1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/fanotify.h>
3 #include <linux/fcntl.h>
4 #include <linux/file.h>
5 #include <linux/fs.h>
6 #include <linux/anon_inodes.h>
7 #include <linux/fsnotify_backend.h>
8 #include <linux/init.h>
9 #include <linux/mount.h>
10 #include <linux/namei.h>
11 #include <linux/poll.h>
12 #include <linux/security.h>
13 #include <linux/syscalls.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <linux/uaccess.h>
17 #include <linux/compat.h>
18 #include <linux/sched/signal.h>
19 #include <linux/memcontrol.h>
20 #include <linux/statfs.h>
21 #include <linux/exportfs.h>
22
23 #include <asm/ioctls.h>
24
25 #include "../fsnotify.h"
26 #include "../fdinfo.h"
27 #include "fanotify.h"
28
29 #define FANOTIFY_DEFAULT_MAX_EVENTS 16384
30 #define FANOTIFY_OLD_DEFAULT_MAX_MARKS 8192
31 #define FANOTIFY_DEFAULT_MAX_GROUPS 128
32 #define FANOTIFY_DEFAULT_FEE_POOL_SIZE 32
33
34 /*
35 * Legacy fanotify marks limits (8192) is per group and we introduced a tunable
36 * limit of marks per user, similar to inotify. Effectively, the legacy limit
37 * of fanotify marks per user is <max marks per group> * <max groups per user>.
38 * This default limit (1M) also happens to match the increased limit of inotify
39 * max_user_watches since v5.10.
40 */
41 #define FANOTIFY_DEFAULT_MAX_USER_MARKS \
42 (FANOTIFY_OLD_DEFAULT_MAX_MARKS * FANOTIFY_DEFAULT_MAX_GROUPS)
43
44 /*
45 * Most of the memory cost of adding an inode mark is pinning the marked inode.
46 * The size of the filesystem inode struct is not uniform across filesystems,
47 * so double the size of a VFS inode is used as a conservative approximation.
48 */
49 #define INODE_MARK_COST (2 * sizeof(struct inode))
50
51 /* configurable via /proc/sys/fs/fanotify/ */
52 static int fanotify_max_queued_events __read_mostly;
53
54 #ifdef CONFIG_SYSCTL
55
56 #include <linux/sysctl.h>
57
58 static long ft_zero = 0;
59 static long ft_int_max = INT_MAX;
60
61 static const struct ctl_table fanotify_table[] = {
62 {
63 .procname = "max_user_groups",
64 .data = &init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS],
65 .maxlen = sizeof(long),
66 .mode = 0644,
67 .proc_handler = proc_doulongvec_minmax,
68 .extra1 = &ft_zero,
69 .extra2 = &ft_int_max,
70 },
71 {
72 .procname = "max_user_marks",
73 .data = &init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS],
74 .maxlen = sizeof(long),
75 .mode = 0644,
76 .proc_handler = proc_doulongvec_minmax,
77 .extra1 = &ft_zero,
78 .extra2 = &ft_int_max,
79 },
80 {
81 .procname = "max_queued_events",
82 .data = &fanotify_max_queued_events,
83 .maxlen = sizeof(int),
84 .mode = 0644,
85 .proc_handler = proc_dointvec_minmax,
86 .extra1 = SYSCTL_ZERO
87 },
88 };
89
fanotify_sysctls_init(void)90 static void __init fanotify_sysctls_init(void)
91 {
92 register_sysctl("fs/fanotify", fanotify_table);
93 }
94 #else
95 #define fanotify_sysctls_init() do { } while (0)
96 #endif /* CONFIG_SYSCTL */
97
98 /*
99 * All flags that may be specified in parameter event_f_flags of fanotify_init.
100 *
101 * Internal and external open flags are stored together in field f_flags of
102 * struct file. Only external open flags shall be allowed in event_f_flags.
103 * Internal flags like FMODE_EXEC shall be excluded.
104 */
105 #define FANOTIFY_INIT_ALL_EVENT_F_BITS ( \
106 O_ACCMODE | O_APPEND | O_NONBLOCK | \
107 __O_SYNC | O_DSYNC | O_CLOEXEC | \
108 O_LARGEFILE | O_NOATIME )
109
110 extern const struct fsnotify_ops fanotify_fsnotify_ops;
111
112 struct kmem_cache *fanotify_mark_cache __ro_after_init;
113 struct kmem_cache *fanotify_fid_event_cachep __ro_after_init;
114 struct kmem_cache *fanotify_path_event_cachep __ro_after_init;
115 struct kmem_cache *fanotify_perm_event_cachep __ro_after_init;
116
117 #define FANOTIFY_EVENT_ALIGN 4
118 #define FANOTIFY_FID_INFO_HDR_LEN \
119 (sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle))
120 #define FANOTIFY_PIDFD_INFO_LEN \
121 sizeof(struct fanotify_event_info_pidfd)
122 #define FANOTIFY_ERROR_INFO_LEN \
123 (sizeof(struct fanotify_event_info_error))
124 #define FANOTIFY_RANGE_INFO_LEN \
125 (sizeof(struct fanotify_event_info_range))
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 (fanotify_is_error_event(event->mask))
164 event_len += FANOTIFY_ERROR_INFO_LEN;
165
166 if (fanotify_event_has_any_dir_fh(event)) {
167 event_len += fanotify_dir_name_info_len(event);
168 } else if ((info_mode & FAN_REPORT_NAME) &&
169 (event->mask & FAN_ONDIR)) {
170 /*
171 * With group flag FAN_REPORT_NAME, if name was not recorded in
172 * event on a directory, we will report the name ".".
173 */
174 dot_len = 1;
175 }
176
177 if (fanotify_event_has_object_fh(event)) {
178 fh_len = fanotify_event_object_fh_len(event);
179 event_len += fanotify_fid_info_len(fh_len, dot_len);
180 }
181
182 if (info_mode & FAN_REPORT_PIDFD)
183 event_len += FANOTIFY_PIDFD_INFO_LEN;
184
185 if (fanotify_event_has_access_range(event))
186 event_len += FANOTIFY_RANGE_INFO_LEN;
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 provide an fd for the userspace program, so it could access the
263 * file without generating fanotify events itself.
264 */
265 new_file = dentry_open_nonotify(path, group->fanotify_data.f_flags,
266 current_cred());
267 if (IS_ERR(new_file)) {
268 put_unused_fd(client_fd);
269 client_fd = PTR_ERR(new_file);
270 } else {
271 *file = new_file;
272 }
273
274 return client_fd;
275 }
276
process_access_response_info(const char __user * info,size_t info_len,struct fanotify_response_info_audit_rule * friar)277 static int process_access_response_info(const char __user *info,
278 size_t info_len,
279 struct fanotify_response_info_audit_rule *friar)
280 {
281 if (info_len != sizeof(*friar))
282 return -EINVAL;
283
284 if (copy_from_user(friar, info, sizeof(*friar)))
285 return -EFAULT;
286
287 if (friar->hdr.type != FAN_RESPONSE_INFO_AUDIT_RULE)
288 return -EINVAL;
289 if (friar->hdr.pad != 0)
290 return -EINVAL;
291 if (friar->hdr.len != sizeof(*friar))
292 return -EINVAL;
293
294 return info_len;
295 }
296
297 /*
298 * Finish processing of permission event by setting it to ANSWERED state and
299 * drop group->notification_lock.
300 */
finish_permission_event(struct fsnotify_group * group,struct fanotify_perm_event * event,u32 response,struct fanotify_response_info_audit_rule * friar)301 static void finish_permission_event(struct fsnotify_group *group,
302 struct fanotify_perm_event *event, u32 response,
303 struct fanotify_response_info_audit_rule *friar)
304 __releases(&group->notification_lock)
305 {
306 bool destroy = false;
307
308 assert_spin_locked(&group->notification_lock);
309 event->response = response & ~FAN_INFO;
310 if (response & FAN_INFO)
311 memcpy(&event->audit_rule, friar, sizeof(*friar));
312
313 if (event->state == FAN_EVENT_CANCELED)
314 destroy = true;
315 else
316 event->state = FAN_EVENT_ANSWERED;
317 spin_unlock(&group->notification_lock);
318 if (destroy)
319 fsnotify_destroy_event(group, &event->fae.fse);
320 }
321
process_access_response(struct fsnotify_group * group,struct fanotify_response * response_struct,const char __user * info,size_t info_len)322 static int process_access_response(struct fsnotify_group *group,
323 struct fanotify_response *response_struct,
324 const char __user *info,
325 size_t info_len)
326 {
327 struct fanotify_perm_event *event;
328 int fd = response_struct->fd;
329 u32 response = response_struct->response;
330 int errno = fanotify_get_response_errno(response);
331 int ret = info_len;
332 struct fanotify_response_info_audit_rule friar;
333
334 pr_debug("%s: group=%p fd=%d response=%x errno=%d buf=%p size=%zu\n",
335 __func__, group, fd, response, errno, 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 if (errno)
347 return -EINVAL;
348 break;
349 case FAN_DENY:
350 /* Custom errno is supported only for pre-content groups */
351 if (errno && group->priority != FSNOTIFY_PRIO_PRE_CONTENT)
352 return -EINVAL;
353
354 /*
355 * Limit errno to values expected on open(2)/read(2)/write(2)
356 * of regular files.
