Network Working Group                                           A. DeKok
INTERNET-DRAFT                                                FreeRADIUS
Category: Proposed Standard
Expires: June 11, 2009
11 December 2008

                            RADIUS Over TCP

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   The list of current Internet-Drafts can be accessed at

   The list of Internet-Draft Shadow Directories can be accessed at

   This Internet-Draft will expire on June 11, 2009.

Copyright Notice

   This Internet-Draft is submitted to IETF in full conformance with the
   provisions of BCP 78 and BCP 79.

   Copyright (c) 2008 IETF Trust and the persons identified as the
   document authors.  All rights reserved.  This document is subject to
   BCP 78 and the IETF Trust's Legal Provisions Relating to IETF
   Documents ( in effect on the
   date of publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.


   The Remote Authentication Dial In User Server (RADIUS) Protocol has
   traditionally used the User Datagram Protocol (UDP) as it's
   underlying transport layer.  This document defines RADIUS over the
   Transport Control Protocol (TCP).

DeKok, Alan                 Proposed Standard                   [Page 1]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

Table of Contents

1.  Introduction .............................................    3
   1.1.  Benefits of Reliable Transport ......................    3
   1.2.  Drawbacks of Reliable Transport .....................    4
   1.3.  Terminology .........................................    4
   1.4.  Requirements Language ...............................    5
2.  Changes to RADIUS ........................................    5
   2.1.  Packet Format .......................................    5
   2.2.  TCP Ports ...........................................    5
   2.3.  Management Information Base (MIB) ...................    6
   2.4.  Interaction with RadSec .............................    6
      2.4.1.  Applicability ..................................    6
   2.5.  RADIUS Proxies ......................................    7
   2.6.  TCP Specific Issues .................................    7
      2.6.1.  Duplicates and Retransmissions .................    8
      2.6.2.  Shared Secrets .................................    9
      2.6.3.  Malformed Packets and Unknown Clients ..........    9
      2.6.4.  Limits of the ID Field .........................   10
      2.6.5.  EAP Sessions ...................................   10
      2.6.6.  TCP Applications are not UDP Applications ......   11
3.  Diameter Considerations ..................................   11
4.  IANA Considerations ......................................   11
5.  Security Considerations ..................................   11
6.  References ...............................................   12
   6.1.  Normative References ................................   12
   6.2.  Informative References ..............................   12

DeKok, Alan                 Proposed Standard                   [Page 2]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

1.  Introduction

   The RADIUS Protocol has been defined in [RFC2865] as using the User
   Datagram Protocol (UDP) for the underlying transport layer.  While
   there are a number of benefits to using UDP as outlined in [RFC2865]
   Section 2.4, there are also some limitations:

      * Unreliable transport.  As a result, systems using RADIUS have to
      implement application-layer timers and re-transmissions, as
      described in [RFC5080] Section 2.2.1.

      * Packet fragmentation.  [RFC2865] Section 3 permits RADIUS
      packets to up to 4096 octets in length.  These packets are larger
      than the default Internet MTU (576), resulting in fragmentation of
      the packets at the IP layer.  Transport of fragmented UDP packets
      appearsto be a poorly tested code path on network devices.  Some
      devices appear to be incapable of transporting fragmented UDP
      packets, making it difficult to deploy RADIUS in a network where
      those devices are deployed.

      * Connectionless transport.  Neither clients no servers can
      reliably detect when the other is down.  This information has be
      deduced from the absence of a reply to a request.

   As RADIUS is widely deployed, and has been widely deployed for well
   over a decade, these issues are relatively minor.  However, new
   systems may be interested in choosing a different set of trade-offs
   than those outlined in [RFC2865] Section 2.4.  For those systems, we
   define RADIUS over TCP.

1.1.  Benefits of Reliable Transport

   There are a number of benefits to using a reliable transport.  For
   example, when RADIUs is used to carry EAP conversions [RFC3579], the
   EAP exchanges may involve 10 round trips at the RADIUS application
   layer.  If we assume a 0.1% probability of packet loss in each
   direction, then approximately 2% (1 - 0.999^20) of the authentication
   attempts will have a lost packet.  If we assume a 0.01% packet loss,
   then 0.2% of authentication attempts will result in a lost packet.

   These lost packets require the supplicant and/or the NAS to re-
   transmit packets at the application layer.  The difficulty with this
   approach is that retransmission implementations have historically
   been poor.  Some implementations retransmit packets, others do not.
   Some implementations are incapable of detecting EAP retransmissions,
   and will instead treat the retransmitted packet as an error.

