Re: setting rcvbuf

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

On Jun 15, 2004, at 2:49 PM, John Heffner wrote:

> I've found that on my linux machines, setting the rcvbuf on the packet
> socket bigger helps to reduce drops at high rates.
>
> I implemented the following, though I'm not sure how this works on BSD 
> and
> other UNIXes.

It *doesn't* work on BSD, unfortunately - you can't set the buffer size 
after binding a network interface to a BPF device, and, as that's done 
at open time, you can't set the buffer size on an open pcap_t.

The ultimate fix is probably to have a new open routine that takes the 
buffer size as an argument (but don't implement that - there's a bunch 
of other stuff for which we need a new open API, and we don't want to 
introduce one new API and then keep introducing new ones; I'm thinking 
of an API that takes an attribute/value list of open parameters, so we 
can add new parameters as needed without having to introduce a whole 
new API).

A short-term fix would be to implement "pcap_setbuff()" routines for 
all platforms - but have, on platforms where you can't set it, either 
no-op routines or routines that return an error - and, when we have the 
new open routine, deprecate "pcap_setbuff()" in favor of the new open 
routine.

-
This is the tcpdump-workers list.
Visit https://lists.sandelman.ca/ to unsubscribe.