1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/fanotify.h>
3 #include <linux/fdtable.h>
4 #include <linux/fsnotify_backend.h>
5 #include <linux/init.h>
6 #include <linux/jiffies.h>
7 #include <linux/kernel.h> /* UINT_MAX */
8 #include <linux/mount.h>
9 #include <linux/sched.h>
10 #include <linux/sched/user.h>
11 #include <linux/sched/signal.h>
12 #include <linux/types.h>
13 #include <linux/wait.h>
14 #include <linux/audit.h>
15 #include <linux/sched/mm.h>
16 #include <linux/statfs.h>
20 static bool fanotify_path_equal(struct path *p1, struct path *p2)
22 return p1->mnt == p2->mnt && p1->dentry == p2->dentry;
25 static inline bool fanotify_fsid_equal(__kernel_fsid_t *fsid1,
26 __kernel_fsid_t *fsid2)
28 return fsid1->val[0] == fsid2->val[0] && fsid1->val[1] == fsid2->val[1];
31 static bool fanotify_fh_equal(struct fanotify_fh *fh1,
32 struct fanotify_fh *fh2)
34 if (fh1->type != fh2->type || fh1->len != fh2->len)
37 /* Do not merge events if we failed to encode fh */
38 if (fh1->type == FILEID_INVALID)
42 !memcmp(fanotify_fh_buf(fh1), fanotify_fh_buf(fh2), fh1->len);
45 static bool fanotify_fid_event_equal(struct fanotify_fid_event *ffe1,
46 struct fanotify_fid_event *ffe2)
48 /* Do not merge fid events without object fh */
49 if (!ffe1->object_fh.len)
52 return fanotify_fsid_equal(&ffe1->fsid, &ffe2->fsid) &&
53 fanotify_fh_equal(&ffe1->object_fh, &ffe2->object_fh);
56 static bool fanotify_name_event_equal(struct fanotify_name_event *fne1,
57 struct fanotify_name_event *fne2)
60 * Do not merge name events without dir fh.
61 * FAN_DIR_MODIFY does not encode object fh, so it may be empty.
63 if (!fne1->dir_fh.len)
66 if (fne1->name_len != fne2->name_len ||
67 !fanotify_fh_equal(&fne1->dir_fh, &fne2->dir_fh))
70 return !memcmp(fne1->name, fne2->name, fne1->name_len);
73 static bool fanotify_should_merge(struct fsnotify_event *old_fsn,
74 struct fsnotify_event *new_fsn)
76 struct fanotify_event *old, *new;
78 pr_debug("%s: old=%p new=%p\n", __func__, old_fsn, new_fsn);
79 old = FANOTIFY_E(old_fsn);
80 new = FANOTIFY_E(new_fsn);
82 if (old_fsn->objectid != new_fsn->objectid ||
83 old->type != new->type || old->pid != new->pid)
87 * We want to merge many dirent events in the same dir (i.e.
88 * creates/unlinks/renames), but we do not want to merge dirent
89 * events referring to subdirs with dirent events referring to
90 * non subdirs, otherwise, user won't be able to tell from a
91 * mask FAN_CREATE|FAN_DELETE|FAN_ONDIR if it describes mkdir+
92 * unlink pair or rmdir+create pair of events.
94 if ((old->mask & FS_ISDIR) != (new->mask & FS_ISDIR))
98 case FANOTIFY_EVENT_TYPE_PATH:
99 return fanotify_path_equal(fanotify_event_path(old),
100 fanotify_event_path(new));
101 case FANOTIFY_EVENT_TYPE_FID:
102 return fanotify_fid_event_equal(FANOTIFY_FE(old),
104 case FANOTIFY_EVENT_TYPE_FID_NAME:
105 return fanotify_name_event_equal(FANOTIFY_NE(old),
114 /* and the list better be locked by something too! */
115 static int fanotify_merge(struct list_head *list, struct fsnotify_event *event)
117 struct fsnotify_event *test_event;
118 struct fanotify_event *new;
120 pr_debug("%s: list=%p event=%p\n", __func__, list, event);
121 new = FANOTIFY_E(event);
124 * Don't merge a permission event with any other event so that we know
125 * the event structure we have created in fanotify_handle_event() is the
126 * one we should check for permission response.
