,size_t count)505 static int copy_pidfd_info_to_user(int pidfd,
506 				   char __user *buf,
507 				   size_t count)
508 {
509 	struct fanotify_event_info_pidfd info = { };
510 	size_t info_len = FANOTIFY_PIDFD_INFO_HDR_LEN;
511 
512 	if (WARN_ON_ONCE(info_len > count))
513 		return -EFAULT;
514 
515 	info.hdr.info_type = FAN_EVENT_INFO_TYPE_PIDFD;
516 	info.hdr.len = info_len;
517 	info.pidfd = pidfd;
518 
519 	if (copy_to_user(buf, &info, info_len))
520 		return -EFAULT;
521 
522 	return info_len;
523 }
524 
copy_info_records_to_user(struct fanotify_event * event,struct fanotify_info * info,unsigned int info_mode,int pidfd,char __user * buf,size_t count)525 static int copy_info_records_to_user(struct fanotify_event *event,
526 				     struct fanotify_info *info,
527 				     unsigned int info_mode, int pidfd,
528 				     char __user *buf, size_t count)
529 {
530 	int ret, total_bytes = 0, info_type = 0;
531 	unsigned int fid_mode = info_mode & FANOTIFY_FID_BITS;
532 	unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
533 
534 	/*
535 	 * Event info records order is as follows:
536 	 * 1. dir fid + name
537 	 * 2. (optional) new dir fid + new name
538 	 * 3. (optional) child fid
539 	 */
540 	if (fanotify_event_has_dir_fh(event)) {
541 		info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
542 					     FAN_EVENT_INFO_TYPE_DFID;
543 
544 		/* FAN_RENAME uses special info types */
545 		if (event->mask & FAN_RENAME)
546 			info_type = FAN_EVENT_INFO_TYPE_OLD_DFID_NAME;
547 
548 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
549 					    fanotify_info_dir_fh(info),
550 					    info_type,
551 					    fanotify_info_name(info),
552 					    info->name_len, buf, count);
553 		if (ret < 0)
554 			return ret;
555 
556 		buf += ret;
557 		count -= ret;
558 		total_bytes += ret;
559 	}
560 
561 	/* New dir fid+name may be reported in addition to old dir fid+name */
562 	if (fanotify_event_has_dir2_fh(event)) {
563 		info_type = FAN_EVENT_INFO_TYPE_NEW_DFID_NAME;
564 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
565 					    fanotify_info_dir2_fh(info),
566 					    info_type,
567 					    fanotify_info_name2(info),
568 					    info->name2_len, buf, count);
569 		if (ret < 0)
570 			return ret;
571 
572 		buf += ret;
573 		count -= ret;
574 		total_bytes += ret;
575 	}
576 
577 	if (fanotify_event_has_object_fh(event)) {
578 		const char *dot = NULL;
579 		int dot_len = 0;
580 
581 		if (fid_mode == FAN_REPORT_FID || info_type) {
582 			/*
583 			 * With only group flag FAN_REPORT_FID only type FID is
584 			 * reported. Second info record type is always FID.
585 			 */
586 			info_type = FAN_EVENT_INFO_TYPE_FID;
587 		} else if ((fid_mode & FAN_REPORT_NAME) &&
588 			   (event->mask & FAN_ONDIR)) {
589 			/*
590 			 * With group flag FAN_REPORT_NAME, if name was not
591 			 * recorded in an event on a directory, report the name
592 			 * "." with info type DFID_NAME.
593 			 */
594 			info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
595 			dot = ".";
596 			dot_len = 1;
597 		} else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
598 			   (event->mask & FAN_ONDIR)) {
599 			/*
600 			 * With group flag FAN_REPORT_DIR_FID, a single info
601 			 * record has type DFID for directory entry modification
602 			 * event and for event on a directory.
603 			 */
604 			info_type = FAN_EVENT_INFO_TYPE_DFID;
605 		} else {
606 			/*
607 			 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
608 			 * a single info record has type FID for event on a
609 			 * non-directory, when there is no directory to report.
610 			 * For example, on FAN_DELETE_SELF event.
611 			 */
612 			info_type = FAN_EVENT_INFO_TYPE_FID;
613 		}
614 
615 		ret = copy_fid_info_to_user(fanotify_event_fsid(event),
616 					    fanotify_event_object_fh(event),
617 					    info_type, dot, dot_len,
618 					    buf, count);
619 		if (ret < 0)
620 			return ret;
621 
622 		buf += ret;
623 		count -= ret;
624 		total_bytes += ret;
625 	}
626 
627 	if (pidfd_mode) {
628 		ret = copy_pidfd_info_to_user(pidfd, buf, count);
629 		if (ret < 0)
630 			return ret;
631 
632 		buf += ret;
633 		count -= ret;
634 		total_bytes += ret;
635 	}
636 
637 	if (fanotify_is_error_event(event->mask)) {
638 		ret = copy_error_info_to_user(event, buf, count);
639 		if (ret < 0)
640 			return ret;
641 		buf += ret;
642 		count -= ret;
643 		total_bytes += ret;
644 	}
645 
646 	return total_bytes;
647 }
648 
copy_event_to_user(struct fsnotify_group * group,struct fanotify_event * event,char __user * buf,size_t count)649 static ssize_t copy_event_to_user(struct fsnotify_group *group,
650 				  struct fanotify_event *event,
651 				  char __user *buf, size_t count)
652 {
653 	struct fanotify_event_metadata metadata;
654 	const struct path *path = fanotify_event_path(event);
655 	struct fanotify_info *info = fanotify_event_info(event);
656 	unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
657 	unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
658 	struct file *f = NULL, *pidfd_file = NULL;
659 	int ret, pidfd = -ESRCH, fd = -EBADF;
660 
661 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
662 
663 	metadata.event_len = fanotify_event_len(info_mode, event);
664 	metadata.metadata_len = FAN_EVENT_METADATA_LEN;
665 	metadata.vers = FANOTIFY_METADATA_VERSION;
666 	metadata.reserved = 0;
667 	metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
668 	metadata.pid = pid_vnr(event->pid);
669 	/*
670 	 * For an unprivileged listener, event->pid can be used to identify the
671 	 * events generated by the listener process itself, without disclosing
672 	 * the pids of other processes.
673 	 */
674 	if (FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
675 	    task_tgid(current) != event->pid)
676 		metadata.pid = 0;
677 
678 	/*
679 	 * For now, fid mode is required for an unprivileged listener and
680 	 * fid mode does not report fd in events.  Keep this check anyway
681 	 * for safety in case fid mode requirement is relaxed in the future
682 	 * to allow unprivileged listener to get events with no fd and no fid.
683 	 */
684 	if (!FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
685 	    path && path->mnt && path->dentry) {
686 		fd = create_fd(group, path, &f);
687 		/*
688 		 * Opening an fd from dentry can fail for several reasons.
689 		 * For example, when tasks are gone and we try to open their
690 		 * /proc files or we try to open a WRONLY file like in sysfs
691 		 * or when trying to open a file that was deleted on the
692 		 * remote network server.
693 		 *
694 		 * For a group with FAN_REPORT_FD_ERROR, we will send the
695 		 * event with the error instead of the open fd, otherwise
696 		 * Userspace may not get the error at all.
