LXD Container Home Server Networking For Dummies

Why?

If you're going to operate a fleet of LXD containers for home entertainment, you probably want some of them exposed with their own ip addresses on your home network, so that you can use them as containerized servers for various applications.

Others containers, you might want to be inaccessable from the lan, in a natted subnet, where they can solicit connections to the outside world from within their natted subnet, but are not addressable from the outside. A database server that you connect a web app to, for instance, or a web app that you have a reverse proxy in front of.

But these are two separate address spaces, so ideally all of the containers would have a second interface of their own, by which they could connect to a third network, that would be a private network that all of the containers can use to talk directly to each other (or the host machine).

It's pretty straightforward, you just have to glue all the pieces together.

Three Part Overview.

Define and create some bridges.
Define profiles that combine the network interfaces in different combinations. In addition to two bridges you will have a macvlan with which to expose the containers that you want exposed, but the macvlan doesn't come into play until here in step two when you define profiles.
Assign each container which profile it should use, and then configure the containers to use the included network interfaces correctly.

Build Sum Moar Bridges

The containers will all have two network interfaces from their own internal point of view, eth0 and eth1.

In this scheme we create a bridge for a natted subnet and a bridge for a non-natted subnet. All of the containers will connect to the non-natted subnet on their second interface, eth1, and some of the containers will connect to the natted subnet on their first interface eth0. The containers that don't connect to the natted subnet will instead connect to a macvlan on their first interface eth0, but that isn't part of this step.

bridge for a natted subnet

If you haven't used lxd before, you'll want to run the command lxd init. By default this creates exactly the bridge we want, called lxdbr0.

Otherwise you would use the following command to create lxdbr0.

lxc network create lxdbr0

To generate a table of all the existing interfaces.

lxd network list

This bridge is for our natted subnet, so we just want to go with the default configuration.

lxc network show lxdbr0

This cats a yaml file where you can see the randomly generated network for lxdbr0.

config:
  ipv4.address: 10.99.153.1/24
  ipv4.nat: "true"
  ipv6.address: fd42:211e:e008:954b::1/64
  ipv6.nat: "true"
description: ""
name: lxdbr0
type: bridge
used_by: []
managed: true

bridge for a non-natted subnet

Create lxdbr1

lxc network create lxdbr1

Use the following commands to remove nat from lxdbr1.

lxc network set lxdbr1 ipv4.nat false
lxc network set lxdbr1 ipv6.nat false

Of if you use this next command, your favourite text editor will pop open, preloaded with the complete yaml file and you can edit the configuration there.

lxc network edit lxdbr1

Either way you're looking for a result such as the following. Notice that the randomly generated address space is different that the one for lxdbr0, and that the *nat keys are set to "false".

config:
  ipv4.address: 10.151.18.1/24
  ipv4.nat: "false"
  ipv6.address: fd42:89d4:f465:1b20::1/64
  ipv6.nat: "false"
description: ""
name: lxdbr1
type: bridge
used_by: []
managed: true

Profiles

recycle the default

When you first ran lxd init, that created a default profile. Confirm with the following.

lxc profile list

To see what the default profile looks like.

lxc profile show default

config:
  environment.http_proxy: ""
  security.privileged: "true"
  user.network_mode: ""
description: Default LXD profile
devices:
  eth0:
    nictype: bridged
    parent: lxdbr0
    type: nic
  root:
    path: /
    pool: default
    type: disk
name: default
used_by: []

profile the natted

The easiest way to create a new profile is start by copying another one.

lxc profile copy default natted

edit the new natted profile

lxc profile edit natted

And add an eth1 interface attached to lxdbr1. eth0 and eth1 will be the interfaces visible from the container's point of view.

config:
  environment.http_proxy: ""
  security.privileged: "true"
  user.network_mode: ""
description: Natted LXD profile
devices:
  eth0:
    nictype: bridged
    parent: lxdbr0
    type: nic
  eth1:
    nictype: bridged
    parent: lxdbr1
    type: nic
  root:
    path: /
    pool: default
    type: disk
name: natted
used_by: []

Any container assigned to the natted profile, will have an interface eth0 connected to a natted subnet, and a second interface eth1 connected to a non-natted subnet, with a static ip on which it will be able to talk directly to the other containers and the host machine.

profile the exposed

Create the exposed profile

lxc profile copy natted exposed

and edit the new exposed profile

lxc profile edit exposed

change the nictype for eth0 from bridged to macvlan, and the parent should be the name of the physical ethernet connection on the host machine, instead of a bridge.

config:
  environment.http_proxy: ""
  security.privileged: "true"
  user.network_mode: ""
description: Exposed LXD profile
devices:
  eth0:
    nictype: macvlan
    parent: eno1
    type: nic
  eth1:
    nictype: bridged
    parent: lxdbr1
    type: nic
  root:
    path: /
    pool: default
    type: disk
name: exposed
used_by: []

