Paul: Thanks for the explanation, it clears up a fair bit for me. Replies inline.
On 03/20/2018 09:48 AM, Paul Wouters wrote: > On Tue, 20 Mar 2018, Michael Casadevall wrote: > >> Without the RRtypes logged, I'm not seeing how you're supposed to be >> able to audit them. In the case where a KSK is compromised, you could > > Neither this nor certificate transparency protects against a (non-known) > private key compromise. In the case of CA's there are multiple entities > that can create a certificate that need to be audited. In DNSSEC, it is > only the zone or its parents that could create a rogue TLSA record. > In my mind, the use case I was making was one of that of an intermediate certificate being misused, but didn't quite make the mental leap to equating that to the TLD/root zone. In that context, this proposal as well as logging DS/NS changes makes considerably more sense to me than it did before. I still suspect there's value for full logging of all record types, but that's a separate discussion. >> I'm likely missing something obvious, by only logging DS/NS, it would be >> impossible to determine if a private key is misused to serve fraudulent >> records which in my mind is a bigger and much more likely/common threat >> since one can simply attack a target domain and not try to compromise >> the root. > > This is similar to the TLS server private key being compromised without > knowledge. That is not what is being protected. > See above. >> From a technical point of view, aren't there TLDs that as authoritative >> for second level domains as well? The specification likely needs a way >> to denote how many levels deep delegation can go. This also would be >> true in the case of delegation_only being used at a level above both the >> root and TLD zones. > > In trying to keep this as simple as possible, the idea was to make this > concept as simple as possible. Making a promise about "every zone cut" > is much easier then making a promise to point to some more complicated > policy that might change over time, and then you have to log/audit the > changes to that policy as well. > > For instance, you could have the DNSKEY flag mean "look for the > DELEGATE RRtype policy" and in there specify exceptions for your empty > non-terminals, but it just makes everything much harder. > I can appreciate simplicity in a specification. Obviously having a hypothetical DELEGATE RRtype would considerable more work both in design and actual implementation/validation. I'm just slightly concerned that it might make real world deployment more difficult. My point is here though is it might be a relatively simple extension to say "I may delegate X levels down" vs. a rather complex system of delegation/termination for cases where the TLDs disallow second level registration. I'm not qualified enough to say definitively if that's actually a valuable enough case to be worth extending the specification, but I did want to highlight it for consideration. >> At a minimum though, the one case I know off hand of a TLD being >> authoritative for a second-level domain is that ICANN requires new gTLDs >> are required to publish a wildcard for 90 days when they're added to the >> root zone to help catch any name collisions. > > If the software allows signing the wildcard and setting the > delegation-only flag that would mean the wildcard must be > treated as BOGUS, that would not be a problem. It would still > cause problems to be found with name collisions, except now the > name SERVFAILs instead of resolving to 127.0.0.53. Both will show > the name collision as a problem. > This likely should be noted as a operational consideration at a minimum as it impacts how Name Collision tracking is done. Thanks for your time, Michael _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop
