Network Working Group                                        M. Holdrege
Request for Comments: 3027                                       ipVerse
Category: Informational                                     P. Srisuresh
                                                        Jasmine Networks
                                                            January 2001

Protocol Complications with the IP Network Address Translator

Status of this Memo

   This memo provides information for the Internet community.  It does
   not specify an Internet standard of any kind.  Distribution of this
   memo is unlimited.

Copyright Notice

Abstract

   Many internet applications can be adversely affected when end nodes
   are not in the same address realm and seek the assistance of an IP
   Network Address Translator (NAT) enroute to bridge the realms.  The
   NAT device alone cannot provide the necessary application/protocol
   transparency in all cases and seeks the assistance of Application
   Level Gateways (ALGs) where possible, to provide transparency.  The
   purpose of this document is to identify the protocols and
   applications that break with NAT enroute.  The document also attempts
   to identify any known workarounds.  It is not possible to capture all
   applications that break with NAT in a single document.  This document
   attempts to capture as much information as possible, but is by no
   means a comprehensive coverage.  We hope the coverage provides
   sufficient clues for applications not covered.

   1.0 Introduction ..............................................  2
   2.0 Common Characteristics of Protocols broken by NAT .........  2
   3.0 Protocols that cannot work with NAT enroute ...............  4
   4.0 Protocols that can work with the aid of an ALG ............  8
   5.0 Protocols designed explicitly to work with NAT enroute .... 16
   6.0 Acknowledgements .......................................... 17
   7.0 Security Considerations ................................... 17
   8.0 References ................................................ 17
   9.0 Authors' Addresses ........................................ 19
   10.0 Full Copyright Statement  ................................ 20

1.0 Introduction

   This document requires the reader to be familiar with the terminology
   and function of NAT devices as described in [NAT-TERM].  In a
   nutshell, NAT attempts to provide a transparent routing solution to
   end hosts requiring communication to disparate address realms.  NAT
   modifies end node addresses (within the IP header of a packet) en-
   route and maintains state for these updates so that datagrams
   pertaining to a session are transparently routed to the right end-
   node in either realm.  Where possible, application specific ALGs may
   be used in conjunction with NAT to provide application level
   transparency.  Unlike NAT, the function of ALG is application
   specific and would likely require examination and recomposition of IP
   payload.

   The following sections attempt to list applications that are known to
   have been impacted by NAT devices enroute.  However, this is by no
   means a comprehensive list of all known protocols and applications
   that have complications with NAT - rather just a subset of the list
   gathered by the authors.  It is also important to note that this
   document is not intended to advocate NAT, but rather to point out the
   complications with protocols and applications when NAT devices are
   enroute.

2.0 Common Characteristics of Protocols broken by NAT

   [NAT-TERM] and [NAT-TRAD] have sections listing the specific nature
   of problems and limitations to NAT devices.  Some of these
   limitations are being restated in this section to summarize
   characteristics of protocols that are broken by NAT.

2.1 Realm-specific IP address information in payload

   A wide range of applications fail with NAT enroute when IP packets
   contain realm-specific IP address or port information in payload.  An
   ALG may be able to provide work around in some cases.  But, if the
   packet payload is IPsec secured (or secure by a transport or
   application level security mechanisms), the application is bound to
   fail.

2.2 Bundled session applications

   Bundled session applications such as FTP, H.323, SIP and RTSP, which
   use a control connection to establish a data flow are also usually
   broken by NAT devices enroute.  This is because these applications
   exchange address and port parameters within control session to
   establish data sessions and session orientations.  NAT cannot know
   the inter-dependency of the bundled sessions and would treat each

   session to be unrelated to one another.  Applications in this case
   can fail for a variety of reasons.  Two most likely reasons for
   failures are:  (a) addressing information in control payload is
   realm-specific and is not valid once packet crosses the originating
   realm, (b) control session permits data session(s) to originate in a
   direction that NAT might not permit.

   When DNS names are used in control payload, NAT device in conjunction
   with a DNS-ALG might be able to offer the necessary application level
   transparency, if NAT has no contention with data session orientation.
   However, using DNS names in place of realm-specific IP addresses may
   not be an option to many of these applications (e.g., FTP).

   When realm-specific addressing is specified in payload, and the
   payload is not encrypted, an ALG may in some cases be able to provide
   the work around necessary to make the applications run transparently
   across realms.  The complexity of ALG depends on the application
   level knowledge required to process payload and maintain state.

