| DNSOP | G. Huston |
| Internet-Draft | J. Damas |
| Intended status: Standards Track | APNIC |
| Expires: December 13, 2018 | W. Kumari |
| June 11, 2018 |
A Root Key Trust Anchor Sentinel for DNSSEC
draft-ietf-dnsop-kskroll-sentinel-14
The DNS Security Extensions (DNSSEC) were developed to provide origin authentication and integrity protection for DNS data by using digital signatures. These digital signatures can be verified by building a chain of trust starting from a trust anchor and proceeding down to a particular node in the DNS. This document specifies a mechanism that will allow an end user and third parties to determine the trusted key state for the root key of the resolvers that handle that user's DNS queries. Note that this method is only applicable for determining which keys are in the trust store for the root key.
[ This document is being collaborated on in Github at: https://github.com/APNIC-Labs/draft-kskroll-sentinel. The most recent version of the document, open issues, etc should all be available here. The authors (gratefully) accept pull requests. RFC Editor, please remove text in square brackets before publication. ]
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 December 13, 2018.
Copyright (c) 2018 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 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.
The DNS Security Extensions (DNSSEC) [RFC4033], [RFC4034] and [RFC4035] were developed to provide origin authentication and integrity protection for DNS data by using digital signatures. DNSSEC uses Key Tags to efficiently match signatures to the keys from which they are generated. The Key Tag is a 16-bit value computed from the RDATA portion of a DNSKEY RR using a formula found in "Key Tag Calculation" (Appendix B of "Resource Records for the DNS Security Extensions" [RFC4034]), a formula similar to a ones-complement checksum. RRSIG RRs contain a Key Tag field whose value is equal to the Key Tag of the DNSKEY RR that validates the signature.
This document specifies how security-aware DNS resolvers that perform validation of their responses can respond to certain queries in a manner that allows an agent performing the queries to deduce whether a particular key for the root has been loaded into that resolver's trusted key store. This document also describes a procedure where a collection of resolvers can be tested to determine if at least one of these resolvers has loaded a given key into its trusted key store. These tests can be used to determine whether a certain root zone KSK is ready to be used as a trusted key, within the context of a planned root zone KSK key roll.
There are two primary use cases for this mechanism:
The mechanism described in this document meets both of these use cases. This new mechanism is OPTIONAL to implement and use, although for reasons of supporting broad-based measurement techniques, it is strongly preferred that configurations of DNSSEC-validating resolvers enabled this mechanism by default, allowing for local configuration directives to disable this mechanism if desired.
The KSK sentinel tests described in this document use a test comprising of a set of DNS queries to domain names that have special values for the left-most label. The test relies on recursive resolvers supporting a mechanism that recognises this special name pattern in queries, and under certain defined circumstances will return a DNS SERVFAIL response code (RCODE 2), mimicking the response code that is returned by security-aware resolvers when DNSSEC validation fails.
If a browser or operating system is configured with multiple resolvers, and those resolvers have different properties (for example, one performs DNSSEC validation and one does not), the sentinel test described in this document can still be used, but it makes a number of assumptions about DNS resolution behaviour that may not necessarily hold in all environments. If these assumptions do not hold (such as, for example, requiring the stub resolver to query the next recursive resolver in the locally configured set upon receipt of a SERVFAIL response code) then this test may produce indeterminate or inconsistent results. In some cases where these assumptions do not hold, repeating the same test query set may generate different results.
Note that the sentinel mechanism described here measures a very different (and likely more useful) metric than [RFC8145]. RFC 8145 relies on resolvers reporting towards the root servers a list of locally cached trust anchors for the root zone. Those reports can be used to infer how many resolvers may be impacted by a KSK roll, but not what the user impact of the KSK roll will be.
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.
DNSSEC-Validating resolvers that implement this mechanism MUST perform validation of responses in accordance with the DNSSEC response validation specification [RFC4035].
This sentinel mechanism makes use of two special labels:[RFC4034], section 5.3), but zero-padded to five digits (for example, a Key Tag value of 42 would be represented in the label as 00042).
