Re: 2 forwarded messages...DNSEXT discussion of Day and Kaminsky

Dean Anderson <[email protected]> wrote:
> On Fri, 20 Feb 2009, Paul Jarc wrote:
>> You say that Kevin Day is malicious, based on his contact with Dan
>> Kaminsky.
>
> Please keep the facts straight, particularly when you attribute to
> others words that they didn't use. I didn't say they are malicious--I
> have no knowledge of actual malice or actual criminal activity.

Fair enough.  But I'd suggest not using the word "blackhat", then,
since it strongly connotes malice.

> The patch has a couple weaknesses that has been described.

I assume you're talking about these attacks (but please, please,
please, I wish you wouldn't make me guess):
1. The attack I detailed, where the attacker makes dnscache forget
   about its previous outstanding queries, and then forges responses
   for the remaining, attacker-controlled query/queries.
2. DOS the authoritative server, or otherwise stop the genuine
   responses from reaching dnscache, to allow more time for the attack
   to succeed.

Both of these attacks work no better against a qmerge-patched
dnscache.  The first runs into the problem that the attacker can't
distinguish recently-forgotten ports from still-waiting-for-response
ports to reduce the number of ports under consideration, as well as
the lack of any benefit in reducing that number, as Jeff explained.

The second attack works exactly the same regardless of whether the
qmerge patch is involved - each packet that the attacker disrupts
gives them one window that's open longer.  An attacker using a
birthday attack doesn't benefit from having just one long-open window
with qmerge, versus one or more long-open windows and several
short-open windows with an unpatched dnscache.

> you (Jarc et al) inexplicibly oppose a patch that doesn't inject any
> weakness by not tinkering with more than is necessary.

I'm not opposed to such a patch, and I don't think I've seen anyone
else voice any opposition.  Yours might be just as good an approach.
So far, I haven't thought about it in much detail.  Once you actually
have a patch that we can look at, then we can see if it introduces any
new problems.

> Well, indeed, I __can't__ help it if you don't understand basic concepts
> of random numbers, and don't understand the the port number selection in
> DNScache isn't perfectly random.

I understand that.  I don't understand how an attacker can practically
benefit from that.

>> It could be that everyone else here is too stupid to measure up to
>> you, but it seems rather more likely that either you simply haven't
>> done a good enough job of explaining the problems, or you're mistaken
>> about their existence.
>
> Nice try at "everyone else here".

I meant everyone participating in this discussion, not everyone who
might be reading it.  Sorry if that was unclear.

> What is certain is that there is no mistake about the randomness of
> UDP port selection (not strongly random); no mistake about the
> existance of introduced weaknesses by increasing the dependence on
> weakly-random UDP port randomness;

This sort of language is the reason it's so hard to understand you.
Rather than saying what the weakness is, you only say that there is a
weakness, referring to some unidentified previous description.  Since
you know that your previous description was not understood, it would
be tremendously more helpful if you could give a new, different
explanation.


paul