vfs: add notifications for mount attach and detach

Add notifications for attaching and detaching mounts to fs/namespace.c

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Link: https://lore.kernel.org/r/20250129165803.72138-4-mszeredi@redhat.com
Signed-off-by: Christian Brauner <brauner@kernel.org>

fs/mount.h

@@ -5,6 +5,8 @@
 #include <linux/ns_common.h>
 #include <linux/fs_pin.h>
 
+extern struct list_head notify_list;
+
 struct mnt_namespace {
 	struct ns_common	ns;
 	struct mount *	root;
@@ -80,6 +82,8 @@ struct mount {
 #ifdef CONFIG_FSNOTIFY
 	struct fsnotify_mark_connector __rcu *mnt_fsnotify_marks;
 	__u32 mnt_fsnotify_mask;
+	struct list_head to_notify;	/* need to queue notification */
+	struct mnt_namespace *prev_ns;	/* previous namespace (NULL if none) */
 #endif
 	int mnt_id;			/* mount identifier, reused */
 	u64 mnt_id_unique;		/* mount ID unique until reboot */
@@ -182,4 +186,20 @@ static inline struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
 	return container_of(ns, struct mnt_namespace, ns);
 }
 
+#ifdef CONFIG_FSNOTIFY
+static inline void mnt_notify_add(struct mount *m)
+{
+	/* Optimize the case where there are no watches */
+	if ((m->mnt_ns && m->mnt_ns->n_fsnotify_marks) ||
+	    (m->prev_ns && m->prev_ns->n_fsnotify_marks))
+		list_add_tail(&m->to_notify, &notify_list);
+	else
+		m->prev_ns = m->mnt_ns;
+}
+#else
+static inline void mnt_notify_add(struct mount *m)
+{
+}
+#endif
+
 struct mnt_namespace *mnt_ns_from_dentry(struct dentry *dentry);

fs/namespace.c

@@ -81,6 +81,9 @@ static HLIST_HEAD(unmounted);	/* protected by namespace_sem */
 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
 static DEFINE_SEQLOCK(mnt_ns_tree_lock);
 
+#ifdef CONFIG_FSNOTIFY
+LIST_HEAD(notify_list); /* protected by namespace_sem */
+#endif
 static struct rb_root mnt_ns_tree = RB_ROOT; /* protected by mnt_ns_tree_lock */
 static LIST_HEAD(mnt_ns_list); /* protected by mnt_ns_tree_lock */
 
@@ -163,6 +166,7 @@ static void mnt_ns_release(struct mnt_namespace *ns)
 {
 	/* keep alive for {list,stat}mount() */
 	if (refcount_dec_and_test(&ns->passive)) {
+		fsnotify_mntns_delete(ns);
 		put_user_ns(ns->user_ns);
 		kfree(ns);
 	}
@@ -1176,6 +1180,8 @@ static void mnt_add_to_ns(struct mnt_namespace *ns, struct mount *mnt)
 		ns->mnt_first_node = &mnt->mnt_node;
 	rb_link_node(&mnt->mnt_node, parent, link);
 	rb_insert_color(&mnt->mnt_node, &ns->mounts);
+
+	mnt_notify_add(mnt);
 }
 
 /*
@@ -1723,6 +1729,50 @@ int may_umount(struct vfsmount *mnt)
 
 EXPORT_SYMBOL(may_umount);
 
+#ifdef CONFIG_FSNOTIFY
+static void mnt_notify(struct mount *p)
+{
+	if (!p->prev_ns && p->mnt_ns) {
+		fsnotify_mnt_attach(p->mnt_ns, &p->mnt);
+	} else if (p->prev_ns && !p->mnt_ns) {
+		fsnotify_mnt_detach(p->prev_ns, &p->mnt);
+	} else if (p->prev_ns == p->mnt_ns) {
+		fsnotify_mnt_move(p->mnt_ns, &p->mnt);
+	} else {
+		fsnotify_mnt_detach(p->prev_ns, &p->mnt);
+		fsnotify_mnt_attach(p->mnt_ns, &p->mnt);
+	}
+	p->prev_ns = p->mnt_ns;
+}
+
+static void notify_mnt_list(void)
+{
+	struct mount *m, *tmp;
+	/*
+	 * Notify about mounts that were added/reparented/detached/remain
+	 * connected after unmount.
+	 */
+	list_for_each_entry_safe(m, tmp, &notify_list, to_notify) {
+		mnt_notify(m);
+		list_del_init(&m->to_notify);
+	}
+}
+
+static bool need_notify_mnt_list(void)
+{
+	return !list_empty(&notify_list);
+}
+#else
+static void notify_mnt_list(void)
+{
+}
+
+static bool need_notify_mnt_list(void)
+{
+	return false;
+}
+#endif
+
 static void namespace_unlock(void)
 {
 	struct hlist_head head;
@@ -1733,7 +1783,18 @@ static void namespace_unlock(void)
 	hlist_move_list(&unmounted, &head);
 	list_splice_init(&ex_mountpoints, &list);
 
-	up_write(&namespace_sem);
+	if (need_notify_mnt_list()) {
+		/*
+		 * No point blocking out concurrent readers while notifications
+		 * are sent. This will also allow statmount()/listmount() to run
+		 * concurrently.
+		 */
+		downgrade_write(&namespace_sem);
+		notify_mnt_list();
+		up_read(&namespace_sem);
+	} else {
+		up_write(&namespace_sem);
+	}
 
 	shrink_dentry_list(&list);
 
@@ -1846,6 +1907,19 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
 		change_mnt_propagation(p, MS_PRIVATE);
 		if (disconnect)
 			hlist_add_head(&p->mnt_umount, &unmounted);
+
+		/*
+		 * At this point p->mnt_ns is NULL, notification will be queued
+		 * only if
+		 *
+		 *  - p->prev_ns is non-NULL *and*
+		 *  - p->prev_ns->n_fsnotify_marks is non-NULL
+		 *
+		 * This will preclude queuing the mount if this is a cleanup
+		 * after a failed copy_tree() or destruction of an anonymous
+		 * namespace, etc.
+		 */
+		mnt_notify_add(p);
 	}
 }
 
@@ -2555,6 +2629,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 			dest_mp = smp;
 		unhash_mnt(source_mnt);
 		attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
+		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
 		if (source_mnt->mnt_ns) {
@@ -4476,6 +4551,8 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	list_del_init(&new_mnt->mnt_expire);
 	put_mountpoint(root_mp);
 	unlock_mount_hash();
+	mnt_notify_add(root_mnt);
+	mnt_notify_add(new_mnt);
 	chroot_fs_refs(&root, &new);
 	error = 0;
 out4:

fs/pnode.c

@@ -549,8 +549,10 @@ static void restore_mounts(struct list_head *to_restore)
 			mp = parent->mnt_mp;
 			parent = parent->mnt_parent;
 		}
-		if (parent != mnt->mnt_parent)
+		if (parent != mnt->mnt_parent) {
 			mnt_change_mountpoint(parent, mp, mnt);
+			mnt_notify_add(mnt);
+		}
 	}
 }