| Internet-Draft | DNS Delegation Revalidation | July 2024 |
| Huque, et al. | Expires 9 January 2025 | [Page] |
This document recommends improved DNS [RFC1034] [RFC1035] resolver behavior with respect to the processing of Name Server (NS) resource record (RR) sets (RRsets) during iterative resolution. When following a referral response from an authoritative server to a child zone, DNS resolvers should explicitly query the authoritative NS RRset at the apex of the child zone and cache this in preference to the NS RRset on the parent side of the zone cut. The (A and AAAA) address RRsets in the additional section from referral responses and authoritative NS answers for the names of the NS RRset, should similarly be re-queried and used to replace the entries with the lower trustworthiness ranking in cache. Resolvers should also periodically revalidate the child delegation by re-querying the parent zone at the expiration of the TTL of the parent side NS RRset.¶
This note is to be removed before publishing as an RFC.¶
Status information for this document may be found at https://datatracker.ietf.org/doc/draft-ietf-dnsop-ns-revalidation/.¶
Discussion of this document takes place on the DNSOP Working Group mailing list (mailto:[email protected]), which is archived at https://mailarchive.ietf.org/arch/browse/dnsop/. Subscribe at https://www.ietf.org/mailman/listinfo/dnsop/.¶
Source for this draft and an issue tracker can be found at https://github.com/shuque/ns-revalidation.¶
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 https://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 9 January 2025.¶
Copyright (c) 2024 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 (https://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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
This document recommends improved DNS resolver behavior with respect to the processing of NS record sets during iterative resolution. The first recommendation is that resolvers, when following a referral response from an authoritative server to a child zone, should explicitly query the authoritative NS RRset at the apex of the child zone and cache this in preference to the NS RRset on the parent side of the zone cut. The address records in the additional section from the referral response (as glue) or authoritative NS response that match the names of the NS RRset should similarly be required if they are cached non-authoritatively. The authoritative answers from those queries should replace the cached non-authoritative A and AAAA RRsets. The second recommendation is to revalidate the delegation by re-querying the parent zone at the expiration of the TTL of the parent side NS RRset.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
There is wide variability in the behavior of deployed DNS resolvers today with respect to how they process delegation records. Some of them prefer the parent NS set, some prefer the child, and for others, what they preferentially cache depends on the dynamic state of queries and responses they have processed. This document aims to bring more commonality and predictability by standardizing the behavior in a way that comports with the DNS protocol. Another goal is to improve DNS security.¶
The delegation NS RRset at the bottom of the parent zone and the apex NS RRset in the child zone are unsynchronized in the DNS protocol. Section 4.2.2 of [RFC1034] says "The administrators of both zones should insure that the NS and glue RRs which mark both sides of the cut are consistent and remain so.". But for a variety of reasons they could not be. Officially, a child zone's apex NS RRset is authoritative and thus has a higher cache credibility than the parent's delegation NS RRset, which is non-authoritative glue (Sections 5.4.1 (Ranking data) and 6.1 (Zone authority) of [RFC2181]). Hence the NS RRset "below the zone cut" should immediately replace the parent's delegating NS RRset in cache when an iterative caching DNS resolver crosses a zone boundary. However, this can only happen if (1) the resolver receives the authoritative NS RRset in the Authority section of a response from the child zone, which is not mandatory, or (2) if the resolver explicitly issues an NS RRset query to the child zone as part of its iterative resolution algorithm. In the absence of this, it is possible for an iterative caching resolver to never learn the authoritative NS RRset for a zone, unless a downstream client of the resolver explicitly issues such an NS query, which is not something that normal enduser applications do, and thus cannot be relied upon to occur with any regularity.¶
Increasingly, there is a trend towards minimizing unnecessary data in DNS responses. Several popular DNS implementations default to such a configuration (see "minimal-responses" in BIND and NSD). So, they may never include the authoritative NS RRset in the Authority section of their responses.¶
