idnits 2.17.1 draft-ietf-dnsop-kskroll-sentinel-17.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- == There are 1 instance of lines with non-RFC2606-compliant FQDNs in the document. ** The document seems to lack a both a reference to RFC 2119 and the recommended RFC 2119 boilerplate, even if it appears to use RFC 2119 keywords. RFC 2119 keyword, line 121: '...his mechanism is OPTIONAL to implement...' RFC 2119 keyword, line 122: '..., this mechanism SHOULD be enabled by ...' RFC 2119 keyword, line 123: '...urement. Configuration options MAY be...' RFC 2119 keyword, line 168: '...ers that implement this mechanism MUST...' RFC 2119 keyword, line 221: '...ions is not met, the resolver MUST NOT...' (6 more instances...) Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (October 20, 2018) is 2008 days in the past. Is this intentional? Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Missing Reference: 'RFC2119' is mentioned on line 162, but not defined == Missing Reference: 'RFC8174' is mentioned on line 162, but not defined -- Obsolete informational reference (is this intentional?): RFC 7719 (Obsoleted by RFC 8499) Summary: 1 error (**), 0 flaws (~~), 4 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 DNSOP G. Huston 3 Internet-Draft J. Damas 4 Intended status: Standards Track APNIC 5 Expires: April 23, 2019 W. Kumari 6 Google 7 October 20, 2018 9 A Root Key Trust Anchor Sentinel for DNSSEC 10 draft-ietf-dnsop-kskroll-sentinel-17 12 Abstract 14 The DNS Security Extensions (DNSSEC) were developed to provide origin 15 authentication and integrity protection for DNS data by using digital 16 signatures. These digital signatures can be verified by building a 17 chain of trust starting from a trust anchor and proceeding down to a 18 particular node in the DNS. This document specifies a mechanism that 19 will allow an end user and third parties to determine the trusted key 20 state for the root key of the resolvers that handle that user's DNS 21 queries. Note that this method is only applicable for determining 22 which keys are in the trust store for the root key. 24 [ This document is being collaborated on in Github at: 25 https://github.com/APNIC-Labs/draft-kskroll-sentinel. The most 26 recent version of the document, open issues, etc should all be 27 available here. The authors (gratefully) accept pull requests. RFC 28 Editor, please remove text in square brackets before publication. ] 30 Status of This Memo 32 This Internet-Draft is submitted in full conformance with the 33 provisions of BCP 78 and BCP 79. 35 Internet-Drafts are working documents of the Internet Engineering 36 Task Force (IETF). Note that other groups may also distribute 37 working documents as Internet-Drafts. The list of current Internet- 38 Drafts is at https://datatracker.ietf.org/drafts/current/. 40 Internet-Drafts are draft documents valid for a maximum of six months 41 and may be updated, replaced, or obsoleted by other documents at any 42 time. It is inappropriate to use Internet-Drafts as reference 43 material or to cite them other than as "work in progress." 45 This Internet-Draft will expire on April 23, 2019. 47 Copyright Notice 49 Copyright (c) 2018 IETF Trust and the persons identified as the 50 document authors. All rights reserved. 52 This document is subject to BCP 78 and the IETF Trust's Legal 53 Provisions Relating to IETF Documents 54 (https://trustee.ietf.org/license-info) in effect on the date of 55 publication of this document. Please review these documents 56 carefully, as they describe your rights and restrictions with respect 57 to this document. Code Components extracted from this document must 58 include Simplified BSD License text as described in Section 4.e of 59 the Trust Legal Provisions and are provided without warranty as 60 described in the Simplified BSD License. 62 Table of Contents 64 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 65 1.1. Terminology . . . . . . . . . . . . . . . . . . . . . . . 4 66 2. Sentinel Mechanism in Resolvers . . . . . . . . . . . . . . . 4 67 2.1. Preconditions . . . . . . . . . . . . . . . . . . . . . . 5 68 2.2. Special Processing . . . . . . . . . . . . . . . . . . . 5 69 3. Sentinel Tests for a Single DNS Resolver . . . . . . . . . . 6 70 3.1. Forwarders . . . . . . . . . . . . . . . . . . . . . . . 9 71 4. Sentinel Tests for Multiple Resolvers . . . . . . . . . . . . 10 72 4.1. Test Scenario and Objective . . . . . . . . . . . . . . . 10 73 4.2. Test Assumptions . . . . . . . . . . . . . . . . . . . . 10 74 4.3. Test Procedure . . . . . . . . . . . . . . . . . . . . . 11 75 5. Security Considerations . . . . . . . . . . . . . . . . . . . 12 76 6. Privacy Considerations . . . . . . . . . . . . . . . . . . . 13 77 7. Implementation Experience . . . . . . . . . . . . . . . . . . 13 78 8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 14 79 9. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 14 80 10. Change Log . . . . . . . . . . . . . . . . . . . . . . . . . 15 81 11. References . . . . . . . . . . . . . . . . . . . . . . . . . 19 82 11.1. Normative References . . . . . . . . . . . . . . . . . . 19 83 11.2. Informative References . . . . . . . . . . . . . . . . . 19 84 Appendix A. Protocol Walkthrough Example . . . . . . . . . . . . 20 85 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 23 87 1. Introduction 89 The DNS Security Extensions (DNSSEC) [RFC4033], [RFC4034] and 90 [RFC4035] were developed to provide origin authentication and 91 integrity protection for DNS data by using digital signatures. 92 DNSSEC uses Key Tags to efficiently match signatures to the keys from 93 which they are generated. The Key Tag is a 16-bit value computed 94 from the RDATA of a DNSKEY RR as described in Appendix B of 96 [RFC4034]. RRSIG RRs contain a Key Tag field whose value is equal to 97 the Key Tag of the DNSKEY RR that was used to generate the 98 corresponding signature. 