Merge tag 'fsnotify_for_v6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs

Pull fsnotify updates from Jan Kara:

 - reduce overhead of fsnotify infrastructure when no permission events
   are in use

 - a few small cleanups

* tag 'fsnotify_for_v6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs:
  fsnotify: fix UAF from FS_ERROR event on a shutting down filesystem
  fsnotify: optimize the case of no permission event watchers
  fsnotify: use an enum for group priority constants
  fsnotify: move s_fsnotify_connectors into fsnotify_sb_info
  fsnotify: lazy attach fsnotify_sb_info state to sb
  fsnotify: create helper fsnotify_update_sb_watchers()
  fsnotify: pass object pointer and type to fsnotify mark helpers
  fanotify: merge two checks regarding add of ignore mark
  fsnotify: create a wrapper fsnotify_find_inode_mark()
  fsnotify: create helpers to get sb and connp from object
  fsnotify: rename fsnotify_{get,put}_sb_connectors()
  fsnotify: Avoid -Wflex-array-member-not-at-end warning
  fanotify: remove unneeded sub-zero check for unsigned value

fs/nfsd/filecache.c

@@ -159,8 +159,8 @@ nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode)
 
 	do {
 		fsnotify_group_lock(nfsd_file_fsnotify_group);
-		mark = fsnotify_find_mark(&inode->i_fsnotify_marks,
-					  nfsd_file_fsnotify_group);
+		mark = fsnotify_find_inode_mark(inode,
+						nfsd_file_fsnotify_group);
 		if (mark) {
 			nfm = nfsd_file_mark_get(container_of(mark,
 						 struct nfsd_file_mark,

fs/notify/dnotify/dnotify.c

@@ -162,7 +162,7 @@ void dnotify_flush(struct file *filp, fl_owner_t id)
 	if (!S_ISDIR(inode->i_mode))
 		return;
 
-	fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, dnotify_group);
+	fsn_mark = fsnotify_find_inode_mark(inode, dnotify_group);
 	if (!fsn_mark)
 		return;
 	dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
@@ -326,7 +326,7 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned int arg)
 	fsnotify_group_lock(dnotify_group);
 
 	/* add the new_fsn_mark or find an old one. */
-	fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, dnotify_group);
+	fsn_mark = fsnotify_find_inode_mark(inode, dnotify_group);
 	if (fsn_mark) {
 		dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
 		spin_lock(&fsn_mark->lock);

fs/notify/fanotify/fanotify_user.c

@@ -502,7 +502,7 @@ static int copy_fid_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh,
 	}
 
 	/* Pad with 0's */
-	WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN);
+	WARN_ON_ONCE(len >= FANOTIFY_EVENT_ALIGN);
 	if (len > 0 && clear_user(buf, len))
 		return -EFAULT;
 
@@ -1076,7 +1076,7 @@ static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 }
 
 static int fanotify_remove_mark(struct fsnotify_group *group,
-				fsnotify_connp_t *connp, __u32 mask,
+				void *obj, unsigned int obj_type, __u32 mask,
 				unsigned int flags, __u32 umask)
 {
 	struct fsnotify_mark *fsn_mark = NULL;
@@ -1084,7 +1084,7 @@ static int fanotify_remove_mark(struct fsnotify_group *group,
 	int destroy_mark;
 
 	fsnotify_group_lock(group);
-	fsn_mark = fsnotify_find_mark(connp, group);
+	fsn_mark = fsnotify_find_mark(obj, obj_type, group);
 	if (!fsn_mark) {
 		fsnotify_group_unlock(group);
 		return -ENOENT;
@@ -1105,30 +1105,6 @@ static int fanotify_remove_mark(struct fsnotify_group *group,
 	return 0;
 }
 
-static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
-					 struct vfsmount *mnt, __u32 mask,
-					 unsigned int flags, __u32 umask)
-{
-	return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
-				    mask, flags, umask);
-}
-
-static int fanotify_remove_sb_mark(struct fsnotify_group *group,
-				   struct super_block *sb, __u32 mask,
-				   unsigned int flags, __u32 umask)
-{
-	return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask,
-				    flags, umask);
-}
-
-static int fanotify_remove_inode_mark(struct fsnotify_group *group,
-				      struct inode *inode, __u32 mask,
-				      unsigned int flags, __u32 umask)
-{
-	return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask,
-				    flags, umask);
-}
-
 static bool fanotify_mark_update_flags(struct fsnotify_mark *fsn_mark,
 				       unsigned int fan_flags)
 {
@@ -1249,7 +1225,7 @@ static int fanotify_set_mark_fsid(struct fsnotify_group *group,
 }
 
 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
-						   fsnotify_connp_t *connp,
+						   void *obj,
 						   unsigned int obj_type,
 						   unsigned int fan_flags,
 						   struct fan_fsid *fsid)
@@ -1288,7 +1264,7 @@ static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
 		fan_mark->fsid.val[0] = fan_mark->fsid.val[1] = 0;
 	}
 
-	ret = fsnotify_add_mark_locked(mark, connp, obj_type, 0);
+	ret = fsnotify_add_mark_locked(mark, obj, obj_type, 0);
 	if (ret)
 		goto out_put_mark;
 
