fsnotify: add mount notification infrastructure

This is just the plumbing between the event source (fs/namespace.c) and the
event consumer (fanotify). In itself it does nothing.

Signed-off-by: Miklos Szeredi <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Christian Brauner <[e-mail]>

fs/mount.h

···
 		struct rcu_head		mnt_ns_rcu;
 	};
 	u64 event;
+#ifdef CONFIG_FSNOTIFY
+	__u32			n_fsnotify_mask;
+	struct fsnotify_mark_connector __rcu *n_fsnotify_marks;
+#endif
 	unsigned int		nr_mounts; /* # of mounts in the namespace */
 	unsigned int		pending_mounts;
 	struct rb_node		mnt_ns_tree_node; /* node in the mnt_ns_tree */

fs/notify/fsnotify.c

···
 	fsnotify_clear_marks_by_mount(mnt);
 }
 
+void __fsnotify_mntns_delete(struct mnt_namespace *mntns)
+{
+	fsnotify_clear_marks_by_mntns(mntns);
+}
+
 /**
  * fsnotify_unmount_inodes - an sb is unmounting.  handle any watched inodes.
  * @sb: superblock being unmounted.
···
 				     file_name, cookie, iter_info);
 }
 
-static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
+static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector *const *connp)
 {
 	struct fsnotify_mark_connector *conn;
 	struct hlist_node *node = NULL;
···
 {
 	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);
+	const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type);
+	struct fsnotify_sb_info *sbinfo = sb ? fsnotify_sb_info(sb) : NULL;
 	struct fsnotify_iter_info iter_info = {};
 	struct mount *mnt = NULL;
 	struct inode *inode2 = NULL;
 	struct dentry *moved;
 	int inode2_type;
 	int ret = 0;
-	__u32 test_mask, marks_mask;
+	__u32 test_mask, marks_mask = 0;
 
 	if (path)
 		mnt = real_mount(path->mnt);
···
 	if ((!sbinfo || !sbinfo->sb_marks) &&
 	    (!mnt || !mnt->mnt_fsnotify_marks) &&
 	    (!inode || !inode->i_fsnotify_marks) &&
-	    (!inode2 || !inode2->i_fsnotify_marks))
+	    (!inode2 || !inode2->i_fsnotify_marks) &&
+	    (!mnt_data || !mnt_data->ns->n_fsnotify_marks))
 		return 0;
 
-	marks_mask = READ_ONCE(sb->s_fsnotify_mask);
+	if (sb)
+		marks_mask |= READ_ONCE(sb->s_fsnotify_mask);
 	if (mnt)
 		marks_mask |= READ_ONCE(mnt->mnt_fsnotify_mask);
 	if (inode)
 		marks_mask |= READ_ONCE(inode->i_fsnotify_mask);
 	if (inode2)
 		marks_mask |= READ_ONCE(inode2->i_fsnotify_mask);
-
+	if (mnt_data)
+		marks_mask |= READ_ONCE(mnt_data->ns->n_fsnotify_mask);
 
 	/*
 	 * If this is a modify event we may need to clear some ignore masks.
···
 	if (inode2) {
 		iter_info.marks[inode2_type] =
 			fsnotify_first_mark(&inode2->i_fsnotify_marks);
+	}
+	if (mnt_data) {
+		iter_info.marks[FSNOTIFY_ITER_TYPE_MNTNS] =
+			fsnotify_first_mark(&mnt_data->ns->n_fsnotify_marks);
 	}
 
 	/*
···
 }
 #endif
 
+void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt)
+{
+	struct fsnotify_mnt data = {
+		.ns = ns,
+		.mnt_id = real_mount(mnt)->mnt_id_unique,
+	};
+
+	if (WARN_ON_ONCE(!ns))
+		return;
+
+	/*
+	 * This is an optimization as well as making sure fsnotify_init() has
+	 * been called.
+	 */
+	if (!ns->n_fsnotify_marks)
+		return;
+
+	fsnotify(mask, &data, FSNOTIFY_EVENT_MNT, NULL, NULL, NULL, 0);
+}
+
 static __init int fsnotify_init(void)
 {
 	int ret;
 