357 */
358 switch (errno) {
359 case 0:
360 case EIO:
361 case EPERM:
362 case EBUSY:
363 case ETXTBSY:
364 case EAGAIN:
365 case ENOSPC:
366 case EDQUOT:
367 break;
368 default:
369 return -EINVAL;
370 }
371 break;
372 default:
373 return -EINVAL;
374 }
375
376 if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
377 return -EINVAL;
378
379 if (response & FAN_INFO) {
380 ret = process_access_response_info(info, info_len, &friar);
381 if (ret < 0)
382 return ret;
383 if (fd == FAN_NOFD)
384 return ret;
385 } else {
386 ret = 0;
387 }
388
389 if (fd < 0)
390 return -EINVAL;
391
392 spin_lock(&group->notification_lock);
393 list_for_each_entry(event, &group->fanotify_data.access_list,
394 fae.fse.list) {
395 if (event->fd != fd)
396 continue;
397
398 list_del_init(&event->fae.fse.list);
399 finish_permission_event(group, event, response, &friar);
400 wake_up(&group->fanotify_data.access_waitq);
401 return ret;
402 }
403 spin_unlock(&group->notification_lock);
404
405 return -ENOENT;
406 }
407
copy_error_info_to_user(struct fanotify_event * event,char __user * buf,int count)408 static size_t copy_error_info_to_user(struct fanotify_event *event,
409 char __user *buf, int count)
410 {
411 struct fanotify_event_info_error info = { };
412 struct fanotify_error_event *fee = FANOTIFY_EE(event);
413
414 info.hdr.info_type = FAN_EVENT_INFO_TYPE_ERROR;
415 info.hdr.len = FANOTIFY_ERROR_INFO_LEN;
416
417 if (WARN_ON(count < info.hdr.len))
418 return -EFAULT;
419
420 info.error = fee->error;
421 info.error_count = fee->err_count;
422
423 if (copy_to_user(buf, &info, sizeof(info)))
424 return -EFAULT;
425
426 return info.hdr.len;
427 }
428
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)429 static int copy_fid_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
430 int info_type, const char *name,
431 size_t name_len,
432 char __user *buf, size_t count)
433 {
434 struct fanotify_event_info_fid info = { };
435 struct file_handle handle = { };
436 unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
437 size_t fh_len = fh ? fh->len : 0;
438 size_t info_len = fanotify_fid_info_len(fh_len, name_len);
439 size_t len = info_len;
440
441 pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
442 __func__, fh_len, name_len, info_len, count);
443
444 if (WARN_ON_ONCE(len < sizeof(info) || len > count))
445 return -EFAULT;
446
447 /*
448 * Copy event info fid header followed by variable sized file handle
449 * and optionally followed by variable sized filename.
450 */
451 switch (info_type) {
452 case FAN_EVENT_INFO_TYPE_FID:
453 case FAN_EVENT_INFO_TYPE_DFID:
454 if (WARN_ON_ONCE(name_len))
455 return -EFAULT;
456 break;
457 case FAN_EVENT_INFO_TYPE_DFID_NAME:
458 case FAN_EVENT_INFO_TYPE_OLD_DFID_NAME:
459 case FAN_EVENT_INFO_TYPE_NEW_DFID_NAME:
460 if (WARN_ON_ONCE(!name || !name_len))
461 return -EFAULT;
462 break;
463 default:
464 return -EFAULT;
465 }
466
467 info.hdr.info_type = info_type;
468 info.hdr.len = len;
469 info.fsid = *fsid;
470 if (copy_to_user(buf, &info, sizeof(info)))
471 return -EFAULT;
472
473 buf += sizeof(info);
474 len -= sizeof(info);
475 if (WARN_ON_ONCE(len < sizeof(handle)))
476 return -EFAULT;
477
478 handle.handle_type = fh->type;
479 handle.handle_bytes = fh_len;
480
481 /* Mangle handle_type for bad file_handle */
482 if (!fh_len)
483 handle.handle_type = FILEID_INVALID;
484
485 if (copy_to_user(buf, &handle, sizeof(handle)))
486 return -EFAULT;
487
488 buf += sizeof(handle);
489 len -= sizeof(handle);
490 if (WARN_ON_ONCE(len < fh_len))
491 return -EFAULT;
492
493 /*
494 * For an inline fh and inline file name, copy through stack to exclude
495 * the copy from usercopy hardening protections.
496 */
497 fh_buf = fanotify_fh_buf(fh);
498 if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
499 memcpy(bounce, fh_buf, fh_len);
500 fh_buf = bounce;
501 }
502 if (copy_to_user(buf, fh_buf, fh_len))
503 return -EFAULT;
504
505 buf += fh_len;
506 len -= fh_len;
507
508 if (name_len) {
509 /* Copy the filename with terminating null */
510 name_len++;
511 if (WARN_ON_ONCE(len < name_len))
512 return -EFAULT;
513
514 if (copy_to_user(buf, name, name_len))
515 return -EFAULT;
516
517 buf += name_len;
518 len -= name_len;
519 }
520
521 /* Pad with 0's */
522 WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
523 if (len > 0 && clear_user(buf, len))
524 return -EFAULT;
525
526 return info_len;
527 }
528
copy_pidfd_info_to_user(int pidfd,char __user * buf,size_t count)529 static int copy_pidfd_info_to_user(int pidfd,
530 char __user *buf,
531 size_t count)
532 {
533 struct fanotify_event_info_pidfd info = { };
534 size_t info_len = FANOTIFY_PIDFD_INFO_LEN;
535
536 if (WARN_ON_ONCE(info_len > count))
537 return -EFAULT;
538
539 info.hdr.info_type = FAN_EVENT_INFO_TYPE_PIDFD;
540 info.hdr.len = info_len;
541 info.pidfd = pidfd;
542
543 if (copy_to_user(buf, &info, info_len))
544 return -EFAULT;
545
546 return info_len;
547 }
548
copy_range_info_to_user(struct fanotify_event * event,char __user * buf,int count)549 static size_t copy_range_info_to_user(struct fanotify_event *event,
550 char __user *buf, int count)
551 {
552 struct fanotify_perm_event *pevent = FANOTIFY_PERM(event);
553 struct fanotify_event_info_range info = { };
554 size_t info_len = FANOTIFY_RANGE_INFO_LEN;
555
556 if (WARN_ON_ONCE(info_len > count))
557 return -EFAULT;
558
559 if (WARN_ON_ONCE(!pevent->ppos))
560 return -EINVAL;
561
562 info.hdr.info_type = FAN_EVENT_INFO_TYPE_RANGE;
563 info.hdr.len = info_len;
564 info.offset = *(pevent->ppos);
565 info.count = pevent->count;
566
567 if (copy_to_user(buf, &info, info_len))
568 return -EFAULT;
569
570 return info_len;
571 }
572
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)573 static int copy_info_records_to_user(struct fanotify_event *event,
574 struct fanotify_info *info,
575 unsigned int info_mode, int pidfd,
576 char __user *buf, size_t count)
577 {
578 int ret, total_bytes = 0, info_type = 0;
579 unsigned int fid_mode = info_mode & FANOTIFY_FID_BITS;
580 unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
581
582 /*
583 * Event info records order is as follows:
584 * 1. dir fid + name
585 * 2. (optional) new dir fid + new name
586 * 3. (optional) child fid
587 */
588 if (fanotify_event_has_dir_fh(event)) {
589 info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
590 FAN_EVENT_INFO_TYPE_DFID;
591
592 /* FAN_RENAME uses special info types */
593 if (event->mask & FAN_RENAME)
594 info_type = FAN_EVENT_INFO_TYPE_OLD_DFID_NAME;
595
596 ret = copy_fid_info_to_user(fanotify_event_fsid(event),
597 fanotify_info_dir_fh(info),
598 info_type,
599 fanotify_info_name(info),
600 info->name_len, buf, count);
601 if (ret < 0)
602 return ret;
603
604 buf += ret;
605 count -= ret;
606 total_bytes += ret;
607 }
608
609 /* New dir fid+name may be reported in addition to old dir fid+name */
610 if (fanotify_event_has_dir2_fh(event)) {
611 info_type = FAN_EVENT_INFO_TYPE_NEW_DFID_NAME;
612 ret = copy_fid_info_to_user(fanotify_event_fsid(event),
613 fanotify_info_dir2_fh(info),
614 info_type,
615 fanotify_info_name2(info),
616 info->name2_len, buf, count);
617 if (ret < 0)
618 return ret;
619
620 buf += ret;
621 count -= ret;
622 total_bytes += ret;
623 }
624
625 if (fanotify_event_has_object_fh(event)) {
626 const char *dot = NULL;
627 int dot_len = 0;
628
629 if (fid_mode == FAN_REPORT_FID || info_type) {
630 /*
631 * With only group flag FAN_REPORT_FID only type FID is
632 * reported. Second info record type is always FID.
