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>
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>
24 #include <asm/ioctls.h>
26 #include "../fsnotify.h"
27 #include "../fdinfo.h"
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
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.
42 #define FANOTIFY_DEFAULT_MAX_USER_MARKS \
43 (FANOTIFY_OLD_DEFAULT_MAX_MARKS * FANOTIFY_DEFAULT_MAX_GROUPS)
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.
50 #define INODE_MARK_COST (2 * sizeof(struct inode))
52 /* configurable via /proc/sys/fs/fanotify/ */
53 static int fanotify_max_queued_events __read_mostly;
57 #include <linux/sysctl.h>
59 static long ft_zero = 0;
60 static long ft_int_max = INT_MAX;
62 static struct ctl_table fanotify_table[] = {
64 .procname = "max_user_groups",
65 .data = &init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS],
66 .maxlen = sizeof(long),
68 .proc_handler = proc_doulongvec_minmax,
70 .extra2 = &ft_int_max,
73 .procname = "max_user_marks",
74 .data = &init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS],
75 .maxlen = sizeof(long),
77 .proc_handler = proc_doulongvec_minmax,
79 .extra2 = &ft_int_max,
82 .procname = "max_queued_events",
83 .data = &fanotify_max_queued_events,
84 .maxlen = sizeof(int),
86 .proc_handler = proc_dointvec_minmax,
91 static void __init fanotify_sysctls_init(void)
93 register_sysctl("fs/fanotify", fanotify_table);
96 #define fanotify_sysctls_init() do { } while (0)
97 #endif /* CONFIG_SYSCTL */
100 * All flags that may be specified in parameter event_f_flags of fanotify_init.
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
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 )
112 extern const struct fsnotify_ops fanotify_fsnotify_ops;
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;
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))
127 static int fanotify_fid_info_len(int fh_len, int name_len)
129 int info_len = fh_len;
132 info_len += name_len + 1;
134 return roundup(FANOTIFY_FID_INFO_HDR_LEN + info_len,
135 FANOTIFY_EVENT_ALIGN);
138 /* FAN_RENAME may have one or two dir+name info records */
139 static int fanotify_dir_name_info_len(struct fanotify_event *event)
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);
147 info_len += fanotify_fid_info_len(dir_fh_len,
150 info_len += fanotify_fid_info_len(dir2_fh_len,
156 static size_t fanotify_event_len(unsigned int info_mode,
157 struct fanotify_event *event)
159 size_t event_len = FAN_EVENT_METADATA_LEN;
166 if (fanotify_is_error_event(event->mask))
167 event_len += FANOTIFY_ERROR_INFO_LEN;
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)) {
174 * With group flag FAN_REPORT_NAME, if name was not recorded in
175 * event on a directory, we will report the name ".".
180 if (info_mode & FAN_REPORT_PIDFD)
181 event_len += FANOTIFY_PIDFD_INFO_HDR_LEN;
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);
192 * Remove an hashed event from merge hash table.
194 static void fanotify_unhash_event(struct fsnotify_group *group,
195 struct fanotify_event *event)
197 assert_spin_locked(&group->notification_lock);
199 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
200 group, event, fanotify_event_hash_bucket(group, event));
202 if (WARN_ON_ONCE(hlist_unhashed(&event->merge_list)))
205 hlist_del_init(&event->merge_list);
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.
214 static struct fanotify_event *get_one_event(struct fsnotify_group *group,
218 struct fanotify_event *event = NULL;
219 struct fsnotify_event *fsn_event;
220 unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
222 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
224 spin_lock(&group->notification_lock);
225 fsn_event = fsnotify_peek_first_event(group);
229 event = FANOTIFY_E(fsn_event);
230 event_size = fanotify_event_len(info_mode, event);
232 if (event_size > count) {
233 event = ERR_PTR(-EINVAL);
238 * Held the notification_lock the whole time, so this is the
239 * same event we peeked above.
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);
247 spin_unlock(&group->notification_lock);
251 static int create_fd(struct fsnotify_group *group, const struct path *path,
255 struct file *new_file;
257 client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
262 * we need a new file handle for the userspace program so it can read even if it was
263 * originally opened O_WRONLY.
265 new_file = dentry_open(path,
266 group->fanotify_data.f_flags | __FMODE_NONOTIFY,
268 if (IS_ERR(new_file)) {
270 * we still send an event even if we can't open the file. this
271 * can happen when say tasks are gone and we try to open their
272 * /proc files or we try to open a WRONLY file like in sysfs
273 * we just send the errno to userspace since there isn't much
276 put_unused_fd(client_fd);
277 client_fd = PTR_ERR(new_file);
285 static int process_access_response_info(const char __user *info,
287 struct fanotify_response_info_audit_rule *friar)
289 if (info_len != sizeof(*friar))
292 if (copy_from_user(friar, info, sizeof(*friar)))
295 if (friar->hdr.type != FAN_RESPONSE_INFO_AUDIT_RULE)
297 if (friar->hdr.pad != 0)
299 if (friar->hdr.len != sizeof(*friar))
306 * Finish processing of permission event by setting it to ANSWERED state and
307 * drop group->notification_lock.
