← home

NixOS Wireguard VPN setup

Recently I finally decided to configure VPN for my main machine. My choice fell on Mullvad VPN which is amazing. They don't even need your email for an account (first time seeing this :D).

This guide is for configuring Wireguard tunnel using Mullvad VPN servers, but you can apply this configuration to any Wireguard VPN.

The NixOS wiki shows starting ground for running Wireguard tunnel. The wiki shows multiple example with networking.wireguard, networking.wg-quick and systemd-networkd for both server and client. Moreover you can find mullvad-vpn app and configure VPN through the clean GUI, but it will not be declarative (not in your configuration.nix.

I recommend going with networking.wg-quick as it's probably the easiest one.

First of all, let's create a separate file, so it won't be a problem in the future to use this module on another machine.

# modules/wireguard.nix
{ pkgs, ... }: {
}

Don't forget to add this file to the imports list in the configuration.nix:

{ config, pkgs, ... }: {
  imports = [
    ./hardware-configuration.nix
    ./modules/wireguard.nix
  ];
  ...
}

In that file copy-paste the same configuration suggested in wiki's "Client" section:

{ pkgs, ... }: {
  networking.wg-quick.interfaces = let
    server_ip = "18.19.23.66";
  in {
    wg0 = {
      # IP address of this machine in the *tunnel network*
      address = [
        "10.64.186.60/32"
        "fdc9:281f:04d7:9ee9::2/64"
      ];

      # To match firewall allowedUDPPorts (without this wg
      # uses random port numbers).
      listenPort = 51820;

      # Path to the private key file.
      privateKeyFile = "/etc/mullvad-vpn.key";

      peers = [{
        publicKey = "1493vtFUbIfSpQKRBki/1d0YgWIQwMV4AQAvGxjCNVM=";
        allowedIPs = [ "0.0.0.0/0" ];
        endpoint = "${server_ip}:51820";
        persistentKeepalive = 25;
      }];
    };
  };
}

On Mullvad website go to "WireGuard configuration" in the left sidebar. Pick country, server, port and any content blockers you wish, enable killswitch checkbox. Download *.conf file, mine is dk-cph-wg-401.conf.

[Interface]
# Device: Fast Basset
PrivateKey = SL/xaxaFRogeNoDOOontGolvdIJ5x8mgLw0U/+1McG4=
Address = 10.75.130.74/32,fc00:bbbb:bbbb:bb01::4:be49/128
DNS = 100.64.0.3
PostUp = iptables -I OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT && ip6tables -I OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT
PreDown = iptables -D OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT && ip6tables -D OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT

[Peer]
PublicKey = Jjml2TSqKlgzW6UzPiJszaun743QYpyl5jQk8UOQYg0=
AllowedIPs = 0.0.0.0/0,::0/0
Endpoint = 146.70.197.194:51820

Private VPN configuration files

We can not put Mullvad VPN configuration file to the Nix configuration directly. The dk-cph-wg-401.conf contains private key which should not be shared. /nix/store is world readable, by putting this file in the *.nix any user would be able to read your private key.

Let's put private key to the /etc/mullvad-vpn.key:

cat dk-cph-wg-401.conf | grep "PrivateKey" | awk '{ print $3 }' | \
    sudo tee /etc/mullvad-vpn.key

Let's make it readable only by the owner (root):

sudo chmod 400 /etc/mullvad-vpn.key

Copy configuration from dk-cph-wg-401.conf to modules/wireguard.nix

Now you need to copy values from Mullvad config to Nix configuration so it's declarative. Each parameter has a line above it, which describes corresponding item inf Mullvad's dk-cph-wg-401.conf file.

{ pkgs, ... }: {
  networking.wireguard.interfaces = let
    # [Peer] section -> Endpoint
    server_ip = "18.19.23.66";
  in {
    wg0 = {
      # [Interface] section -> Address
      ips = [ "10.75.130.74/32" ];

      # [Peer] section -> Endpoint:port
      listenPort = 51820;

      # Path to the private key file.
      privateKeyFile = "/etc/mullvad-vpn.key";

      peers = [{
        # [Peer] section -> PublicKey
        publicKey = "1493vtFUbIfSpQKRBki/1d0YgWIQwMV4AQAvGxjCNVM=";
        # [Peer] section -> AllowedIPs
        allowedIPs = [ "0.0.0.0/0" ];
        # [Peer] section -> Endpoint:port
        endpoint = "${server_ip}:51820";
        persistentKeepalive = 25;
      }];
    };
  };
}

Now, that's enough to have a VPN tunnel. But, to be on a safe side you need a killswitch. The killswitch is networking filter which will allow traffic go only through VPN. So, when VPN tunnel suddenly goes down you won't expose your real IP address.