633 */
634 info_type = FAN_EVENT_INFO_TYPE_FID;
635 } else if ((fid_mode & FAN_REPORT_NAME) &&
636 (event->mask & FAN_ONDIR)) {
637 /*
638 * With group flag FAN_REPORT_NAME, if name was not
639 * recorded in an event on a directory, report the name
640 * "." with info type DFID_NAME.
641 */
642 info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
643 dot = ".";
644 dot_len = 1;
645 } else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
646 (event->mask & FAN_ONDIR)) {
647 /*
648 * With group flag FAN_REPORT_DIR_FID, a single info
649 * record has type DFID for directory entry modification
650 * event and for event on a directory.
651 */
652 info_type = FAN_EVENT_INFO_TYPE_DFID;
653 } else {
654 /*
655 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
656 * a single info record has type FID for event on a
657 * non-directory, when there is no directory to report.
658 * For example, on FAN_DELETE_SELF event.
659 */
660 info_type = FAN_EVENT_INFO_TYPE_FID;
661 }
662
663 ret = copy_fid_info_to_user(fanotify_event_fsid(event),
664 fanotify_event_object_fh(event),
665 info_type, dot, dot_len,
666 buf, count);
667 if (ret < 0)
668 return ret;
669
670 buf += ret;
671 count -= ret;
672 total_bytes += ret;
673 }
674
675 if (pidfd_mode) {
676 ret = copy_pidfd_info_to_user(pidfd, buf, count);
677 if (ret < 0)
678 return ret;
679
680 buf += ret;
681 count -= ret;
682 total_bytes += ret;
683 }
684
685 if (fanotify_is_error_event(event->mask)) {
686 ret = copy_error_info_to_user(event, buf, count);
687 if (ret < 0)
688 return ret;
689 buf += ret;
690 count -= ret;
691 total_bytes += ret;
692 }
693
694 if (fanotify_event_has_access_range(event)) {
695 ret = copy_range_info_to_user(event, buf, count);
696 if (ret < 0)
697 return ret;
698 buf += ret;
699 count -= ret;
700 total_bytes += ret;
701 }
702
703 return total_bytes;
704 }
705
copy_event_to_user(struct fsnotify_group * group,struct fanotify_event * event,char __user * buf,size_t count)706 static ssize_t copy_event_to_user(struct fsnotify_group *group,
707 struct fanotify_event *event,
708 char __user *buf, size_t count)
709 {
710 struct fanotify_event_metadata metadata;
711 const struct path *path = fanotify_event_path(event);
712 struct fanotify_info *info = fanotify_event_info(event);
713 unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
714 unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
715 struct file *f = NULL, *pidfd_file = NULL;
716 int ret, pidfd = -ESRCH, fd = -EBADF;
717
718 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
719
720 metadata.event_len = fanotify_event_len(info_mode, event);
721 metadata.metadata_len = FAN_EVENT_METADATA_LEN;
722 metadata.vers = FANOTIFY_METADATA_VERSION;
723 metadata.reserved = 0;
724 metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
725 metadata.pid = pid_vnr(event->pid);
726 /*
727 * For an unprivileged listener, event->pid can be used to identify the
728 * events generated by the listener process itself, without disclosing
729 * the pids of other processes.
730 */
731 if (FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
732 task_tgid(current) != event->pid)
733 metadata.pid = 0;
734
735 /*
736 * For now, fid mode is required for an unprivileged listener and
737 * fid mode does not report fd in events. Keep this check anyway
738 * for safety in case fid mode requirement is relaxed in the future
739 * to allow unprivileged listener to get events with no fd and no fid.
740 */
741 if (!FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
742 path && path->mnt && path->dentry) {
743 fd = create_fd(group, path, &f);
744 /*
745 * Opening an fd from dentry can fail for several reasons.
746 * For example, when tasks are gone and we try to open their
747 * /proc files or we try to open a WRONLY file like in sysfs
748 * or when trying to open a file that was deleted on the
749 * remote network server.
750 *
751 * For a group with FAN_REPORT_FD_ERROR, we will send the
752 * event with the error instead of the open fd, otherwise
753 * Userspace may not get the error at all.
754 * In any case, userspace will not know which file failed to
755 * open, so add a debug print for further investigation.
756 */
757 if (fd < 0) {
758 pr_debug("fanotify: create_fd(%pd2) failed err=%d\n",
759 path->dentry, fd);
760 if (!FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR)) {
761 /*
762 * Historically, we've handled EOPENSTALE in a
763 * special way and silently dropped such
764 * events. Now we have to keep it to maintain
765 * backward compatibility...
766 */
767 if (fd == -EOPENSTALE)
768 fd = 0;
769 return fd;
770 }
771 }
772 }
773 if (FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR))
774 metadata.fd = fd;
775 else
776 metadata.fd = fd >= 0 ? fd : FAN_NOFD;
777
778 if (pidfd_mode) {
779 /*
780 * Complain if the FAN_REPORT_PIDFD and FAN_REPORT_TID mutual
781 * exclusion is ever lifted. At the time of incoporating pidfd
782 * support within fanotify, the pidfd API only supported the
783 * creation of pidfds for thread-group leaders.
784 */
785 WARN_ON_ONCE(FAN_GROUP_FLAG(group, FAN_REPORT_TID));
786
787 /*
788 * The PIDTYPE_TGID check for an event->pid is performed
789 * preemptively in an attempt to catch out cases where the event
790 * listener reads events after the event generating process has
791 * already terminated. Depending on flag FAN_REPORT_FD_ERROR,
792 * report either -ESRCH or FAN_NOPIDFD to the event listener in
793 * those cases with all other pidfd creation errors reported as
794 * the error code itself or as FAN_EPIDFD.
795 */
796 if (metadata.pid && pid_has_task(event->pid, PIDTYPE_TGID))
797 pidfd = pidfd_prepare(event->pid, 0, &pidfd_file);
798
799 if (!FAN_GROUP_FLAG(group, FAN_REPORT_FD_ERROR) && pidfd < 0)
800 pidfd = pidfd == -ESRCH ? FAN_NOPIDFD : FAN_EPIDFD;
801 }
802
803 ret = -EFAULT;
804 /*
805 * Sanity check copy size in case get_one_event() and
806 * event_len sizes ever get out of sync.
807 */
808 if (WARN_ON_ONCE(metadata.event_len > count))
809 goto out_close_fd;
810
811 if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
812 goto out_close_fd;
813
814 buf += FAN_EVENT_METADATA_LEN;
815 count -= FAN_EVENT_METADATA_LEN;
816
817 ret = copy_info_records_to_user(event, info, info_mode, pidfd,
818 buf, count);
819 if (ret < 0)
820 goto out_close_fd;
821
822 if (f)
823 fd_install(fd, f);
824
825 if (pidfd_file)
826 fd_install(pidfd, pidfd_file);
827
828 if (fanotify_is_perm_event(event->mask))
829 FANOTIFY_PERM(event)->fd = fd;
830
831 return metadata.event_len;
832
833 out_close_fd:
834 if (f) {
835 put_unused_fd(fd);
836 fput(f);
837 }
838
839 if (pidfd_file) {
840 put_unused_fd(pidfd);
841 fput(pidfd_file);
842 }
843
844 return ret;
845 }
846
847 /* intofiy userspace file descriptor functions */
fanotify_poll(struct file * file,poll_table * wait)848 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
849 {
850 struct fsnotify_group *group = file->private_data;
851 __poll_t ret = 0;
852
853 poll_wait(file, &group->notification_waitq, wait);
854 spin_lock(&group->notification_lock);
855 if (!fsnotify_notify_queue_is_empty(group))
856 ret = EPOLLIN | EPOLLRDNORM;
857 spin_unlock(&group->notification_lock);
858
859 return ret;
860 }
861
fanotify_read(struct file * file,char __user * buf,size_t count,loff_t * pos)862 static ssize_t fanotify_read(struct file *file, char __user *buf,
863 size_t count, loff_t *pos)
864 {
865 struct fsnotify_group *group;
866 struct fanotify_event *event;
867 char __user *start;
868 int ret;
869 DEFINE_WAIT_FUNC(wait, woken_wake_function);
870
871 start = buf;
872 group = file->private_data;
873
874 pr_debug("%s: group=%p\n", __func__, group);
875
876 add_wait_queue(&group->notification_waitq, &wait);
877 while (1) {
878 /*
879 * User can supply arbitrarily large buffer. Avoid softlockups
880 * in case there are lots of available events.