A common reason that zone owners want to ensure that resolvers place the authoritative NS RRset preferentially in their cache is that the TTLs may differ between the parent and child side of the zone cut. Some DNS Top Level Domains (TLDs) only support long fixed TTLs in their delegation NS sets. This inhibits a child zone owner's ability to make more rapid changes to their nameserver configuration using a shorter TTL, if resolvers have no systematic mechanism to observe and cache the child NS RRset.¶
Similarly, a child zone owner may also choose to have longer TTLs in their delegation NS sets and address records to decrease the attack window for cache poisoning attacks. For example, at the time of writing, root-servers.net has a TTL of 6 weeks for the root server identifier address records, where the TTL in the priming response is 6 days.¶
A child zone's delegation still needs to be periodically revalidated at the parent to make sure that the parent zone has not legitimately re-delegated the zone to a different set of nameservers, or even removed the delegation. Otherwise, resolvers that refresh the TTL of a child NS RRset on subsequent queries or due to pre-fetching, may cling to those nameservers long after they have been re-delegated elsewhere. This leads to the second recommendation in this document, "Delegation Revalidation" - Resolvers should record the TTL of the parent's delegating NS RRset, and use it to trigger a revalidation action. Attacks exploiting lack of this revalidation have been described in [GHOST1], [GHOST2].¶
When a referral response is received during iteration, a validation query SHOULD be sent in parallel with the resolution of the triggering query, to the delegated nameservers for the newly discovered zone cut. Note that validating resolvers today, when following a secure referral, already need to dispatch a query to the delegated nameservers for the DNSKEY RRset, so this validation query could be sent in parallel with that DNSKEY query.¶
A validation query consists of a query for the child's apex NS RRset, sent to the newly discovered delegation's nameservers. Normal iterative logic applies to the processing of responses to validation queries, including storing the results in cache, trying the next server on SERVFAIL or timeout, and so on. Positive responses to this validation query will be cached with an authoritative data ranking. Successive queries directed to the same zone will be directed to the nameservers listed in the child's apex, due to the ranking of this answer. If the validation query fails, the parent NS RRset will remain the one with the highest ranking and will be used for successive queries.¶
The triggering query to the child may contain the NS RRset in the authority section as well. This NS RRset however has a lower trustworthiness than the set from the direct query (Section 5.4.1 of [RFC2181]), so regardless of the order in which the responses are received, the NS RRset from the answer section from the direct child's apex NS RRset query will be stored in the cache eventually.¶
When a resolver detects that the child's apex NS RRset contains different nameservers than the non-authoritative version at the parent side of the zone cut, it MAY report the mismatch using DNS Error Reporting [RFC9567] on the Report-Channel for the child zone, as well as on the Report-Channel for the parent zone, with an extended DNS error code of TBD Section 6.¶
Additional validation queries for the "glue" address RRs of referral responses (if not already authoritatively present in cache) SHOULD be sent with the validation query for the NS RRset as well. Positive responses will be cached authoritatively and replace the non authoritative "glue" entries. Successive queries directed to the same zone will be directed to the authoritative nameservers denoted in the referral response.¶
The names from the NS RRset from a validating NS query may differ from the names in NS RRset in the referral response. Outstanding validation queries for "glue" address RRs that do not match names in the authoritative NS RRset may be discarded, or they may be left running to completion. Their result will no longer be used in queries for the zone. Outstanding validation queries for "glue" address RRs that do match names in the authoritative NS RRset MUST be left running to completion. They do not need to be re-queried after reception of the authoritative NS RRset (see Section 4).¶
Resolvers may choose to delay the response to the triggering query until both the triggering query and the validation query have been answered. In practice, we expect many implementations may answer the triggering query in advance of the validation query for performance reasons. An additional reason is that there are unfortunately a number of nameservers in the field that (incorrectly) fail to properly answer explicit queries for zone apex NS records, and thus the revalidation logic may need to be applied lazily and opportunistically to deal with them. In cases where the delegated nameservers respond incorrectly to an NS query, the resolver SHOULD abandon this algorithm for the zone in question and fall back to using only the information from the parent's referral response.¶
