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

On Wed, Feb 18, 2009 at 09:19:40PM -0500, Dean Anderson wrote:

[apologies for the long quote, but all levels are relevant to what I
will say]

> On Wed, 18 Feb 2009, Matthew Dempsky wrote:
> 
> > On Wed, Feb 18, 2009 at 1:12 PM, Dean Anderson <[email protected]> wrote:
> > > Yes, I do know that qmerge doesn't make THAT change. My example was
> > > to demonstrate that changes affect the attacks that are possible.
> > 
> > That's a straw man.  No one suggested that changes never affect attack
> > success probabilities.
> 
> You said:
> 
> > So you understand that if an attacker can carry out an attack against
> > dnscache with a single outstanding UDP query, then he can apply the
> > same technique when dnscache has 200 outstanding UDP queries and have
> > the same (or better) chance of success, right?  I.e., you understand
> > that any attack that an attacker can carry out with the qmerge patches
> > applied, he can still carry out without it applied, right?
> 
> To which I said:
> 
>   No, the above isn't true in any non-trivial sense.  One can obviously
>   attempt any attack at any time (the trivial sense).  But the chance of
>   success of any given attack depends on the changes made to DNScache.
>   For example, if you change it so to bind the port to 53, the chances 
>   of a brute force attack change to 1 in 65536. Doh.
> 
> It appears that I gave the correct analysis.

Here's why your strawman doesn't apply. Matthew indicated that one
_particular_ class of changes (e.g., the one in the qmerge patches) has
this property. This class of changes includes ones which reduce the
number of queries sent by dnscache. Your strawman of turning off
dnscache's port randomization has nothing to do with that.

> Untrue. You have merely failed to understand my arguments. You have not
> discredited any of my arguments.

Please count me in the list of people who do not understand your
arguments. They make no sense to me.

> That's your opinion of my arguments, and it seems to be you who is
> waving arms.  You haven't shown my math to be wrong. In fact, I _have_
> provided sufficient mathematical argument to show that entropy of port
> numbers may be affected by Kaminsky's changes, and is unaffected by my
> changes.  You just don't follow the math, I'm not sure that you follow
> any of the math.

You are talking about _total_ entropy of port numbers. As in, before the
patches, I might be using 200 different ports to send queries. With the
patches, I might use only 1 (if all the queries have identical qtypes,
queries, and destinations). But what matters is the entropy of _each_
port number. Each spoofed packet you send can only match a given (qtype,
query, destination) set. So if in the original example all 200 queries
were different, your chances of that spoofed packet matching are
_exactly the same_. And if they were not all different before the patch,
then that is the very definition of the birthday attack, and the qmerge
patches make it _worse_.

But you keep insisting that somehow the attacker's odds get better.
Please describe a sequence of attacker actions that demonstrate this.

> But math is only half the argument against Kaminsky/Day/Kings changes.  
> Trust is the other part, and the changes originate from an untrustworthy
> source, a known blackhat, who has been discredited on all of the claims
> of discovering a flaw in DNS.

No, the math is the entire argument. Either the patches impact the
birthday attack, or they don't. Either they introduce a new attack, or
they don't.

> I am quite taken aback by the notion that character doesn't matter. I
> see that a lot in some quarters---it is the common assertion of the
> dishonest---but I didn't expect it here.  It is said that character is
> to people what carbon is to steel.  One must first _have_ good character
> and good judgement to be trusted for their good character and judgement.
> Who one associates with is part of their trustworthiness and character.
> I won't trust people who naively but knowingly associate with
> disreputable people.

The character and reputation of the source of a message are a useful
guess when you are too lazy or too short on time to look at the
facts of the message (or when you refuse to open the PDF containing the
message). But they are not a valid argument against the _facts_ of the
message. Bad people saying true things does not make them untrue, nor
do good people saying untrue things make them true.

> But I am concerned because they have already proposed a change that
> would create a serious flaw by reusing QIDs, and every claim of

How are QIDs reused? Each outgoing (i.e., spoofable) packet should get a
single unique QID. If it doesn't, then there is a bug in the patch, and
I would be happy for you to point it out.

-Peff