2.3 Peer-to-peer applications

   Peer-to-peer applications more than client-server based applications
   are likely to break with NAT enroute.  Unlike Client-server
   applications, Peer-to-peer applications can be originated by any of
   the peers.  When peers are distributed across private and public
   realms, a session originated from an external realm is just as likely
   as the session from  a host in private realm.  External peers will be
   able to locate their peers in private realm only when they know the
   externally assigned IP address or the FQDN ahead of time.  FQDN name
   to assigned address mapping can happen only so long as the enroute
   NAT device supports DNS-ALG.  Examples of Peer-to-peer applications
   include interactive games, Internet telephony and event-based
   protocols (such as Instant-Messaging).

   This is particularly a problem with traditional NAT and may be less
   of an issue with bi-directional NAT, where sessions are permitted in
   both directions.

   A possible work-around for this type of problem with traditional-NAT
   is for private hosts to maintain an outbound connection with a server
   acting as their representative to the globally routed Internet.

2.4 IP fragmentation with NAPT enroute

   IP fragmentation with NAPT enroute is not an issue with any single
   application, but pervades across all TCP/UDP applications.  The
   problem is described in detail in [NAT-TRAD].  Briefly, the problem
   goes as follows.  Say, two private hosts originated fragmented

   TCP/UDP packets to the same destination host.  And, they happened to
   use the same fragmentation identifier.  When the target host receives
   the two unrelated datagrams, carrying same fragmentation id, and from
   the same assigned host address, the target host is unable to
   determine which of the two sessions the datagrams belong to.
   Consequently, both sessions will be corrupted.

2.5 Applications requiring retention of address mapping

   NAT will most likely break applications that require address mapping
   to be retained across contiguous sessions.  These applications
   require the private-to-external address mapping to be retained
   between sessions so the same external address may be reused for
   subsequent session interactions.  NAT cannot know this requirement
   and may reassign external address to different hosts between
   sessions.

   Trying to keep NAT from discarding an address mapping would require a
   NAT extension protocol to the application that would allow the
   application to inform the NAT device to retain the mappings.
   Alternately, an ALG may be required to interact with NAT to keep the
   address mapping from being discarded by NAT.

2.6 Applications requiring more public addresses than available

   This is a problem when the number of private hosts is larger than the
   external addresses available to map the private addresses into.  Take
   for example the rlogin service initiated from a host in private realm
   supported by NAPT.  Rlogin service clients use well-known rlogin port
   512 as their TCP port ID.  No more than one host in private realm can
   initiate the service.  This is a case of trying to use a service that
   fundamentally requires more public addresses than are available.  NAT
   devices can conserve addresses, but they cannot create more
   addresses.

3.0 Protocols that cannot work with NAT enroute

3.1 IPsec and IKE

   NAT fundamentally operates by modifying end node addresses (within
   the IP header) en-route.  The IPsec AH standard [IPsec-AH] on the
   other hand is explicitly designed to detect alterations to IP packet
   header.  So when NAT alters the address information enroute in IP
   header, the destination host receiving the altered packet will
   invalidate the packet since the contents of the headers have been
   altered.  The IPsec AH secure packet traversing NAT will simply not
   reach the target application, as a result.

   IPsec ESP ([IPsec-ESP]) encrypted packets may be altered by NAT
   device enroute only in a limited number of cases.  In the case of
   TCP/UDP packets, NAT would need to update the checksum in TCP/UDP
   headers, when an address in IP header is changed.  However, as the
   TCP/UDP header is encrypted by the ESP, NAT would not be able to make
   this checksum update.  As a result, TCP/UDP packets encrypted in
   transport mode ESP, traversing a NAT device will fail the TCP/UDP
   checksum validation on the receiving end and will simply not reach
   the target application.

   Internet Key Exchange Protocol IKE can potentially pass IP addresses
   as node identifiers during Main, Aggressive and Quick Modes.  In
   order for an IKE negotiation to correctly pass through a NAT, these
   payloads would need to be modified.  However, these payloads are
   often protected by hash or obscured by encryption.  Even in the case
   where IP addresses are not used in IKE payloads and an IKE
   negotiation could occur uninterrupted, there is difficulty with
   retaining the private-to-external address mapping on NAT from the
   time IKE completed negotiation to the time IPsec uses the key on an
   application.  In the end, the use of end-to-end IPsec is severely
   hampered anyway, as described earlier.

   For all practical purposes, end-to-end IPsec is impossible to
   accomplish with NAT enroute.

3.2 Kerberos 4

   Kerberos 4 tickets are encrypted.  Therefore, an ALG cannot be
   written.  When the KDC receives a ticket request, it includes the
   source IP address in the returned ticket.  Not all Kerberos 4
   services actually check source IP addresses.  AFS is a good example
   of a Kerberos 4 service which does not.  Services which don't check
   are not picky about NAT devices enroute.  Kerberos tickets are tied
   to the IP address that requested the ticket and the service with
   which to use the ticket.

