The Xen hypervisor is a virtualisation system: it allows you to run multiple virtual ("guest") machines concurrently on a single physical ("host") machine. It supports both unmodified guest operating systems on hosts with hardware virtualisation support, and paravirtualised guest operating systems that communicate more efficiently with the host system.

When Xen is enabled, the host system boots directly into the Xen hypervisor, which then starts and manages the guests. Virtual machines running on Xen are called domains. There always is a special, privileged domain called Domain 0 (Dom0), which is responsible for managing and accessing the host hardware (as the Xen kernel itself has no disk or network drivers). The other, unprivileged domains are called DomU. NixOS can run both in Dom0 and DomU.

edit Enabling Xen

To enable the Xen hypervisor in a NixOS system, just enable the following option in configuration.nix:

virtualisation.xen.enable = true;

Then run nixos-rebuild boot and reboot. The system should now boot Xen, with your NixOS system running as the Dom0. You may also want to set the option virtualisation.xen.domain0MemorySize to limit the amount of memory reserved for Dom0, e.g.

virtualisation.xen.domain0MemorySize = 512;

edit Creating NixOS guests

You can create a paravirtualised NixOS guest (DomU) from Dom0 as follows. First, create a sufficiently large disk image to hold the root filesystem:

$ dd if=/dev/zero of=/vmdisks/mrkitty-root.img bs=1M count=4096
$ mkfs.ext3 -F -L nixos /vmdisks/mrkitty-root.img

(For sparse allocation, do dd if=/dev/zero of=/vmdisks/mrkitty-root.img bs=1M seek=4096 count=0.)

Warning: because we're going to use pv-grub to boot the domain, the filesystem that holds /boot must be one that is supported by pv-grub. This includes ext2 and ext3, but excludes ext4. Thus, if you want to use ext4 as the root filesystem, you should set up a separate /boot disk.

Now we install NixOS on this filesystem in the usual manner (see the NixOS manual for details):

$ mount -o loop /vmdisks/mrkitty-root.img /mnt
$ mkdir -p /mnt/etc/nixos
$ nano /mnt/etc/nixos/configuration.nix

Here is a minimal configuration.nix:

{ config, pkgs, modulesPath, ... }:

{
  require = [ "${modulesPath}/virtualisation/xen-domU.nix" ];

  fileSystems = 
    [ { mountPoint = "/";
        label = "nixos";
      }
    ];
}

Note that you need to include the module xen-domU.nix, which defines the necessary configuration for running as a DomU system on Xen. For instance, it enables a mingetty on the Xen console (/dev/hvc0) and disables NTP.

You can now perform the installation:

$ nixos-install

If it succeeds:

$ umount /mnt

To actually create the domain, we need to define a Xen domain configuration file. It's customary to place these in /etc/xen or /etc/xen/auto (if you want it to be booted automatically when Dom0 starts), e.g., /etc/xen/auto/mrkitty:

from xen.util.path import *
memory = 512
kernel = XENFIRMWAREDIR + '/pv-grub-x86_64.gz'
extra = '(hd0)/boot/grub/menu.lst'
disk = [ 'file:/vmdisks/mrkitty-root.img,xvda1,w' ]
vif = [ '' ]

Now we can start the domain:

$ xm create -c auto/mrkitty

You should see GRUB appear, which, after a few seconds, will boot NixOS. Eventually you should get a login prompt. (You can exit the Xen console by pressing Ctrl-]. You can reconnect by running xm console domain.)

edit Managing domains

The xendomains Upstart job automatically starts all domains defined in /etc/xen/auto at boot time. It also saves and restores all running domains when the system is rebooted. You can manually save and restore running domains by doing

$ stop xendomains

and

$ start xendomains

The state of the domains is saved in /var/lib/xen/save/.

• About us