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>
17
18#include "fanotify.h"
19
20static bool fanotify_path_equal(struct path *p1, struct path *p2)
21{
22	return p1->mnt == p2->mnt && p1->dentry == p2->dentry;
23}
24
25static inline bool fanotify_fsid_equal(__kernel_fsid_t *fsid1,
26				       __kernel_fsid_t *fsid2)
27{
28	return fsid1->val[0] == fsid2->val[0] && fsid1->val[1] == fsid2->val[1];
29}
30
31static bool fanotify_fh_equal(struct fanotify_fh *fh1,
32			      struct fanotify_fh *fh2)
33{
34	if (fh1->type != fh2->type || fh1->len != fh2->len)
35		return false;
36
37	return !fh1->len ||
38		!memcmp(fanotify_fh_buf(fh1), fanotify_fh_buf(fh2), fh1->len);
39}
40
41static bool fanotify_fid_event_equal(struct fanotify_fid_event *ffe1,
42				     struct fanotify_fid_event *ffe2)
43{
44	/* Do not merge fid events without object fh */
45	if (!ffe1->object_fh.len)
46		return false;
47
48	return fanotify_fsid_equal(&ffe1->fsid, &ffe2->fsid) &&
49		fanotify_fh_equal(&ffe1->object_fh, &ffe2->object_fh);
50}
51
52static bool fanotify_info_equal(struct fanotify_info *info1,
53				struct fanotify_info *info2)
54{
55	if (info1->dir_fh_totlen != info2->dir_fh_totlen ||
56	    info1->file_fh_totlen != info2->file_fh_totlen ||
57	    info1->name_len != info2->name_len)
58		return false;
59
60	if (info1->dir_fh_totlen &&
61	    !fanotify_fh_equal(fanotify_info_dir_fh(info1),
62			       fanotify_info_dir_fh(info2)))
63		return false;
64
65	if (info1->file_fh_totlen &&
66	    !fanotify_fh_equal(fanotify_info_file_fh(info1),
67			       fanotify_info_file_fh(info2)))
68		return false;
69
70	return !info1->name_len ||
71		!memcmp(fanotify_info_name(info1), fanotify_info_name(info2),
72			info1->name_len);
73}
74
75static bool fanotify_name_event_equal(struct fanotify_name_event *fne1,
76				      struct fanotify_name_event *fne2)
77{
78	struct fanotify_info *info1 = &fne1->info;
79	struct fanotify_info *info2 = &fne2->info;
80
81	/* Do not merge name events without dir fh */
82	if (!info1->dir_fh_totlen)
83		return false;
84
85	if (!fanotify_fsid_equal(&fne1->fsid, &fne2->fsid))
86		return false;
87
88	return fanotify_info_equal(info1, info2);
89}
90
91static bool fanotify_should_merge(struct fsnotify_event *old_fsn,
92				  struct fsnotify_event *new_fsn)
93{
94	struct fanotify_event *old, *new;
95
96	pr_debug("%s: old=%p new=%p\n", __func__, old_fsn, new_fsn);
97	old = FANOTIFY_E(old_fsn);
98	new = FANOTIFY_E(new_fsn);
99
100	if (old_fsn->objectid != new_fsn->objectid ||
101	    old->type != new->type || old->pid != new->pid)
102		return false;
103
104	/*
105	 * We want to merge many dirent events in the same dir (i.e.
106	 * creates/unlinks/renames), but we do not want to merge dirent
107	 * events referring to subdirs with dirent events referring to
108	 * non subdirs, otherwise, user won't be able to tell from a
109	 * mask FAN_CREATE|FAN_DELETE|FAN_ONDIR if it describes mkdir+
110	 * unlink pair or rmdir+create pair of events.