@@ -1344,7 +1320,7 @@ static int fanotify_may_update_existing_mark(struct fsnotify_mark *fsn_mark,
 }
 
 static int fanotify_add_mark(struct fsnotify_group *group,
-			     fsnotify_connp_t *connp, unsigned int obj_type,
+			     void *obj, unsigned int obj_type,
 			     __u32 mask, unsigned int fan_flags,
 			     struct fan_fsid *fsid)
 {
@@ -1353,9 +1329,9 @@ static int fanotify_add_mark(struct fsnotify_group *group,
 	int ret = 0;
 
 	fsnotify_group_lock(group);
-	fsn_mark = fsnotify_find_mark(connp, group);
+	fsn_mark = fsnotify_find_mark(obj, obj_type, group);
 	if (!fsn_mark) {
-		fsn_mark = fanotify_add_new_mark(group, connp, obj_type,
+		fsn_mark = fanotify_add_new_mark(group, obj, obj_type,
 						 fan_flags, fsid);
 		if (IS_ERR(fsn_mark)) {
 			fsnotify_group_unlock(group);
@@ -1392,42 +1368,6 @@ static int fanotify_add_mark(struct fsnotify_group *group,
 	return ret;
 }
 
-static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
-				      struct vfsmount *mnt, __u32 mask,
-				      unsigned int flags, struct fan_fsid *fsid)
-{
-	return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks,
-				 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid);
-}
-
-static int fanotify_add_sb_mark(struct fsnotify_group *group,
-				struct super_block *sb, __u32 mask,
-				unsigned int flags, struct fan_fsid *fsid)
-{
-	return fanotify_add_mark(group, &sb->s_fsnotify_marks,
-				 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid);
-}
-
-static int fanotify_add_inode_mark(struct fsnotify_group *group,
-				   struct inode *inode, __u32 mask,
-				   unsigned int flags, struct fan_fsid *fsid)
-{
-	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
-
-	/*
-	 * If some other task has this inode open for write we should not add
-	 * an ignore mask, unless that ignore mask is supposed to survive
-	 * modification changes anyway.
-	 */
-	if ((flags & FANOTIFY_MARK_IGNORE_BITS) &&
-	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
-	    inode_is_open_for_write(inode))
-		return 0;
-
-	return fanotify_add_mark(group, &inode->i_fsnotify_marks,
-				 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid);
-}
-
 static struct fsnotify_event *fanotify_alloc_overflow_event(void)
 {
 	struct fanotify_event *oevent;
@@ -1576,13 +1516,13 @@ SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 	INIT_LIST_HEAD(&group->fanotify_data.access_list);
 	switch (class) {
 	case FAN_CLASS_NOTIF:
-		group->priority = FS_PRIO_0;
+		group->priority = FSNOTIFY_PRIO_NORMAL;
 		break;
 	case FAN_CLASS_CONTENT:
-		group->priority = FS_PRIO_1;
+		group->priority = FSNOTIFY_PRIO_CONTENT;
 		break;
 	case FAN_CLASS_PRE_CONTENT:
-		group->priority = FS_PRIO_2;
+		group->priority = FSNOTIFY_PRIO_PRE_CONTENT;
 		break;
 	default:
 		fd = -EINVAL;
@@ -1750,6 +1690,7 @@ static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
 	unsigned int mark_cmd = flags & FANOTIFY_MARK_CMD_BITS;
 	unsigned int ignore = flags & FANOTIFY_MARK_IGNORE_BITS;
 	unsigned int obj_type, fid_mode;
+	void *obj;
 	u32 umask = 0;
 	int ret;
 
@@ -1833,12 +1774,11 @@ static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
 		goto fput_and_out;
 
 	/*
-	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
-	 * allowed to set permissions events.
+	 * Permission events require minimum priority FAN_CLASS_CONTENT.
 	 */
 	ret = -EINVAL;
 	if (mask & FANOTIFY_PERM_EVENTS &&
-	    group->priority == FS_PRIO_0)
+	    group->priority < FSNOTIFY_PRIO_CONTENT)
 		goto fput_and_out;
 
 	if (mask & FAN_FS_ERROR &&
@@ -1908,17 +1848,34 @@ static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
 	}
 
 	/* inode held in place by reference to path; group by fget on fd */
-	if (mark_type == FAN_MARK_INODE)
+	if (mark_type == FAN_MARK_INODE) {
 		inode = path.dentry->d_inode;
-	else
+		obj = inode;
+	} else {
 		mnt = path.mnt;
+		if (mark_type == FAN_MARK_MOUNT)
+			obj = mnt;
+		else
+			obj = mnt->mnt_sb;
+	}
 
-	ret = mnt ? -EINVAL : -EISDIR;
-	/* FAN_MARK_IGNORE requires SURV_MODIFY for sb/mount/dir marks */
-	if (mark_cmd == FAN_MARK_ADD && ignore == FAN_MARK_IGNORE &&
-	    (mnt || S_ISDIR(inode->i_mode)) &&
-	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY))
-		goto path_put_and_out;
+	/*
+	 * If some other task has this inode open for write we should not add
+	 * an ignore mask, unless that ignore mask is supposed to survive
+	 * modification changes anyway.
+	 */
+	if (mark_cmd == FAN_MARK_ADD && (flags & FANOTIFY_MARK_IGNORE_BITS) &&
+	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY)) {
+		ret = mnt ? -EINVAL : -EISDIR;
+		/* FAN_MARK_IGNORE requires SURV_MODIFY for sb/mount/dir marks */
+		if (ignore == FAN_MARK_IGNORE &&
+		    (mnt || S_ISDIR(inode->i_mode)))
+			goto path_put_and_out;
+
+		ret = 0;
+		if (inode && inode_is_open_for_write(inode))
+			goto path_put_and_out;
+	}
 