100 This document specifies how security-aware DNS resolvers that perform 101 validation of their responses can respond to certain queries in a 102 manner that allows an agent performing the queries to deduce whether 103 a particular key for the root has been loaded into that resolver's 104 trusted key store. This document also describes a procedure where a 105 collection of resolvers can be tested to determine if at least one of 106 these resolvers has loaded a given key into its trusted key store. 107 These tests can be used to determine whether a certain root zone Key 108 Signing Key (KSK) is ready to be used as a trusted key, within the 109 context of a planned root zone KSK key roll. 111 There are two primary use cases for this mechanism: 113 o Users may wish to ascertain whether their DNS resolution 114 environment's resolver is ready for an upcoming root KSK rollover. 116 o Researchers want to perform Internet-wide studies about the 117 proportion of users who will be negatively impacted by an upcoming 118 root KSK rollover. 120 The mechanism described in this document satisfy the requirements of 121 both these use-cases. This mechanism is OPTIONAL to implement and 122 use. If implemented, this mechanism SHOULD be enabled by default to 123 facilitate Internet-wide measurement. Configuration options MAY be 124 provided to disable the mechanism for reasons of local policy. 126 The KSK sentinel tests described in this document use a test 127 comprising of a set of DNS queries to domain names that have special 128 values for the left-most label. The test relies on recursive 129 resolvers supporting a mechanism that recognises this special name 130 pattern in queries, and under certain defined circumstances will 131 return a DNS SERVFAIL response code (RCODE 2), mimicking the response 132 code that is returned by security-aware resolvers when DNSSEC 133 validation fails. 135 If a browser or operating system is configured with multiple 136 resolvers, and those resolvers have different properties (for 137 example, one performs DNSSEC validation and one does not), the 138 sentinel test described in this document can still be used. The 139 sentinel test makes a number of assumptions about DNS resolution 140 behaviour that may not necessarily hold in all environments; if these 141 assumptions do not hold (such as, for example, requiring the stub 142 resolver to query the next recursive resolver in the locally 143 configured set upon receipt of a SERVFAIL response code) then this 144 test may produce indeterminate or inconsistent results. In some 145 cases where these assumptions do not hold, repeating the same test 146 query set may generate different results. 148 Note that the measurements facilitated by the mechanism described in 149 this document are different from those of [RFC8145]. RFC 8145 relies 150 on resolvers reporting towards the root servers a list of locally 151 cached trust anchors for the root zone. Those reports can be used to 152 infer how many resolvers may be impacted by a KSK roll, but not what 153 the user impact of the KSK roll will be. 155 1.1. Terminology 157 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 158 "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and 159 "OPTIONAL" in this document are to be interpreted as described in BCP 160 14 [RFC2119] [RFC8174] when, and only when, they appear in all 161 capitals, as shown here. 163 This document contains a number of terms related to the DNS. The 164 current definitions of these terms can be found in [RFC7719]. 166 2. Sentinel Mechanism in Resolvers 168 DNSSEC-Validating resolvers that implement this mechanism MUST 169 perform validation of responses in accordance with the DNSSEC 170 response validation specification [RFC4035]. 172 This sentinel mechanism makes use of two special labels: 174 o root-key-sentinel-is-ta- 176 o root-key-sentinel-not-ta- 178 These labels trigger special processing in the validating DNS 179 resolver when responses from authoritative servers are received. 180 Labels containing "root-key-sentinel-is-ta-" is used to 181 answer the question "Is this the Key Tag of a key which the 182 validating DNS resolver is currently trusting as a trust anchor?" 183 Labels containing "root-key-sentinel-not-ta-" is used to 184 answer the question "Is this the Key Tag of a key which the 185 validating DNS resolver is *not* currently trusting as a trust 186 anchor?". 188 The special labels defined here came after extensive IETF evaluation 189 of alternative patterns and approaches in light of the desired 190 behaviour (sections 2.1, 2.2) within the resolver and the applied 191 testing methodology (section 4.3). As one example, underscore 192 prefixed names were rejected because some browsers and operating 193 systems would not fetch them because they domain names but not valid 194 hostnames (see [RFC7719] for these definitions). Attention was paid 195 to the consideration of local collisions and the reservation of 196 leftmost labels of a domain name, and the impact upon zone operators 197 who might desire to use a similarly constructed hostname for a 198 purpose other than as documented here. Therefore, it is important to 199 note that the reservation of the labels in this manner is definitely 200 not considered "best practice". 202 2.1. Preconditions 204 All of the following conditions must be met to trigger special 205 processing inside resolver code: 207 o The DNS response is DNSSEC validated. 209 o The result of validation is "Secure". 211 o The EDNS(0) Checking Disabled (CD) bit in the query is not set. 213 o The QTYPE is either A or AAAA (Query Type value 1 or 28). 215 o The OPCODE is QUERY. 217 o The leftmost label of the original QNAME (the name sent in the 218 Question Section in the original query) is either "root-key- 219 sentinel-is-ta-" or "root-key-sentinel-not-ta-". 221 If any one of the preconditions is not met, the resolver MUST NOT 222 alter the DNS response based on the mechanism in this document. 