~hc/RK356X_SDK_RELEASE.git

..	..	@@ -17,6 +17,8 @@
17	17	#include <linux/compat.h>
18	18	#include <linux/sched/signal.h>
19	19	#include <linux/memcontrol.h>
	20	+#include <linux/statfs.h>
	21	+#include <linux/exportfs.h>
20	22
21	23	#include <asm/ioctls.h>
22	24
..	..	@@ -44,42 +46,90 @@
44	46	extern const struct fsnotify_ops fanotify_fsnotify_ops;
45	47
46	48	struct kmem_cache *fanotify_mark_cache __read_mostly;
47		-struct kmem_cache *fanotify_event_cachep __read_mostly;
	49	+struct kmem_cache *fanotify_fid_event_cachep __read_mostly;
	50	+struct kmem_cache *fanotify_path_event_cachep __read_mostly;
48	51	struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
49	52
	53	+#define FANOTIFY_EVENT_ALIGN 4
	54	+#define FANOTIFY_INFO_HDR_LEN \
	55	+ (sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle))
	56	+
	57	+static int fanotify_fid_info_len(int fh_len, int name_len)
	58	+{
	59	+ int info_len = fh_len;
	60	+
	61	+ if (name_len)
	62	+ info_len += name_len + 1;
	63	+
	64	+ return roundup(FANOTIFY_INFO_HDR_LEN + info_len, FANOTIFY_EVENT_ALIGN);
	65	+}
	66	+
	67	+static int fanotify_event_info_len(unsigned int fid_mode,
	68	+ struct fanotify_event *event)
	69	+{
	70	+ struct fanotify_info *info = fanotify_event_info(event);
	71	+ int dir_fh_len = fanotify_event_dir_fh_len(event);
	72	+ int fh_len = fanotify_event_object_fh_len(event);
	73	+ int info_len = 0;
	74	+ int dot_len = 0;
	75	+
	76	+ if (dir_fh_len) {
	77	+ info_len += fanotify_fid_info_len(dir_fh_len, info->name_len);
	78	+ } else if ((fid_mode & FAN_REPORT_NAME) && (event->mask & FAN_ONDIR)) {
	79	+ /*
	80	+ * With group flag FAN_REPORT_NAME, if name was not recorded in
	81	+ * event on a directory, we will report the name ".".
	82	+ */
	83	+ dot_len = 1;
	84	+ }
	85	+
	86	+ if (fh_len)
	87	+ info_len += fanotify_fid_info_len(fh_len, dot_len);
	88	+
	89	+ return info_len;
	90	+}
	91	+
50	92	/*
51		- * Get an fsnotify notification event if one exists and is small
	93	+ * Get an fanotify notification event if one exists and is small
52	94	* enough to fit in "count". Return an error pointer if the count
53		- * is not large enough.
54		- *
55		- * Called with the group->notification_lock held.
	95	+ * is not large enough. When permission event is dequeued, its state is
	96	+ * updated accordingly.
56	97	*/
57		-static struct fsnotify_event get_one_event(struct fsnotify_group group,
	98	+static struct fanotify_event get_one_event(struct fsnotify_group group,
58	99	size_t count)
59	100	{
60		- assert_spin_locked(&group->notification_lock);
	101	+ size_t event_size = FAN_EVENT_METADATA_LEN;
	102	+ struct fanotify_event *event = NULL;
	103	+ unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
61	104
62	105	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
63	106
	107	+ spin_lock(&group->notification_lock);
64	108	if (fsnotify_notify_queue_is_empty(group))
65		- return NULL;
	109	+ goto out;
66	110
67		- if (FAN_EVENT_METADATA_LEN > count)
68		- return ERR_PTR(-EINVAL);
	111	+ if (fid_mode) {
	112	+ event_size += fanotify_event_info_len(fid_mode,
	113	+ FANOTIFY_E(fsnotify_peek_first_event(group)));
	114	+ }
69	115
70		- /* held the notification_lock the whole time, so this is the
71		- * same event we peeked above */
72		- return fsnotify_remove_first_event(group);
	116	+ if (event_size > count) {
	117	+ event = ERR_PTR(-EINVAL);
	118	+ goto out;
	119	+ }
	120	+ event = FANOTIFY_E(fsnotify_remove_first_event(group));
	121	+ if (fanotify_is_perm_event(event->mask))
	122	+ FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED;
	123	+out:
	124	+ spin_unlock(&group->notification_lock);
	125	+ return event;
73	126	}
74	127
75		-static int create_fd(struct fsnotify_group *group,
76		- struct fanotify_event_info *event,
	128	+static int create_fd(struct fsnotify_group group, struct path path,
77	129	struct file **file)
78	130	{
79	131	int client_fd;
80	132	struct file *new_file;
81		-
82		- pr_debug("%s: group=%p event=%p\n", __func__, group, event);
83	133
84	134	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
85	135	if (client_fd < 0)
..	..	@@ -89,14 +139,9 @@
89	139	* we need a new file handle for the userspace program so it can read even if it was
90	140	* originally opened O_WRONLY.
91	141	*/
92		- /* it's possible this event was an overflow event. in that case dentry and mnt
93		- * are NULL; That's fine, just don't call dentry open */
94		- if (event->path.dentry && event->path.mnt)
95		- new_file = dentry_open(&event->path,
96		- group->fanotify_data.f_flags \| FMODE_NONOTIFY,
97		- current_cred());
98		- else
99		- new_file = ERR_PTR(-EOVERFLOW);
	142	+ new_file = dentry_open(path,
	143	+ group->fanotify_data.f_flags \| FMODE_NONOTIFY,
	144	+ current_cred());
100	145	if (IS_ERR(new_file)) {
101	146	/*
102	147	* we still send an event even if we can't open the file. this
..	..	@@ -114,62 +159,32 @@
114	159	return client_fd;
115	160	}
116	161
117		-static int fill_event_metadata(struct fsnotify_group *group,
118		- struct fanotify_event_metadata *metadata,
119		- struct fsnotify_event *fsn_event,
120		- struct file **file)
	162	+/*
	163	+ * Finish processing of permission event by setting it to ANSWERED state and
	164	+ * drop group->notification_lock.