 	/* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */
 	if (mnt || !S_ISDIR(inode->i_mode)) {
@@ -1936,26 +1893,12 @@ static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
 	/* create/update an inode mark */
 	switch (mark_cmd) {
 	case FAN_MARK_ADD:
-		if (mark_type == FAN_MARK_MOUNT)
-			ret = fanotify_add_vfsmount_mark(group, mnt, mask,
-							 flags, fsid);
-		else if (mark_type == FAN_MARK_FILESYSTEM)
-			ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask,
-						   flags, fsid);
-		else
-			ret = fanotify_add_inode_mark(group, inode, mask,
-						      flags, fsid);
+		ret = fanotify_add_mark(group, obj, obj_type, mask, flags,
+					fsid);
 		break;
 	case FAN_MARK_REMOVE:
-		if (mark_type == FAN_MARK_MOUNT)
-			ret = fanotify_remove_vfsmount_mark(group, mnt, mask,
-							    flags, umask);
-		else if (mark_type == FAN_MARK_FILESYSTEM)
-			ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask,
-						      flags, umask);
-		else
-			ret = fanotify_remove_inode_mark(group, inode, mask,
-							 flags, umask);
+		ret = fanotify_remove_mark(group, obj, obj_type, mask, flags,
+					   umask);
 		break;
 	default:
 		ret = -EINVAL;

fs/notify/fdinfo.c

@@ -41,29 +41,25 @@ static void show_fdinfo(struct seq_file *m, struct file *f,
 #if defined(CONFIG_EXPORTFS)
 static void show_mark_fhandle(struct seq_file *m, struct inode *inode)
 {
-	struct {
-		struct file_handle handle;
-		u8 pad[MAX_HANDLE_SZ];
-	} f;
+	DEFINE_FLEX(struct file_handle, f, f_handle, handle_bytes, MAX_HANDLE_SZ);
 	int size, ret, i;
 
-	f.handle.handle_bytes = sizeof(f.pad);
-	size = f.handle.handle_bytes >> 2;
+	size = f->handle_bytes >> 2;
 
-	ret = exportfs_encode_fid(inode, (struct fid *)f.handle.f_handle, &size);
+	ret = exportfs_encode_fid(inode, (struct fid *)f->f_handle, &size);
 	if ((ret == FILEID_INVALID) || (ret < 0)) {
 		WARN_ONCE(1, "Can't encode file handler for inotify: %d\n", ret);
 		return;
 	}
 
-	f.handle.handle_type = ret;
-	f.handle.handle_bytes = size * sizeof(u32);
+	f->handle_type = ret;
+	f->handle_bytes = size * sizeof(u32);
 
 	seq_printf(m, "fhandle-bytes:%x fhandle-type:%x f_handle:",
-		   f.handle.handle_bytes, f.handle.handle_type);
+		   f->handle_bytes, f->handle_type);
 
-	for (i = 0; i < f.handle.handle_bytes; i++)
-		seq_printf(m, "%02x", (int)f.handle.f_handle[i]);
+	for (i = 0; i < f->handle_bytes; i++)
+		seq_printf(m, "%02x", (int)f->f_handle[i]);
 }
 #else
 static void show_mark_fhandle(struct seq_file *m, struct inode *inode)

fs/notify/fsnotify.c

@@ -89,11 +89,25 @@ static void fsnotify_unmount_inodes(struct super_block *sb)
 
 void fsnotify_sb_delete(struct super_block *sb)
 {
+	struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
+
+	/* Were any marks ever added to any object on this sb? */
+	if (!sbinfo)
+		return;
+
 	fsnotify_unmount_inodes(sb);
 	fsnotify_clear_marks_by_sb(sb);
 	/* Wait for outstanding object references from connectors */
-	wait_var_event(&sb->s_fsnotify_connectors,
-		       !atomic_long_read(&sb->s_fsnotify_connectors));
+	wait_var_event(fsnotify_sb_watched_objects(sb),
+		       !atomic_long_read(fsnotify_sb_watched_objects(sb)));
+	WARN_ON(fsnotify_sb_has_priority_watchers(sb, FSNOTIFY_PRIO_CONTENT));
+	WARN_ON(fsnotify_sb_has_priority_watchers(sb,
+						  FSNOTIFY_PRIO_PRE_CONTENT));
+}
+
+void fsnotify_sb_free(struct super_block *sb)
+{
+	kfree(sb->s_fsnotify_info);
 }
 
 /*
@@ -489,6 +503,7 @@ int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
 {
 	const struct path *path = fsnotify_data_path(data, data_type);
 	struct super_block *sb = fsnotify_data_sb(data, data_type);
+	struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
 	struct fsnotify_iter_info iter_info = {};
 	struct mount *mnt = NULL;
 	struct inode *inode2 = NULL;
@@ -525,7 +540,7 @@ int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
 	 * SRCU because we have no references to any objects and do not
 	 * need SRCU to keep them "alive".
 	 */
-	if (!sb->s_fsnotify_marks &&
+	if ((!sbinfo || !sbinfo->sb_marks) &&
 	    (!mnt || !mnt->mnt_fsnotify_marks) &&
 	    (!inode || !inode->i_fsnotify_marks) &&
 	    (!inode2 || !inode2->i_fsnotify_marks))
@@ -552,8 +567,10 @@ int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
 
 	iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
 
-	iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
-		fsnotify_first_mark(&sb->s_fsnotify_marks);
+	if (sbinfo) {
+		iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
+			fsnotify_first_mark(&sbinfo->sb_marks);
+	}
 	if (mnt) {
 		iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
 			fsnotify_first_mark(&mnt->mnt_fsnotify_marks);

fs/notify/fsnotify.h

@@ -9,39 +9,58 @@
 
 #include "../mount.h"
 
