1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/fanotify.h>
3 #include <linux/fcntl.h>
4 #include <linux/file.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>
23 #include <asm/ioctls.h>
25 #include "../../mount.h"
26 #include "../fdinfo.h"
29 #define FANOTIFY_DEFAULT_MAX_EVENTS 16384
30 #define FANOTIFY_OLD_DEFAULT_MAX_MARKS 8192
31 #define FANOTIFY_DEFAULT_MAX_GROUPS 128
34 * Legacy fanotify marks limits (8192) is per group and we introduced a tunable
35 * limit of marks per user, similar to inotify. Effectively, the legacy limit
36 * of fanotify marks per user is <max marks per group> * <max groups per user>.
37 * This default limit (1M) also happens to match the increased limit of inotify
38 * max_user_watches since v5.10.
40 #define FANOTIFY_DEFAULT_MAX_USER_MARKS \
41 (FANOTIFY_OLD_DEFAULT_MAX_MARKS * FANOTIFY_DEFAULT_MAX_GROUPS)
44 * Most of the memory cost of adding an inode mark is pinning the marked inode.
45 * The size of the filesystem inode struct is not uniform across filesystems,
46 * so double the size of a VFS inode is used as a conservative approximation.
48 #define INODE_MARK_COST (2 * sizeof(struct inode))
50 /* configurable via /proc/sys/fs/fanotify/ */
51 static int fanotify_max_queued_events __read_mostly;
55 #include <linux/sysctl.h>
57 struct ctl_table fanotify_table[] = {
59 .procname = "max_user_groups",
60 .data = &init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS],
61 .maxlen = sizeof(int),
63 .proc_handler = proc_dointvec_minmax,
64 .extra1 = SYSCTL_ZERO,
67 .procname = "max_user_marks",
68 .data = &init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS],
69 .maxlen = sizeof(int),
71 .proc_handler = proc_dointvec_minmax,
72 .extra1 = SYSCTL_ZERO,
75 .procname = "max_queued_events",
76 .data = &fanotify_max_queued_events,
77 .maxlen = sizeof(int),
79 .proc_handler = proc_dointvec_minmax,
84 #endif /* CONFIG_SYSCTL */
87 * All flags that may be specified in parameter event_f_flags of fanotify_init.
89 * Internal and external open flags are stored together in field f_flags of
90 * struct file. Only external open flags shall be allowed in event_f_flags.
91 * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
94 #define FANOTIFY_INIT_ALL_EVENT_F_BITS ( \
95 O_ACCMODE | O_APPEND | O_NONBLOCK | \
96 __O_SYNC | O_DSYNC | O_CLOEXEC | \
97 O_LARGEFILE | O_NOATIME )
99 extern const struct fsnotify_ops fanotify_fsnotify_ops;
101 struct kmem_cache *fanotify_mark_cache __read_mostly;
102 struct kmem_cache *fanotify_fid_event_cachep __read_mostly;
103 struct kmem_cache *fanotify_path_event_cachep __read_mostly;
104 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
106 #define FANOTIFY_EVENT_ALIGN 4
107 #define FANOTIFY_INFO_HDR_LEN \
108 (sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle))
110 static int fanotify_fid_info_len(int fh_len, int name_len)
112 int info_len = fh_len;
115 info_len += name_len + 1;
117 return roundup(FANOTIFY_INFO_HDR_LEN + info_len, FANOTIFY_EVENT_ALIGN);
120 static int fanotify_event_info_len(unsigned int fid_mode,
121 struct fanotify_event *event)
123 struct fanotify_info *info = fanotify_event_info(event);
124 int dir_fh_len = fanotify_event_dir_fh_len(event);
125 int fh_len = fanotify_event_object_fh_len(event);
130 info_len += fanotify_fid_info_len(dir_fh_len, info->name_len);
131 } else if ((fid_mode & FAN_REPORT_NAME) && (event->mask & FAN_ONDIR)) {
133 * With group flag FAN_REPORT_NAME, if name was not recorded in
134 * event on a directory, we will report the name ".".
140 info_len += fanotify_fid_info_len(fh_len, dot_len);
146 * Remove an hashed event from merge hash table.
148 static void fanotify_unhash_event(struct fsnotify_group *group,
149 struct fanotify_event *event)
151 assert_spin_locked(&group->notification_lock);
153 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
154 group, event, fanotify_event_hash_bucket(group, event));
156 if (WARN_ON_ONCE(hlist_unhashed(&event->merge_list)))
159 hlist_del_init(&event->merge_list);
163 * Get an fanotify notification event if one exists and is small
164 * enough to fit in "count". Return an error pointer if the count
165 * is not large enough. When permission event is dequeued, its state is
166 * updated accordingly.
