| Unlike GRE (a simple encapsulating header) L2TP is a full-fledged communications protocol with control channel, data channels, and a robust command structure. In addition to PPP, other link layer types (called pseudowires) can be and are defined for delivery in L2TP by separate RFC documents. Further complexity is created by the capability to define vender-specific parameters beyond those defined in the L2TP specifications.
The endpoint devices of an L2TP connection can be an L2TP Access Concentrator (LAC) in which case it inputs/outputs the layer 2 protocol to/from the L2TP tunnel. Otherwise it is an L2TP Network Server (LNS), in which case it inputs/outputs the layer 3 (IP) protocol to/from the L2TP tunnel. The specifications describe three reference models: LAC-LNS, LAC-LAC, and LNS-LNS, the first of which is the most common case. The LAC-LNS model allows a remote access user to reach his home network or ISP from a remote location. The remote access user either dials (or otherwise connects via layer 2) to a LAC device which tunnels his connection home to an awaiting LNS. The LAC could also be located on the remote user's laptop which connects to an LNS at home using some generic internet connection. The other reference models may be used for more obscure scenarios.
Although the L2TP protocol does not contain encryption capability, it can be operated over IPSEC which would provide authentication and confidentiality. A remote user in the LAC-LNS model would most likely obtain a dynamically assigned IP address from the home network to ultimately use through the tunnel back to the home network. Secondly, the outer IP source address used to send the L2TP tunnel packet to the home network is likely to be unknown or highly variable. Thirdly, since the LNS provides the remote user with a dynamic IP address to use, the firewall at the home network would have to be dynamically updated to accept this address in conjunction with the outer tunnel address. Finally, there is also the issue of authentication of the remote user prior to divulging an acceptable IP address. As a result of all of these complications, the strict filtering rules applied to the IP-in-IP and GRE tunneling cases will likely not be possible in the L2TP scenario.
In addition to the difficulty of enforcing addresses and endpoints (as explained above), the L2TP protocol itself is a security concern if allowed through a security boundary. In particular:
1) L2TP potentially allows link layer protocols to be delivered from afar. These protocols were intended for link-local scope only, are less defended, and not as well-known
2) The L2TP tunnels can carry IP packets that are very difficult to see and filter because of the additional layer 2 overhead
3) L2TP is highly complex and variable (vender-specific variability) and therefore would be a viable target that is difficult to defend. It is better left outside of the main firewall where less damage occurs if the L2TP-processing node is compromised.
4) Filtering cannot be used to detect and prevent other unintended layer 2 protocols from being tunneled. The strength of the application layer code would have to be relied on to achieve this task.
5) Regardless of whether the L2TP is handled inside or outside of the main network, a secondary layer of IP filtering is required, therefore bringing it inside doesn't save resources.
Therefore, it is not recommended to allow unencrypted L2TP packets across the security boundary into the network's protected areas. Reference the Backbone Transport STIG for additional L2TP guidance and use. |