Any container assigned to the exposed profile, will have an interface eth0 connected to a macvlan, addressable from your lan, just like any other arbitrary computer on your home network, and a second interface eth1 connected to a non-natted subnet, with a static ip on which it will be able to talk directly to the other containers and the host machine.

exposed profile with a regular linux br0 interface bridge

You can configure an Ubuntu server with a br0 interface

# /etc/network/interfaces
auto lo
iface lo inet loopback

# br0 bridge in dhcp configuration with ethernet
# port ens2 added to it.
auto br0
iface br0 inet dhcp
      bridge_ports ens2
      bridge_stp off
      bridge_maxwait 0

and a cooresponding profile....

config: {}
description: exposed LXD profile
devices:
  eth0:
    nictype: bridged
    parent: br0
    type: nic
  eth1:
    nictype: bridged
    parent: lxdbr1
    type: nic
  root:
    path: /
    pool: default
    type: disk
name: exposed
used_by: []

Assign Containers to Profiles and configure them to connect correctly.

There are a lot of different ways that a Linux instance can solicit network services. So for now I will just describe a method that will work here for a lxc container from ubuntu:16.04, as well as a debian stretch container from images.linuxcontainers.org.

Start a new container and assign the profile. We'll use an arbitrary whimsical container name, quick-joey. This process is the same for either the natted profile or the exposed profile.

lxc init ubuntu:16.04 quick-joey
# assign the profile
lxc profile assign quick-joey exposed
# start quick-joey
lxc start quick-joey
# and start a bash shell
lxc exec quick-joey bash

With either an ubuntu:16.04 container, or a debian stretch container, for either the natted or exposed profile, because of all the above configuration work they will automatically connect on their eth0 interfaces and be able to talk to the internet. You need to edit /etc/network/interfaces, the main difference being what that file looks like before you edit it.

You need to tell these containers how to connect to the non-natted subnet on eth1.

ubuntu:16.04

If you start a shell on an ubuntu:16.04 container, you see that /etc/network/interfaces describes the loopback device for localhost, then sources /etc/network/interfaces.d/*.cfg where some magical cloud-config jazz is going on. You just want to add a static ip description for eth1 to the file /etc/network/interfaces. And obviously take care that the static ip address you assign is unique and on the same subnet with lxdbr1.

Reminder: the address for lxdbr1 is 10.151.18.1/24, (but it will be different on your machine).

auto lo
iface lo inet loopback

source /etc/network/interfaces.d/*.cfg
# what you add goes below here
auto eth1
iface eth1 inet static
   address 10.151.18.123
   netmask 255.255.255.0
   broadcast 255.255.255.255 
   network 10.151.18.0

ubuntu:16.04 using only dhcp for two nics

So the example here is tested with eth0 and eth1 connected to br0 and lxdbr1 respectively. You need post-up hooks for both eth0 and eth1 inside the containers, in order to specify the default route, eth0 gets it's configuration dynamically by default from cloud-init. So disable cloud-init by creating the following file on the container.

# /etc/cloud/cloud.cfg.d/99-disable-network-config.cfg
network: {config: disabled}

Then, on the container describe the interfaces.

# /etc/network/interfaces
auto lo
iface lo inet loopback

auto eth1
iface eth1 inet dhcp
post-up route del default dev eth1

auto eth0
iface eth0 inet dhcp
post-up route add default dev eth0 via 192.168.1.1

and delete /etc/network/interfaces.d/50-cloud-init.cfg

rm /etc/network/interfaces.d/50-cloud-init.cfg

The advantage to this scenario is now you can make copies of the container without having to update the network descriptions, because both interfaces will solicit addresses via dhcp.

debian stretch

The configuration for a debian stretch container is the same, except the the file /etc/network/interfaces will also describe eth0, but you only have to add the description for eth1.

systemd-networkd

This seems to work.

# eth0.network
[Match]
Name=eth0

[Network]
DHCP=ipv4

# eth1.network
[Match]
Name=eth1

[Network]
DHCP=ipv4

[DHCP]
UseRoutes=false

the /etc/hosts file

Once you assign the containers static ip addresses for their eth1 interfaces, you can use the /etc/hosts file on each container to make them aware of where the other containers and the host machine are.

For instance, if you want the container quick-joey to talk directly to the host machine, which will be at the ip address of lxdbr1, start a shell on the container quick-joey

lxc exec quick-joey bash

and edit /etc/hosts

# /etc/hosts
10.151.18.1    mothership

Or you have a container named fat-cinderella, that needs to be able to talk directly quick-joey.

lxc exec fat-cinderella bash
vim /etc/hosts

# /etc/hosts
10.151.18.123  quick-joey

etcetera