168 static struct fanotify_event *get_one_event(struct fsnotify_group *group,
171 size_t event_size = FAN_EVENT_METADATA_LEN;
172 struct fanotify_event *event = NULL;
173 struct fsnotify_event *fsn_event;
174 unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
176 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
178 spin_lock(&group->notification_lock);
179 fsn_event = fsnotify_peek_first_event(group);
183 event = FANOTIFY_E(fsn_event);
185 event_size += fanotify_event_info_len(fid_mode, event);
187 if (event_size > count) {
188 event = ERR_PTR(-EINVAL);
193 * Held the notification_lock the whole time, so this is the
194 * same event we peeked above.
196 fsnotify_remove_first_event(group);
197 if (fanotify_is_perm_event(event->mask))
198 FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED;
199 if (fanotify_is_hashed_event(event->mask))
200 fanotify_unhash_event(group, event);
202 spin_unlock(&group->notification_lock);
206 static int create_fd(struct fsnotify_group *group, struct path *path,
210 struct file *new_file;
212 client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
217 * we need a new file handle for the userspace program so it can read even if it was
218 * originally opened O_WRONLY.
220 new_file = dentry_open(path,
221 group->fanotify_data.f_flags | FMODE_NONOTIFY,
223 if (IS_ERR(new_file)) {
225 * we still send an event even if we can't open the file. this
226 * can happen when say tasks are gone and we try to open their
227 * /proc files or we try to open a WRONLY file like in sysfs
228 * we just send the errno to userspace since there isn't much
231 put_unused_fd(client_fd);
232 client_fd = PTR_ERR(new_file);
241 * Finish processing of permission event by setting it to ANSWERED state and
242 * drop group->notification_lock.
244 static void finish_permission_event(struct fsnotify_group *group,
245 struct fanotify_perm_event *event,
246 unsigned int response)
247 __releases(&group->notification_lock)
249 bool destroy = false;
251 assert_spin_locked(&group->notification_lock);
252 event->response = response;
253 if (event->state == FAN_EVENT_CANCELED)
256 event->state = FAN_EVENT_ANSWERED;
257 spin_unlock(&group->notification_lock);
259 fsnotify_destroy_event(group, &event->fae.fse);
262 static int process_access_response(struct fsnotify_group *group,
263 struct fanotify_response *response_struct)
265 struct fanotify_perm_event *event;
266 int fd = response_struct->fd;
267 int response = response_struct->response;
269 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
272 * make sure the response is valid, if invalid we do nothing and either
273 * userspace can send a valid response or we will clean it up after the
276 switch (response & ~FAN_AUDIT) {
287 if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
290 spin_lock(&group->notification_lock);
291 list_for_each_entry(event, &group->fanotify_data.access_list,
296 list_del_init(&event->fae.fse.list);
297 finish_permission_event(group, event, response);
298 wake_up(&group->fanotify_data.access_waitq);
301 spin_unlock(&group->notification_lock);
306 static int copy_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
307 int info_type, const char *name, size_t name_len,
308 char __user *buf, size_t count)
310 struct fanotify_event_info_fid info = { };
311 struct file_handle handle = { };
312 unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
313 size_t fh_len = fh ? fh->len : 0;
314 size_t info_len = fanotify_fid_info_len(fh_len, name_len);
315 size_t len = info_len;
317 pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
318 __func__, fh_len, name_len, info_len, count);
323 if (WARN_ON_ONCE(len < sizeof(info) || len > count))
327 * Copy event info fid header followed by variable sized file handle
328 * and optionally followed by variable sized filename.
331 case FAN_EVENT_INFO_TYPE_FID:
332 case FAN_EVENT_INFO_TYPE_DFID:
333 if (WARN_ON_ONCE(name_len))
336 case FAN_EVENT_INFO_TYPE_DFID_NAME:
337 if (WARN_ON_ONCE(!name || !name_len))
344 info.hdr.info_type = info_type;
347 if (copy_to_user(buf, &info, sizeof(info)))
352 if (WARN_ON_ONCE(len < sizeof(handle)))
355 handle.handle_type = fh->type;
356 handle.handle_bytes = fh_len;
357 if (copy_to_user(buf, &handle, sizeof(handle)))
360 buf += sizeof(handle);
361 len -= sizeof(handle);
362 if (WARN_ON_ONCE(len < fh_len))
366 * For an inline fh and inline file name, copy through stack to exclude
367 * the copy from usercopy hardening protections.
