TLS Extension for Optimizing Application Protocols, Specifically SASL with GSS-API mechanisms
draft-williams-tls-app-sasl-opt-02.txt

Status of this Memo

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

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-Drafts.

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 http://www.ietf.org/ietf/1id-abstracts.txt.

The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html.

This Internet-Draft will expire on October 16, 2009.

Copyright Notice

Copyright (c) 2009 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 (http://trustee.ietf.org/license-info). Please review these documents carefully, as they describe your rights and restrictions with respect to this document.

Abstract

This document specifies an extension to Transport Layer Security (TLS) for carrying application data which is suitable for delayed integrity protection and does not require privacy protection. In particular we describe how to use this extension to reduce the number of round-trips needed for application-layer authentication, specifically Simple Authentication (SASL), and through it, Generic Security Services (GSS-API). The use of this extension to optimize SASL/GSS-API authentication is termed "TLS/SA".

This extension can also be used to optimize application protocols. Optimizations for Simple Mail transfer Protocol (SMTP) and Light-weight Directory Access Protocol (LDAP) are described.

Table of Contents

1.  Introduction
1.1.  Conventions used in this document
2.  TLS Extensions for Optimization of SASL and Application protocols
3.  Using TLS/SA
3.1.  Optimizing SASL Mechanism Negotiation
3.2.  Optimizing Authentication
3.2.1.  Channel Binding
4.  Non-SASL Optimizations of Existing Application Protocols
4.1.  LDAP
4.2.  SMTP
5.  IANA Considerations
6.  Security Considerations
7.  References
7.1.  Normative References
7.2.  Informative References
§  Author's Address

1.  Introduction

Many applications use TLS [RFC5246] (Dierks, T. and E. Rescorla, “The Transport Layer Security (TLS) Protocol Version 1.2,” August 2008.) and then Simple Authentication and Security Layers (SASL) [RFC4422] (Melnikov, A. and K. Zeilenga, “Simple Authentication and Security Layer (SASL),” June 2006.) on top of TLS. This requires at least two round-trips for TLS, then one round-trip for SASL mechanism negotiation, then as many round-trips as the negotiated SASL mechanism requires. One and a half of the TLS round-trips can carry extensions such that we could piggyback some application data on those TLS messages to save up to two round-trips. This document specifies how to take advantage of TLS extensions to reduce the number of round-trips needed altogether.

First we define a TLS extension for use in Client Hello and Handshake messages. This extension will carry typed application data. Then we describe how to reduce the number of round-trips for SASL applications. And through the new SASL/GSS-API bridge [I‑D.ietf‑sasl‑gs2] (Josefsson, S. and N. Williams, “Using GSS-API Mechanisms in SASL: The GS2 Mechanism Family,” January 2010.) we obtain support for use of GSS-API [RFC2743] (Linn, J., “Generic Security Service Application Program Interface Version 2, Update 1,” January 2000.) mechanisms as well. [RFC2743] (Linn, J., “Generic Security Service Application Program Interface Version 2, Update 1,” January 2000.) applications. Altogether we achieve a one and a half round-trip reduction for SASL applications.

In the case of SASL applications we use the first TLS round-trip to optimize the SASL mechanism negotiation. Then we use the client's handshake message to send the first authentication message of the selected SASL mechanism. Note that the TLS channel binding [RFC5056] (Williams, N., “On the Use of Channel Bindings to Secure Channels,” November 2007.) can be made available at that time, thus no special considerations apply to how channel binding is done (but see Section 3.2.1 (Channel Binding) below). Use of channel binding protects against man-in-the-middle attacks, including downgrade attacks on mechanism negotiation.

This extension is motivated by:

• a desire to reduce the number of round-trips needed by SASL and GSS-API applications running over TLS;
• a desire to replace an earlier proposal for "TLS/GSS" with one that passes muster at the TLS WG;
• a desire to provide a profile that new applications may use for TLS with GSS-API for user authentication.

The use of this extension to optimize SASL/GSS-API authentication is termed "TLS/SA".