881 */
882 cond_resched();
883 event = get_one_event(group, count);
884 if (IS_ERR(event)) {
885 ret = PTR_ERR(event);
886 break;
887 }
888
889 if (!event) {
890 ret = -EAGAIN;
891 if (file->f_flags & O_NONBLOCK)
892 break;
893
894 ret = -ERESTARTSYS;
895 if (signal_pending(current))
896 break;
897
898 if (start != buf)
899 break;
900
901 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
902 continue;
903 }
904
905 ret = copy_event_to_user(group, event, buf, count);
906
907 /*
908 * Permission events get queued to wait for response. Other
909 * events can be destroyed now.
910 */
911 if (!fanotify_is_perm_event(event->mask)) {
912 fsnotify_destroy_event(group, &event->fse);
913 } else {
914 if (ret <= 0 || FANOTIFY_PERM(event)->fd < 0) {
915 spin_lock(&group->notification_lock);
916 finish_permission_event(group,
917 FANOTIFY_PERM(event), FAN_DENY, NULL);
918 wake_up(&group->fanotify_data.access_waitq);
919 } else {
920 spin_lock(&group->notification_lock);
921 list_add_tail(&event->fse.list,
922 &group->fanotify_data.access_list);
923 spin_unlock(&group->notification_lock);
924 }
925 }
926 if (ret < 0)
927 break;
928 buf += ret;
929 count -= ret;
930 }
931 remove_wait_queue(&group->notification_waitq, &wait);
932
933 if (start != buf && ret != -EFAULT)
934 ret = buf - start;
935 return ret;
936 }
937
fanotify_write(struct file * file,const char __user * buf,size_t count,loff_t * pos)938 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
939 {
940 struct fanotify_response response;
941 struct fsnotify_group *group;
942 int ret;
943 const char __user *info_buf = buf + sizeof(struct fanotify_response);
944 size_t info_len;
945
946 if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
947 return -EINVAL;
948
949 group = file->private_data;
950
951 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
952
953 if (count < sizeof(response))
954 return -EINVAL;
955
956 if (copy_from_user(&response, buf, sizeof(response)))
957 return -EFAULT;
958
959 info_len = count - sizeof(response);
960
961 ret = process_access_response(group, &response, info_buf, info_len);
962 if (ret < 0)
963 count = ret;
964 else
965 count = sizeof(response) + ret;
966
967 return count;
968 }
969
fanotify_release(struct inode * ignored,struct file * file)970 static int fanotify_release(struct inode *ignored, struct file *file)
971 {
972 struct fsnotify_group *group = file->private_data;
973 struct fsnotify_event *fsn_event;
974
975 /*
976 * Stop new events from arriving in the notification queue. since
977 * userspace cannot use fanotify fd anymore, no event can enter or
978 * leave access_list by now either.
979 */
980 fsnotify_group_stop_queueing(group);
981
982 /*
983 * Process all permission events on access_list and notification queue
984 * and simulate reply from userspace.
985 */
986 spin_lock(&group->notification_lock);
987 while (!list_empty(&group->fanotify_data.access_list)) {
988 struct fanotify_perm_event *event;
989
990 event = list_first_entry(&group->fanotify_data.access_list,
991 struct fanotify_perm_event, fae.fse.list);
992 list_del_init(&event->fae.fse.list);
993 finish_permission_event(group, event, FAN_ALLOW, NULL);
994 spin_lock(&group->notification_lock);
995 }
996
997 /*
998 * Destroy all non-permission events. For permission events just
999 * dequeue them and set the response. They will be freed once the
1000 * response is consumed and fanotify_get_response() returns.
1001 */
1002 while ((fsn_event = fsnotify_remove_first_event(group))) {
1003 struct fanotify_event *event = FANOTIFY_E(fsn_event);
1004
1005 if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
1006 spin_unlock(&group->notification_lock);
1007 fsnotify_destroy_event(group, fsn_event);
1008 } else {
1009 finish_permission_event(group, FANOTIFY_PERM(event),
1010 FAN_ALLOW, NULL);
1011 }
1012 spin_lock(&group->notification_lock);
1013 }
1014 spin_unlock(&group->notification_lock);
1015
1016 /* Response for all permission events it set, wakeup waiters */
1017 wake_up(&group->fanotify_data.access_waitq);
1018
1019 /* matches the fanotify_init->fsnotify_alloc_group */
1020 fsnotify_destroy_group(group);
1021
1022 return 0;
1023 }
1024
fanotify_ioctl(struct file * file,unsigned int cmd,unsigned long arg)1025 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1026 {
1027 struct fsnotify_group *group;
1028 struct fsnotify_event *fsn_event;
1029 void __user *p;
1030 int ret = -ENOTTY;
1031 size_t send_len = 0;
1032
1033 group = file->private_data;
1034
1035 p = (void __user *) arg;
1036
1037 switch (cmd) {
1038 case FIONREAD:
1039 spin_lock(&group->notification_lock);
1040 list_for_each_entry(fsn_event, &group->notification_list, list)
1041 send_len += FAN_EVENT_METADATA_LEN;
1042 spin_unlock(&group->notification_lock);
1043 ret = put_user(send_len, (int __user *) p);
1044 break;
1045 }
1046
1047 return ret;
1048 }
1049
1050 static const struct file_operations fanotify_fops = {
1051 .show_fdinfo = fanotify_show_fdinfo,
1052 .poll = fanotify_poll,
1053 .read = fanotify_read,
1054 .write = fanotify_write,
1055 .fasync = NULL,
1056 .release = fanotify_release,
1057 .unlocked_ioctl = fanotify_ioctl,
1058 .compat_ioctl = compat_ptr_ioctl,
1059 .llseek = noop_llseek,
1060 };
1061
fanotify_find_path(int dfd,const char __user * filename,struct path * path,unsigned int flags,__u64 mask,unsigned int obj_type)1062 static int fanotify_find_path(int dfd, const char __user *filename,
1063 struct path *path, unsigned int flags, __u64 mask,
1064 unsigned int obj_type)
1065 {
1066 int ret;
1067
1068 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
1069 dfd, filename, flags);
1070
1071 if (filename == NULL) {
1072 CLASS(fd, f)(dfd);
1073
1074 if (fd_empty(f))
1075 return -EBADF;
1076
1077 if ((flags & FAN_MARK_ONLYDIR) &&
1078 !(S_ISDIR(file_inode(fd_file(f))->i_mode)))
1079 return -ENOTDIR;
1080
1081 *path = fd_file(f)->f_path;
1082 path_get(path);
1083 } else {
1084 unsigned int lookup_flags = 0;
1085
1086 if (!(flags & FAN_MARK_DONT_FOLLOW))
1087 lookup_flags |= LOOKUP_FOLLOW;
1088 if (flags & FAN_MARK_ONLYDIR)
1089 lookup_flags |= LOOKUP_DIRECTORY;
1090
1091 ret = user_path_at(dfd, filename, lookup_flags, path);
1092 if (ret)
1093 goto out;
1094 }
1095
1096 /* you can only watch an inode if you have read permissions on it */
1097 ret = path_permission(path, MAY_READ);
1098 if (ret) {
1099 path_put(path);
1100 goto out;
1101 }
1102
1103 ret = security_path_notify(path, mask, obj_type);
1104 if (ret)
1105 path_put(path);
1106
1107 out:
1108 return ret;
1109 }
1110
fanotify_mark_remove_from_mask(struct fsnotify_mark * fsn_mark,__u32 mask,unsigned int flags,__u32 umask,int * destroy)1111 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
1112 __u32 mask, unsigned int flags,
1113 __u32 umask, int *destroy)
1114 {
1115 __u32 oldmask, newmask;
1116
1117 /* umask bits cannot be removed by user */
1118 mask &= ~umask;
1119 spin_lock(&fsn_mark->lock);
1120 oldmask = fsnotify_calc_mask(fsn_mark);
1121 if (!(flags & FANOTIFY_MARK_IGNORE_BITS)) {
1122 fsn_mark->mask &= ~mask;
1123 } else {
1124 fsn_mark->ignore_mask &= ~mask;
1125 }
1126 newmask = fsnotify_calc_mask(fsn_mark);
1127 /*
1128 * We need to keep the mark around even if remaining mask cannot
1129 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
1130 * changes to the mask.
1131 * Destroy mark when only umask bits remain.
