Re: DNS Speculation

[marc_bevand () rapid7 com]

Philippe Biondi <phil () secdev org> wrote on 07/26/2008 01:00:57 PM:
> Did you do something like:
>
> for i in range(500):
>      send(pkt)
> or
> while True:
>      send(pkt)
>
> In this case the "optimised" way to do it would be (resp.) :
>
> send(pkt*500)

Passing a list of packets to send() was the first optimization I tried.
But it was still too slow.

> At last resort, you have the sendpfast() function which Bryan
> spoke about.

I didn't try sendpfast(), but, yes, according to Bryan this seems to
be the solution.

> Or did the problem was that the time between creating the packet and 
> having it hit the wire is too big because most of the initialisation 
> done by send() could have been done before ?
>
> In this case, the way to do it would be to avoid send()'
> high-levelness and directly use scapy layer 3 socket:
>
> s = conf.L3socket()
> while True:
>    ** do stuff, create pkt **
>    s.send(pkt)

This is the 2nd optimization I tried, and the fastest scapy exploit I
could come up with. But even when stracing it, I could see extra system
calls being called between the sendto(). IIRC this 2nd optimization
allowed me to send packets at a rate about 2 or 3 times faster than my
first Python version, but it was still about twice as slow as my C
implementation.

> In any cases, I'm interested by feedback on improvements this could 
> provide.

I think sendpfast() is a good solution for this case.

-marc
_______________________________________________
Dailydave mailing list
Dailydave () lists immunitysec com
http://lists.immunitysec.com/mailman/listinfo/dailydave