Debian Bug - 982277 - zfs-dkms: ZFS is not validating checksums on file read

[Kyle]

per https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=982277

I was wondering if anyone else is has a moment to test and confirm this bug and share their pveversion and zfs versions.
I think it would be possible to test this by creating a pool from a file, which is likely easier for most than creating a device partition.

question to proxmox staff to improve my general pve knowledsge: are the pve zfs versions packaged/forked from debian sources? or are you packing/forking from OpenZFS directly?

I'll post my test results in the coming days.

 

[Kyle]

I've done some testing of the bug report with some confusing results, so I don't have anything ready to share for comments just yet.

I noted via the README via pve-kernel repo on GitHub that Ubuntu kernel sources are being used and zfs modules via https://github.com/zfsonlinux/. However the submodules zfsonlinux links to a 404? Is that a private repo?

I'd be interesting in learning more about how proxmox team is packaging zfs.

edit: ok so it looks like its because the main repo is https://git.proxmox.com and github is just selective mirror of some projects. So submodules/zfsonlinux links to zfsonlinux. I'll have a closer look at the log to see what I can figure out.

 

[Stoiko Ivanov]

question to proxmox staff to improve my general pve knowledsge: are the pve zfs versions packaged/forked from debian sources? or are you packing/forking from OpenZFS directly?

As you've found out by now - Proxmox packages ZFS from upstream
https://github.com/openzfs/zfs
the repository is mirrored at https://git.proxmox.com (which is also the authoritative source for our sources)

The README in the pve-kernel repo should provide the instruction for building:
https://git.proxmox.com/?p=pve-kern...e116a33ef67fc63da9a483836863da4d34deb;hb=HEAD
(check the part about submodules)

As for the debian bug - not sure I follow the argumentation there or that it breaks any kind of promise ZFS does - the error is detected after a scrub - and scrubs are there to detect problems/bitflips on disks (which is also the reason why most ZFS packages (including the ones from Proxmox) do include a monthly timer for scrubbing all pools)?

I hope this helps!

 

[Kyle]

Thanks for the details @Stoiko Ivanov, I'll have a closer look.

As for the debian bug - not sure I follow the argumentation there or that it breaks any kind of promise ZFS does

Per https://openzfs.github.io/openzfs-docs/Basic Concepts/Checksums.html I think the person reporting the bug is referring to:

(zfs) detects data corruption upon reading from media

This is a little vague, one might interpret this as "upon reading during normal file access?" or it could be "upon pool import?"
The documentation could be updated to be less ambiguous.

My impression has always been that ZFS verifies checksums as its reading data (normal file read access etc)?
Is that wrong? I've not go into the code to check yet.