111	 */
112	if ((old->mask & FS_ISDIR) != (new->mask & FS_ISDIR))
113		return false;
114
115	switch (old->type) {
116	case FANOTIFY_EVENT_TYPE_PATH:
117		return fanotify_path_equal(fanotify_event_path(old),
118					   fanotify_event_path(new));
119	case FANOTIFY_EVENT_TYPE_FID:
120		return fanotify_fid_event_equal(FANOTIFY_FE(old),
121						FANOTIFY_FE(new));
122	case FANOTIFY_EVENT_TYPE_FID_NAME:
123		return fanotify_name_event_equal(FANOTIFY_NE(old),
124						 FANOTIFY_NE(new));
125	default:
126		WARN_ON_ONCE(1);
127	}
128
129	return false;
130}
131
132/* Limit event merges to limit CPU overhead per event */
133#define FANOTIFY_MAX_MERGE_EVENTS 128
134
135/* and the list better be locked by something too! */
136static int fanotify_merge(struct list_head *list, struct fsnotify_event *event)
137{
138	struct fsnotify_event *test_event;
139	struct fanotify_event *new;
140	int i = 0;
141
142	pr_debug("%s: list=%p event=%p\n", __func__, list, event);
143	new = FANOTIFY_E(event);
144
145	/*
146	 * Don't merge a permission event with any other event so that we know
147	 * the event structure we have created in fanotify_handle_event() is the
148	 * one we should check for permission response.
149	 */
150	if (fanotify_is_perm_event(new->mask))
151		return 0;
152
153	list_for_each_entry_reverse(test_event, list, list) {
154		if (++i > FANOTIFY_MAX_MERGE_EVENTS)
155			break;
156		if (fanotify_should_merge(test_event, event)) {
157			FANOTIFY_E(test_event)->mask |= new->mask;
158			return 1;
159		}
160	}
161
162	return 0;
163}
164
165/*
166 * Wait for response to permission event. The function also takes care of
167 * freeing the permission event (or offloads that in case the wait is canceled
168 * by a signal). The function returns 0 in case access got allowed by userspace,
169 * -EPERM in case userspace disallowed the access, and -ERESTARTSYS in case
170 * the wait got interrupted by a signal.
171 */
172static int fanotify_get_response(struct fsnotify_group *group,
173				 struct fanotify_perm_event *event,
174				 struct fsnotify_iter_info *iter_info)
175{
176	int ret;
177
178	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
179
180	ret = wait_event_killable(group->fanotify_data.access_waitq,
181				  event->state == FAN_EVENT_ANSWERED);
182	/* Signal pending? */
183	if (ret < 0) {
184		spin_lock(&group->notification_lock);
185		/* Event reported to userspace and no answer yet? */
186		if (event->state == FAN_EVENT_REPORTED) {
187			/* Event will get freed once userspace answers to it */
188			event->state = FAN_EVENT_CANCELED;
189			spin_unlock(&group->notification_lock);
190			return ret;
191		}
192		/* Event not yet reported? Just remove it. */
193		if (event->state == FAN_EVENT_INIT)
194			fsnotify_remove_queued_event(group, &event->fae.fse);
195		/*
196		 * Event may be also answered in case signal delivery raced
197		 * with wakeup. In that case we have nothing to do besides
198		 * freeing the event and reporting error.
199		 */
200		spin_unlock(&group->notification_lock);
201		goto out;
202	}
203
204	/* userspace responded, convert to something usable */
205	switch (event->response & ~FAN_AUDIT) {
206	case FAN_ALLOW:
207		ret = 0;
208		break;
209	case FAN_DENY:
210	default:
211		ret = -EPERM;
212	}
213
214	/* Check if the response should be audited */
215	if (event->response & FAN_AUDIT)
216		audit_fanotify(event->response & ~FAN_AUDIT);
217
218	pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
219		 group, event, ret);
220out:
221	fsnotify_destroy_event(group, &event->fae.fse);
222
223	return ret;
224}
225
226/*
227 * This function returns a mask for an event that only contains the flags
228 * that have been specifically requested by the user. Flags that may have
229 * been included within the event mask, but have not been explicitly
230 * requested by the user, will not be present in the returned mask.
