Making dnscrypt-proxy and Docker play well together
This post is available in Russian here.
DNSCrypt is a fairly popular way of protecting DNS traffic that is usually left unencrypted from other people. dnscrypt-proxy, a client program that implements DNSCrypt, also supports the DNS-over-HTTPS protocol, allowing name resolution over DoH.
Unfortunately, leaving dnscrypt-proxy with its default settings while setting it as the default resolver breaks name resolution in Docker containers. Fixing this while not exposing a DNS resolver on the LAN is what’s described below.
This post in short
- Create a dummy adapter, assign a private network IP to it.
- Make dnscrypt-proxy listen on this interface, change system DNS settings accordingly.
Docker sets up DNS in containers by copying host DNS settings,
/etc/resolv.conf from the host.
dnscrypt-proxy and other custom resolvers usually bind to
/etc/resolv.conf entry is
nameserver 127.0.0.1. This setting
is propagated to the containers, but, as the containers belong to a different
network namespace, host’s
127.0.0.1 and container’s mean two different things.
The easiest solution is to run dnscrypt-proxy on host’s public IP address and
then add this address to
/etc/resolv.conf. This means we expose a DNS
resolver to the network, and we’d rather not.
Instead we’ll create a
dummy network adapter that is routable yet doesn’t
actually send any packets; it is routable from the container since Docker
containers use the host machine as their default gateway.
Creating the adapter
dummy kernel module is not loaded yet (
% lsmod | grep dummy displays
nothing), load it and enable its autostart:
# modprobe dummy # echo "dummy" >> /etc/modules-load.d/net_dummy.conf
Creating and setting up a dummy adapter is as simple as running these two commands on any modern Linux system with iproute2 installed:
# ip link add type dummy name dummy0 # ip addr add dev dummy0 10.0.197.1/24
Making this permanent will vary between network configuration software. With systemd-networkd you’ll need two config files:
[NetDev] Name=dummy0 Kind=dummy Description=Dummy network for dnscrypt-proxy
[Match] Name=dummy0 [Network] DHCP=no Address=10.0.197.1/24 DefaultRouteOnDevice=false
Changing DNS settings
To bind dnscrypt-proxy to a new address, edit
/etc/dnscrypt-proxy/dnscrypt-proxy.toml and make it look like this:
listen_addresses = ['127.0.0.1:53', '[::1]:53', '10.0.197.1:53']
Restart dnscrypt-proxy and then replace the text in
(or wherever your network configurator stores DNS settings) with this:
Run a new container:
% docker run -it --rm alpine:3.12 # cat /etc/resolv.conf nameserver 10.0.197.1 # ping -c 1 ya.ru
If you use a firewall (and you should be), then allow incoming traffic to
10.0.197.1:53 from the subnets Docker uses for containers.
If you use systemd-resolved as your caching resolver of choice with
dnscrypt-proxy set as its upstream, then you’re still fine even though
systemd-resolved won’t allow you to listen on anything besides
Docker detects the use of systemd-resolved and copies
/run/systemd/resolve/resolv.conf, which is generated from resolved settings,
instead of using