369 fh_buf = fanotify_fh_buf(fh);
370 if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
371 memcpy(bounce, fh_buf, fh_len);
374 if (copy_to_user(buf, fh_buf, fh_len))
381 /* Copy the filename with terminating null */
383 if (WARN_ON_ONCE(len < name_len))
386 if (copy_to_user(buf, name, name_len))
394 WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
395 if (len > 0 && clear_user(buf, len))
401 static ssize_t copy_event_to_user(struct fsnotify_group *group,
402 struct fanotify_event *event,
403 char __user *buf, size_t count)
405 struct fanotify_event_metadata metadata;
406 struct path *path = fanotify_event_path(event);
407 struct fanotify_info *info = fanotify_event_info(event);
408 unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
409 struct file *f = NULL;
410 int ret, fd = FAN_NOFD;
413 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
415 metadata.event_len = FAN_EVENT_METADATA_LEN +
416 fanotify_event_info_len(fid_mode, event);
417 metadata.metadata_len = FAN_EVENT_METADATA_LEN;
418 metadata.vers = FANOTIFY_METADATA_VERSION;
419 metadata.reserved = 0;
420 metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
421 metadata.pid = pid_vnr(event->pid);
423 if (path && path->mnt && path->dentry) {
424 fd = create_fd(group, path, &f);
432 * Sanity check copy size in case get_one_event() and
433 * event_len sizes ever get out of sync.
435 if (WARN_ON_ONCE(metadata.event_len > count))
438 if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
441 buf += FAN_EVENT_METADATA_LEN;
442 count -= FAN_EVENT_METADATA_LEN;
444 if (fanotify_is_perm_event(event->mask))
445 FANOTIFY_PERM(event)->fd = fd;
450 /* Event info records order is: dir fid + name, child fid */
451 if (fanotify_event_dir_fh_len(event)) {
452 info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
453 FAN_EVENT_INFO_TYPE_DFID;
454 ret = copy_info_to_user(fanotify_event_fsid(event),
455 fanotify_info_dir_fh(info),
456 info_type, fanotify_info_name(info),
457 info->name_len, buf, count);
465 if (fanotify_event_object_fh_len(event)) {
466 const char *dot = NULL;
469 if (fid_mode == FAN_REPORT_FID || info_type) {
471 * With only group flag FAN_REPORT_FID only type FID is
472 * reported. Second info record type is always FID.
474 info_type = FAN_EVENT_INFO_TYPE_FID;
475 } else if ((fid_mode & FAN_REPORT_NAME) &&
476 (event->mask & FAN_ONDIR)) {
478 * With group flag FAN_REPORT_NAME, if name was not
479 * recorded in an event on a directory, report the
480 * name "." with info type DFID_NAME.
482 info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
485 } else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
486 (event->mask & FAN_ONDIR)) {
488 * With group flag FAN_REPORT_DIR_FID, a single info
489 * record has type DFID for directory entry modification
490 * event and for event on a directory.
492 info_type = FAN_EVENT_INFO_TYPE_DFID;
495 * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID,
496 * a single info record has type FID for event on a
497 * non-directory, when there is no directory to report.
498 * For example, on FAN_DELETE_SELF event.
500 info_type = FAN_EVENT_INFO_TYPE_FID;
503 ret = copy_info_to_user(fanotify_event_fsid(event),
504 fanotify_event_object_fh(event),
505 info_type, dot, dot_len, buf, count);
513 return metadata.event_len;
516 if (fd != FAN_NOFD) {
523 /* intofiy userspace file descriptor functions */
524 static __poll_t fanotify_poll(struct file *file, poll_table *wait)
526 struct fsnotify_group *group = file->private_data;
529 poll_wait(file, &group->notification_waitq, wait);
530 spin_lock(&group->notification_lock);
531 if (!fsnotify_notify_queue_is_empty(group))
532 ret = EPOLLIN | EPOLLRDNORM;
533 spin_unlock(&group->notification_lock);
538 static ssize_t fanotify_read(struct file *file, char __user *buf,
539 size_t count, loff_t *pos)
541 struct fsnotify_group *group;
542 struct fanotify_event *event;
545 DEFINE_WAIT_FUNC(wait, woken_wake_function);
548 group = file->private_data;
550 pr_debug("%s: group=%p\n", __func__, group);
552 add_wait_queue(&group->notification_waitq, &wait);
555 * User can supply arbitrarily large buffer. Avoid softlockups
556 * in case there are lots of available events.