231 */
232static u32 fanotify_group_event_mask(struct fsnotify_group *group,
233				     struct fsnotify_iter_info *iter_info,
234				     u32 event_mask, const void *data,
235				     int data_type, struct inode *dir)
236{
237	__u32 marks_mask = 0, marks_ignored_mask = 0;
238	__u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS |
239				     FANOTIFY_EVENT_FLAGS;
240	const struct path *path = fsnotify_data_path(data, data_type);
241	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
242	struct fsnotify_mark *mark;
243	int type;
244
245	pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
246		 __func__, iter_info->report_mask, event_mask, data, data_type);
247
248	if (!fid_mode) {
249		/* Do we have path to open a file descriptor? */
250		if (!path)
251			return 0;
252		/* Path type events are only relevant for files and dirs */
253		if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry))
254			return 0;
255	} else if (!(fid_mode & FAN_REPORT_FID)) {
256		/* Do we have a directory inode to report? */
257		if (!dir && !(event_mask & FS_ISDIR))
258			return 0;
259	}
260
261	fsnotify_foreach_obj_type(type) {
262		if (!fsnotify_iter_should_report_type(iter_info, type))
263			continue;
264		mark = iter_info->marks[type];
265
266		/* Apply ignore mask regardless of ISDIR and ON_CHILD flags */
267		marks_ignored_mask |= mark->ignored_mask;
268
269		/*
270		 * If the event is on dir and this mark doesn't care about
271		 * events on dir, don't send it!
272		 */
273		if (event_mask & FS_ISDIR && !(mark->mask & FS_ISDIR))
274			continue;
275
276		/*
277		 * If the event is on a child and this mark is on a parent not
278		 * watching children, don't send it!
279		 */
280		if (type == FSNOTIFY_OBJ_TYPE_PARENT &&
281		    !(mark->mask & FS_EVENT_ON_CHILD))
282			continue;
283
284		marks_mask |= mark->mask;
285	}
286
287	test_mask = event_mask & marks_mask & ~marks_ignored_mask;
288
289	/*
290	 * For dirent modification events (create/delete/move) that do not carry
291	 * the child entry name information, we report FAN_ONDIR for mkdir/rmdir
292	 * so user can differentiate them from creat/unlink.
293	 *
294	 * For backward compatibility and consistency, do not report FAN_ONDIR
295	 * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR
296	 * to user in fid mode for all event types.
297	 *
298	 * We never report FAN_EVENT_ON_CHILD to user, but we do pass it in to
299	 * fanotify_alloc_event() when group is reporting fid as indication
300	 * that event happened on child.
301	 */
302	if (fid_mode) {
303		/* Do not report event flags without any event */
304		if (!(test_mask & ~FANOTIFY_EVENT_FLAGS))
305			return 0;
306	} else {
307		user_mask &= ~FANOTIFY_EVENT_FLAGS;
308	}
309
310	return test_mask & user_mask;
311}
312
313/*
314 * Check size needed to encode fanotify_fh.
315 *
316 * Return size of encoded fh without fanotify_fh header.
317 * Return 0 on failure to encode.
318 */
319static int fanotify_encode_fh_len(struct inode *inode)
320{
321	int dwords = 0;
322
323	if (!inode)
324		return 0;
325
326	exportfs_encode_inode_fh(inode, NULL, &dwords, NULL);
327
328	return dwords << 2;
329}
330
331/*
332 * Encode fanotify_fh.
333 *
334 * Return total size of encoded fh including fanotify_fh header.
335 * Return 0 on failure to encode.
336 */
337static int fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode,
338			      unsigned int fh_len, gfp_t gfp)
339{
340	int dwords, type = 0;
341	char *ext_buf = NULL;
342	void *buf = fh->buf;
343	int err;
344
345	fh->type = FILEID_ROOT;
346	fh->len = 0;
347	fh->flags = 0;
348	if (!inode)
349		return 0;
350
351	/*
352	 * !gpf means preallocated variable size fh, but fh_len could
353	 * be zero in that case if encoding fh len failed.