Note that the <key-tag> is specified in the DNS label as unsigned decimal integer (as described in
These labels trigger special processing in the validating DNS resolver when responses from authoritative servers are received. Labels containing "root-key-sentinel-is-ta-<key-tag>" is used to answer the question "Is this the Key Tag of a key which the validating DNS resolver is currently trusting as a trust anchor?" Labels containing "root-key-sentinel-not-ta-<key-tag>" is used to answer the question "Is this the Key Tag of a key which the validating DNS resolver is *not* currently trusting as a trust anchor?"
All of the following conditions must be met to trigger special processing inside resolver code:
If any one of the preconditions is not met, the resolver MUST NOT alter the DNS response based on the mechanism in this document.
Responses which fulfil all of the preconditions in Section 2.1 require special processing, depending on leftmost label in the QNAME.
First, the resolver determines if the numerical value of <key-tag> is equal to any of the Key Tag values of an active root zone KSK which is currently trusted by the local resolver and is stored in its store of trusted keys. An active root zone KSK is one which could currently be used for validation (that is, a key that is not in either the AddPend or Revoked state as described in [RFC5011]).
Second, the resolver alters the response being sent to the original query based on both the left-most label and the presence of a key with given Key Tag in the trust anchor store. Two labels and two possible states of the corresponding key generate four possible combinations summarized in the table:
Label | Key is trusted | Key is not trusted ------------------------------------------------------------------ is-ta | return original answer | return SERVFAIL not-ta | return SERVFAIL | return original answer
Instruction "return SERVFAIL" means that the resolver MUST set RCODE=SERVFAIL (value 2) and the ANSWER section of the DNS response MUST be empty, ignoring all other documents which specify content of the ANSWER section.
This section describes the use of the sentinel detection mechanism against a single DNS recursive resolver in order to determine whether this resolver is using a particular trust anchor to validate DNSSEC-signed responses.
Note that the test in this section applies to a single DNS resolver. The test described in Section 4 applies instead to a collection of DNS resolvers, as might be found in the DNS configuration of an end-user environment.
The critical aspect of the DNS names used in this mechanism is that they contain the specified label for either the positive and negative test as the left-most label in the query name.
The sentinel detection procedure can test a DNS resolver using three queries:
The responses received from queries to resolve each of these names can be evaluated to infer a trust key state of the DNS resolver.
An essential assumption here is that this technique relies on security-aware (DNSSEC validating) resolvers responding with a SERVFAIL response code to queries where DNSSEC checking is requested and the response cannot be validated. Note that a slew of other issues can also cause SERVFAIL responses, and so the sentinel processing may sometimes result in incorrect or indeterminate conclusions.
To describe this process of classification, DNS resolvers are classified by five distinct behavior types using the labels: "Vnew", "Vold", "Vind", "nonV", and "other". These labels correspond to resolver system behaviour types as follows:
Note that SERVFAIL might be cached according to Section 7 of [RFC2308] for up to 5 minutes and a positive answer for up to its TTL.
If a client directs these three queries to a single resolver, the responses should allow the client to determine the capability of the resolver, and if it supports this sentinel mechanism, whether or not it has a particular key in its trust anchor store, as in the following table:
Query
+----------+-----------+------------+
| is-ta | not-ta | bogus |
+-------+----------+-----------+------------+
| Vnew | A | SERVFAIL | SERVFAIL |
| Vold | SERVFAIL | A | SERVFAIL |
Type | Vind | A | A | SERVFAIL |
| nonV | A | A | A |
| other | * | * | * |
+-------+----------+-----------+------------+
There is also the common case of a recursive resolver using a forwarder.
If the resolver is non-validating, and it has a single forwarder, then the resolver will presumably mirror the capabilities of the forwarder target resolver.
If the validating resolver has a forwarding configuration, and uses the CD bit on all forwarded queries, then this resolver is acting in a manner that is identical to a standalone resolver.
A more complex case is where all of the following conditions hold:
In such a case, either the outcome is indeterminate validating ("Vind"), or a case of mixed signals such as SERVFAIL in all three responses, ("other") which is similarly an indeterminate response with respect to the trusted key state.
The description in Section 3 describes a trust anchor test that can be used in the simple situation where the test queries were being passed to a single recursive resolver that directly queries authoritative name servers.
However, the common end user scenario is where a user's local DNS resolution environment is configured to use a set of recursive resolvers. The single resolver test technique will not function reliably in such cases, as a a SERVFAIL response from one resolver may cause the local stub resolver to repeat the query against one of the other configured resolvers and the results may be inconclusive.
