Re: djbdns/dnscache poisoning weakness

On Wed, Feb 11, 2009 at 03:32:23PM -0600, David Nicol wrote:

> I have not looked at your patch.  I had imagined a state machine,
> where a query, either from a client or intermediate, is either (1)
> cached within timeout or (2) unknown.  And I imagined the combination
> process could be implemented by introducing a third state, (3) query
> in process for this information with attendant query attachment
> mechanism.

OK, that is more or less how it works. But instead of a state machine,
it is actually a merging layer between the rest of dnscache and the
dns_transmit library. I.e.,

  dnscache ------> dns_transmit
        send Q to IP
           <------
           answer

becomes

  dnscache -----> qmerge ------> dns_transmit
                   | ^
                   v |
            array of outstanding
              transmissions

I'm sure there are probably better or more efficient ways to implement
it (e.g., using a state machine in the cache), but I wanted to keep the
changes very minimal and easy to read.

-Peff