354	 */
355	err = -ENOENT;
356	if (fh_len < 4 || WARN_ON_ONCE(fh_len % 4))
357		goto out_err;
358
359	/* No external buffer in a variable size allocated fh */
360	if (gfp && fh_len > FANOTIFY_INLINE_FH_LEN) {
361		/* Treat failure to allocate fh as failure to encode fh */
362		err = -ENOMEM;
363		ext_buf = kmalloc(fh_len, gfp);
364		if (!ext_buf)
365			goto out_err;
366
367		*fanotify_fh_ext_buf_ptr(fh) = ext_buf;
368		buf = ext_buf;
369		fh->flags |= FANOTIFY_FH_FLAG_EXT_BUF;
370	}
371
372	dwords = fh_len >> 2;
373	type = exportfs_encode_inode_fh(inode, buf, &dwords, NULL);
374	err = -EINVAL;
375	if (!type || type == FILEID_INVALID || fh_len != dwords << 2)
376		goto out_err;
377
378	fh->type = type;
379	fh->len = fh_len;
380
381	return FANOTIFY_FH_HDR_LEN + fh_len;
382
383out_err:
384	pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n",
385			    type, fh_len, err);
386	kfree(ext_buf);
387	*fanotify_fh_ext_buf_ptr(fh) = NULL;
388	/* Report the event without a file identifier on encode error */
389	fh->type = FILEID_INVALID;
390	fh->len = 0;
391	return 0;
392}
393
394/*
395 * The inode to use as identifier when reporting fid depends on the event.
396 * Report the modified directory inode on dirent modification events.
397 * Report the "victim" inode otherwise.
398 * For example:
399 * FS_ATTRIB reports the child inode even if reported on a watched parent.
400 * FS_CREATE reports the modified dir inode and not the created inode.
401 */
402static struct inode *fanotify_fid_inode(u32 event_mask, const void *data,
403					int data_type, struct inode *dir)
404{
405	if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
406		return dir;
407
408	return fsnotify_data_inode(data, data_type);
409}
410
411/*
412 * The inode to use as identifier when reporting dir fid depends on the event.
413 * Report the modified directory inode on dirent modification events.
414 * Report the "victim" inode if "victim" is a directory.
415 * Report the parent inode if "victim" is not a directory and event is
416 * reported to parent.
417 * Otherwise, do not report dir fid.
418 */
419static struct inode *fanotify_dfid_inode(u32 event_mask, const void *data,
420					 int data_type, struct inode *dir)
421{
422	struct inode *inode = fsnotify_data_inode(data, data_type);
423
424	if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
425		return dir;
426
427	if (S_ISDIR(inode->i_mode))
428		return inode;
429
430	return dir;
431}
432
433static struct fanotify_event *fanotify_alloc_path_event(const struct path *path,
434							gfp_t gfp)
435{
436	struct fanotify_path_event *pevent;
437
438	pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp);
439	if (!pevent)
440		return NULL;
441
442	pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH;
443	pevent->path = *path;
444	path_get(path);
445
446	return &pevent->fae;
447}
448
449static struct fanotify_event *fanotify_alloc_perm_event(const struct path *path,
450							gfp_t gfp)
451{
452	struct fanotify_perm_event *pevent;
453
454	pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
455	if (!pevent)
456		return NULL;
457
458	pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
459	pevent->response = 0;
460	pevent->state = FAN_EVENT_INIT;
461	pevent->path = *path;
462	path_get(path);
463
464	return &pevent->fae;
465}
466
467static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id,
468						       __kernel_fsid_t *fsid,