In describing a test procedure that can be used in this environment of a set of DNS resolvers there are some necessary changes to the nature of the question that this test can answer, the assumptions about the behaviour of the DNS resolution environment, and some further observations about potential variability in the test outcomes.
This test is not intended to expose which trust anchors are used by any single DNS resolver.
The test scenario is explicitly restricted to that of the KSK environment where a current active KSK (called "KSK-current") is to be replaced with a new KSK (called "KSK-new"). The test is designed to be run between when KSK-new is introduced into the root zone and when the root zone is signed with KSK-new.
The objective of the test is to determine if the user will be negatively impacted by the KSK roll. A "negative impact" for the user is defined such that all the configured resolvers are security-aware resolvers that perform validation of DNSSEC-signed responses, and none of these resolvers have loaded KSK-new into their local trust anchor set. In this situation, it is anticipated that once the KSK is rolled the entire set of the user's resolvers will not be able to validate the contents of the root zone and the user is likely to lose DNS service as a result of this inability to perform successful DNSSEC validation.
There are a number of assumptions about the DNS environment used in this test. Where these assumptions do not hold, the results of the test will be indeterminate.
There is no current published measurement data that indicates to what extent the first two assumptions listed here are valid, and how many end users may be impacted by these assumptions. In particular, the first assumption, that a consistent SERFAIL response will cause the local stub DNS resolution environment to query all of its configured recursive resolvers before concluding that the name cannot be resolved, is a very critical assumption for this test.
The sentinel detection process tests a DNS resolution environment with three query names:
The responses received from queries to resolve each of these names can be evaluated to infer a trust key state of the user's DNS resolution environment.
The responses to these queries are described using a simplified notation. Each query will either result in a SERFVAIL response (denoted as "S"), indicating that all of the resolvers in the recursive resolver set returned the SERVFAIL response code, or result in a response with the desire RRset value (denoted as "A"). The queries are ordered by the "invalid" name, the "not-ta" label, then the "is-ta" label, and a triplet notation denotes a particular response. For example, the triplet "(S S A)" denotes a SERVFAIL response to the invalid query, a SERVFAIL response to the "not-ta" query and a RRset response to the "is-ta" query.
The set of all possible responses to these three queries are:
This document describes a mechanism to allow users to determine the trust anchor state of root zone key signing keys in the DNS resolution system that they use. If the user executes third party code, then this information may also be available to the third party.
The mechanism does not require resolvers to set otherwise unauthenticated responses to be marked as authenticated, and does not alter the security properties of DNSSEC with respect to the interpretation of the authenticity of responses that are so marked.
The mechanism does not require any further significant processing of DNS responses, and queries of the form described in this document do not impose any additional load that could be exploited in an attack over the normal DNSSEC validation processing load.
The mechanism in this document enables third parties (with either good or bad intentions) to learn something about the security configuration of recursive DNS resolvers. That is, someone who can cause an Internet user to make specific DNS queries (e.g. via web-based advertisements or javascript in web pages), can, under certain specific circumstances that includes additional knowledge of the resolvers that are invoked by the user, determine which trust anchors are configured in these resolvers. Without this additional knowledge, the third party can infer the aggregate capabilities of the user's DNS resolution environment, but cannot necessarily infer the trust configuration of any recursive name server.
[ RFC Editor: Please remove before publication. As this section will be removed, it is more conversational than would appear in a published doc. ]
List of known resolver implementations (alphabetical):
A (partial) list of "client" / user side implementations (the author was keeping a more complete list of implementations, but has misplaced it - apologies, I'm happy to re-add them if you send me a note.):
[Note to IANA, to be removed prior to publication: there are no IANA considerations stated in this version of the document.]
This document has borrowed extensively from [RFC8145] for the introductory text, and the authors would like to acknowledge and thank the authors of that document both for some text excerpts and for the more general stimulation of thoughts about monitoring the progress of a roll of the KSK of the root zone of the DNS.
