DNSEXT Working Group Olafur Gudmundsson INTERNET-DRAFT July 2001 Updates: RFC 1035, RFC 2535, RFC 3008. Delegation Signer record in parent. Status of this Memo This document is an Internet-Draft and is in full conformance with 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/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Comments should be sent to the authors or the DNSEXT WG mailing list namedroppers@ops.ietf.org This draft expires on January 15, 2002. Copyright Notice Copyright (C) The Internet Society (2001). All rights reserved. Abstract One of the biggest problems in DNS occur when records of the same type can appear on both sides of an delegation. If the contents of these sets differs clients can get confused. RFC2535 KEY records follows the same model as for NS records, parent is responsible for the records but the child is responsible for the contents. This document Gudmundsson Expires January 2002 [Page 1] INTERNET-DRAFT Delegation Signer Record July 2001 proposes to store a different record in the parent that specifies which one of the child's keys are authorized to sign the child's KEY set. This change is not backwards compatible with RFC2535 but simplifies DNSSEC operation. 1 - Introduction Familiarity with the DNS system [RFC1035], DNS security extensions [RFC2535] and DNSSEC terminology [RFC3090] is important. When the same data can reside in two administratively different DNS zones sources it is common that the data gets out of sync. NS record in a zone indicates that there is a delegation at this name and the NS record lists the authorative servers for the real zone. Based on actual measurements 10-30% of all delegations in the Internet have differing NS sets at parent and child. There are number of reasons for this, including lack of communication between parent and child and bogus name-servers being listed to meet registrar requirements. DNSSEC [RFC2535,RFC3008,RFC3090] specifies that child must have its KEY set signed by the parent to create a verifiable chain of KEYs. There is some debate, where the signed KEY set should reside, parent[Parent] or child[RFC2535]. If the KEY set resides at the child, frequent communication is needed between the two parties, to transmit key sets up to parent and then the signed set or signatures down to child. If the KEY set resides at the parent[Parent] the communication is reduced having only child send updated key sets to parent. DNSSEC[RFC2535] requires that the parent store NULL key set for unsecure children, this complicates resolution process as in many cases as servers for both parent and child need to be queried for KEY set the [Parent] proposal simplifies this. Further complication of the DNSSEC KEY model is that KEY record is used to store DNS zone keys and public keys for other protocols. This can lead to large key sets at delegation points. There are number of potential problems with this including: 1. KEY set may become quite large if many applications/protocols store their keys at the zone apex. Example of protocols are IPSEC, HTTP, SMTP, SSH etc. 2. Key set may require frequent updates. 3. Probability of compromised/lost keys increases and triggers emergency key rollover. 4. Parent may refuse sign key sets with NON DNS zone keys. 5. Parent may not have QoS on key changes that meets child's expectations. Gudmundsson Expires January 2002 [Page 2] INTERNET-DRAFT Delegation Signer Record July 2001 Given these and other reasons there is good reason to explore alternatives to using only KEY records to create chain of trust. Some of these problems can be reduced or eliminated by operational rules or protocol changes. To reduce the number of keys at apex, rule to require applications to store their KEY records at the SRV name for that application is one possibility. Another is to restrict KEY record to DNS keys only and create a new type for all non DNS keys. Third possible solution is to ban the storage of non DNS related keys at zone apex. There are other possible solutions but they are outside the scope of this document. 1.1 - Delegation Signer Record model This document proposes an alternative to the KEY record chain of trust, that uses a special record that can only reside at the parent. This record will identify the key(s) that child will use to self sign its own KEY set. The chain of trust is now established by verifying the parent KEY set, the DS set from the parent and then the KEY set at the child. This is cryptographically equivalent to just using KEY records. Communication between the parent and child is reduced as the parent only needs to know of changes in DNS zone KEY(s) used to sign the apex KEY set. If other KEY records are stored at the zone apex, the parent does not need to be aware of them. This approach has the advantage that it minimizes the