Reimaging a VM from the inside with Debian

Recently I was in a situation where I wanted to spin up a virtual machine in a cloud computing service but the provided operating system selection was meager. The two options were Ubuntu 11.10 (i.e. October 2011) or RHEL 6.3 (June 2012). Worse, these were the desktop versions of these platforms, so they had X.Org installed along with a bunch of desktop applications. These images were not really intended as servers.

As always, I strongly prefer Debian. Ubuntu is derived from Debian so it isn’t far from ideal, but I didn’t want to deal with such an old version, plus all the desktop cruft. Also, I really just want to use Debian Wheezy (currently stable). It’s extremely solid and I know it through and through.

Since there’s no way for me to provide my own image, my only option is to install Debian from within the virtual machine. A significant difficulty in this is that I also have no way to provide alternate boot media. I would need to install Debian from within Ubuntu, over top of Ubuntu while it’s running. Within these restrictions this may sound like an impossible task.

Fortunately, Debian, being the universal operating system and living up to its name, has a way to do this, and do it cleanly. When I’m done there will be absolutely no trace of Ubuntu left. I could do the same from within the Red Hat system, but working from Ubuntu takes one fewer step. The magic tool for solving this problem is debootstrap. It installs a Debian base system into an arbitrary subdirectory. This is what the official Debian installer uses and it’s used to build my live CD. Ubuntu offers this as a normal package, so I don’t need to download anything special.

Creating a Pivot

So I’ve got a way to install Debian from within another Linux installation, but now how can I install Debian over top of Ubuntu? I will ultimately need to use debootstrap on the root of a fresh filesystem. I can’t wipe Ubuntu’s root partition while it’s running. I can’t even resize it since that’s where / is mounted.

Fortunately for me the person who set up this VM just went through Ubuntu’s defaults. This means there’s a single primary partition holding the entire Ubuntu installation (sda1), a second extended partition (sda2), and within the extended partition there’s a swap partition (sda5).

The swap partition is 1GB, which, while cramped, is plenty of room for a Debian base system. I have no use for an extended partition so I can just blow the whole thing away and install Debian to sda2.

ubuntu# swapoff /dev/sda5
ubuntu# fdisk /dev/sda  # (fix up partitions)
ubuntu# mkfs.ext4 /dev/sda2
ubuntu# mkdir /mnt/debian
ubuntu# mount /dev/sda2 /mnt/debian
ubuntu# debootstrap wheezy /mnt/debian <local-deb-mirror>

Now I have a proper Debian base system installed on a second partition. At this point I could configure Grub to boot into it by default rather than Ubuntu. However, as it stands so far, I would have no way to access it remotely (or at all) since I haven’t set anything up. From here I just follow the guide in appendix D and configure it from a chroot,

ubuntu# LANG=C.UTF-8 chroot /mnt/debian /bin/bash
chroot# mount -t proc proc /proc
chroot# dpkg-reconfigure tzdata
chroot# apt-get install linux-image-amd64 locales openssh-server
chroot# passwd

I just copy Ubuntu’s /etc/network/interfaces so that the new system uses the same network configuration (DHCP in this case).

ubuntu# cp /etc/network/interfaces /mnt/debian/etc/network/

A create a one-line /etc/fstab, mounting /dev/sda2 as /.

/dev/sda2  /  ext4  defaults  0  1

Finally I overwrite the Ubuntu-installed Grub. I need to set up the devices in order to actually install Grub.

chroot# apt-get install makedev
chroot# cd /dev
chroot# MAKEDEV generic
chroot# apt-get install grub-pc
chroot# update-grub

The Ubuntu system isn’t visible to the Grub installer, so the VM will now boot into my tiny Debian system by default.

ubuntu# reboot

Taking Over the Host

After about a minute I can SSH into the VM again. This time I’m in a proper Debian Wheezy system running from sda2! Now the problem is making use of that large partition that still houses Ubuntu. I could try slicing it up and mounting parts of it for Debian, but there’s something simpler I can do: repeat the same exact process on the first partition. The Debian system I just set up becomes a pivot for the real installation.

pivot# mkfs.ext4 /dev/sda1
pivot# mkdir /mnt/debian
pivot# mount /dev/sda1 /mnt/debian
pivot# debootstrap wheezy /mnt/debian
(... etc ...)

After installing Grub again, reboot. There may be some way to simply copy the pivot system directly but I’m not confident enough to trust a straight cp -r.

Final Touches

Now I’m running Debian on the large partition. The pivot can be converted back into a swap partition.

debian# mkswap /dev/sda2
debian# swapon /dev/sda2
debian# echo /dev/sda2 swap swap defaults 0 0 >> /etc/fstab

This is now as good as a fresh installation of a Debian base system (i.e. what the cloud provider should have offered in the first place). If it was possible on this cloud, this is where I’d make a snapshot for cloning future VMs. Since that’s not an option, should I need more Debian VMs in the future I’ll write a pair of shell scripts to do all this for me. An automated conversion process would probably take about 5 minutes.

I love Debian.

Have a comment on this article? Start a discussion in my public inbox by sending an email to ~skeeto/ [mailing list etiquette] , or see existing discussions.

null program

Chris Wellons (PGP)
~skeeto/ (view)