   The K4 ticket (response) contains a single IP address describing the
   interface used by the  client to retrieve the ticket from the TGT
   from the perspective of KDC.  This works fine if the KDC is across a
   NAT gateway and as long as all of the Kerberos services are also
   across a NAT gateway.  The end user on private network will not
   notice any problems.

   There is also the caveat that NAT uses the same address mapping for
   the private host for the connection between the client and the KDC as
   for the connection between the client and the application server.  A
   work around this problem would be to keep an arbitrary connection
   open to remote server during throughout the ticket lifetime, so as

   not to let NAT drop the address binding.  Alternately, an ALG will
   need to be deployed to ensure that NAT would not change address
   bindings during the lifetime of a ticket and between the time a
   ticket is issued to private host and the time the ticket is used by
   private host.

   But, the ticket will be valid from any host within the private realm
   of NAPT.  Without NAPT, an attacker needs to be able to spoof the
   source IP addresses of a connection that is being made in order to
   use a stolen ticket on a different host.  With NAPT, all the attacker
   needs to do from the private realm of NAPT is to simply gain
   possession of a ticket.  Of course, this assumes, NAPT private domain
   is not a trusted network - not surprisingly, since many attacks occur
   from inside the organization.

3.3 Kerberos 5

   Just as with Kerberos 4, Kerberos 5 tickets are encrypted.
   Therefore, an ALG cannot be written.

   In Kerberos 5, the client specifies a list of IP addresses which the
   ticket should be valid for, or it can ask for a ticket valid for all
   IP addresses.  By asking for an all-IP-addresses ticket or a ticket
   containing the NAPT device address, you can get krb5 to work with an
   NAPT device, although it isn't very transparent (it requires the
   clients to behave differently than they otherwise would).  The MIT
   krb5 1.0 implementation didn't have any configurability for what IP
   addresses the client asked for (it always asked for the set of its
   interface addresses) and did not interact well with NAT.  The MIT
   krb5 1.1 implementation allows you to put "noaddresses" somewhere in
   krb5.conf to request all-IP-addresses-valid tickets.

   The K5 ticket (response) contains IP addresses, as requested by the
   client node, from which the ticket is to be considered valid.  If the
   services being accessed with Kerberos authentication are on the
   public side of the NAT, then the Kerberos authentication will fail
   because the IP address used by the NAT (basic NAT or NAPT) is not in
   the list of acceptable addresses.

   There are two workarounds in Kerberos 5 both of which reduce the
   security of the tickets.  The first is to have the clients in NAPT
   private realm specify the public IP address of the NAPT in the
   ticket's IP list.  But this leads to the same security problem
   detailed for K4.  Plus, it is not obvious for the client in the
   private domain to find out the public IP Address of the NAPT.  That
   would be a change of application behavior on end-host.

   The second method is to remove all IP addresses from the K5 tickets
   but this now makes theft of the ticket even worse since the tickets
   can be used from anywhere.  Not just from within the private network.

3.4 The X Windowing System and X-term/Telnet

   The X Windowing system is TCP based.  However, the client-server
   relationship with these applications is reverse compared to most
   other applications.  The X server or Open-windows server is the
   display/mouse/keyboard unit (i.e., the one that controls the actual
   Windows interface).  The clients are the application programs driving
   the Windows interface.

   Some machines run multiple X Windows servers on the same machine.
   The first X Windows server is at TCP port 6000.  The first Open
   Windows server can be at port 6000 or port 2000 (more flexible).  We
   will mainly refer X windowing system for illustration purposes here.

   X-term Transmits IP addresses from the client to the server for the
   purpose of setting the DISPLAY variable.  When set the DISPLAY
   variable is used for subsequent connections from X clients on the
   host to an X server on the workstation.  The DISPLAY variable is sent
   inline during the TELNET negotiations as

     DISPLAY=<local-ip-addr>:<server>.<display>

   where the <local-ip-addr> is retrieved by looking at the local ip
   address associated with the socket used to connect to <server>.  The
   <server> determines which port (6000 + <server>) should be used to
   make the connection.  <display> is used to indicate which monitor
   attached to the X server should be used but is not important to this
   discussion.

   The <local-ip-addr> used is not a DNS name because:

    . The is no ability for the local machine to know its DNS name
      without performing a reverse DNS lookup on the local-ip-addr

    . There is no guarantee that the name returned by a reverse
      DNS lookup actually maps back to the local IP address.

    . Lastly, without DNSSEC, it may not be safe to use DNS addresses
      because they can easily be spoofed.  NAT and DNS-ALG cannot work
      unless DNSSEC is disabled.

