On Fri, Jul 07, 2017 at 09:58:19AM +0000, Shane Kerr wrote:
> Hugo,
> 
> I'm curious what you mean by this. Do you really mean to propose an
> option to pad every query and response message to 65K bytes? I guess I
> don't object it for the sake of completion, but it seems a bit crazy.
> 
> OTOH, people use Tor for browsing, so maybe someone will actually
> want to do this? ;)
> 
> Seriously though, on the query side padding beyond a few hundred bytes
> is not helpful, because no queries are longer than that. Maybe on the
> response side it is indeed more privacy-protecting. 

On query side, one could always pad to 286 (288 for TCP) bytes, AFAICT
the maximal query in practice is:

xx xx: Query length (TCP only).
xx xx: Query ID.
01: Query, requesting recursion.
00: DNSSEC errors are fatal.
00 01: 1 query
00 00: No answers
00 00: No authority
00 01: 1 additional record
<255 bytes>: QNAME
xx xx: QTYPE
00 01: QCLASS (1=IN).
00: Dummy domain (root)
00 29: OPT
04 B0: Maximum UDP response size (1200 bytes).
00 00 80 00: EDNS0, DNSSEC supported.
00 04: 4 bytes of EDNS data
00 12: Padding
00 00: 0 bytes of padding

However, responses are thornier issue. This is recursive, so it might
need to relay all kinds of responses. I could provke one ccTLD to
return a 3651 byte response with normal QTYPE (ZSK rollover plus
healthy amount of authoritative nameservers, most available over
IPv6). That kind of response when sent over UDP gets fragmented at
IP layer (not good) or triggers a fallback to TCP (not good).


-Ilari

_______________________________________________
dns-privacy mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dns-privacy

Reply via email to