224 Note that the is specified in the DNS label as unsigned 225 decimal integer (as described in [RFC4034], section 5.3), but zero- 226 padded to five digits (for example, a Key Tag value of 42 would be 227 represented in the label as 00042). The precise specification of the 228 special labels above should be followed exactly. For example, a 229 label that does not include a Key Tag zero-padded to five digits does 230 not match this specification, and should not be processed as if they 231 did -- in other words, such queries should be handled as any other 232 label and not according to Section 2.2. 234 2.2. Special Processing 236 Responses which fulfil all of the preconditions in Section 2.1 237 require special processing, depending on leftmost label in the QNAME. 239 First, the resolver determines if the numerical value of is 240 equal to any of the Key Tag values of an active root zone KSK which 241 is currently trusted by the local resolver and is stored in its store 242 of trusted keys. An active root zone KSK is one which could 243 currently be used for validation (that is, a key that is not in 244 either the AddPend or Revoked state as described in [RFC5011]). 246 Second, the resolver alters the response being sent to the original 247 query based on both the left-most label and the presence of a key 248 with given Key Tag in the trust anchor store. Two labels and two 249 possible states of the corresponding key generate four possible 250 combinations summarized in the table: 252 Label | Key is trusted | Key is not trusted 253 ------------------------------------------------------------------ 254 is-ta | return original answer | return SERVFAIL 255 not-ta | return SERVFAIL | return original answer 257 Instruction "return SERVFAIL" means that the resolver MUST set 258 RCODE=SERVFAIL (value 2) and the ANSWER section of the DNS response 259 MUST be empty, ignoring all other documents which specify content of 260 the ANSWER section. 262 Instruction "return original answer" means that the resolver MUST 263 process the query without any further special processing; that is, 264 exactly as if the mechanism described in this document was not 265 implemented or was disabled. The answer for the A or AAAA query is 266 sent on to the client. 268 3. Sentinel Tests for a Single DNS Resolver 270 This section describes the use of the sentinel detection mechanism 271 against a single DNS recursive resolver in order to determine whether 272 this resolver is using a particular trust anchor to validate DNSSEC- 273 signed responses. 275 Note that the test in this section applies to a single DNS resolver. 276 The test described in Section 4 applies instead to a collection of 277 DNS resolvers, as might be found in the DNS configuration of an end- 278 user environment. 280 The critical aspect of the DNS names used in this mechanism is that 281 they contain the specified label for either the positive and negative 282 test as the left-most label in the query name. 284 The sentinel detection procedure can test a DNS resolver using three 285 queries: 287 o A query name containing the left-most label "root-key-sentinel-is- 288 ta-". This corresponds to a validly-signed name in the 289 parent zone, so that responses associated with this query name can 290 be authenticated by a DNSSEC-validating resolver. Any validly- 291 signed DNS zone can be used as the parent zone for this test. 293 o A query name containing the left-most label "root-key-sentinel- 294 not-ta-". This also corresponds to a validly-signed 295 name. Any validly-signed DNS zone can be used as the parent zone 296 for this test. 298 o A query name that is signed with a DNSSEC signature that cannot be 299 validated (described as a "bogus" RRset in Section 5 of [RFC4033], 300 when, for example, an RRset associated with a zone that is not 301 signed with a valid RRSIG record). 303 The responses received from queries to resolve each of these query 304 names can be evaluated to infer a trust key state of the DNS 305 resolver. 307 An essential assumption here is that this technique relies on 308 security-aware (DNSSEC validating) resolvers responding with a 309 SERVFAIL response code to queries where DNSSEC checking is requested 310 and the response cannot be validated. Note that other issues can 311 also cause a resolver to return SERVFAIL responses, and so the 312 sentinel processing may sometimes result in incorrect or 313 indeterminate conclusions. 315 To describe this process of classification, DNS resolvers are 316 classified by five distinct behavior types using the labels: "Vnew", 317 "Vold", "Vind", "nonV", and "other". These labels correspond to 318 resolver system behaviour types as follows: 320 Vnew: A DNS resolver that is configured to implement this mechanism 321 and has loaded the nominated key into their local trusted key 322 stores will respond with an A or AAAA RRset response for the 323 associated "root-key-sentinel-is-ta" queries, SERVFAIL for "root- 324 key-sentinel-not-ta" queries and SERVFAIL for the signed name 325 queries that return "bogus" validation status. 327 Vold: A DNS resolver that is configured to implement this mechanism 328 and has not loaded the nominated key into their local trusted key 329 stores will respond with an SERVFAIL for the associated "root-key- 330 sentinel-is-ta" queries, an A or AAAA RRset response for "root- 331 key-sentinel-not-ta" queries and SERVFAIL for the signed name 332 queries that return "bogus" validation status. 334 Vind: A DNS resolver that has is not configured to implement this 335 mechanism will respond with an A or AAAA RRset response for "root- 336 key-sentinel-is-ta", an A or AAAA RRset response for "root-key- 337 sentinel-not-ta" and SERVFAIL for the name that returns "bogus" 338 validation status. This set of responses does not give any 339 information about the trust anchors used by this resolver. 341 nonV: A non-security-aware DNS resolver will respond with an A or 342 AAAA RRset response for "root-key-sentinel-is-ta", an A or AAAA 343 RRset response for "root-key-sentinel-not-ta" and an A or AAAA 344 RRset response for the name that returns "bogus" validation 345 status. 347 other: There is the potential to admit other combinations of 348 responses to these three queries. While this may appear self- 349 contradictory, there are cases where such an outcome is possible. 