SSH tunnel with single hop, using systemd-networkd and autossh

Recently I had the pleasure of setting up a SSH tunnel between two virtual machines that share no route and are located in two different subnets.
They can however reach each other via SSH, hopping their host.
Let's assume the following setup:
  • client1 (Arch Linux) has
  • client2 (Arch Linux) has
  • host (Debian) is to client1 and to client2
As I needed the two clients to be able to send mail to each other and reach each others' services, I did some digging and opted for a SSH connection using TUN devices (aka. "poor man's VPN").
The following is needed to set this up:
  • root access on both virtual machines (client1 & client2)
  • a user account on the host system
  • SSH (OpenSSH assumed) installed on all three machines

Connect the clients

Change sshd_config

The following two settings have to be made in each clients /etc/ssh/sshd_config (to allow root login and the creation of TUN devices):
PermitRootLogin yes
PermitTunnel yes
I hope it is needless to say, that permitting root access via SSH has its caveats. You should make sure to set a very secure password, or only allow SSH keys for login.

Generate and exchange keys

Generate SSH keys on client1 (you can of course use other key types, if your OpenSSH installation allows and supports it):
ssh-keygen -t rsa -b 4096 -C "$(whoami)@$(hostname)-$(date -I)"
Here you can choose between setting a password for the key (to unlock the key with ssh-add yourself) or not setting one (to be able to use the key on system boot with an automated service).
Add them to your user at host like this:
ssh-copy-id -i .ssh/id_rsa user@host
Also add it to /root/.ssh/authorized_keys on client2.

Use ProxyCommand to connect

To make a first connection between the clients, one can use the following settings in /root/.ssh/config of client1 to hop host and connect to client2:
Host client2
  ProxyCommand ssh user@ -W
  ForwardAgent yes
  User root
  ServerAliveInterval 120
  Compression yes
  ControlMaster auto
  ControlPath ~/.ssh/socket-%r@%h:%p
The ForwardAgent yes setting here is especially interesting, as it forwards the SSH key of client1 to client2.
On client1 a simple
ssh client2 -v
should now directly connect to client2 by hopping host.


Start the tunnel

Now to the fun part: Creating the tunnel.
OpenSSH supports a feature similar to VPN, that creates a TUN device on both ends of the connection. As the "direct" (hopping host) connection between client1 and client2 has been setup already, let's try the tunnel:
ssh -w5:5 client2 -v
The -w switch will create a TUN device (tun5 to be exact) on each client.
Now, to start the tunnel without executing a remote command (-N), compression of the data (-C) and disabling pseudo-tty allocation (-T), one can use the following:
ssh -NCTv -w5:5 client2

Setting up the TUN devices

A short
ip a s
on client1 and client2 shows, that the tun5 devices have been created on both clients. However they don't feature a link yet.
This can be achieved by setting up a systemd network with the help of systemd-networkd. By placing a .network file in /etc/systemd/network/, the TUN device will be configured as soon as it shows up.
Here I chose the subnet, but you could use any other private subnet (that's still available in your setup).
On client1 (/etc/systemd/network/


On client2 (/etc/systemd/network/


After adding the files a restart of the systemd-networkd service on both machines is necessary.
systemctl restart systemd-networkd
Now starting the tunnel again should give a fully working point-to-point TCP connection between the two (virtual) machines using the TUN devices.
If you need a more complex setup (i.e. to access the other clients' subnet), you will have to apply some routes (either using netfilter or systemd-networkd), depending on your individual setup.


To make both hosts know about each other by hostname (and domain, if any), too, those can be added to the clients' /etc/hosts files.
On client1 (/etc/hosts): client2
On client2 (/etc/hosts): client1


If using postfix as MTA, the service has to be configured to use /etc/hosts before resolving to your networks DNS resolving.
On client1 and client2 (/etc/postfix/
lmtp_host_lookup = native
smtp_host_lookup = native
ignore_mx_lookup_error = yes

Autossh and system boot

Wrapping it all up, it's usually intended to have a tunnel service be started on system boot. SSH tunnels are supposedly known for their poor connectivity. One way to get around this issue is to manage them with autossh .
A simple systemd service file can then be used to manage this behavior.
On client1 (/etc/systemd/system/tunnel@.service):
Description=AutoSSH tunnel to a host

ExecStart=/usr/bin/autossh -M 0 -NCTv -o ServerAliveInterval=45 -o ServerAliveCountMax=2 -o TCPKeepAlive=yes -w 5:5 %I

Enable the service with
systemctl enable tunnel@client2
Start the service with
systemctl start tunnel@client2