1.1.  Conventions used in this document

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119] (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

2.  TLS Extensions for Optimization of SASL and Application protocols

When a client application wishes to exchange one or more application messages prior to the conclusion of a TLS exchange it uses the TLS Client Hello message extension to a) indicate its intention to the server, and b) optionally send the first application message to the server. These messages will not have any privacy or integrity protection applied by TLS unless a ChangeCipherSpec has been done earlier (i.e., unless the application has already done one TLS handshake).

When this message is received the server MUST either ignore the extension or pass it to the application, which then MUST respond to that application data via a new handshake message (see below). If the server ignores it then the client will discover that the server does not support this extension when the client receives the server's handshake messages. Otherwise there must be a corresponding application data handshake message in the server's response, and that indicates that the server TLS and application implementations support this extension.

The extension contents are defined by the application. In order to save the application having to encode application data types and lengths we define two application data extension types and we allow the Client Hello to carry one of each of these extensions:

• pfapp_data (<TBD>)
• sasl_sml_req (<TBD>)

The "pf" prefix indicates "pre-Finished message exchange". It is the application's responsibility to define the contents of the pfapp_data extension.

The sasl_sml_req (SASL server mechanism list request) message contains an empty payload.

We also define new Handshake messages that may be used after the Client Hello messages:

      enum {
          finished(20), pfapp_data(<TBD>),
          sasl_sml(<TBD>), sasl_msg(<TBD>), (255)
      } HandshakeType;

      struct {
          HandshakeType msg_type;    /* handshake type */
          uint24 length;             /* bytes in message */
          select (HandshakeType) {
              case hello_request:       HelloRequest;
              ...
              /* Application pre-Finished message data */
              case pfapp_data:         PFAppData;
              /* SASL server mechanism list */
              case sasl_sml:           SaslSML;
              /* SASL mechanism message */
              case sasl_msg:           SaslMsg;
          } body;
      } Handshake;

      opaque PFAppData<2^16-1>;
      opaque SaslSML<2^16-1>;
      opaque SaslMsg<2^16-1>;

A generic application protocol using these extensions might look like:

   Client                                               Server

   ClientHello w/ sasl_sml_req -------->
                                                   ServerHello
                                                      SaslSML*
                                                  Certificate*
                                            ServerKeyExchange*
                                           CertificateRequest*
                                <--------      ServerHelloDone
   Certificate*
   ClientKeyExchange
   CertificateVerify*
   SaslMsg*
   [ChangeCipherSpec]
   Finished                     -------->
                                            [ChangeCipherSpec]
                                <--------             Finished
   SASL auth messages           <------->   SASL auth messages
   Application Data             <------->     Application Data

Note that the SaslMsg in the client's handshake MUST NOT be included in the client's Finisshed message construction. See Section 3.2.1 (Channel Binding) for more information. [An alternative to this oddity would be to allow application data to be sent by the client after sending the client's Finished message but before receiving the server's Finished message. I'm not sure which of these two options might be the most controversial, but I'd be happy with either. -Nico]

3.  Using TLS/SA

"TLS/SA" is the name given to the use of the sasl_sml_req, SaslSML and SaslMsg TLS extensions to perform SASL authentication in a round-trip optimized manner over TLS. Only SASL/GS2 [I‑D.ietf‑sasl‑gs2] (Josefsson, S. and N. Williams, “Using GSS-API Mechanisms in SASL: The GS2 Mechanism Family,” January 2010.) mechanisms may be optimized.

In order to use SASL via this extension it's important to define how the client requests authentication, how it sends its first message (SASL/GS2 mechanisms are always client-speaks-first mechanisms), and how the SASL exchange continues after the TLS handshake finishes. It is also important to explain how existing SASL applications, such as IMAP, POP3, SMTP, LDAP, etcetera, will use this extension. We do this in the following two sections.

Application protocols may differ from the protocol described below where specifically allowed, but all of the following is NORMATIVE for IMAP, POP3, SMTP, and LDAP.

3.1.  Optimizing SASL Mechanism Negotiation

A client wishing to optimize SASL mechanism negotiation MUST send a sasl_sml_req extension in the client's TLS Hello message. The client MUST NOT send a payload in its sasl_sml_req client hello extension.

If the server supports SASL mechanism negotiation optimization and the server's mechanisl list is shorter than 2^16 - 1 bytes then the server MUST include a SaslSML message in its reply to the client. The payload of the server's SaslSML message MUST be a comma-separated list of SASL mechanism names (note: no NUL terminator is needed, but if present the client MUST ignore the NUL).