350 For example, in DNS resolver farms what appears to be a single DNS 351 resolver that responds to queries passed to a single IP address is 352 in fact constructed as a a collection of slave resolvers, and the 353 query is passed to one of these internal resolver engines. If 354 these individual slave resolvers in the farm do not behave 355 identically, then other sets of results can be expected from these 356 three queries. In such a case, no determination about the 357 capabilities of this DNS resolver farm can be made. 359 Note that SERVFAIL might be cached according to Section 7 of 360 [RFC2308] for up to 5 minutes and a positive answer for up to its 361 TTL. 363 If a client directs these three queries to a single resolver, the 364 responses should allow the client to determine the capability of the 365 resolver, and if it supports this sentinel mechanism, whether or not 366 it has a particular key in its trust anchor store, as in the 367 following table: 369 Query 370 +----------+-----------+------------+ 371 | is-ta | not-ta | bogus | 372 +-------+----------+-----------+------------+ 373 | Vnew | Y | SERVFAIL | SERVFAIL | 374 | Vold | SERVFAIL | Y | SERVFAIL | 375 Type | Vind | Y | Y | SERVFAIL | 376 | nonV | Y | Y | Y | 377 | other | * | * | * | 378 +-------+----------+-----------+------------+ 380 In this table, the "Y" response denotes an A or AAAA RRset response 381 (depending on the query type of A or AAAA records), "SERVFAIL" 382 denotes a DNS SERVFAIL response code (RCODE 2), and "*" denotes 383 either response. 385 Vnew: The nominated key is trusted by the resolver. 387 Vold: The nominated key is not yet trusted by the resolver. 389 Vind: There is no information about the trust anchors of the 390 resolver. 392 nonV: The resolver does not perform DNSSEC validation. 394 other: The properties of the resolver cannot be analyzed by this 395 protocol. 397 3.1. Forwarders 399 Some resolvers are configured not to answer queries using the 400 recursive algorithm first described in [RFC1034] section 4.3.2, but 401 instead relay queries to one or more other resolvers. Resolvers 402 configured in this manner are referred to in this document as 403 "forwarders". 405 If the resolver is non-validating, and it has a single forwarder, 406 then the resolver will presumably mirror the capabilities of the 407 forwarder's target resolver. 409 If the validating resolver has a forwarding configuration, and it 410 sets the EDNS(0) Checking Disabled (CD) bit as described in 411 Section 3.2.2 of [RFC4035] on all forwarded queries, then this 412 resolver is acting in a manner that is identical to a standalone 413 resolver. 415 A more complex case is where all of the following conditions hold: 417 o Both the validating resolver and the forwarder target resolver 418 support this trusted key sentinel mechanism 420 o The local resolver's queries do not have the EDNS(0) CD bit set 422 o The trusted key state differs between the forwarding resolver and 423 the forwarder's target resolver 425 In such a case, either the outcome is indeterminate validating 426 ("Vind"), or a case of mixed signals such as SERVFAIL in all three 427 responses, ("other") which is similarly an indeterminate response 428 with respect to the trusted key state. 430 4. Sentinel Tests for Multiple Resolvers 432 The description in Section 3 describes a trust anchor test that can 433 be used in the simple situation where the test queries were being 434 passed to a single recursive resolver that directly queried 435 authoritative name servers. 437 However, the common end-user scenario is where a user's local DNS 438 resolution environment is configured to use more than one recursive 439 resolver. The single resolver test technique will not function 440 reliably in such cases, as a a SERVFAIL response from one resolver 441 may cause the local stub resolver to repeat the query against one of 442 the other configured resolvers and the results may be inconclusive. 444 In describing a test procedure that can be used in this environment 445 of a set of DNS resolvers there are some necessary changes to the 446 nature of the question that this test can answer, the assumptions 447 about the behaviour of the DNS resolution environment, and some 448 further observations about potential variability in the test 449 outcomes. 451 4.1. Test Scenario and Objective 453 This test is not intended to expose which trust anchors are used by 454 any single DNS resolver. 456 The test scenario is explicitly restricted to that of the KSK 457 environment where a current active KSK (called "KSK-current") is to 458 be replaced with a new KSK (called "KSK-new"). The test is designed 459 to be run between when KSK-new is introduced into the root zone and 460 when the root zone is signed with KSK-new. 462 The objective of the test is to determine if the user will be 463 negatively impacted by the KSK roll. A "negative impact" for the 464 user is defined such that all the configured resolvers are security- 465 aware resolvers that perform validation of DNSSEC-signed responses, 466 and none of these resolvers have loaded KSK-new into their local 467 trust anchor set. In this situation, it is anticipated that once the 468 KSK is rolled the entire set of the user's resolvers will not be able 469 to validate the contents of the root zone and the user is likely to 470 lose DNS service as a result of this inability to perform successful 471 DNSSEC validation. 473 4.2. Test Assumptions 475 There are a number of assumptions about the DNS environment used in 476 this test. Where these assumptions do not hold, the results of the 477 test will be indeterminate. 479 o When a recursive resolver returns SERVFAIL to the user's stub 480 resolver, the stub resolver will send the same query to the next 481 resolver in the locally configured resolver set. It will continue 482 to do this until it gets a non-SERVFAIL response or until it runs 483 out of resolvers to try. 485 o When the user's stub resolver passes a query to a resolver in the 486 configured resolver set, it will get a consistent answer over the 487 timeframe of the queries. This assumption implies that if the 488 same query is asked by the same stub resolver multiple times in 489 succession to the same recursive resolver, the recursive 490 resolver's response will be the same for each of these queries. 