NAT and Protocol Compatibility

Some protocols do not work well with NAT and others will not work at all. Problematic protocols embed IP addresses and/or port numbers within packets (e.g. SIP and FTP), some do not work properly if the source port is rewritten (SIP from a PBX, IPsec), and some are difficult because of limitations of pf (PPTP). This section covers a sampling of protocols that have difficulties with NAT in pfSense, and how to work around these issues where possible.

FTP

FTP poses problems with both NAT and firewalls because of the design of the protocol. FTP was initially designed in the 1970s, and the current standard defining the specifications of the protocol was written in 1985. Since FTP was created more than a decade prior to NAT, and long before firewalls were common, it acts in ways that are very unfriendly toward NAT and firewalls.

pfSense does not include an FTP proxy by default, but there is a client proxy available as an add-on package.

FTP Limitations

Because pf lacks the ability to properly handle FTP traffic without a proxy, and the FTP proxy package implementation is somewhat lacking, there are some restrictions on the usage of FTP.

FTP servers behind NAT

For FTP servers behind NAT, all relevant ports must be manually forwarded in to the server and allowed in firewall rules. Or in the case of 1:1 NAT, only the firewall rules are necessary. Depending on the FTP mode, server software, and client software, some server configuration may also be required.

FTP modes

FTP can act in multiple modes that change the behavior of the client and server, and which side listens for incoming connections. The complications of NAT and firewall rules depend on these modes and whether a remote client is attempting to reach a server behind pfSense, or if a client behind pfSense is attempting to reach a remote server.

Active Mode

With Active Mode FTP, when a file transfer is requested, the client listens on a local port and then tells the server the client IP address and port. The server will then connect back to that IP address and port in order to transfer the data. This is a problem for firewalls because the port is typically random, though modern clients allow for limiting the range that is used. In the case of a client behind NAT, the IP address given would be a local address, unreachable from the server. Not only that, but a firewall rule would need to be added along with a port forward allowing traffic into this port.

When the FTP proxy package is in use and a client is behind pfSense connecting to a remote server, the proxy attempts to do three major things: First, it will rewrite the FTP PORT command so that the IP address is the WAN IP address of the firewall, and a randomly chosen port on that IP address. Next, it adds a port forward that connects the translated IP address and port to the original IP address and port specified by the FTP client. Finally, it allows traffic from the FTP server to connect to that “public” port. With Multi-WAN, the proxy will only function on the WAN containing the default gateway.

When everything is working properly, this all happens transparently. The server never knows it is talking to a client behind NAT, and the client never knows that the server isn’t connecting directly.

In the case of a server behind NAT, active mode is not usually a problem since the server will only be listening for connections on the standard FTP ports and then making outbound connections back to the clients. The outbound firewall rules must allow the server to make arbitrary outbound connections, and the rules must not policy route those connections out a WAN other than the one that accepted the inbound FTP connection.

Passive Mode

Passive Mode (PASV) acts somewhat in reverse. For clients, it is more NAT and firewall friendly because the server listens on a port when a file transfer is requested, not the client. Typically, PASV mode will work for FTP clients behind NAT without using any proxy or special handling at all.

Similar to the situation in the previous section, when a client requests PASV mode the server will provide the client with its IP address and a random port to which the client can attempt to connect. Since the server is on a private network, that IP address and port will need to be translated and allowed through the firewall. See FTP Servers and Port Forwards below for rule requirements. The FTP server must provide the public IP address to which clients connect, but some clients such as Filezilla are smart enough to ignore a given IP address if it is private, and will connect to the original server IP address instead.

Extended Passive Mode

Extended Passive Mode (EPSV) works similar to PASV mode but makes allowances for use on IPv6. When a client requests a transfer, the server will reply with the port to which the client should connect. The same caveats for servers in PASV mode apply here.

FTP Servers and Port Forwards

For FTP servers providing passive mode to clients, the configuration of the FTP server must define a passive port range and must also set the external NAT address, typically the WAN IP address of the firewall. The means of setting these values varies depending on the FTP server software implementation. Consult the FTP server documentation for more information. On the firewall, the passive port range must be forwarded in with port forwards along with TCP port 21.

For FTP servers providing active mode to clients, a port forward is only required for TCP port 21.

FTP Servers and 1:1 NAT

With 1:1 NAT, firewall rules must allow port 21 and the passive port range.

TFTP

Standard TCP and UDP traffic initiates connections to remote hosts using a random source port in the ephemeral port range, which varies by operating system but falls within 1024-65535, and the destination port of the protocol in use. Replies from server to client reverse that: The source port is the client destination port, and the destination port is the client source port. This is how pf associates the reply traffic with connections initiated from inside a network.

TFTP (Trivial File Transfer Protocol) does not follow this convention, however. The standard defining TFTP, RFC 1350, specifies the reply from the TFTP server to client will be sourced from a pseudo-random port number. The TFTP client may choose a source port of 10325 (as an example) and use the destination port for TFTP, port 69. The server for other protocols would then send the reply using source port 69 and destination port 10325. Since TFTP instead uses a pseudo- random source port, the reply traffic will not match the state pf has created for this traffic. Hence the replies will be blocked because they appear to be unsolicited traffic from the Internet.

TFTP is not a commonly used protocol across the Internet. The only situation that occasionally comes up where this is an issue is with some IP phones that connect to outside VoIP providers on the Internet using TFTP to pull configuration and other information. Most VoIP providers do not require this.

If TFTP traffic must pass through the firewall, a TFTP proxy is available which is configured under System > Advanced on the Firewall & NAT tab. See TFTP Proxy for more information.

PPTP / GRE

The limitations with PPTP in pfSense are caused by limitations in the ability of pf to NAT the GRE protocol. As such, the limitations apply to any use of the GRE protocol, however PPTP has been the most common use of GRE in the wild.

The state tracking code in pf for the GRE protocol can only track a single session per public IP address per external server. This means if a PPTP VPN connection is in place, only one internal machine can connect simultaneously to the same a PPTP server on the Internet. A thousand machines can connect simultaneously to a thousand different PPTP servers, but only one simultaneously to a single server. A single client can also connect to an unlimited number of outside PPTP servers.

The only available work around is to use multiple public IP addresses on the firewall, one per client via Outbound or 1:1 NAT, or to use multiple public IP addresses on the external PPTP server. This is not a problem with other types of VPN connections.

Due to the extremely flawed security in PPTP (See PPTP Warning), including a complete compromise of the entire protocol, its usage should be discontinued as soon as possible, so this issue is not relevant given the current security standards.

Online Games

Games typically are NAT friendly aside from a couple caveats. This section refers to both PC games and console gaming systems with online capabilities. This section provides an overview of the experiences of numerous pfSense users. Visit the Gaming board on the pfSense forum to find more information.

Static Port

Some games do not work properly unless static port is enabled on outbound NAT rules. If a game has problems establishing a connection, the best thing to try first is enabling static port for traffic coming from the console. See Static Port for more information.

Multiple players or devices behind one NAT device

Some games have issues where multiple players or devices are behind a single NAT device. These issues appear to be specific to NAT, not pfSense, as users who have tried other firewalls experience the same problems with them as well. Search the Gaming board on the pfSense forum for the game or system to find information from others with similar experiences.

Overcome NAT issues with UPnP

Many modern game systems support Universal Plug-and-Play (UPnP) to automatically configure any required NAT port forwards and firewall rules. Enabling UPnP on pfSense will typically allow games to work with little or no intervention. See UPnP & NAT-PMP for more information on configuring and using UPnP, and for information on potential security concerns.