128 if (fanotify_is_perm_event(new->mask))
131 list_for_each_entry_reverse(test_event, list, list) {
132 if (fanotify_should_merge(test_event, event)) {
133 FANOTIFY_E(test_event)->mask |= new->mask;
142 * Wait for response to permission event. The function also takes care of
143 * freeing the permission event (or offloads that in case the wait is canceled
144 * by a signal). The function returns 0 in case access got allowed by userspace,
145 * -EPERM in case userspace disallowed the access, and -ERESTARTSYS in case
146 * the wait got interrupted by a signal.
148 static int fanotify_get_response(struct fsnotify_group *group,
149 struct fanotify_perm_event *event,
150 struct fsnotify_iter_info *iter_info)
154 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
156 ret = wait_event_killable(group->fanotify_data.access_waitq,
157 event->state == FAN_EVENT_ANSWERED);
158 /* Signal pending? */
160 spin_lock(&group->notification_lock);
161 /* Event reported to userspace and no answer yet? */
162 if (event->state == FAN_EVENT_REPORTED) {
163 /* Event will get freed once userspace answers to it */
164 event->state = FAN_EVENT_CANCELED;
165 spin_unlock(&group->notification_lock);
168 /* Event not yet reported? Just remove it. */
169 if (event->state == FAN_EVENT_INIT)
170 fsnotify_remove_queued_event(group, &event->fae.fse);
172 * Event may be also answered in case signal delivery raced
173 * with wakeup. In that case we have nothing to do besides
174 * freeing the event and reporting error.
176 spin_unlock(&group->notification_lock);
180 /* userspace responded, convert to something usable */
181 switch (event->response & ~FAN_AUDIT) {
190 /* Check if the response should be audited */
191 if (event->response & FAN_AUDIT)
192 audit_fanotify(event->response & ~FAN_AUDIT);
194 pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
197 fsnotify_destroy_event(group, &event->fae.fse);
203 * This function returns a mask for an event that only contains the flags
204 * that have been specifically requested by the user. Flags that may have
205 * been included within the event mask, but have not been explicitly
206 * requested by the user, will not be present in the returned mask.
208 static u32 fanotify_group_event_mask(struct fsnotify_group *group,
209 struct fsnotify_iter_info *iter_info,
210 u32 event_mask, const void *data,
213 __u32 marks_mask = 0, marks_ignored_mask = 0;
214 __u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS |
215 FANOTIFY_EVENT_FLAGS;
216 const struct path *path = fsnotify_data_path(data, data_type);
217 struct fsnotify_mark *mark;
220 pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
221 __func__, iter_info->report_mask, event_mask, data, data_type);
223 if (!FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
224 /* Do we have path to open a file descriptor? */
227 /* Path type events are only relevant for files and dirs */
228 if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry))
232 fsnotify_foreach_obj_type(type) {
233 if (!fsnotify_iter_should_report_type(iter_info, type))
235 mark = iter_info->marks[type];
237 /* Apply ignore mask regardless of ISDIR and ON_CHILD flags */
238 marks_ignored_mask |= mark->ignored_mask;
241 * If the event is on dir and this mark doesn't care about
242 * events on dir, don't send it!
244 if (event_mask & FS_ISDIR && !(mark->mask & FS_ISDIR))
248 * If the event is for a child and this mark doesn't care about
249 * events on a child, don't send it!
251 if (event_mask & FS_EVENT_ON_CHILD &&
252 (type != FSNOTIFY_OBJ_TYPE_INODE ||
253 !(mark->mask & FS_EVENT_ON_CHILD)))
256 marks_mask |= mark->mask;
259 test_mask = event_mask & marks_mask & ~marks_ignored_mask;
262 * For dirent modification events (create/delete/move) that do not carry
263 * the child entry name information, we report FAN_ONDIR for mkdir/rmdir
264 * so user can differentiate them from creat/unlink.
266 * For backward compatibility and consistency, do not report FAN_ONDIR
267 * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR
268 * to user in fid mode for all event types.
270 * We never report FAN_EVENT_ON_CHILD to user, but we do pass it in to
271 * fanotify_alloc_event() when group is reporting fid as indication
272 * that event happened on child.