   A common use of this application is people dialing in to corporate
   offices from their X terminals at home.  Say, the X client is running
   on a host on the public side of the NAT and X server is running on a

   host on the private side of the NAT.  The DISPLAY variable is
   transmitted inline to the host the X client is running in some way.
   The process transmitting the contents of the DISPLAY variable does
   not know the address of the NAT.

   If the channel transmitting the DISPLAY variable is not encrypted,
   NAT device might solicit the help of an ALG to replace the IP address
   and configure a port in the valid display range (ports 6000 and
   higher) to act as a gateway.  Alternately, NAT may be configured to
   listen for incoming connections to provide access to the X Server(s),
   without requiring an ALG.  But, this approach increases the security
   risk by providing access to the X server that would not otherwise be
   available.  As the ALG tampers with the IP addresses it will also not
   be possible for X Authorization methods other than MIT-MAGIC-COOKIE-1
   to be used.  MIT-MAGIC-COOKIE-1 is the least secure of all the
   documented X Authorization methods.

   When START_TLS is used there may be client certificate verification
   problems caused by the NAT depending on the information provided in
   the certificate.

3.5 RSH/RLOGIN

   RSH uses multiple sessions to support separate streams for stdout and
   stderr.  A random port number is transmitted inline from the client
   to the server for use as stderr port.  The stderr socket is a
   connection back from the server to the client.  And unlike FTP, there
   is no equivalent to PASV mode.  For traditional NAT, this is a
   problem as traditional NAT would not permit incoming sessions.

   RLOGIN does not use multiple sessions.  But the Kerberos protected
   versions of RSH and RLOGIN will not work in a NAT environment due to
   the ticket problems and the use of multiple sessions.

4.0 Protocols that can work with the aid of an ALG

This document predominantly addresses problems associated with
Traditional NAT, especially NAPT.

4.1 FTP

   FTP is a TCP based application, used to reliably transfer files
   between two hosts.  FTP uses bundled session approach to accomplish
   this.

   FTP is initiated by a client accessing a well-known port number 21 on
   the FTP server.  This is called the FTP control session.  Often, an
   additional data session accompanies the control session.  By default,

   the data session would be from TCP port 20 on server to the TCP port
   client used to initiate control session.  However, the data session
   ports may be altered within the FTP control sessions using ASCII
   encoded PORT and PASV commands and responses.

   Say, an FTP client is in a NAT supported private network.  An FTP ALG
   will be required to monitor the FTP control session (for both PORT
   and PASV modes) to identify the FTP data session port numbers and
   modify the private address and port number with the externally valid
   address and port number.  In addition, the sequence and
   acknowledgement numbers, TCP checksum, IP packet length and checksum
   have to be updated.  Consequently the sequence numbers in all
   subsequent packets for that stream must be adjusted as well as TCP
   ACK fields and checksums.

   In rare cases, increasing the size of the packet could cause it to
   exceed the MTU of a given transport link.  The packet would then have
   to be fragmented which could affect performance.  Or, if the packet
   has the DF bit set, it would be ICMP rejected and the originating
   host would then have to perform Path MTU Discovery.  This could have
   an adverse effect on performance.

   Note however, if the control command channel is secured, it will be
   impossible for an ALG to update the IP addresses in the command
   exchange.

   When AUTH is used with Kerberos 4, Kerberos 5, and TLS, the same
   problems that occur with X-Term/Telnet occur with FTP.

   Lastly, it is of interest to note section 4 of RFC 2428 (FTP
   extensions for IPv6 and NATs) which describes how a new FTP port
   command (EPSV ALL) can be used to allow NAT devices to fast-track the
   FTP protocol, eliminating further processing through ALG, if the
   remote server accepts "EPSV ALL".

4.2 RSVP

   RSVP is positioned in the protocol stack at the transport layer,
   operating on top of IP (either IPv4 or IPv6).  However, unlike other
   transport protocols, RSVP does not transport application data but
   instead acts like other Internet control protocols (for example,
   ICMP, IGMP, routing protocols).  RSVP messages are sent hop-by-hop
   between RSVP-capable routers as raw IP datagrams using protocol
   number 46.  It is intended that raw IP datagrams should be used
   between the end systems and the first (or last) hop router.  However,
   this may not always be possible as not all systems can do raw network
   I/O.  Because of this, it is possible to encapsulate RSVP messages
   within UDP datagrams for end-system communication.  UDP-encapsulated

   RSVP messages are sent to either port 1698 (if sent by an end system)
   or port 1699 (if sent by an RSVP-enabled router).  For more
   information concerning UDP encapsulation of RSVP messages; consult
   Appendix C of RFC 2205.

   An RSVP session, a data flow with a particular destination and
   transport-layer protocol, is defined by:

   Destination Address - the destination IP address for the data
   packets.  This may be either a unicast or a multicast address.

