Re: UDP bugs
Richard Gooch (rgooch@atnf.csiro.au)
Tue, 15 Dec 1998 01:36:30 +1100
Jamie Lokier writes:
> On Mon, Dec 14, 1998 at 12:53:33PM +1100, Richard Gooch wrote:
> > kuznet@ms2.inr.ac.ru writes:
> > > rmem_max == 65535 by default.
> > >
> > > Taking into account 16 bytes of mac header and 4+alignment
> > > for skb refcnt socket with default rmem_max can receive
> > > only 65476 byte frames.
> > >
> > > So that, to play with such frames or with IPv6 jumbograms
> > > rcvbuf must be increased. By user/admin, rather than by default,
> > > certainly.
> >
> > Yes, I've just tried:
> > int buflen = 65536 + 32;
> > setsockopt (fd, SOL_SOCKET, SO_RCVBUF, &buflen, sizeof buflen);
> >
> > and now I can send and receive packets with payload size of 65507
> > bytes with Linux 2.1.126+.
>
> I would have expected SO_RCVBUF to adjust the application-visible buffer
> size for a UDP socket, the TCP window size for a TCP socket, and the raw
> buffer size for a raw socket.
>
> The reason I expect this is that an application dealing with a UDP
> socket shouldn't have to know the UDP, IP and MAC header sizes. What if
> you pass it a non-ethernet bound address, and IPV6 bound socket, some
> other datagram type, an IP with source route, etc.?
I would agree with this. However, the default is 65535, and that is
clearly not big enough to receive 65507 bytes.
> So perhaps the SO_RCVBUF sockopt should be changed to reflect this?
> Or another sockopt, SO_APPRCVBUF?
I'd rather have SO_RCVBUF changed.
Regards,
Richard....
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/