697 		 * In any case, userspace will not know which file failed to
698 		 * open, so add a debug print for further investigation.
699 		 */
700 		if (fd < 0) {
701 			pr_debug("fanotify: create_fd(%pd2) failed err=%d\n",
702 				 path->dentry, fd);
703 			if (!FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR)) {
704 				/*
705 				 * Historically, we've handled EOPENSTALE in a
706 				 * special way and silently dropped such
707 				 * events. Now we have to keep it to maintain
708 				 * backward compatibility...
709 				 */
710 				if (fd == -EOPENSTALE)
711 					fd = 0;
712 				return fd;
713 			}
714 		}
715 	}
716 	if (FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR))
717 		metadata.fd = fd;
718 	else
719 		metadata.fd = fd >= 0 ? fd : FAN_NOFD;
720 
721 	if (pidfd_mode) {
722 		/*
723 		 * Complain if the FAN_REPORT_PIDFD and FAN_REPORT_TID mutual
724 		 * exclusion is ever lifted. At the time of incoporating pidfd
725 		 * support within fanotify, the pidfd API only supported the
726 		 * creation of pidfds for thread-group leaders.
727 		 */
728 		WARN_ON_ONCE(FAN_GROUP_FLAG(group, FAN_REPORT_TID));
729 
730 		/*
731 		 * The PIDTYPE_TGID check for an event->pid is performed
732 		 * preemptively in an attempt to catch out cases where the event
733 		 * listener reads events after the event generating process has
734 		 * already terminated.  Depending on flag FAN_REPORT_FD_ERROR,
735 		 * report either -ESRCH or FAN_NOPIDFD to the event listener in
736 		 * those cases with all other pidfd creation errors reported as
737 		 * the error code itself or as FAN_EPIDFD.
738 		 */
739 		if (metadata.pid && pid_has_task(event->pid, PIDTYPE_TGID))
740 			pidfd = pidfd_prepare(event->pid, 0, &pidfd_file);
741 
742 		if (!FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR) && pidfd < 0)
743 			pidfd = pidfd == -ESRCH ? FAN_NOPIDFD : FAN_EPIDFD;
744 	}
745 
746 	ret = -EFAULT;
747 	/*
748 	 * Sanity check copy size in case get_one_event() and
749 	 * event_len sizes ever get out of sync.
750 	 */
751 	if (WARN_ON_ONCE(metadata.event_len > count))
752 		goto out_close_fd;
753 
754 	if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
755 		goto out_close_fd;
756 
757 	buf += FAN_EVENT_METADATA_LEN;
758 	count -= FAN_EVENT_METADATA_LEN;
759 
760 	if (info_mode) {
761 		ret = copy_info_records_to_user(event, info, info_mode, pidfd,
762 						buf, count);
763 		if (ret < 0)
764 			goto out_close_fd;
765 	}
766 
767 	if (f)
768 		fd_install(fd, f);
769 
770 	if (pidfd_file)
771 		fd_install(pidfd, pidfd_file);
772 
773 	if (fanotify_is_perm_event(event->mask))
774 		FANOTIFY_PERM(event)->fd = fd;
775 
776 	return metadata.event_len;
777 
778 out_close_fd:
779 	if (f) {
780 		put_unused_fd(fd);
781 		fput(f);
782 	}
783 
784 	if (pidfd_file) {
785 		put_unused_fd(pidfd);
786 		fput(pidfd_file);
787 	}
788 
789 	return ret;
790 }
791 
792 /* intofiy userspace file descriptor functions */
fanotify_poll(struct file * file,poll_table * wait)793 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
794 {
795 	struct fsnotify_group *group = file->private_data;
796 	__poll_t ret = 0;
797 
798 	poll_wait(file, &group->notification_waitq, wait);
799 	spin_lock(&group->notification_lock);
800 	if (!fsnotify_notify_queue_is_empty(group))
801 		ret = EPOLLIN | EPOLLRDNORM;
802 	spin_unlock(&group->notification_lock);
803 
804 	return ret;
805 }
806 
fanotify_read(struct file * file,char __user * buf,size_t count,loff_t * pos)807 static ssize_t fanotify_read(struct file *file, char __user *buf,
808 			     size_t count, loff_t *pos)
809 {
810 	struct fsnotify_group *group;
811 	struct fanotify_event *event;
812 	char __user *start;
813 	int ret;
814 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
815 
816 	start = buf;
817 	group = file->private_data;
818 
819 	pr_debug("%s: group=%p\n", __func__, group);
820 
821 	add_wait_queue(&group->notification_waitq, &wait);
822 	while (1) {
823 		/*
824 		 * User can supply arbitrarily large buffer. Avoid softlockups
825 		 * in case there are lots of available events.
826 		 */
827 		cond_resched();
828 		event = get_one_event(group, count);
829 		if (IS_ERR(event)) {
830 			ret = PTR_ERR(event);
831 			break;
832 		}
833 
834 		if (!event) {
835 			ret = -EAGAIN;
836 			if (file->f_flags & O_NONBLOCK)
837 				break;
838 
839 			ret = -ERESTARTSYS;
840 			if (signal_pending(current))
841 				break;
842 
843 			if (start != buf)
844 				break;
845 
846 			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
847 			continue;
848 		}
849 
850 		ret = copy_event_to_user(group, event, buf, count);
851 
852 		/*
853 		 * Permission events get queued to wait for response.  Other
854 		 * events can be destroyed now.
855 		 */
856 		if (!fanotify_is_perm_event(event->mask)) {
857 			fsnotify_destroy_event(group, &event->fse);
858 		} else {
859 			if (ret <= 0 || FANOTIFY_PERM(event)->fd < 0) {
860 				spin_lock(&group->notification_lock);
861 				finish_permission_event(group,
862 					FANOTIFY_PERM(event), FAN_DENY, NULL);
863 				wake_up(&group->fanotify_data.access_waitq);
864 			} else {
865 				spin_lock(&group->notification_lock);
866 				list_add_tail(&event->fse.list,
867 					&group->fanotify_data.access_list);
868 				spin_unlock(&group->notification_lock);
869 			}
870 		}
871 		if (ret < 0)
872 			break;
873 		buf += ret;
874 		count -= ret;
875 	}
876 	remove_wait_queue(&group->notification_waitq, &wait);
877 
878 	if (start != buf && ret != -EFAULT)
879 		ret = buf - start;
880 	return ret;
881 }
882 
fanotify_write(struct file * file,const char __user * buf,size_t count,loff_t * pos)883 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
884 {
885 	struct fanotify_response response;
886 	struct fsnotify_group *group;
887 	int ret;
888 	const char __user *info_buf = buf + sizeof(struct fanotify_response);
889 	size_t info_len;
890 
891 	if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
892 		return -EINVAL;
893 
894 	group = file->private_data;
895 
896 	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
897 
898 	if (count < sizeof(response))
899 		return -EINVAL;
900 
901 	if (copy_from_user(&response, buf, sizeof(response)))
902 		return -EFAULT;
903 
904 	info_len = count - sizeof(response);
905 
906 	ret = process_access_response(group, &response, info_buf, info_len);
907 	if (ret < 0)
908 		count = ret;
909 	else
910 		count = sizeof(response) + ret;
911 
912 	return count;
913 }
914 
fanotify_release(struct inode * ignored,struct file * file)915 static int fanotify_release(struct inode *ignored, struct file *file)
916 {
917 	struct fsnotify_group *group = file->private_data;
918 	struct fsnotify_event *fsn_event;
919 
920 	/*
921 	 * Stop new events from arriving in the notification queue. since
922 	 * userspace cannot use fanotify fd anymore, no event can enter or
923 	 * leave access_list by now either.