   These retransmissions have a high likelihood of causing the entire

DeKok, Alan                 Proposed Standard                   [Page 3]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

   authentication session to fail.  For systems with millions to tens of
   millions of users, such a high authentication failure rate (0.2% to
   2%) may be unacceptable.

   Using TCP as an underlying reliable transport means that the RADIUS
   implementations can remove all of the application-layer
   retransmissions, and instead rely on the Operating System (OS)
   kernel's well-tested TCP transport.

1.2.  Drawbacks of Reliable Transport

   No protocol is perfect for all uses.  RADIUS over TCP has some
   drawbacks, as noted in [RFC2865] Section 2.4.  [RFC3539] Section 2
   discusses further issues with using TCP as a transport for
   Authentication, Authorization, and/or Accounting (AAA) protocols such
   as RADIUS.

   The impact of these issues is dicussed in more detail, below.

1.3.  Terminology

   This document uses the following terms:

Network Access Server (NAS)
     A device that provides an access service for a user to a network.

RADIUS server
     A RADIUS authentication, authorization, and/or accounting (AAA)
     server is an entity that provides one or more AAA services to a

RADIUS proxy
     A RADIUS proxy acts as a RADIUS server to the NAS, and a RADIUS
     client to the RADIUS server.

RADIUS request packet
     A packet originated by a RADIUS client to a RADIUS server.  e.g.
     Access-Request, Accounting-Request, CoA-Request, or Disconnect-

RADIUS response packet
     A packet sent by a RADIUS server to a RADIUS client, in response to
     a RADIUS request packet.  e.g. Access-Accept, Access-Reject,
     Access-Challenge, Accounting-Response, CoA-ACK, etc.

DeKok, Alan                 Proposed Standard                   [Page 4]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

1.4.  Requirements Language

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   document are to be interpreted as described in [RFC2119].

2.  Changes to RADIUS

   Adding TCP as a RADIUS transport has a number of impacts on the
   protocol, on applications using the protocol, and on networks that
   deploy the protocol.  This section outlines those impacts, and
   defines behaviors.

2.1.  Packet Format

   The RADIUS packet format is unchanged from [RFC2865], [RFC2866], and
   [RFC5176].  Specifically, all of the following portions of RADIUS
   MUST be unchanged when using RADIUS over TCP:

      * Packet format
      * Permitted codes
      * Request Authenticator calculation
      * Response Authenticator calculation
      * Minimum packet length
      * Maximum packet length
      * Attribute format
      * Vendor-Specific Attribute (VSA) format
      * Permitted data types
      * Calculations of dynamic attributes such as CHAP-Challenge,
      or Message-Authenticator.

   The changes to RADIUS implementations required to implement this
   specification are largely limited to the code that sends and receives
   packets on the network.

2.2.  TCP Ports

   IANA has already assigned TCP ports for RADIUS transport, as outlined

      * radius          1812/udp
      * radius-acct     1813/tcp
      * radius-dynauth  3799/tcp

   These ports are unused by existing RADIUS applications.
   Implementations SHOULD use the assigned values as the default ports
   for RADIUS over TCP.

DeKok, Alan                 Proposed Standard                   [Page 5]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

   The early deployment of RADIUS was done using UDP port number 1645,
   which conflicts with the "datametrics" service.  Implementations
   using RADIUS over TCP MUST NOT use TCP ports 1645 or 1646 as the
   default ports for this specification.

2.3.  Management Information Base (MIB)

   The MIB definitions in [RFC4668], [RFC4669], [RFC4670], [RFC4671],
   [RFC4672], and [RFC4673] each contain only one reference to UDP.
   These references are in the DESCRIPTION field of the MIB definition,
   and are in the form of "The UDP port" or "the UDP destination port".

   Implementations of RADIUS over TCP MAY re-use these MIBs to perform
   statistics counting for RADIUS over TCP connections.  However,
   implementors are warned that there is no way for these MIBs to
   distinguish between packets sent over UDP or over TCP transport.
   Similarly, there is no requirement in RADIUS that the RADIUS services
   offered over UDP on a particular IP address and port are identical to
   the RADIUS services offered over TCP on a particular IP address and
   the same (numerical) port.