1132 */
1133 *destroy = !((fsn_mark->mask | fsn_mark->ignore_mask) & ~umask);
1134 spin_unlock(&fsn_mark->lock);
1135
1136 return oldmask & ~newmask;
1137 }
1138
fanotify_remove_mark(struct fsnotify_group * group,void * obj,unsigned int obj_type,__u32 mask,unsigned int flags,__u32 umask)1139 static int fanotify_remove_mark(struct fsnotify_group *group,
1140 void *obj, unsigned int obj_type, __u32 mask,
1141 unsigned int flags, __u32 umask)
1142 {
1143 struct fsnotify_mark *fsn_mark = NULL;
1144 __u32 removed;
1145 int destroy_mark;
1146
1147 fsnotify_group_lock(group);
1148 fsn_mark = fsnotify_find_mark(obj, obj_type, group);
1149 if (!fsn_mark) {
1150 fsnotify_group_unlock(group);
1151 return -ENOENT;
1152 }
1153
1154 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
1155 umask, &destroy_mark);
1156 if (removed & fsnotify_conn_mask(fsn_mark->connector))
1157 fsnotify_recalc_mask(fsn_mark->connector);
1158 if (destroy_mark)
1159 fsnotify_detach_mark(fsn_mark);
1160 fsnotify_group_unlock(group);
1161 if (destroy_mark)
1162 fsnotify_free_mark(fsn_mark);
1163
1164 /* matches the fsnotify_find_mark() */
1165 fsnotify_put_mark(fsn_mark);
1166 return 0;
1167 }
1168
fanotify_mark_update_flags(struct fsnotify_mark * fsn_mark,unsigned int fan_flags)1169 static bool fanotify_mark_update_flags(struct fsnotify_mark *fsn_mark,
1170 unsigned int fan_flags)
1171 {
1172 bool want_iref = !(fan_flags & FAN_MARK_EVICTABLE);
1173 unsigned int ignore = fan_flags & FANOTIFY_MARK_IGNORE_BITS;
1174 bool recalc = false;
1175
1176 /*
1177 * When using FAN_MARK_IGNORE for the first time, mark starts using
1178 * independent event flags in ignore mask. After that, trying to
1179 * update the ignore mask with the old FAN_MARK_IGNORED_MASK API
1180 * will result in EEXIST error.
1181 */
1182 if (ignore == FAN_MARK_IGNORE)
1183 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS;
1184
1185 /*
1186 * Setting FAN_MARK_IGNORED_SURV_MODIFY for the first time may lead to
1187 * the removal of the FS_MODIFY bit in calculated mask if it was set
1188 * because of an ignore mask that is now going to survive FS_MODIFY.
1189 */
1190 if (ignore && (fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1191 !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) {
1192 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
1193 if (!(fsn_mark->mask & FS_MODIFY))
1194 recalc = true;
1195 }
1196
1197 if (fsn_mark->connector->type != FSNOTIFY_OBJ_TYPE_INODE ||
1198 want_iref == !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1199 return recalc;
1200
1201 /*
1202 * NO_IREF may be removed from a mark, but not added.
1203 * When removed, fsnotify_recalc_mask() will take the inode ref.
1204 */
1205 WARN_ON_ONCE(!want_iref);
1206 fsn_mark->flags &= ~FSNOTIFY_MARK_FLAG_NO_IREF;
1207
1208 return true;
1209 }
1210
fanotify_mark_add_to_mask(struct fsnotify_mark * fsn_mark,__u32 mask,unsigned int fan_flags)1211 static bool fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
1212 __u32 mask, unsigned int fan_flags)
1213 {
1214 bool recalc;
1215
1216 spin_lock(&fsn_mark->lock);
1217 if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS))
1218 fsn_mark->mask |= mask;
1219 else
1220 fsn_mark->ignore_mask |= mask;
1221
1222 recalc = fsnotify_calc_mask(fsn_mark) &
1223 ~fsnotify_conn_mask(fsn_mark->connector);
1224
1225 recalc |= fanotify_mark_update_flags(fsn_mark, fan_flags);
1226 spin_unlock(&fsn_mark->lock);
1227
1228 return recalc;
1229 }
1230
1231 struct fan_fsid {
1232 struct super_block *sb;
1233 __kernel_fsid_t id;
1234 bool weak;
1235 };
1236
fanotify_set_mark_fsid(struct fsnotify_group * group,struct fsnotify_mark * mark,struct fan_fsid * fsid)1237 static int fanotify_set_mark_fsid(struct fsnotify_group *group,
1238 struct fsnotify_mark *mark,
1239 struct fan_fsid *fsid)
1240 {
1241 struct fsnotify_mark_connector *conn;
1242 struct fsnotify_mark *old;
1243 struct super_block *old_sb = NULL;
1244
1245 FANOTIFY_MARK(mark)->fsid = fsid->id;
1246 mark->flags |= FSNOTIFY_MARK_FLAG_HAS_FSID;
1247 if (fsid->weak)
1248 mark->flags |= FSNOTIFY_MARK_FLAG_WEAK_FSID;
1249
1250 /* First mark added will determine if group is single or multi fsid */
1251 if (list_empty(&group->marks_list))
1252 return 0;
1253
1254 /* Find sb of an existing mark */
1255 list_for_each_entry(old, &group->marks_list, g_list) {
1256 conn = READ_ONCE(old->connector);
1257 if (!conn)
1258 continue;
1259 old_sb = fsnotify_connector_sb(conn);
1260 if (old_sb)
1261 break;
1262 }
1263
1264 /* Only detached marks left? */
1265 if (!old_sb)
1266 return 0;
1267
1268 /* Do not allow mixing of marks with weak and strong fsid */
1269 if ((mark->flags ^ old->flags) & FSNOTIFY_MARK_FLAG_WEAK_FSID)
1270 return -EXDEV;
1271
1272 /* Allow mixing of marks with strong fsid from different fs */
1273 if (!fsid->weak)
1274 return 0;
1275
1276 /* Do not allow mixing marks with weak fsid from different fs */
1277 if (old_sb != fsid->sb)
1278 return -EXDEV;
1279
1280 /* Do not allow mixing marks from different btrfs sub-volumes */
1281 if (!fanotify_fsid_equal(&FANOTIFY_MARK(old)->fsid,
1282 &FANOTIFY_MARK(mark)->fsid))
1283 return -EXDEV;
1284
1285 return 0;
1286 }
1287
fanotify_add_new_mark(struct fsnotify_group * group,void * obj,unsigned int obj_type,unsigned int fan_flags,struct fan_fsid * fsid)1288 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
1289 void *obj,
1290 unsigned int obj_type,
1291 unsigned int fan_flags,
1292 struct fan_fsid *fsid)
1293 {
1294 struct ucounts *ucounts = group->fanotify_data.ucounts;
1295 struct fanotify_mark *fan_mark;
1296 struct fsnotify_mark *mark;
1297 int ret;
1298
1299 /*
1300 * Enforce per user marks limits per user in all containing user ns.
1301 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
1302 * in the limited groups account.
1303 */
1304 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
1305 !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
1306 return ERR_PTR(-ENOSPC);
1307
1308 fan_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
1309 if (!fan_mark) {
1310 ret = -ENOMEM;
1311 goto out_dec_ucounts;
1312 }
1313
1314 mark = &fan_mark->fsn_mark;
1315 fsnotify_init_mark(mark, group);
1316 if (fan_flags & FAN_MARK_EVICTABLE)
1317 mark->flags |= FSNOTIFY_MARK_FLAG_NO_IREF;
1318
1319 /* Cache fsid of filesystem containing the marked object */
1320 if (fsid) {
1321 ret = fanotify_set_mark_fsid(group, mark, fsid);
1322 if (ret)
1323 goto out_put_mark;
1324 } else {
1325 fan_mark->fsid.val[0] = fan_mark->fsid.val[1] = 0;
1326 }
1327
1328 ret = fsnotify_add_mark_locked(mark, obj, obj_type, 0);
1329 if (ret)
1330 goto out_put_mark;
1331
1332 return mark;
1333
1334 out_put_mark:
1335 fsnotify_put_mark(mark);
1336 out_dec_ucounts:
1337 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1338 dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
1339 return ERR_PTR(ret);
1340 }
1341
fanotify_group_init_error_pool(struct fsnotify_group * group)1342 static int fanotify_group_init_error_pool(struct fsnotify_group *group)
1343 {
1344 if (mempool_initialized(&group->fanotify_data.error_events_pool))
1345 return 0;
1346
1347 return mempool_init_kmalloc_pool(&group->fanotify_data.error_events_pool,
1348 FANOTIFY_DEFAULT_FEE_POOL_SIZE,
1349 sizeof(struct fanotify_error_event));
1350 }
1351
fanotify_may_update_existing_mark(struct fsnotify_mark * fsn_mark,__u32 mask,unsigned int fan_flags)1352 static int fanotify_may_update_existing_mark(struct fsnotify_mark *fsn_mark,
1353 __u32 mask, unsigned int fan_flags)
1354 {
1355 /*
1356 * Non evictable mark cannot be downgraded to evictable mark.
1357 */
1358 if (fan_flags & FAN_MARK_EVICTABLE &&
1359 !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1360 return -EEXIST;
1361
1362 /*
1363 * New ignore mask semantics cannot be downgraded to old semantics.