If the resolver chooses to delay the response, and there are no nameserver names in common between the child's apex NS RRset and the parent's delegation NS RRset, then the responses received from forwarding the triggering query to the parent's delegated nameservers SHOULD be discarded after validation, and this query should be forwarded again to the child's apex nameservers.¶
Authoritative responses for a zone's NS RRset at the apex can contain additional addresses. A NS RRset validation response is such an example of such responses. A priming response is another example of an authoritative zone's NS RRset response [RFC8109].¶
Additional addresses in authoritative NS RRset responses SHOULD be validated if they are not already authoritatively in cache. Only when such additional addresses are DNSSEC verifiable, (because the complete RRset is included, including a verifiable signature for the RRset), such additional addresses can be cached authoritatively immediately without additional validation queries. DNSSEC validation is enough validation in those cases. Otherwise, the addresses cannot be assumed to be complete or even authoritatively present in the same zone, and additional validation queries SHOULD be made for these addresses.¶
Note that there may be outstanding address validation queries for the names of the authoritative NS RRset (from referral address validation queries). In those cases no new validation queries need to be made.¶
Resolvers may choose to delay the response to a triggering query until it can be verified that the answer came from a name server listening on an authoritatively acquired address for an authoritatively acquired name. This would offer the most trustworthy responses with the least risk for forgery or eavesdropping.¶
The essence of this mechanism is re-validation of all delegation metadata that directly or indirectly supports an owner name in cache. This requires a cache to remember the delegated name server names for each zone cut as received from the parent (delegating) zone's name servers, and also the TTL of that NS RRset, and the TTL of the associated DS RRset (if seen).¶
A delegation under re-validation is called a "re-validation point" and is "still valid" if its parent zone's servers still respond to an in-zone question with a referral to the re-validation point, and if that referral overlaps with the previously cached referral by at least one name server name, and the DS RRset (if seen) overlaps the previously cached DS RRset (if also seen) by at least one delegated signer.¶
If the response is not a referral or refers to a different zone than before, then the shape of the delegation hierarchy has changed. If the response is a referral to the re-validation point but to a wholly novel NS RRset or a wholly novel DS RRset, then the authority for that zone has changed. For clarity, this includes transitions between empty and non-empty DS RRsets.¶
If the shape of the delegation hierarchy or the authority for a zone has been found to change, then currently cached data whose owner names are at or below that re-validation point MUST NOT be used. Such non-use can be by directed garbage collection or lazy generational garbage collection or some other method befitting the architecture of the cache. What matters is that the cache behave as though this data was removed.¶
Since re-validation can discover changes in the shape of the delegation hierarchy it is more efficient to re-validate from the top (root) downward (to the owner name) since an upper level re-validation may obviate lower level re-validations. What matters is that the supporting chain of delegations from the root to the owner name be demonstrably valid; further specifics are implementation details.¶
Re-validation MUST be triggered when delegation meta-data has been cached for a period at most exceeding the delegating NS or DS (if seen) RRset TTL. If the corresponding child zone's apex NS RRset TTL is smaller than the delegating NS RRset TTL, revalidation MUST happen at that interval instead. However, resolvers SHOULD impose a sensitive minimum TTL floor they are willing to endure to avoid potential computational DoS attacks inflicted by zones with very short TTLs.¶
In normal operations this meta-data can be quickly re-validated with no further work. However, when re-delegation or take-down occurs, a re-validating cache will discover this within one delegation TTL period, allowing the rapid expulsion of old data from the cache.¶
IANA is requested to assign a value to the "Extended DNS Error Codes" registry [RFC8914].¶
| INFO-CODE | Purpose | Reference |
|---|---|---|
| TBD | referral NS RRset mismatch | [this document] |