	165	+ */
	166	+static void finish_permission_event(struct fsnotify_group *group,
	167	+ struct fanotify_perm_event *event,
	168	+ unsigned int response)
	169	+ __releases(&group->notification_lock)
121	170	{
122		- int ret = 0;
123		- struct fanotify_event_info *event;
	171	+ bool destroy = false;
124	172
125		- pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
126		- group, metadata, fsn_event);
127		-
128		- *file = NULL;
129		- event = container_of(fsn_event, struct fanotify_event_info, fse);
130		- metadata->event_len = FAN_EVENT_METADATA_LEN;
131		- metadata->metadata_len = FAN_EVENT_METADATA_LEN;
132		- metadata->vers = FANOTIFY_METADATA_VERSION;
133		- metadata->reserved = 0;
134		- metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
135		- metadata->pid = pid_vnr(event->tgid);
136		- if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
137		- metadata->fd = FAN_NOFD;
138		- else {
139		- metadata->fd = create_fd(group, event, file);
140		- if (metadata->fd < 0)
141		- ret = metadata->fd;
142		- }
143		-
144		- return ret;
145		-}
146		-
147		-static struct fanotify_perm_event_info *dequeue_event(
148		- struct fsnotify_group *group, int fd)
149		-{
150		- struct fanotify_perm_event_info event, return_e = NULL;
151		-
152		- spin_lock(&group->notification_lock);
153		- list_for_each_entry(event, &group->fanotify_data.access_list,
154		- fae.fse.list) {
155		- if (event->fd != fd)
156		- continue;
157		-
158		- list_del_init(&event->fae.fse.list);
159		- return_e = event;
160		- break;
161		- }
	173	+ assert_spin_locked(&group->notification_lock);
	174	+ event->response = response;
	175	+ if (event->state == FAN_EVENT_CANCELED)
	176	+ destroy = true;
	177	+ else
	178	+ event->state = FAN_EVENT_ANSWERED;
162	179	spin_unlock(&group->notification_lock);
163		-
164		- pr_debug("%s: found return_re=%p\n", __func__, return_e);
165		-
166		- return return_e;
	180	+ if (destroy)
	181	+ fsnotify_destroy_event(group, &event->fae.fse);
167	182	}
168	183
169	184	static int process_access_response(struct fsnotify_group *group,
170	185	struct fanotify_response *response_struct)
171	186	{
172		- struct fanotify_perm_event_info *event;
	187	+ struct fanotify_perm_event *event;
173	188	int fd = response_struct->fd;
174	189	int response = response_struct->response;
175	190
..	..	@@ -191,45 +206,233 @@
191	206	if (fd < 0)
192	207	return -EINVAL;
193	208
194		- if ((response & FAN_AUDIT) && !group->fanotify_data.audit)
	209	+ if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
195	210	return -EINVAL;
196	211
197		- event = dequeue_event(group, fd);
198		- if (!event)
199		- return -ENOENT;
	212	+ spin_lock(&group->notification_lock);
	213	+ list_for_each_entry(event, &group->fanotify_data.access_list,
	214	+ fae.fse.list) {
	215	+ if (event->fd != fd)
	216	+ continue;
200	217
201		- event->response = response;
202		- wake_up(&group->fanotify_data.access_waitq);
	218	+ list_del_init(&event->fae.fse.list);
	219	+ finish_permission_event(group, event, response);
	220	+ wake_up(&group->fanotify_data.access_waitq);
	221	+ return 0;
	222	+ }
	223	+ spin_unlock(&group->notification_lock);
203	224
204		- return 0;
	225	+ return -ENOENT;
	226	+}
	227	+
	228	+static int copy_info_to_user(__kernel_fsid_t fsid, struct fanotify_fh fh,
	229	+ int info_type, const char *name, size_t name_len,
	230	+ char __user *buf, size_t count)
	231	+{
	232	+ struct fanotify_event_info_fid info = { };
	233	+ struct file_handle handle = { };
	234	+ unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf;
	235	+ size_t fh_len = fh ? fh->len : 0;
	236	+ size_t info_len = fanotify_fid_info_len(fh_len, name_len);
	237	+ size_t len = info_len;
	238	+
	239	+ pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n",
	240	+ __func__, fh_len, name_len, info_len, count);
	241	+
	242	+ if (!fh_len)
	243	+ return 0;
	244	+
	245	+ if (WARN_ON_ONCE(len < sizeof(info) \|\| len > count))
	246	+ return -EFAULT;
	247	+
	248	+ /*
	249	+ * Copy event info fid header followed by variable sized file handle
	250	+ * and optionally followed by variable sized filename.
	251	+ */
	252	+ switch (info_type) {
	253	+ case FAN_EVENT_INFO_TYPE_FID:
	254	+ case FAN_EVENT_INFO_TYPE_DFID:
	255	+ if (WARN_ON_ONCE(name_len))
	256	+ return -EFAULT;
	257	+ break;
	258	+ case FAN_EVENT_INFO_TYPE_DFID_NAME:
	259	+ if (WARN_ON_ONCE(!name \|\| !name_len))
	260	+ return -EFAULT;
	261	+ break;
	262	+ default:
	263	+ return -EFAULT;
	264	+ }
	265	+
	266	+ info.hdr.info_type = info_type;
	267	+ info.hdr.len = len;
	268	+ info.fsid = *fsid;
	269	+ if (copy_to_user(buf, &info, sizeof(info)))
	270	+ return -EFAULT;
	271	+
	272	+ buf += sizeof(info);
	273	+ len -= sizeof(info);
	274	+ if (WARN_ON_ONCE(len < sizeof(handle)))
	275	+ return -EFAULT;
	276	+
	277	+ handle.handle_type = fh->type;
	278	+ handle.handle_bytes = fh_len;
	279	+ if (copy_to_user(buf, &handle, sizeof(handle)))
	280	+ return -EFAULT;
	281	+
	282	+ buf += sizeof(handle);
	283	+ len -= sizeof(handle);
	284	+ if (WARN_ON_ONCE(len < fh_len))
	285	+ return -EFAULT;
	286	+
	287	+ /*
	288	+ * For an inline fh and inline file name, copy through stack to exclude
	289	+ * the copy from usercopy hardening protections.
