KX509 Kerberized Certificate Issuance Protocol
draft-hotz-kx509-01.txt

Abstract

This rfc describes a protocol, called kx509, for using Kerberos tickets to acquire X.509 certificates.

While not (previously) standardized, this protocol is already in use at several large organizations, and certificates issued with this protocol are recognized by TAGPMA (The Americas Grid Policy Management Authority).

Status of this Memo

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

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.

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.”

This Internet-Draft will expire on May 13, 2011.

Copyright Notice

Copyright (c) 2010 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 (http://trustee.ietf.org/license-info) 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. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

1.  Introduction
    1.1.  Requirements Language
2.  Protocol Data
    2.1.  Request Packet
    2.2.  Reply Packet
3.  Protocol Operation
4.  Acknowledgements
5.  IANA Considerations
6.  Security Considerations
7.  References
    7.1.  Normative References
    7.2.  Informative References
Appendix A.  Certificate Cacheing and Deployment Considerations
Appendix B.  Known Issues with This Draft
§  Author's Address

1.  Introduction

The two primary ways of providing cryptographically secure identification on the Internet are Kerberos tickets [RFC4120] (Neuman, C., Yu, T., Hartman, S., and K. Raeburn, “The Kerberos Network Authentication Service (V5),” July 2005.), and X.509 [RFC5280] (Cooper, D., Santesson, S., Farrell, S., Boeyen, S., Housley, R., and W. Polk, “Internet X.509 Public Key Infrastructure Certificate and Certificate Revocation List (CRL) Profile,” May 2008.) and [X.509] (International Telecommunications Union, “Recommendation X.509: The Directory: Public-key and attribute certificate framework,” November 2008.) certificates.

In practical IT infrastructure where both are in use, it's highly desirable to deploy their support in a way which guarantees they both authoritatively refer to the same entities. There is already a widely-adopted standard for using X.509 certificates to acquire corresponding Kerberos tickets called PKINIT [RFC4556] (Zhu, L. and B. Tung, “Public Key Cryptography for Initial Authentication in Kerberos (PKINIT),” June 2006.). This rfc describes the kx509 protocol for supporting the symmetric operation of acquiring X.509 certificates using Kerberos tickets.

In normal operation kx509 can be used after a Kerberos ticket-granting-ticket (TGT) is acquired, which is most likely during user login. First, the client generates a RSA public/private key-pair. Next, using the Kerberos ticket-granting-ticket, it acquires a Kerberos service ticket for the KCA (Kerberized Certificate Authority), and uses this to send the public half of its key-pair. The KCA will decrypt the service ticket, verify the integrity of the incoming packet, determine the identity of the user, and use the session key to send back a corresponding X.509 certificate.

1.1.  Requirements Language

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 RFC 2119 (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.) [RFC2119].

2.  Protocol Data

The protocol consists of a single request/reply exchange using UDP.

Both the request and the reply packet begin with four bytes of version ID information, followed by a DER encoded ASN.1 message. The first two bytes of the version ID are reserved. They MUST be set to zero when sent, and SHOULD be ignored when received. The third and fourth bytes are the major and minor version numbers. The version of the protocol described in this document is designated 2.0, so the first four bytes of the packet are 0, 0, 2, 0.

Incompatible variations of this protocol MUST use a different major version number.

2.1.  Request Packet

The request consists of a version ID, a Kerberos AP_REQ, integrity check data on the request, and a public key generated by the client. The ASN.1 encoding is:

KX509Request ::= SEQUENCE {
        ap-req OCTET STRING,
        pk-hash OCTET STRING,
        pk-key OCTET STRING
}

The ap-req is as described in [RFC4120] (Neuman, C., Yu, T., Hartman, S., and K. Raeburn, “The Kerberos Network Authentication Service (V5),” July 2005.) Section 5.5.1.

The pk-hash is HMAC using SHA-1 as the underlying hash. All 160 bits are sent. The key used is the Kerberos session key. The data is the 4-byte version ID and the octet string for pk-key.

The pk-key contains a public key. This key and its corresponding private key are generated by the client before contacting the server. Implementations of this protocol MUST support RSA keys, in which case the key is a DER encoded RSAPublicKey as defined in [RFC3447] (Jonsson, J. and B. Kaliski, “Public-Key Cryptography Standards (PKCS) #1: RSA Cryptography Specifications Version 2.1,” February 2003.), section A.1.1, and then stored in this octet string in the request. Use of other public-key types is not defined.

2.2.  Reply Packet

The reply consists of a version ID, an error code, and an optional authentication hash, optional certificate, and optional error text. The service SHOULD return replies of the same version as the request where possible.