559 event = get_one_event(group, count);
561 ret = PTR_ERR(event);
567 if (file->f_flags & O_NONBLOCK)
571 if (signal_pending(current))
577 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
581 ret = copy_event_to_user(group, event, buf, count);
582 if (unlikely(ret == -EOPENSTALE)) {
584 * We cannot report events with stale fd so drop it.
585 * Setting ret to 0 will continue the event loop and
586 * do the right thing if there are no more events to
587 * read (i.e. return bytes read, -EAGAIN or wait).
593 * Permission events get queued to wait for response. Other
594 * events can be destroyed now.
596 if (!fanotify_is_perm_event(event->mask)) {
597 fsnotify_destroy_event(group, &event->fse);
600 spin_lock(&group->notification_lock);
601 finish_permission_event(group,
602 FANOTIFY_PERM(event), FAN_DENY);
603 wake_up(&group->fanotify_data.access_waitq);
605 spin_lock(&group->notification_lock);
606 list_add_tail(&event->fse.list,
607 &group->fanotify_data.access_list);
608 spin_unlock(&group->notification_lock);
616 remove_wait_queue(&group->notification_waitq, &wait);
618 if (start != buf && ret != -EFAULT)
623 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
625 struct fanotify_response response = { .fd = -1, .response = -1 };
626 struct fsnotify_group *group;
629 if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
632 group = file->private_data;
634 if (count < sizeof(response))
637 count = sizeof(response);
639 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
641 if (copy_from_user(&response, buf, count))
644 ret = process_access_response(group, &response);
651 static int fanotify_release(struct inode *ignored, struct file *file)
653 struct fsnotify_group *group = file->private_data;
654 struct fsnotify_event *fsn_event;
657 * Stop new events from arriving in the notification queue. since
658 * userspace cannot use fanotify fd anymore, no event can enter or
659 * leave access_list by now either.
661 fsnotify_group_stop_queueing(group);
664 * Process all permission events on access_list and notification queue
665 * and simulate reply from userspace.
667 spin_lock(&group->notification_lock);
668 while (!list_empty(&group->fanotify_data.access_list)) {
669 struct fanotify_perm_event *event;
671 event = list_first_entry(&group->fanotify_data.access_list,
672 struct fanotify_perm_event, fae.fse.list);
673 list_del_init(&event->fae.fse.list);
674 finish_permission_event(group, event, FAN_ALLOW);
675 spin_lock(&group->notification_lock);
679 * Destroy all non-permission events. For permission events just
680 * dequeue them and set the response. They will be freed once the
681 * response is consumed and fanotify_get_response() returns.
683 while ((fsn_event = fsnotify_remove_first_event(group))) {
684 struct fanotify_event *event = FANOTIFY_E(fsn_event);
686 if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
687 spin_unlock(&group->notification_lock);
688 fsnotify_destroy_event(group, fsn_event);
690 finish_permission_event(group, FANOTIFY_PERM(event),
693 spin_lock(&group->notification_lock);
695 spin_unlock(&group->notification_lock);
697 /* Response for all permission events it set, wakeup waiters */
698 wake_up(&group->fanotify_data.access_waitq);
700 /* matches the fanotify_init->fsnotify_alloc_group */
701 fsnotify_destroy_group(group);
706 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
708 struct fsnotify_group *group;
709 struct fsnotify_event *fsn_event;
714 group = file->private_data;
716 p = (void __user *) arg;
720 spin_lock(&group->notification_lock);
721 list_for_each_entry(fsn_event, &group->notification_list, list)
722 send_len += FAN_EVENT_METADATA_LEN;
723 spin_unlock(&group->notification_lock);
724 ret = put_user(send_len, (int __user *) p);
731 static const struct file_operations fanotify_fops = {