924 	 */
925 	fsnotify_group_stop_queueing(group);
926 
927 	/*
928 	 * Process all permission events on access_list and notification queue
929 	 * and simulate reply from userspace.
930 	 */
931 	spin_lock(&group->notification_lock);
932 	while (!list_empty(&group->fanotify_data.access_list)) {
933 		struct fanotify_perm_event *event;
934 
935 		event = list_first_entry(&group->fanotify_data.access_list,
936 				struct fanotify_perm_event, fae.fse.list);
937 		list_del_init(&event->fae.fse.list);
938 		finish_permission_event(group, event, FAN_ALLOW, NULL);
939 		spin_lock(&group->notification_lock);
940 	}
941 
942 	/*
943 	 * Destroy all non-permission events. For permission events just
944 	 * dequeue them and set the response. They will be freed once the
945 	 * response is consumed and fanotify_get_response() returns.
946 	 */
947 	while ((fsn_event = fsnotify_remove_first_event(group))) {
948 		struct fanotify_event *event = FANOTIFY_E(fsn_event);
949 
950 		if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
951 			spin_unlock(&group->notification_lock);
952 			fsnotify_destroy_event(group, fsn_event);
953 		} else {
954 			finish_permission_event(group, FANOTIFY_PERM(event),
955 						FAN_ALLOW, NULL);
956 		}
957 		spin_lock(&group->notification_lock);
958 	}
959 	spin_unlock(&group->notification_lock);
960 
961 	/* Response for all permission events it set, wakeup waiters */
962 	wake_up(&group->fanotify_data.access_waitq);
963 
964 	/* matches the fanotify_init->fsnotify_alloc_group */
965 	fsnotify_destroy_group(group);
966 
967 	return 0;
968 }
969 
fanotify_ioctl(struct file * file,unsigned int cmd,unsigned long arg)970 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
971 {
972 	struct fsnotify_group *group;
973 	struct fsnotify_event *fsn_event;
974 	void __user *p;
975 	int ret = -ENOTTY;
976 	size_t send_len = 0;
977 
978 	group = file->private_data;
979 
980 	p = (void __user *) arg;
981 
982 	switch (cmd) {
983 	case FIONREAD:
984 		spin_lock(&group->notification_lock);
985 		list_for_each_entry(fsn_event, &group->notification_list, list)
986 			send_len += FAN_EVENT_METADATA_LEN;
987 		spin_unlock(&group->notification_lock);
988 		ret = put_user(send_len, (int __user *) p);
989 		break;
990 	}
991 
992 	return ret;
993 }
994 
995 static const struct file_operations fanotify_fops = {
996 	.show_fdinfo	= fanotify_show_fdinfo,
997 	.poll		= fanotify_poll,
998 	.read		= fanotify_read,
999 	.write		= fanotify_write,
1000 	.fasync		= NULL,
1001 	.release	= fanotify_release,
1002 	.unlocked_ioctl	= fanotify_ioctl,
1003 	.compat_ioctl	= compat_ptr_ioctl,
1004 	.llseek		= noop_llseek,
1005 };
1006 
fanotify_find_path(int dfd,const char __user * filename,struct path * path,unsigned int flags,__u64 mask,unsigned int obj_type)1007 static int fanotify_find_path(int dfd, const char __user *filename,
1008 			      struct path *path, unsigned int flags, __u64 mask,
1009 			      unsigned int obj_type)
1010 {
1011 	int ret;
1012 
1013 	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
1014 		 dfd, filename, flags);
1015 
1016 	if (filename == NULL) {
1017 		struct fd f = fdget(dfd);
1018 
1019 		ret = -EBADF;
1020 		if (!fd_file(f))
1021 			goto out;
1022 
1023 		ret = -ENOTDIR;
1024 		if ((flags & FAN_MARK_ONLYDIR) &&
1025 		    !(S_ISDIR(file_inode(fd_file(f))->i_mode))) {
1026 			fdput(f);
1027 			goto out;
1028 		}
1029 
1030 		*path = fd_file(f)->f_path;
1031 		path_get(path);
1032 		fdput(f);
1033 	} else {
1034 		unsigned int lookup_flags = 0;
1035 
1036 		if (!(flags & FAN_MARK_DONT_FOLLOW))
1037 			lookup_flags |= LOOKUP_FOLLOW;
1038 		if (flags & FAN_MARK_ONLYDIR)
1039 			lookup_flags |= LOOKUP_DIRECTORY;
1040 
1041 		ret = user_path_at(dfd, filename, lookup_flags, path);
1042 		if (ret)
1043 			goto out;
1044 	}
1045 
1046 	/* you can only watch an inode if you have read permissions on it */
1047 	ret = path_permission(path, MAY_READ);
1048 	if (ret) {
1049 		path_put(path);
1050 		goto out;
1051 	}
1052 
1053 	ret = security_path_notify(path, mask, obj_type);
1054 	if (ret)
1055 		path_put(path);
1056 
1057 out:
1058 	return ret;
1059 }
1060 
fanotify_mark_remove_from_mask(struct fsnotify_mark * fsn_mark,__u32 mask,unsigned int flags,__u32 umask,int * destroy)1061 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
1062 					    __u32 mask, unsigned int flags,
1063 					    __u32 umask, int *destroy)
1064 {
1065 	__u32 oldmask, newmask;
1066 
1067 	/* umask bits cannot be removed by user */
1068 	mask &= ~umask;
1069 	spin_lock(&fsn_mark->lock);
1070 	oldmask = fsnotify_calc_mask(fsn_mark);
1071 	if (!(flags & FANOTIFY_MARK_IGNORE_BITS)) {
1072 		fsn_mark->mask &= ~mask;
1073 	} else {
1074 		fsn_mark->ignore_mask &= ~mask;
1075 	}
1076 	newmask = fsnotify_calc_mask(fsn_mark);
1077 	/*
1078 	 * We need to keep the mark around even if remaining mask cannot
1079 	 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
1080 	 * changes to the mask.
1081 	 * Destroy mark when only umask bits remain.
