zokier
commented
6 months ago If I'm understanding the situation correctly, the problem you are facing might be mke2fs changing atimes of files when it reads them from source directory, and when you rebuild the FS the timestamps differ? The simple workaround might be to run touch to set timestamps in the source directory to fixed values before running mke2fs. With this in mind I'm not sure if this is something that should be addressed in e2fsprogs, considering that the FS build process as I understand it is already reproducible as long as the input stays fixed.
That all being said, I suspect that #118 will be the best way forwards to get reproducible results in the long term.
tytso
There's a small issue that has to be fixed if you'd like to be able to create a reproducible ext4 FS from the source directory. By default, https://reproducible-builds.org/docs/system-images/ suggests you:
Yes, make_ext4fs can create a reproducible FSes, but it lacks the proper hard links support (I'm about to send patches that fix this issue as well, so in the end everyone has a choice to use either mkfs.ext4 or make_ext4fs, but the latter to the best of my knowledge has been phased out ).
You can try to test these changes by first trying to create a reproducible FS from the stand alone structure 2 times in a row and compare SHA256 hashes of the result image (or block device, whatever you choose to use):
You'll notice that the result images will differ.
Then apply this tiny change and repeat, this time you can see that both images are identical.
P.S. There might be cases (say, by using different options) that could still lead to FS being not reproducible. In that case I suggest to handle them on the need-to-do basis (once the problem is reported).