[tor-talk] Tor DNS Deanonymization

[Philipp Winter]

On Sun, Oct 16, 2016 at 01:15:32AM -0400, Nick Mathewson wrote:
> On Fri, Oct 14, 2016 at 11:09 AM, Philipp Winter <phw at nymity.ch> wrote:
>  [...]
> > There are two ways to mitigate the issue.  First, we need better
> > defences against website fingerprinting, so an attacker learns less by
> > observing the connection to your guard relay.  Second, we need to
> > improve the DNS setup of exit relays.  I would like to see less relays
> > use Google's resolver, and we need to move towards encrypted DNS.
> 
> Thanks, Philipp!
> 
> Could you comment at all about whether our current exit side dns
> caching approach makes the attack harder, easier, or doesn't matter?

Generally, the longer exit relays cache domains, the less precise the
attack.  The trade-off is illustrated in Figure 10b in our paper [0].
At the moment, exit relays cache domains for only 60 seconds [1],
regardless of the domain's TTL.  If that bug is fixed, the attack
becomes a bit harder to mount.  It can become even harder if exit relays
were to cache each domain for, say, 10 minutes or more.

[0] <https://nymity.ch/tor-dns/tor-dns.pdf>
[1] <https://bugs.torproject.org/19025>