+/*
+ * fsnotify_connp_t is what we embed in objects which connector can be attached
+ * to.
+ */
+typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t;
+
 static inline struct inode *fsnotify_conn_inode(
 				struct fsnotify_mark_connector *conn)
 {
-	return container_of(conn->obj, struct inode, i_fsnotify_marks);
+	return conn->obj;
 }
 
 static inline struct mount *fsnotify_conn_mount(
 				struct fsnotify_mark_connector *conn)
 {
-	return container_of(conn->obj, struct mount, mnt_fsnotify_marks);
+	return real_mount(conn->obj);
 }
 
 static inline struct super_block *fsnotify_conn_sb(
 				struct fsnotify_mark_connector *conn)
 {
-	return container_of(conn->obj, struct super_block, s_fsnotify_marks);
+	return conn->obj;
 }
 
-static inline struct super_block *fsnotify_connector_sb(
-				struct fsnotify_mark_connector *conn)
+static inline struct super_block *fsnotify_object_sb(void *obj,
+			enum fsnotify_obj_type obj_type)
 {
-	switch (conn->type) {
+	switch (obj_type) {
 	case FSNOTIFY_OBJ_TYPE_INODE:
-		return fsnotify_conn_inode(conn)->i_sb;
+		return ((struct inode *)obj)->i_sb;
 	case FSNOTIFY_OBJ_TYPE_VFSMOUNT:
-		return fsnotify_conn_mount(conn)->mnt.mnt_sb;
+		return ((struct vfsmount *)obj)->mnt_sb;
 	case FSNOTIFY_OBJ_TYPE_SB:
-		return fsnotify_conn_sb(conn);
+		return (struct super_block *)obj;
 	default:
 		return NULL;
 	}
 }
 
+static inline struct super_block *fsnotify_connector_sb(
+				struct fsnotify_mark_connector *conn)
+{
+	return fsnotify_object_sb(conn->obj, conn->type);
+}
+
+static inline fsnotify_connp_t *fsnotify_sb_marks(struct super_block *sb)
+{
+	struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
+
+	return sbinfo ? &sbinfo->sb_marks : NULL;
+}
+
 /* destroy all events sitting in this groups notification queue */
 extern void fsnotify_flush_notify(struct fsnotify_group *group);
 
@@ -67,7 +86,7 @@ static inline void fsnotify_clear_marks_by_mount(struct vfsmount *mnt)
 /* run the list of all marks associated with sb and destroy them */
 static inline void fsnotify_clear_marks_by_sb(struct super_block *sb)
 {
-	fsnotify_destroy_marks(&sb->s_fsnotify_marks);
+	fsnotify_destroy_marks(fsnotify_sb_marks(sb));
 }
 
 /*

fs/notify/inotify/inotify_user.c

@@ -544,7 +544,7 @@ static int inotify_update_existing_watch(struct fsnotify_group *group,
 	int create = (arg & IN_MASK_CREATE);
 	int ret;
 
-	fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
+	fsn_mark = fsnotify_find_inode_mark(inode, group);
 	if (!fsn_mark)
 		return -ENOENT;
 	else if (create) {

fs/notify/mark.c

@@ -97,6 +97,21 @@ void fsnotify_get_mark(struct fsnotify_mark *mark)
 	refcount_inc(&mark->refcnt);
 }
 
+static fsnotify_connp_t *fsnotify_object_connp(void *obj,
+				enum fsnotify_obj_type obj_type)
+{
+	switch (obj_type) {
+	case FSNOTIFY_OBJ_TYPE_INODE:
+		return &((struct inode *)obj)->i_fsnotify_marks;
+	case FSNOTIFY_OBJ_TYPE_VFSMOUNT:
+		return &real_mount(obj)->mnt_fsnotify_marks;
+	case FSNOTIFY_OBJ_TYPE_SB:
+		return fsnotify_sb_marks(obj);
+	default:
+		return NULL;
+	}
+}
+
 static __u32 *fsnotify_conn_mask_p(struct fsnotify_mark_connector *conn)
 {
 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
@@ -116,10 +131,69 @@ __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn)
 	return *fsnotify_conn_mask_p(conn);
 }
 