communication between the parent and child and the child is the authority for the KEY set with full control over the contents. This enables each to operate and maintain each zone independent of each other. Thus if child wants to have frequent key rollover for its DNS keys parent does not need to be aware of it as the child can use one key to only sign its apex KEY set and other keys to sign the other record sets in the zone. The child can just as well use the same key to sign all records in its zone. Another advantage is that this model fits well with slow rollout of DNSSEC and islands of security model. In the islands of security model someone that trusts "good.example." preconfigures a key from "good.example." as a trusted keys and from then on trusts any data that is signed by that key or has a chain of trust to that key. If "example." starts advertising DS records "good.example." does not have to change operations, by suspending self-signing. If DS records can also be used to identify trusted keys instead of KEY records. Gudmundsson Expires January 2002 [Page 3] INTERNET-DRAFT Delegation Signer Record July 2001 The main disadvantage of this approach is double the number of signatures that need to be verified for the each delegation KEY set. There is no impact on verifying other record sets. 1.2 - Reserved words The key words "MUST", "MUST NOT", "SHOULD", and "MAY" in this document are to be interpreted as described in RFC2119. 2 - DS (Delegation KEY signer) record: 2.1 Protocol change DS record MUST only appear at secure delegations in the parent zone. The record lists the child's keys that SHOULD sign the child's key set. Insecure delegation MUST NOT have a DS record, the presence of DS record SHOULD be considered a hint that the child might be secure. Resolver MUST only trust KEY records that match a DS record. NOTE: It has been suggested that NULL DS record for insecure children is better than no record. The advantage is to have authenticated denial of child's security status, the drawback is for large delegating zones there will be many NULL DS records. If parent uses NXT records adding NXT record to the authority section in the cases when no DS record exists at delegation will give the same result as NULL DS record. WG please comment on which approach is better. Updates RFC2535 sections 2.3.4 and 3.4, as well as RFC3008 section 2.7: Delegating zones MUST NOT store KEY records for delegations. The only records that can appear at delegation in parent are NS, SIG, NXT and DS. Zone MUST self sign its apex KEY set, it SHOULD sign it with a key that corresponds to a DS record in the parent. The KEY used to sign the apex KEY RRset CAN sign other RRsets in the zone. If child apex KEY RRset is not signed with one of the keys specified in the DS record the child is locally secure[RFC3090] and SHOULD only be considered secure the resolver has been instructed to trust the key used, via preconfiguration. Authorative server answering a query, that has the OK bit set[OKbit], MUST include the DS set in the additional section if the answer is a referral and there is space. Caching servers SHOULD return the DS record in the additional section under the same condition. Gudmundsson Expires January 2002 [Page 4] INTERNET-DRAFT Delegation Signer Record July 2001 2.1.1 - Comments on protocol change Over the years there has been various discussions on that the delegation model in DNS is broken as there is no real good way to assert if delegation exists. In RFC2535 version of DNSSEC the authentication of a delegation is the NS bit in the NXT bitmap at the delegation point. Something more explicit is needed and the DS record addresses this for secure delegations. DS record is the first DNS record to be only stored at the upper side of a delegation. NS records appear at both sides as do SIG and NXT. All other records can only appear at the lower side. This will cause some problems as servers authorative for parent may reject DS record even if the server understands unknown types, or not hand them out unless explicitly asked. Similarly a nameserver acting as a authorative for child and as a caching recursive server may never return the DS record. A caching server does not care from which side DS record comes from and thus should not have to be changed if it supports unknown types. Different TTL values on the childs NS set and parents DS set may cause the DS set to expire before the NS set. In this case an non-DS aware server would ask the child server for the DS set and get NXDOMAIN answer. DS aware server will know to ask the parent for the DS record. Secure resolvers need to know about the DS record and how to interpret it. In the worst case, introducing the DS record, doubles the signatures that need to be checked to validate a KEY set. Note: The working