INTERNET-DRAFT Samuel Weiler Expires: November 2003 May 22, 2003 Legacy Resolver Compatibility for Delegation Signer draft-ietf-dnsext-dnssec-2535typecode-change-01.txt Status of this Memo This document is an Internet-Draft and is subject to all provisions of Section 10 of RFC2026. 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/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Comments should be sent to the author or to the DNSEXT WG mailing list: namedroppers@ops.ietf.org Abstract As the DNS Security (DNSSEC) specifications have evolved, the syntax and semantics of the DNSSEC resource records (RRs) have changed. Many deployed nameservers understand variants of these semantics. Dangerous interactions can occur when a resolver that understands an earlier version of these semantics queries an authoritative server that understands the new delegation signer semantics, including at least one failure scenario that will cause an unsecured zone to be unresolvable. This document proposes that these interactions be avoided by changing the type codes and mnemonics of the DNSSEC RRs (SIG, KEY, and NXT). 1. Introduction The DNSSEC protocol has been through many iterations whose syntax and semantics are not completely compatible. This has occurred as part of the ordinary process of proposing a protocol, implementing it, testing it in the increasingly complex and diverse environment of the Internet, and refining the definitions of the initial Proposed Standard. In the case of DNSSEC, the process has been complicated by DNS's criticality and wide deployment and the need to add security while minimizing daily operational complexity. A weak area for previous DNS specifications has been lack of detail in specifying resolver behavior, leaving implementors largely on their own to determine many details of resolver function. This, combined with the number of iterations the DNSSEC spec has been through, has resulted in fielded code with a wide variety of behaviors. This variety makes it difficult to predict how a protocol change will be handled by all deployed resolvers. The risk that a change will cause unacceptable or even catastrophic failures makes it difficult to design and deploy a protocol change. One strategy for managing that risk is to structure protocol changes so that existing resolvers can completely ignore input that might confuse them or trigger undesirable failure modes. This document addresses a specific problem caused by Delegation Signer's [DS] introduction of new semantics for the NXT RR that are incompatible with the semantics in [RFC2535]. Answers provided by DS-aware servers can trigger an unacceptable failure mode in some resolvers that implement RFC 2535, which provides a great disincentive to sign zones with DS. The proposed solution allows for the incremental deployment of DS. 1.1 The Problem Delegation signer [DS] introduces new semantics for the NXT RR that are incompatible with the semantics in [RFC2535]. In [RFC2535], NXT records were only required to be returned as part of a non-existence proof. In [DS], an unsecure referral returns, in addition to the NS, a proof of non-existence of a DS RR in the form of an NXT and SIG(NXT). RFC 2535 didn't specify how a resolver was to interpret a response with both an NS and an NXT in the authority section and with NOERROR or NODATA set. Some widely deployed 2535-aware resolvers interpret any answer with an NXT as a proof of non-existence of the requested record. This results in unsecure delegations being invisible to 2535-aware resolvers and violates the basic architectural principle that DNSSEC must do no harm -- the signing of zones must not prevent the resolution of unsecured names. 2. Possible Solutions This section presents several possible solutions. Section 3 recommends one and describes it in more detail. 2.1. Change SIG, KEY, and NXT To avoid the problem described above, legacy (RFC2535-aware) resolvers need to be kept from seeing unsecure referrals that include NXT records in the authority section. The simplest way to do that is to change the type codes for SIG, KEY, and NXT. The obvious drawback to this is that new resolvers will not be able to validate zones signed with the old RRs. This problem already exists, however, because of the changes made by DS, and resolvers that understand the old RRs (and have compatibility issues with DS) are far more prevalent than 2535-signed zones. 2.2. Change a subset of type codes The observed problem with unsecure referrals could be addressed by changing only the NXT type code or another subset of the type codes that includes NXT. This has the virtue of apparent simplicity, but it risks introducing new problems or not going far enough. It's quite possible that more incompatibilities exist between DS and earlier semantics. Legacy resolvers may also be confused by seeing records they recognize (SIG and KEY) while being unable to find NXTs. Although it may seem unnecessary to fix that which is not obviously broken, it's far cleaner to change all of the type codes at once. This will leave legacy resolvers and tools completely blinded to DNSSEC -- they will see only unknown RRs. 2.3. Replace the DO bit Another way to keep legacy resolvers from ever seeing DNSSEC records with DS semantics is to have authoritative servers only send that data to DS-aware resolvers. It's been proposed that assigning a new EDNS0 flag bit to signal DS-awareness (tentatively called "DA"), and having authoritative servers send DNSSEC data only in response to queries with the DA bit set, would accomplish this. This bit would presumably supplant the DO bit described in [RFC3225]. This solution is sufficient only if all 2535-aware resolvers zero out EDNS0 flags that they don't understand. If one passed through the DA bit unchanged, it would still see the new semantics, and it would probably fail to see unsecure delegations. Since it's impractical to know how every DNS implementation handles unknown EDNS0 flags, this is not a universal solution. It could, though, be considered in addition to changing the RR type codes. 