3.2.  Optimizing Authentication

A client wishing to optimize a SASL/GS2 mechanism whose first message is shorter than 2^16 - 2 - length of SASL mechanism name bytes MUST: a) use channel binding, b) send the mechanism's first message in the client's SaslMsg handshake message, and c) it MUST prefix the mechanism's first message with the SASL name of the mechanism and a zero-valued byte: "<mech-name><NUL><mech-message>. Note that any replies to this message will be in the form of application data in the record protocol, the form of which may differ by application, though below we define a generic form.

After the TLS handshake finishes the application must continue exchanging SASL messages: any mechanism messages and, finally, the outcome of authentication exchange message. SASL requires that applications define how to frame and encode these messages. Here we provide an example of how applications SHOULD do it, though applications are free to use their own framing and encoding. IMAP, POP3, SMTP and LDAP MUST, when using this extension, do as described below.

For any SASL mechanism authentication messages subsequent to the initial message the application MUST send a network byte order, four byte unsigned binary length of the mechanism message followed by the mechanism message as-is. Messages longer than 2^24 MUST NOT be sent. Where empty messages are required by the SASL mechanism the application should send a zero-valued length and an empty message.

The server's successful "outcome of authentication exchange" message MUST consist of four bytes with all bits set followed by a network byte order four byte unsigned binary length of supplementary information to be defined by the application.

Whereas a server's failed authentication message MUST consist of four bytes in network byte order with the high bit set and the remaining bits cleared, followed by a network byte order four byte unsigned binary length of supplementary information to be defined by the application.

If the last message of the SASL mechanism used is sent by the server then the server's outcome of authentication message MUST immediately follow the last mechanism message. That is: there is no need for the client to send an empty message in response to the last mechanism message just to get the outcome of authentication message. This saves another round-trip.

Typically the supplementary information will be a character string meant for the user to read; the language and encoding may be application dependent or negotiated by the SASL mechanism but, unless specified otherwise by the application, the encoding MUST be UTF-8.

If the SASL authentication exchange ends successfully then the application protocol takes over as it is normally specified, but with the user already authenticated, thus there should be no need to use SASL authentication as normally specified for the application (unless there is a need to re-authenticate, possibly as a different user). If the SASL authentication exchange ends unsuccessfully then the application protocol takes over as it is normally specified, with the user not authenticated, at which point the client MAY re-try authentication.

With a one round trip SASL/GS2 mechanism the protocol then looks like:

   Client                                               Server

   ClientHello w/ sasl_sml_req -------->
                                                   ServerHello
                                                       SaslSML
                                                  Certificate*
                                            ServerKeyExchange*
                                           CertificateRequest*
                                <--------      ServerHelloDone
   Certificate*
   ClientKeyExchange
   CertificateVerify*
   SaslMsg
   [ChangeCipherSpec]
   Finished                     -------->
                                            [ChangeCipherSpec]
                                <--------             Finished
                                <--------        SASL auth msg
                                                   as app data
                                <--------      Outcome of SASL
                                                authentication
   Application Data             <------->     Application Data

With a one and one half round trip mechanism the protocol looks like:

   Client                                               Server

   ClientHello w/ sasl_sml_req -------->
                                                   ServerHello
                                                       SaslSML
                                                  Certificate*
                                            ServerKeyExchange*
                                           CertificateRequest*
                                <--------      ServerHelloDone
   Certificate*
   ClientKeyExchange
   CertificateVerify*
   SaslMsg
   [ChangeCipherSpec]
   Finished                     -------->
                                            [ChangeCipherSpec]
                                <--------             Finished
                                <--------        SASL auth msg
                                                   as app data
   SASL auth msg as app data    -------->
                                <--------      Outcome of SASL
                                                authentication
   Application Data             <------->     Application Data

And with a two round trip mechanism the protocol looks like:

