Re: [dns-privacy] DS Hacks

[Shane Kerr <[e-mail]>]

On 30/07/2021 17.30, Ben Schwartz wrote:
> It seems like there's still interest in DS hacks.  Here's how I would do 
> a DS hack.
> 
> 1. Use the VERBATIM hash from 
> https://datatracker.ietf.org/doc/html/draft-vandijk-dnsop-ds-digest-verbatim 
> <https://datatracker.ietf.org/doc/html/draft-vandijk-dnsop-ds-digest-verbatim>
> 
> 2. Use a new Algorithm Number like 
> https://datatracker.ietf.org/doc/html/draft-fujiwara-dnsop-delegation-information-signer-00 
> <https://datatracker.ietf.org/doc/html/draft-fujiwara-dnsop-delegation-information-signer-00>.  
> This number is named "DS Glue", and MUST only be used with the VERBATIM 
> hash.
> 
> 3. In the digest field, encode one "DS glue" record as follows:
> 
> struct DSGlueFakeDigest {
>    uint16 rr_type;
>    DNSName prefix;  // Labels prepended to the DS owner name.
>    uint8[] rdata;
> }
> 
> DS-glue-aware resolvers ignore any unsigned glue with this RR type and 
> owner name.

So this is generic, and can store any record. We don't need TTL since 
this is just for glue and really shouldn't be cached, and we don't need 
class because we don't ever need class. 😉

This can't store out-of-bailiwick data, which means we can't secure an 
arbitrary NS RRset this way. Converting DNSName from "prefix" to just 
"name" would allow that. This seems useful as it would eliminate a 
potential means for an attacker to get you to connect to their servers 
(which is, spoofing a reply with a bogus NS RRset, since delegation NS 
are not signed in DNSSEC because DNS is stupid).

> 4. For ADoX, place NS records (with a prefix like "ns3.") and SVCB 
> records (with a prefix like "_dns.ns3.") in the DS glue.


--
Shane