309 static void finish_permission_event(struct fsnotify_group *group,
310 struct fanotify_perm_event *event, u32 response,
311 struct fanotify_response_info_audit_rule *friar)
312 __releases(&group->notification_lock)
314 bool destroy = false;
316 assert_spin_locked(&group->notification_lock);
317 event->response = response & ~FAN_INFO;
318 if (response & FAN_INFO)
319 memcpy(&event->audit_rule, friar, sizeof(*friar));
321 if (event->state == FAN_EVENT_CANCELED)
324 event->state = FAN_EVENT_ANSWERED;
325 spin_unlock(&group->notification_lock);
327 fsnotify_destroy_event(group, &event->fae.fse);
330 static int process_access_response(struct fsnotify_group *group,
331 struct fanotify_response *response_struct,
332 const char __user *info,
335 struct fanotify_perm_event *event;
336 int fd = response_struct->fd;
337 u32 response = response_struct->response;
339 struct fanotify_response_info_audit_rule friar;
341 pr_debug("%s: group=%p fd=%d response=%u buf=%p size=%zu\n", __func__,
342 group, fd, response, info, info_len);
344 * make sure the response is valid, if invalid we do nothing and either
345 * userspace can send a valid response or we will clean it up after the
348 if (response & ~FANOTIFY_RESPONSE_VALID_MASK)
351 switch (response & FANOTIFY_RESPONSE_ACCESS) {
359 if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
362 if (response & FAN_INFO) {
363 ret = process_access_response_info(info, info_len, &friar);
375 spin_lock(&group->notification_lock);
376 list_for_each_entry(event, &group->fanotify_data.access_list,
381 list_del_init(&event->fae.fse.list);
382 finish_permission_event(group, event, response, &friar);
383 wake_up(&group->fanotify_data.access_waitq);
386 spin_unlock(&group->notification_lock);
391 static size_t copy_error_info_to_user(struct fanotify_event *event,
392 char __user *buf, int count)
394 struct fanotify_event_info_error info = { };
395 struct fanotify_error_event *fee = FANOTIFY_EE(event);
397 info.hdr.info_type = FAN_EVENT_INFO_TYPE_ERROR;
398 info.hdr.len = FANOTIFY_ERROR_INFO_LEN;
400 if (WARN_ON(count < info.hdr.len))
403 info.error = fee->error;
404 info.error_count = fee->err_count;
406 if (copy_to_user(buf, &info, sizeof(info)))
412 static int copy_fid_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
413 int info_type, const char *name,
415 char __user *buf, size_t count)
417 struct fanotify_event_info_fid info = { };
418 struct file_handle handle = { };
419 unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
420 size_t fh_len = fh ? fh->len : 0;
421 size_t info_len = fanotify_fid_info_len(fh_len, name_len);
422 size_t len = info_len;
424 pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
425 __func__, fh_len, name_len, info_len, count);
427 if (WARN_ON_ONCE(len < sizeof(info) || len > count))
431 * Copy event info fid header followed by variable sized file handle
432 * and optionally followed by variable sized filename.
435 case FAN_EVENT_INFO_TYPE_FID:
436 case FAN_EVENT_INFO_TYPE_DFID:
437 if (WARN_ON_ONCE(name_len))
440 case FAN_EVENT_INFO_TYPE_DFID_NAME:
441 case FAN_EVENT_INFO_TYPE_OLD_DFID_NAME:
442 case FAN_EVENT_INFO_TYPE_NEW_DFID_NAME:
443 if (WARN_ON_ONCE(!name || !name_len))
450 info.hdr.info_type = info_type;
453 if (copy_to_user(buf, &info, sizeof(info)))
458 if (WARN_ON_ONCE(len < sizeof(handle)))
461 handle.handle_type = fh->type;
462 handle.handle_bytes = fh_len;
464 /* Mangle handle_type for bad file_handle */
466 handle.handle_type = FILEID_INVALID;
468 if (copy_to_user(buf, &handle, sizeof(handle)))
471 buf += sizeof(handle);
472 len -= sizeof(handle);
473 if (WARN_ON_ONCE(len < fh_len))
477 * For an inline fh and inline file name, copy through stack to exclude
478 * the copy from usercopy hardening protections.
480 fh_buf = fanotify_fh_buf(fh);
481 if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
482 memcpy(bounce, fh_buf, fh_len);
485 if (copy_to_user(buf, fh_buf, fh_len))
492 /* Copy the filename with terminating null */
494 if (WARN_ON_ONCE(len < name_len))
497 if (copy_to_user(buf, name, name_len))
505 WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
506 if (len > 0 && clear_user(buf, len))
512 static int copy_pidfd_info_to_user(int pidfd,
516 struct fanotify_event_info_pidfd info = { };
517 size_t info_len = FANOTIFY_PIDFD_INFO_HDR_LEN;
519 if (WARN_ON_ONCE(info_len > count))
522 info.hdr.info_type = FAN_EVENT_INFO_TYPE_PIDFD;
523 info.hdr.len = info_len;
526 if (copy_to_user(buf, &info, info_len))
532 static int copy_info_records_to_user(struct fanotify_event *event,
533 struct fanotify_info *info,
534 unsigned int info_mode, int pidfd,
535 char __user *buf, size_t count)
537 int ret, total_bytes = 0, info_type = 0;
538 unsigned int fid_mode = info_mode & FANOTIFY_FID_BITS;
539 unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
542 * Event info records order is as follows:
544 * 2. (optional) new dir fid + new name
545 * 3. (optional) child fid
547 if (fanotify_event_has_dir_fh(event)) {
548 info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
549 FAN_EVENT_INFO_TYPE_DFID;
551 /* FAN_RENAME uses special info types */
552 if (event->mask & FAN_RENAME)
553 info_type = FAN_EVENT_INFO_TYPE_OLD_DFID_NAME;
555 ret = copy_fid_info_to_user(fanotify_event_fsid(event),
556 fanotify_info_dir_fh(info),
558 fanotify_info_name(info),
559 info->name_len, buf, count);
568 /* New dir fid+name may be reported in addition to old dir fid+name */
569 if (fanotify_event_has_dir2_fh(event)) {
570 info_type = FAN_EVENT_INFO_TYPE_NEW_DFID_NAME;
571 ret = copy_fid_info_to_user(fanotify_event_fsid(event),
572 fanotify_info_dir2_fh(info),
574 fanotify_info_name2(info),
575 info->name2_len, buf, count);
584 if (fanotify_event_has_object_fh(event)) {
585 const char *dot = NULL;
588 if (fid_mode == FAN_REPORT_FID || info_type) {
590 * With only group flag FAN_REPORT_FID only type FID is
591 * reported. Second info record type is always FID.
593 info_type = FAN_EVENT_INFO_TYPE_FID;
594 } else if ((fid_mode & FAN_REPORT_NAME) &&
595 (event->mask & FAN_ONDIR)) {
597 * With group flag FAN_REPORT_NAME, if name was not
598 * recorded in an event on a directory, report the name
599 * "." with info type DFID_NAME.
601 info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
604 } else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
605 (event->mask & FAN_ONDIR)) {
607 * With group flag FAN_REPORT_DIR_FID, a single info
608 * record has type DFID for directory entry modification
609 * event and for event on a directory.
611 info_type = FAN_EVENT_INFO_TYPE_DFID;
614 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
615 * a single info record has type FID for event on a
616 * non-directory, when there is no directory to report.
617 * For example, on FAN_DELETE_SELF event.
619 info_type = FAN_EVENT_INFO_TYPE_FID;
622 ret = copy_fid_info_to_user(fanotify_event_fsid(event),
623 fanotify_event_object_fh(event),
624 info_type, dot, dot_len,
635 ret = copy_pidfd_info_to_user(pidfd, buf, count);
644 if (fanotify_is_error_event(event->mask)) {
645 ret = copy_error_info_to_user(event, buf, count);
656 static ssize_t copy_event_to_user(struct fsnotify_group *group,
657 struct fanotify_event *event,
658 char __user *buf, size_t count)
660 struct fanotify_event_metadata metadata;
661 const struct path *path = fanotify_event_path(event);
662 struct fanotify_info *info = fanotify_event_info(event);
663 unsigned int info_mode = FAN_GROUP_FLAG(group, FANOTIFY_INFO_MODES);
664 unsigned int pidfd_mode = info_mode & FAN_REPORT_PIDFD;
665 struct file *f = NULL, *pidfd_file = NULL;
666 int ret, pidfd = FAN_NOPIDFD, fd = FAN_NOFD;
668 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
670 metadata.event_len = fanotify_event_len(info_mode, event);
671 metadata.metadata_len = FAN_EVENT_METADATA_LEN;
672 metadata.vers = FANOTIFY_METADATA_VERSION;
673 metadata.reserved = 0;
674 metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
675 metadata.pid = pid_vnr(event->pid);
677 * For an unprivileged listener, event->pid can be used to identify the
678 * events generated by the listener process itself, without disclosing
679 * the pids of other processes.
681 if (FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
682 task_tgid(current) != event->pid)
686 * For now, fid mode is required for an unprivileged listener and
687 * fid mode does not report fd in events. Keep this check anyway
688 * for safety in case fid mode requirement is relaxed in the future
689 * to allow unprivileged listener to get events with no fd and no fid.
691 if (!FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV) &&
692 path && path->mnt && path->dentry) {
693 fd = create_fd(group, path, &f);
701 * Complain if the FAN_REPORT_PIDFD and FAN_REPORT_TID mutual
702 * exclusion is ever lifted. At the time of incoporating pidfd
703 * support within fanotify, the pidfd API only supported the
704 * creation of pidfds for thread-group leaders.
706 WARN_ON_ONCE(FAN_GROUP_FLAG(group, FAN_REPORT_TID));
709 * The PIDTYPE_TGID check for an event->pid is performed
710 * preemptively in an attempt to catch out cases where the event
711 * listener reads events after the event generating process has
712 * already terminated. Report FAN_NOPIDFD to the event listener
713 * in those cases, with all other pidfd creation errors being
714 * reported as FAN_EPIDFD.
716 if (metadata.pid == 0 ||
717 !pid_has_task(event->pid, PIDTYPE_TGID)) {
720 pidfd = pidfd_prepare(event->pid, 0, &pidfd_file);
728 * Sanity check copy size in case get_one_event() and
729 * event_len sizes ever get out of sync.
731 if (WARN_ON_ONCE(metadata.event_len > count))
734 if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
737 buf += FAN_EVENT_METADATA_LEN;
738 count -= FAN_EVENT_METADATA_LEN;
740 if (fanotify_is_perm_event(event->mask))
741 FANOTIFY_PERM(event)->fd = fd;
744 ret = copy_info_records_to_user(event, info, info_mode, pidfd,
754 fd_install(pidfd, pidfd_file);
756 return metadata.event_len;
759 if (fd != FAN_NOFD) {
765 put_unused_fd(pidfd);
772 /* intofiy userspace file descriptor functions */
773 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
775 struct fsnotify_group *group = file->private_data;
778 poll_wait(file, &group->notification_waitq, wait);
779 spin_lock(&group->notification_lock);
780 if (!fsnotify_notify_queue_is_empty(group))
781 ret = EPOLLIN | EPOLLRDNORM;
782 spin_unlock(&group->notification_lock);
787 static ssize_t fanotify_read(struct file *file, char __user *buf,
788 size_t count, loff_t *pos)
790 struct fsnotify_group *group;
791 struct fanotify_event *event;
794 DEFINE_WAIT_FUNC(wait, woken_wake_function);
797 group = file->private_data;
799 pr_debug("%s: group=%p\n", __func__, group);
801 add_wait_queue(&group->notification_waitq, &wait);
804 * User can supply arbitrarily large buffer. Avoid softlockups
805 * in case there are lots of available events.
808 event = get_one_event(group, count);
810 ret = PTR_ERR(event);
816 if (file->f_flags & O_NONBLOCK)
820 if (signal_pending(current))
826 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
830 ret = copy_event_to_user(group, event, buf, count);
831 if (unlikely(ret == -EOPENSTALE)) {
833 * We cannot report events with stale fd so drop it.
834 * Setting ret to 0 will continue the event loop and
835 * do the right thing if there are no more events to
836 * read (i.e. return bytes read, -EAGAIN or wait).
842 * Permission events get queued to wait for response. Other
843 * events can be destroyed now.
845 if (!fanotify_is_perm_event(event->mask)) {
846 fsnotify_destroy_event(group, &event->fse);
849 spin_lock(&group->notification_lock);
850 finish_permission_event(group,
851 FANOTIFY_PERM(event), FAN_DENY, NULL);
852 wake_up(&group->fanotify_data.access_waitq);
854 spin_lock(&group->notification_lock);
855 list_add_tail(&event->fse.list,
856 &group->fanotify_data.access_list);
857 spin_unlock(&group->notification_lock);
865 remove_wait_queue(&group->notification_waitq, &wait);
867 if (start != buf && ret != -EFAULT)
872 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
874 struct fanotify_response response;
875 struct fsnotify_group *group;
877 const char __user *info_buf = buf + sizeof(struct fanotify_response);
880 if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
883 group = file->private_data;
885 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
887 if (count < sizeof(response))
890 if (copy_from_user(&response, buf, sizeof(response)))
893 info_len = count - sizeof(response);
895 ret = process_access_response(group, &response, info_buf, info_len);
899 count = sizeof(response) + ret;
904 static int fanotify_release(struct inode *ignored, struct file *file)
906 struct fsnotify_group *group = file->private_data;
907 struct fsnotify_event *fsn_event;
910 * Stop new events from arriving in the notification queue. since
911 * userspace cannot use fanotify fd anymore, no event can enter or
912 * leave access_list by now either.
914 fsnotify_group_stop_queueing(group);
917 * Process all permission events on access_list and notification queue
918 * and simulate reply from userspace.
920 spin_lock(&group->notification_lock);
921 while (!list_empty(&group->fanotify_data.access_list)) {
922 struct fanotify_perm_event *event;
924 event = list_first_entry(&group->fanotify_data.access_list,
925 struct fanotify_perm_event, fae.fse.list);
926 list_del_init(&event->fae.fse.list);
927 finish_permission_event(group, event, FAN_ALLOW, NULL);
928 spin_lock(&group->notification_lock);
932 * Destroy all non-permission events. For permission events just
933 * dequeue them and set the response. They will be freed once the
934 * response is consumed and fanotify_get_response() returns.
936 while ((fsn_event = fsnotify_remove_first_event(group))) {
937 struct fanotify_event *event = FANOTIFY_E(fsn_event);
939 if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
940 spin_unlock(&group->notification_lock);
941 fsnotify_destroy_event(group, fsn_event);
943 finish_permission_event(group, FANOTIFY_PERM(event),
946 spin_lock(&group->notification_lock);
948 spin_unlock(&group->notification_lock);
950 /* Response for all permission events it set, wakeup waiters */
951 wake_up(&group->fanotify_data.access_waitq);
953 /* matches the fanotify_init->fsnotify_alloc_group */
954 fsnotify_destroy_group(group);
959 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
961 struct fsnotify_group *group;
962 struct fsnotify_event *fsn_event;
967 group = file->private_data;
969 p = (void __user *) arg;
973 spin_lock(&group->notification_lock);
974 list_for_each_entry(fsn_event, &group->notification_list, list)
975 send_len += FAN_EVENT_METADATA_LEN;
976 spin_unlock(&group->notification_lock);
977 ret = put_user(send_len, (int __user *) p);
984 static const struct file_operations fanotify_fops = {
985 .show_fdinfo = fanotify_show_fdinfo,
986 .poll = fanotify_poll,
987 .read = fanotify_read,
988 .write = fanotify_write,
990 .release = fanotify_release,
991 .unlocked_ioctl = fanotify_ioctl,
992 .compat_ioctl = compat_ptr_ioctl,
993 .llseek = noop_llseek,
996 static int fanotify_find_path(int dfd, const char __user *filename,
997 struct path *path, unsigned int flags, __u64 mask,
998 unsigned int obj_type)
1002 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
1003 dfd, filename, flags);
1005 if (filename == NULL) {
1006 struct fd f = fdget(dfd);
1013 if ((flags & FAN_MARK_ONLYDIR) &&
1014 !(S_ISDIR(file_inode(f.file)->i_mode))) {
1019 *path = f.file->f_path;
1023 unsigned int lookup_flags = 0;
1025 if (!(flags & FAN_MARK_DONT_FOLLOW))
1026 lookup_flags |= LOOKUP_FOLLOW;
1027 if (flags & FAN_MARK_ONLYDIR)
1028 lookup_flags |= LOOKUP_DIRECTORY;
1030 ret = user_path_at(dfd, filename, lookup_flags, path);
1035 /* you can only watch an inode if you have read permissions on it */
1036 ret = path_permission(path, MAY_READ);
1042 ret = security_path_notify(path, mask, obj_type);
1050 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
1051 __u32 mask, unsigned int flags,
1052 __u32 umask, int *destroy)
1054 __u32 oldmask, newmask;
1056 /* umask bits cannot be removed by user */
1058 spin_lock(&fsn_mark->lock);
1059 oldmask = fsnotify_calc_mask(fsn_mark);
1060 if (!(flags & FANOTIFY_MARK_IGNORE_BITS)) {
1061 fsn_mark->mask &= ~mask;
1063 fsn_mark->ignore_mask &= ~mask;
1065 newmask = fsnotify_calc_mask(fsn_mark);
1067 * We need to keep the mark around even if remaining mask cannot
1068 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
1069 * changes to the mask.
1070 * Destroy mark when only umask bits remain.
1072 *destroy = !((fsn_mark->mask | fsn_mark->ignore_mask) & ~umask);
1073 spin_unlock(&fsn_mark->lock);
1075 return oldmask & ~newmask;
1078 static int fanotify_remove_mark(struct fsnotify_group *group,
1079 fsnotify_connp_t *connp, __u32 mask,
1080 unsigned int flags, __u32 umask)
1082 struct fsnotify_mark *fsn_mark = NULL;
1086 fsnotify_group_lock(group);
1087 fsn_mark = fsnotify_find_mark(connp, group);
1089 fsnotify_group_unlock(group);
1093 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
1094 umask, &destroy_mark);
1095 if (removed & fsnotify_conn_mask(fsn_mark->connector))
1096 fsnotify_recalc_mask(fsn_mark->connector);
1098 fsnotify_detach_mark(fsn_mark);
1099 fsnotify_group_unlock(group);
1101 fsnotify_free_mark(fsn_mark);
1103 /* matches the fsnotify_find_mark() */
1104 fsnotify_put_mark(fsn_mark);
1108 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
1109 struct vfsmount *mnt, __u32 mask,
1110 unsigned int flags, __u32 umask)
1112 return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
1113 mask, flags, umask);
1116 static int fanotify_remove_sb_mark(struct fsnotify_group *group,
1117 struct super_block *sb, __u32 mask,
1118 unsigned int flags, __u32 umask)
1120 return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
1124 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
1125 struct inode *inode, __u32 mask,
1126 unsigned int flags, __u32 umask)
1128 return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
1132 static bool fanotify_mark_update_flags(struct fsnotify_mark *fsn_mark,
1133 unsigned int fan_flags)
1135 bool want_iref = !(fan_flags & FAN_MARK_EVICTABLE);
1136 unsigned int ignore = fan_flags & FANOTIFY_MARK_IGNORE_BITS;
1137 bool recalc = false;
1140 * When using FAN_MARK_IGNORE for the first time, mark starts using
1141 * independent event flags in ignore mask. After that, trying to
1142 * update the ignore mask with the old FAN_MARK_IGNORED_MASK API
1143 * will result in EEXIST error.