	290	+ */
	291	+ fh_buf = fanotify_fh_buf(fh);
	292	+ if (fh_len <= FANOTIFY_INLINE_FH_LEN) {
	293	+ memcpy(bounce, fh_buf, fh_len);
	294	+ fh_buf = bounce;
	295	+ }
	296	+ if (copy_to_user(buf, fh_buf, fh_len))
	297	+ return -EFAULT;
	298	+
	299	+ buf += fh_len;
	300	+ len -= fh_len;
	301	+
	302	+ if (name_len) {
	303	+ /* Copy the filename with terminating null */
	304	+ name_len++;
	305	+ if (WARN_ON_ONCE(len < name_len))
	306	+ return -EFAULT;
	307	+
	308	+ if (copy_to_user(buf, name, name_len))
	309	+ return -EFAULT;
	310	+
	311	+ buf += name_len;
	312	+ len -= name_len;
	313	+ }
	314	+
	315	+ /* Pad with 0's */
	316	+ WARN_ON_ONCE(len < 0 \|\| len >= FANOTIFY_EVENT_ALIGN);
	317	+ if (len > 0 && clear_user(buf, len))
	318	+ return -EFAULT;
	319	+
	320	+ return info_len;
205	321	}
206	322
207	323	static ssize_t copy_event_to_user(struct fsnotify_group *group,
208		- struct fsnotify_event *event,
209		- char __user *buf)
	324	+ struct fanotify_event *event,
	325	+ char __user *buf, size_t count)
210	326	{
211		- struct fanotify_event_metadata fanotify_event_metadata;
212		- struct file *f;
213		- int fd, ret;
	327	+ struct fanotify_event_metadata metadata;
	328	+ struct path *path = fanotify_event_path(event);
	329	+ struct fanotify_info *info = fanotify_event_info(event);
	330	+ unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
	331	+ struct file *f = NULL;
	332	+ int ret, fd = FAN_NOFD;
	333	+ int info_type = 0;
214	334
215	335	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
216	336
217		- ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
218		- if (ret < 0)
219		- return ret;
	337	+ metadata.event_len = FAN_EVENT_METADATA_LEN +
	338	+ fanotify_event_info_len(fid_mode, event);
	339	+ metadata.metadata_len = FAN_EVENT_METADATA_LEN;
	340	+ metadata.vers = FANOTIFY_METADATA_VERSION;
	341	+ metadata.reserved = 0;
	342	+ metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS;
	343	+ metadata.pid = pid_vnr(event->pid);
220	344
221		- fd = fanotify_event_metadata.fd;
	345	+ if (path && path->mnt && path->dentry) {
	346	+ fd = create_fd(group, path, &f);
	347	+ if (fd < 0)
	348	+ return fd;
	349	+ }
	350	+ metadata.fd = fd;
	351	+
222	352	ret = -EFAULT;
223		- if (copy_to_user(buf, &fanotify_event_metadata,
224		- fanotify_event_metadata.event_len))
	353	+ /*
	354	+ * Sanity check copy size in case get_one_event() and
	355	+ * event_len sizes ever get out of sync.
	356	+ */
	357	+ if (WARN_ON_ONCE(metadata.event_len > count))
225	358	goto out_close_fd;
226	359
227		- if (fanotify_is_perm_event(event->mask))
228		- FANOTIFY_PE(event)->fd = fd;
	360	+ if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN))
	361	+ goto out_close_fd;
229	362
230		- if (fd != FAN_NOFD)
	363	+ buf += FAN_EVENT_METADATA_LEN;
	364	+ count -= FAN_EVENT_METADATA_LEN;
	365	+
	366	+ if (fanotify_is_perm_event(event->mask))
	367	+ FANOTIFY_PERM(event)->fd = fd;
	368	+
	369	+ /* Event info records order is: dir fid + name, child fid */
	370	+ if (fanotify_event_dir_fh_len(event)) {
	371	+ info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME :
	372	+ FAN_EVENT_INFO_TYPE_DFID;
	373	+ ret = copy_info_to_user(fanotify_event_fsid(event),
	374	+ fanotify_info_dir_fh(info),
	375	+ info_type, fanotify_info_name(info),
	376	+ info->name_len, buf, count);
	377	+ if (ret < 0)
	378	+ goto out_close_fd;
	379	+
	380	+ buf += ret;
	381	+ count -= ret;
	382	+ }
	383	+
	384	+ if (fanotify_event_object_fh_len(event)) {
	385	+ const char *dot = NULL;
	386	+ int dot_len = 0;
	387	+
	388	+ if (fid_mode == FAN_REPORT_FID \|\| info_type) {
	389	+ /*
	390	+ * With only group flag FAN_REPORT_FID only type FID is
	391	+ * reported. Second info record type is always FID.
	392	+ */
	393	+ info_type = FAN_EVENT_INFO_TYPE_FID;
	394	+ } else if ((fid_mode & FAN_REPORT_NAME) &&
	395	+ (event->mask & FAN_ONDIR)) {
	396	+ /*
	397	+ * With group flag FAN_REPORT_NAME, if name was not
	398	+ * recorded in an event on a directory, report the
	399	+ * name "." with info type DFID_NAME.
	400	+ */
	401	+ info_type = FAN_EVENT_INFO_TYPE_DFID_NAME;
	402	+ dot = ".";
	403	+ dot_len = 1;
	404	+ } else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) \|\|
	405	+ (event->mask & FAN_ONDIR)) {
	406	+ /*
	407	+ * With group flag FAN_REPORT_DIR_FID, a single info
	408	+ * record has type DFID for directory entry modification
	409	+ * event and for event on a directory.
	410	+ */
	411	+ info_type = FAN_EVENT_INFO_TYPE_DFID;
	412	+ } else {
	413	+ /*
	414	+ * With group flags FAN_REPORT_DIR_FID\|FAN_REPORT_FID,
	415	+ * a single info record has type FID for event on a
	416	+ * non-directory, when there is no directory to report.
	417	+ * For example, on FAN_DELETE_SELF event.
	418	+ */
	419	+ info_type = FAN_EVENT_INFO_TYPE_FID;
	420	+ }
	421	+
	422	+ ret = copy_info_to_user(fanotify_event_fsid(event),
	423	+ fanotify_event_object_fh(event),
	424	+ info_type, dot, dot_len, buf, count);
	425	+ if (ret < 0)
	426	+ goto out_close_fd;
	427	+
	428	+ buf += ret;
	429	+ count -= ret;
	430	+ }
	431	+
	432	+ if (f)
231	433	fd_install(fd, f);
232		- return fanotify_event_metadata.event_len;
	434	+
	435	+ return metadata.event_len;
233	436
234	437	out_close_fd:
235	438	if (fd != FAN_NOFD) {
..	..	@@ -258,7 +461,7 @@
258	461	size_t count, loff_t *pos)
259	462	{
260	463	struct fsnotify_group *group;
261		- struct fsnotify_event *kevent;
	464	+ struct fanotify_event *event;
262	465	char __user *start;
263	466	int ret;
264	467	DEFINE_WAIT_FUNC(wait, woken_wake_function);
..	..	@@ -270,16 +473,18 @@
270	473
271	474	add_wait_queue(&group->notification_waitq, &wait);
272	475	while (1) {
273		- spin_lock(&group->notification_lock);
274		- kevent = get_one_event(group, count);
275		- spin_unlock(&group->notification_lock);
276		-
277		- if (IS_ERR(kevent)) {
278		- ret = PTR_ERR(kevent);
	476	+ /*
	477	+ * User can supply arbitrarily large buffer. Avoid softlockups
	478	+ * in case there are lots of available events.