Killswitch! All traffic through VPN

If you enabled killswitch checkbox on Mullvad's configuration page, then, your *.conf file will have PostUp and PreDown fields. These are shell commands run before and after VPN is started.

PostUp = iptables -I OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT && ip6tables -I OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT
PreDown = iptables -D OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT && ip6tables -D OUTPUT ! -o %i -m mark ! --mark $(wg show %i fwmark) -m addrtype ! --dst-type LOCAL -j REJECT

In my config, Wireguard runs iptables to tell network stack that any packet which goes not through wg0 interface should be REJECTed. iptables is utility used to create network filters. Now copy those rules to the postUp and postDown parametrs in modules/wireguard.nix.

postUp = ''
  # Mark packets on the wg0 interface
  wg set wg0 fwmark 51820

  # Forbid anything else which doesn't go through wireguard VPN on
  # ipV4 and ipV6
  ${pkgs.iptables}/bin/iptables -A OUTPUT \
    ! -d 192.168.0.0/16 \
    ! -o wg0 \
    -m mark ! --mark $(wg show wg0 fwmark) \
    -m addrtype ! --dst-type LOCAL \
    -j REJECT
  ${pkgs.iptables}/bin/ip6tables -A OUTPUT \
    ! -o wg0 \
    -m mark ! --mark $(wg show wg0 fwmark) \
    -m addrtype ! --dst-type LOCAL \
    -j REJECT
'';

Note that Nix configuration has a few differences:

Same for PreDown -> postDown conversion:

postDown = ''
  ${pkgs.iptables}/bin/iptables -D OUTPUT \
    ! -o wg0 \
    -m mark ! --mark $(wg show wg0 fwmark) \
    -m addrtype ! --dst-type LOCAL \
    -j REJECT
  ${pkgs.iptables}/bin/ip6tables -D OUTPUT \
    ! -o wg0 -m mark \
    ! --mark $(wg show wg0 fwmark) \
    -m addrtype ! --dst-type LOCAL \
    -j REJECT
'';

Exclude traffic/port/IP from VPN

Local network application don't need VPN. We can exclude particular IP addresses or ports. For example to exclude kdeconnect ports (range 1714-1764 for UDP and TCP) from VPN add following iptables rules into postSetup:

# Accept kdeconnect connections
${pkgs.iptables}/bin/iptables -A INPUT -i wg0 -p udp \
    --dport 1714:1764 -m state --state NEW,ESTABLISHED -j ACCEPT
${pkgs.iptables}/bin/iptables -A INPUT -i wg0 -p tcp \
    --dport 1714:1764 -m state --state NEW,ESTABLISHED -j ACCEPT
${pkgs.iptables}/bin/iptables -A OUTPUT -o wg0 -p udp \
    --sport 1714:1764 -m state --state NEW,ESTABLISHED -j ACCEPT
${pkgs.iptables}/bin/iptables -A OUTPUT -o wg0 -p tcp \
    --sport 1714:1764 -m state --state NEW,ESTABLISHED -j ACCEPT

To exclude local port (in this case Deluge web client is running in container on port 8112):

# Allow deluge web gui
${pkgs.iptables}/bin/iptables -I OUTPUT -o lo -p tcp \
    --dport 8112 -m state --state NEW,ESTABLISHED -j ACCEPT

Another example is to exclude subnet for Nix containers (container has IP of 10.233.1.2 and host 10.233.1.1):

${pkgs.iptables}/bin/iptables -I OUTPUT -s 10.233.1.0/24 -d 10.233.1.0/24 \
    -j ACCEPT

Split Tunnel

Ok, but what if I want to route some traffic through the Wireguard and other traffic through other tunnel. For example, I have a company VPN which is needed for work. There is a way.

In this section I will describe how to add OpenVPN tunnel in addition to setup above. With the tunnel my machine would be connected to another subnet without going through Wireguard tunnel.

First of all, killswitch/iptables needs to know that traffic going to OpenVPN subnet should not be rejected. To accept new connection add these rules:

${pkgs.iptables}/bin/iptables -A INPUT -s 19.29.79.10 -d 192.168.0.100 \
  -m state --state NEW,ESTABLISHED -j ACCEPT
${pkgs.iptables}/bin/iptables -I OUTPUT -s 192.168.0.100 -d 19.29.79.10 \
  -m state --state NEW,ESTABLISHED -j ACCEPT

Where:

Note that I explicitly specified source -s and -d destination IPs instead of subnets. In my case I want node to node connection. For subnets IP mask need to be used /32.