1145 if (ignore == FAN_MARK_IGNORE)
1146 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS;
1149 * Setting FAN_MARK_IGNORED_SURV_MODIFY for the first time may lead to
1150 * the removal of the FS_MODIFY bit in calculated mask if it was set
1151 * because of an ignore mask that is now going to survive FS_MODIFY.
1153 if (ignore && (fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1154 !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) {
1155 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
1156 if (!(fsn_mark->mask & FS_MODIFY))
1160 if (fsn_mark->connector->type != FSNOTIFY_OBJ_TYPE_INODE ||
1161 want_iref == !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1165 * NO_IREF may be removed from a mark, but not added.
1166 * When removed, fsnotify_recalc_mask() will take the inode ref.
1168 WARN_ON_ONCE(!want_iref);
1169 fsn_mark->flags &= ~FSNOTIFY_MARK_FLAG_NO_IREF;
1174 static bool fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
1175 __u32 mask, unsigned int fan_flags)
1179 spin_lock(&fsn_mark->lock);
1180 if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS))
1181 fsn_mark->mask |= mask;
1183 fsn_mark->ignore_mask |= mask;
1185 recalc = fsnotify_calc_mask(fsn_mark) &
1186 ~fsnotify_conn_mask(fsn_mark->connector);
1188 recalc |= fanotify_mark_update_flags(fsn_mark, fan_flags);
1189 spin_unlock(&fsn_mark->lock);
1195 struct super_block *sb;
1200 static int fanotify_set_mark_fsid(struct fsnotify_group *group,
1201 struct fsnotify_mark *mark,
1202 struct fan_fsid *fsid)
1204 struct fsnotify_mark_connector *conn;
1205 struct fsnotify_mark *old;
1206 struct super_block *old_sb = NULL;
1208 FANOTIFY_MARK(mark)->fsid = fsid->id;
1209 mark->flags |= FSNOTIFY_MARK_FLAG_HAS_FSID;
1211 mark->flags |= FSNOTIFY_MARK_FLAG_WEAK_FSID;
1213 /* First mark added will determine if group is single or multi fsid */
1214 if (list_empty(&group->marks_list))
1217 /* Find sb of an existing mark */
1218 list_for_each_entry(old, &group->marks_list, g_list) {
1219 conn = READ_ONCE(old->connector);
1222 old_sb = fsnotify_connector_sb(conn);
1227 /* Only detached marks left? */
1231 /* Do not allow mixing of marks with weak and strong fsid */
1232 if ((mark->flags ^ old->flags) & FSNOTIFY_MARK_FLAG_WEAK_FSID)
1235 /* Allow mixing of marks with strong fsid from different fs */
1239 /* Do not allow mixing marks with weak fsid from different fs */
1240 if (old_sb != fsid->sb)
1243 /* Do not allow mixing marks from different btrfs sub-volumes */
1244 if (!fanotify_fsid_equal(&FANOTIFY_MARK(old)->fsid,
1245 &FANOTIFY_MARK(mark)->fsid))
1251 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
1252 fsnotify_connp_t *connp,
1253 unsigned int obj_type,
1254 unsigned int fan_flags,
1255 struct fan_fsid *fsid)
1257 struct ucounts *ucounts = group->fanotify_data.ucounts;
1258 struct fanotify_mark *fan_mark;
1259 struct fsnotify_mark *mark;
1263 * Enforce per user marks limits per user in all containing user ns.
1264 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
1265 * in the limited groups account.
1267 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
1268 !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
1269 return ERR_PTR(-ENOSPC);
1271 fan_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
1274 goto out_dec_ucounts;
1277 mark = &fan_mark->fsn_mark;
1278 fsnotify_init_mark(mark, group);
1279 if (fan_flags & FAN_MARK_EVICTABLE)
1280 mark->flags |= FSNOTIFY_MARK_FLAG_NO_IREF;
1282 /* Cache fsid of filesystem containing the marked object */
1284 ret = fanotify_set_mark_fsid(group, mark, fsid);
1288 fan_mark->fsid.val[0] = fan_mark->fsid.val[1] = 0;
1291 ret = fsnotify_add_mark_locked(mark, connp, obj_type, 0);
1298 fsnotify_put_mark(mark);
1300 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1301 dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
1302 return ERR_PTR(ret);
1305 static int fanotify_group_init_error_pool(struct fsnotify_group *group)
1307 if (mempool_initialized(&group->fanotify_data.error_events_pool))
1310 return mempool_init_kmalloc_pool(&group->fanotify_data.error_events_pool,
1311 FANOTIFY_DEFAULT_FEE_POOL_SIZE,
1312 sizeof(struct fanotify_error_event));
1315 static int fanotify_may_update_existing_mark(struct fsnotify_mark *fsn_mark,
1316 unsigned int fan_flags)
1319 * Non evictable mark cannot be downgraded to evictable mark.
1321 if (fan_flags & FAN_MARK_EVICTABLE &&
1322 !(fsn_mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
1326 * New ignore mask semantics cannot be downgraded to old semantics.
1328 if (fan_flags & FAN_MARK_IGNORED_MASK &&
1329 fsn_mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS)
1333 * An ignore mask that survives modify could never be downgraded to not
1334 * survive modify. With new FAN_MARK_IGNORE semantics we make that rule
1335 * explicit and return an error when trying to update the ignore mask
1336 * without the original FAN_MARK_IGNORED_SURV_MODIFY value.