   Client                                               Server

   ClientHello w/ sasl_sml_req -------->
                                                   ServerHello
                                                       SaslSML
                                                  Certificate*
                                            ServerKeyExchange*
                                           CertificateRequest*
                                <--------      ServerHelloDone
   Certificate*
   ClientKeyExchange
   CertificateVerify*
   SaslMsg
   [ChangeCipherSpec]
   Finished                     -------->
                                            [ChangeCipherSpec]
                                <--------             Finished
                                <--------        SASL auth msg
                                                   as app data
   SASL auth msg as app data    -------->
                                <--------        SASL auth msg
                                                   as app data
                                <--------      Outcome of SASL
                                                authentication
   Application Data             <------->     Application Data

3.2.1.  Channel Binding

The TLS channel binding types that are suitable for use with SASL in this facility are:

• tls-server-end-point
• tls-unique

See the IANA channel binding type registry for more information about these channel binding types. The channel binding type to use is to be selected as described in [I‑D.ietf‑sasl‑channel‑bindings] (Williams, N., “SASL And Channel Binding,” April 2009.) (namely: if there is a server certificate, then use tls-server-end-point, else use tls-unique).

Note that the application has to construct its first SASL authentication message for sending in the same half-round trip as the client's Finished message, yet the client's Finished message is used in the tls-unique channel binding type. This means that the Finished message MUST be constructed before the client's SaslMsg, and the SaslMsg is not integrity protected by the client's Finished message, though it will be integrity protected by the server's Finished message.

4.  Non-SASL Optimizations of Existing Application Protocols

In this section and its sub-sections we INFORMATIVELY how a number of existing TLS application protocols might be modified to take advantage of the application data extension for optimization of the application protocol.

It is crucial that clients only use the pfapp_data extension for operations that do not require that the user already be authenticated (the server application MUST reject such uses of pfapp_data) or that require privacy protection. There are no operations in IMAP and POP3, for example, which are suitable for optimization via pfapp_data, but there are for SMTP and LDAP. That's because IMAP and POP3 deal exclusively with user data, while SMTP and LDAP have some operations or objects which can be executed publicly and without user authentication (see below).

4.1.  LDAP

In the case of LDAP the pfapp_data extension can be used to send a single LDAP message, typically a search for the root DSE object. If the server supports this extension then the pfapp_data handshake message can be used to return the result. If the server does not support this extension then the client can repeat its search after the TLS handshake is completed and the TLS record protocol begins to operate.

4.2.  SMTP

Clients may use the pfapp_data extension to send a EHLO SMTP command to the server, and the server may send the SMTP reply to it in a pfapp_data handshake message.

5.  IANA Considerations

When this document is approved for the Standards-Track the &lgt;TBD> values above will be filled in and the IANA TLS ExtensionType and HandshakeType registries will have to be updated to reflect these assignments. (These registries require IETF Consensus and Standards action, respectively.)

6.  Security Considerations

The security considerations of [RFC4422] (Melnikov, A. and K. Zeilenga, “Simple Authentication and Security Layer (SASL),” June 2006.), [I‑D.ietf‑sasl‑channel‑bindings] (Williams, N., “SASL And Channel Binding,” April 2009.), [RFC5246] (Dierks, T. and E. Rescorla, “The Transport Layer Security (TLS) Protocol Version 1.2,” August 2008.) and [RFC5056] (Williams, N., “On the Use of Channel Bindings to Secure Channels,” November 2007.) apply, as do those of [RFC2743] (Linn, J., “Generic Security Service Application Program Interface Version 2, Update 1,” January 2000.) when used via the SASL/GS2 bridge [I‑D.ietf‑sasl‑gs2] (Josefsson, S. and N. Williams, “Using GSS-API Mechanisms in SASL: The GS2 Mechanism Family,” January 2010.).

As usual with TLS there is no privacy protection of client identity unless the client first completes a handshake without authenticating itself, changes the cipher spec, then initiates a new handshake where it does authenticate itself. In this case, client authentication being done via SASL, this means not sending a SaslMsg until after the initial ChangeCipherSpec exchange.

The initial SASL authentication message is not protected by the TLS client's Finished message, but it is protected by the server's Finished message. Channel binding must be used in the optimized authentication case. Therefore the server can detect modifications to the initial SASL authentication message to the best of the selected SASL mechanism's ability, and the client can detect modifications to its initial SASL authentication message through the server's TLS Finished message.

The SASL mechanism negotiation is protected by the TLS Finished messages.

7.  References

7.1. Normative References

7.2. Informative References

Author's Address