+static void fsnotify_get_sb_watched_objects(struct super_block *sb)
+{
+	atomic_long_inc(fsnotify_sb_watched_objects(sb));
+}
+
+static void fsnotify_put_sb_watched_objects(struct super_block *sb)
+{
+	if (atomic_long_dec_and_test(fsnotify_sb_watched_objects(sb)))
+		wake_up_var(fsnotify_sb_watched_objects(sb));
+}
+
 static void fsnotify_get_inode_ref(struct inode *inode)
 {
 	ihold(inode);
-	atomic_long_inc(&inode->i_sb->s_fsnotify_connectors);
+	fsnotify_get_sb_watched_objects(inode->i_sb);
+}
+
+static void fsnotify_put_inode_ref(struct inode *inode)
+{
+	fsnotify_put_sb_watched_objects(inode->i_sb);
+	iput(inode);
+}
+
+/*
+ * Grab or drop watched objects reference depending on whether the connector
+ * is attached and has any marks attached.
+ */
+static void fsnotify_update_sb_watchers(struct super_block *sb,
+					struct fsnotify_mark_connector *conn)
+{
+	struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
+	bool is_watched = conn->flags & FSNOTIFY_CONN_FLAG_IS_WATCHED;
+	struct fsnotify_mark *first_mark = NULL;
+	unsigned int highest_prio = 0;
+
+	if (conn->obj)
+		first_mark = hlist_entry_safe(conn->list.first,
+					      struct fsnotify_mark, obj_list);
+	if (first_mark)
+		highest_prio = first_mark->group->priority;
+	if (WARN_ON(highest_prio >= __FSNOTIFY_PRIO_NUM))
+		highest_prio = 0;
+
+	/*
+	 * If the highest priority of group watching this object is prio,
+	 * then watched object has a reference on counters [0..prio].
+	 * Update priority >= 1 watched objects counters.
+	 */
+	for (unsigned int p = conn->prio + 1; p <= highest_prio; p++)
+		atomic_long_inc(&sbinfo->watched_objects[p]);
+	for (unsigned int p = conn->prio; p > highest_prio; p--)
+		atomic_long_dec(&sbinfo->watched_objects[p]);
+	conn->prio = highest_prio;
+
+	/* Update priority >= 0 (a.k.a total) watched objects counter */
+	BUILD_BUG_ON(FSNOTIFY_PRIO_NORMAL != 0);
+	if (first_mark && !is_watched) {
+		conn->flags |= FSNOTIFY_CONN_FLAG_IS_WATCHED;
+		fsnotify_get_sb_watched_objects(sb);
+	} else if (!first_mark && is_watched) {
+		conn->flags &= ~FSNOTIFY_CONN_FLAG_IS_WATCHED;
+		fsnotify_put_sb_watched_objects(sb);
+	}
 }
 
 /*
@@ -213,35 +287,12 @@ static void fsnotify_connector_destroy_workfn(struct work_struct *work)
 	}
 }
 
-static void fsnotify_put_inode_ref(struct inode *inode)
-{
-	struct super_block *sb = inode->i_sb;
-
-	iput(inode);
-	if (atomic_long_dec_and_test(&sb->s_fsnotify_connectors))
-		wake_up_var(&sb->s_fsnotify_connectors);
-}
-
-static void fsnotify_get_sb_connectors(struct fsnotify_mark_connector *conn)
-{
-	struct super_block *sb = fsnotify_connector_sb(conn);
-
-	if (sb)
-		atomic_long_inc(&sb->s_fsnotify_connectors);
-}
-
-static void fsnotify_put_sb_connectors(struct fsnotify_mark_connector *conn)
-{
-	struct super_block *sb = fsnotify_connector_sb(conn);
-
-	if (sb && atomic_long_dec_and_test(&sb->s_fsnotify_connectors))
-		wake_up_var(&sb->s_fsnotify_connectors);
-}
-
 static void *fsnotify_detach_connector_from_object(
 					struct fsnotify_mark_connector *conn,
 					unsigned int *type)
 {
+	fsnotify_connp_t *connp = fsnotify_object_connp(conn->obj, conn->type);
+	struct super_block *sb = fsnotify_connector_sb(conn);
 	struct inode *inode = NULL;
 
 	*type = conn->type;
@@ -261,10 +312,10 @@ static void *fsnotify_detach_connector_from_object(
 		fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
 	}
 
-	fsnotify_put_sb_connectors(conn);
-	rcu_assign_pointer(*(conn->obj), NULL);
+	rcu_assign_pointer(*connp, NULL);
 	conn->obj = NULL;
 	conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
+	fsnotify_update_sb_watchers(sb, conn);
 
 	return inode;
 }
@@ -316,6 +367,11 @@ void fsnotify_put_mark(struct fsnotify_mark *mark)
 		objp = fsnotify_detach_connector_from_object(conn, &type);
 		free_conn = true;
 	} else {
+		struct super_block *sb = fsnotify_connector_sb(conn);
+
+		/* Update watched objects after detaching mark */
+		if (sb)
+			fsnotify_update_sb_watchers(sb, conn);
 		objp = __fsnotify_recalc_mask(conn);
 		type = conn->type;
 	}
@@ -536,8 +592,28 @@ int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
 	return -1;
 }
 
+static int fsnotify_attach_info_to_sb(struct super_block *sb)
+{
+	struct fsnotify_sb_info *sbinfo;
+
+	/* sb info is freed on fsnotify_sb_delete() */
+	sbinfo = kzalloc(sizeof(*sbinfo), GFP_KERNEL);
+	if (!sbinfo)
+		return -ENOMEM;
+
+	/*
+	 * cmpxchg() provides the barrier so that callers of fsnotify_sb_info()
+	 * will observe an initialized structure
+	 */
+	if (cmpxchg(&sb->s_fsnotify_info, NULL, sbinfo)) {
+		/* Someone else created sbinfo for us */
+		kfree(sbinfo);
+	}
+	return 0;
+}
+
 static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
-					       unsigned int obj_type)
+					       void *obj, unsigned int obj_type)
 {
 	struct fsnotify_mark_connector *conn;
 