group must determine if the tradeoff between more work in resolvers is justified by the operational simplification of this model. The author think this is a small price to pay to have a cleaner delegations structure. One argument put forward is that DNS should be optimized for read when ever possible, and on the face of it adding the DS record makes reading data from DNS more expensive. The operational complexities and legal hurdles that KEY records in parents or children make prevent DNSSEC to ever get deployed. Gudmundsson Expires January 2002 [Page 5] INTERNET-DRAFT Delegation Signer Record July 2001 2.2 Wire format of DS record The DS record consists of algorithm, size, key tag and SHA-1 digest of the public key KEY record allowed to sign the child's delegation. 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | key tag | size | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | algorithm | SHA-1 digest | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | (20 bytes) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-| | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-| | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-| | | +-+-+-+-+-+-+-+-+ The key tag is calculated as specified in RFC2535, the size is the size of the public key in bits as specified in the document specifying the algorithm. Algorithm MUST be an algorithm number assigned in the range 1..251. The SHA-1 digest is calculated over the canonical name of the delegation followed by the RDATA of the KEY record. The size of the DS RDATA is 25 bytes, regardless of the key size. NOTE: if 160 bits is to large the SHA-1 digest can be shortened but that weakens the overall security of the system. 2.2.1 Justifications for fields The algorithm and size fields are here to allow resolvers to quickly identify the candidate KEY records to examine. Key Tag is to allow quick check if this is a good candidate. The key tag is redundant but provides some greater assurance than SHA-1 digest on its own. SHA-1 is a strong cryptographic checksum, it is hard for attacker to generate a KEY record that has the same SHA-1 digest. Making sure that the KEY record is a valid public key is much harder. Combining the name of the key and the key data as input to the digest provides stronger assurance of the binding. Combining the SHA-1 with the other fields makes the task of the attacker is as hard breaking the public key. Even if someone creates a database of all SHA-1 key hashes seen so far, the addition of the name renders that database useless for attacks against random zones. Gudmundsson Expires January 2002 [Page 6] INTERNET-DRAFT Delegation Signer Record July 2001 2.3 Presentation format of DS record The presentation format of DS record consists of 2 numbers, followed by either the name of the signature algorithm or the algorithm number. The digest is to be presented in hex. 2.4 Justifications for compact format This format allows concise representation of the keys that child will use, thus keeping down the size of the answer for the delegation, reducing the probability of packet overflow. The SHA-1 hash is strong enough to uniquely identify the key. This is similar to the PGP footprint. Each DS record has RDATA size of 25, regardless of the size of the keys, keeping the answers from the parent smaller than if public key was used. The smallest currently defined KEY record RDATA is 70 bytes. Compact DS format is also better suited to list trusted keys for islands of security in configuration files. 2.5 Transition issues for installed base RFC2535 compliant resolver will assume that all DS secured delegations are locally secure. This is a bad thing, thus it might be necessary for a transition period to support both DS and SIG@Child. The cost is one more signatures in the answer and that early adopters have to cumbersome communications that DS is supposed to solve. Resolvers will not get confused as they will select signatures with the KEY they trust and ignore the other one. 3 Resolver Example To create a chain of trust resolver goes from trusted KEY to DS to KEY. Assume the key for domain example. is trusted. In zone "example." we have example. KEY secure.example. DS tag=12345 size=1024 alg=dsa secure.example. NS ns1.secure.example. NS ns1.secure.example. s secure.example. NXT NS SIG NXT DS tail.example. secure.example. SIG(NXT) secure.example. SIG(DS) Gudmundsson Expires January 2002 [Page 7] INTERNET-DRAFT Delegation Signer Record July 2001 In zone "secure.example." we have secure.example. SOA secure.example. NS ns1.secure.example. NS ns1.secure.example. secure.example. KEY KEY KEY secure.example. SIG(KEY) secure.example. SIG(SOA) secure.example. SIG(NS) In this example the trusted key for example signs the DS record for "secure.example.", making that a trusted record. The DS record states what key is supposed to sign the KEY record at secure.example. In this example "secure.example." has three different KEY records and the one corresponding to the KEY identified in the DS record signs the KEY set, thus the key set is validated and trusted. Note that one of the other keys in the keyset actually signs the zone data, and resolvers will trust the signatures as the key appears in the KEY set. This example has only one DS record for the child but there no reason to outlaw multiple DS records. More than one DS record is needed during signing key rollover. It is strongly recommended that the DS set be kept small. 