Systemd-nspawn

From CSCWiki
Jump to navigation Jump to search

systemd-nspawn is a simpler alternative to LXC which works well on modern versions of Debian (and, unlike LXC, it does not break very critical systemd services running in containers). For "pet" containers, we should be using systemd-nspawn; for "cattle" containers, Podman is more appropriate.

Some light reading:

Quickstart

In the example below, we will create a container called 'machine1'.

Create a directory for the rootfs:

mkdir /var/lib/machines/machine1

Or, if you are using an LVM volume, just create a symlink in /var/lib/machines to where the LV is mounted:

ln -s /vm/machine1 /var/lib/machines/machine1

Now bootstrap the rootfs:

debootstrap --variant=minbase --include=systemd,systemd-sysv,systemd-container,iproute2,inetutils-ping,ifupdown,procps,less,nano bullseye /var/lib/machines/machine1 http://mirror.csclub.uwaterloo.ca/debian

Note that the systemd-container package must be installed in the guest.

Now do a bit of setup in the rootfs:

chroot /var/lib/machines/machine1
# Only do this if you want to use `machinectl login`
passwd -d root
cat <<EOF >>/etc/securetty
pts/0
pts/1
pts/2
pts/3
EOF
# set hostname
echo machine1 > /etc/hostname
# set FQDN
nano /etc/hosts
# set up network config
nano /etc/network/interfaces
exit

Now paste the following into /etc/systemd/nspawn/machine1.nspawn:

[Exec]
Boot=yes
Hostname=machine1
PrivateUsers=no

[Network]
Bridge=br0

Replace 'br0' by the bridge interface on the host to which the container should be attached (a veth pair will be created when the container starts up).

Also make sure to set 'PrivateUsers=no', because by default systemd-nspawn uses some randomized UID/GID mapping which makes it difficult to migrate the container to a different system.

Now start the container:

systemctl start systemd-nspawn@machine1

Or alternatively, using machinectl:

machinectl start machine1

To login to a container via an emulated serial console (I don't recommend doing this, since the TTY gets screwed up):

machinectl login machine1

Attach to a running container (similar to lxc-attach):

machinectl shell machine1

Note: if you see the error sh: 2: exec: : Permission denied, append /bin/bash to the end of the command:

machinectl shell machine1 /bin/bash

Important: make sure the container starts up at boot:

systemctl enable systemd-nspawn@machine1

Multiple network interfaces

Unfortunately systemd does not have a built-in way to create multiple bridged network interfaces. Thankfully, it's not too difficult to accomplish this using the VirtualEthernetExtra option and a systemd drop-in; the idea is to create some extra veth pairs and then manually attach them to the bridge.

Let's say you have three bridges on the host: br0, br1 and br2, and you want the container to be attached to all three. Make your nspawn file look like this:

...
[Network]
Bridge=br0
# These will be manually bridged to the host
VirtualEthernetExtra=ve-machine1-1:veth1
VirtualEthernetExtra=ve-machine1-2:veth2

Now run systemctl edit systemd-nspawn@machine1 and paste the following:

[Service]
ExecStartPost=/usr/sbin/ip link set dev ve-machine1-1 master br1
ExecStartPost=/usr/sbin/ip link set dev ve-machine1-1 up
ExecStartPost=/usr/sbin/ip link set dev ve-machine1-2 master br2
ExecStartPost=/usr/sbin/ip link set dev ve-machine1-2 up

In the container, there will be three interfaces:

  • host0, which is attached to br0 on the host
  • veth1, which is attached to br1 on the host
  • veth2, which is attached to br2 on the host

Make sure you update /etc/network/interfaces in the container accordingly.