732 .show_fdinfo = fanotify_show_fdinfo,
733 .poll = fanotify_poll,
734 .read = fanotify_read,
735 .write = fanotify_write,
737 .release = fanotify_release,
738 .unlocked_ioctl = fanotify_ioctl,
739 .compat_ioctl = compat_ptr_ioctl,
740 .llseek = noop_llseek,
743 static int fanotify_find_path(int dfd, const char __user *filename,
744 struct path *path, unsigned int flags, __u64 mask,
745 unsigned int obj_type)
749 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
750 dfd, filename, flags);
752 if (filename == NULL) {
753 struct fd f = fdget(dfd);
760 if ((flags & FAN_MARK_ONLYDIR) &&
761 !(S_ISDIR(file_inode(f.file)->i_mode))) {
766 *path = f.file->f_path;
770 unsigned int lookup_flags = 0;
772 if (!(flags & FAN_MARK_DONT_FOLLOW))
773 lookup_flags |= LOOKUP_FOLLOW;
774 if (flags & FAN_MARK_ONLYDIR)
775 lookup_flags |= LOOKUP_DIRECTORY;
777 ret = user_path_at(dfd, filename, lookup_flags, path);
782 /* you can only watch an inode if you have read permissions on it */
783 ret = path_permission(path, MAY_READ);
789 ret = security_path_notify(path, mask, obj_type);
797 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
798 __u32 mask, unsigned int flags,
799 __u32 umask, int *destroy)
803 /* umask bits cannot be removed by user */
805 spin_lock(&fsn_mark->lock);
806 if (!(flags & FAN_MARK_IGNORED_MASK)) {
807 oldmask = fsn_mark->mask;
808 fsn_mark->mask &= ~mask;
810 fsn_mark->ignored_mask &= ~mask;
813 * We need to keep the mark around even if remaining mask cannot
814 * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
815 * changes to the mask.
816 * Destroy mark when only umask bits remain.
818 *destroy = !((fsn_mark->mask | fsn_mark->ignored_mask) & ~umask);
819 spin_unlock(&fsn_mark->lock);
821 return mask & oldmask;
824 static int fanotify_remove_mark(struct fsnotify_group *group,
825 fsnotify_connp_t *connp, __u32 mask,
826 unsigned int flags, __u32 umask)
828 struct fsnotify_mark *fsn_mark = NULL;
832 mutex_lock(&group->mark_mutex);
833 fsn_mark = fsnotify_find_mark(connp, group);
835 mutex_unlock(&group->mark_mutex);
839 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
840 umask, &destroy_mark);
841 if (removed & fsnotify_conn_mask(fsn_mark->connector))
842 fsnotify_recalc_mask(fsn_mark->connector);
844 fsnotify_detach_mark(fsn_mark);
845 mutex_unlock(&group->mark_mutex);
847 fsnotify_free_mark(fsn_mark);
849 /* matches the fsnotify_find_mark() */
850 fsnotify_put_mark(fsn_mark);
854 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
855 struct vfsmount *mnt, __u32 mask,
856 unsigned int flags, __u32 umask)
858 return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
862 static int fanotify_remove_sb_mark(struct fsnotify_group *group,
863 struct super_block *sb, __u32 mask,
864 unsigned int flags, __u32 umask)
866 return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
870 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
871 struct inode *inode, __u32 mask,
872 unsigned int flags, __u32 umask)
874 return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
878 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
884 spin_lock(&fsn_mark->lock);
885 if (!(flags & FAN_MARK_IGNORED_MASK)) {
886 oldmask = fsn_mark->mask;
887 fsn_mark->mask |= mask;
889 fsn_mark->ignored_mask |= mask;
890 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
891 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
893 spin_unlock(&fsn_mark->lock);
895 return mask & ~oldmask;
898 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
899 fsnotify_connp_t *connp,
901 __kernel_fsid_t *fsid)
903 struct ucounts *ucounts = group->fanotify_data.ucounts;
904 struct fsnotify_mark *mark;
908 * Enforce per user marks limits per user in all containing user ns.
909 * A group with FAN_UNLIMITED_MARKS does not contribute to mark count
910 * in the limited groups account.