492 o All DNSSEC-validating resolvers have KSK-current in their local 493 trust anchor cache. 495 There is no current published measurement data that indicates to what 496 extent the first two assumptions listed here are valid, and how many 497 end users may be impacted by these assumptions. In particular, the 498 first assumption, that a consistent SERFAIL response will cause the 499 local stub DNS resolution environment to query all of its configured 500 recursive resolvers before concluding that the name cannot be 501 resolved, is a very critical assumption for this test. 503 Note that additional precision / determinism may be achievable by 504 bypassing the normal OS behavior and explicitly testing using each 505 configured recursive resolver (e.g using 'dig'). 507 4.3. Test Procedure 509 The sentinel detection process tests a DNS resolution environment 510 with three query names. Note that these same general categories of 511 query as in Section 3 but the key tag used is different for some 512 queries: 514 o A query name that is signed with a DNSSEC signature that cannot be 515 validated (described as a "bogus" RRset in Section 5 of [RFC4033], 516 when, for example, an RRset is not signed with a valid RRSIG 517 record). 519 o A query name containing the left-most label "root-key-sentinel- 520 not-ta-". This name MUST be a validly- 521 signed name. Any validly-signed DNS zone can be used for this 522 test. 524 o A query name containing the left-most label "root-key-sentinel-is- 525 ta-". This name MUST be a validly-signed 526 name. Any validly-signed DNS zone can be used for this test. 528 The responses received from queries to resolve each of these names 529 can be evaluated to infer a trust key state of the user's DNS 530 resolution environment. 532 The responses to these queries are described using a simplified 533 notation. Each query will either result in a SERFVAIL response 534 (denoted as "S"), indicating that all of the resolvers in the 535 recursive resolver set returned the SERVFAIL response code, or result 536 in a response with the desire RRset value (denoted as "A"). The 537 queries are ordered by the "invalid" name, the "root-key-sentinel- 538 not-ta" label, then the "root-key-sentinel-is-ta" label, and a 539 triplet notation denotes a particular response. For example, the 540 triplet "(S S A)" denotes a SERVFAIL response to the invalid query, a 541 SERVFAIL response to the "root-key-sentinel-not-ta" query and a RRset 542 response to the "root-key-sentinel-is-ta" query. 544 The set of all possible responses to these three queries are: 546 (A * *): If any resolver returns an "A" response for the query for 547 the invalid name, then the resolver set contains at least one non- 548 validating DNS resolver, and the user will not be impacted by the 549 KSK roll. 551 (S A *): If any of the resolvers returns an "A" response the the 552 "root-key-sentinel-not-ta" query, then at least one of the 553 resolvers does not recognise the sentinel mechanism, and the 554 behaviour of the collection of resolvers during the KSK roll 555 cannot be reliably determined. 557 (S S A): This case implies that all of the resolvers in the set 558 perform DNSSEC-validation, all of the resolvers are aware of the 559 sentinel mechanism, and at least one resolver has loaded KSK-new 560 as a local trust anchor. The user will not be impacted by the KSK 561 roll. 563 (S S S): This case implies that all of the resolvers in the set 564 perform DNSSEC-validation, all of the resolvers are aware of the 565 sentinel mechanism, and none of the resolvers has loaded KSK-new 566 as a local trust anchor. The user will be negatively impacted by 567 the KSK roll. 569 5. Security Considerations 571 This document describes a mechanism to allow users to determine the 572 trust anchor state of root zone key signing keys in the DNS 573 resolution system that they use. If the user executes third party 574 code, then this information may also be available to the third party. 576 The mechanism does not require resolvers to set otherwise 577 unauthenticated responses to be marked as authenticated, and does not 578 alter the security properties of DNSSEC with respect to the 579 interpretation of the authenticity of responses that are so marked. 581 The mechanism does not require any further significant processing of 582 DNS responses, and queries of the form described in this document do 583 not impose any additional load that could be exploited in an attack 584 over the normal DNSSEC validation processing load. 586 6. Privacy Considerations 588 The mechanism in this document enables third parties (with either 589 good or bad intentions) to learn something about the security 590 configuration of recursive DNS resolvers. That is, someone who can 591 cause an Internet user to make specific DNS queries (e.g. via web- 592 based advertisements or javascript in web pages), can, under certain 593 specific circumstances that includes additional knowledge of the 594 resolvers that are invoked by the user, determine which trust anchors 595 are configured in these resolvers. Without this additional 596 knowledge, the third party can infer the aggregate capabilities of 597 the user's DNS resolution environment, but cannot necessarily infer 598 the trust configuration of any recursive name server. 