tinc

July 2018

In this article I will describe how to use my NixOS tinc-module. I had to write my own, because the standard services.tinc module misses a lot of features, for example forwarding to sub-networks and network wise activation and deactivation. I designed it to work nicely with NixOps.

You have to enable and disable every network you define, instead of enable tinc which enables all defined networks. This should make it easy to define all your networks in one file (to keep track about everything), and micromanage in the computer specific definitions.

How to import

You can use fetchgit to import it without downloading it yourself. To find the newest rev and sha256 just call

nix-shell -p nix-prefetch-git --run "nix-prefetch-git https://github.com/mrVanDalo/nixos-tinc.git"
{pkgs, ... }:
let
  tincModule = pkgs.fetchgit {                                                     
    "url" = "https://github.com/mrVanDalo/nixos-tinc.git";             
    "rev" = "1b8c822a20be38ca19120f4c5d7f9f3236851674";              
    "sha256" = "0a99rxhrqwh454q6w1znb0icdpm8fl4rnr1dr7xblacqjmghhyh1";
  };

in {
  imports = [
    "${tincModule}"
  ];
}

Now you are ready to go!

Create Keys

First, we have to create tinc keys for every computer in the VPN Mesh. To put the keys in the current folder we use the '--config .' option,

$> nix-shell -p tinc_pre --run "tinc --config . generate-keys 4096"

After that is done we create the hostfile by

$> cat *.pub > hostfile ; rm *.pub

Key-File-structure for these Examples

In the following examples I expect the following file-structure for keys and config files.

|-- public
|   |-- Gibson
|   |   `-- hostfile
|   |-- Hackbardt
|   |   `-- hostfile
|   `-- HAL
|       `-- hostfile
`-- secrets
    |-- Gibson
    |   |-- ed25519_key.priv
    |   `-- rsa_key.priv
    |-- Hackbardt
    |   |-- ed25519_key.priv
    |   `-- rsa_key.priv
    `-- HAL
        |-- ed25519_key.priv
        `-- rsa_key.priv

Connect 3 computers

We want to connect 3 computers in a private network of range 10.1.1.0/24. One computer needs to be accessible from the internet, it will be the computer that connects all the other computer. You can have multiple computers which are reachable from the internet but for this example we only have one.

Here is the configuration.nix. First we define the whole topology in default and then we enable and configure secret-keys for every computer in Gibson, Hackbardt and HAL

let

  includePrivateKeys = host: {
    deployment.keys."rsa_key" = {
      keyFile = toString (./secrets/ + "/${host}/rsa_key.priv");
      destDir = "/root/secrets";
    };
    deployment.keys."ed25519_key" = {
      keyFile = toString (./secrets/ + "/${host}/ed25519_key.priv");
      destDir = "/root/secrets";
    };
  };


in {

# for all machines
# ----------------
default = { config, pkgs, lib, ... }:
{
  with lib;
  services.custom.tinc =
  let
    publicHostFile = host: lib.fileContents (./public + "/${host}/hostfile");
  in {
    "private" = {
      debugLevel    = 0;
      port          = 655;
      networkSubnet = "10.1.1.0/24";
      hosts = {
        Gibson = {
          realAddress = [ "my.awesome.dns.com" ];
          tincIp      = "10.1.1.1";
          publicKey   = publicHostFile "Gibson";
        };
        Hackbardt = {
          tincIp      = "10.1.1.2";
          publicKey   = publicHostFile "Hackbardt";
        };
        HAL = {
          tincIp      = "10.1.1.3";
          publicKey   = publicHostFile "HAL";
        };
      };
    };
  };
}

# Hackbardt specific
# ------------------
Gibson =
      enable                = true;
      connectTo             = [ "Gibson" ];
      privateRsaKeyFile     = config.deployment.keys."rsa_key".path;
      privateEd25519KeyFile = config.deployment.keys."ed25519_key".path;
    };
  };
}

# HAL specific
# ------------
HAL =
{config, pkgs, ... }:
includePrivateKeys "HAL" // {
  services.custom.tinc = {
    "private" = {
      enable                = true;
      connectTo             = [ "Gibson" ];
      privateRsaKeyFile     = config.deployment.keys."rsa_key".path;
      privateEd25519KeyFile = config.deployment.keys."ed25519_key".path;
    };
  };
}

}

If we deploy that and check the servers, we can see that tinc creates interfaces called tinc.private. Observing the routes we see that tinc sets up everything which is needed for proper routing.

$Gibson> ip addr show dev tinc.private
4: tinc.private: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP group default qlen 1000
    link/ether 3f:ac:bd:c2:f6:9c brd ff:ff:ff:ff:ff:ff
    inet 10.1.1.1/32 scope global tinc.private
       valid_lft forever preferred_lft forever
    inet 169.254.22.60/16 brd 169.254.255.255 scope global tinc.private
       valid_lft forever preferred_lft forever
    inet6 ffa1::2afc:b2ff:fcf2:f97a/64 scope link
       valid_lft forever preferred_lft forever
$Gibson> ip route show dev tinc.private
10.1.1.0/24 scope link
169.254.0.0/16 proto kernel scope link src 169.254.22.60 metric 204

It also creates /etc/host entries <computername>.<networkname> so you don’t have to remember the IPs.

$Gibson> ping HAL.private  -c 1
PING HAL.private (10.1.1.3) 56(84) bytes of data.
64 bytes from HAL.private (10.1.1.1): icmp_seq=1 ttl=64 time=5.27 ms

Awesome! That was easy!

Connect 2 sub-networks

So far so good, but let’s imagine we have some virtual machines running on 2 computers and want to make these virtual machines see each other. This is a very common problem in Kubernetes. It can be resolved by using the tincSubnet parameter, to configure sub-network routing.

Achieving this is very simple, just add the tincSubnet parameter in the hosts attribute and your done.

...
default = { config, pkgs, lib, ... }:
{
  with lib;
  services.custom.tinc =
  let
    publicHostFile = name: lib.fileContents  (./public + "/${host}/hostfile");
  in {
    "private" = {
      debugLevel    = 0;
      port          = 655;
      networkSubnet = "10.1.1.0/24";
      hosts = {
        Gibson = {
          realAddress = [ "my.awesome.dns.com" ];
          tincIp      = "10.1.1.1";
          publicKey   = publicHostFile "Gibson";
        };
        Hackbardt = {
          tincIp      = "10.1.1.2";
          subnetIp    = "10.2.2.0/24";
          publicKey   = publicHostFile "Hackbardt";
        };
        HAL = {
          tincIp      = "10.1.1.3";
          subnetIp    = "10.2.3.0/24";
          publicKey   = publicHostFile "HAL";
        };
      };
    };
  };
}

...

After deployment, we can see that Gibson has proper routing to the configured tincSubnet ranges as well as to 10.1.1.0/24 to reach the other network-nodes.

$Gibson> ip route show dev tinc.private
10.1.1.0/24 scope link
10.2.2.0/24 scope link
10.2.3.0/24 scope link
169.254.0.0/16 proto kernel scope link src 169.254.116.112 metric 203

Hackbardt has routing to the network provided by HAL, but has no routing (on the tinc.private interface) to the network it provides it self.

$Hackbardt> ip route show dev tinc.private
10.1.1.0/24 scope link
10.2.2.0/24 scope link
169.254.0.0/16 proto kernel scope link src 169.254.116.112 metric 203

The module also sets the sysctl parameter net.ipv4.config.tinc/private.forwarding and net.ipv6.config.tinc/private.forwarding to make sure the tinc.private interface forwards the traffic to the configured sub-networks.