KX509Response ::= SEQUENCE {
        error-code[0] INTEGER DEFAULT 0,
        hash[1] OCTET STRING OPTIONAL,
        certificate[2] OCTET STRING OPTIONAL,
        e-text[3] VisibleString OPTIONAL
}

Although the format of the reply contains independently optional objects, the server MUST only generate replies with one of the following allowed combinations.

The first case is returned when the server successfully generates a certificate for the user. The certificate is a DER encoded Certificate as defined in [RFC5280] (Cooper, D., Santesson, S., Farrell, S., Boeyen, S., Housley, R., and W. Polk, “Internet X.509 Public Key Infrastructure Certificate and Certificate Revocation List (CRL) Profile,” May 2008.) Section A, page 116.

The second case is returned when the server successfully authenticates the user and their key, but is unable for some other reason to generate a certificate.

The third case MAY be returned if the server is unable to successfully authenticate the user and intends to return some unauthenticated information to the client.

The hash on a response is computed using SHA-1 HMAC as for the request. The data that is hashed is the concatenation of the 4-byte version ID at the beginning of the packet, the error-code (if present), and all other optional fields which are present except the hash itself. In other words, the hash is computed on the fields which are present exclusive of the overall ASN.1 wrapping. The e-text MAY be translated into other character sets for display purposes, but the hash is computed on the e-text in its VisibleString representation.

If the e-text contains NUL characters, the client MAY ignore any part of the error message after the first NUL character for display purposes.

As implied by the above table, if the reply does not contain a certificate it MUST contain an error message and a non-zero error code. Conversely, if a certificate is returned then the error code MUST be zero. The server SHOULD NOT send a zero error-code. The client MUST treat a missing error-code as if it were zero.

If error-code 1 or 2 is returned, the client SHOULD NOT retry the request unless some remedial action is first taken. If error-code 3 is returned, the client MAY retry with any or all known servers before giving up.

If a server error is returned, it is RECOMMENDED that the client retry the request with a different server if one is known. If all known servers have returned server errors, the client MAY retry with servers that returned an error-code of 5 before giving up.

Since all KCAs serving a Kerberos realm are intended to be equivalent, in accordance with [RFC5280] (Cooper, D., Santesson, S., Farrell, S., Boeyen, S., Housley, R., and W. Polk, “Internet X.509 Public Key Infrastructure Certificate and Certificate Revocation List (CRL) Profile,” May 2008.) Section 4.1.2.2, the certificates returned from different KCAs serving the same Kerberos realm MUST NOT contain duplicate serial numbers.

The returned certificate SHOULD identify the Kerberos client principal from the ap-req in the original KX509Request in the subject of the cert, or in a subjectAltName extension. It is RECOMMENDED that the extension be of type id-pkinit-san as described in [RFC4556] (Zhu, L. and B. Tung, “Public Key Cryptography for Initial Authentication in Kerberos (PKINIT),” June 2006.) Section 3.2.2. Note that the id-pkinit-san is simply a standard representation of a Kerberos principal, and has no other implications with respect to PKINIT.

Other extensions MAY be added according to local policy. For example a subjectAltName othername extension of type kcaAuthRealm (OID value 1.3.6.1.4.1.250.42.1) is frequently used to include the client’s realm as an ASN.1 octet string, and the Microsoft userPrincipalName has sometimes been used for the same purpose as the id-pkinit-san.

3.  Protocol Operation

Absent errors, the protocol consists of a single request, sent via UDP, and a single reply, also sent via UDP.

There is no provision for requests or replies which exceed the allowable size of a UDP packet. Furthermore, if the request or reply exceeds the MTU size of a UDP packet for the infrastructure in use, then the reliability of the exchange will decrease significantly. For “normal” Kerberos ap-req structures, and “normal” X.509 certificates, this is unlikely unless the Kerberos service ticket contains large amounts of authorization data. For this reason, it is RECOMMENDED that service tickets for the KCA be issued without authorization data, and that the KCA perform authorization by other means.

Before constructing the request, the client must know the canonical name(s) and port(s) of the server(s) to contact. It MAY determine them by looking up the service's SRV record as described in[RFC2782] (Gulbrandsen, A., Vixie, P., and L. Esibov, “A DNS RR for specifying the location of services (DNS SRV),” February 2000.). The entry to be used is _kca._udp.realm, where realm is the Kerberos realm, used as part of the DNS name.

The client must then acquire a service ticket in order to construct the ap-req for the service. The Kerberos service principal name to use for this service has a first component of "kca_service". The second component and the realm of the principal follow normal Kerberos conventions.