600 7. Implementation Experience 602 [ RFC Editor: Please remove before publication. As this section will 603 be removed, it is more conversational than would appear in a 604 published doc. ] 606 List of known resolver implementations (alphabetical): 608 BIND Ondrej Sury of ISC reported to the DNSOP Working Group in 609 April 2018 that this technique was peer-reviewed and merged into 610 BIND master branch with the intent to backport the feature into 611 older release branches. The merge request: 612 https://gitlab.isc.org/isc-projects/bind9/merge_requests/123 613 Information on configuring this can be found in the BIND 9.13.0 614 Administrator Reference Manual (ARM), available at 615 https://ftp.isc.org/isc/bind9/9.13.0/doc/arm/Bv9ARM.pdf 617 Knot resolver Petr Spacek implemented early versions of this 618 technique into the Knot resolver, identified a number of places 619 where it wasn't clear, and provided very helpful text to address 620 these issues and make the document mode clear. Petr also 621 identified an embarrassingly large number of typos (and similar) 622 in the ksk-test setup. More information is at http://knot- 623 resolver.readthedocs.io/en/stable/modules.html#sentinel-for- 624 detecting-trusted-keys 626 Unbound Benno Overeinder of NLnet Labs reported to the DNSOP Working 627 Group in April 2018 an intention to support this technique in 628 Unbound in the near future. This is now implemented in Unbound 629 version 1.7.1, available from http://unbound.nlnetlabs.nl/ 630 download.html . Configuration information is at 631 http://unbound.nlnetlabs.nl/documentation/unbound.conf.html 633 A (partial) list of "client" / user side implementations (the author 634 was keeping a more complete list of implementations, but has 635 misplaced it - apologies, I'm happy to re-add them if you send me a 636 note.): 638 http://www.ksk-test.net An Javascript implementation of the client 639 side of this protocol is available at: http://www.ksk-test.net 641 http://test.kskroll.dnssec.lab.nic.cl/ Hugo Salgado-Hernandez has 642 created an implementation at 643 http://test.kskroll.dnssec.lab.nic.cl/ 645 http://sentinel.research.icann.org/ The code for this implementation 646 is published at https://github.com/paulehoffman/sentinel-testbed 648 http://www.bellis.me.uk/sentinel/ Ray Bellis client implementation - 649 http://www.bellis.me.uk/sentinel/ 651 8. IANA Considerations 653 This document has no IANA actions. 655 9. Acknowledgements 657 This document has borrowed extensively from [RFC8145] for the 658 introductory text, and the authors would like to acknowledge and 659 thank the authors of that document both for some text excerpts and 660 for the more general stimulation of thoughts about monitoring the 661 progress of a roll of the KSK of the root zone of the DNS. 663 The authors would like to thank Joe Abley, Mehmet Akcin, Mark 664 Andrews, Richard Barnes, Ray Bellis, Stephane Bortzmeyer, David 665 Conrad, Ralph Dolmans, John Dickinson, Steinar Haug, Bob Harold, Wes 666 Hardaker, Paul Hoffman, Matt Larson, Jinmei Tatuya, Edward Lewis, 667 George Michaelson, Benno Overeinder, Matthew Pounsett, Hugo Salgado- 668 Hernandez, Andreas Schulze, Mukund Sivaraman, Petr Spacek, Job 669 Snijders, Andrew Sullivan, Ondrej Sury, Paul Vixie, Duane Wessels and 670 Paul Wouters for their helpful feedback. 672 The authors would like to especially call out Paul Hoffman and Duane 673 Wessels for providing comments in the form of pull requests. Joe 674 Abley also helpfully provided extensive review and OLD / NEW text. 676 Petr Spacek wrote some very early implementations, and provided 677 significant feedback (including pointing out when the test bed didn't 678 match the document!) 680 10. Change Log 682 RFC Editor: Please remove this section! 684 Note that this document is being worked on in GitHub - see Abstract. 685 The below is mainly large changes, and is not authoritative. 687 From -16 to -17: 689 o Thank to Paul Hoffman for "Lots of editorial fixes for post-IESG 690 draft" ( https://github.com/APNIC-Labs/draft-kskroll-sentinel/ 691 pull/28 ) 693 o Repeat after me: Do not drive git while on cold meds... 695 From -15 to -16: 697 o Addressed IESG comments 699 o Benjamin Kaduk's Discuss on draft-ietf-dnsop-kskroll-sentinel 701 o Also added Terry's "This a bad design pattern, but we decided the 702 benefits outweigh the costs this time." text. 704 o Suggestion from Adam to clarify that bypassing e.g gethostbyname() 705 can provide better testing. 707 o Nit: Forgot 'name' in 'This name MUST be a validly-signed name.' 709 o Clarified that 'bogus.example.com' is intentionally DNSSEC bogus / 710 invalid. 712 From -14 to -15: 714 o Addressed Joe Abley's thorough review, at: 715 https://mailarchive.ietf.org/arch/msg/dnsop/8ZnN1xj55Yimet2cg- 716 LrdoJafEA 718 From -13 to -14: 720 o Addressed nits from Bob Harold - 721 https://mailarchive.ietf.org/arch/msg/dnsop/ 722 j4Serw0z24o470AnlD8ISo8o9k4 724 o Formatting changes (and a bit more text) in the implementation 725 section. 727 o Closes PR #21: Clarify indeterminate and resolution systems, 729 o Closes PR #22: Updates to -13 describing the test procedure for a 730 set of resolvers 732 o Closes PR #23: Fix sundry typos, 734 o Closes PR #24: Editorial and clarifications to the new text 736 o Closes PR #25: Clarified when the test can be run 738 From -12 to -13: 740 o Merged Paul Hoffmans PR#19, PR#20. 742 o Moved toy ksk-test.net to implementation section. 744 o Split the test procedures between the test of a single DNS 745 resolvers and the test of a collection of DNS resolvers as would 746 be found in an end user environment. 748 From -11 to -12: 750 o Moved the Walkthrough Example to the end of the document as an 751 appendix. 753 o Incorporated changes as proposed by Ondrej Sury, relating to a 754 consistent use of Key Tag and a reference to the definition of a 755 Bogus RRset. 757 o Corrected minor typos. 759 o Revised the Privacy Considerations. 761 o In response to a request from DNSOP Working Group chairs, a 762 section on reported Implementation Experience has been added, 763 based on postings to the DNSOP Working Group mailing list. 765 From -10 to -11: 767 o Clarified the preconditions for this mechanism as per Working 768 Group mailing list discussion. 770 o Corrected minor typo. 772 From -09 to -10: 774 o Clarified the precondition list to specify that the resolver had 775 performed DNSSEC-validation by setting the AD bit in the response 777 o Clarified the language referring to the operation of RFC8145 778 signalling. 