ostree/debian/ostree-boot-examples

Testing ostree-boot on an existing system

Start from an amd64 Debian system (9 or newer, as long as a backported ostree-boot package is available) - it will be switched to a Debian unstable (sid) OSTree-based installation as part of following these instructions. A VM is obviously most convenient, but bare metal should work equally.

The installation needs to satisfy the following requirements:

• It must have a separate /boot partition (this is a general libostree limitation, see ostree issue 1452).
• The root partition must not be encrypted.
• It must use GRUB and BIOS booting. Other bootloaders or EFI may require different bootloader setup steps; see old Fedora instructions

We need the bootloader integration files on the non-OSTree system from which we are switching, as well as in the OSTree-based installation, so the bootloader gets configured properly when we "ostree admin deploy":

# apt-get update
# apt-get install ostree ostree-boot multistrap

Create the ostree system repository and a stateroot:

# ostree admin init-fs /
# ostree admin os-init debian

Copy modified-deb-ostree-builder, ostree-1.conf and ostree-2.conf from debian/ostree-boot-examples/ to the test machine, and run the builder script:

# chmod +x ./deb-ostree-builder
# ./deb-ostree-builder ./ostree-1.conf sid-1 /ostree/repo
# ./deb-ostree-builder ./ostree-2.conf sid-2 /ostree/repo

If ostree-boot is not available in the target suite in the Debian archive yet, then you will need to edit ostree-1.conf and ostree-2.conf to remove ostree-boot from the bootstrap, and instead put the ostree-boot, ostree and libostree-1-1 packages in /root/extra-packages, and use modified-deb-ostree-builder instead of deb-ostree-builder. This is a temporary hack to solve the chicken-and-egg situation of not adding ostree-boot to the Debian archive until it is testable, but not being able to test it until it is in the archive.

Then we deploy the first of those commits:

# ostree admin deploy --karg-proc-cmdline --os=debian os/debian/amd64/sid-1
# deploy=$(find /ostree/deploy/debian/deploy/* -maxdepth 0 -type d)

Next, we set the root password and copy a few essential configuration files into the initial deployment:

# chroot $deploy passwd root
# : > $deploy/etc/machine-id
# for f in etc/fstab etc/default/grub; do cp /$f $deploy/$f; done

Finally, we set up the bootloader by pointing GRUB at the configuration file managed by ostree. Alternatively, you can run update-grub by hand after every new ostree deployment. This step may be different or unnecessary for other bootloaders.

# mv /boot/grub/grub.cfg /boot/grub/grub.cfg.backup
# ln -s ../loader/grub.cfg /boot/grub/grub.cfg

Now reboot. Make sure to select the new ostree entry in the bootloader. Log in as root using the password you selected before. The system is rather unconfigured (network access can be set up in /etc/network/interfaces). This is the commit built from the ostree-1.conf file, so it doesn't have the hello package:

# hello
-bash: hello: command not found

Now we can deploy the second commit from inside the ostree system:

# ostree admin deploy --karg-proc-cmdline --os=debian os/debian/amd64/sid-2

Reboot again. Note that your new deployment will again be the first ostree entry in the menu, labelled ostree:0. Your old deployment has been moved down the menu to ostree:1. The root password and the other configuration was copied from the first deployment, so you can log in as before.

Now try running the hello command again:

# hello
Hello, world!

This shows we've now booted the second commit built from ostree-2.conf, which includes the hello package.