912 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS) &&
913 !inc_ucount(ucounts->ns, ucounts->uid, UCOUNT_FANOTIFY_MARKS))
914 return ERR_PTR(-ENOSPC);
916 mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
919 goto out_dec_ucounts;
922 fsnotify_init_mark(mark, group);
923 ret = fsnotify_add_mark_locked(mark, connp, type, 0, fsid);
925 fsnotify_put_mark(mark);
926 goto out_dec_ucounts;
932 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
933 dec_ucount(ucounts, UCOUNT_FANOTIFY_MARKS);
938 static int fanotify_add_mark(struct fsnotify_group *group,
939 fsnotify_connp_t *connp, unsigned int type,
940 __u32 mask, unsigned int flags,
941 __kernel_fsid_t *fsid)
943 struct fsnotify_mark *fsn_mark;
946 mutex_lock(&group->mark_mutex);
947 fsn_mark = fsnotify_find_mark(connp, group);
949 fsn_mark = fanotify_add_new_mark(group, connp, type, fsid);
950 if (IS_ERR(fsn_mark)) {
951 mutex_unlock(&group->mark_mutex);
952 return PTR_ERR(fsn_mark);
955 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
956 if (added & ~fsnotify_conn_mask(fsn_mark->connector))
957 fsnotify_recalc_mask(fsn_mark->connector);
958 mutex_unlock(&group->mark_mutex);
960 fsnotify_put_mark(fsn_mark);
964 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
965 struct vfsmount *mnt, __u32 mask,
966 unsigned int flags, __kernel_fsid_t *fsid)
968 return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
969 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
972 static int fanotify_add_sb_mark(struct fsnotify_group *group,
973 struct super_block *sb, __u32 mask,
974 unsigned int flags, __kernel_fsid_t *fsid)
976 return fanotify_add_mark(group, &sb->s_fsnotify_marks,
977 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
980 static int fanotify_add_inode_mark(struct fsnotify_group *group,
981 struct inode *inode, __u32 mask,
982 unsigned int flags, __kernel_fsid_t *fsid)
984 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
987 * If some other task has this inode open for write we should not add
988 * an ignored mark, unless that ignored mark is supposed to survive
989 * modification changes anyway.
991 if ((flags & FAN_MARK_IGNORED_MASK) &&
992 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
993 inode_is_open_for_write(inode))
996 return fanotify_add_mark(group, &inode->i_fsnotify_marks,
997 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
1000 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
1002 struct fanotify_event *oevent;
1004 oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
1008 fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
1009 oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
1011 return &oevent->fse;
1014 static struct hlist_head *fanotify_alloc_merge_hash(void)
1016 struct hlist_head *hash;
1018 hash = kmalloc(sizeof(struct hlist_head) << FANOTIFY_HTABLE_BITS,
1019 GFP_KERNEL_ACCOUNT);
1023 __hash_init(hash, FANOTIFY_HTABLE_SIZE);
1028 /* fanotify syscalls */
1029 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
1031 struct fsnotify_group *group;
1033 unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
1034 unsigned int class = flags & FANOTIFY_CLASS_BITS;
1036 pr_debug("%s: flags=%x event_f_flags=%x\n",
1037 __func__, flags, event_f_flags);
1039 if (!capable(CAP_SYS_ADMIN))
1042 #ifdef CONFIG_AUDITSYSCALL
1043 if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT))
1045 if (flags & ~FANOTIFY_INIT_FLAGS)
1049 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
1052 switch (event_f_flags & O_ACCMODE) {
1061 if (fid_mode && class != FAN_CLASS_NOTIF)
1065 * Child name is reported with parent fid so requires dir fid.
1066 * We can report both child fid and dir fid with or without name.
1068 if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
1071 f_flags = O_RDWR | FMODE_NONOTIFY;
1072 if (flags & FAN_CLOEXEC)
1073 f_flags |= O_CLOEXEC;
1074 if (flags & FAN_NONBLOCK)
1075 f_flags |= O_NONBLOCK;
1077 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
1078 group = fsnotify_alloc_user_group(&fanotify_fsnotify_ops);
1079 if (IS_ERR(group)) {
1080 return PTR_ERR(group);
1083 /* Enforce groups limits per user in all containing user ns */
1084 group->fanotify_data.ucounts = inc_ucount(current_user_ns(),
1086 UCOUNT_FANOTIFY_GROUPS);
1087 if (!group->fanotify_data.ucounts) {
1089 goto out_destroy_group;
1092 group->fanotify_data.flags = flags;
1093 group->memcg = get_mem_cgroup_from_mm(current->mm);
1095 group->fanotify_data.merge_hash = fanotify_alloc_merge_hash();
1096 if (!group->fanotify_data.merge_hash) {
1098 goto out_destroy_group;
1101 group->overflow_event = fanotify_alloc_overflow_event();
1102 if (unlikely(!group->overflow_event)) {
1104 goto out_destroy_group;
1107 if (force_o_largefile())
1108 event_f_flags |= O_LARGEFILE;
1109 group->fanotify_data.f_flags = event_f_flags;
1110 init_waitqueue_head(&group->fanotify_data.access_waitq);
1111 INIT_LIST_HEAD(&group->fanotify_data.access_list);
1113 case FAN_CLASS_NOTIF:
1114 group->priority = FS_PRIO_0;
1116 case FAN_CLASS_CONTENT:
1117 group->priority = FS_PRIO_1;
1119 case FAN_CLASS_PRE_CONTENT:
1120 group->priority = FS_PRIO_2;
1124 goto out_destroy_group;
1127 if (flags & FAN_UNLIMITED_QUEUE) {
1129 if (!capable(CAP_SYS_ADMIN))
1130 goto out_destroy_group;
1131 group->max_events = UINT_MAX;
1133 group->max_events = fanotify_max_queued_events;
1136 if (flags & FAN_UNLIMITED_MARKS) {
1138 if (!capable(CAP_SYS_ADMIN))
1139 goto out_destroy_group;
1142 if (flags & FAN_ENABLE_AUDIT) {
1144 if (!capable(CAP_AUDIT_WRITE))
1145 goto out_destroy_group;
1148 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
1150 goto out_destroy_group;
1155 fsnotify_destroy_group(group);
1159 /* Check if filesystem can encode a unique fid */
1160 static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
1162 __kernel_fsid_t root_fsid;
1166 * Make sure path is not in filesystem with zero fsid (e.g. tmpfs).
