The current code uses kpartx -d on the image file to remove the partition mapping. This does not work and can not work, as the same image could be mapped several times and kpartx would not know what "our" mapping is. Therefore kpartx has to be invoked with the loop-device as argument to remove the partition mappings and then losetup must be executed to get rid of the loopback device created by kpartx at the beginning.