-	BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 24);
+	BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 26);
 
 	ret = init_srcu_struct(&fsnotify_mark_srcu);
 	if (ret)

fs/notify/fsnotify.h

···
 	return conn->obj;
 }
 
+static inline struct mnt_namespace *fsnotify_conn_mntns(
+				struct fsnotify_mark_connector *conn)
+{
+	return conn->obj;
+}
+
 static inline struct super_block *fsnotify_object_sb(void *obj,
 			enum fsnotify_obj_type obj_type)
 {
···
 static inline void fsnotify_clear_marks_by_sb(struct super_block *sb)
 {
 	fsnotify_destroy_marks(fsnotify_sb_marks(sb));
+}
+
+static inline void fsnotify_clear_marks_by_mntns(struct mnt_namespace *mntns)
+{
+	fsnotify_destroy_marks(&mntns->n_fsnotify_marks);
 }
 
 /*

fs/notify/mark.c

···
 		return &real_mount(obj)->mnt_fsnotify_marks;
 	case FSNOTIFY_OBJ_TYPE_SB:
 		return fsnotify_sb_marks(obj);
+	case FSNOTIFY_OBJ_TYPE_MNTNS:
+		return &((struct mnt_namespace *)obj)->n_fsnotify_marks;
 	default:
 		return NULL;
 	}
···
 		return &fsnotify_conn_mount(conn)->mnt_fsnotify_mask;
 	else if (conn->type == FSNOTIFY_OBJ_TYPE_SB)
 		return &fsnotify_conn_sb(conn)->s_fsnotify_mask;
+	else if (conn->type == FSNOTIFY_OBJ_TYPE_MNTNS)
+		return &fsnotify_conn_mntns(conn)->n_fsnotify_mask;
 	return NULL;
 }
 
···
 		fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
 	} else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
 		fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
+	} else if (conn->type == FSNOTIFY_OBJ_TYPE_MNTNS) {
+		fsnotify_conn_mntns(conn)->n_fsnotify_mask = 0;
 	}
 
 	rcu_assign_pointer(*connp, NULL);
 	conn->obj = NULL;
 	conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
-	fsnotify_update_sb_watchers(sb, conn);
+	if (sb)
+		fsnotify_update_sb_watchers(sb, conn);
 
 	return inode;
 }
···
 	 * 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)) {
+	if (sb && !fsnotify_sb_info(sb)) {
 		err = fsnotify_attach_info_to_sb(sb);
 		if (err)
 			return err;
···
 	/* 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);
+	if (sb)
+		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

include/linux/fsnotify.h

···
 	__fsnotify_vfsmount_delete(mnt);
 }
 
+static inline void fsnotify_mntns_delete(struct mnt_namespace *mntns)
+{
+	__fsnotify_mntns_delete(mntns);
+}
+
 /*
  * fsnotify_inoderemove - an inode is going away
  */
···
 
 	return fsnotify(FS_ERROR, &report, FSNOTIFY_EVENT_ERROR,
 			NULL, NULL, NULL, 0);
+}
+
+static inline void fsnotify_mnt_attach(struct mnt_namespace *ns, struct vfsmount *mnt)
+{
+	fsnotify_mnt(FS_MNT_ATTACH, ns, mnt);
+}
+
+static inline void fsnotify_mnt_detach(struct mnt_namespace *ns, struct vfsmount *mnt)
+{
+	fsnotify_mnt(FS_MNT_DETACH, ns, mnt);
+}
+
+static inline void fsnotify_mnt_move(struct mnt_namespace *ns, struct vfsmount *mnt)
+{
+	fsnotify_mnt(FS_MNT_MOVE, ns, mnt);
 }
 