   Protocol ID - the IP protocol ID (for example, UDP or TCP).

   Destination Port - a generalized destination port that is used for
   demultiplexing at a layer above the IP layer.

   NAT devices are presented with unique problems when it comes to
   supporting RSVP.  Two issues are:

   1. RSVP message objects may contain IP addresses.  The result is that
   an RSVP-ALG must be able to replace the IP addresses based upon the
   direction and type of the message.  For example, if an external
   sender were to send an RSVP Path message to an internal receiver, the
   RSVP session will specify the IP address that the external sender
   believes is the IP address of the internal receiver.  However, when
   the RSVP Path message reaches the NAT device, the RSVP session must
   be changed to reflect the IP address that is used internally for the
   receiver.  Similar actions must be taken for all message objects that
   contain IP addresses.

   2. RSVP provides a means, the RSVP Integrity object, to guarantee the
   integrity of RSVP messages.  The problem is that because of the first
   point, a NAT device must be able to change IP addresses within the
   RSVP messages.  However, when this is done, the RSVP Integrity object
   is no longer valid as the RSVP message has been changed.  Therefore
   an RSVP-ALG will not work when RSVP Integrity Object is used.

4.3 DNS

   DNS is a TCP/UDP based protocol.  Domain Names are an issue for hosts
   which use local DNS servers in NAT private realm.  DNS name to
   address mapping for hosts in private domain should be configured on
   an authoritative name server within private domain.  This server
   would be accessed by external and internal hosts alike for name
   resolutions.  A DNS-ALG would be required to perform address to name
   conversions on DNS queries and responses.  [DNS-ALG] describes DNS-
   ALG

   in detail.  If DNS packets are encrypted/authenticated per DNSSEC,
   then DNS_ALG will fail because it won't be able to perform payload
   modifications.

   Applications using DNS resolver to resolve a DNS name into an IP
   address, assume availability of address assignment for reuse by the
   application specific session.  As a result, DNS-ALG will be required
   to keep the address assignment (between private and external
   addresses) valid for a pre-configured period of time, past the DNS
   query.

   Alternately, if there isn't a need for a name server within private
   domain, private domain hosts could simply point to an external name
   server for external name lookup.  No ALG is required when the name
   server is located in external domain.

4.4 SMTP

   SMTP is a TCP based protocol ([SMTP]), used by Internet email
   programs such as sendmail to send TCP-based email messages to well-
   known port 25.  The mail server may be located within or outside
   private domain.  But, the server must be assigned a global name and
   address, accessible by external hosts.  When mail server is located
   within private domain, inbound SMTP sessions must be redirected to
   the private host from its externally assigned address.  No special
   mapping is required when Mail server is located in external domain.

   Generally speaking, mail systems are configured such that all users
   specify a single centralized address (such as fooboo@company.com),
   instead of including individual hosts (such as
   fooboo@hostA.company.com).  The central address must have an MX
   record specified in the DNS name server accessible by external hosts.

   In the majority of cases, mail messages do not contain reference to
   private IP addresses or links to content data via names that are not
   visible to outside.  However, some mail messages do contain IP
   addresses of the MTAs that relay the message in the "Received: "
   field.  Some mail messages use IP addresses in place of FQDN for
   debug purposes or due to lack of a DNS record, in "Mail From: "
   field.

   If one or more MTAs were to be located behind NAT in a private
   domain, and the mail messages are not secured by signature or
   cryptographic keys, an SMTP-ALG may be used to translate the IP
   address information registered by the MTAs.  If the MTAs have static
   address mapping, the translation would be valid across realms for
   long periods of time.

   The ability to trace the mail route may be hampered or prevented by
   NAT alone, without the ALG.  This can cause problems when debugging
   mail problems or tracking down abusive users of mail.

4.5 SIP

   SIP (Refer [SIP]) can run on either TCP or UDP, but by default on the
   same port 5060.

   When used with UDP, a response to a SIP request does not go to the
   source port the request came from.  Rather the SIP message contains
   the port number the response should be sent to.  SIP makes use of
   ICMP port unreachable errors in the response to request
   transmissions.  Request messages are usually sent on the connected
   socket.  If responses are sent to the source port in the request,
   each thread handling a request would have to listen on the socket it
   sent the request on.  However, by allowing responses to come to a
   single port, a single thread can be used for listening instead.

   A server may prefer to place the source port of each connected socket
   in the message.  Then each thread can listen for responses
   separately.  Since the port number for a response may not go to the
   source port of the request, SIP will not normally traverse a NAT and
   would require a SIP-ALG.