469						       gfp_t gfp)
470{
471	struct fanotify_fid_event *ffe;
472
473	ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp);
474	if (!ffe)
475		return NULL;
476
477	ffe->fae.type = FANOTIFY_EVENT_TYPE_FID;
478	ffe->fsid = *fsid;
479	fanotify_encode_fh(&ffe->object_fh, id, fanotify_encode_fh_len(id),
480			   gfp);
481
482	return &ffe->fae;
483}
484
485static struct fanotify_event *fanotify_alloc_name_event(struct inode *id,
486							__kernel_fsid_t *fsid,
487							const struct qstr *file_name,
488							struct inode *child,
489							gfp_t gfp)
490{
491	struct fanotify_name_event *fne;
492	struct fanotify_info *info;
493	struct fanotify_fh *dfh, *ffh;
494	unsigned int dir_fh_len = fanotify_encode_fh_len(id);
495	unsigned int child_fh_len = fanotify_encode_fh_len(child);
496	unsigned int size;
497
498	size = sizeof(*fne) + FANOTIFY_FH_HDR_LEN + dir_fh_len;
499	if (child_fh_len)
500		size += FANOTIFY_FH_HDR_LEN + child_fh_len;
501	if (file_name)
502		size += file_name->len + 1;
503	fne = kmalloc(size, gfp);
504	if (!fne)
505		return NULL;
506
507	fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME;
508	fne->fsid = *fsid;
509	info = &fne->info;
510	fanotify_info_init(info);
511	dfh = fanotify_info_dir_fh(info);
512	info->dir_fh_totlen = fanotify_encode_fh(dfh, id, dir_fh_len, 0);
513	if (child_fh_len) {
514		ffh = fanotify_info_file_fh(info);
515		info->file_fh_totlen = fanotify_encode_fh(ffh, child, child_fh_len, 0);
516	}
517	if (file_name)
518		fanotify_info_copy_name(info, file_name);
519
520	pr_debug("%s: ino=%lu size=%u dir_fh_len=%u child_fh_len=%u name_len=%u name='%.*s'\n",
521		 __func__, id->i_ino, size, dir_fh_len, child_fh_len,
522		 info->name_len, info->name_len, fanotify_info_name(info));
523
524	return &fne->fae;
525}
526
527static struct fanotify_event *fanotify_alloc_event(struct fsnotify_group *group,
528						   u32 mask, const void *data,
529						   int data_type, struct inode *dir,
530						   const struct qstr *file_name,
531						   __kernel_fsid_t *fsid)
532{
533	struct fanotify_event *event = NULL;
534	gfp_t gfp = GFP_KERNEL_ACCOUNT;
535	struct inode *id = fanotify_fid_inode(mask, data, data_type, dir);
536	struct inode *dirid = fanotify_dfid_inode(mask, data, data_type, dir);
537	const struct path *path = fsnotify_data_path(data, data_type);
538	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
539	struct mem_cgroup *old_memcg;
540	struct inode *child = NULL;
541	bool name_event = false;
542
543	if ((fid_mode & FAN_REPORT_DIR_FID) && dirid) {
544		/*
545		 * With both flags FAN_REPORT_DIR_FID and FAN_REPORT_FID, we
546		 * report the child fid for events reported on a non-dir child
547		 * in addition to reporting the parent fid and maybe child name.
548		 */
549		if ((fid_mode & FAN_REPORT_FID) &&
550		    id != dirid && !(mask & FAN_ONDIR))
551			child = id;
552
553		id = dirid;
554
555		/*
556		 * We record file name only in a group with FAN_REPORT_NAME
557		 * and when we have a directory inode to report.
558		 *
559		 * For directory entry modification event, we record the fid of
560		 * the directory and the name of the modified entry.
561		 *
562		 * For event on non-directory that is reported to parent, we
563		 * record the fid of the parent and the name of the child.
564		 *
565		 * Even if not reporting name, we need a variable length
566		 * fanotify_name_event if reporting both parent and child fids.