The authors would like to thank Joe Abley, Mehmet Akcin, Mark Andrews, Richard Barnes, Ray Bellis, Stephane Bortzmeyer, David Conrad, Ralph Dolmans, John Dickinson, Steinar Haug, Bob Harold, Wes Hardaker, Paul Hoffman, Matt Larson, Jinmei Tatuya, Edward Lewis, George Michaelson, Benno Overeinder, Matthew Pounsett, Hugo Salgado-Hernández, Andreas Schulze, Mukund Sivaraman, Petr Spacek, Job Snijders, Andrew Sullivan, Ondrej Sury, Paul Vixie, Duane Wessels and Paul Wouters for their helpful feedback.
The authors would like to especially call out Paul Hoffman and Duane Wessels for providing comments in the form of a pull request.
RFC Editor: Please remove this section!
Note that this document is being worked on in GitHub - see Abstract. The below is mainly large changes, and is not authoritative.
From -13 to -14:
From -12 to -13:
From -11 to -12:
From -10 to -11:
From -09 to -10:
From -08 to -09:
From -08 to -07:
From -07 to -06:
From -05 to -06:
From -04 to -05:
From -03 to -04:
From -02 to -03:
From -01 to 02:
From -00 to 01:
| [RFC2308] | Andrews, M., "Negative Caching of DNS Queries (DNS NCACHE)", RFC 2308, DOI 10.17487/RFC2308, March 1998. |
| [RFC4033] | Arends, R., Austein, R., Larson, M., Massey, D. and S. Rose, "DNS Security Introduction and Requirements", RFC 4033, DOI 10.17487/RFC4033, March 2005. |
| [RFC4034] | Arends, R., Austein, R., Larson, M., Massey, D. and S. Rose, "Resource Records for the DNS Security Extensions", RFC 4034, DOI 10.17487/RFC4034, March 2005. |
| [RFC4035] | Arends, R., Austein, R., Larson, M., Massey, D. and S. Rose, "Protocol Modifications for the DNS Security Extensions", RFC 4035, DOI 10.17487/RFC4035, March 2005. |
| [RFC5011] | StJohns, M., "Automated Updates of DNS Security (DNSSEC) Trust Anchors", STD 74, RFC 5011, DOI 10.17487/RFC5011, September 2007. |
| [RFC8145] | Wessels, D., Kumari, W. and P. Hoffman, "Signaling Trust Anchor Knowledge in DNS Security Extensions (DNSSEC)", RFC 8145, DOI 10.17487/RFC8145, April 2017. |
This Appendix provides a non-normative example of how the sentinel mechanism could be used, and what each participant does. It is provided in a conversational tone to be easier to follow. The examples here all assume that each person has just one resolver, or a system of resolvers that have the same properties.
Alice is in charge of the DNS root KSK (Key Signing Key), and would like to roll / replace the key with a new one. She publishes the new KSK, but would like to be able to predict / measure what the impact will be before removing/revoking the old key. The current KSK has a Key Tag of 11112, the new KSK has a Key Tag of 02323. Users want to verify that their resolver will not break after Alice rolls the root KSK key (that is, starts signing with just the KSK whose Key Tag is 02323).
Bob, Charlie, Dave, Ed are all users. They use the DNS recursive resolvers supplied by their ISPs. They would like to confirm that their ISPs have picked up the new KSK. Bob's ISP does not perform validation. Charlie's ISP does validate, but the resolvers have not yet been upgraded to support this mechanism. Dave and Ed's resolvers have been upgraded to support this mechanism; Dave's resolver has the new KSK, Ed's resolver hasn't managed to install the 02323 KSK in its trust store yet.
Geoff is a researcher, and would like to both provide a means for Bob, Charlie, Dave and Ed to be able to perform tests, and also would like to be able to perform Internet-wide measurements of what the impact will be (and report this back to Alice).
Geoff sets an authoritative DNS server for example.com, and also a webserver (www.example.com). He adds three address records to example.com:
Note that the use of "example.com" names and the addresses here are examples. In a real deployment, the domain names need to be under control of the researcher, and the addresses must be real, reachable addresses.
Geoff then DNSSEC signs the example.com zone, and intentionally makes the bogus.example.com record have bogus validation status (for example, by editing the signed zone and entering garbage for the signature). Geoff also configures his webserver to listen on 2001:db8::1 and serve a resource (for example, a 1x1 GIF, 1x1.gif) for all of these names. The webserver also serves a webpage (www.example.com) which contains links to these 3 resources (http://bogus.example.com/1x1.gif, http://root-key-sentinel-is-ta-02323.example.com/1x1.gif, http://root-key-sentinel-not-ta-11112.example.com/1x1.gif).
