Miroslav Lichvar writes:
> On Thu, Mar 31, 2016 at 07:50:08AM +0000, Harlan Stenn wrote:
> > Miroslav Lichvar writes:
> > > Any state that the protocol would require on the server is a potential
> > > vector for a DoS attack. How easy is for an attacker to make new
> > > "connections" and allocate memory on the server?
> > 
> > While great hunks of this are implementation-dependent, I suspect no
> > additional memory will be required to answer "now" as opposed to
> > "later".  The difference is "duration" and it's easy enough to deal with
> > that.  There are other steps that can be taken, too.
> 
> If you want to respond later, you need to save that response
> somewhere. And when you have too many reponses saved, you need to drop
> some of them.

How is that significantly different from having lots of NTS connections
starting up at the same time?

Regardless, this is a pretty easy situation to address.  I'm not worried
about it at all.

> > > Can the attacker prevent other clients from fully forming the NTS
> > > association?
> > 
> > How?
> 
> By forcing the server to drop the saved response which the client will
> need. If the server doesn't save any state, there is nothing for the
> attacker to play with.

Same as above.  This is something that's pretty easy to address.

> > > Ok, but why would you not send the result to the client when the
> > > operation is finished? Are you expecting it will take longer than some
> > > minimum expected polling interval of the client (e.g. 1 second)?
> > 
> > In a client/server situation, how would you propose to do that before
> > the next poll?
> 
> Like with any other NTP packet? Respond immediately when the request
> is processed. The crypto operations should be much faster than the 2s
> iburst interval, I don't think we need to worry about clients sending
> new request before they get the reply. If there are more requests than
> the server can handle, then some of them will be dropped.

Again, that's a local policy choice.  If you choose that approach and
the crypto work is "expensive" you've just traded one attack vector for
another.

H

_______________________________________________
TICTOC mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tictoc

Reply via email to