1082 	 */
1083 	*destroy = !((fsn_mark->mask | fsn_mark->ignore_mask) & ~umask);
1084 	spin_unlock(&fsn_mark->lock);
1085 
1086 	return oldmask & ~newmask;
1087 }
1088 
fanotify_remove_mark(struct fsnotify_group * group,void * obj,unsigned int obj_type,__u32 mask,unsigned int flags,__u32 umask)1089 static int fanotify_remove_mark(struct fsnotify_group *group,
1090 				void *obj, unsigned int obj_type, __u32 mask,
1091 				unsigned int flags, __u32 umask)
1092 {
1093 	struct fsnotify_mark *fsn_mark = NULL;
1094 	__u32 removed;
1095 	int destroy_mark;
1096 
1097 	fsnotify_group_lock(group);
1098 	fsn_mark = fsnotify_find_mark(obj, obj_type, group);
1099 	if (!fsn_mark) {
1100 		fsnotify_group_unlock(group);
1101 		return -ENOENT;
1102 	}
1103 
1104 	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
1105 						 umask, &destroy_mark);
1106 	if (removed & fsnotify_conn_mask(fsn_mark->connector))
1107 		fsnotify_recalc_mask(fsn_mark->connector);
1108 	if (destroy_mark)
1109 		fsnotify_detach_mark(fsn_mark);
1110 	fsnotify_group_unlock(group);
1111 	if (destroy_mark)
1112 		fsnotify_free_mark(fsn_mark);
1113 
1114 	/* matches the fsnotify_find_mark() */
1115 	fsnotify_put_mark(fsn_mark);
1116 	return 0;
1117 }
1118 
fanotify_mark_update_flags(struct fsnotify_mark * fsn_mark,unsigned int fan_flags)1119 static bool fanotify_mark_update_flags(struct fsnotify_mark *fsn_mark,
1120 				       unsigned int fan_flags)
1121 {
1122 	bool want_iref = !(fan_flags & FAN_MARK_EVICTABLE);
1123 	unsigned int ignore = fan_flags & FANOTIFY_MARK_IGNORE_BITS;
1124 	bool recalc = false;
1125 
1126 	/*
1127 	 * When using FAN_MARK_IGNORE for the first time, mark starts using
1128 	 * independent event flags in ignore mask.  After that, trying to
1129 	 * update the ignore mask with the old FAN_MARK_IGNORED_MASK API
1130 	 * will result in EEXIST error.
1131 	 */
1132 	if (ignore == FAN_MARK_IGNORE)
1133 		fsn_mark->flags |= FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS;
1134 
1135 	/*
1136 	 * Setting FAN_MARK_IGNORED_SURV_MODIFY for the first time may lead to
1137 	 * the removal of the FS_MODIFY bit in calculated mask if it was set
1138 	 * because of an ignore mask that is now going to survive FS_MODIFY.
1139 	 */
1140 	if (ignore && (fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1141 	    !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) {
1142 		fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
1143 		if (!(fsn_mark->mask & FS_MODIFY))
1144 			recalc = true;
1145 	}
1146 
1147 	if (fsn_mark->connector->type != FSNOTIFY_OBJ_TYPE_INODE ||
1148 	    want_iref == !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1149 		return recalc;
1150 
1151 	/*
1152 	 * NO_IREF may be removed from a mark, but not added.
1153 	 * When removed, fsnotify_recalc_mask() will take the inode ref.
1154 	 */
1155 	WARN_ON_ONCE(!want_iref);
1156 	fsn_mark->flags &= ~FSNOTIFY_MARK_FLAG_NO_IREF;
1157 
1158 	return true;
1159 }
1160 
fanotify_mark_add_to_mask(struct fsnotify_mark * fsn_mark,__u32 mask,unsigned int fan_flags)1161 static bool fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
1162 				      __u32 mask, unsigned int fan_flags)
1163 {
1164 	bool recalc;
1165 
1166 	spin_lock(&fsn_mark->lock);
1167 	if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS))
1168 		fsn_mark->mask |= mask;
1169 	else
1170 		fsn_mark->ignore_mask |= mask;
1171 
1172 	recalc = fsnotify_calc_mask(fsn_mark) &
1173 		~fsnotify_conn_mask(fsn_mark->connector);
1174 
1175 	recalc |= fanotify_mark_update_flags(fsn_mark, fan_flags);
1176 	spin_unlock(&fsn_mark->lock);
1177 
1178 	return recalc;
1179 }
1180 
1181 struct fan_fsid {
1182 	struct super_block *sb;
1183 	__kernel_fsid_t id;
1184 	bool weak;
1185 };
1186 
fanotify_set_mark_fsid(struct fsnotify_group * group,struct fsnotify_mark * mark,struct fan_fsid * fsid)1187 static int fanotify_set_mark_fsid(struct fsnotify_group *group,
1188 				  struct fsnotify_mark *mark,
1189 				  struct fan_fsid *fsid)
1190 {
1191 	struct fsnotify_mark_connector *conn;
1192 	struct fsnotify_mark *old;
1193 	struct super_block *old_sb = NULL;
1194 
1195 	FANOTIFY_MARK(mark)->fsid = fsid->id;
1196 	mark->flags |= FSNOTIFY_MARK_FLAG_HAS_FSID;
1197 	if (fsid->weak)
1198 		mark->flags |= FSNOTIFY_MARK_FLAG_WEAK_FSID;
1199 
1200 	/* First mark added will determine if group is single or multi fsid */
1201 	if (list_empty(&group->marks_list))
1202 		return 0;
1203 
1204 	/* Find sb of an existing mark */
1205 	list_for_each_entry(old, &group->marks_list, g_list) {
1206 		conn = READ_ONCE(old->connector);
1207 		if (!conn)
1208 			continue;
1209 		old_sb = fsnotify_connector_sb(conn);
1210 		if (old_sb)
1211 			break;
1212 	}
1213 
1214 	/* Only detached marks left? */
1215 	if (!old_sb)
1216 		return 0;
1217 
1218 	/* Do not allow mixing of marks with weak and strong fsid */
1219 	if ((mark->flags ^ old->flags) & FSNOTIFY_MARK_FLAG_WEAK_FSID)
1220 		return -EXDEV;
1221 
1222 	/* Allow mixing of marks with strong fsid from different fs */
1223 	if (!fsid->weak)
1224 		return 0;
1225 
1226 	/* Do not allow mixing marks with weak fsid from different fs */
1227 	if (old_sb != fsid->sb)
1228 		return -EXDEV;
1229 
1230 	/* Do not allow mixing marks from different btrfs sub-volumes */
1231 	if (!fanotify_fsid_equal(&FANOTIFY_MARK(old)->fsid,
1232 				 &FANOTIFY_MARK(mark)->fsid))
1233 		return -EXDEV;
1234 
1235 	return 0;
1236 }
1237 
fanotify_add_new_mark(struct fsnotify_group * group,void * obj,unsigned int obj_type,unsigned int fan_flags,struct fan_fsid * fsid)1238 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
1239 						   void *obj,
1240 						   unsigned int obj_type,
1241 						   unsigned int fan_flags,
1242 						   struct fan_fsid *fsid)
1243 {
1244 	struct ucounts *ucounts = group->fanotify_data.ucounts;
1245 	struct fanotify_mark *fan_mark;
1246 	struct fsnotify_mark *mark;
1247 	int ret;
1248 
1249 	/*
1250 	 * Enforce per user marks limits per user in all containing user ns.
1251 	 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
1252 	 * in the limited groups account.