@@ -547,10 +623,9 @@ static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
 	spin_lock_init(&conn->lock);
 	INIT_HLIST_HEAD(&conn->list);
 	conn->flags = 0;
+	conn->prio = 0;
 	conn->type = obj_type;
-	conn->obj = connp;
-	conn->flags = 0;
-	fsnotify_get_sb_connectors(conn);
+	conn->obj = obj;
 
 	/*
 	 * cmpxchg() provides the barrier so that readers of *connp can see
@@ -558,10 +633,8 @@ static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
 	 */
 	if (cmpxchg(connp, NULL, conn)) {
 		/* Someone else created list structure for us */
-		fsnotify_put_sb_connectors(conn);
 		kmem_cache_free(fsnotify_mark_connector_cachep, conn);
 	}
-
 	return 0;
 }
 
@@ -598,24 +671,36 @@ static struct fsnotify_mark_connector *fsnotify_grab_connector(
  * to which group and for which inodes. These marks are ordered according to
  * priority, highest number first, and then by the group's location in memory.
  */
-static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
-				  fsnotify_connp_t *connp,
+static int fsnotify_add_mark_list(struct fsnotify_mark *mark, void *obj,
 				  unsigned int obj_type, int add_flags)
 {
+	struct super_block *sb = fsnotify_object_sb(obj, obj_type);
 	struct fsnotify_mark *lmark, *last = NULL;
 	struct fsnotify_mark_connector *conn;
+	fsnotify_connp_t *connp;
 	int cmp;
 	int err = 0;
 
 	if (WARN_ON(!fsnotify_valid_obj_type(obj_type)))
 		return -EINVAL;
 
+	/*
+	 * Attach the sb info before attaching a connector to any object on sb.
+	 * The sb info will remain attached as long as sb lives.
+	 */
+	if (!fsnotify_sb_info(sb)) {
+		err = fsnotify_attach_info_to_sb(sb);
+		if (err)
+			return err;
+	}
+
+	connp = fsnotify_object_connp(obj, obj_type);
 restart:
 	spin_lock(&mark->lock);
 	conn = fsnotify_grab_connector(connp);
 	if (!conn) {
 		spin_unlock(&mark->lock);
-		err = fsnotify_attach_connector_to_object(connp, obj_type);
+		err = fsnotify_attach_connector_to_object(connp, obj, obj_type);
 		if (err)
 			return err;
 		goto restart;
@@ -649,6 +734,7 @@ static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
 	/* mark should be the last entry.  last is the current last entry */
 	hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
 added:
+	fsnotify_update_sb_watchers(sb, conn);
 	/*
 	 * Since connector is attached to object using cmpxchg() we are
 	 * guaranteed that connector initialization is fully visible by anyone
@@ -667,7 +753,7 @@ static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
  * event types should be delivered to which group.
  */
 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
-			     fsnotify_connp_t *connp, unsigned int obj_type,
+			     void *obj, unsigned int obj_type,
 			     int add_flags)
 {
 	struct fsnotify_group *group = mark->group;
@@ -688,7 +774,7 @@ int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
 	fsnotify_get_mark(mark); /* for g_list */
 	spin_unlock(&mark->lock);
 
-	ret = fsnotify_add_mark_list(mark, connp, obj_type, add_flags);
+	ret = fsnotify_add_mark_list(mark, obj, obj_type, add_flags);
 	if (ret)
 		goto err;
 
@@ -706,14 +792,14 @@ int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
 	return ret;
 }
 
-int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp,
+int fsnotify_add_mark(struct fsnotify_mark *mark, void *obj,
 		      unsigned int obj_type, int add_flags)
 {
 	int ret;
 	struct fsnotify_group *group = mark->group;
 
 	fsnotify_group_lock(group);
-	ret = fsnotify_add_mark_locked(mark, connp, obj_type, add_flags);
+	ret = fsnotify_add_mark_locked(mark, obj, obj_type, add_flags);
 	fsnotify_group_unlock(group);
 	return ret;
 }
@@ -723,12 +809,16 @@ EXPORT_SYMBOL_GPL(fsnotify_add_mark);
  * Given a list of marks, find the mark associated with given group. If found
  * take a reference to that mark and return it, else return NULL.
  */
-struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
+struct fsnotify_mark *fsnotify_find_mark(void *obj, unsigned int obj_type,
 					 struct fsnotify_group *group)
 {
+	fsnotify_connp_t *connp = fsnotify_object_connp(obj, obj_type);
 	struct fsnotify_mark_connector *conn;
 	struct fsnotify_mark *mark;
 
+	if (!connp)
+		return NULL;
+
 	conn = fsnotify_grab_connector(connp);
 	if (!conn)
 		return NULL;

fs/super.c

@@ -274,6 +274,7 @@ static void destroy_super_work(struct work_struct *work)
 {
 	struct super_block *s = container_of(work, struct super_block,
 							destroy_work);
+	fsnotify_sb_free(s);
 	security_sb_free(s);
 	put_user_ns(s->s_user_ns);
 	kfree(s->s_subtype);

include/linux/fs.h

@@ -73,6 +73,8 @@ struct fscrypt_inode_info;
 struct fscrypt_operations;
 struct fsverity_info;
 struct fsverity_operations;
+struct fsnotify_mark_connector;
+struct fsnotify_sb_info;
 struct fs_context;
 struct fs_parameter_spec;
 struct fileattr;
@@ -618,8 +620,6 @@ is_uncached_acl(struct posix_acl *acl)
 #define IOP_XATTR	0x0008
 #define IOP_DEFAULT_READLINK	0x0010
 