   SIP messages carry arbitrary content, which is defined by a MIME
   type.  For multimedia sessions, this is usually the Session
   Description Protocol (SDP RFC 2327).  SDP may specify IP addresses or
   ports to be used for the exchange of multimedia.  These may loose
   significance when traversing a NAT.  Thus a SIP-ALG would need the
   intelligence to decipher and translate realm-relevant information.

   SIP carries URL's in its Contact, To and From fields that specify
   signaling addresses.  These URL's can contain IP addresses or domain
   names in the host port portion of the URL.  These may not be valid
   once they traverse a NAT.

   As an alternative to an SIP-ALG, SIP supports a proxy server which
   could co-reside with NAT and function on the globally significant NAT
   port.  Such a proxy would have a locally specific configuration.

4.6 RealAudio

   In default mode, RealAudio clients (say, in a private domain) access
   TCP port 7070 to initiate conversation with a real-audio server (say,
   located an external domain) and to exchange control messages during
   playback (ex: pausing or stopping the audio stream).  Audio session
   parameters are embedded in the TCP control session as byte stream.

   The actual audio traffic is carried in the opposite direction on
   incoming UDP based packets (originated from the server) directed to
   ports in the range of 6970-7170.

   As a result, RealAudio is broken by default on a traditional NAT
   device.  A work around for this would be for the ALG to examine the
   TCP traffic to determine the audio session parameters and selectively
   enable inbound UDP sessions for the ports agreed upon in the TCP
   control session.  Alternately, the ALG could simply redirect all
   inbound UDP sessions directed to ports 6970-7170 to the client
   address in the private domain.

   For bi-Directional NAT, you will not need an ALG.  Bi-directional NAT
   could simply treat each of the TCP and UDP sessions as 2 unrelated
   sessions and perform IP and TCP/UDP header level translations.

   The readers may contact RealNetworks for detailed guidelines on how
   their applications can be made to work, traversing through NAT and
   firewall devices.

4.7 H.323

H.323 is complex, uses dynamic ports, and includes multiple UDP
streams. Here is a summary of the relevant issues:

   An H.323 call is made up of many different simultaneous connections.
   At least two of the connections are TCP.  For an audio-only
   conference, there may be up to 4 different UDP 'connections' made.

   All connections except one are made to ephemeral (dynamic) ports.

   Calls can be initiated from the private as well as the external
   domain.  For conferencing to be useful, external users need to be
   able to establish calls directly with internal users' desktop
   systems.

   The addresses and port numbers are exchanged within the data stream
   of the 'next higher' connection.  For example, the port number for
   the H.245 connection is established within the Q.931 data stream.
   (This makes it particularly difficult for the ALG, which will be
   required to modify the addresses inside these data streams.)  To make
   matters worse, it is possible in Q.931, for example, to specify that
   the H.245 connection should be secure (encrypted).  If a session is
   encrypted, it is impossible for the ALG to decipher the data stream,
   unless it has access to the shared key.

   Most of the control information is encoded in ASN.1 (only the User-
   User Information within Q.931 Protocol Data Units, or PDUs, is

   ASN.1-encoded (other parts of each Q.931 PDU are not encoded).  For
   those unfamiliar with ASN.1, suffice it to say that it is a complex
   encoding scheme, which does not end up with fixed byte offsets for
   address information.  In fact, the same version of the same
   application connecting to the same destination may negotiate to
   include different options, changing the byte offsets.

   Below is the protocol exchange for a typical H.323 call between User
   A and User B.  A's IP address is 88.88.88.88 and B's IP address is
   99.99.99.99.  Note that the Q.931 and H.245 messages are encoded in
   ASN.1 in the payload of an RTP packet.  So to accomplish a connection
   through a NAT device, an H.323-ALG will be required to examine the
   packet, decode the ASN.1, and translate the various H.323 control IP
   addresses.

   User A                                                  User B
         A establishes connection to B on well-
         known Q.931 port (1720)

         ----------------------------------------------->
         Q.931 Setup caller address = 88.88.88.88
                     caller port    = 1120
                     callee address = 99.99.99.99
                     callee port    = 1720
         <-----------------------------------------------
         Q.931 Alerting
         <-----------------------------------------------
         Q.931 Connect H.245 address = 99.99.99.99
                       H.245 port    = 1092

         User A establishes connection to User B at
         99.99.99.99, port 1092

         <---------------------------------------------->
         Several H.245 messages are exchanged (Terminal
         Capability Set, Master Slave Determination and
         their respective ACKs)

         <-----------------------------------------------
         H.245 Open Logical Channel, channel = 257
                   RTCP address = 99.99.99.99
                   RTCP port    = 1093
         ----------------------------------------------->
         H.245 Open Logical Channel Ack, channel = 257
                   RTP address = 88.88.88.88
                   RTP port    = 2002
                   (This is where User A would like RTP
                    data sent to)