780 From -08 to -09: 782 o Incorporated Paul Hoffman's PR # 15 (Two issues from the 783 Hackathon) - https://github.com/APNIC-Labs/draft-kskroll-sentinel/ 784 pull/15 786 o Clarifies that the match is on the *original* QNAME. 788 From -08 to -07: 790 o Changed title from "A Sentinel for Detecting Trusted Keys in 791 DNSSEC" to "A Root Key Trust Anchor Sentinel for DNSSEC". 793 o Changed magic string from "kskroll-sentinel-" to "root-key- 794 sentinel-" -- this time for sure, Rocky! 796 From -07 to -06: 798 o Addressed GitHub PR #14: Clarifications regarding caching and 799 SERVFAIL responses 801 o Addressed GitHub PR #12, #13: Clarify situation with multiple 802 resolvers, Fix editorial nits. 804 From -05 to -06: 806 o Paul improved my merging of Petr's text to make it more readable. 807 Minor change, but this is just before the cut-off, so I wanted it 808 maximally readable. 810 From -04 to -05: 812 o Incorporated Duane's #10 813 o Integrated Petr Spacek's Issue - https://github.com/APNIC-Labs/ 814 draft-kskroll-sentinel/issues/9 (note that commit-log incorrectly 815 referred to Duane's PR as number 9, it is actually 10). 817 From -03 to -04: 819 o Addressed GitHub pull requests #4, #5, #6, #7 #8. 821 o Added Duane's privacy concerns 823 o Makes the use cases clearer 825 o Fixed some A/AAAA stuff 827 o Changed the example numbers 829 o Made it clear that names and addresses must be real 831 From -02 to -03: 833 o Integrated / published comments from Paul in GitHub PR #2 - 834 https://github.com/APNIC-Labs/draft-kskroll-sentinel/pull/2 836 o Made the Key Tag be decimal, not hex (thread / consensus in 837 https://mailarchive.ietf.org/arch/msg/dnsop/ 838 Kg7AtDhFRNw31He8n0_bMr9hBuE ) 840 From -01 to 02: 842 o Removed Address Record definition. 844 o Clarified that many things can cause SERVFAIL. 846 o Made examples FQDN. 848 o Fixed a number of typos. 850 o Had accidentally said that Charlie was using a non-validating 851 resolver in example. 853 o [ TODO(WK): Doc says Key Tags are hex, is this really what the WG 854 wants? ] 856 o And active key is one that can be used *now* (not e.g AddPend) 858 From -00 to 01: 860 o Added a conversational description of how the system is intended 861 to work. 863 o Clarification that this is for the root. 865 o Changed the label template from _is-ta- to kskroll- 866 sentinel-is-ta-. This is because BIND (at least) will 867 not allow records which start with an underscore to have address 868 records (CNAMEs, yes, A/AAAA no). Some browsers / operating 869 systems also will not fetch resources from names which start with 870 an underscore. 872 11. References 874 11.1. Normative References 876 [RFC1034] Mockapetris, P., "Domain names - concepts and facilities", 877 STD 13, RFC 1034, DOI 10.17487/RFC1034, November 1987, 878 . 880 [RFC2308] Andrews, M., "Negative Caching of DNS Queries (DNS 881 NCACHE)", RFC 2308, DOI 10.17487/RFC2308, March 1998, 882 . 884 [RFC4033] Arends, R., Austein, R., Larson, M., Massey, D., and S. 885 Rose, "DNS Security Introduction and Requirements", 886 RFC 4033, DOI 10.17487/RFC4033, March 2005, 887 . 889 [RFC4034] Arends, R., Austein, R., Larson, M., Massey, D., and S. 890 Rose, "Resource Records for the DNS Security Extensions", 891 RFC 4034, DOI 10.17487/RFC4034, March 2005, 892 . 894 [RFC4035] Arends, R., Austein, R., Larson, M., Massey, D., and S. 895 Rose, "Protocol Modifications for the DNS Security 896 Extensions", RFC 4035, DOI 10.17487/RFC4035, March 2005, 897 . 899 [RFC5011] StJohns, M., "Automated Updates of DNS Security (DNSSEC) 900 Trust Anchors", STD 74, RFC 5011, DOI 10.17487/RFC5011, 901 September 2007, . 903 11.2. Informative References 905 [RFC7719] Hoffman, P., Sullivan, A., and K. Fujiwara, "DNS 906 Terminology", RFC 7719, DOI 10.17487/RFC7719, December 907 2015, . 909 [RFC8145] Wessels, D., Kumari, W., and P. Hoffman, "Signaling Trust 910 Anchor Knowledge in DNS Security Extensions (DNSSEC)", 911 RFC 8145, DOI 10.17487/RFC8145, April 2017, 912 . 914 Appendix A. Protocol Walkthrough Example 916 This Appendix provides a non-normative example of how the sentinel 917 mechanism could be used, and what each participant does. It is 918 provided in a conversational tone to be easier to follow. The 919 examples here all assume that each person has just one resolver, or a 920 system of resolvers that have the same properties. 922 Alice is in charge of the DNS root KSK (Key Signing Key), and would 923 like to roll / replace the key with a new one. She publishes the new 924 KSK, but would like to be able to predict / measure what the impact 925 will be before removing/revoking the old key. The current KSK has a 926 Key Tag of 11112, the new KSK has a Key Tag of 02323. Users want to 927 verify that their resolver will not break after Alice rolls the root 928 KSK key (that is, starts signing with just the KSK whose Key Tag is 929 02323). 931 Bob, Charlie, Dave, Ed are all users. They use the DNS recursive 932 resolvers supplied by their ISPs. They would like to confirm that 933 their ISPs have picked up the new KSK. Bob's ISP does not perform 934 validation. Charlie's ISP does validate, but the resolvers have not 935 yet been upgraded to support this mechanism. Dave and Ed's resolvers 936 have been upgraded to support this mechanism; Dave's resolver has the 937 new KSK, Ed's resolver hasn't managed to install the 02323 KSK in its 938 trust store yet. 940 Geoff is a researcher, and would like to both provide a means for 941 Bob, Charlie, Dave and Ed to be able to perform tests, and also would 942 like to be able to perform Internet-wide measurements of what the 943 impact will be (and report this back to Alice). 945 Geoff sets an authoritative DNS server for example.com, and also a 946 webserver (www.example.com). He adds three address records to 947 example.com: 949 bogus.example.com. IN AAAA 2001:db8::1 951 root-key-sentinel-is-ta-02323.example.com. IN AAAA 2001:db8::1 953 root-key-sentinel-not-ta-11112.example.com. IN AAAA 2001:db8::1 955 Note that the use of "example.com" names and the addresses here are 956 examples, and "bogus" intentionally has invalid DNSSEC signatures. 