	479	+ */
	480	+ cond_resched();
	481	+ event = get_one_event(group, count);
	482	+ if (IS_ERR(event)) {
	483	+ ret = PTR_ERR(event);
279	484	break;
280	485	}
281	486
282		- if (!kevent) {
	487	+ if (!event) {
283	488	ret = -EAGAIN;
284	489	if (file->f_flags & O_NONBLOCK)
285	490	break;
..	..	@@ -295,7 +500,7 @@
295	500	continue;
296	501	}
297	502
298		- ret = copy_event_to_user(group, kevent, buf);
	503	+ ret = copy_event_to_user(group, event, buf, count);
299	504	if (unlikely(ret == -EOPENSTALE)) {
300	505	/*
301	506	* We cannot report events with stale fd so drop it.
..	..	@@ -310,15 +515,17 @@
310	515	* Permission events get queued to wait for response. Other
311	516	* events can be destroyed now.
312	517	*/
313		- if (!fanotify_is_perm_event(kevent->mask)) {
314		- fsnotify_destroy_event(group, kevent);
	518	+ if (!fanotify_is_perm_event(event->mask)) {
	519	+ fsnotify_destroy_event(group, &event->fse);
315	520	} else {
316	521	if (ret <= 0) {
317		- FANOTIFY_PE(kevent)->response = FAN_DENY;
	522	+ spin_lock(&group->notification_lock);
	523	+ finish_permission_event(group,
	524	+ FANOTIFY_PERM(event), FAN_DENY);
318	525	wake_up(&group->fanotify_data.access_waitq);
319	526	} else {
320	527	spin_lock(&group->notification_lock);
321		- list_add_tail(&kevent->list,
	528	+ list_add_tail(&event->fse.list,
322	529	&group->fanotify_data.access_list);
323	530	spin_unlock(&group->notification_lock);
324	531	}
..	..	@@ -346,8 +553,10 @@
346	553
347	554	group = file->private_data;
348	555
349		- if (count > sizeof(response))
350		- count = sizeof(response);
	556	+ if (count < sizeof(response))
	557	+ return -EINVAL;
	558	+
	559	+ count = sizeof(response);
351	560
352	561	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
353	562
..	..	@@ -364,8 +573,6 @@
364	573	static int fanotify_release(struct inode ignored, struct file file)
365	574	{
366	575	struct fsnotify_group *group = file->private_data;
367		- struct fanotify_perm_event_info event, next;
368		- struct fsnotify_event *fsn_event;
369	576
370	577	/*
371	578	* Stop new events from arriving in the notification queue. since
..	..	@@ -379,13 +586,14 @@
379	586	* and simulate reply from userspace.
380	587	*/
381	588	spin_lock(&group->notification_lock);
382		- list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
383		- fae.fse.list) {
384		- pr_debug("%s: found group=%p event=%p\n", __func__, group,
385		- event);
	589	+ while (!list_empty(&group->fanotify_data.access_list)) {
	590	+ struct fanotify_perm_event *event;
386	591
	592	+ event = list_first_entry(&group->fanotify_data.access_list,
	593	+ struct fanotify_perm_event, fae.fse.list);
387	594	list_del_init(&event->fae.fse.list);
388		- event->response = FAN_ALLOW;
	595	+ finish_permission_event(group, event, FAN_ALLOW);
	596	+ spin_lock(&group->notification_lock);
389	597	}
390	598
391	599	/*
..	..	@@ -394,14 +602,17 @@
394	602	* response is consumed and fanotify_get_response() returns.
395	603	*/
396	604	while (!fsnotify_notify_queue_is_empty(group)) {
397		- fsn_event = fsnotify_remove_first_event(group);
398		- if (!(fsn_event->mask & FAN_ALL_PERM_EVENTS)) {
	605	+ struct fanotify_event *event;
	606	+
	607	+ event = FANOTIFY_E(fsnotify_remove_first_event(group));
	608	+ if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
399	609	spin_unlock(&group->notification_lock);
400		- fsnotify_destroy_event(group, fsn_event);
401		- spin_lock(&group->notification_lock);
	610	+ fsnotify_destroy_event(group, &event->fse);
402	611	} else {
403		- FANOTIFY_PE(fsn_event)->response = FAN_ALLOW;
	612	+ finish_permission_event(group, FANOTIFY_PERM(event),
	613	+ FAN_ALLOW);
404	614	}
	615	+ spin_lock(&group->notification_lock);
405	616	}
406	617	spin_unlock(&group->notification_lock);
407	618
..	..	@@ -447,12 +658,13 @@
447	658	.fasync = NULL,
448	659	.release = fanotify_release,
449	660	.unlocked_ioctl = fanotify_ioctl,
450		- .compat_ioctl = fanotify_ioctl,
	661	+ .compat_ioctl = compat_ptr_ioctl,
451	662	.llseek = noop_llseek,
452	663	};
453	664
454	665	static int fanotify_find_path(int dfd, const char __user *filename,
455		- struct path *path, unsigned int flags)
	666	+ struct path *path, unsigned int flags, __u64 mask,
	667	+ unsigned int obj_type)
456	668	{
457	669	int ret;
458	670
..	..	@@ -490,36 +702,42 @@
490	702	}
491	703
492	704	/* you can only watch an inode if you have read permissions on it */
493		- ret = inode_permission2(path->mnt, path->dentry->d_inode, MAY_READ);
	705	+ ret = inode_permission(path->dentry->d_inode, MAY_READ);
	706	+ if (ret) {
	707	+ path_put(path);
	708	+ goto out;
	709	+ }
	710	+
	711	+ ret = security_path_notify(path, mask, obj_type);
494	712	if (ret)
495	713	path_put(path);
	714	+
496	715	out:
497	716	return ret;
498	717	}
499	718
500	719	static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
501		- __u32 mask,
502		- unsigned int flags,
503		- int *destroy)
	720	+ __u32 mask, unsigned int flags,
	721	+ __u32 umask, int *destroy)
504	722	{
505	723	__u32 oldmask = 0;
506	724
	725	+ /* umask bits cannot be removed by user */
	726	+ mask &= ~umask;
507	727	spin_lock(&fsn_mark->lock);
508	728	if (!(flags & FAN_MARK_IGNORED_MASK)) {
509		- __u32 tmask = fsn_mark->mask & ~mask;
510		-
511		- if (flags & FAN_MARK_ONDIR)
512		- tmask &= ~FAN_ONDIR;
513		-
514	729	oldmask = fsn_mark->mask;
515		- fsn_mark->mask = tmask;
	730	+ fsn_mark->mask &= ~mask;
516	731	} else {
517		- __u32 tmask = fsn_mark->ignored_mask & ~mask;
518		- if (flags & FAN_MARK_ONDIR)
519		- tmask &= ~FAN_ONDIR;
520		- fsn_mark->ignored_mask = tmask;
	732	+ fsn_mark->ignored_mask &= ~mask;
521	733	}
522		- *destroy = !(fsn_mark->mask \| fsn_mark->ignored_mask);
	734	+ /*
	735	+ * We need to keep the mark around even if remaining mask cannot
	736	+ * result in any events (e.g. mask == FAN_ONDIR) to support incremenal
	737	+ * changes to the mask.