274 if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
275 /* Do not report event flags without any event */
276 if (!(test_mask & ~FANOTIFY_EVENT_FLAGS))
279 user_mask &= ~FANOTIFY_EVENT_FLAGS;
282 return test_mask & user_mask;
285 static void fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode,
288 int dwords, type, bytes = 0;
289 char *ext_buf = NULL;
298 type = exportfs_encode_inode_fh(inode, NULL, &dwords, NULL);
303 if (bytes > FANOTIFY_INLINE_FH_LEN) {
304 /* Treat failure to allocate fh as failure to allocate event */
306 ext_buf = kmalloc(bytes, gfp);
310 *fanotify_fh_ext_buf_ptr(fh) = ext_buf;
314 type = exportfs_encode_inode_fh(inode, buf, &dwords, NULL);
316 if (!type || type == FILEID_INVALID || bytes != dwords << 2)
325 pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n",
328 *fanotify_fh_ext_buf_ptr(fh) = NULL;
330 /* Report the event without a file identifier on encode error */
331 fh->type = FILEID_INVALID;
336 * The inode to use as identifier when reporting fid depends on the event.
337 * Report the modified directory inode on dirent modification events.
338 * Report the "victim" inode otherwise.
340 * FS_ATTRIB reports the child inode even if reported on a watched parent.
341 * FS_CREATE reports the modified dir inode and not the created inode.
343 static struct inode *fanotify_fid_inode(u32 event_mask, const void *data,
344 int data_type, struct inode *dir)
346 if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
349 return fsnotify_data_inode(data, data_type);
352 static struct fanotify_event *fanotify_alloc_path_event(const struct path *path,
355 struct fanotify_path_event *pevent;
357 pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp);
361 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH;
362 pevent->path = *path;
368 static struct fanotify_event *fanotify_alloc_perm_event(const struct path *path,
371 struct fanotify_perm_event *pevent;
373 pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
377 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
378 pevent->response = 0;
379 pevent->state = FAN_EVENT_INIT;
380 pevent->path = *path;
386 static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id,
387 __kernel_fsid_t *fsid,
390 struct fanotify_fid_event *ffe;
392 ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp);
396 ffe->fae.type = FANOTIFY_EVENT_TYPE_FID;
398 fanotify_encode_fh(&ffe->object_fh, id, gfp);
403 static struct fanotify_event *fanotify_alloc_name_event(struct inode *id,
404 __kernel_fsid_t *fsid,
405 const struct qstr *file_name,
408 struct fanotify_name_event *fne;
410 fne = kmalloc(sizeof(*fne) + file_name->len + 1, gfp);
414 fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME;
416 fanotify_encode_fh(&fne->dir_fh, id, gfp);
417 fne->name_len = file_name->len;
418 strcpy(fne->name, file_name->name);
423 static struct fanotify_event *fanotify_alloc_event(struct fsnotify_group *group,
424 u32 mask, const void *data,
425 int data_type, struct inode *dir,
426 const struct qstr *file_name,
427 __kernel_fsid_t *fsid)
429 struct fanotify_event *event = NULL;
430 gfp_t gfp = GFP_KERNEL_ACCOUNT;
431 struct inode *id = fanotify_fid_inode(mask, data, data_type, dir);
432 const struct path *path = fsnotify_data_path(data, data_type);
433 bool name_event = false;
436 * For queues with unlimited length lost events are not expected and
437 * can possibly have security implications. Avoid losing events when
438 * memory is short. For the limited size queues, avoid OOM killer in the
439 * target monitoring memcg as it may have security repercussion.
441 if (group->max_events == UINT_MAX)
444 gfp |= __GFP_RETRY_MAYFAIL;
446 /* Whoever is interested in the event, pays for the allocation. */
447 memalloc_use_memcg(group->memcg);
449 if (fanotify_is_perm_event(mask)) {
450 event = fanotify_alloc_perm_event(path, gfp);
451 } else if (name_event && file_name) {
452 event = fanotify_alloc_name_event(id, fsid, file_name, gfp);
453 } else if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
454 event = fanotify_alloc_fid_event(id, fsid, gfp);
456 event = fanotify_alloc_path_event(path, gfp);
463 * Use the victim inode instead of the watching inode as the id for
464 * event queue, so event reported on parent is merged with event
465 * reported on child when both directory and child watches exist.
467 fanotify_init_event(event, (unsigned long)id, mask);
468 if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
469 event->pid = get_pid(task_pid(current));
471 event->pid = get_pid(task_tgid(current));
474 memalloc_unuse_memcg();
479 * Get cached fsid of the filesystem containing the object from any connector.