1168 err = vfs_get_fsid(path->dentry, fsid);
1172 if (!fsid->val[0] && !fsid->val[1])
1176 * Make sure path is not inside a filesystem subvolume (e.g. btrfs)
1177 * which uses a different fsid than sb root.
1179 err = vfs_get_fsid(path->dentry->d_sb->s_root, &root_fsid);
1183 if (root_fsid.val[0] != fsid->val[0] ||
1184 root_fsid.val[1] != fsid->val[1])
1188 * We need to make sure that the file system supports at least
1189 * encoding a file handle so user can use name_to_handle_at() to
1190 * compare fid returned with event to the file handle of watched
1191 * objects. However, name_to_handle_at() requires that the
1192 * filesystem also supports decoding file handles.
1194 if (!path->dentry->d_sb->s_export_op ||
1195 !path->dentry->d_sb->s_export_op->fh_to_dentry)
1201 static int fanotify_events_supported(struct path *path, __u64 mask)
1204 * Some filesystems such as 'proc' acquire unusual locks when opening
1205 * files. For them fanotify permission events have high chances of
1206 * deadlocking the system - open done when reporting fanotify event
1207 * blocks on this "unusual" lock while another process holding the lock
1208 * waits for fanotify permission event to be answered. Just disallow
1209 * permission events for such filesystems.
1211 if (mask & FANOTIFY_PERM_EVENTS &&
1212 path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
1217 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
1218 int dfd, const char __user *pathname)
1220 struct inode *inode = NULL;
1221 struct vfsmount *mnt = NULL;
1222 struct fsnotify_group *group;
1225 __kernel_fsid_t __fsid, *fsid = NULL;
1226 u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS;
1227 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
1228 bool ignored = flags & FAN_MARK_IGNORED_MASK;
1229 unsigned int obj_type, fid_mode;
1233 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
1234 __func__, fanotify_fd, flags, dfd, pathname, mask);
1236 /* we only use the lower 32 bits as of right now. */
1237 if (mask & ((__u64)0xffffffff << 32))
1240 if (flags & ~FANOTIFY_MARK_FLAGS)
1243 switch (mark_type) {
1244 case FAN_MARK_INODE:
1245 obj_type = FSNOTIFY_OBJ_TYPE_INODE;
1247 case FAN_MARK_MOUNT:
1248 obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
1250 case FAN_MARK_FILESYSTEM:
1251 obj_type = FSNOTIFY_OBJ_TYPE_SB;
1257 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
1259 case FAN_MARK_REMOVE:
1263 case FAN_MARK_FLUSH:
1264 if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH))
1271 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
1272 valid_mask |= FANOTIFY_PERM_EVENTS;
1274 if (mask & ~valid_mask)
1277 /* Event flags (ONDIR, ON_CHILD) are meaningless in ignored mask */
1279 mask &= ~FANOTIFY_EVENT_FLAGS;
1281 f = fdget(fanotify_fd);
1282 if (unlikely(!f.file))
1285 /* verify that this is indeed an fanotify instance */
1287 if (unlikely(f.file->f_op != &fanotify_fops))
1289 group = f.file->private_data;
1292 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
1293 * allowed to set permissions events.