2.4.  Interaction with RadSec

   IANA has already assigned TCP ports for RadSec transport, as outlined

      * radsec          2083/tcp

   This value SHOULD be used as the default port for RADIUS over TLS
   (i.e. RadSec).  The "radius" port (1812/tcp) SHOULD NOT be used for

2.4.1.  Applicability

   As noted in [RFC3539] Section 2.1, for systems originating low
   numbers of RADIUS request packets, inter-packet spacing is often
   larger than the RTT.  In those situations, RADIUS over TCP SHOULD NOT
   be used.

   In general, RADIUS clients generating small amounts of RADIUS traffic
   SHOULD NOT use TCP.  This suggestion will usually apply to most
   NASes, and to most clients that originate CoA-Request and Disconnect-
   Request packets.

   RADIUS over TCP is most applicable to RADIUS proxies that exchange a
   large volume of packets with RADIUS clients and servers (10's to
   1000's of packets per second).  In those situations, RADIUS over TCP
   is a good fit, and may result in increased network stability and

DeKok, Alan                 Proposed Standard                   [Page 6]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008


2.5.  RADIUS Proxies

   As RADIUS is a "hop by hop" protocol, a RADIUS proxy effectively
   shields the client from any information about downstream servers.
   While the client may be able to deduce the operational state of the
   local server (i.e. proxy), it cannot make any determination about the
   operational state of the downstream servers.

   If a request is proxied through intermediate proxies, it is not
   possible to detect which of the later hops is responsible for the
   absence of a reply.  An intermediate proxy also cannot signal that
   the outage lies in a later hop because RADIUS does not have the
   ability to carry such signalling information.  This issue is further
   exacerbated by some proxy implementations that do not reply to a
   client if they do not recieve a reply to a proxied request.

   When UDP was used as a transport protocol, the absence of a reply can
   cause a client to deduce (incorrectly) that the proxy is unavailable.
   The client could then fail over to another server, or conclude that
   no "live" servers are available.  This situation is made even worse
   when requests are sent through a proxy to multiple destinations.
   Failures in one destination may result in service outages for other
   destinations, if the client erroneously believes that the proxy is

   For RADIUS over TCP, the continued existence of the TCP connection
   SHOULD be used to deduce that the service on the other end of the
   connection is still responsive.  Further, the application layer
   watchdog defined in [RFC3539] Section 3.4 enables clients to
   determine that the server is "live", even though it may not have
   responded recently to other, non-watchdog requests.

   RADIUS clients using RADIUS over TCP MUST NOT decide that a
   connection is down until the application layer watchdog algorithm has
   marked it DOWN ([RFC3539] Appendix A).  RADIUS clients using RADIUS
   over TCP MUST NOT decide that a RADIUS server is unresponsive until
   all TCP connections to it have been marked DOWN.

2.6.  TCP Specific Issues

   The guidelines defined in [RFC3539] for implementing an AAA protocol
   operating over a reliable transport MUST be followed by implementors
   of this specification.

   The Application Layer Watchdog defined in [RFC3539] Section 3.4 MUST
   be used.  The Status-Server packet [STATUS] MUST be used as the

DeKok, Alan                 Proposed Standard                   [Page 7]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

   application layer watchdog message.  Implementations MUST reserve one
   RADIUS ID per connection for the application layer watchdog message.
   This restriction is described further below.

   Implementations MUST NOT confuse UDP and TCP transport.  That is,
   RADIUS clients and servers MUST be treated as unique based on a key
   of (IP address, port, transport protocol).  Implementations MUST be
   configurable to have different shared secrets for UDP and TCP to the
   same destination IP address and numerical port.

   This requirement does not forbid the traditional practice of using
   primary and secondary servers in a fail-over relationship.  Instead,
   it requires that two services sharing an IP address and numerical
   port, but differing in transport protocol, MUST be treated as
   independent services for the purpose of fail-over, load-balancing,

   Whenever the underlying operating system permits the use of TCP
   keepalive socket options, their use is RECOMMENDED.

2.6.1.  Duplicates and Retransmissions

   As TCP is a reliable transport, implementors of this specification
   MUST NOT retransmit RADIUS packets over the same TCP connection.
   Similarly, if there is no response to a RADIUS packet over one TCP
   connection, implementations MUST NOT retransmit that packet over a
   different TCP connection to the same destination IP address and port.