1338 if (fan_flags & FAN_MARK_IGNORE &&
1339 !(fan_flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1340 fsn_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)
1346 static int fanotify_add_mark(struct fsnotify_group *group,
1347 fsnotify_connp_t *connp, unsigned int obj_type,
1348 __u32 mask, unsigned int fan_flags,
1349 struct fan_fsid *fsid)
1351 struct fsnotify_mark *fsn_mark;
1355 fsnotify_group_lock(group);
1356 fsn_mark = fsnotify_find_mark(connp, group);
1358 fsn_mark = fanotify_add_new_mark(group, connp, obj_type,
1360 if (IS_ERR(fsn_mark)) {
1361 fsnotify_group_unlock(group);
1362 return PTR_ERR(fsn_mark);
1367 * Check if requested mark flags conflict with an existing mark flags.
1369 ret = fanotify_may_update_existing_mark(fsn_mark, fan_flags);
1374 * Error events are pre-allocated per group, only if strictly
1375 * needed (i.e. FAN_FS_ERROR was requested).
1377 if (!(fan_flags & FANOTIFY_MARK_IGNORE_BITS) &&
1378 (mask & FAN_FS_ERROR)) {
1379 ret = fanotify_group_init_error_pool(group);
1384 recalc = fanotify_mark_add_to_mask(fsn_mark, mask, fan_flags);
1386 fsnotify_recalc_mask(fsn_mark->connector);
1389 fsnotify_group_unlock(group);
1391 fsnotify_put_mark(fsn_mark);
1395 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
1396 struct vfsmount *mnt, __u32 mask,
1397 unsigned int flags, struct fan_fsid *fsid)
1399 return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
1400 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
1403 static int fanotify_add_sb_mark(struct fsnotify_group *group,
1404 struct super_block *sb, __u32 mask,
1405 unsigned int flags, struct fan_fsid *fsid)
1407 return fanotify_add_mark(group, &sb->s_fsnotify_marks,
1408 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
1411 static int fanotify_add_inode_mark(struct fsnotify_group *group,
1412 struct inode *inode, __u32 mask,
1413 unsigned int flags, struct fan_fsid *fsid)
1415 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
1418 * If some other task has this inode open for write we should not add
1419 * an ignore mask, unless that ignore mask is supposed to survive
1420 * modification changes anyway.
1422 if ((flags & FANOTIFY_MARK_IGNORE_BITS) &&
1423 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
1424 inode_is_open_for_write(inode))
1427 return fanotify_add_mark(group, &inode->i_fsnotify_marks,
1428 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
1431 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1433 struct fanotify_event *oevent;
1435 oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1439 fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1440 oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1442 return &oevent->fse;
1445 static struct hlist_head *fanotify_alloc_merge_hash(void)
1447 struct hlist_head *hash;
1449 hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1450 GFP_KERNEL_ACCOUNT);
1454 __hash_init(hash, FANOTIFY_HTABLE_SIZE);
1459 /* fanotify syscalls */
1460 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1462 struct fsnotify_group *group;
1464 unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1465 unsigned int class = flags & FANOTIFY_CLASS_BITS;
1466 unsigned int internal_flags = 0;
1468 pr_debug("%s: flags=%x event_f_flags=%x\n",
1469 __func__, flags, event_f_flags);
1471 if (!capable(CAP_SYS_ADMIN)) {
1473 * An unprivileged user can setup an fanotify group with
1474 * limited functionality - an unprivileged group is limited to
1475 * notification events with file handles and it cannot use
1476 * unlimited queue/marks.
1478 if ((flags & FANOTIFY_ADMIN_INIT_FLAGS) || !fid_mode)
1482 * Setting the internal flag FANOTIFY_UNPRIV on the group
1483 * prevents setting mount/filesystem marks on this group and
1484 * prevents reporting pid and open fd in events.
1486 internal_flags |= FANOTIFY_UNPRIV;
1489 #ifdef CONFIG_AUDITSYSCALL
1490 if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1492 if (flags & ~FANOTIFY_INIT_FLAGS)
1497 * A pidfd can only be returned for a thread-group leader; thus
1498 * FAN_REPORT_PIDFD and FAN_REPORT_TID need to remain mutually
1501 if ((flags & FAN_REPORT_PIDFD) && (flags & FAN_REPORT_TID))
1504 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1507 switch (event_f_flags & O_ACCMODE) {
1516 if (fid_mode && class != FAN_CLASS_NOTIF)
1520 * Child name is reported with parent fid so requires dir fid.
1521 * We can report both child fid and dir fid with or without name.
1523 if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1527 * FAN_REPORT_TARGET_FID requires FAN_REPORT_NAME and FAN_REPORT_FID
1528 * and is used as an indication to report both dir and child fid on all
1531 if ((fid_mode & FAN_REPORT_TARGET_FID) &&
1532 (!(fid_mode & FAN_REPORT_NAME) || !(fid_mode & FAN_REPORT_FID)))
1535 f_flags = O_RDWR | __FMODE_NONOTIFY;
1536 if (flags & FAN_CLOEXEC)
1537 f_flags |= O_CLOEXEC;
1538 if (flags & FAN_NONBLOCK)
1539 f_flags |= O_NONBLOCK;
1541 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
1542 group = fsnotify_alloc_group(&fanotify_fsnotify_ops,
1543 FSNOTIFY_GROUP_USER | FSNOTIFY_GROUP_NOFS);
1544 if (IS_ERR(group)) {
1545 return PTR_ERR(group);
1548 /* Enforce groups limits per user in all containing user ns */
1549 group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1551 UCOUNT_FANOTIFY_GROUPS);
1552 if (!group->fanotify_data.ucounts) {
1554 goto out_destroy_group;
1557 group->fanotify_data.flags = flags | internal_flags;
1558 group->memcg = get_mem_cgroup_from_mm(current->mm);
1560 group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1561 if (!group->fanotify_data.merge_hash) {
1563 goto out_destroy_group;
1566 group->overflow_event = fanotify_alloc_overflow_event();
1567 if (unlikely(!group->overflow_event)) {
1569 goto out_destroy_group;
1572 if (force_o_largefile())
1573 event_f_flags |= O_LARGEFILE;
1574 group->fanotify_data.f_flags = event_f_flags;
1575 init_waitqueue_head(&group->fanotify_data.access_waitq);
1576 INIT_LIST_HEAD(&group->fanotify_data.access_list);
1578 case FAN_CLASS_NOTIF:
1579 group->priority = FS_PRIO_0;
1581 case FAN_CLASS_CONTENT:
1582 group->priority = FS_PRIO_1;
1584 case FAN_CLASS_PRE_CONTENT:
1585 group->priority = FS_PRIO_2;
1589 goto out_destroy_group;
1592 if (flags & FAN_UNLIMITED_QUEUE) {
1594 if (!capable(CAP_SYS_ADMIN))
1595 goto out_destroy_group;
1596 group->max_events = UINT_MAX;
1598 group->max_events = fanotify_max_queued_events;
1601 if (flags & FAN_UNLIMITED_MARKS) {
1603 if (!capable(CAP_SYS_ADMIN))
1604 goto out_destroy_group;
1607 if (flags & FAN_ENABLE_AUDIT) {
1609 if (!capable(CAP_AUDIT_WRITE))
1610 goto out_destroy_group;
1613 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
1615 goto out_destroy_group;
1620 fsnotify_destroy_group(group);
1624 static int fanotify_test_fsid(struct dentry *dentry, unsigned int flags,
1625 struct fan_fsid *fsid)
1627 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1628 __kernel_fsid_t root_fsid;
1632 * Make sure dentry is not of a filesystem with zero fsid (e.g. fuse).
