Re: [DNSOP] Incremental zone hash - XHASH

[Mark Andrews <[e-mail]>]

> On 20 Jul 2018, at 10:37 pm, George Michaelson <[e-mail]> wrote:
> 
> The intent (to me at least) is to be able to use exterior fetch, *not*
> DNS, to source this as a file. curl. wget. ncftp. rsync.

Nothing in this proposal stops these transport mechanisms.

> the "thing" is a file object. It almost certainly is in near-canonical
> sort order already. Its a stream of characters, probably in
> bind-normal form.

Which is basically DNSSEC order.

> If you can compute the path through the labels and the chain of NSEC
> regions and the expected hadda-yadda-dadda.. you don't *need* a
> digest.
 
The difference between METAMD and XHASH is that you do XHASH in smaller
chunks on such a ordered zone and verify each chunk as you go through.

> If you have a digest, and its already in near canonical order, then
> *cost* to compute "is the file exactly as the publisher wrote it" is
> low. And, since its a signature under the ZSK, its not just "its as
> the publisher said" its "the publisher knows the ZSK" which is strong
> enough to say: "just load it"
> 
> So, I ask: is this incremental method applicable to this model?

Yes

> Sure, works for giant zone. What about a root zone? Do I need this?

Yes, it works for the root zone.

> Also.. glue.

Yes it hashes glue and delegating NS record.

> -G
> 
> On Fri, Jul 20, 2018 at 6:31 AM, Mark Andrews <[e-mail]> wrote:
>> Rather than having a full zone hash this can be done as a chain
>> of hashes (XHASH).
>> 
>> The XHASH would include all records at a signed name (where a signed
>> name is NOT an NSEC3 name) up until the next signed name (where a
>> signed name is NOT a NSEC3 name) in DNSSEC order similar to ZONEMD.
>> If there is a NSEC3 record and its RRSIGs in this range it is included
>> in the hash computation.  Where a NSEC3 record matches the name of a
>> record that exists in the zone it is hashed with that name. The record
>> type appears at both top and bottom of zone similar to NS.
>> 
>> The chain is only deemed to be complete if there is a hash record at
>> the zone apex. This allows for incremental construction and destruction
>> of the XHASH chain similar to the way the presence of NSEC at the zone
>> apex indicates that chain is complete.
>> 
>> If there are records that are not at or under the zone apex they are included
>> in the final XHASH of the zone sorting from the zone apex to the end of the
>> namespace then from the start of the namespace to the zone apex. Such records
>> at not normally visible to queries other than AXFR/IXFR.  AXFR/IXFR permit such
>> records.
>> 
>> XHASH would allow for UPDATE to incrementally adjust the chain without
>> having to hash the entire zone at once.
>> 
>> XHASH would allow for a slave server to verify a zone is still complete
>> after a IXFR by just checking the areas of the zone impacted by the IXFR.
>> 
>> e.g.
>> 
>>        example.com SOA
>>        example.com NS ns.example.com
>>        example.com DNSKEY …
>>        example.com NSEC a.example.com NS SOA RRSIG NSEC DNSKEY XHASH
>>        example.com XHASH …
>> 
>>        a.example.com NS ns.a.example.com
>>        a.example.com NSEC b.example.com NS RRSIG NSEC XHASH
>>        a.example.com XHASH …
>>        ns.a.example.com A …
>> 
>>        b.example.com NS ns.b.example.com
>>        b.example.com NSEC ns.example.com NS RRSIG NSEC XHASH
>>        b.example.com XHASH …
>>        ns.b.example.com A …
>> 
>>        ns.example.com A …
>>        ns.example.com AAAA …
>>        ns.example.com NSEC example.com A AAAA RRSIG NSEC XHASH
>>        ns.example.com XHASH …
>> 
>> Each of the groupings shows which records plus RRSIGs that are
>> included in the XHASH calculation.
>> 
>> To prevent removal/introduction of RRSIGs of XHASH records a DNSKEY
>> flag bit is be needed to indicate which RRSIG(XHASH) should/should not
>> be present once the chain is complete.  The same applies to RRSIG(ZONEMD).
>> 
>> Verification of a AXFR would be slightly slower than with ZONEMD as there
>> are more RRSIG records to be processed,
>> 
>> 
>> --
>> Mark Andrews, ISC
>> 1 Seymour St., Dundas Valley, NSW 2117, Australia
>> PHONE: [phone]              INTERNET: [e-mail]
>> 
>> _______________________________________________
>> DNSOP mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/dnsop

-- 
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: [phone]              INTERNET: [e-mail]