   However, if the TCP connection is broken or closed, the above
   requirement can be relaxed somewhat.  RADIUS request packets that
   have not yet received a response MAY be transmitted by a RADIUS
   client over a new TCP connection.  As this procedure involves using a
   new source port, the ID of the packet MAY change.  If the ID changes,
   any security attributes such as Message-Authenticator MUST be

   If a TCP connection is broken or closed, any cached RADIUS response
   packets ([RFC5080] Section 2.2.2) associated with that connection
   MUST be discarded.  A RADIUS server SHOULD stop processing any "live"
   requests associated with that TCP connection.  No response to these
   requests cannot be sent over the TCP connection, so any further
   processing is pointless.  A RADIUS proxy that has a client close it's
   TCP connection SHOULD silently discard any responses it recieves to a
   proxied requests that is associated with the original client request.

   Despite the above requirement, RADIUS servers SHOULD still perform
   duplicate detection on received packets, as described in [RFC5080]
   Section 2.2.2.  This effort can prevent duplicate processing of

DeKok, Alan                 Proposed Standard                   [Page 8]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

   packets from non-conformant clients.

   As noted above, RADIUS packets SHOULD NOT be re-transmitted to the
   same destination IP and numerical port, but over a different
   transport layer.  There is no guarantee in RADIUS that the two ports
   are in any way related.  This requirement does not forbid the
   practice of putting multiple servers into a fail-over or load-balance

   Much of the discussion in this section can be summarized by the
   following requirement.  RADIUS requests MAY be re-transmitted
   verbatim only if the following 5-tuple (Client IP address, Client
   port, Transport Protocol, Server IP address, Server port) remains the
   same.  If any field of that 5-typle changes, the packet MUST NOT be
   considered to be a re-transmission.  Instead, the packet MUST be
   considered to be a new request, and be treated accordingly.  (e.g.
   header calculations, packet signatures, associated timers and
   counters, etc.)

   The above requirement is necessary, but not sufficient in all cases.
   Other specifications give additional situations where the packet is
   to be considered as a new request.  Those recommendations MUST be

2.6.2.  Shared Secrets

   The use of shared secrets in calculating the Response Authenticator,
   and other attributes such as User-Password or Message-Authenticator
   [RFC3579] MUST be unchanged from previous specifications.

   Clients and servers MUST be able to store and manage shared secrets
   based on the key described above, of (IP address, port, transport).

2.6.3.  Malformed Packets and Unknown Clients

   The original specifications say that an implement should "silently
   discard" a packet in a number of circumstances.  This action has no
   further consequences for UDP transport, as the "next" packet is
   completely independent of the previous one.

   When TCP is used as a transport, decoding the "next" packet on a
   connection depends on the proper decoding of the previous packet.  As
   a result, the behavior with respect to discarded packets has to

   Implementations of this specification SHOULD treat the "silently
   discard" texts referenced above as "silently discard and close the
   connection."  Specifically, the TCP connection MUST be closed if any

DeKok, Alan                 Proposed Standard                   [Page 9]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

   of the following circumstances are seen:

      * Packet from an unknown client (using the key as defined above)
      * Packet with an invalid code field
      * Packet that is less than the minimim RADIUS packet length
      * Packet that is more than the minimim RADIUS packet length
      * A packet that is otherwise malformed, e.g. Attribute Length of
        0 or 1
      * Packet where the Request Authenticator fails validation
        (if applicable)
      * Packet where the Response Authenticator fails validation
      * Packet where the Message-Authenticator fails validation
      * Response packets that do not match any outstanding request

   These requirements minimize the possibilty for a misbehaving client
   or server to wreak havoc on the network.

2.6.4.  Limits of the ID Field

   The RADIUS ID field is one octet in size.  As a result, any one TCP
   connection can have only 256 "in flight" RADIUS packets at a time.
   If more than 256 simultaneous "in flight" packets are required,
   additional TCP connections will need to be opened.  This limitation
   is also noted in [RFC3539] Section 2.4.

   An additional limit is the requirement to send a Status-Server packet
   over the same TCP connection as is used for normal requests.  As
   noted in [STATUS], the response to a Status-Server packet is either
   an Access-Accept, or an Accounting-Response.  If all IDs were
   allocated to normal requests, then there would be no free Id to use
   for the Status-Server packet, and it could not be sent over the

   Implementations SHOULD reserve ID zero on each TCP connection for
   Status-Server packets.  This value was picked arbitrarily, as there
   is no reason to choose any one value over another for this use.

   It is tempting to extend RADIUS to permit more than 256 outstanding
   packets on one connection.  However, doing so will likely require
   fundamental changes to the RADIUS protocol, and as such, are outside
   of the scope of this specification.