-struct fsnotify_mark_connector;
-
 /*
  * Keep mostly read-only and often accessed (especially for
  * the RCU path lookup and 'stat' data) fields at the beginning
@@ -1248,7 +1248,7 @@ struct super_block {
 
 	/*
 	 * Keep s_fs_info, s_time_gran, s_fsnotify_mask, and
-	 * s_fsnotify_marks together for cache efficiency. They are frequently
+	 * s_fsnotify_info together for cache efficiency. They are frequently
 	 * accessed and rarely modified.
 	 */
 	void			*s_fs_info;	/* Filesystem private info */
@@ -1260,7 +1260,7 @@ struct super_block {
 	time64_t		   s_time_max;
 #ifdef CONFIG_FSNOTIFY
 	__u32			s_fsnotify_mask;
-	struct fsnotify_mark_connector __rcu	*s_fsnotify_marks;
+	struct fsnotify_sb_info	*s_fsnotify_info;
 #endif
 
 	/*
@@ -1301,12 +1301,6 @@ struct super_block {
 	/* Number of inodes with nlink == 0 but still referenced */
 	atomic_long_t s_remove_count;
 
-	/*
-	 * Number of inode/mount/sb objects that are being watched, note that
-	 * inodes objects are currently double-accounted.
-	 */
-	atomic_long_t s_fsnotify_connectors;
-
 	/* Read-only state of the superblock is being changed */
 	int s_readonly_remount;
 

include/linux/fsnotify.h

@@ -17,10 +17,23 @@
 #include <linux/slab.h>
 #include <linux/bug.h>
 
+/* Are there any inode/mount/sb objects watched with priority prio or above? */
+static inline bool fsnotify_sb_has_priority_watchers(struct super_block *sb,
+						     int prio)
+{
+	struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
+
+	/* Were any marks ever added to any object on this sb? */
+	if (!sbinfo)
+		return false;
+
+	return atomic_long_read(&sbinfo->watched_objects[prio]);
+}
+
 /* Are there any inode/mount/sb objects that are being watched at all? */
 static inline bool fsnotify_sb_has_watchers(struct super_block *sb)
 {
-	return atomic_long_read(&sb->s_fsnotify_connectors);
+	return fsnotify_sb_has_priority_watchers(sb, 0);
 }
 
 /*
@@ -103,6 +116,12 @@ static inline int fsnotify_file(struct file *file, __u32 mask)
 		return 0;
 
 	path = &file->f_path;
+	/* Permission events require group prio >= FSNOTIFY_PRIO_CONTENT */
+	if (mask & ALL_FSNOTIFY_PERM_EVENTS &&
+	    !fsnotify_sb_has_priority_watchers(path->dentry->d_sb,
+					       FSNOTIFY_PRIO_CONTENT))
+		return 0;
+
 	return fsnotify_parent(path->dentry, mask, path, FSNOTIFY_EVENT_PATH);
 }
 

include/linux/fsnotify_backend.h

@@ -176,6 +176,17 @@ struct fsnotify_event {
 	struct list_head list;
 };
 
+/*
+ * fsnotify group priorities.
+ * Events are sent in order from highest priority to lowest priority.
+ */
+enum fsnotify_group_prio {
+	FSNOTIFY_PRIO_NORMAL = 0,	/* normal notifiers, no permissions */
+	FSNOTIFY_PRIO_CONTENT,		/* fanotify permission events */
+	FSNOTIFY_PRIO_PRE_CONTENT,	/* fanotify pre-content events */
+	__FSNOTIFY_PRIO_NUM
+};
+
 /*
  * A group is a "thing" that wants to receive notification about filesystem
  * events.  The mask holds the subset of event types this group cares about.
@@ -201,14 +212,7 @@ struct fsnotify_group {
 	wait_queue_head_t notification_waitq;	/* read() on the notification file blocks on this waitq */
 	unsigned int q_len;			/* events on the queue */
 	unsigned int max_events;		/* maximum events allowed on the list */
-	/*
-	 * Valid fsnotify group priorities.  Events are send in order from highest
-	 * priority to lowest priority.  We default to the lowest priority.
-	 */
-	#define FS_PRIO_0	0 /* normal notifiers, no permissions */
-	#define FS_PRIO_1	1 /* fanotify content based access control */
-	#define FS_PRIO_2	2 /* fanotify pre-content access */
-	unsigned int priority;
+	enum fsnotify_group_prio priority;	/* priority for sending events */
 	bool shutdown;		/* group is being shut down, don't queue more events */
 
 #define FSNOTIFY_GROUP_USER	0x01 /* user allocated group */
@@ -456,13 +460,6 @@ FSNOTIFY_ITER_FUNCS(sb, SB)
 	     type < FSNOTIFY_ITER_TYPE_COUNT; \
 	     type++)
 
-/*
- * fsnotify_connp_t is what we embed in objects which connector can be attached
- * to. fsnotify_connp_t * is how we refer from connector back to object.
- */
-struct fsnotify_mark_connector;
-typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t;
-
 /*
  * Inode/vfsmount/sb point to this structure which tracks all marks attached to
  * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this
@@ -471,18 +468,51 @@ typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t;
  */
 struct fsnotify_mark_connector {
 	spinlock_t lock;
-	unsigned short type;	/* Type of object [lock] */
+	unsigned char type;	/* Type of object [lock] */
+	unsigned char prio;	/* Highest priority group */
+#define FSNOTIFY_CONN_FLAG_IS_WATCHED	0x01
 #define FSNOTIFY_CONN_FLAG_HAS_IREF	0x02
 	unsigned short flags;	/* flags [lock] */
 	union {
 		/* Object pointer [lock] */
-		fsnotify_connp_t *obj;
+		void *obj;
 		/* Used listing heads to free after srcu period expires */
 		struct fsnotify_mark_connector *destroy_next;
 	};
 	struct hlist_head list;
 };
 