1364 */
1365 if (fan_flags & FAN_MARK_IGNORED_MASK &&
1366 fsn_mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS)
1367 return -EEXIST;
1368
1369 /*
1370 * An ignore mask that survives modify could never be downgraded to not
1371 * survive modify. With new FAN_MARK_IGNORE semantics we make that rule
1372 * explicit and return an error when trying to update the ignore mask
1373 * without the original FAN_MARK_IGNORED_SURV_MODIFY value.
1374 */
1375 if (fan_flags & FAN_MARK_IGNORE &&
1376 !(fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1377 fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)
1378 return -EEXIST;
1379
1380 /* For now pre-content events are not generated for directories */
1381 mask |= fsn_mark->mask;
1382 if (mask & FANOTIFY_PRE_CONTENT_EVENTS && mask & FAN_ONDIR)
1383 return -EEXIST;
1384
1385 return 0;
1386 }
1387
fanotify_add_mark(struct fsnotify_group * group,void * obj,unsigned int obj_type,__u32 mask,unsigned int fan_flags,struct fan_fsid * fsid)1388 static int fanotify_add_mark(struct fsnotify_group *group,
1389 void *obj, unsigned int obj_type,
1390 __u32 mask, unsigned int fan_flags,
1391 struct fan_fsid *fsid)
1392 {
1393 struct fsnotify_mark *fsn_mark;
1394 bool recalc;
1395 int ret = 0;
1396
1397 fsnotify_group_lock(group);
1398 fsn_mark = fsnotify_find_mark(obj, obj_type, group);
1399 if (!fsn_mark) {
1400 fsn_mark = fanotify_add_new_mark(group, obj, obj_type,
1401 fan_flags, fsid);
1402 if (IS_ERR(fsn_mark)) {
1403 fsnotify_group_unlock(group);
1404 return PTR_ERR(fsn_mark);
1405 }
1406 }
1407
1408 /*
1409 * Check if requested mark flags conflict with an existing mark flags.
1410 */
1411 ret = fanotify_may_update_existing_mark(fsn_mark, mask, fan_flags);
1412 if (ret)
1413 goto out;
1414
1415 /*
1416 * Error events are pre-allocated per group, only if strictly
1417 * needed (i.e. FAN_FS_ERROR was requested).
1418 */
1419 if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS) &&
1420 (mask & FAN_FS_ERROR)) {
1421 ret = fanotify_group_init_error_pool(group);
1422 if (ret)
1423 goto out;
1424 }
1425
1426 recalc = fanotify_mark_add_to_mask(fsn_mark, mask, fan_flags);
1427 if (recalc)
1428 fsnotify_recalc_mask(fsn_mark->connector);
1429
1430 out:
1431 fsnotify_group_unlock(group);
1432
1433 fsnotify_put_mark(fsn_mark);
1434 return ret;
1435 }
1436
fanotify_alloc_overflow_event(void)1437 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1438 {
1439 struct fanotify_event *oevent;
1440
1441 oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1442 if (!oevent)
1443 return NULL;
1444
1445 fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1446 oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1447
1448 return &oevent->fse;
1449 }
1450
fanotify_alloc_merge_hash(void)1451 static struct hlist_head *fanotify_alloc_merge_hash(void)
1452 {
1453 struct hlist_head *hash;
1454
1455 hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1456 GFP_KERNEL_ACCOUNT);
1457 if (!hash)
1458 return NULL;
1459
1460 __hash_init(hash, FANOTIFY_HTABLE_SIZE);
1461
1462 return hash;
1463 }
1464
1465 /* fanotify syscalls */
SYSCALL_DEFINE2(fanotify_init,unsigned int,flags,unsigned int,event_f_flags)1466 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1467 {
1468 struct fsnotify_group *group;
1469 int f_flags, fd;
1470 unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1471 unsigned int class = flags & FANOTIFY_CLASS_BITS;
1472 unsigned int internal_flags = 0;
1473 struct file *file;
1474
1475 pr_debug("%s: flags=%x event_f_flags=%x\n",
1476 __func__, flags, event_f_flags);
1477
1478 if (!capable(CAP_SYS_ADMIN)) {
1479 /*
1480 * An unprivileged user can setup an fanotify group with
1481 * limited functionality - an unprivileged group is limited to
1482 * notification events with file handles and it cannot use
1483 * unlimited queue/marks.
1484 */
1485 if ((flags & FANOTIFY_ADMIN_INIT_FLAGS) || !fid_mode)
1486 return -EPERM;
1487
1488 /*
1489 * Setting the internal flag FANOTIFY_UNPRIV on the group
1490 * prevents setting mount/filesystem marks on this group and
1491 * prevents reporting pid and open fd in events.
1492 */
1493 internal_flags |= FANOTIFY_UNPRIV;
1494 }
1495
1496 #ifdef CONFIG_AUDITSYSCALL
1497 if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1498 #else
1499 if (flags & ~FANOTIFY_INIT_FLAGS)
1500 #endif
1501 return -EINVAL;
1502
1503 /*
1504 * A pidfd can only be returned for a thread-group leader; thus
1505 * FAN_REPORT_PIDFD and FAN_REPORT_TID need to remain mutually
1506 * exclusive.
1507 */
1508 if ((flags & FAN_REPORT_PIDFD) && (flags & FAN_REPORT_TID))
1509 return -EINVAL;
1510
1511 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1512 return -EINVAL;
1513
1514 switch (event_f_flags & O_ACCMODE) {
1515 case O_RDONLY:
1516 case O_RDWR:
1517 case O_WRONLY:
1518 break;
1519 default:
1520 return -EINVAL;
1521 }
1522
1523 if (fid_mode && class != FAN_CLASS_NOTIF)
1524 return -EINVAL;
1525
1526 /*
1527 * Child name is reported with parent fid so requires dir fid.
1528 * We can report both child fid and dir fid with or without name.
1529 */
1530 if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1531 return -EINVAL;
1532
1533 /*
1534 * FAN_REPORT_TARGET_FID requires FAN_REPORT_NAME and FAN_REPORT_FID
1535 * and is used as an indication to report both dir and child fid on all
1536 * dirent events.
1537 */
1538 if ((fid_mode & FAN_REPORT_TARGET_FID) &&
1539 (!(fid_mode & FAN_REPORT_NAME) || !(fid_mode & FAN_REPORT_FID)))
1540 return -EINVAL;
1541
1542 f_flags = O_RDWR;
1543 if (flags & FAN_CLOEXEC)
1544 f_flags |= O_CLOEXEC;
1545 if (flags & FAN_NONBLOCK)
1546 f_flags |= O_NONBLOCK;
1547
1548 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
1549 group = fsnotify_alloc_group(&fanotify_fsnotify_ops,
1550 FSNOTIFY_GROUP_USER);
1551 if (IS_ERR(group)) {
1552 return PTR_ERR(group);
1553 }
1554
1555 /* Enforce groups limits per user in all containing user ns */
1556 group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1557 current_euid(),
1558 UCOUNT_FANOTIFY_GROUPS);
1559 if (!group->fanotify_data.ucounts) {
1560 fd = -EMFILE;
1561 goto out_destroy_group;
1562 }
1563
1564 group->fanotify_data.flags = flags | internal_flags;
1565 group->memcg = get_mem_cgroup_from_mm(current->mm);
1566
1567 group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1568 if (!group->fanotify_data.merge_hash) {
1569 fd = -ENOMEM;
1570 goto out_destroy_group;
1571 }
1572
1573 group->overflow_event = fanotify_alloc_overflow_event();
1574 if (unlikely(!group->overflow_event)) {
1575 fd = -ENOMEM;
1576 goto out_destroy_group;
1577 }
1578
1579 if (force_o_largefile())
1580 event_f_flags |= O_LARGEFILE;
1581 group->fanotify_data.f_flags = event_f_flags;
1582 init_waitqueue_head(&group->fanotify_data.access_waitq);
1583 INIT_LIST_HEAD(&group->fanotify_data.access_list);
1584 switch (class) {
1585 case FAN_CLASS_NOTIF:
1586 group->priority = FSNOTIFY_PRIO_NORMAL;
1587 break;
1588 case FAN_CLASS_CONTENT:
1589 group->priority = FSNOTIFY_PRIO_CONTENT;
1590 break;
1591 case FAN_CLASS_PRE_CONTENT:
1592 group->priority = FSNOTIFY_PRIO_PRE_CONTENT;
1593 break;
1594 default:
1595 fd = -EINVAL;
1596 goto out_destroy_group;
1597 }
1598
1599 if (flags & FAN_UNLIMITED_QUEUE) {
1600 fd = -EPERM;
1601 if (!capable(CAP_SYS_ADMIN))
1602 goto out_destroy_group;
1603 group->max_events = UINT_MAX;
1604 } else {
1605 group->max_events = fanotify_max_queued_events;
1606 }
1607
1608 if (flags & FAN_UNLIMITED_MARKS) {
1609 fd = -EPERM;
1610 if (!capable(CAP_SYS_ADMIN))
1611 goto out_destroy_group;
1612 }
1613
1614 if (flags & FAN_ENABLE_AUDIT) {
1615 fd = -EPERM;
1616 if (!capable(CAP_AUDIT_WRITE))
1617 goto out_destroy_group;
1618 }
1619
1620 fd = get_unused_fd_flags(f_flags);
1621 if (fd < 0)
1622 goto out_destroy_group;
1623
1624 file = anon_inode_getfile_fmode("[fanotify]", &fanotify_fops, group,
1625 f_flags, FMODE_NONOTIFY);
1626 if (IS_ERR(file)) {
1627 put_unused_fd(fd);
1628 fd = PTR_ERR(file);
1629 goto out_destroy_group;
1630 }
1631 fd_install(fd, file);
1632 return fd;
1633
1634 out_destroy_group:
1635 fsnotify_destroy_group(group);
1636 return fd;
1637 }
1638
fanotify_test_fsid(struct dentry * dentry,unsigned int flags,struct fan_fsid * fsid)1639 static int fanotify_test_fsid(struct dentry *dentry, unsigned int flags,
1640 struct fan_fsid *fsid)
1641 {
1642 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1643 __kernel_fsid_t root_fsid;
1644 int err;
1645
1646 /*
1647 * Make sure dentry is not of a filesystem with zero fsid (e.g. fuse).
