Felix Zielcke writes ("Re: Bug#532896: debian-installer: add support for using
whole disk device instead of partitioning"):
> Am Freitag, den 12.06.2009, 17:17 +0200 schrieb Sascha Silbe:
> > Please add support for using whole disk devices for file systems instead
> > of having to partition them. This is especially useful on virtual
> > machines where the underlying disk device can be resized easily.
I just ran into this. I am trying to install a VM under Xen. I have
allocated separate LVM volumes in the host for the guest's various
"partitions" but the d-i partitioner won't let me use them. It wants
to put a partition table on each one.
That's undesirable because then, if I want to resize the volume later,
I have to wrestle a resized partition table.
Whereas, using the whole virtual block device would be fine.
> grub would then use blocklists for the stage2 or core.img like LILO for
> the installed kernels.
In my configuration I'm not using grub at all. The VM host directly
reads the kernel from the guest filesystem using a program called
"pygrub".
Another common configuration is to have the host provide a copy of
grub, incliuding the whole of the stage2.
In both these cases, there is no need to use any kind of "embedding"
technique.
> And second, you can't rely that the filesystem even leaves enough space
> in the beginning for some bootcode.
> For example XFS stores it's superblock right at the beginning of it.
> The first 4 bytes of an XFS filesystem contain the magic XFSB.
No, no bootcode is needed. That will work just fine.
I have several existing Xen VMs which are set up this way. It is much
better than using multiple "partitions" in a single block device.
Partition tables only make sense on a volume that (a) needs to be
booted from directly without assitance from an outside hypervisor and
(b) can't be resized (ie, baremetal or simulations thereof).
For scenarios where bootcode is wanted, I think possibly using a
single virtual volume as a physical volume for LVM within the guest,
would probably work, but I'm no bootloader expert so I'm not sure.
> If that gets implemented there should be a big warning.
So at the very least if the underlying block device is a Xen virtual
block device, this is very normal and readily supported. It does not
deserve a warniong.
For now I guess I will let d-i have its "partition" and then shut down
the guest and reoeganise the storage from the outside shortly after
the installation, which I can do by copying. That's easy to do now
before the volumes become too large.
Ian.
--
Ian Jackson <[email protected]> These opinions are my own.
Pronouns: they/he. If I emailed you from @fyvzl.net or @evade.org.uk,
that is a private address which bypasses my fierce spamfilter.