[PATCH] lock exclusively in collect_mounts() and drop_collected_mounts() (1a60a280) · Commits · 戴 / test

Taking namespace_sem shared there isn't worth the trouble, especially with vfsmount ID allocation about to be added. That way we know that umount_tree(), copy_tree() and clone_mnt() are _always_ serialized by namespace_sem. umount_tree() still needs vfsmount_lock (it manipulates hash chains, among other things), but that's a separate story. Signed-off-by:

Al Viro <viro@zeniv.linux.org.uk>

fs/namespace.c

+4 −4

Original line number	Diff line number	Diff line
		@@ -1091,20 +1091,20 @@ Enomem:
		struct vfsmount collect_mounts(struct vfsmount mnt, struct dentry *dentry)
		{
		struct vfsmount *tree;
		down_read(&namespace_sem);
		down_write(&namespace_sem);
		tree = copy_tree(mnt, dentry, CL_COPY_ALL \| CL_PRIVATE);
		up_read(&namespace_sem);
		up_write(&namespace_sem);
		return tree;
		}

		void drop_collected_mounts(struct vfsmount *mnt)
		{
		LIST_HEAD(umount_list);
		down_read(&namespace_sem);
		down_write(&namespace_sem);
		spin_lock(&vfsmount_lock);
		umount_tree(mnt, 0, &umount_list);
		spin_unlock(&vfsmount_lock);
		up_read(&namespace_sem);
		up_write(&namespace_sem);
		release_mounts(&umount_list);
		}

Admin message