	738	+ * Destroy mark when only umask bits remain.
	739	+ */
	740	+ *destroy = !((fsn_mark->mask \| fsn_mark->ignored_mask) & ~umask);
523	741	spin_unlock(&fsn_mark->lock);
524	742
525	743	return mask & oldmask;
..	..	@@ -527,7 +745,7 @@
527	745
528	746	static int fanotify_remove_mark(struct fsnotify_group *group,
529	747	fsnotify_connp_t *connp, __u32 mask,
530		- unsigned int flags)
	748	+ unsigned int flags, __u32 umask)
531	749	{
532	750	struct fsnotify_mark *fsn_mark = NULL;
533	751	__u32 removed;
..	..	@@ -541,7 +759,7 @@
541	759	}
542	760
543	761	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
544		- &destroy_mark);
	762	+ umask, &destroy_mark);
545	763	if (removed & fsnotify_conn_mask(fsn_mark->connector))
546	764	fsnotify_recalc_mask(fsn_mark->connector);
547	765	if (destroy_mark)
..	..	@@ -557,18 +775,26 @@
557	775
558	776	static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
559	777	struct vfsmount *mnt, __u32 mask,
560		- unsigned int flags)
	778	+ unsigned int flags, __u32 umask)
561	779	{
562	780	return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
563		- mask, flags);
	781	+ mask, flags, umask);
	782	+}
	783	+
	784	+static int fanotify_remove_sb_mark(struct fsnotify_group *group,
	785	+ struct super_block *sb, __u32 mask,
	786	+ unsigned int flags, __u32 umask)
	787	+{
	788	+ return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
	789	+ flags, umask);
564	790	}
565	791
566	792	static int fanotify_remove_inode_mark(struct fsnotify_group *group,
567	793	struct inode *inode, __u32 mask,
568		- unsigned int flags)
	794	+ unsigned int flags, __u32 umask)
569	795	{
570	796	return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
571		- flags);
	797	+ flags, umask);
572	798	}
573	799
574	800	static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
..	..	@@ -579,19 +805,10 @@
579	805
580	806	spin_lock(&fsn_mark->lock);
581	807	if (!(flags & FAN_MARK_IGNORED_MASK)) {
582		- __u32 tmask = fsn_mark->mask \| mask;
583		-
584		- if (flags & FAN_MARK_ONDIR)
585		- tmask \|= FAN_ONDIR;
586		-
587	808	oldmask = fsn_mark->mask;
588		- fsn_mark->mask = tmask;
	809	+ fsn_mark->mask \|= mask;
589	810	} else {
590		- __u32 tmask = fsn_mark->ignored_mask \| mask;
591		- if (flags & FAN_MARK_ONDIR)
592		- tmask \|= FAN_ONDIR;
593		-
594		- fsn_mark->ignored_mask = tmask;
	811	+ fsn_mark->ignored_mask \|= mask;
595	812	if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
596	813	fsn_mark->flags \|= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
597	814	}
..	..	@@ -602,7 +819,8 @@
602	819
603	820	static struct fsnotify_mark fanotify_add_new_mark(struct fsnotify_group group,
604	821	fsnotify_connp_t *connp,
605		- unsigned int type)
	822	+ unsigned int type,
	823	+ __kernel_fsid_t *fsid)
606	824	{
607	825	struct fsnotify_mark *mark;
608	826	int ret;
..	..	@@ -615,7 +833,7 @@
615	833	return ERR_PTR(-ENOMEM);
616	834
617	835	fsnotify_init_mark(mark, group);
618		- ret = fsnotify_add_mark_locked(mark, connp, type, 0);
	836	+ ret = fsnotify_add_mark_locked(mark, connp, type, 0, fsid);
619	837	if (ret) {
620	838	fsnotify_put_mark(mark);
621	839	return ERR_PTR(ret);
..	..	@@ -627,7 +845,8 @@
627	845
628	846	static int fanotify_add_mark(struct fsnotify_group *group,
629	847	fsnotify_connp_t *connp, unsigned int type,
630		- __u32 mask, unsigned int flags)
	848	+ __u32 mask, unsigned int flags,
	849	+ __kernel_fsid_t *fsid)
631	850	{
632	851	struct fsnotify_mark *fsn_mark;
633	852	__u32 added;
..	..	@@ -635,7 +854,7 @@
635	854	mutex_lock(&group->mark_mutex);
636	855	fsn_mark = fsnotify_find_mark(connp, group);
637	856	if (!fsn_mark) {
638		- fsn_mark = fanotify_add_new_mark(group, connp, type);
	857	+ fsn_mark = fanotify_add_new_mark(group, connp, type, fsid);
639	858	if (IS_ERR(fsn_mark)) {
640	859	mutex_unlock(&group->mark_mutex);
641	860	return PTR_ERR(fsn_mark);
..	..	@@ -652,15 +871,23 @@
652	871
653	872	static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
654	873	struct vfsmount *mnt, __u32 mask,
655		- unsigned int flags)
	874	+ unsigned int flags, __kernel_fsid_t *fsid)
656	875	{
657	876	return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
658		- FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags);
	877	+ FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
	878	+}
	879	+
	880	+static int fanotify_add_sb_mark(struct fsnotify_group *group,
	881	+ struct super_block *sb, __u32 mask,
	882	+ unsigned int flags, __kernel_fsid_t *fsid)
	883	+{
	884	+ return fanotify_add_mark(group, &sb->s_fsnotify_marks,
