sha3sum gives different hash between file on same partition or on a mount point

[maandree]

sha3sum gives different hash between file on same partition or on a mount point

[maandree]

Are you using the latest version of libkeccak, does rolling back to the previous version of libkeccak change anything? How big is the file, and how big is the I/O block size?

[Bonnietwin]

Arch Linux is a rolling distro so is very up to date usually.
sha3sum is 1.2.1
libkeccak is 1.3

I rolled back to 1.2.2 of libkeccak and the problem is no longer there. The sha3sum is correct for the file wherever it is located.

The sqlite tar file is 2.9MiB and the bacula tar file is 4.0MiB
I will also do a test with a much smaller file and see if there is a difference and find out what the I/O block sizes are for the files evaluated.
Rolling back to the previous version of libkeccak is a working solution for me until the issue is solved.

[Bonnietwin]

I am not sure if I have found what you were asking for with the I/O Block Size but after googling, I have run the stat command on the file.
On the same partition as I was working on that gave the correct sha3 hash the IO Block number was 4096
On the nfs mounted fileserver (running on a raid array) the number was 262144

[Bonnietwin]

I forgot to mention in the last post that the partition I was working on that gave the correct sha3 hash is also a raid array, so both locations are raid array based.

[maandree]

Great, then I'm rather sure where the problem is located. I'll get back to you when I've looked at it.

[Bonnietwin]

I installed the 1.3 version of libkeccak again and ran the test on different file sizes.

The same different sha3sum occurred with files of 1016.5KiB and 158.5KiB but with a file of 2.7KiB the sha3sum was the same for the file in both locations.

[maandree]

I'm certain it is fixed now, would you mind checking with the lasted commit to libkeccak?

[Bonnietwin]

Sorry for delay but I was a bit busy today.

Just patched the source and installed it using the Arch Linux Build System and tested it out on the sqlite package and now I get the same sha3-256sum as they have on the sqlite website.

Thanks very much. Marking this issue as closed. Look forward to it ending up in the Arch Linux system but the patched version works so no problems any more for me.