IETF-55 Atlanta DNSEXT WG Minutes
Chair: Olafur Gudmundsson ogud@ogud.com
Randy Bush (absent)
Note takers: Jun-ichiro itojun Hagino
Scott Rose
Date: 2002/November/19 13:00-14:00 EST (18:00-19:00 UTC)

Agenda Bash

WG document status
    RFC editor: Restrict KEY, Roadmap, Obsolete IQuery
    IESG: AD is secure, AXFR clarify, Unknown Types, DS
    WG last call complete pending changes: Opcode Discover
    WG last call: RFC1886bis, DNSSEC Opt-In, TKEY Renewal


GSSAPI and TSIG conflict:
    DNSEXT wg generated TSIG RFC
    DNSEXT wg processed gssapi TSIG
    just before rfc editor started 48hour period we got a report that 
there was a conflict between two the documents.
    TSIG specifies that TSIG can only be used if original query 
contains TSIG
    GSSAPI specifies that last message in TKEY exchange has TSIG last 
message is empty, and this proves the key negotiated is working from 
security point of view this is a good thing

    TSIG needs minor updates before advancing to draft standard:  is this 
extensions one of them?

The sense of the room was that this was a reasonable extensions and the 
chair is instructed to take this question to the mailing list.


RFC1886/AAAAbis  Vladimir Ksinant  presenting
    RFC 1886 Update:
    goal:  push AAAA to draft standard
    tests done in July '02
    Results:
        minor changes in 1886 needed
    Status of draft:  00 in Sept, now -01 in Oct.
        Now in WG last call - comment now please.


dnssec protocol changes
opt-in David Blacka
    03 to 04: editorial
    02 to 03:
        interaction with wild-cards clarification
        ad-bit clarification
        validation process changes made more explicit

    increases code complexity
    2 independent code exists
    decreased cost in big registries (signing/whatever)
    way NXDOMAIN works in presence of DS and opt-in
    how NXDOMAIN change affect application?

There was a comment that security section needs stronger language.
Clearly list the need for zone transfer protection and risk from zone 
hijacking of opt-outed delegations.
Authors agreed examine if the current text is insufficient .

Opt-in issues: Rob Austein
opinions about OPT-IN from a study done by Rob and Roy A. -
        1. Corner cases are actually DNSSEC issues, not OPT-in 
specific.
        2. DNSSEC pushes a change to the cache algorithm that is 
considered standard in DNS today.  You have to keep things together under 
query names now in order to build auth chain.  Especially when NXT RRs are 
involved.
        3. OPT IN does make the code more complex - resolvers have to be 
more intelligent now.
        4. No workshop experience dealing with OPT-IN.  That would be 
nice, since DS has shown problems.
        5. It seems to help large delegation heavy zones, but not 
resolvers.
        6. From a technical (coding) viewpoint:  the costs are larger than 
the benefits.  However, higher, operational interests may rule the day.

Mark Kosters stated there are 2 independent implementations of opt-in 
authorative server code. and 2 "independent" implementations of 
resolver, both done by the same person in two different code bases.  There 
where some discussion on if Opt-in is delaying or enabling the roll-out of 
DNSSEC. All the speakers stressed the need for some kind of DNSSEC Real 
Soon Now.

DS status Olafur Gudmundsson
Implementations:
    1 resolver (or 2)
    2 server implementation
    3 different management tools in development
    3 workshops since Yokohama

    3 under specified corner cases found
      - need to specify what child server returns for DS query at apex
           parent not found
          - if child is served by the same server as ancestor other
           than parent
      - RFC2535 capable caches have problem with DS
    are there more undiscovered?


    one more document update to deal with ancestor problem
    solution: resolver detects this from authority section and asks for 
delegation information on parent nameservers and then asks them the 
question.

    TIME TO DECIDE if DS goes forward, is close.


There's no way to determine if all relays are DS capable, does this 
require flag to state if DNS entity is DS aware?
Discussion addressed the problems experienced with DNSSEC failing due to 
middle boxes that do not understand DNSSEC types or DS processing.
Is it acceptable during deployment to not be able to do DNSSEC 
resolution, or should DNSSEC require "clear path"?
The sense of the room was that broken middle-boxes need to be updated and as 
long as DNSSEC answers are not messing up middle boxes it is 
acceptable to be only able to do DNSSEC resolution in places where 
resolvers can talk directly to authorative servers.

KS flag in KEY: Olaf Kolkman
   version 03 of draft.
   Using a bit in the flags of the KEY record to denote a Key signing
   key.  No protocol value assigned, for user/operator use only.
   Going to WG last call.

Wild-card optimization: Olaf Kolkman
   assumption: one needs proof for non-existence of a wild-cards
    the proof is to be supplied in the common case where there is
        no wild-card in a zone that proof is complicated and somewhat
        expensive in terms of computational resources and packet size
    is there a way to signal that there is no wild-card in a zone?

    proposal:
    use a bit in the NXT RR's type bitmap to signal non-existence
    of wild-cards
    the meaning:
    there is no wild-card expansion possible of any name in the zone's 
namespace spanned by a NXT RR simplest algorithm to set the bit turn it on on 
all NXT RRs in the zone if there are no wild-cards in the zone turn it off on 
all NXT RRs in the zone if there is any wild-card in the zone


    pro:
    shortcut to a simple and fast code path for server and resolver 
smaller footprint
    con:
    bypassed in the common case, the complicated verification code is 
still needed RR type code without RR backward incompatible

    current version of doc
    the suggested algorithm for setting the bit contain an error
    clarifications are needed

    what's next
    update the draft
    does the working group think this is a sane path
    dnssec protocol specs need to describe algorithm for denial of 
authentication of wild-cards
        if not, resolver implementers will do it wrong
        no need for more than 2 NXT RRs?

    if it makes things more complicated, object.
There where some questions about savings on the wire (about 150 bytes in the 
common case). Requires more code, not a lot in resolvers, some in 
servers, a lot in signers.
Take issues to the mailing list.


Domain Name Auto Registration for plugged in IPv6 nodes: Hiroshi 
Kitamura
       This document is in a need of home for working group review is 
DNSEXT the right home ?
       The DNSEXT chairs have suggested advancing the document either as 
informational or experimental RFC. Authors would like a working group last 
call for that status.
Erik Nordmark commented that he thinks Experimental is a better status for 
this as it might be promoted to standards track if it becomes widely used.
After author updates document chair is to start working group last call.


End of meeting.