1634 err = vfs_get_fsid(dentry, &fsid->id);
1638 fsid->sb = dentry->d_sb;
1639 if (!fsid->id.val[0] && !fsid->id.val[1]) {
1645 * Make sure dentry is not of a filesystem subvolume (e.g. btrfs)
1646 * which uses a different fsid than sb root.
1648 err = vfs_get_fsid(dentry->d_sb->s_root, &root_fsid);
1652 if (!fanotify_fsid_equal(&root_fsid, &fsid->id)) {
1661 /* Allow weak fsid when marking inodes */
1663 return (mark_type == FAN_MARK_INODE) ? 0 : err;
1666 /* Check if filesystem can encode a unique fid */
1667 static int fanotify_test_fid(struct dentry *dentry, unsigned int flags)
1669 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1670 const struct export_operations *nop = dentry->d_sb->s_export_op;
1673 * We need to make sure that the filesystem supports encoding of
1674 * file handles so user can use name_to_handle_at() to compare fids
1675 * reported with events to the file handle of watched objects.
1677 if (!exportfs_can_encode_fid(nop))
1681 * For sb/mount mark, we also need to make sure that the filesystem
1682 * supports decoding file handles, so user has a way to map back the
1683 * reported fids to filesystem objects.
1685 if (mark_type != FAN_MARK_INODE && !exportfs_can_decode_fh(nop))
1691 static int fanotify_events_supported(struct fsnotify_group *group,
1692 const struct path *path, __u64 mask,
1695 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1696 /* Strict validation of events in non-dir inode mask with v5.17+ APIs */
1697 bool strict_dir_events = FAN_GROUP_FLAG(group, FAN_REPORT_TARGET_FID) ||
1698 (mask & FAN_RENAME) ||
1699 (flags & FAN_MARK_IGNORE);
1702 * Some filesystems such as 'proc' acquire unusual locks when opening
1703 * files. For them fanotify permission events have high chances of
1704 * deadlocking the system - open done when reporting fanotify event
1705 * blocks on this "unusual" lock while another process holding the lock
1706 * waits for fanotify permission event to be answered. Just disallow
1707 * permission events for such filesystems.
1709 if (mask & FANOTIFY_PERM_EVENTS &&
1710 path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1714 * mount and sb marks are not allowed on kernel internal pseudo fs,
1715 * like pipe_mnt, because that would subscribe to events on all the
1716 * anonynous pipes in the system.
1718 * SB_NOUSER covers all of the internal pseudo fs whose objects are not
1719 * exposed to user's mount namespace, but there are other SB_KERNMOUNT
1720 * fs, like nsfs, debugfs, for which the value of allowing sb and mount
1721 * mark is questionable. For now we leave them alone.
1723 if (mark_type != FAN_MARK_INODE &&
1724 path->mnt->mnt_sb->s_flags & SB_NOUSER)
1728 * We shouldn't have allowed setting dirent events and the directory
1729 * flags FAN_ONDIR and FAN_EVENT_ON_CHILD in mask of non-dir inode,
1730 * but because we always allowed it, error only when using new APIs.
1732 if (strict_dir_events && mark_type == FAN_MARK_INODE &&
1733 !d_is_dir(path->dentry) && (mask & FANOTIFY_DIRONLY_EVENT_BITS))
1739 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1740 int dfd, const char __user *pathname)
1742 struct inode *inode = NULL;
1743 struct vfsmount *mnt = NULL;
1744 struct fsnotify_group *group;
1747 struct fan_fsid __fsid, *fsid = NULL;
1748 u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1749 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1750 unsigned int mark_cmd = flags & FANOTIFY_MARK_CMD_BITS;
1751 unsigned int ignore = flags & FANOTIFY_MARK_IGNORE_BITS;
1752 unsigned int obj_type, fid_mode;
1756 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1757 __func__, fanotify_fd, flags, dfd, pathname, mask);
1759 /* we only use the lower 32 bits as of right now. */
1760 if (upper_32_bits(mask))
1763 if (flags & ~FANOTIFY_MARK_FLAGS)
1766 switch (mark_type) {
1767 case FAN_MARK_INODE:
1768 obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1770 case FAN_MARK_MOUNT:
1771 obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1773 case FAN_MARK_FILESYSTEM:
1774 obj_type = FSNOTIFY_OBJ_TYPE_SB;
1782 case FAN_MARK_REMOVE:
1786 case FAN_MARK_FLUSH:
1787 if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1794 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1795 valid_mask |= FANOTIFY_PERM_EVENTS;
1797 if (mask & ~valid_mask)
1801 /* We don't allow FAN_MARK_IGNORE & FAN_MARK_IGNORED_MASK together */
1802 if (ignore == (FAN_MARK_IGNORE | FAN_MARK_IGNORED_MASK))
1806 * Event flags (FAN_ONDIR, FAN_EVENT_ON_CHILD) have no effect with
1807 * FAN_MARK_IGNORED_MASK.
