SSH access to office host behind NAT router
Solution 1:
youatwork@officepc$ autossh -R 12345:localhost:22 notroot@serverpc
Later:
you@homepc$ autossh -L 23456:localhost:12345 notroot@serverpc
you@homepc$ ssh youatwork@localhost -p 23456
What you could do is this: in step 1 forward a remote port from the office PC to the server (12345
is used as an example, any port >1024 should do). Now connecting to 12345 on the server should connect you to port 22 on officepc.
In step 2, forward the port 23456 from your home machine to 12345 on the server (whence it gets forwarded to officepc:22, as set up in step 1)
In step 3, you connect to the local port 23456 with your office PC login. This is forwarded by step 2 to port 12345 on your server, and by step 1 to your office PC.
Note that I'm using autossh for the forwardings, as it's a ssh wrapper which automatically reconnects the tunnel should it be disconnected; however normal ssh would work as well, as long as the connection doesn't drop.
There is a possible vulnerability: anyone who can connect to localhost:12345 on serverpc can now connect to officepc:22, and try to hack into it. (Note that if you're running a SSH server, you should anyway secure it above the basic protections which are on by default; I recommend at least disabling root login and disabling password authentication - see e.g. this)
Edit: I have verified this with the same config, and it works. GatewayPorts no
only affects the ports that are open to the world at large, not local tunnels. This is what the forwarded ports are:
homepc:
outgoing ssh to serverpc:22
listening localhost:23456 forwarded through ssh tunnel
serverpc:
listening ssh at *:22
incoming localhost ssh tunnel (from homepc) forwarded to localhost:12345
listening localhost ssh tunnel (from officepc) forwarded from localhost:12345
officepc:
outgoing ssh to serverpc:22
incoming localhost through ssh tunnel (from serverpc) forwarded to localhost:22
So, as far as the network stack is concerned, it's all local traffic on the respective loopback interfaces (plus ssh connections to serverpc); therefore, GatewayPorts
is not checked at all.
There is, however, the directive AllowTcpForwarding
: if that is no
, this setup will fail as no forwarding is allowed at all, not even across the loopback interface.
Caveats:
-
if using autossh and recent ssh, you may want to use ssh's
ServerAliveInterval
andServerAliveCountMax
for keeping the tunnel up. Autossh has a built-in check, but apparently it has some issues on Fedora.-M0
disables that, and-oServerAliveInterval=20 -oServerAliveCountMax=3
checks if the connection is up - tries each 20 sec, if it fails 3x in a row, stops ssh (and autossh makes a new one):autossh -M0 -R 12345:localhost:22 -oServerAliveInterval=20 -oServerAliveCountMax=3 notroot@serverpc autossh -M0 -L 23456:localhost:12345 -oServerAliveInterval=20 -oServerAliveCountMax=3 notroot@serverpc
it might be useful to restart ssh tunnel if the forward fails, using
-oExitOnForwardFailure=yes
- if the port is already bound, you might get a working SSH connection, but no forwarded tunnel.-
using
~/.ssh/config
for the options (and ports) is advisable, else the command lines get too verbose. For example:Host fwdserverpc Hostname serverpc User notroot ServerAliveInterval 20 ServerAliveCountMax 3 ExitOnForwardFailure yes LocalForward 23456 localhost:12345
Then you can use just the server alias:
autossh -M0 fwdserverpc
Solution 2:
If you can ssh to the internal server from home and from the internal server to your office Linux machine, then from home you can use the ssh ProxyCommand
to bounce silently through the server to the internal machine via nc
(netcat)
# ~/.ssh/config on your home machine:
Host internalpc
ForwardAgent yes
ProxyCommand ssh [email protected] exec nc internal.pc.example.com %p
Then you just ssh user@internalpc
and you are silently forwarded through the server machine, no opening of ports or tunnels required on either end.