	885	+ FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
659	886	}
660	887
661	888	static int fanotify_add_inode_mark(struct fsnotify_group *group,
662	889	struct inode *inode, __u32 mask,
663		- unsigned int flags)
	890	+ unsigned int flags, __kernel_fsid_t *fsid)
664	891	{
665	892	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
666	893
..	..	@@ -671,11 +898,25 @@
671	898	*/
672	899	if ((flags & FAN_MARK_IGNORED_MASK) &&
673	900	!(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
674		- (atomic_read(&inode->i_writecount) > 0))
	901	+ inode_is_open_for_write(inode))
675	902	return 0;
676	903
677	904	return fanotify_add_mark(group, &inode->i_fsnotify_marks,
678		- FSNOTIFY_OBJ_TYPE_INODE, mask, flags);
	905	+ FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
	906	+}
	907	+
	908	+static struct fsnotify_event *fanotify_alloc_overflow_event(void)
	909	+{
	910	+ struct fanotify_event *oevent;
	911	+
	912	+ oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT);
	913	+ if (!oevent)
	914	+ return NULL;
	915	+
	916	+ fanotify_init_event(oevent, 0, FS_Q_OVERFLOW);
	917	+ oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW;
	918	+
	919	+ return &oevent->fse;
679	920	}
680	921
681	922	/* fanotify syscalls */
..	..	@@ -684,18 +925,19 @@
684	925	struct fsnotify_group *group;
685	926	int f_flags, fd;
686	927	struct user_struct *user;
687		- struct fanotify_event_info *oevent;
	928	+ unsigned int fid_mode = flags & FANOTIFY_FID_BITS;
	929	+ unsigned int class = flags & FANOTIFY_CLASS_BITS;
688	930
689		- pr_debug("%s: flags=%d event_f_flags=%d\n",
690		- __func__, flags, event_f_flags);
	931	+ pr_debug("%s: flags=%x event_f_flags=%x\n",
	932	+ __func__, flags, event_f_flags);
691	933
692	934	if (!capable(CAP_SYS_ADMIN))
693	935	return -EPERM;
694	936
695	937	#ifdef CONFIG_AUDITSYSCALL
696		- if (flags & ~(FAN_ALL_INIT_FLAGS \| FAN_ENABLE_AUDIT))
	938	+ if (flags & ~(FANOTIFY_INIT_FLAGS \| FAN_ENABLE_AUDIT))
697	939	#else
698		- if (flags & ~FAN_ALL_INIT_FLAGS)
	940	+ if (flags & ~FANOTIFY_INIT_FLAGS)
699	941	#endif
700	942	return -EINVAL;
701	943
..	..	@@ -710,6 +952,16 @@
710	952	default:
711	953	return -EINVAL;
712	954	}
	955	+
	956	+ if (fid_mode && class != FAN_CLASS_NOTIF)
	957	+ return -EINVAL;
	958	+
	959	+ /*
	960	+ * Child name is reported with parent fid so requires dir fid.
	961	+ * We can report both child fid and dir fid with or without name.
	962	+ */
	963	+ if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID))
	964	+ return -EINVAL;
713	965
714	966	user = get_current_user();
715	967	if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
..	..	@@ -731,22 +983,22 @@
731	983	}
732	984
733	985	group->fanotify_data.user = user;
	986	+ group->fanotify_data.flags = flags;
734	987	atomic_inc(&user->fanotify_listeners);
735	988	group->memcg = get_mem_cgroup_from_mm(current->mm);
736	989
737		- oevent = fanotify_alloc_event(group, NULL, FS_Q_OVERFLOW, NULL);
738		- if (unlikely(!oevent)) {
	990	+ group->overflow_event = fanotify_alloc_overflow_event();
	991	+ if (unlikely(!group->overflow_event)) {
739	992	fd = -ENOMEM;
740	993	goto out_destroy_group;
741	994	}
742		- group->overflow_event = &oevent->fse;
743	995
744	996	if (force_o_largefile())
745	997	event_f_flags \|= O_LARGEFILE;
746	998	group->fanotify_data.f_flags = event_f_flags;
747	999	init_waitqueue_head(&group->fanotify_data.access_waitq);
748	1000	INIT_LIST_HEAD(&group->fanotify_data.access_list);
749		- switch (flags & FAN_ALL_CLASS_BITS) {
	1001	+ switch (class) {
750	1002	case FAN_CLASS_NOTIF:
751	1003	group->priority = FS_PRIO_0;
752	1004	break;
..	..	@@ -783,7 +1035,6 @@
783	1035	fd = -EPERM;
784	1036	if (!capable(CAP_AUDIT_WRITE))
785	1037	goto out_destroy_group;
786		- group->fanotify_data.audit = true;
787	1038	}
788	1039
789	1040	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
..	..	@@ -797,6 +1048,82 @@
797	1048	return fd;
798	1049	}
799	1050
	1051	+/* Check if filesystem can encode a unique fid */
	1052	+static int fanotify_test_fid(struct path path, __kernel_fsid_t fsid)
	1053	+{
	1054	+ __kernel_fsid_t root_fsid;
	1055	+ int err;
	1056	+
	1057	+ /*
	1058	+ * Make sure path is not in filesystem with zero fsid (e.g. tmpfs).
	1059	+ */
	1060	+ err = vfs_get_fsid(path->dentry, fsid);
	1061	+ if (err)
	1062	+ return err;
	1063	+
	1064	+ if (!fsid->val[0] && !fsid->val[1])
	1065	+ return -ENODEV;
	1066	+
	1067	+ /*
	1068	+ * Make sure path is not inside a filesystem subvolume (e.g. btrfs)
	1069	+ * which uses a different fsid than sb root.
