NAME nat-traverse - NAT gateway traversal utility SYNOPSIS To create a simple text-only tunnel, use the commands user@left $ nat-traverse 40000:natgw-of-right:40001 user@right $ nat-traverse 40001:natgw-of-left:40000 where 40000 is an unused UDP port on "left" and 40001 is an unused port on "right". See "EXAMPLES" for more. VERSION This document describes nat-traverse v0.7. DESCRIPTION nat-traverse establishes connections between nodes which are behind NAT gateways, i.e. hosts which do *not* have public IP addresses. Additionally, you can setup a small VPN by using pppd on top of nat-traverse (see "EXAMPLES"). nat-traverse does *not* need an external server on the Internet, and it isn't necessary to reconfigure the involved NAT gateways, either. *nat-traverse works out-of-the-box.* See "TECHNIQUE" for how this is achieved. Limitation: nat-traverse does not work with gateways which change the port numbers. This is a fundamental problem of nat-traverse's design, as the changed port numbers are (in general) not predictable. OPTIONS "*local_port*:*peer*:*remote_port*" (required) Sets the local port to use and the remote address to connect to. Note that you have to give the IP address or hostname of the *NAT gateway* of the host you want to connect to, as the target host doesn't have a public IP address. "--cmd="*pppd...*"" Runs the specified command after establishing the connection. The command will be run with its STDIN and STDOUT bound to the socket, i.e. everything the command writes to STDOUT will be forwarded to the peer. If no command is specified, nat-traverse will relay input from STDIN to the peer and vice versa, i.e. nat-traverse degrades to netcat. "--window=*10*" Sets the number of initial garbage packets to send. The default, 10, should work with most firewalls. "--timeout=*10*" Sets the maximum number of seconds to wait for an acknowledgement by the peer. "--quit-after-connect" Quits nat-traverse after the tunnel has been established successfully. nat-traverse returns a non-0 statuscode to indicate that it wasn't able to establish the tunnel. "--quit-after-connect" is useful if you want another program to use the tunnel. For example, you could configure OpenVPN to use the the same ports as nat-traverse -- thus OpenVPN would be able to cross NAT gateways. "--version", "--help" TECHNIQUE nat-traverse establishes connections between hosts behind NAT gateways without need for reconfiguration of the involved NAT gateways. 1. Firstly, nat-traverse on host "left" sends garbage UDP packets to the NAT gateway of "right". These packets are, of course, discarded by the firewall. 2. Then "right"'s nat-traverse sends garbage UDP packets to the NAT gateway of "left". These packets are *not* discarded, as "left"'s NAT gateway thinks these packets are replies to the packets sent in step 1! 3. "left"'s nat-traverse continues to send garbage packets to "right"'s NAT gateway. These packets are now not dropped either, as the NAT gateway thinks the packets are replies to the packets sent in step 2. 4. Finally, both hosts send an acknowledgement packet to signal readiness. When these packets are received, the connection is established and nat-traverse can either relay STDIN/STDOUT to the socket or execute a program. EXAMPLES Setup of a small VPN with PPP It's easy to setup a VPN (Virtual Private Network) by using the Point-to-Point Protocol Daemon, "pppd": root@left # nat-traverse \ --cmd="pppd updetach noauth passive notty \ ipparam vpn 10.0.0.1:10.0.0.2" 40000:natgw-of-right:40001 root@right # nat-traverse \ --cmd="pppd nodetach notty noauth" 40001:natgw-of-left:40000 "pppd" creates a new interface, typically "ppp0". Using this interface, you can ping 10.0.0.1 or 10.0.0.2. As you can see, "pppd" upgrades the data-only tunnel nat-traverse provides to a full IP tunnel. Thus you can establish reliable TCP connections over the tunnel, even though the tunnel uses UDP! Furthermore, you could even add IPv6 addresses to "ppp0" by running "ip -6 addr add..."! Note though that although this VPN *is* arguably a private network, it is *not* secured in any way. You may want to use SSH to encrypt the connection. Port Forwarding with netcat You can use "netcat" to forward one of your local UDP or TCP ports to an arbitrary UDP or TCP port of the remote host, similar to "ssh -L" or "ssh -R": user@left $ nat-traverse 10001:natgw-of-right:10002 \ --cmd="nc -vl 20000" user@right $ nat-traverse 10002:natgw-of-left:10001 \ --cmd="nc -v localhost 22" As soon as the tunnel is established (using UDP ports 10001 and 10002), "left"'s TCP port 20000 is forwarded to "right"'s SSH Daemon (TCP port 22): user@some-other-host $ ssh -p 20000 user@left # Will connect to right's SSH daemon! But do note that you lose the reliability of TCP in this example, as the actual data is transported via UDP; so this is only a toy example. If you want reliable streams, use PPP on top of nat-traverse, as described above. Setup of a VPN with OpenVPN You can use OpenVPN over nat-traverse if you want to have a *secure* VPN. Using OpenVPN over nat-traverse requires only one change to OpenVPN's configuration file, presuming that you don't want to use OpenVPN's multi-client mode: You have to adjust the "code" and "lport" options accordingly, for example: # Options to add to left's and right's OpenVPN config: port 60001 lport 60001 # Command to execute on left resp. right: root@left # until \ nat-traverse --quit-after-connect 60001:right:60001 \ do \ sleep 5 \ done; \ openvpn [...] root@right # until \ nat-traverse --quit-after-connect 60001:left:60001 \ do \ sleep 5 \ done; \ openvpn [...]