How to wake on lan a remote host on demand using systemd's sockets

Some years ago I wrote an article on how to wake on lan a (SSHFS) fileserver on demand using autofs.

Today I want to describe a more generic way to do the same. This should not only work for SSH but for every services that communicates via TCP. Everything you need is systemd, netcat (nc) and a wake-on-lan tool like etherwake.

For the impatient

The first step is to create a systemd socket that listens on a local port for new connections. Since SSH normally uses port 22 I'm going to bind this on localhost:2222. Of course you can also listen on a public interface if you are building a public gateway to a server within the local net.




The second file we have to create is a service file that will be invoked by the socket. The basename must be identical to that of the socket, but followed by an @ sign and the ".service" suffix.


Description=Forwards request to another host. Sends WOL if not reachable.

ExecStartPre=/bin/sh -c 'for n in `seq 1 6`; do nc -z 22 && break || (wakeonlan D8:F4:62:67:86:3A >&2 && sleep 10); done'
ExecStart=/bin/nc -q0 22

Replace the the IP "", Port "22" and MAC address "D8:F4:62:67:86:3A" with those of the server that you want to connect to.

Next, reload and enable the daemon (Note: You have to enable the socket, not the service):

$ systemctl daemon-reload && systemctl start ssh-forward.socket && systemctl enable ssh-forward.socket

So, what does this do?

The socket file tells systemd to listen in Port 2222 for incoming connections. Whenever (this is what "Accept=true" means) a new connection to this port is established the service is invoked.

The service file first launches the command in ExecStartPre. You can ignore the "/bin/sh -c '...'" wrapper for the moment, it is only required because systemd does not allow shell operators within the command:

for n in `seq 1 6`; do nc -z 22 && break || (wakeonlan D8:F4:62:67:86:3A >&2 && sleep 10); done

"nc -z 22" checks if the remote port is reachable. This is more reliable than "ping" (as I used in my old article) since ICMP might be blocked by a firewall. If the port is reachable ExecStartPre returns immediately. If not, it sends a Wake-on-Lan package to the server's MAC address and waits 10 seconds for it to wake up. After that the check is done again, up to 5 times (`seq 1 6`), so the server has to wake up and boot within 60 seconds.

When ExecStartPre returns ExecStart is invoked:

ExecStart=/bin/nc -q0 22

This opens a TCP connection to on port 22, and pipes the incoming socket's connection to STDIN/STDOUT. When the connection to localhost:2222 finally is closed the "nc" process will end too, so the remote server can go to sleep again.

And what is this good for?

My fileserver goes to sleep after an idle time of 15 minutes. I have a Raspberry Pi behind my router that is running Nextcloud. Within Nextcloud I've defined a remote connection to the fileserver. This trick allows me to wake up the fileserver only if Nextcloud demands access to the fileserver.

Since Nextcloud only waits up to 10 seconds for an SFTP connection you might also want to apply this small patch to the the Nextcloud source that increases the timeout to 60 seconds:

$ patch -p1 <<EOF
--- a/apps/files_external/lib/Lib/Storage/SFTP.php	Tue Feb 07 21:48:03 2017 +0100
+++ b/apps/files_external/lib/Lib/Storage/SFTP.php	Tue Feb 07 21:49:28 2017 +0100
@@ -124,7 +124,7 @@
 		$hostKeys = $this->readHostKeys();
-		$this->client = new \phpseclib\Net\SFTP($this->host, $this->port);
+		$this->client = new \phpseclib\Net\SFTP($this->host, $this->port, 60);
 		// The SSH Host Key MUST be verified before login().
 		$currentHostKey = $this->client->getServerPublicHostKey();

But why not running the fileserver directly on the Pi?

  • RAID 5
  • Gigabit-Ethernet
  • BananaPi sucks (sorry, but that's true. I've tried it. Terrible software support. Really.)

And from time to time I need a potent system to do thinks like video transcoding.