[Soc-coordination] Status Report of ZFS on Linux Integration, Week 4

Aron Xu happyaron.xu at gmail.com
Fri Jul 19 18:36:56 UTC 2013

Hi everyone,

This is the fourth report of ZFS on Linux Integration GSoC project.
This week, I have worked on debian-installer support of ZFS on Linux,
and has got some progress so that very very interested person may want
to have a "preview" (not exactly working for "try" yet).

As I assumed that support ZFS on Linux could be mostly involved in
modifying d-i, but after a deeper look the problem appears in partman
side, necessarily partman-base and of course partman-zfs. The troubles
are first caused by the detailed differences of the modules behavior
on loading, presenting block devices, etc.

I have made some experimental patches that would let it work for some
parts, and with my generated netboot mini.iso, d-i can properly create
zpool, and zvols, also it can made to configure how a zvol is going to
be used (similar to choose a traditional partition for /). It cannot
save the mount point information in the last step correctly right now,
which should be not too difficult to deal with, but requires some time
for debugging.

The patches for partman-* aren't well tested, and even has several
known issues to be worked on. The quality is still a bit far from
being able to get merged, so I'm not submitting any of them for now.
Here are the git repositories where I put the patches:

udeb packages I've generated using last week's work, and they are used
in my mini.iso:

Generated netboot mini.iso, wheezy based, amd64 only, binary modules included:

The image can be tested easily using qemu-kvm on amd64 host machine,
make sure you have connectivity to a debian mirror:
# qemu-img create -f raw hda.img 20G   // Create an raw disk image for rootfs
# kvm -cdrom zol-mini-amd64-20130719.iso -hda hda.img  // Boot the VM from ISO

So next step I'll continue to work on improving the quality of
packaging and patches, so that they can be pushed to
experimental/unstable soon. As a file system is something really
complex, I do not expect current stuff can work well for most common
environments without lots of more work.

Aron Xu

More information about the Soc-coordination mailing list