567		 */
568		if (!(fid_mode & FAN_REPORT_NAME)) {
569			name_event = !!child;
570			file_name = NULL;
571		} else if ((mask & ALL_FSNOTIFY_DIRENT_EVENTS) ||
572			   !(mask & FAN_ONDIR)) {
573			name_event = true;
574		}
575	}
576
577	/*
578	 * For queues with unlimited length lost events are not expected and
579	 * can possibly have security implications. Avoid losing events when
580	 * memory is short. For the limited size queues, avoid OOM killer in the
581	 * target monitoring memcg as it may have security repercussion.
582	 */
583	if (group->max_events == UINT_MAX)
584		gfp |= __GFP_NOFAIL;
585	else
586		gfp |= __GFP_RETRY_MAYFAIL;
587
588	/* Whoever is interested in the event, pays for the allocation. */
589	old_memcg = set_active_memcg(group->memcg);
590
591	if (fanotify_is_perm_event(mask)) {
592		event = fanotify_alloc_perm_event(path, gfp);
593	} else if (name_event && (file_name || child)) {
594		event = fanotify_alloc_name_event(id, fsid, file_name, child,
595						  gfp);
596	} else if (fid_mode) {
597		event = fanotify_alloc_fid_event(id, fsid, gfp);
598	} else {
599		event = fanotify_alloc_path_event(path, gfp);
600	}
601
602	if (!event)
603		goto out;
604
605	/*
606	 * Use the victim inode instead of the watching inode as the id for
607	 * event queue, so event reported on parent is merged with event
608	 * reported on child when both directory and child watches exist.
609	 */
610	fanotify_init_event(event, (unsigned long)id, mask);
611	if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
612		event->pid = get_pid(task_pid(current));
613	else
614		event->pid = get_pid(task_tgid(current));
615
616out:
617	set_active_memcg(old_memcg);
618	return event;
619}
620
621/*
622 * Get cached fsid of the filesystem containing the object from any connector.
623 * All connectors are supposed to have the same fsid, but we do not verify that
624 * here.
625 */
626static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
627{
628	int type;
629	__kernel_fsid_t fsid = {};
630
631	fsnotify_foreach_obj_type(type) {
632		struct fsnotify_mark_connector *conn;
633
634		if (!fsnotify_iter_should_report_type(iter_info, type))
635			continue;
636
637		conn = READ_ONCE(iter_info->marks[type]->connector);
638		/* Mark is just getting destroyed or created? */
639		if (!conn)
640			continue;
641		if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID))
642			continue;
643		/* Pairs with smp_wmb() in fsnotify_add_mark_list() */
644		smp_rmb();
645		fsid = conn->fsid;
646		if (WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
647			continue;
648		return fsid;
649	}
650
651	return fsid;
652}
653
654static int fanotify_handle_event(struct fsnotify_group *group, u32 mask,
655				 const void *data, int data_type,
656				 struct inode *dir,
657				 const struct qstr *file_name, u32 cookie,
658				 struct fsnotify_iter_info *iter_info)
659{
660	int ret = 0;
661	struct fanotify_event *event;
662	struct fsnotify_event *fsn_event;
663	__kernel_fsid_t fsid = {};
664
665	BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
666	BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
667	BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB);
668	BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
669	BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
670	BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
671	BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO);
672	BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM);
673	BUILD_BUG_ON(FAN_CREATE != FS_CREATE);
674	BUILD_BUG_ON(FAN_DELETE != FS_DELETE);
675	BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF);
676	BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF);
677	BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
678	BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
679	BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
680	BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
681	BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
682	BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
683	BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
684
685	BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 19);
686
687	mask = fanotify_group_event_mask(group, iter_info, mask, data,
688					 data_type, dir);
689	if (!mask)
690		return 0;
691
692	pr_debug("%s: group=%p mask=%x\n", __func__, group, mask);
693
694	if (fanotify_is_perm_event(mask)) {
695		/*
696		 * fsnotify_prepare_user_wait() fails if we race with mark
697		 * deletion.  Just let the operation pass in that case.