1253 	 */
1254 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
1255 	    !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
1256 		return ERR_PTR(-ENOSPC);
1257 
1258 	fan_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
1259 	if (!fan_mark) {
1260 		ret = -ENOMEM;
1261 		goto out_dec_ucounts;
1262 	}
1263 
1264 	mark = &fan_mark->fsn_mark;
1265 	fsnotify_init_mark(mark, group);
1266 	if (fan_flags & FAN_MARK_EVICTABLE)
1267 		mark->flags |= FSNOTIFY_MARK_FLAG_NO_IREF;
1268 
1269 	/* Cache fsid of filesystem containing the marked object */
1270 	if (fsid) {
1271 		ret = fanotify_set_mark_fsid(group, mark, fsid);
1272 		if (ret)
1273 			goto out_put_mark;
1274 	} else {
1275 		fan_mark->fsid.val[0] = fan_mark->fsid.val[1] = 0;
1276 	}
1277 
1278 	ret = fsnotify_add_mark_locked(mark, obj, obj_type, 0);
1279 	if (ret)
1280 		goto out_put_mark;
1281 
1282 	return mark;
1283 
1284 out_put_mark:
1285 	fsnotify_put_mark(mark);
1286 out_dec_ucounts:
1287 	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1288 		dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
1289 	return ERR_PTR(ret);
1290 }
1291 
fanotify_group_init_error_pool(struct fsnotify_group * group)1292 static int fanotify_group_init_error_pool(struct fsnotify_group *group)
1293 {
1294 	if (mempool_initialized(&group->fanotify_data.error_events_pool))
1295 		return 0;
1296 
1297 	return mempool_init_kmalloc_pool(&group->fanotify_data.error_events_pool,
1298 					 FANOTIFY_DEFAULT_FEE_POOL_SIZE,
1299 					 sizeof(struct fanotify_error_event));
1300 }
1301 
fanotify_may_update_existing_mark(struct fsnotify_mark * fsn_mark,unsigned int fan_flags)1302 static int fanotify_may_update_existing_mark(struct fsnotify_mark *fsn_mark,
1303 					      unsigned int fan_flags)
1304 {
1305 	/*
1306 	 * Non evictable mark cannot be downgraded to evictable mark.
1307 	 */
1308 	if (fan_flags & FAN_MARK_EVICTABLE &&
1309 	    !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1310 		return -EEXIST;
1311 
1312 	/*
1313 	 * New ignore mask semantics cannot be downgraded to old semantics.
1314 	 */
1315 	if (fan_flags & FAN_MARK_IGNORED_MASK &&
1316 	    fsn_mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS)
1317 		return -EEXIST;
1318 
1319 	/*
1320 	 * An ignore mask that survives modify could never be downgraded to not
1321 	 * survive modify.  With new FAN_MARK_IGNORE semantics we make that rule
1322 	 * explicit and return an error when trying to update the ignore mask
1323 	 * without the original FAN_MARK_IGNORED_SURV_MODIFY value.
1324 	 */
1325 	if (fan_flags & FAN_MARK_IGNORE &&
1326 	    !(fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1327 	    fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)
1328 		return -EEXIST;
1329 
1330 	return 0;
1331 }
1332 
fanotify_add_mark(struct fsnotify_group * group,void * obj,unsigned int obj_type,__u32 mask,unsigned int fan_flags,struct fan_fsid * fsid)1333 static int fanotify_add_mark(struct fsnotify_group *group,
1334 			     void *obj, unsigned int obj_type,
1335 			     __u32 mask, unsigned int fan_flags,
1336 			     struct fan_fsid *fsid)
1337 {
1338 	struct fsnotify_mark *fsn_mark;
1339 	bool recalc;
1340 	int ret = 0;
1341 
1342 	fsnotify_group_lock(group);
1343 	fsn_mark = fsnotify_find_mark(obj, obj_type, group);
1344 	if (!fsn_mark) {
1345 		fsn_mark = fanotify_add_new_mark(group, obj, obj_type,
1346 						 fan_flags, fsid);
1347 		if (IS_ERR(fsn_mark)) {
1348 			fsnotify_group_unlock(group);
1349 			return PTR_ERR(fsn_mark);
1350 		}
1351 	}
1352 
1353 	/*
1354 	 * Check if requested mark flags conflict with an existing mark flags.
1355 	 */
1356 	ret = fanotify_may_update_existing_mark(fsn_mark, fan_flags);
1357 	if (ret)
1358 		goto out;
1359 
1360 	/*
1361 	 * Error events are pre-allocated per group, only if strictly
1362 	 * needed (i.e. FAN_FS_ERROR was requested).
1363 	 */
1364 	if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS) &&
1365 	    (mask & FAN_FS_ERROR)) {
1366 		ret = fanotify_group_init_error_pool(group);
1367 		if (ret)
1368 			goto out;
1369 	}
1370 
1371 	recalc = fanotify_mark_add_to_mask(fsn_mark, mask, fan_flags);
1372 	if (recalc)
1373 		fsnotify_recalc_mask(fsn_mark->connector);
1374 
1375 out:
1376 	fsnotify_group_unlock(group);
1377 
1378 	fsnotify_put_mark(fsn_mark);
1379 	return ret;
1380 }
1381 
fanotify_alloc_overflow_event(void)1382 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1383 {
1384 	struct fanotify_event *oevent;
1385 
1386 	oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1387 	if (!oevent)
1388 		return NULL;
1389 
1390 	fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1391 	oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1392 
1393 	return &oevent->fse;
1394 }
1395 
fanotify_alloc_merge_hash(void)1396 static struct hlist_head *fanotify_alloc_merge_hash(void)
1397 {
1398 	struct hlist_head *hash;
1399 
1400 	hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1401 		       GFP_KERNEL_ACCOUNT);
1402 	if (!hash)
1403 		return NULL;
1404 
1405 	__hash_init(hash, FANOTIFY_HTABLE_SIZE);
1406 
1407 	return hash;
1408 }
1409 
1410 /* fanotify syscalls */
SYSCALL_DEFINE2(fanotify_init,unsigned int,flags,unsigned int,event_f_flags)1411 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1412 {
1413 	struct fsnotify_group *group;
1414 	int f_flags, fd;
1415 	unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1416 	unsigned int class = flags & FANOTIFY_CLASS_BITS;
1417 	unsigned int internal_flags = 0;
1418 
1419 	pr_debug("%s: flags=%x event_f_flags=%x\n",
1420 		 __func__, flags, event_f_flags);
1421 
1422 	if (!capable(CAP_SYS_ADMIN)) {
1423 		/*
1424 		 * An unprivileged user can setup an fanotify group with
1425 		 * limited functionality - an unprivileged group is limited to
1426 		 * notification events with file handles and it cannot use
1427 		 * unlimited queue/marks.
1428 		 */
1429 		if ((flags & FANOTIFY_ADMIN_INIT_FLAGS) || !fid_mode)
1430 			return -EPERM;
1431 
1432 		/*
1433 		 * Setting the internal flag FANOTIFY_UNPRIV on the group
1434 		 * prevents setting mount/filesystem marks on this group and
1435 		 * prevents reporting pid and open fd in events.
1436 		 */
1437 		internal_flags |= FANOTIFY_UNPRIV;
1438 	}
1439 
1440 #ifdef CONFIG_AUDITSYSCALL
1441 	if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1442 #else
1443 	if (flags & ~FANOTIFY_INIT_FLAGS)
1444 #endif
1445 		return -EINVAL;
1446 
1447 	/*
1448 	 * A pidfd can only be returned for a thread-group leader; thus
1449 	 * FAN_REPORT_PIDFD and FAN_REPORT_TID need to remain mutually
1450 	 * exclusive.