3.1 Resolver cost estimates for DS records From a RFC2535 resolver point of view for each delegation followed to chase down an answer one KEY record has to be verified and possibly some other records based on policy, for example the contents of the NS set. Once the resolver gets to the appropriate delegation validating the answer may require verifying one or more signatures. For a simple A record lookup requires at least N delegations to be verified and 1 RRset. For a DS enabled resolver the cost is 2N+1. For MX record the cost where the target of the MX record is in the same zone as the MX record the costs are N+2 and 2N+2. In the case of negative answer the same holds ratios hold true. Resolver may require an extra query to get the DS record but and this may add to the overall cost of the query, but this is never worse than chasing down NULL KEY records from the parent in RFC2535 DNSSEC. DS adds processing overhead on resolvers, increases the size of delegation answers. DS requires much less storage in large delegation zones than SIG@Parent. Gudmundsson Expires January 2002 [Page 8] INTERNET-DRAFT Delegation Signer Record July 2001 4 Acknowledgments Number of people have over the last few years contributed number of ideas that are captured in this document. The core idea of using one key to that has only the role of signing a key set, comes from discussions with Bill Manning and Perry Metzger on how to put in a single root key in all resolver that lives for a long time. Brian Wellington, Dan Massey, Edward Lewis, Havard Eidnes, Jakob Schlyter, Mark Kosters, Miek Gieben, Roy Arens, Scott Rosen have provided usefull comments. 4 - Security Considerations: This document proposes a change to the validation chain of KEY records in DNS. The change in is not believed to reduce security in the overall system, in RFC2535 DNSSEC child must communicate keys to parent and prudent parents will require some authentication on that handshake. The modified protocol will require same authentication but allows the child to exert more local control over its own KEY set. In the representation of DS record, there is a possibility that an attacker can generate an valid KEY that matches all the checks thus starting to forge data from the child. This is considered impractical as on average more than 2**80 keys must be generated before one is found that will match. DS record is a change to DNSSEC protocol and there is some installed base of implementations, as well as text books on how to set up secured delegations. Implementations that do not understand DS record will not be able to follow the KEY to DS to KEY chain and consider all zone secured that way insecure. 5 - IANA Considerations: IANA needs to allocate RR type code for DS from the standard RR type space. Gudmundsson Expires January 2002 [Page 9] INTERNET-DRAFT Delegation Signer Record July 2001 References: [RFC1035] P. Mockapetris, ``Domain Names - Implementation and Specification'', STD 13, RFC 1035, November 1987. [RFC2535] D. Eastlake, ``Domain Name System Security Extensions'', RFC 2535, March 1999. [RFC3008] B. Wellington, ``Domain Name System Security (DNSSEC) Signing Authority'', RFC 3008, November 2000. [RFC3090] E. Lewis `` DNS Security Extension Clarification on Zone Status'', RFC 3090, March 2001. [OKbit] D. Conrad, ``Indicating Resolver Support of DNSSEC'', work in progress , April 2001. [Parent] R. Gieben, T. Lindgreen, ``Parent stores the child's zone KEYs'', work in progress , May 2001. Author Address Olafur Gudmundsson 3826 Legation Street, NW Washington, DC, 20015 USA Appendix A: Changes from Prior versions Changes from version 00 Changed name from DK to DS based on working group comments. Dropped verbose format based on WG comments. Added text about TTL issue/problem in caching servers. Added text about islands of security and clarified the cost impact. Major editing of arguments and some reordering of text for clarity. Added section on transition issues. Gudmundsson Expires January 2002 [Page 10] INTERNET-DRAFT Delegation Signer Record July 2001 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." Gudmundsson Expires January 2002 [Page 11]