1296 if (mask & FANOTIFY_PERM_EVENTS &&
1297 group->priority == FS_PRIO_0)
1301 * Events with data type inode do not carry enough information to report
1302 * event->fd, so we do not allow setting a mask for inode events unless
1303 * group supports reporting fid.
1304 * inode events are not supported on a mount mark, because they do not
1305 * carry enough information (i.e. path) to be filtered by mount point.
1307 fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
1308 if (mask & FANOTIFY_INODE_EVENTS &&
1309 (!fid_mode || mark_type == FAN_MARK_MOUNT))
1312 if (flags & FAN_MARK_FLUSH) {
1314 if (mark_type == FAN_MARK_MOUNT)
1315 fsnotify_clear_vfsmount_marks_by_group(group);
1316 else if (mark_type == FAN_MARK_FILESYSTEM)
1317 fsnotify_clear_sb_marks_by_group(group);
1319 fsnotify_clear_inode_marks_by_group(group);
1323 ret = fanotify_find_path(dfd, pathname, &path, flags,
1324 (mask & ALL_FSNOTIFY_EVENTS), obj_type);
1328 if (flags & FAN_MARK_ADD) {
1329 ret = fanotify_events_supported(&path, mask);
1331 goto path_put_and_out;
1335 ret = fanotify_test_fid(&path, &__fsid);
1337 goto path_put_and_out;
1342 /* inode held in place by reference to path; group by fget on fd */
1343 if (mark_type == FAN_MARK_INODE)
1344 inode = path.dentry->d_inode;
1348 /* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
1349 if (mnt || !S_ISDIR(inode->i_mode)) {
1350 mask &= ~FAN_EVENT_ON_CHILD;
1351 umask = FAN_EVENT_ON_CHILD;
1353 * If group needs to report parent fid, register for getting
1354 * events with parent/name info for non-directory.
1356 if ((fid_mode & FAN_REPORT_DIR_FID) &&
1357 (flags & FAN_MARK_ADD) && !ignored)
1358 mask |= FAN_EVENT_ON_CHILD;
1361 /* create/update an inode mark */
1362 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
1364 if (mark_type == FAN_MARK_MOUNT)
1365 ret = fanotify_add_vfsmount_mark(group, mnt, mask,
1367 else if (mark_type == FAN_MARK_FILESYSTEM)
1368 ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
1371 ret = fanotify_add_inode_mark(group, inode, mask,
1374 case FAN_MARK_REMOVE:
1375 if (mark_type == FAN_MARK_MOUNT)
1376 ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
1378 else if (mark_type == FAN_MARK_FILESYSTEM)
1379 ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
1382 ret = fanotify_remove_inode_mark(group, inode, mask,
1396 #ifndef CONFIG_ARCH_SPLIT_ARG64
1397 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
1398 __u64, mask, int, dfd,
1399 const char __user *, pathname)
1401 return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
1405 #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT)
1406 SYSCALL32_DEFINE6(fanotify_mark,
1407 int, fanotify_fd, unsigned int, flags,
1408 SC_ARG64(mask), int, dfd,
1409 const char __user *, pathname)
1411 return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
1417 * fanotify_user_setup - Our initialization function. Note that we cannot return
1418 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
1419 * must result in panic().
1421 static int __init fanotify_user_setup(void)
1428 * Allow up to 1% of addressable memory to be accounted for per user
1429 * marks limited to the range [8192, 1048576]. mount and sb marks are
1430 * a lot cheaper than inode marks, but there is no reason for a user
1431 * to have many of those, so calculate by the cost of inode marks.
1433 max_marks = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
1435 max_marks = clamp(max_marks, FANOTIFY_OLD_DEFAULT_MAX_MARKS,
1436 FANOTIFY_DEFAULT_MAX_USER_MARKS);
1438 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 10);
1439 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9);
1441 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark,
1442 SLAB_PANIC|SLAB_ACCOUNT);
1443 fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
1445 fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
1447 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
1448 fanotify_perm_event_cachep =
1449 KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
1452 fanotify_max_queued_events = FANOTIFY_DEFAULT_MAX_EVENTS;
1453 init_user_ns.ucount_max[UCOUNT_FANOTIFY_GROUPS] =
1454 FANOTIFY_DEFAULT_MAX_GROUPS;
1455 init_user_ns.ucount_max[UCOUNT_FANOTIFY_MARKS] = max_marks;
1459 device_initcall(fanotify_user_setup);