On Mon 22 Jun 2020 05:06:53 PM CEST, Max Reitz wrote: >>> No, this is wrong. This still wouldn’t fix the problem of having a >>> device file as the external data file, when it already has non-zero >>> data during creation. (Reading the qcow2 file would return zeroes, >>> but reading the device would not.) >> >> But you wouldn't fix that preallocating the metadata either, you >> would need to fill the device with zeroes. > > What it fixes is that reading the qcow2 image and the raw device > returns the same data. > > Initially, I also thought that we should initialize raw data files to > be zero during creation, but Eric changed my mind: > > https://lists.nongnu.org/archive/html/qemu-block/2020-04/msg00223.html
Then in that case you would indeed need to preallocate all the metadata, both on creation and on resize. Then if you read the image with an older version of QEMU you would get the exact same contents in both cases. You would still have problems with images created with raw data files but without preallocation. Coincidentally, data_file and data_file_raw were both introduced in QEMU 4.0 so in practice there's no official QEMU release that can read an image with an external data file but does not support data_file_raw. But in theory it could happen of course. Berto