Referral response NS RRsets and glue, and the additional addresses from authoritative NS RRset responses (such as the root priming response), are not protected with DNSSEC signatures. An attacker that is able to alter the unsigned A and AAAA RRsets in the additional section of referral and authoritative NS RRset responses, can fool a resolver into taking addresses under the control of the attacker to be authoritative for the zone. Such an attacker can redirect all traffic to the zone (of the referral or authoritative NS RRset response) to a rogue name server.¶
A rogue name server can view all queries from the resolver to that zone and alter all unsigned parts of responses, such as the parent side NS RRsets and glue of further referral responses. Resolvers following referrals from a rogue name server, that do not revalidate those referral responses, can subsequently be fooled into taking addresses under the control of the attacker to be authoritative for those delegations as well. The higher up the DNS tree, the more impact such an attack has. An attacker controlling a rogue name server for the root has potentially complete control over the entire domain name space and can alter all unsigned parts undetected.¶
Revalidating referral and authoritative NS RRset responses enables to defend against the above described attack with DNSSEC signed infrastructure RRsets. Unlike cache poisoning defences that leverage increase entropy to protect the transaction, revalidation of NS RRsets and addresses also provides protection against on-path attacks.¶
Since December 6, 2023, the root zone contains a DNSSEC signed cryptographic message digest [RFC8976][ROOT-ZONEMD], covering all root zone data. This includes all non-authoritative data such as the A and AAAA RRsets for the IP addresses of the root server identifiers, as well as the NS RRsets and glue that make up the delegations. A root zone local to the resolver [RFC8806] with a verified and validated ZONEMD RRset, would provide protection similarly strong to revalidating the root and the top level domains.¶
In [DNS-CACHE-INJECTIONS] an overview is given of 18 cache poisoning attacks from which 13 can be remedied with delegation revalidation. The paper provides recommendations for handling records in DNS response with respect to an earlier version of the idea presented in this document [I-D.wijngaards-dnsext-resolver-side-mitigation].¶
Upgrading NS RRset Credibility (Section 3) allows resolvers to cache and utilize the authoritative child apex NS RRset in preference to the non-authoritative parent NS RRset. However, it is important to implement the steps described in Delegation Revalidation (Section 5) at the expiration of the parent's delegating TTL. Otherwise, the operator of a malicious child zone, originally delegated to, but subsequently delegated away from, can cause resolvers that refresh TTLs on subsequent NS set queries, or that pre-fetch NS queries, to never learn of the redelegated zone [GHOST1], [GHOST2].¶
An implementation may wish to consider limiting revalidation to delegations that cross administrative boundaries such as anywhere in .ip6.arpa and .in-addr.arpa as well as any so-called "public suffix" such as the root zone, top level zones such as ".com" or ".net", and effective top level zones such as ".ad.jp" or ".co.uk".¶
Some resolvers do not adhere to Sections 5.4.1 and 6.1 of [RFC2181], and only use the non-authoritative parent side NS RRsets and glue returned in referral responses for contacting authoritative name servers [I-D.fujiwara-dnsop-resolver-update]. As a consequence, they are not susceptible to many of the cache poisoning attacks enumerated in [DNS-CACHE-INJECTIONS] that are based upon the relative trustworthiness of DNS data. Such resolvers are also not susceptible to the GHOST domain attacks [GHOST1], [GHOST2]. Such resolvers will however never benefit from DNSSEC protection of infrastructure RRsets and are susceptible to query redirection attacks.¶
Wouter Wijngaards proposed explicitly obtaining authoritative child NS data in [I-D.wijngaards-dnsext-resolver-side-mitigation]. This behavior has been implemented in the Unbound DNS resolver via the "harden-referral-path" option. The combination of child NS fetch and revalidating the child delegation was originally proposed in [I-D.vixie-dnsext-resimprove], by Paul Vixie, Rodney Joffe, and Frederico Neves.¶
The authors would like to thank Ralph Dolmans who was an early collaborator on this work, as well as the many members of the IETF DNS Operations Working Group for helpful comments and discussion.¶
Note to the RFC Editor: please remove this entire appendix before publication.¶
The Unbound resolver software has delegation revalidation as described in this document implemented since version 1.1 (released August 29, 2008).
It is enabled with a configuration option harden-referral-path: yes which is disabled by default.¶
"Redhat Enterprise Linux has been running Unbound with the harden-referral-path: option set to yes for years without problems", as mentioned by Paul Wouters during dnsop workgroup session at the IETF 119.¶
The Knot Resolver software revalidates the priming response as part of priming the root zone since version 1.5.1 (released December 12, 2017)¶