When the server receives a request, it MUST make sanity checks including at least the following:

• The AP-REQ can be decoded and is not expired.
• If the request uses cross-realm authentication, then it satisfies the requirements of local policy and [RFC4120] (Neuman, C., Yu, T., Hartman, S., and K. Raeburn, “The Kerberos Network Authentication Service (V5),” July 2005.) Sections 1.2 and 2.7.
• The request's hash is valid.

The server SHOULD make other sanity checks, such as a minimum public key length, to the extent feasible.

The server MAY decline to respond to an erroneous request. If it does not receive a response a client MAY retry its request, but the client SHOULD wait at least one second before doing so.

The client MUST verify any hash in the reply, and MUST NOT use any certificate in a reply whose hash does not verify. The client MAY display the e-text if the hash is absent or does not verify, but SHOULD indicate the message is not authenticated.

4.  Acknowledgements

The original version of kx509 was implemented using Kerberos 4 at the University of Michigan, and was nicely documented in [KX509] (Doster, W., Watts, M., and D. Hyde, “The KX509 Protocol,” September 2001.). Many thanks to them for their original work.

While developing this document I received important corrections and comments from Jeffrey Altman, and Love Hornquist Astrand. I also received many helpful comments and corrections from Doug Engert, Jeffrey Hutzelman, Sam Hartman, and Timothy J. Miller. Example network traffic was provided by Doug Engert, Marcus Watts, and Matt Crawford from their deployments, and was extremely useful to verify the reality of this specification.

5.  IANA Considerations

This service is conventionally run on UDP port 9878, but this memo includes no request to IANA.

6.  Security Considerations

The only encrypted information in the protocol is that used by Kerberos itself. The considerations for any Kerberized service apply here.

The public key in the request is sent in the clear, and without any guarantees that the requestor actually possesses the corresponding private key. Therefore the only appropriate uses of the returned certificate are those where the subsequent use independently guarantees that the user possesses the private key. In particular digitalSignature MUST NOT be an allowed key usage.

Some information, such as the public key and certificate, is transmitted in the clear but (as the name implies) were designed to be publicly available. However their visibility could still raise privacy concerns. The hash is used to protect their integrity.

The policies for issuing Kerberos tickets and X.509 certificates are usually expressed very differently. An implementation of this protocol should not provide a mechanism for bypassing ticket or certificate policies.

In particular, if the issued certificate can be used with PKINIT, this authentication loop should not bypass policy limits for either X.509 certificates or Kerberos tickets. Since a PKINIT request must be signed and digitalSignature is not an allowed usage for the issued certificate, this loop should not occur.

X.509 certificates are usually issued with considerably longer validity times than Kerberos tickets. Care should be taken that the issued certificate is not valid for longer than the intended policy should allow. Note that[RFC4556] (Zhu, L. and B. Tung, “Public Key Cryptography for Initial Authentication in Kerberos (PKINIT),” June 2006.) Section 3.2.3.1 REQUIRES that the lifetime of an issued ticket not exceed the lifetime of the predecessor certificate. By analogy it is RECOMMENDED that the lifetime of an issued certificate not exceed the lifetime of the predecessor Kerberos ticket.

7.  References

7.1. Normative References

7.2. Informative References

Appendix A.  Certificate Cacheing and Deployment Considerations

As noted in the Security Considerations section, the functional lifetime of the acquired X.509 certificate should match the lifetime of its predecessor Kerberos ticket. Therefore, it is likely that X.509 certificates issued with this protocol should be deleted when the supporting Kerberos tickets are deleted. That makes the Kerberos ticket cache a reasonable location to store the certificate (and its private key).

On the other hand applications, such as web browsers, probably expect certificates in different stores.

A widely used solution to this dichotomy is to implement a PKCS11 library which supports the KX509-acquired credentials.

Appendix B.  Known Issues with This Draft

1. This draft only describes the existing in-the-wild protocol, which has some warts. We need an updated, standards-track protocol.
2. Should IANA registration of a service port be requested?
3. Is there anything that should be added to clarify how Windows might interact with this service? (e.g. more detail on SRV records in section 3?)
4. Should timeouts be more fully specified? Should retry behavior be less specified? Any other issues with error-handling?
5. Several people suggested more information be given on UDP size restrictions and their effect on reliability. Is the text in section 3 sufficient?
6. The original code has provisions for DSA private keys in addition to RSA. Nobody seems to use DSA. Everyone seems to use RSA. Should we specify DSA as an allowed alternative?

Author's Address