1451 	 */
1452 	if ((flags & FAN_REPORT_PIDFD) && (flags & FAN_REPORT_TID))
1453 		return -EINVAL;
1454 
1455 	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1456 		return -EINVAL;
1457 
1458 	switch (event_f_flags & O_ACCMODE) {
1459 	case O_RDONLY:
1460 	case O_RDWR:
1461 	case O_WRONLY:
1462 		break;
1463 	default:
1464 		return -EINVAL;
1465 	}
1466 
1467 	if (fid_mode && class != FAN_CLASS_NOTIF)
1468 		return -EINVAL;
1469 
1470 	/*
1471 	 * Child name is reported with parent fid so requires dir fid.
1472 	 * We can report both child fid and dir fid with or without name.
1473 	 */
1474 	if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1475 		return -EINVAL;
1476 
1477 	/*
1478 	 * FAN_REPORT_TARGET_FID requires FAN_REPORT_NAME and FAN_REPORT_FID
1479 	 * and is used as an indication to report both dir and child fid on all
1480 	 * dirent events.
1481 	 */
1482 	if ((fid_mode & FAN_REPORT_TARGET_FID) &&
1483 	    (!(fid_mode & FAN_REPORT_NAME) || !(fid_mode & FAN_REPORT_FID)))
1484 		return -EINVAL;
1485 
1486 	f_flags = O_RDWR | __FMODE_NONOTIFY;
1487 	if (flags & FAN_CLOEXEC)
1488 		f_flags |= O_CLOEXEC;
1489 	if (flags & FAN_NONBLOCK)
1490 		f_flags |= O_NONBLOCK;
1491 
1492 	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
1493 	group = fsnotify_alloc_group(&fanotify_fsnotify_ops,
1494 				     FSNOTIFY_GROUP_USER);
1495 	if (IS_ERR(group)) {
1496 		return PTR_ERR(group);
1497 	}
1498 
1499 	/* Enforce groups limits per user in all containing user ns */
1500 	group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1501 						  current_euid(),
1502 						  UCOUNT_FANOTIFY_GROUPS);
1503 	if (!group->fanotify_data.ucounts) {
1504 		fd = -EMFILE;
1505 		goto out_destroy_group;
1506 	}
1507 
1508 	group->fanotify_data.flags = flags | internal_flags;
1509 	group->memcg = get_mem_cgroup_from_mm(current->mm);
1510 
1511 	group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1512 	if (!group->fanotify_data.merge_hash) {
1513 		fd = -ENOMEM;
1514 		goto out_destroy_group;
1515 	}
1516 
1517 	group->overflow_event = fanotify_alloc_overflow_event();
1518 	if (unlikely(!group->overflow_event)) {
1519 		fd = -ENOMEM;
1520 		goto out_destroy_group;
1521 	}
1522 
1523 	if (force_o_largefile())
1524 		event_f_flags |= O_LARGEFILE;
1525 	group->fanotify_data.f_flags = event_f_flags;
1526 	init_waitqueue_head(&group->fanotify_data.access_waitq);
1527 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
1528 	switch (class) {
1529 	case FAN_CLASS_NOTIF:
1530 		group->priority = FSNOTIFY_PRIO_NORMAL;
1531 		break;
1532 	case FAN_CLASS_CONTENT:
1533 		group->priority = FSNOTIFY_PRIO_CONTENT;
1534 		break;
1535 	case FAN_CLASS_PRE_CONTENT:
1536 		group->priority = FSNOTIFY_PRIO_PRE_CONTENT;
1537 		break;
1538 	default:
1539 		fd = -EINVAL;
1540 		goto out_destroy_group;
1541 	}
1542 
1543 	if (flags & FAN_UNLIMITED_QUEUE) {
1544 		fd = -EPERM;
1545 		if (!capable(CAP_SYS_ADMIN))
1546 			goto out_destroy_group;
1547 		group->max_events = UINT_MAX;
1548 	} else {
1549 		group->max_events = fanotify_max_queued_events;
1550 	}
1551 
1552 	if (flags & FAN_UNLIMITED_MARKS) {
1553 		fd = -EPERM;
1554 		if (!capable(CAP_SYS_ADMIN))
1555 			goto out_destroy_group;
1556 	}
1557 
1558 	if (flags & FAN_ENABLE_AUDIT) {
1559 		fd = -EPERM;
1560 		if (!capable(CAP_AUDIT_WRITE))
1561 			goto out_destroy_group;
1562 	}
1563 
1564 	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
1565 	if (fd < 0)
1566 		goto out_destroy_group;
1567 
1568 	return fd;
1569 
1570 out_destroy_group:
1571 	fsnotify_destroy_group(group);
1572 	return fd;
1573 }
1574 
fanotify_test_fsid(struct dentry * dentry,unsigned int flags,struct fan_fsid * fsid)1575 static int fanotify_test_fsid(struct dentry *dentry, unsigned int flags,
1576 			      struct fan_fsid *fsid)
1577 {
1578 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1579 	__kernel_fsid_t root_fsid;
1580 	int err;
1581 
1582 	/*
1583 	 * Make sure dentry is not of a filesystem with zero fsid (e.g. fuse).
1584 	 */
1585 	err = vfs_get_fsid(dentry, &fsid->id);
1586 	if (err)
1587 		return err;
1588 
1589 	fsid->sb = dentry->d_sb;
1590 	if (!fsid->id.val[0] && !fsid->id.val[1]) {
1591 		err = -ENODEV;
1592 		goto weak;
1593 	}
1594 
1595 	/*
1596 	 * Make sure dentry is not of a filesystem subvolume (e.g. btrfs)
1597 	 * which uses a different fsid than sb root.
1598 	 */
1599 	err = vfs_get_fsid(dentry->d_sb->s_root, &root_fsid);
1600 	if (err)
1601 		return err;
1602 
1603 	if (!fanotify_fsid_equal(&root_fsid, &fsid->id)) {
1604 		err = -EXDEV;
1605 		goto weak;
1606 	}
1607 
1608 	fsid->weak = false;
1609 	return 0;
1610 
1611 weak:
1612 	/* Allow weak fsid when marking inodes */
1613 	fsid->weak = true;
1614 	return (mark_type == FAN_MARK_INODE) ? 0 : err;
1615 }
1616 
1617 /* Check if filesystem can encode a unique fid */
fanotify_test_fid(struct dentry * dentry,unsigned int flags)1618 static int fanotify_test_fid(struct dentry *dentry, unsigned int flags)
1619 {
1620 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1621 	const struct export_operations *nop = dentry->d_sb->s_export_op;
1622 
1623 	/*
1624 	 * We need to make sure that the filesystem supports encoding of
1625 	 * file handles so user can use name_to_handle_at() to compare fids
1626 	 * reported with events to the file handle of watched objects.
1627 	 */
1628 	if (!exportfs_can_encode_fid(nop))
1629 		return -EOPNOTSUPP;
1630 
1631 	/*
1632 	 * For sb/mount mark, we also need to make sure that the filesystem
1633 	 * supports decoding file handles, so user has a way to map back the
1634 	 * reported fids to filesystem objects.