1809 if (ignore == FAN_MARK_IGNORED_MASK) {
1810 mask &= ~FANOTIFY_EVENT_FLAGS;
1811 umask = FANOTIFY_EVENT_FLAGS;
1814 f = fdget(fanotify_fd);
1815 if (unlikely(!f.file))
1818 /* verify that this is indeed an fanotify instance */
1820 if (unlikely(f.file->f_op != &fanotify_fops))
1822 group = f.file->private_data;
1825 * An unprivileged user is not allowed to setup mount nor filesystem
1826 * marks. This also includes setting up such marks by a group that
1827 * was initialized by an unprivileged user.
1830 if ((!capable(CAP_SYS_ADMIN) ||
1831 FAN_GROUP_FLAG(group, FANOTIFY_UNPRIV)) &&
1832 mark_type != FAN_MARK_INODE)
1836 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
1837 * allowed to set permissions events.
1840 if (mask & FANOTIFY_PERM_EVENTS &&
1841 group->priority == FS_PRIO_0)
1844 if (mask & FAN_FS_ERROR &&
1845 mark_type != FAN_MARK_FILESYSTEM)
1849 * Evictable is only relevant for inode marks, because only inode object
1850 * can be evicted on memory pressure.
1852 if (flags & FAN_MARK_EVICTABLE &&
1853 mark_type != FAN_MARK_INODE)
1857 * Events that do not carry enough information to report
1858 * event->fd require a group that supports reporting fid. Those
1859 * events are not supported on a mount mark, because they do not
1860 * carry enough information (i.e. path) to be filtered by mount
1863 fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1864 if (mask & ~(FANOTIFY_FD_EVENTS|FANOTIFY_EVENT_FLAGS) &&
1865 (!fid_mode || mark_type == FAN_MARK_MOUNT))
1869 * FAN_RENAME uses special info type records to report the old and
1870 * new parent+name. Reporting only old and new parent id is less
1871 * useful and was not implemented.
1873 if (mask & FAN_RENAME && !(fid_mode & FAN_REPORT_NAME))
1876 if (mark_cmd == FAN_MARK_FLUSH) {
1878 if (mark_type == FAN_MARK_MOUNT)
1879 fsnotify_clear_vfsmount_marks_by_group(group);
1880 else if (mark_type == FAN_MARK_FILESYSTEM)
1881 fsnotify_clear_sb_marks_by_group(group);
1883 fsnotify_clear_inode_marks_by_group(group);
1887 ret = fanotify_find_path(dfd, pathname, &path, flags,
1888 (mask & ALL_FSNOTIFY_EVENTS), obj_type);
1892 if (mark_cmd == FAN_MARK_ADD) {
1893 ret = fanotify_events_supported(group, &path, mask, flags);
1895 goto path_put_and_out;
1899 ret = fanotify_test_fsid(path.dentry, flags, &__fsid);
1901 goto path_put_and_out;
1903 ret = fanotify_test_fid(path.dentry, flags);
1905 goto path_put_and_out;
1910 /* inode held in place by reference to path; group by fget on fd */
1911 if (mark_type == FAN_MARK_INODE)
1912 inode = path.dentry->d_inode;
1916 ret = mnt ? -EINVAL : -EISDIR;
1917 /* FAN_MARK_IGNORE requires SURV_MODIFY for sb/mount/dir marks */
1918 if (mark_cmd == FAN_MARK_ADD && ignore == FAN_MARK_IGNORE &&
1919 (mnt || S_ISDIR(inode->i_mode)) &&
1920 !(flags & FAN_MARK_IGNORED_SURV_MODIFY))
1921 goto path_put_and_out;
1923 /* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1924 if (mnt || !S_ISDIR(inode->i_mode)) {
1925 mask &= ~FAN_EVENT_ON_CHILD;
1926 umask = FAN_EVENT_ON_CHILD;
1928 * If group needs to report parent fid, register for getting
1929 * events with parent/name info for non-directory.
1931 if ((fid_mode & FAN_REPORT_DIR_FID) &&
1932 (flags & FAN_MARK_ADD) && !ignore)
1933 mask |= FAN_EVENT_ON_CHILD;
1936 /* create/update an inode mark */
1939 if (mark_type == FAN_MARK_MOUNT)
1940 ret = fanotify_add_vfsmount_mark(group, mnt, mask,
1942 else if (mark_type == FAN_MARK_FILESYSTEM)
1943 ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
1946 ret = fanotify_add_inode_mark(group, inode, mask,
1949 case FAN_MARK_REMOVE:
1950 if (mark_type == FAN_MARK_MOUNT)
1951 ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
1953 else if (mark_type == FAN_MARK_FILESYSTEM)
1954 ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
1957 ret = fanotify_remove_inode_mark(group, inode, mask,
1971 #ifndef CONFIG_ARCH_SPLIT_ARG64
1972 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1973 __u64, mask, int, dfd,
1974 const char __user *, pathname)
1976 return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1980 #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT)
1981 SYSCALL32_DEFINE6(fanotify_mark,
1982 int, fanotify_fd, unsigned int, flags,
1983 SC_ARG64(mask), int, dfd,
1984 const char __user *, pathname)
1986 return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
1992 * fanotify_user_setup - Our initialization function. Note that we cannot return
1993 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
1994 * must result in panic().
1996 static int __init fanotify_user_setup(void)
2003 * Allow up to 1% of addressable memory to be accounted for per user
2004 * marks limited to the range [8192, 1048576]. mount and sb marks are
2005 * a lot cheaper than inode marks, but there is no reason for a user
2006 * to have many of those, so calculate by the cost of inode marks.
2008 max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
2010 max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
2011 FANOTIFY_DEFAULT_MAX_USER_MARKS);
2013 BUILD_BUG_ON(FANOTIFY_INIT_FLAGS & FANOTIFY_INTERNAL_GROUP_FLAGS);
2014 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 12);
2015 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 11);
2017 fanotify_mark_cache = KMEM_CACHE(fanotify_mark,
2018 SLAB_PANIC|SLAB_ACCOUNT);
2019 fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
2021 fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
2023 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
2024 fanotify_perm_event_cachep =
2025 KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
2028 fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
2029 init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
2030 FANOTIFY_DEFAULT_MAX_GROUPS;
2031 init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
2032 fanotify_sysctls_init();
2036 device_initcall(fanotify_user_setup);