Very well said.  +Lots.

On Oct 22, 2014, at 4:19 PM, Ted Hardie 
<[email protected]<mailto:[email protected]>> wrote:

On Mon, Oct 20, 2014 at 6:10 PM, Paul Vixie 
<[email protected]<mailto:[email protected]>> wrote:

 query minimization i see no benefit or need for secrecy since dns is a 
publication system -- if you don't want people to see your information, don't 
put it into the dns. cache miss transactions (#2 above) expose only server IP 
addresses, not end-user IP addresses, and no re-use events. as such it has 
always been my position that there is no PII and no privacy impact from having 
these recorded and analyzed. i would not like to see the IETF work in that 
area.ld be classified into:

At the STRINT workshop, we discussed the extent to which combining queries at a 
proxy provides protection against analysis by a monitor.  It is certainly the 
case that it provides some benefit, but we there are limits, especially when 
the resolver serves a relatively small number of users.

To take an example:  Joe belongs a club, example.org<http://example.org/>, and 
is using their network to look up a sensitiveblog.journal.example.  
example.org<http://example.org/> hosts it own DNS resolver, but few users 
connect to journal.example (having mostly moved to 
face-example.org<http://face-example.org/>), so when Joe makes his look up, 
it's a cache miss.  A surveillance entity will therefore see 
sensitiveblog.journal.example as a target of the DNS query if it has a tap on 
path between example.org<http://example.org/> and journal.example, and it now 
knows that the traffic between example.org<http://example.org/> and 
journal.example is for the senstiveblog.

Note that it would not necessarily have known that otherwise.  Because 
journal.example uses the HOST header to disambiguate traffic to different 
hosted journals, thus consuming a very small number of IP addresses, the 
connection itself would not have identified which blog was at issue. [There are 
other potential sources of leaks, here, such as the unencrypted SNI in TLS, but 
that has to be worked elsewhere.  I do not personally accept the idea that we 
can't fix information leakage piece by piece, just because other leaks are 
possible].

Clearly, had Joe used OpenDNS, Google's DNS, or a large ISP DNS service which 
aggregates many more users, the chance of a cache miss would be lower and the 
correlation would be much, much harder even in the case of a cache miss.   But 
the DNS protocol cannot presume that sort of deployment, and they have other 
consequences in terms of trust between users and large services.

As long as the DNS continues to support deployment of recursive resolvers in 
topologies that cannot effectively provide aggregation as a mitigation, I 
believe that the recursive to authoritative problem should stay on the table.  
It is a lower priority, but it is a valuable piece of work.  I personally also 
think that it is a substantially different problem from the stub to recursive 
resolver problem, so we should treat them independently.

My two cents, in any case.

Ted Hardie


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

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

Reply via email to