2.4. Increment the EDNS version Another proposed solution is to increment the EDNS version number as defined in [RFC2671], on the assumption that all existing implementations will reject higher versions than they support, and retain the DO bit as the signal for DNSSEC awareness. This approach has not been tested. 2.5. Do nothing There is a large deployed base of DNS resolvers that understand DNSSEC as defined by the standards track [RFC2535] and [RFC2065] and, due to underspecification in those documents, interpret any answer with an NXT as a non-existence proof. So long as that is the case, zone owners will have a strong incentive to not sign any zones that contain unsecure delegations, lest those delegations be invisible to such a large installed base. This will dramatically slow DNSSEC adoption. Unfortunately, without signed zones there's no clear incentive for operators of resolvers to upgrade their software to support the new version of DNSSEC, as defined in [DS]. Historical data suggests that resolvers are rarely upgraded, and that old nameserver code never dies. Rather than wait years for resolvers to be upgraded through natural processes before signing zones with unsecure delegations, addressing this problem with a protocol change will immediately remove the disincentive for signing zones and allow widespread deployment of DNSSEC. 3. Protocol changes This document proposes changing the type codes of SIG, KEY, and NXT. This solution is the cleanest and safest, largely because the behavior of resolvers that receive unknown type codes is well understood. This approach has also received the most testing. To avoid operational confusion, it's also necessary to change the mnemonics for these RRs. DNSKEY will be the replacement for KEY, with the mnemonic indicating that these keys are not for application use, per [RFC3445]. RRSIG (Resource Record SIGnature) will replace SIG, and NSEC (Next SECure) will replace NXT. The new types will have exactly the same syntax and semantics as specified for SIG, KEY, and NXT in [RFC2535] and [DS], and they completely replace the old types. A resolver, if it receives the old types, SHOULD treat them as unknown RRs, and SHOULD NOT assign any special semantic value to them. It MUST NOT use them for DNSSEC validations or other DNS operational decision making. For example, a resolver MUST NOT use DNSKEYs to validate SIGs or use KEYs to validate RRSIGs. Authoritative servers SHOULD NOT serve SIG, KEY, or NXT records. If those records are included, they MUST NOT receive special treatment. As an example, if a SIG is included in a signed zone, there MUST be an RRSIG for it. As a clarification to previous documents, many positive responses, including wildcard proofs and insecure referrals, will contain NSEC RRs. As a result, resolvers MUST NOT treat answers with NSEC RRs as negative answers merely because they contain an NSEC. A resolver SHOULD either ignore the NSEC, as a DNSSEC-unaware (or 2535-aware) resolver would, or validate the NSEC and check its applicability and interpretation as described in [RFC2535] and [DS]. 4. IANA Considerations This document updates the IANA registry for DNS Resource Record Types by assigning types 46, 47, and 48 to the DNSKEY, RRSIG, and NSEC RRs, respectively. Types 24, 25, and 30 (SIG, KEY, and NXT) should be marked as Obsolete. 5. Security Considerations The change proposed here does not materially effect security. The implications of trying to use both new and legacy types together are not well understood, and attempts to do so would probably lead to unintended and dangerous results. Changing type codes will leave code paths in legacy resolvers that are never exercised. Unexercised code paths are a frequent source of security holes, largely because those code paths do not get frequent scrutiny. Doing nothing, as described in 3.1, will slow DNSSEC deployment. While this does not decrease security, it also fails to increase it. 6. Normative references [RFC2065] Eastlake, D. and C. Kaufman, "Domain Name System Security Extensions", RFC 2065, January 1997. [RFC2535] Eastlake, D., "Domain Name System Security Extensions", RFC 2535, March 1999. [DS] Gudmundsson, O., "Delegation Signer Resource Record", draft-ietf-dnsext-delegation-signer-14.txt, work in progress, May 2003. 7. Informative References [RFC2671] Vixie, P., "Extension Mechanisms for DNS (EDNS0)", RFC 2671, August 1999. [RFC3225] Conrad, D., "Indicating Resolver Support of DNSSEC", RFC 3225, December 2001. [RFC2929] Eastlake, D., E. Brunner-Williams, and B. Manning. Domain Name System (DNS) IANA Considerations. BCP 42, RFC 2929, September 2000. [RFC3445] Massey, D., and S. Rose. Limiting the Scope of the KEY Resource Record (RR). RFC 3445, December 2002. 8. Acknowledgments The proposed solution and the analysis of alternatives had many contributors. With apologies to anyone overlooked, those include: Micheal Graff, John Ihren, Olaf Kolkman, Mark Kosters, Ed Lewis, Bill Manning, and Suzanne Woolf. Thanks to Jakob Schlyter and Mark Andrews for identifying the incompatibility described in section 1.1. In addition to the above, the author would like to thank Scott Rose, Olafur Gudmundsson, and Sandra Murphy for their substantive comments. 9. Author's Address Samuel Weiler Network Associates Laboratories 15204 Omega Dr., Suite 300 Rockville, MD 20850 USA weiler@tislabs.com