	1070	+ */
	1071	+ err = vfs_get_fsid(path->dentry->d_sb->s_root, &root_fsid);
	1072	+ if (err)
	1073	+ return err;
	1074	+
	1075	+ if (root_fsid.val[0] != fsid->val[0] \|\|
	1076	+ root_fsid.val[1] != fsid->val[1])
	1077	+ return -EXDEV;
	1078	+
	1079	+ /*
	1080	+ * We need to make sure that the file system supports at least
	1081	+ * encoding a file handle so user can use name_to_handle_at() to
	1082	+ * compare fid returned with event to the file handle of watched
	1083	+ * objects. However, name_to_handle_at() requires that the
	1084	+ * filesystem also supports decoding file handles.
	1085	+ */
	1086	+ if (!path->dentry->d_sb->s_export_op \|\|
	1087	+ !path->dentry->d_sb->s_export_op->fh_to_dentry)
	1088	+ return -EOPNOTSUPP;
	1089	+
	1090	+ return 0;
	1091	+}
	1092	+
	1093	+static int fanotify_events_supported(struct path *path, __u64 mask,
	1094	+ unsigned int flags)
	1095	+{
	1096	+ unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
	1097	+
	1098	+ /*
	1099	+ * Some filesystems such as 'proc' acquire unusual locks when opening
	1100	+ * files. For them fanotify permission events have high chances of
	1101	+ * deadlocking the system - open done when reporting fanotify event
	1102	+ * blocks on this "unusual" lock while another process holding the lock
	1103	+ * waits for fanotify permission event to be answered. Just disallow
	1104	+ * permission events for such filesystems.
	1105	+ */
	1106	+ if (mask & FANOTIFY_PERM_EVENTS &&
	1107	+ path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM)
	1108	+ return -EINVAL;
	1109	+
	1110	+ /*
	1111	+ * mount and sb marks are not allowed on kernel internal pseudo fs,
	1112	+ * like pipe_mnt, because that would subscribe to events on all the
	1113	+ * anonynous pipes in the system.
	1114	+ *
	1115	+ * SB_NOUSER covers all of the internal pseudo fs whose objects are not
	1116	+ * exposed to user's mount namespace, but there are other SB_KERNMOUNT
	1117	+ * fs, like nsfs, debugfs, for which the value of allowing sb and mount
	1118	+ * mark is questionable. For now we leave them alone.
	1119	+ */
	1120	+ if (mark_type != FAN_MARK_INODE &&
	1121	+ path->mnt->mnt_sb->s_flags & SB_NOUSER)
	1122	+ return -EINVAL;
	1123	+
	1124	+ return 0;
	1125	+}
	1126	+
800	1127	static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
801	1128	int dfd, const char __user *pathname)
802	1129	{
..	..	@@ -805,7 +1132,12 @@
805	1132	struct fsnotify_group *group;
806	1133	struct fd f;
807	1134	struct path path;
808		- u32 valid_mask = FAN_ALL_EVENTS \| FAN_EVENT_ON_CHILD;
	1135	+ __kernel_fsid_t __fsid, *fsid = NULL;
	1136	+ u32 valid_mask = FANOTIFY_EVENTS \| FANOTIFY_EVENT_FLAGS;
	1137	+ unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS;
	1138	+ bool ignored = flags & FAN_MARK_IGNORED_MASK;
	1139	+ unsigned int obj_type, fid_mode;
	1140	+ u32 umask = 0;
809	1141	int ret;
810	1142
811	1143	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
..	..	@@ -815,32 +1147,46 @@
815	1147	if (mask & ((__u64)0xffffffff << 32))
816	1148	return -EINVAL;
817	1149
818		- if (flags & ~FAN_ALL_MARK_FLAGS)
	1150	+ if (flags & ~FANOTIFY_MARK_FLAGS)
819	1151	return -EINVAL;
	1152	+
	1153	+ switch (mark_type) {
	1154	+ case FAN_MARK_INODE:
	1155	+ obj_type = FSNOTIFY_OBJ_TYPE_INODE;
	1156	+ break;
	1157	+ case FAN_MARK_MOUNT:
	1158	+ obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT;
	1159	+ break;
	1160	+ case FAN_MARK_FILESYSTEM:
	1161	+ obj_type = FSNOTIFY_OBJ_TYPE_SB;
	1162	+ break;
	1163	+ default:
	1164	+ return -EINVAL;
	1165	+ }
	1166	+
820	1167	switch (flags & (FAN_MARK_ADD \| FAN_MARK_REMOVE \| FAN_MARK_FLUSH)) {
821		- case FAN_MARK_ADD: /* fallthrough */
	1168	+ case FAN_MARK_ADD:
822	1169	case FAN_MARK_REMOVE:
823	1170	if (!mask)
824	1171	return -EINVAL;
825	1172	break;
826	1173	case FAN_MARK_FLUSH:
827		- if (flags & ~(FAN_MARK_MOUNT \| FAN_MARK_FLUSH))
	1174	+ if (flags & ~(FANOTIFY_MARK_TYPE_BITS \| FAN_MARK_FLUSH))
828	1175	return -EINVAL;
829	1176	break;
830	1177	default:
831	1178	return -EINVAL;
832	1179	}
833	1180
834		- if (mask & FAN_ONDIR) {
835		- flags \|= FAN_MARK_ONDIR;
836		- mask &= ~FAN_ONDIR;
837		- }
838		-
839	1181	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
840		- valid_mask \|= FAN_ALL_PERM_EVENTS;
	1182	+ valid_mask \|= FANOTIFY_PERM_EVENTS;
841	1183
842	1184	if (mask & ~valid_mask)
843	1185	return -EINVAL;
	1186	+
	1187	+ /* Event flags (ONDIR, ON_CHILD) are meaningless in ignored mask */
	1188	+ if (ignored)
	1189	+ mask &= ~FANOTIFY_EVENT_FLAGS;
844	1190
845	1191	f = fdget(fanotify_fd);
846	1192	if (unlikely(!f.file))
..	..	@@ -857,73 +1203,123 @@
857	1203	* allowed to set permissions events.
858	1204	*/
859	1205	ret = -EINVAL;
860		- if (mask & FAN_ALL_PERM_EVENTS &&
	1206	+ if (mask & FANOTIFY_PERM_EVENTS &&
861	1207	group->priority == FS_PRIO_0)
	1208	+ goto fput_and_out;
	1209	+
	1210	+ /*
	1211	+ * Events with data type inode do not carry enough information to report
	1212	+ * event->fd, so we do not allow setting a mask for inode events unless
	1213	+ * group supports reporting fid.
	1214	+ * inode events are not supported on a mount mark, because they do not
	1215	+ * carry enough information (i.e. path) to be filtered by mount point.