480 * All connectors are supposed to have the same fsid, but we do not verify that
483 static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
486 __kernel_fsid_t fsid = {};
488 fsnotify_foreach_obj_type(type) {
489 struct fsnotify_mark_connector *conn;
491 if (!fsnotify_iter_should_report_type(iter_info, type))
494 conn = READ_ONCE(iter_info->marks[type]->connector);
495 /* Mark is just getting destroyed or created? */
498 if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID))
500 /* Pairs with smp_wmb() in fsnotify_add_mark_list() */
503 if (WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
511 static int fanotify_handle_event(struct fsnotify_group *group, u32 mask,
512 const void *data, int data_type,
514 const struct qstr *file_name, u32 cookie,
515 struct fsnotify_iter_info *iter_info)
518 struct fanotify_event *event;
519 struct fsnotify_event *fsn_event;
520 __kernel_fsid_t fsid = {};
522 BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
523 BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
524 BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB);
525 BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
526 BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
527 BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
528 BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO);
529 BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM);
530 BUILD_BUG_ON(FAN_CREATE != FS_CREATE);
531 BUILD_BUG_ON(FAN_DELETE != FS_DELETE);
532 BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF);
533 BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF);
534 BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
535 BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
536 BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
537 BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
538 BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
539 BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
540 BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
542 BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 19);
544 mask = fanotify_group_event_mask(group, iter_info, mask, data,
549 pr_debug("%s: group=%p mask=%x\n", __func__, group, mask);
551 if (fanotify_is_perm_event(mask)) {
553 * fsnotify_prepare_user_wait() fails if we race with mark
554 * deletion. Just let the operation pass in that case.
556 if (!fsnotify_prepare_user_wait(iter_info))
560 if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
561 fsid = fanotify_get_fsid(iter_info);
562 /* Racing with mark destruction or creation? */
563 if (!fsid.val[0] && !fsid.val[1])
567 event = fanotify_alloc_event(group, mask, data, data_type, dir,
570 if (unlikely(!event)) {
572 * We don't queue overflow events for permission events as
573 * there the access is denied and so no event is in fact lost.
575 if (!fanotify_is_perm_event(mask))
576 fsnotify_queue_overflow(group);
580 fsn_event = &event->fse;
581 ret = fsnotify_add_event(group, fsn_event, fanotify_merge);
583 /* Permission events shouldn't be merged */
584 BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
585 /* Our event wasn't used in the end. Free it. */
586 fsnotify_destroy_event(group, fsn_event);
589 } else if (fanotify_is_perm_event(mask)) {
590 ret = fanotify_get_response(group, FANOTIFY_PERM(event),
594 if (fanotify_is_perm_event(mask))
595 fsnotify_finish_user_wait(iter_info);
600 static void fanotify_free_group_priv(struct fsnotify_group *group)
602 struct user_struct *user;
604 user = group->fanotify_data.user;
605 atomic_dec(&user->fanotify_listeners);
609 static void fanotify_free_path_event(struct fanotify_event *event)
611 path_put(fanotify_event_path(event));
612 kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event));
615 static void fanotify_free_perm_event(struct fanotify_event *event)
617 path_put(fanotify_event_path(event));
618 kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
621 static void fanotify_free_fid_event(struct fanotify_event *event)
623 struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
625 if (fanotify_fh_has_ext_buf(&ffe->object_fh))
626 kfree(fanotify_fh_ext_buf(&ffe->object_fh));
627 kmem_cache_free(fanotify_fid_event_cachep, ffe);
630 static void fanotify_free_name_event(struct fanotify_event *event)
632 struct fanotify_name_event *fne = FANOTIFY_NE(event);
634 if (fanotify_fh_has_ext_buf(&fne->dir_fh))
635 kfree(fanotify_fh_ext_buf(&fne->dir_fh));
639 static void fanotify_free_event(struct fsnotify_event *fsn_event)
641 struct fanotify_event *event;
643 event = FANOTIFY_E(fsn_event);
645 switch (event->type) {
646 case FANOTIFY_EVENT_TYPE_PATH:
647 fanotify_free_path_event(event);
649 case FANOTIFY_EVENT_TYPE_PATH_PERM:
650 fanotify_free_perm_event(event);
652 case FANOTIFY_EVENT_TYPE_FID:
653 fanotify_free_fid_event(event);
655 case FANOTIFY_EVENT_TYPE_FID_NAME:
656 fanotify_free_name_event(event);
658 case FANOTIFY_EVENT_TYPE_OVERFLOW:
666 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
668 kmem_cache_free(fanotify_mark_cache, fsn_mark);
671 const struct fsnotify_ops fanotify_fsnotify_ops = {
672 .handle_event = fanotify_handle_event,
673 .free_group_priv = fanotify_free_group_priv,
674 .free_event = fanotify_free_event,
675 .free_mark = fanotify_free_mark,