Re: Compatibility Libpcap/Winpcap and timout of pcap_next_ex().

[Guy Harris <guy () alum mit edu>]

On Jan 27, 2009, at 9:17 AM, Benoit wrote:

> I've start a simple protocol to communicate with FPGA using only MAC  
> layer.
> This software should run under linux and windows, however i've a  
> problem
> with the timeout of pcap_next_ex() function under linux.

There is no guarantee that, with a timeout value of N milliseconds,  
pcap_next_ex() will return within N milliseconds.  It might *never*  
return if no packets arrive.

The only *guarantee* that libpcap/WinPcap offers is that it will  
return within N milliseconds of the arrival of a packet.  On some  
platforms , the timer starts when pcap_next() or pcap_next_ex() or  
pcap_dispatch() or pcap_loop() is called.  On other platforms, the  
timer starts when the first packet arrives.  On other platforms, there  
isn't any timer.

The timeout is *NOT* intended to be used to make sure you don't block  
forever waiting for a packet.  The timeout is intended to be used so  
that, on platforms where the underlying packet capture mechanism will  
wait some amount of time to try to deliver multiple packets in one  
read, rather than waking up the process for every packet (to reduce  
the number of wakeups and user->kernel and kernel->user transitions),  
it won't wait forever if at least one packet has arrived.

If you want a timeout, so you don't wait forever for a packet to  
arrived, I'd suggest using select() or poll() on UN*X systems (other  
than *BSD and Mac OS X, where select() and poll() might not work  
correctly on BPF devices - and, in the case of Mac OS X, where poll()  
doesn't work *at all* on "character" devices such as BPF devices or  
ttys).

> Moreover sometimes I need to loop 10 times to be sure to receive
> correctly the packet,

Why?  Does pcap_next_ex() return 0 or a negative value?
-
This is the tcpdump-workers list.
Visit https://cod.sandelman.ca/ to unsubscribe.