                   RTCP address = 88.88.88.88
                   RTCP port    = 2003
         ----------------------------------------------->
         H.245 Open Logical Channel, channel = 257
                   RTCP address = 88.88.88.88
                   RTCP port    = 2003
         <-----------------------------------------------
         H.245 Open Logical Channel Ack, channel = 257
                   RTP address = 99.99.99.99
                   RTP port    = 1092
                   (This is where User B would like RTP data
                    sent to)
                   RTCP address = 99.99.99.99
                   RTP port     = 1093

   Also note that if an H.323 Gateway resided inside a NAT boundary, the
   ALG would have to be cognizant of the various gateway discovery
   schemes and adapt to those schemes as well.  Or if just the H.323
   host/terminal was inside the NAT boundary and tried to register with
   a Gatekeeper, the IP information in the registration messages would
   have to be translated by NAT.

4.8 SNMP

   SNMP is a network management protocol based on UDP.  SNMP payload may
   contain IP addresses or may refer IP addresses through an index into
   a table.  As a result, when devices within a private network are
   managed by an external node, SNMP packets transiting a NAT device may
   contain information that is not relevant in external domain.  In some
   cases, as described in [SNMP-ALG], an SNMP ALG may be used to
   transparently convert realm-specific addresses into globally unique
   addresses.  Such an ALG assumes static address mapping and bi-
   directional NAT.  It can only work for the set of data types (textual
   conventions) understood by the SNMP-ALG implementation and for a
   given set of MIB modules.  Furthermore, replacing IP addresses in the
   SNMP payload may lead to communication failures due to changes in
   message size or changes in the lexicographic ordering.

   Making SNMP ALGs completely transparent to all management
   applications is not an achievable task.  The ALGs will run into
   problems with SNMPv3 security features, when authentication (and
   optionally privacy) is enabled, unless the ALG has access to security
   keys.  [NAT-ARCH] also hints at potential issues with SNMP management
   via NAT.

   Alternately,  SNMP proxies, as defined in [SNMP-APPL], may be used in
   conjunction with NAT to forward SNMP messages to external SNMP
   engines (and vice versa).  SNMP proxies are tailored to the private

   domain context and can hence operate independent of the specific
   managed object types being accessed.  The proxy solution will require
   the external management application to be aware of the proxy
   forwarder and the individual nodes being managed will need to be
   configured to direct their SNMP traffic (notifications and requests)
   to the proxy forwarder.

5.0 Protocols designed explicitly to work with NAT enroute

5.1 Activision Games

   Activision Games were designed to be NAT-friendly so as not to
   require an ALG for the games to work transparently through
   traditional NAT devices.  Game players within a private domain can
   play with other players in the same domain or external domain.
   Activision gaming protocol is proprietary and is based on UDP.  The
   address server uses UDP port number 21157 and is expected to be
   located in the global address realm.

   Game players connect to the address server first, and send their
   private IP address information (such as private IP address and UDP
   port number) in the initial connect message.  The server notes
   private address information from the connect message and external
   address information from the IP and UDP headers.  The server then
   sends both the private and external address information of the game
   player to all the other peer players.  At this point, each game
   player knows the private and public address information of every
   other peer.  Subsequent to this, each client opens up symmetrical
   direct connection to each other and uses whichever address (private
   or external) works first.

   Now, the clients can have a session directly with other clients (or)
   they can have session with other clients via the gaming server.  The
   key is to allow reuse of the same (global address, assigned UDP port)
   tuple used for initial connection to the game server for all
   subsequent connections to the client.  A game player is recognized by
   one of (private address, UDP port) or (global address, assigned UDP
   port) by all other peer players.  So, the binding between tuples
   should remain unchanged on NAT, so long as the gaming player is in
   session with one or multiple other players.

   Opening a connection to a game server in external realm from a
   private host is no problem.  All NAT would have to do is provide
   routing transparency and retain the same private-to-external address
   binding so long as there is a minimum of one gaming session with an
   external node.  But, an NAPT configuration must allow multiple
   simultaneous UDP connections on the same assigned global
   address/port.

   The above approach has some problems.  For example, a client could
   try contacting a private address, but that private address could be
   in use locally, when the private address at some other realm is
   meant. If the node that was contacted wrongfully has some other
   service or no service registered for the UDP port, the Activision
   connect messages are expected to be simply dropped.  In the unlikely
   event, a registered application chooses to interpret the message, the
   results can be unpredictable.

   The readers may refer to Activision for the proprietary, detailed
   information on the function and design of this protocol.

6.0 Acknowledgements

   The authors would like to express sincere thanks to Bernard Aboba,
   Bill Sommerfield, Dave Cridland, Greg Hudson, Henning Schulzrine,
   Jeffrey Altman, Keith Moore, Thomas Narten, Vernon Shryver and others
   that had provided valuable input in preparing this document.  Special
   thanks to Dan Kegel for sharing the Activision games design
   methodology.