1648 */
1649 err = vfs_get_fsid(dentry, &fsid->id);
1650 if (err)
1651 return err;
1652
1653 fsid->sb = dentry->d_sb;
1654 if (!fsid->id.val[0] && !fsid->id.val[1]) {
1655 err = -ENODEV;
1656 goto weak;
1657 }
1658
1659 /*
1660 * Make sure dentry is not of a filesystem subvolume (e.g. btrfs)
1661 * which uses a different fsid than sb root.
1662 */
1663 err = vfs_get_fsid(dentry->d_sb->s_root, &root_fsid);
1664 if (err)
1665 return err;
1666
1667 if (!fanotify_fsid_equal(&root_fsid, &fsid->id)) {
1668 err = -EXDEV;
1669 goto weak;
1670 }
1671
1672 fsid->weak = false;
1673 return 0;
1674
1675 weak:
1676 /* Allow weak fsid when marking inodes */
1677 fsid->weak = true;
1678 return (mark_type == FAN_MARK_INODE) ? 0 : err;
1679 }
1680
1681 /* Check if filesystem can encode a unique fid */
fanotify_test_fid(struct dentry * dentry,unsigned int flags)1682 static int fanotify_test_fid(struct dentry *dentry, unsigned int flags)
1683 {
1684 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1685 const struct export_operations *nop = dentry->d_sb->s_export_op;
1686
1687 /*
1688 * We need to make sure that the filesystem supports encoding of
1689 * file handles so user can use name_to_handle_at() to compare fids
1690 * reported with events to the file handle of watched objects.
1691 */
1692 if (!exportfs_can_encode_fid(nop))
1693 return -EOPNOTSUPP;
1694
1695 /*
1696 * For sb/mount mark, we also need to make sure that the filesystem
1697 * supports decoding file handles, so user has a way to map back the
1698 * reported fids to filesystem objects.
1699 */
1700 if (mark_type != FAN_MARK_INODE && !exportfs_can_decode_fh(nop))
1701 return -EOPNOTSUPP;
1702
1703 return 0;
1704 }
1705
fanotify_events_supported(struct fsnotify_group * group,const struct path * path,__u64 mask,unsigned int flags)1706 static int fanotify_events_supported(struct fsnotify_group *group,
1707 const struct path *path, __u64 mask,
1708 unsigned int flags)
1709 {
1710 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1711 bool is_dir = d_is_dir(path->dentry);
1712 /* Strict validation of events in non-dir inode mask with v5.17+ APIs */
1713 bool strict_dir_events = FAN_GROUP_FLAG(group, FAN_REPORT_TARGET_FID) ||
1714 (mask & FAN_RENAME) ||
1715 (flags & FAN_MARK_IGNORE);
1716
1717 /*
1718 * Filesystems need to opt-into pre-content evnets (a.k.a HSM)
1719 * and they are only supported on regular files and directories.
1720 */
1721 if (mask & FANOTIFY_PRE_CONTENT_EVENTS) {
1722 if (!(path->mnt->mnt_sb->s_iflags & SB_I_ALLOW_HSM))
1723 return -EOPNOTSUPP;
1724 if (!is_dir && !d_is_reg(path->dentry))
1725 return -EINVAL;
1726 }
1727
1728 /*
1729 * Some filesystems such as 'proc' acquire unusual locks when opening
1730 * files. For them fanotify permission events have high chances of
1731 * deadlocking the system - open done when reporting fanotify event
1732 * blocks on this "unusual" lock while another process holding the lock
1733 * waits for fanotify permission event to be answered. Just disallow
1734 * permission events for such filesystems.
1735 */
1736 if (mask & FANOTIFY_PERM_EVENTS &&
1737 path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1738 return -EINVAL;
1739
1740 /*
1741 * mount and sb marks are not allowed on kernel internal pseudo fs,
1742 * like pipe_mnt, because that would subscribe to events on all the
1743 * anonynous pipes in the system.
1744 *
1745 * SB_NOUSER covers all of the internal pseudo fs whose objects are not
1746 * exposed to user's mount namespace, but there are other SB_KERNMOUNT
1747 * fs, like nsfs, debugfs, for which the value of allowing sb and mount
1748 * mark is questionable. For now we leave them alone.
1749 */
1750 if (mark_type != FAN_MARK_INODE &&
1751 path->mnt->mnt_sb->s_flags & SB_NOUSER)
1752 return -EINVAL;
1753
1754 /*
1755 * We shouldn't have allowed setting dirent events and the directory
1756 * flags FAN_ONDIR and FAN_EVENT_ON_CHILD in mask of non-dir inode,
1757 * but because we always allowed it, error only when using new APIs.
1758 */
1759 if (strict_dir_events && mark_type == FAN_MARK_INODE &&
1760 !is_dir && (mask & FANOTIFY_DIRONLY_EVENT_BITS))
1761 return -ENOTDIR;
1762
1763 return 0;
1764 }
1765
do_fanotify_mark(int fanotify_fd,unsigned int flags,__u64 mask,int dfd,const char __user * pathname)1766 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1767 int dfd, const char __user *pathname)
1768 {
1769 struct inode *inode = NULL;
1770 struct vfsmount *mnt = NULL;
1771 struct fsnotify_group *group;
1772 struct path path;
1773 struct fan_fsid __fsid, *fsid = NULL;
1774 u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1775 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1776 unsigned int mark_cmd = flags & FANOTIFY_MARK_CMD_BITS;
1777 unsigned int ignore = flags & FANOTIFY_MARK_IGNORE_BITS;
1778 unsigned int obj_type, fid_mode;
1779 void *obj;
1780 u32 umask = 0;
1781 int ret;
1782
1783 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1784 __func__, fanotify_fd, flags, dfd, pathname, mask);
1785
1786 /* we only use the lower 32 bits as of right now. */
1787 if (upper_32_bits(mask))
1788 return -EINVAL;
1789
1790 if (flags & ~FANOTIFY_MARK_FLAGS)
1791 return -EINVAL;
1792
1793 switch (mark_type) {
1794 case FAN_MARK_INODE:
1795 obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1796 break;
1797 case FAN_MARK_MOUNT:
1798 obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1799 break;
1800 case FAN_MARK_FILESYSTEM:
1801 obj_type = FSNOTIFY_OBJ_TYPE_SB;
1802 break;
1803 default:
1804 return -EINVAL;
1805 }
1806
1807 switch (mark_cmd) {
1808 case FAN_MARK_ADD:
1809 case FAN_MARK_REMOVE:
1810 if (!mask)
1811 return -EINVAL;
1812 break;
1813 case FAN_MARK_FLUSH:
1814 if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1815 return -EINVAL;
1816 break;
1817 default:
1818 return -EINVAL;
1819 }
1820
1821 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1822 valid_mask |= FANOTIFY_PERM_EVENTS;
1823
1824 if (mask & ~valid_mask)
1825 return -EINVAL;
1826
1827
1828 /* We don't allow FAN_MARK_IGNORE & FAN_MARK_IGNORED_MASK together */
1829 if (ignore == (FAN_MARK_IGNORE | FAN_MARK_IGNORED_MASK))
1830 return -EINVAL;
1831
1832 /*
1833 * Event flags (FAN_ONDIR, FAN_EVENT_ON_CHILD) have no effect with
1834 * FAN_MARK_IGNORED_MASK.