958 In a real deployment, the domain names need to be under control of 959 the researcher, and the addresses must be real, reachable addresses. 961 Geoff then DNSSEC signs the example.com zone, and intentionally makes 962 the bogus.example.com record have bogus validation status (for 963 example, by editing the signed zone and entering garbage for the 964 signature). Geoff also configures his webserver to listen on 965 2001:db8::1 and serve a resource (for example, a 1x1 GIF, 1x1.gif) 966 for all of these names. The webserver also serves a webpage 967 (www.example.com) which contains links to these 3 resources 968 (http://bogus.example.com/1x1.gif, http://root-key-sentinel-is-ta- 969 02323.example.com/1x1.gif, http://root-key-sentinel-not-ta- 970 11112.example.com/1x1.gif). 972 Geoff then asks Bob, Charlie, Dave and Ed to browse to 973 www.example.com. Using the methods described in this document, the 974 users can figure out what their fate will be when the 11112 KSK is 975 removed. 977 Bob is not using a validating resolver. This means that he will be 978 able to resolve bogus.example.com (and fetch the 1x1 GIF) - this 979 tells him that the KSK roll does not affect him, and so he will be 980 OK. 982 Charlie's resolvers are validating, but they have not been upgraded 983 to support the KSK sentinel mechanism. Charlie will not be able to 984 fetch the http://bogus.example.com/1x1.gif resource (the 985 bogus.example.com record is bogus, and none of his resolvers will 986 resolve it). He is able to fetch both of the other resources - from 987 this he knows (see the logic in the body of this document) that he is 988 using validating resolvers, but at least one of these resolvers is 989 not configured to perform sentinel processing. The KSK sentinel 990 method cannot provide him with a definitive answer to the question of 991 whether he will be impacted by the KSK roll. 993 Dave's resolvers implement the sentinel method, and have picked up 994 the new KSK. For the same reason as Charlie, he cannot fetch the 995 "bogus" resource. His resolver resolves the root-key-sentinel-is-ta- 996 02323.example.com name normally (it contacts the example.com 997 authoritative servers, etc); as it supports the sentinel mechanism, 998 just before Dave's recursive resolver sends the reply to Dave's stub, 999 it performs the KSK Sentinel check. The QNAME starts with "root-key- 1000 sentinel-is-ta-", and the recursive resolver does indeed have a key 1001 with the Key Tag of 02323 in its root trust store. This means that 1002 that this part of the KSK Sentinel check passes (it is true that Key 1003 Tag 02323 is in the trust anchor store), and the recursive resolver 1004 replies normally (with the answer provided by the authoritative 1005 server). Dave's recursive resolver then resolves the root-key- 1006 sentinel-not-ta-11112.example.com name. Once again, it performs the 1007 normal resolution process, but because it implements KSK Sentinel 1008 (and the QNAME starts with "root-key-sentinel-not-ta-"), just before 1009 sending the reply, it performs the KSK Sentinel check. As it has the 1010 key with key-tag 11112 in it's trust anchor store, the answer to "is 1011 this *not* a trust anchor" is false, and so the recursive resolver 1012 does not reply with the answer from the authoritative server - 1013 instead, it replies with a SERVFAIL (note that replying with SERVFAIL 1014 instead of the original answer is the only mechanism that KSK 1015 Sentinel uses). This means that Dave cannot fetch "bogus", he can 1016 fetch "root-key-sentinel-is-ta-02323", but he cannot fetch "root-key- 1017 sentinel-not-ta-11112". From this, Dave knows that he is behind an 1018 collection of resolvers that all validate, all have the key with key 1019 tag 11112 loaded and at least one of these resolvers has loaded the 1020 key with key-tag 02323 into its local trust anchor cache, Dave will 1021 not be impacted by the KSK roll. 1023 Just like Charlie and Dave, Ed cannot fetch the "bogus" record. This 1024 tells him that his resolvers are validating. When his (sentinel- 1025 aware) resolvers performs the KSK Sentinel check for "root-key- 1026 sentinel-is-ta-02323", none of them have loaded the new key with key- 1027 tag 02323 in their local trust anchor store. This means check fails, 1028 and Ed's recursive resolver converts the (valid) answer into a 1029 SERVFAIL error response. It performs the same check for root-key- 1030 sentinel-not-ta-11112.example.com, and as all of Ed's resolvers both 1031 perform DNSSEC validation and recognise the sentinel label Ed will be 1032 unable to fetch the "root-key-sentinel-not-ta-11112" resource. This 1033 tells Ed that his resolvers have not installed the new KSK and he 1034 will be negatively implacted by the KSK roll.. 1036 Geoff would like to do a large scale test and provide the information 1037 back to Alice. He uses some mechanism such as causing users to go to 1038 a web page to cause a large number of users to attempt to resolve the 1039 three resources, and then analyzes the results of the tests to 1040 determine what percentage of users will be affected by the KSK 1041 rollover event. 1043 This description is a simplified example - it is not anticipated that 1044 Bob, Charlie, Dave and Ed will actually look for the absence or 1045 presence of web resources; instead, the webpage that they load would 1046 likely contain JavaScript (or similar) which displays the result of 1047 the tests, sends the results to Geoff, or both. This sentinel 1048 mechanism does not rely on the web: it can equally be used by trying 1049 to resolve the names (for example, using the common "dig" command) 1050 and checking which result in a SERVFAIL. 1052 Authors' Addresses 1054 Geoff Huston 1056 Email: gih@apnic.net 1057 URI: http://www.apnic.net 1059 Joao Silva Damas 1061 Email: joao@apnic.net 1062 URI: http://www.apnic.net 1064 Warren Kumari 1066 Email: warren@kumari.net