1635 	 */
1636 	if (mark_type != FAN_MARK_INODE && !exportfs_can_decode_fh(nop))
1637 		return -EOPNOTSUPP;
1638 
1639 	return 0;
1640 }
1641 
fanotify_events_supported(struct fsnotify_group * group,const struct path * path,__u64 mask,unsigned int flags)1642 static int fanotify_events_supported(struct fsnotify_group *group,
1643 				     const struct path *path, __u64 mask,
1644 				     unsigned int flags)
1645 {
1646 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1647 	/* Strict validation of events in non-dir inode mask with v5.17+ APIs */
1648 	bool strict_dir_events = FAN_GROUP_FLAG(group, FAN_REPORT_TARGET_FID) ||
1649 				 (mask & FAN_RENAME) ||
1650 				 (flags & FAN_MARK_IGNORE);
1651 
1652 	/*
1653 	 * Some filesystems such as 'proc' acquire unusual locks when opening
1654 	 * files. For them fanotify permission events have high chances of
1655 	 * deadlocking the system - open done when reporting fanotify event
1656 	 * blocks on this "unusual" lock while another process holding the lock
1657 	 * waits for fanotify permission event to be answered. Just disallow
1658 	 * permission events for such filesystems.
1659 	 */
1660 	if (mask & FANOTIFY_PERM_EVENTS &&
1661 	    path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1662 		return -EINVAL;
1663 
1664 	/*
1665 	 * mount and sb marks are not allowed on kernel internal pseudo fs,
1666 	 * like pipe_mnt, because that would subscribe to events on all the
1667 	 * anonynous pipes in the system.
1668 	 *
1669 	 * SB_NOUSER covers all of the internal pseudo fs whose objects are not
1670 	 * exposed to user's mount namespace, but there are other SB_KERNMOUNT
1671 	 * fs, like nsfs, debugfs, for which the value of allowing sb and mount
1672 	 * mark is questionable. For now we leave them alone.
1673 	 */
1674 	if (mark_type != FAN_MARK_INODE &&
1675 	    path->mnt->mnt_sb->s_flags & SB_NOUSER)
1676 		return -EINVAL;
1677 
1678 	/*
1679 	 * We shouldn't have allowed setting dirent events and the directory
1680 	 * flags FAN_ONDIR and FAN_EVENT_ON_CHILD in mask of non-dir inode,
1681 	 * but because we always allowed it, error only when using new APIs.
1682 	 */
1683 	if (strict_dir_events && mark_type == FAN_MARK_INODE &&
1684 	    !d_is_dir(path->dentry) && (mask & FANOTIFY_DIRONLY_EVENT_BITS))
1685 		return -ENOTDIR;
1686 
1687 	return 0;
1688 }
1689 
do_fanotify_mark(int fanotify_fd,unsigned int flags,__u64 mask,int dfd,const char __user * pathname)1690 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1691 			    int dfd, const char  __user *pathname)
1692 {
1693 	struct inode *inode = NULL;
1694 	struct vfsmount *mnt = NULL;
1695 	struct fsnotify_group *group;
1696 	struct fd f;
1697 	struct path path;
1698 	struct fan_fsid __fsid, *fsid = NULL;
1699 	u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1700 	unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1701 	unsigned int mark_cmd = flags & FANOTIFY_MARK_CMD_BITS;
1702 	unsigned int ignore = flags & FANOTIFY_MARK_IGNORE_BITS;
1703 	unsigned int obj_type, fid_mode;
1704 	void *obj;
1705 	u32 umask = 0;
1706 	int ret;
1707 
1708 	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1709 		 __func__, fanotify_fd, flags, dfd, pathname, mask);
1710 
1711 	/* we only use the lower 32 bits as of right now. */
1712 	if (upper_32_bits(mask))
1713 		return -EINVAL;
1714 
1715 	if (flags & ~FANOTIFY_MARK_FLAGS)
1716 		return -EINVAL;
1717 
1718 	switch (mark_type) {
1719 	case FAN_MARK_INODE:
1720 		obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1721 		break;
1722 	case FAN_MARK_MOUNT:
1723 		obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1724 		break;
1725 	case FAN_MARK_FILESYSTEM:
1726 		obj_type = FSNOTIFY_OBJ_TYPE_SB;
1727 		break;
1728 	default:
1729 		return -EINVAL;
1730 	}
1731 
1732 	switch (mark_cmd) {
1733 	case FAN_MARK_ADD:
1734 	case FAN_MARK_REMOVE:
1735 		if (!mask)
1736 			return -EINVAL;
1737 		break;
1738 	case FAN_MARK_FLUSH:
1739 		if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1740 			return -EINVAL;
1741 		break;
1742 	default:
1743 		return -EINVAL;
1744 	}
1745 
1746 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1747 		valid_mask |= FANOTIFY_PERM_EVENTS;
1748 
1749 	if (mask & ~valid_mask)
1750 		return -EINVAL;
1751 
1752 
1753 	/* We don't allow FAN_MARK_IGNORE & FAN_MARK_IGNORED_MASK together */
1754 	if (ignore == (FAN_MARK_IGNORE | FAN_MARK_IGNORED_MASK))
1755 		return -EINVAL;
1756 
1757 	/*
1758 	 * Event flags (FAN_ONDIR, FAN_EVENT_ON_CHILD) have no effect with
1759 	 * FAN_MARK_IGNORED_MASK.
1760 	 */
1761 	if (ignore == FAN_MARK_IGNORED_MASK) {
1762 		mask &= ~FANOTIFY_EVENT_FLAGS;
1763 		umask = FANOTIFY_EVENT_FLAGS;
1764 	}
1765 
1766 	f = fdget(fanotify_fd);
1767 	if (unlikely(!fd_file(f)))
1768 		return -EBADF;
1769 
1770 	/* verify that this is indeed an fanotify instance */
1771 	ret = -EINVAL;
1772 	if (unlikely(fd_file(f)->f_op != &fanotify_fops))
1773 		goto fput_and_out;
1774 	group = fd_file(f)->private_data;
1775 
1776 	/*
1777 	 * An unprivileged user is not allowed to setup mount nor filesystem
1778 	 * marks.  This also includes setting up such marks by a group that
1779 	 * was initialized by an unprivileged user.
1780 	 */
1781 	ret = -EPERM;
1782 	if ((!capable(CAP_SYS_ADMIN) ||
1783 	     FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV)) &&
1784 	    mark_type != FAN_MARK_INODE)
1785 		goto fput_and_out;
1786 
1787 	/*
1788 	 * Permission events require minimum priority FAN_CLASS_CONTENT.
1789 	 */
1790 	ret = -EINVAL;
1791 	if (mask & FANOTIFY_PERM_EVENTS &&
1792 	    group->priority < FSNOTIFY_PRIO_CONTENT)
1793 		goto fput_and_out;
1794 
1795 	if (mask & FAN_FS_ERROR &&
1796 	    mark_type != FAN_MARK_FILESYSTEM)
1797 		goto fput_and_out;
1798 
1799 	/*
1800 	 * Evictable is only relevant for inode marks, because only inode object
1801 	 * can be evicted on memory pressure.
1802 	 */
1803 	if (flags & FAN_MARK_EVICTABLE &&
1804 	     mark_type != FAN_MARK_INODE)
1805 		goto fput_and_out;
1806 
1807 	/*
1808 	 * Events that do not carry enough information to report
1809 	 * event->fd require a group that supports reporting fid.  Those
1810 	 * events are not supported on a mount mark, because they do not
1811 	 * carry enough information (i.e. path) to be filtered by mount
1812 	 * point.
