Re: pcap_dispatch() using non-blocking I/O

[Guy Harris <guy () netapp com>]

On Tue, May 06, 2003 at 11:00:35PM -0400, TCPdump wrote:
> I'm having some portability problems with pcap_dispatch() on FreeBSD
> 4.8-STABLE. I doubt the problem is specific to FreeBSD, however, at the
> moment I only have Redhat 9 and FreeBSD 4.8 environments to test on. 
>
> The problem is simply that I can't seem to get pcap_dispatch() to block
> and based on the pcap man page and from what I've read on Google, it
> should be the default (even on FreeBSD).

Do you mean "block forever until a packet arrives"?

Unfortunately, there's no portable way to write code guaranteed to get
that behavior on all platforms.  The problem is that:

        1) the behavior of the underlying platform is exposed by libpcap
           (it's difficult, if not impossible, to do otherwise);

        2) on systems with BPF, either

                1) there's no timeout, in which case a read blocks until
                   the packet buffer fills, even if there are packets
                   waiting to be read;

                2) there's a timeout, in which case a read blocks until
                   the packet buffer fills *or* the timeout expires,
                   even if there are *no* packets waiting to be read
                   when the timeout expires.

The best way to handle that is simply not to worry if "pcap_dispatch()"
returns 0 - just go back into the loop and wait for more packets.


> The goal here (and it works in Linux)

Linux's underlying mechanism differs from BSDs; it doesn't have the same
type of packet buffer - a "recvfrom()" completes as soon as a single
packet arrives - and there is no timeout, so the "recvfrom()" doesn't
complete until a packet is available to read.
-
This is the TCPDUMP workers list. It is archived at
http://www.tcpdump.org/lists/workers/index.html
To unsubscribe use mailto:tcpdump-workers-request () tcpdump org?body=unsubscribe