Geoff then asks Bob, Charlie, Dave and Ed to browse to www.example.com. Using the methods described in this document, the users can figure out what their fate will be when the 11112 KSK is removed.
Bob is not using a validating resolver. This means that he will be able to resolve bogus.example.com (and fetch the 1x1 GIF) - this tells him that the KSK roll does not affect him, and so he will be OK.
Charlie's resolvers are validating, but they have not been upgraded to support the KSK sentinel mechanism. Charlie will not be able to fetch the http://bogus.example.com/1x1.gif resource (the bogus.example.com record is bogus, and none of his resolvers will resolve it). He is able to fetch both of the other resources - from this he knows (see the logic in the body of this document) that he is using validating resolvers, but at least one of these resolvers is not configured to perform sentinel processing. The KSK sentinel method cannot provide him with a definitive answer to the question of whether he will be impacted by the KSK roll.
Dave's resolvers implement the sentinel method, and have picked up the new KSK. For the same reason as Charlie, he cannot fetch the "bogus" resource. His resolver resolves the root-key-sentinel-is-ta-02323.example.com name normally (it contacts the example.com authoritative servers, etc); as it supports the sentinel mechanism, just before Dave's recursive resolver sends the reply to Dave's stub, it performs the KSK Sentinel check. The QNAME starts with "root-key-sentinel-is-ta-", and the recursive resolver does indeed have a key with the Key Tag of 02323 in its root trust store. This means that that this part of the KSK Sentinel check passes (it is true that Key Tag 02323 is in the trust anchor store), and the recursive resolver replies normally (with the answer provided by the authoritative server). Dave's recursive resolver then resolves the root-key-sentinel-not-ta-11112.example.com name. Once again, it performs the normal resolution process, but because it implements KSK Sentinel (and the QNAME starts with "root-key-sentinel-not-ta-"), just before sending the reply, it performs the KSK Sentinel check. As it has the key with key-tag 11112 in it's trust anchor store, the answer to "is this *not* a trust anchor" is false, and so the recursive resolver does not reply with the answer from the authoritative server - instead, it replies with a SERVFAIL (note that replying with SERVFAIL instead of the original answer is the only mechanism that KSK Sentinel uses). This means that Dave cannot fetch "bogus", he can fetch "root-key-sentinel-is-ta-02323", but he cannot fetch "root-key-sentinel-not-ta-11112". From this, Dave knows that he is behind an collection of resolvers that all validate, all have the key with key tag 11112 loaded and at least one of these resolvers has loaded the key with key-tag 02323 into its local trust anchor cache, Dave will not be impacted by the KSK roll.
Just like Charlie and Dave, Ed cannot fetch the "bogus" record. This tells him that his resolvers are validating. When his (sentinel-aware) resolvers performs the KSK Sentinel check for "root-key-sentinel-is-ta-02323", none of them have loaded the new key with key-tag 02323 in their local trust anchor store. This means check fails, and Ed's recursive resolver converts the (valid) answer into a SERVFAIL error response. It performs the same check for root-key-sentinel-not-ta-11112.example.com, and as all of Ed's resolvers both perform DNSSEC validation and recognise the sentinel label Ed will be unable to fetch the "root-key-sentinel-not-ta-11112" resource. This tells Ed that his resolvers have not installed the new KSK and he will be negatively implacted by the KSK roll..
Geoff would like to do a large scale test and provide the information back to Alice. He uses some mechanism such as causing users to go to a web page to cause a large number of users to attempt to resolve the three resources, and then analyzes the results of the tests to determine what percentage of users will be affected by the KSK rollover event.
This description is a simplified example - it is not anticipated that Bob, Charlie, Dave and Ed will actually look for the absence or presence of web resources; instead, the webpage that they load would likely contain JavaScript (or similar) which displays the result of the tests, sends the results to Geoff, or both. This sentinel mechanism does not rely on the web: it can equally be used by trying to resolve the names (for example, using the common "dig" command) and checking which result in a SERVFAIL.