1813 	 */
1814 	fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1815 	if (mask & ~(FANOTIFY_FD_EVENTS|FANOTIFY_EVENT_FLAGS) &&
1816 	    (!fid_mode || mark_type == FAN_MARK_MOUNT))
1817 		goto fput_and_out;
1818 
1819 	/*
1820 	 * FAN_RENAME uses special info type records to report the old and
1821 	 * new parent+name.  Reporting only old and new parent id is less
1822 	 * useful and was not implemented.
1823 	 */
1824 	if (mask & FAN_RENAME && !(fid_mode & FAN_REPORT_NAME))
1825 		goto fput_and_out;
1826 
1827 	if (mark_cmd == FAN_MARK_FLUSH) {
1828 		ret = 0;
1829 		if (mark_type == FAN_MARK_MOUNT)
1830 			fsnotify_clear_vfsmount_marks_by_group(group);
1831 		else if (mark_type == FAN_MARK_FILESYSTEM)
1832 			fsnotify_clear_sb_marks_by_group(group);
1833 		else
1834 			fsnotify_clear_inode_marks_by_group(group);
1835 		goto fput_and_out;
1836 	}
1837 
1838 	ret = fanotify_find_path(dfd, pathname, &path, flags,
1839 			(mask & ALL_FSNOTIFY_EVENTS), obj_type);
1840 	if (ret)
1841 		goto fput_and_out;
1842 
1843 	if (mark_cmd == FAN_MARK_ADD) {
1844 		ret = fanotify_events_supported(group, &path, mask, flags);
1845 		if (ret)
1846 			goto path_put_and_out;
1847 	}
1848 
1849 	if (fid_mode) {
1850 		ret = fanotify_test_fsid(path.dentry, flags, &__fsid);
1851 		if (ret)
1852 			goto path_put_and_out;
1853 
1854 		ret = fanotify_test_fid(path.dentry, flags);
1855 		if (ret)
1856 			goto path_put_and_out;
1857 
1858 		fsid = &__fsid;
1859 	}
1860 
1861 	/* inode held in place by reference to path; group by fget on fd */
1862 	if (mark_type == FAN_MARK_INODE) {
1863 		inode = path.dentry->d_inode;
1864 		obj = inode;
1865 	} else {
1866 		mnt = path.mnt;
1867 		if (mark_type == FAN_MARK_MOUNT)
1868 			obj = mnt;
1869 		else
1870 			obj = mnt->mnt_sb;
1871 	}
1872 
1873 	/*
1874 	 * If some other task has this inode open for write we should not add
1875 	 * an ignore mask, unless that ignore mask is supposed to survive
1876 	 * modification changes anyway.
1877 	 */
1878 	if (mark_cmd == FAN_MARK_ADD && (flags & FANOTIFY_MARK_IGNORE_BITS) &&
1879 	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY)) {
1880 		ret = mnt ? -EINVAL : -EISDIR;
1881 		/* FAN_MARK_IGNORE requires SURV_MODIFY for sb/mount/dir marks */
1882 		if (ignore == FAN_MARK_IGNORE &&
1883 		    (mnt || S_ISDIR(inode->i_mode)))
1884 			goto path_put_and_out;
1885 
1886 		ret = 0;
1887 		if (inode && inode_is_open_for_write(inode))
1888 			goto path_put_and_out;
1889 	}
1890 
1891 	/* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1892 	if (mnt || !S_ISDIR(inode->i_mode)) {
1893 		mask &= ~FAN_EVENT_ON_CHILD;
1894 		umask = FAN_EVENT_ON_CHILD;
1895 		/*
1896 		 * If group needs to report parent fid, register for getting
1897 		 * events with parent/name info for non-directory.
1898 		 */
1899 		if ((fid_mode & FAN_REPORT_DIR_FID) &&
1900 		    (flags & FAN_MARK_ADD) && !ignore)
1901 			mask |= FAN_EVENT_ON_CHILD;
1902 	}
1903 
1904 	/* create/update an inode mark */
1905 	switch (mark_cmd) {
1906 	case FAN_MARK_ADD:
1907 		ret = fanotify_add_mark(group, obj, obj_type, mask, flags,
1908 					fsid);
1909 		break;
1910 	case FAN_MARK_REMOVE:
1911 		ret = fanotify_remove_mark(group, obj, obj_type, mask, flags,
1912 					   umask);
1913 		break;
1914 	default:
1915 		ret = -EINVAL;
1916 	}
1917 
1918 path_put_and_out:
1919 	path_put(&path);
1920 fput_and_out:
1921 	fdput(f);
1922 	return ret;
1923 }
1924 
1925 #ifndef CONFIG_ARCH_SPLIT_ARG64
SYSCALL_DEFINE5(fanotify_mark,int,fanotify_fd,unsigned int,flags,__u64,mask,int,dfd,const char __user *,pathname)1926 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1927 			      __u64, mask, int, dfd,
1928 			      const char  __user *, pathname)
1929 {
1930 	return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1931 }
1932 #endif
1933 
1934 #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT)
SYSCALL32_DEFINE6(fanotify_mark,int,fanotify_fd,unsigned int,flags,SC_ARG64 (mask),int,dfd,const char __user *,pathname)1935 SYSCALL32_DEFINE6(fanotify_mark,
1936 				int, fanotify_fd, unsigned int, flags,
1937 				SC_ARG64(mask), int, dfd,
1938 				const char  __user *, pathname)
1939 {
1940 	return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
1941 				dfd, pathname);
1942 }
1943 #endif
1944 
1945 /*
1946  * fanotify_user_setup - Our initialization function.  Note that we cannot return
1947  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
1948  * must result in panic().
1949  */
fanotify_user_setup(void)1950 static int __init fanotify_user_setup(void)
1951 {
1952 	struct sysinfo si;
1953 	int max_marks;
1954 
1955 	si_meminfo(&si);
1956 	/*
1957 	 * Allow up to 1% of addressable memory to be accounted for per user
1958 	 * marks limited to the range [8192, 1048576]. mount and sb marks are
1959 	 * a lot cheaper than inode marks, but there is no reason for a user
1960 	 * to have many of those, so calculate by the cost of inode marks.
1961 	 */
1962 	max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
1963 		    INODE_MARK_COST;
1964 	max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
1965 				     FANOTIFY_DEFAULT_MAX_USER_MARKS);
1966 
1967 	BUILD_BUG_ON(FANOTIFY_INIT_FLAGS & FANOTIFY_INTERNAL_GROUP_FLAGS);
1968 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 13);
1969 	BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 11);
1970 
1971 	fanotify_mark_cache = KMEM_CACHE(fanotify_mark,
1972 					 SLAB_PANIC|SLAB_ACCOUNT);
1973 	fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
1974 					       SLAB_PANIC);
1975 	fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
1976 						SLAB_PANIC);
1977 	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
1978 		fanotify_perm_event_cachep =
1979 			KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
1980 	}
1981 
1982 	fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
1983 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
1984 					FANOTIFY_DEFAULT_MAX_GROUPS;
1985 	init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
1986 	fanotify_sysctls_init();
1987 
1988 	return 0;
1989 }
1990 device_initcall(fanotify_user_setup);
1991