2.6.5.  EAP Sessions

   When RADIUS clients send EAP requests using RADIUS over TCP, they
   SHOULD choose the same TCP connection for all packets related to one
   EAP conversation.  A simple method that may often work is hashing the
   contents of the Calling-Station-Id attribute, which normally contains

DeKok, Alan                 Proposed Standard                  [Page 10]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

   the MAC accress.  The output of that hash can be used to select a
   particular TCP connection.

   It may be difficult to implement this suggestion in practice, as busy
   servers may allocate all RADIUS IDs in one TCP connection in the time
   between two subsequent EAP packets.  It is difficult to suggest
   simple and reasonable methods to address this issue.

2.6.6.  TCP Applications are not UDP Applications

   Implementors should be aware that programming a robust TCP
   application can be a very different process than programming a robust
   UDP application.  We RECOMMEND that implementors of this
   specification familiarize themselves with TCP application programming
   concepts.  We RECOMMEND also that existing TCP applications be
   examined with an eye to robustness, performance, scalability, etc.

   Clients and servers SHOULD implement configurable connection limits.
   Allowing an unlimited number of connections may result in resource

   Further discussion of implementation issues is outside of the scope
   of this document.

3.  Diameter Considerations

   This document defines TCP as a transport layer for RADIUS.  It
   defines no new RADIUS attributes or codes.  The only interaction with
   Diameter is in a RADIUS to Diameter, or in a Diameter to RADIUS
   gateway.  The RADIUS side of such a gateway MAY implement RADIUS over
   TCP, but this change has no effect on Diameter.

4.  IANA Considerations

   This document requires no action by IANA.

5.  Security Considerations

   As the RADIUS packet format, signing, and client verification are
   unchanged from prior specifications, all of the security issues
   outlined in previous specifications for RADIUS over UDP are also
   applicable here.

   As noted above, clients and servers SHOULD support configurable
   connection limits.  Allowing an unlimited number of connections may
   result in resource exhaustion.

   There are no (at this time) other known security issues for RADIUS

DeKok, Alan                 Proposed Standard                  [Page 11]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

   over TCP transport.

6.  References

6.1.  Normative References

[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
          Requirement Levels", BCP 14, RFC 2119, March 1997.

[RFC2865] Rigney, C., Willens, S., Rubens, A., and W. Simpson, "Remote
          Authentication Dial In User Service (RADIUS)", RFC 2865, June

[RFC3539] Aboba, B. et al., "Authentication, Authorization and
          Accounting (AAA) Transport Profile", RFC 3539, June 2003.

6.2.  Informative References

[RFC2866] Rigney, C., "RADIUS Accounting", RFC 2866, June 2000.

[RFC3579] Aboba, B. and P. Calhoun, "RADIUS (Remote Authentication Dial
          In User Service) Support For Extensible Authentication
          Protocol (EAP)", RFC 3579, September 2003.

[RFC4668] Nelson, D, "RADIUS Authentication Client MIB for IPv6", RFC
          4668, August 2006.

[RFC4669] Nelson, D, "RADIUS Authentication Server MIB for IPv6", RFC
          4669, August 2006.

[RFC4670] Nelson, D, "RADIUS Accounting Client MIB for IPv6", RFC 4670,
          August 2006.

[RFC4671] Nelson, D, "RADIUS Accounting Server MIB for IPv6", RFC 4671,
          August 2006.

[RFC4672] Nelson, D, "RADIUS Dynamic Authorization Client MIB", RFC
          4672, August 2006.

[RFC4673] Nelson, D, "RADIUS Dynamic Authorization Server MIB", RFC
          4673, August 2006.

[RFC5080] Nelson, D. and DeKok, A, "Common Remote Authentication Dial In
          User Service (RADIUS) Implementation Issues and Suggested
          Fixes", RFC 5080, December 2007.

[RFC5176] Chiba, M. et al., "Dynamic Authorization Extensions to Remote
          Authentication Dial In User Service (RADIUS)", RFC 5176,

DeKok, Alan                 Proposed Standard                  [Page 12]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

          January 2008.

[STATUS]  DeKok, A., "Use of Status-Server Packets in the Remote
          Authentication Dial In User Service (RADIUS) Protocol", draft-
          ietf-radext-status-server-02.txt, November 2008.


   None at this time.

Authors' Addresses

   Alan DeKok
   The FreeRADIUS Server Project


DeKok, Alan                 Proposed Standard                  [Page 13]

INTERNET-DRAFT          RADIUS Design Guidelines        11 December 2008

Open issues

   Open issues relating to this document are tracked on the following
   web site:

DeKok, Alan                 Proposed Standard                  [Page 14]