locking in xfs_unmount_flush

Does anybody have an idea why the ilock on the root inode is held over flushing and releasing of the rt and quota inodes in xfs_unmount_flush? I've looked through all that code and can't find any reason why we'd want it.