Next, obtain your OpenVPN configuration file *.ovpn. This file is also secret and can not be shared, so, don't put it into Nix configuration. Copy it somewhere in your system.

I copied mine into /etc/openvpn/jellyfin-tunnel.ovpn. Also, as system already has VPN, OpenVPN needs to know that it should not route all the traffic through itself. Usually it does it with a network route. To tell OpenVPN not to create a global route add following line to your *.ovpn config:

pull-filter ignore redirect-gateway

That's all for preparations phase, now let's create the tunnel in the system:

# MANUAL ACTIONS ARE REQUIRED!
# - Copy your VPN configuration to /etc/openvpn/jellyfin-tunnel.ovpn
# - Add following line to /etc/openvpn/jellyfin-tunnel.ovpn
#
#     pull-filter ignore redirect-gateway
#
#   This means that OpenVPN won't create route which routes all the
#   traffic through OpenVPN tunnel.
{ config, pkgs, lib, ...}: {
  networking.dhcpcd.runHook = ''
    ${pkgs.iproute2}/bin/ip route add 19.29.79.10 via 192.168.0.1 dev enp34s0
  '';

  users.users = {
    openvpn = {
      name = "openvpn";
      group = "openvpn";
      isNormalUser = true;
      uid = 1100;
    };
  };

  users.groups.openvpn = {
    name = "openvpn";
    members = ["openvpn"];
    gid = 1100;
  };

  # Configure our OpenVPN client
  services.openvpn.servers = {
    jellyfin = {
      config = ''config /etc/openvpn/jellyfin-tunnel.ovpn'';
      autoStart = true;
    };
  };
}

Here, new network route is created to route all the traffic addressed to OpenVPN endpoint 19.29.79.10 through LAN gateway 192.168.0.1 (e.g. WiFi router). This is needed as without the route OpenVPN client will try to reach endpoint through Wireguard tunnel, which won't work (not sure why).

Then, new openvpn user and group are created with specific UID/GID. Specific UID/GID could be anything or you can even left it out. I like to specify user ID as then it's convenient in places where ID instead of name is used.

Finally, config declares OpenVPN connection with name openvpn-jellyfin. This name is used as a systemd's service name. You can check status of your VPN tunnel with:

sudo systemctl status openvpn-jellyfin

Enabling/Disabling autoStart is straightforward. Otherwise, you can start your tunnel with systemd:

sudo systemctl start openvpn-jellyfin

I want to use networking.wireguard

This was my initial approach and there's two additional things to handle:

Additional IP route

As configuration specifies allowedIPs = 0.0.0.0/0 all connection on wg0 interface will be routed through VPN tunnel. This creates a routing issue as Wireguard needs to connect to endpoint via public network.

To do so, create a new route to tell network stack to route traffic going to endpoint IP (18.19.23..) through main gateway (192.168.0.1 is my WiFi router):

networking.dhcpcd.runHook = ''
  ${pkgs.iproute2}/bin/ip route add 18.19.23.66/32 via 192.168.0.1 dev enp34s0
'';

You can create the route with networking.interfaces but it will not work just like that! The route will be flushed on suspend.

networking.interfaces.enp34s0.ipv4.routes = [{
    address = "18.19.23.66";
    prefixLength = 32;
    via = "192.168.0.1";
}];

Wireguard VPN doesn't work after suspend/sleep

Unfortunately, dhcpcd will not re-create an additional route created via networking.interfaces. There is similar problem described at Arch Wiki, but I don't use systemd-networkd so that doesn't apply.

On my system dhcpcd.service creates all necessary IP routes. But the one necessary for Wireguard is created by network-addresses-enp34s0.service. This service doesn't restart after suspend. As dhcpcd will remove all routes on wake-up, Wireguard will fail to connect to the endpoint.

Note that by using networking.dhcpcd.runHook this problem is solved as route is created by dhcpcd itself.

To make it work without dhcpcd hook, I decided to go with easy fix by restarting the VPN service after network is established. To do so:

See this code snippet for doing this in code.

The suspend-restart also creates route as described in Additional IP route. This is not a nice way to solve it but I didn't want to continue with this solution as I switched to wg-quick which doesn't need all of this. This is probably the same problem as described in Arch Wiki, so, if you know how to fix it send me a message, I will update the article.

References

Hey👋 I'm Andrey. In this blog I post my personal short tutorials or interesting technical notes. Over the day I work as a Software Engineer developing and testing Linux filesystems. I use free software mainly #NixOS #Neovim #Kitty. Btw I use NixOS. Subscribe for updates on:

telegram • @alberand@mas.to • twitter