+/*
+ * Container for per-sb fsnotify state (sb marks and more).
+ * Attached lazily on first marked object on the sb and freed when killing sb.
+ */
+struct fsnotify_sb_info {
+	struct fsnotify_mark_connector __rcu *sb_marks;
+	/*
+	 * Number of inode/mount/sb objects that are being watched in this sb.
+	 * Note that inodes objects are currently double-accounted.
+	 *
+	 * The value in watched_objects[prio] is the number of objects that are
+	 * watched by groups of priority >= prio, so watched_objects[0] is the
+	 * total number of watched objects in this sb.
+	 */
+	atomic_long_t watched_objects[__FSNOTIFY_PRIO_NUM];
+};
+
+static inline struct fsnotify_sb_info *fsnotify_sb_info(struct super_block *sb)
+{
+#ifdef CONFIG_FSNOTIFY
+	return READ_ONCE(sb->s_fsnotify_info);
+#else
+	return NULL;
+#endif
+}
+
+static inline atomic_long_t *fsnotify_sb_watched_objects(struct super_block *sb)
+{
+	return &fsnotify_sb_info(sb)->watched_objects[0];
+}
+
 /*
  * A mark is simply an object attached to an in core inode which allows an
  * fsnotify listener to indicate they are either no longer interested in events
@@ -546,6 +576,7 @@ extern int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data
 extern void __fsnotify_inode_delete(struct inode *inode);
 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
 extern void fsnotify_sb_delete(struct super_block *sb);
+extern void fsnotify_sb_free(struct super_block *sb);
 extern u32 fsnotify_get_cookie(void);
 
 static inline __u32 fsnotify_parent_needed_mask(__u32 mask)
@@ -758,30 +789,35 @@ extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
 extern void fsnotify_init_mark(struct fsnotify_mark *mark,
 			       struct fsnotify_group *group);
 /* Find mark belonging to given group in the list of marks */
-extern struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
-						struct fsnotify_group *group);
+struct fsnotify_mark *fsnotify_find_mark(void *obj, unsigned int obj_type,
+					 struct fsnotify_group *group);
 /* attach the mark to the object */
-extern int fsnotify_add_mark(struct fsnotify_mark *mark,
-			     fsnotify_connp_t *connp, unsigned int obj_type,
-			     int add_flags);
-extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
-				    fsnotify_connp_t *connp,
-				    unsigned int obj_type, int add_flags);
+int fsnotify_add_mark(struct fsnotify_mark *mark, void *obj,
+		      unsigned int obj_type, int add_flags);
+int fsnotify_add_mark_locked(struct fsnotify_mark *mark, void *obj,
+			     unsigned int obj_type, int add_flags);
 
 /* attach the mark to the inode */
 static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
 					  struct inode *inode,
 					  int add_flags)
 {
-	return fsnotify_add_mark(mark, &inode->i_fsnotify_marks,
-				 FSNOTIFY_OBJ_TYPE_INODE, add_flags);
+	return fsnotify_add_mark(mark, inode, FSNOTIFY_OBJ_TYPE_INODE,
+				 add_flags);
 }
 static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark,
 						 struct inode *inode,
 						 int add_flags)
 {
-	return fsnotify_add_mark_locked(mark, &inode->i_fsnotify_marks,
-					FSNOTIFY_OBJ_TYPE_INODE, add_flags);
+	return fsnotify_add_mark_locked(mark, inode, FSNOTIFY_OBJ_TYPE_INODE,
+					add_flags);
+}
+
+static inline struct fsnotify_mark *fsnotify_find_inode_mark(
+						struct inode *inode,
+						struct fsnotify_group *group)
+{
+	return fsnotify_find_mark(inode, FSNOTIFY_OBJ_TYPE_INODE, group);
 }
 
 /* given a group and a mark, flag mark to be freed when all references are dropped */
@@ -845,6 +881,9 @@ static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
 static inline void fsnotify_sb_delete(struct super_block *sb)
 {}
 
+static inline void fsnotify_sb_free(struct super_block *sb)
+{}
+
 static inline void fsnotify_update_flags(struct dentry *dentry)
 {}
 

kernel/audit_tree.c

@@ -463,7 +463,7 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree)
 	int n;
 
 	fsnotify_group_lock(audit_tree_group);
-	mark = fsnotify_find_mark(&inode->i_fsnotify_marks, audit_tree_group);
+	mark = fsnotify_find_inode_mark(inode, audit_tree_group);
 	if (!mark)
 		return create_chunk(inode, tree);
 

kernel/audit_watch.c

@@ -90,7 +90,7 @@ static inline struct audit_parent *audit_find_parent(struct inode *inode)
 	struct audit_parent *parent = NULL;
 	struct fsnotify_mark *entry;
 
-	entry = fsnotify_find_mark(&inode->i_fsnotify_marks, audit_watch_group);
+	entry = fsnotify_find_inode_mark(inode, audit_watch_group);
 	if (entry)
 		parent = container_of(entry, struct audit_parent, mark);