698		 */
699		if (!fsnotify_prepare_user_wait(iter_info))
700			return 0;
701	}
702
703	if (FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS)) {
704		fsid = fanotify_get_fsid(iter_info);
705		/* Racing with mark destruction or creation? */
706		if (!fsid.val[0] && !fsid.val[1])
707			return 0;
708	}
709
710	event = fanotify_alloc_event(group, mask, data, data_type, dir,
711				     file_name, &fsid);
712	ret = -ENOMEM;
713	if (unlikely(!event)) {
714		/*
715		 * We don't queue overflow events for permission events as
716		 * there the access is denied and so no event is in fact lost.
717		 */
718		if (!fanotify_is_perm_event(mask))
719			fsnotify_queue_overflow(group);
720		goto finish;
721	}
722
723	fsn_event = &event->fse;
724	ret = fsnotify_add_event(group, fsn_event, fanotify_merge);
725	if (ret) {
726		/* Permission events shouldn't be merged */
727		BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
728		/* Our event wasn't used in the end. Free it. */
729		fsnotify_destroy_event(group, fsn_event);
730
731		ret = 0;
732	} else if (fanotify_is_perm_event(mask)) {
733		ret = fanotify_get_response(group, FANOTIFY_PERM(event),
734					    iter_info);
735	}
736finish:
737	if (fanotify_is_perm_event(mask))
738		fsnotify_finish_user_wait(iter_info);
739
740	return ret;
741}
742
743static void fanotify_free_group_priv(struct fsnotify_group *group)
744{
745	struct user_struct *user;
746
747	user = group->fanotify_data.user;
748	atomic_dec(&user->fanotify_listeners);
749	free_uid(user);
750}
751
752static void fanotify_free_path_event(struct fanotify_event *event)
753{
754	path_put(fanotify_event_path(event));
755	kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event));
756}
757
758static void fanotify_free_perm_event(struct fanotify_event *event)
759{
760	path_put(fanotify_event_path(event));
761	kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
762}
763
764static void fanotify_free_fid_event(struct fanotify_event *event)
765{
766	struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
767
768	if (fanotify_fh_has_ext_buf(&ffe->object_fh))
769		kfree(fanotify_fh_ext_buf(&ffe->object_fh));
770	kmem_cache_free(fanotify_fid_event_cachep, ffe);
771}
772
773static void fanotify_free_name_event(struct fanotify_event *event)
774{
775	kfree(FANOTIFY_NE(event));
776}
777
778static void fanotify_free_event(struct fsnotify_event *fsn_event)
779{
780	struct fanotify_event *event;
781
782	event = FANOTIFY_E(fsn_event);
783	put_pid(event->pid);
784	switch (event->type) {
785	case FANOTIFY_EVENT_TYPE_PATH:
786		fanotify_free_path_event(event);
787		break;
788	case FANOTIFY_EVENT_TYPE_PATH_PERM:
789		fanotify_free_perm_event(event);
790		break;
791	case FANOTIFY_EVENT_TYPE_FID:
792		fanotify_free_fid_event(event);
793		break;
794	case FANOTIFY_EVENT_TYPE_FID_NAME:
795		fanotify_free_name_event(event);
796		break;
797	case FANOTIFY_EVENT_TYPE_OVERFLOW:
798		kfree(event);
799		break;
800	default:
801		WARN_ON_ONCE(1);
802	}
803}
804
805static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
806{
807	kmem_cache_free(fanotify_mark_cache, fsn_mark);
808}
809
810const struct fsnotify_ops fanotify_fsnotify_ops = {
811	.handle_event = fanotify_handle_event,
812	.free_group_priv = fanotify_free_group_priv,
813	.free_event = fanotify_free_event,
814	.free_mark = fanotify_free_mark,
815};
816