Hi everyone, I thought a bit longer about this problem, and now understand it properly. I don't think that the problem needs its own draft. An errata seems like the better option to me.
The attack is the following: - Attacker controls example.com. This zone contains: example.com. SOA ... *.example.com. TXT "poison" - Now, the attacker queries this zone for a.example.com., but they set the respective RRSIG record's label count to 1, i.e., a recursive resolver would conclude that the original owner was *.com. - The recursive resolver now caches this RRRSIG record. If the resolver also implements RFC 8198, the attacker can have any TXT query to a non-existent domain in com. return the result "poison". The respective name server would correctly serve the NSEC(3) proofs for that domain, which the resolver would use in conjunction with the previous RRSIG record to synthesize the response. As I said, the fix doesn't require its own draft. I believe the problem lies in RFC 4035, Sec. 5.3.1., which says that a resolver must check: - The RRSIG RR's Signer's Name field MUST be the name of the zone that contains the RRset. This is ambiguous for wildcard records. The synthesized RRset indeed is in the zone of the signer's name. But the original owner is not. Thus, I would suggest changing this text to: - The RRSIG RR's Signer's Name field MUST be the name of the zone that contains the RRset*'s** original owner*. That change should make it clear to implementers that they must consider wildcard synthesis in this step, which is not clear right now. All text regarding wildcard synthesis comes only below. Best, Felix Am Di., 27. Okt. 2026 um 17:52 Uhr schrieb Qifan Zhang < [email protected]>: > Hi Felix, > > Thank you for these helpful comments. I also appreciated your team’s > USENIX Security 2025 work on verifying iMessage PQ3. I read it while > working on DNSSECVerif and found it quite inspiring. > > I agree that §5.3.1 already states the intended authority requirement. The > clarification concerns its application to the original owner reconstructed > using RRSIG Labels, which §5.3.2 does not explicitly compare with the > Signer’s Name. > > For example, a received owner of `a.child.example.`, with signer > `child.example.` and Labels=1, reconstructs to `*.example.`. Checking > containment of the received owner alone misses that the reconstructed owner > lies outside the signer’s zone. > > Our concern stems from confirmed vulnerabilities in all four open-source > resolvers we evaluated that implemented wildcard synthesis from > DNSSEC-validated aggressive caches: BIND 9, Unbound, PowerDNS Recursor, and > Knot Resolver. This feature was enabled by default in all four. Although > exploitation conditions differ, we observed the same validation boundary > missed across every implementation of this feature we examined. In our > view, this provides strong empirical evidence that leaving the > reconstructed-owner check implicit creates a gap in implementation guidance > with real security consequences, strengthening the case for making it > explicit. > > I have prepared a -02 revision [1] addressing your feedback: > > - The abstract and introduction explicitly acknowledge the existing > authority requirement and frame the check as a clarification. > - Section 4.1’s three normative paragraphs are condensed into one > sentence, preserving per-signature handling and the exact asterisk-apex > exception. > - Section 4.2 retains the cache clarification: wildcard-source metadata > must be authenticated by a successfully validated signature, rather than > inherit validation from another signature. > > I would welcome WG guidance on the appropriate publication path for this > clarification. > > Best regards, > Qifan > > [1]: > https://datatracker.ietf.org/doc/draft-zhang-dnssec-wildcard-validation/02/ > > On Tue, Oct 27, 2026 at 7:41 AM Felix Linker <[email protected]> > wrote: > >> Hi Qifan, >> >> This is a very cool finding, and apparently, there is software out there >> that got this wrong, so some action seems to be advised. >> >> I wonder, though, whether a standard update is required. Shouldn't the >> following text cover this already (from >> https://www.rfc-editor.org/info/rfc4035/#section-5.3.1 >> <https://urldefense.proofpoint.com/v2/url?u=https-3A__www.rfc-2Deditor.org_info_rfc4035_-23section-2D5.3.1&d=DwMFaQ&c=V9IgWpI5PvzTw83UyHGVSoW3Uc1MFWe5J8PTfkrzVSo&r=G8Y2PafMWp_pxuBT8os_MllFpqBNVDXvAhGzwXTXJfs&m=jDGktRVZzb32TupJlIZl3KYKyUYZIF68EfpG3QoLk2f21VCuKM_0_9Wi2YPZb5p5&s=zrJQ4ckZSiatVCTTKLFP7C9t175rD0oRZTGjMiZKcWM&e=> >> )? >> >> o The RRSIG RR's Signer's Name field MUST be the name of the zone >> that contains the RRset. >> >> >> This seems to be consistent with your suggestion in 4.1, which updates >> signer's name comparison too. >> >> I struggled with the above sentence in the past too. I think the >> validation obligation to the validating resolver is quite unclear. >> >> In any case, I believe your draft could benefit from some succinctness. >> Your proposed addition to RFC 4035 has three paragraphs, but I think it >> could be a single sentence. That's just an example, though. >> >> Best, >> Felix >> >> Am Sa., 17. Okt. 2026 um 07:00 Uhr schrieb Qifan Zhang <qzhang= >> [email protected]>: >> >>> Hi John, >>> >>> First, thank you for your prompt response and for raising the >>> compatibility concern. However, I would like to clarify that the >>> proposal actually **preserves** existing wildcard matching and >>> expansion, including expansion over multiple labels. TL;DR: It adds a >>> validation check on RRSIG Labels to enforce the signing zone’s authority. >>> >>> For example, for an RRset at `a.child.example.`, a signature from >>> `child.example.` with Labels=1 reconstructs `*.example.`, which the child >>> key cannot authenticate. A legitimate source at `*.child.example.` with >>> Labels=2 remains valid, including expansion to `a.b.child.example.` when >>> existing applicability rules permit it. >>> >>> To address your concern, we submitted revision -01: >>> >>> >>> https://datatracker.ietf.org/doc/draft-zhang-dnssec-wildcard-validation/01/ >>> <https://urldefense.proofpoint.com/v2/url?u=https-3A__datatracker.ietf.org_doc_draft-2Dzhang-2Ddnssec-2Dwildcard-2Dvalidation_01_&d=DwMFaQ&c=V9IgWpI5PvzTw83UyHGVSoW3Uc1MFWe5J8PTfkrzVSo&r=G8Y2PafMWp_pxuBT8os_MllFpqBNVDXvAhGzwXTXJfs&m=jDGktRVZzb32TupJlIZl3KYKyUYZIF68EfpG3QoLk2f21VCuKM_0_9Wi2YPZb5p5&s=eboOii5T8Wjh7BrCKAPtJUOCg98B0U3isb7CjSkWkjc&e=> >>> . >>> >>> It explicitly states the unchanged wildcard semantics in the abstract >>> and introduction, adds valid multi-label and parent-zone expansion >>> examples, and clarifies source authentication versus expanded-answer >>> validation. These are explanatory changes; the normative requirements >>> remain unchanged. >>> >>> Section 3.1 links to the corresponding resolver advisories. Could you >>> point to a legitimate wildcard expansion that you believe the proposed >>> check would reject? >>> >>> Best regards, >>> Qifan >>> >>> On Fri, Oct 16, 2026 at 8:38 PM John Levine <[email protected]> wrote: >>> >>>> It appears that Qifan Zhang <[email protected]> said: >>>> >-=-=-=-=-=- >>>> > >>>> >Dear DNSOP colleagues, >>>> > >>>> >We have submitted a new individual Internet-Draft, “Signer-Zone >>>> Containment >>>> >for DNSSEC Wildcard Validation”: >>>> > >>>> > >>>> https://urldefense.proofpoint.com/v2/url?u=https-3A__datatracker.ietf.org_doc_draft-2Dzhang-2Ddnssec-2Dwildcard-2Dvalidation_&d=DwIDaQ&c=V9IgWpI5PvzTw83UyHGVSoW3Uc1MFWe5J8PTfkrzVSo&r=G8Y2PafMWp_pxuBT8os_MllFpqBNVDXvAhGzwXTXJfs&m=biL346UC13wbVp0UyS0Xom39F8tB4Z6xp08lEff-3Egen14y6HkTnwtJ8aXp7SGv&s=MnP4J7RNPU105m7sy91UHhZi4wH46csZM8uVT4gLAuo&e= >>>> > >>>> >Based on findings from our DNSSECVerif research, accepted at ACM CCS >>>> 2026, >>>> >the draft proposes focused updates to RFCs 4035 and 8198 to: >>>> > >>>> > - Require signer-zone containment of reconstructed RRset owner >>>> names. >>>> > - Preserve authenticated wildcard-source associations during cache >>>> reuse. >>>> >>>> I can understand why you might not like the way that wildcards work, >>>> but changing >>>> the way wildcards are expanded would be a breaking change to a spec >>>> that has been in place for >>>> 40 years. So, no. Also, this would make wildcard expansion with >>>> DNSSEC give different >>>> answers to expansion without DNSSEC. So doubly no. >>>> >>>> R's, >>>> John >>>> >>> _______________________________________________ >>> DNSOP mailing list -- [email protected] >>> To unsubscribe send an email to [email protected] >>> >>
_______________________________________________ DNSOP mailing list -- [email protected] To unsubscribe send an email to [email protected]