1835 */
1836 if (ignore == FAN_MARK_IGNORED_MASK) {
1837 mask &= ~FANOTIFY_EVENT_FLAGS;
1838 umask = FANOTIFY_EVENT_FLAGS;
1839 }
1840
1841 CLASS(fd, f)(fanotify_fd);
1842 if (fd_empty(f))
1843 return -EBADF;
1844
1845 /* verify that this is indeed an fanotify instance */
1846 if (unlikely(fd_file(f)->f_op != &fanotify_fops))
1847 return -EINVAL;
1848 group = fd_file(f)->private_data;
1849
1850 /*
1851 * An unprivileged user is not allowed to setup mount nor filesystem
1852 * marks. This also includes setting up such marks by a group that
1853 * was initialized by an unprivileged user.
1854 */
1855 if ((!capable(CAP_SYS_ADMIN) ||
1856 FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV)) &&
1857 mark_type != FAN_MARK_INODE)
1858 return -EPERM;
1859
1860 /*
1861 * Permission events are not allowed for FAN_CLASS_NOTIF.
1862 * Pre-content permission events are not allowed for FAN_CLASS_CONTENT.
1863 */
1864 if (mask & FANOTIFY_PERM_EVENTS &&
1865 group->priority == FSNOTIFY_PRIO_NORMAL)
1866 return -EINVAL;
1867 else if (mask & FANOTIFY_PRE_CONTENT_EVENTS &&
1868 group->priority == FSNOTIFY_PRIO_CONTENT)
1869 return -EINVAL;
1870
1871 if (mask & FAN_FS_ERROR &&
1872 mark_type != FAN_MARK_FILESYSTEM)
1873 return -EINVAL;
1874
1875 /*
1876 * Evictable is only relevant for inode marks, because only inode object
1877 * can be evicted on memory pressure.
1878 */
1879 if (flags & FAN_MARK_EVICTABLE &&
1880 mark_type != FAN_MARK_INODE)
1881 return -EINVAL;
1882
1883 /*
1884 * Events that do not carry enough information to report
1885 * event->fd require a group that supports reporting fid. Those
1886 * events are not supported on a mount mark, because they do not
1887 * carry enough information (i.e. path) to be filtered by mount
1888 * point.
1889 */
1890 fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1891 if (mask & ~(FANOTIFY_FD_EVENTS|FANOTIFY_EVENT_FLAGS) &&
1892 (!fid_mode || mark_type == FAN_MARK_MOUNT))
1893 return -EINVAL;
1894
1895 /*
1896 * FAN_RENAME uses special info type records to report the old and
1897 * new parent+name. Reporting only old and new parent id is less
1898 * useful and was not implemented.
1899 */
1900 if (mask & FAN_RENAME && !(fid_mode & FAN_REPORT_NAME))
1901 return -EINVAL;
1902
1903 /* Pre-content events are not currently generated for directories. */
1904 if (mask & FANOTIFY_PRE_CONTENT_EVENTS && mask & FAN_ONDIR)
1905 return -EINVAL;
1906
1907 if (mark_cmd == FAN_MARK_FLUSH) {
1908 if (mark_type == FAN_MARK_MOUNT)
1909 fsnotify_clear_vfsmount_marks_by_group(group);
1910 else if (mark_type == FAN_MARK_FILESYSTEM)
1911 fsnotify_clear_sb_marks_by_group(group);
1912 else
1913 fsnotify_clear_inode_marks_by_group(group);
1914 return 0;
1915 }
1916
1917 ret = fanotify_find_path(dfd, pathname, &path, flags,
1918 (mask & ALL_FSNOTIFY_EVENTS), obj_type);
1919 if (ret)
1920 return ret;
1921
1922 if (mark_cmd == FAN_MARK_ADD) {
1923 ret = fanotify_events_supported(group, &path, mask, flags);
1924 if (ret)
1925 goto path_put_and_out;
1926 }
1927
1928 if (fid_mode) {
1929 ret = fanotify_test_fsid(path.dentry, flags, &__fsid);
1930 if (ret)
1931 goto path_put_and_out;
1932
1933 ret = fanotify_test_fid(path.dentry, flags);
1934 if (ret)
1935 goto path_put_and_out;
1936
1937 fsid = &__fsid;
1938 }
1939
1940 /* inode held in place by reference to path; group by fget on fd */
1941 if (mark_type == FAN_MARK_INODE) {
1942 inode = path.dentry->d_inode;
1943 obj = inode;
1944 } else {
1945 mnt = path.mnt;
1946 if (mark_type == FAN_MARK_MOUNT)
1947 obj = mnt;
1948 else
1949 obj = mnt->mnt_sb;
1950 }
1951
1952 /*
1953 * If some other task has this inode open for write we should not add
1954 * an ignore mask, unless that ignore mask is supposed to survive
1955 * modification changes anyway.
1956 */
1957 if (mark_cmd == FAN_MARK_ADD && (flags & FANOTIFY_MARK_IGNORE_BITS) &&
1958 !(flags & FAN_MARK_IGNORED_SURV_MODIFY)) {
1959 ret = mnt ? -EINVAL : -EISDIR;
1960 /* FAN_MARK_IGNORE requires SURV_MODIFY for sb/mount/dir marks */
1961 if (ignore == FAN_MARK_IGNORE &&
1962 (mnt || S_ISDIR(inode->i_mode)))
1963 goto path_put_and_out;
1964
1965 ret = 0;
1966 if (inode && inode_is_open_for_write(inode))
1967 goto path_put_and_out;
1968 }
1969
1970 /* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1971 if (mnt || !S_ISDIR(inode->i_mode)) {
1972 mask &= ~FAN_EVENT_ON_CHILD;
1973 umask = FAN_EVENT_ON_CHILD;
1974 /*
1975 * If group needs to report parent fid, register for getting
1976 * events with parent/name info for non-directory.
1977 */
1978 if ((fid_mode & FAN_REPORT_DIR_FID) &&
1979 (flags & FAN_MARK_ADD) && !ignore)
1980 mask |= FAN_EVENT_ON_CHILD;
1981 }
1982
1983 /* create/update an inode mark */
1984 switch (mark_cmd) {
1985 case FAN_MARK_ADD:
1986 ret = fanotify_add_mark(group, obj, obj_type, mask, flags,
1987 fsid);
1988 break;
1989 case FAN_MARK_REMOVE:
1990 ret = fanotify_remove_mark(group, obj, obj_type, mask, flags,
1991 umask);
1992 break;
1993 default:
1994 ret = -EINVAL;
1995 }
1996
1997 path_put_and_out:
1998 path_put(&path);
1999 return ret;
2000 }
2001
2002 #ifndef CONFIG_ARCH_SPLIT_ARG64
SYSCALL_DEFINE5(fanotify_mark,int,fanotify_fd,unsigned int,flags,__u64,mask,int,dfd,const char __user *,pathname)2003 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
2004 __u64, mask, int, dfd,
2005 const char __user *, pathname)
2006 {
2007 return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
2008 }
2009 #endif
2010
2011 #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)2012 SYSCALL32_DEFINE6(fanotify_mark,
2013 int, fanotify_fd, unsigned int, flags,
2014 SC_ARG64(mask), int, dfd,
2015 const char __user *, pathname)
2016 {
2017 return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
2018 dfd, pathname);
2019 }
2020 #endif
2021
2022 /*
2023 * fanotify_user_setup - Our initialization function. Note that we cannot return
2024 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
2025 * must result in panic().
2026 */
fanotify_user_setup(void)2027 static int __init fanotify_user_setup(void)
2028 {
2029 struct sysinfo si;
2030 int max_marks;
2031
2032 si_meminfo(&si);
2033 /*
2034 * Allow up to 1% of addressable memory to be accounted for per user
2035 * marks limited to the range [8192, 1048576]. mount and sb marks are
2036 * a lot cheaper than inode marks, but there is no reason for a user
2037 * to have many of those, so calculate by the cost of inode marks.
2038 */
2039 max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
2040 INODE_MARK_COST;
2041 max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
2042 FANOTIFY_DEFAULT_MAX_USER_MARKS);
2043
2044 BUILD_BUG_ON(FANOTIFY_INIT_FLAGS & FANOTIFY_INTERNAL_GROUP_FLAGS);
2045 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 13);
2046 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 11);
2047
2048 fanotify_mark_cache = KMEM_CACHE(fanotify_mark,
2049 SLAB_PANIC|SLAB_ACCOUNT);
2050 fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
2051 SLAB_PANIC);
2052 fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
2053 SLAB_PANIC);
2054 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
2055 fanotify_perm_event_cachep =
2056 KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
2057 }
2058
2059 fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
2060 init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
2061 FANOTIFY_DEFAULT_MAX_GROUPS;
2062 init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
2063 fanotify_sysctls_init();
2064
2065 return 0;
2066 }
2067 device_initcall(fanotify_user_setup);
2068