7.0 Security Considerations

   The security considerations outlined in [NAT-TERM] are relevant to
   all NAT devices.  This document does not warrant additional security
   considerations.

8.0 References

   [NAT-TERM]   Srisuresh, P. and M. Holdrege, "IP Network Address
                Translator (NAT) Terminology and Considerations", RFC
                2663, August 1999.

   [NAT-TRAD]   Srisuresh, P. and K. Egevang, "Traditional IP Network
                Address Translator (Traditional NAT)", RFC 3022, January
                2001.

   [H.323]      ITU-T SG16 H.323, Intel white paper, "H.323 and
                Firewalls", Dave Chouinard, John Richardson, Milind
                Khare (with further assistance from Jamie Jason).

   [SNMP-ALG]   Raz, D., Schoenwaelder, J. and B. Sugla, "An SNMP
                Application Level Gateway for Payload Address
                Translation", RFC 2962, October 2000.

   [SNMP-APPL]  Levi, D., Meyer, P. and B. Stewart, "SNMP Applications",
                RFC 2573, April 1999.

   [NAT-ARCH]   Hain, T. "Architectural Implications of NAT", RFC 2993,
                November 2000.

   [SMTP]       Postel, J., "Simple Mail Transfer Protocol", STDl 10,
                RFC 821, August 1982.

   [FTP]        Postel, J. and J. Reynolds, "File Transfer Protocol
                (FTP)", STD 9, RFC 959, October 1985.

   [SIP]        Handley, M., Schulzrinne, H., Schooler, E. and J.
                Rosenberg, "SIP: Session Initiation Protocol", RFC 2543,
                March 1999.

   [X Windows]  Scheifler, B., "FYI on the X Window System", FYI 6, RFC
                1198, January 1991.

   [RSVP]       Braden, R., Zhang. L., Berson. S., Herzog, S. and S.
                Jamin, "Resource ReSerVation Protocol (RSVP) -- Version
                1 Functional Specification", RFC 2205, September 1997.

   [DNS-TERMS]  Mockapetris, P., "Domain Names - Concepts and
                Facilities", STD 13, RFC 1034, November 1987.

   [DNS-IMPL]   Mockapetris, P., "Domain Names - Implementation and
                Specification", STD 13, RFC 1035, November 1987.

   [DNS-ALG]    Srisuresh, P., Tsirtsis, G., Akkiraju, P. and A.
                Heffernan, "DNS extensions to Network Address
                Translators (DNS_ALG)", RFC 2694, September 1999.

   [IPsec]      Kent, S. and R. Atkinson, "Security Architecture for the
                Internet Protocol", RFC 2401, November 1998.

   [IPsec-ESP]  Kent, S. and R. Atkinson, "IP Encapsulating Security
                Payload (ESP)", RFC 2406, November 1998.

   [IPsec-AH]   Kent, S. and R. Atkinson, "IP Authentication Header",
                RFC 2402, November 1998.

   [IPsec-DOCS] Thayer, R., Doraswamy, N. and R. Glenn, "IP Security
                Document Roadmap", RFC 2411, November 1998.

   [NAT-SEC]    Srisuresh, P., "Security Model with Tunnel-mode IPsec
                for NAT Domains", RFC 2709, October 1999.

   [PRIV-ADDR]  Rekhter, Y., Moskowitz, B., Karrenberg, D., G. de Groot,
                and E. Lear, "Address Allocation for Private Internets",
                BCP 5, RFC 1918, February 1996.

[ADDR-BEHA] Carpenter, B., Crowcroft, J. and Y. Rekhter, "IPv4
Address Behaviour Today", RFC 2101, February 1997.

Authors' Addresses

   Matt Holdrege
   ipVerse
   223 Ximeno Ave.
   Long Beach, CA 90803

   EMail: matt@ipverse.com

   Pyda Srisuresh
   Jasmine Networks, Inc.
   3061 Zanker Road, Suite B
   San Jose, CA 95134
   U.S.A.

   Phone: (408) 895-5032
   EMail: srisuresh@yahoo.com

Full Copyright Statement

   Copyright (C) The Internet Society (2001).  All Rights Reserved.

   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implementation may be prepared, copied, published
   and distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph are
   included on all such copies and derivative works.  However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than
   English.

   The limited permissions granted above are perpetual and will not be
   revoked by the Internet Society or its successors or assigns.

   This document and the information contained herein is provided on an
   "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
   TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
   BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
   HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Acknowledgement

Funding for the RFC Editor function is currently provided by the
Internet Society.