Re: [NSE] Raw sockets support

[Diman Todorov <diman.todorov () chello at>]

On 11.12.2006, at 21:02, castorpilot wrote:
> I am not good with Lua, but could coroutine do the job ? Or are they
> not thread enough ?

Coroutines are like threads which cannot be suspended from outside.  
They suspend when they feel like it. That's also how scripts are  
executed in parallel in NSE. When a script wants to do network I/O it  
says "ok, I'm done for now" - it is put in a waiting queue. When the  
I/O is finished the script is moved to a second waiting queue. The  
main loop keeps fetching scripts from this second waiting queue and  
executes them until they are either finished or they block on I/O again.

> > Solutions I can think about:
> > - modify nsock and support listening to pcap in nsock_loop
> >   (I personally don't like this solution. I think that nsock
> >    shouldn't be changed if possible.)
I don't know if there is a reason for which nsock should or should  
not support raw sockets. This might turn out to be the most doable  
option in the end. Fyodor?

> > - run two threads of nmap. in one thread listen to nsock
> >   in other listen to pcap.
> >   (probably there will be huge compatibility problems with threads
> >    and synchronization)
Maybe it is not necessary to synchronize that much. All the thread  
has to do is store the incoming packets in a synchronized buffer. The  
nsock loop can check if there is something in the buffer at each  
iteration. As far as portability is concerned - posix threads anyone?  
I personally think that modifying nsock is preferable to introducing  
a thread library dependence into nmap.

> > - set some custom flags on pcap descriptor and when
> >   packet comes signal SIGIO will be triggered.
> >   In signal handler do something to get out from nsock_loop function.
> >   (this solution is also very os-dependant and some changes to nsock
> >    will be needed)

> > - check for pcap events after nsock_loop() leaves after some time.
> >   (some packets can be lost while nsock_loop() is working)
When you're doing raw sockets you have to expect packet loss :) But I  
agree that this solution is not very elegant.

> > - extract socket descriptors from nsock and run select() on them  
> > and on pcap by hand.
> >   (there will be complications with NSE_TYPE_TIMER events)

just my 2 cents

cheers
Diman



_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://SecLists.Org