 #endif	/* _LINUX_FS_NOTIFY_H */

include/linux/fsnotify_backend.h

···
 
 #define FS_PRE_ACCESS		0x00100000	/* Pre-content access hook */
 
+#define FS_MNT_ATTACH		0x01000000	/* Mount was attached */
+#define FS_MNT_DETACH		0x02000000	/* Mount was detached */
+#define FS_MNT_MOVE		(FS_MNT_ATTACH | FS_MNT_DETACH)
+
 /*
  * Set on inode mark that cares about things that happen to its children.
  * Always set for dnotify and inotify.
···
  * when a directory entry inside a child subdir changes.
  */
 #define ALL_FSNOTIFY_DIRENT_EVENTS (FS_CREATE | FS_DELETE | FS_MOVE | FS_RENAME)
+
+/* Mount namespace events */
+#define FSNOTIFY_MNT_EVENTS (FS_MNT_ATTACH | FS_MNT_DETACH)
 
 /* Content events can be used to inspect file content */
 #define FSNOTIFY_CONTENT_PERM_EVENTS (FS_OPEN_PERM | FS_OPEN_EXEC_PERM | \
···
 
 /* Events that can be reported to backends */
 #define ALL_FSNOTIFY_EVENTS (ALL_FSNOTIFY_DIRENT_EVENTS | \
+			     FSNOTIFY_MNT_EVENTS | \
 			     FS_EVENTS_POSS_ON_CHILD | \
 			     FS_DELETE_SELF | FS_MOVE_SELF | \
 			     FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
···
 	FSNOTIFY_EVENT_PATH,
 	FSNOTIFY_EVENT_INODE,
 	FSNOTIFY_EVENT_DENTRY,
+	FSNOTIFY_EVENT_MNT,
 	FSNOTIFY_EVENT_ERROR,
 };
 
···
 {
 	return range->path;
 }
+
+struct fsnotify_mnt {
+	const struct mnt_namespace *ns;
+	u64 mnt_id;
+};
 
 static inline struct inode *fsnotify_data_inode(const void *data, int data_type)
 {
···
 	}
 }
 
+static inline const struct fsnotify_mnt *fsnotify_data_mnt(const void *data,
+							   int data_type)
+{
+	switch (data_type) {
+	case FSNOTIFY_EVENT_MNT:
+		return data;
+	default:
+		return NULL;
+	}
+}
+
+static inline u64 fsnotify_data_mnt_id(const void *data, int data_type)
+{
+	const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type);
+
+	return mnt_data ? mnt_data->mnt_id : 0;
+}
+
 static inline struct fs_error_report *fsnotify_data_error_report(
 							const void *data,
 							int data_type)
···
 	FSNOTIFY_ITER_TYPE_SB,
 	FSNOTIFY_ITER_TYPE_PARENT,
 	FSNOTIFY_ITER_TYPE_INODE2,
+	FSNOTIFY_ITER_TYPE_MNTNS,
 	FSNOTIFY_ITER_TYPE_COUNT
 };
 
···
 	FSNOTIFY_OBJ_TYPE_INODE,
 	FSNOTIFY_OBJ_TYPE_VFSMOUNT,
 	FSNOTIFY_OBJ_TYPE_SB,
+	FSNOTIFY_OBJ_TYPE_MNTNS,
 	FSNOTIFY_OBJ_TYPE_COUNT,
 	FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT
 };
···
 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_mntns_delete(struct mnt_namespace *mntns);
 extern void fsnotify_sb_free(struct super_block *sb);
 extern u32 fsnotify_get_cookie(void);
+extern void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt);
 
 static inline __u32 fsnotify_parent_needed_mask(__u32 mask)
 {
···
 static inline void fsnotify_sb_delete(struct super_block *sb)
 {}
 
+static inline void __fsnotify_mntns_delete(struct mnt_namespace *mntns)
+{}
+
 static inline void fsnotify_sb_free(struct super_block *sb)
 {}
 
···
 }
 
 static inline void fsnotify_unmount_inodes(struct super_block *sb)
+{}
+
+static inline void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt)
 {}
 
 #endif	/* CONFIG_FSNOTIFY */