	1216	+ */
	1217	+ fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
	1218	+ if (mask & FANOTIFY_INODE_EVENTS &&
	1219	+ (!fid_mode \|\| mark_type == FAN_MARK_MOUNT))
862	1220	goto fput_and_out;
863	1221
864	1222	if (flags & FAN_MARK_FLUSH) {
865	1223	ret = 0;
866		- if (flags & FAN_MARK_MOUNT)
	1224	+ if (mark_type == FAN_MARK_MOUNT)
867	1225	fsnotify_clear_vfsmount_marks_by_group(group);
	1226	+ else if (mark_type == FAN_MARK_FILESYSTEM)
	1227	+ fsnotify_clear_sb_marks_by_group(group);
868	1228	else
869	1229	fsnotify_clear_inode_marks_by_group(group);
870	1230	goto fput_and_out;
871	1231	}
872	1232
873		- ret = fanotify_find_path(dfd, pathname, &path, flags);
	1233	+ ret = fanotify_find_path(dfd, pathname, &path, flags,
	1234	+ (mask & ALL_FSNOTIFY_EVENTS), obj_type);
874	1235	if (ret)
875	1236	goto fput_and_out;
876	1237
	1238	+ if (flags & FAN_MARK_ADD) {
	1239	+ ret = fanotify_events_supported(&path, mask, flags);
	1240	+ if (ret)
	1241	+ goto path_put_and_out;
	1242	+ }
	1243	+
	1244	+ if (fid_mode) {
	1245	+ ret = fanotify_test_fid(&path, &__fsid);
	1246	+ if (ret)
	1247	+ goto path_put_and_out;
	1248	+
	1249	+ fsid = &__fsid;
	1250	+ }
	1251	+
877	1252	/* inode held in place by reference to path; group by fget on fd */
878		- if (!(flags & FAN_MARK_MOUNT))
	1253	+ if (mark_type == FAN_MARK_INODE)
879	1254	inode = path.dentry->d_inode;
880	1255	else
881	1256	mnt = path.mnt;
882	1257
	1258	+ /* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
	1259	+ if (mnt \|\| !S_ISDIR(inode->i_mode)) {
	1260	+ mask &= ~FAN_EVENT_ON_CHILD;
	1261	+ umask = FAN_EVENT_ON_CHILD;
	1262	+ /*
	1263	+ * If group needs to report parent fid, register for getting
	1264	+ * events with parent/name info for non-directory.
	1265	+ */
	1266	+ if ((fid_mode & FAN_REPORT_DIR_FID) &&
	1267	+ (flags & FAN_MARK_ADD) && !ignored)
	1268	+ mask \|= FAN_EVENT_ON_CHILD;
	1269	+ }
	1270	+
883	1271	/* create/update an inode mark */
884	1272	switch (flags & (FAN_MARK_ADD \| FAN_MARK_REMOVE)) {
885	1273	case FAN_MARK_ADD:
886		- if (flags & FAN_MARK_MOUNT)
887		- ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
	1274	+ if (mark_type == FAN_MARK_MOUNT)
	1275	+ ret = fanotify_add_vfsmount_mark(group, mnt, mask,
	1276	+ flags, fsid);
	1277	+ else if (mark_type == FAN_MARK_FILESYSTEM)
	1278	+ ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
	1279	+ flags, fsid);
888	1280	else
889		- ret = fanotify_add_inode_mark(group, inode, mask, flags);
	1281	+ ret = fanotify_add_inode_mark(group, inode, mask,
	1282	+ flags, fsid);
890	1283	break;
891	1284	case FAN_MARK_REMOVE:
892		- if (flags & FAN_MARK_MOUNT)
893		- ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
	1285	+ if (mark_type == FAN_MARK_MOUNT)
	1286	+ ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
	1287	+ flags, umask);
	1288	+ else if (mark_type == FAN_MARK_FILESYSTEM)
	1289	+ ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
	1290	+ flags, umask);
894	1291	else
895		- ret = fanotify_remove_inode_mark(group, inode, mask, flags);
	1292	+ ret = fanotify_remove_inode_mark(group, inode, mask,
	1293	+ flags, umask);
896	1294	break;
897	1295	default:
898	1296	ret = -EINVAL;
899	1297	}
900	1298
	1299	+path_put_and_out:
901	1300	path_put(&path);
902	1301	fput_and_out:
903	1302	fdput(f);
904	1303	return ret;
905	1304	}
906	1305
	1306	+#ifndef CONFIG_ARCH_SPLIT_ARG64
907	1307	SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
908	1308	__u64, mask, int, dfd,
909	1309	const char __user *, pathname)
910	1310	{
911	1311	return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
912	1312	}
	1313	+#endif
913	1314
914		-#ifdef CONFIG_COMPAT
915		-COMPAT_SYSCALL_DEFINE6(fanotify_mark,
	1315	+#if defined(CONFIG_ARCH_SPLIT_ARG64) \|\| defined(CONFIG_COMPAT)
	1316	+SYSCALL32_DEFINE6(fanotify_mark,
916	1317	int, fanotify_fd, unsigned int, flags,
917		- __u32, mask0, __u32, mask1, int, dfd,
	1318	+ SC_ARG64(mask), int, dfd,
918	1319	const char __user *, pathname)
919	1320	{
920		- return do_fanotify_mark(fanotify_fd, flags,
921		-#ifdef __BIG_ENDIAN
922		- ((__u64)mask0 << 32) \| mask1,
923		-#else
924		- ((__u64)mask1 << 32) \| mask0,
925		-#endif
926		- dfd, pathname);
	1321	+ return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask),
	1322	+ dfd, pathname);
927	1323	}
928	1324	#endif
929	1325
..	..	@@ -934,12 +1330,18 @@
934	1330	*/
935	1331	static int __init fanotify_user_setup(void)
936	1332	{
	1333	+ BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 10);
	1334	+ BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9);
	1335	+
937	1336	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark,
938	1337	SLAB_PANIC\|SLAB_ACCOUNT);
939		- fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
	1338	+ fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
	1339	+ SLAB_PANIC);
	1340	+ fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
	1341	+ SLAB_PANIC);
940	1342	if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
941	1343	fanotify_perm_event_cachep =
942		- KMEM_CACHE(fanotify_perm_event_info, SLAB_PANIC);
	1344	+ KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);
943	1345	}
944	1346
945	1347	return 0;