[738] in linux-net channel archive
Re: UDP fixes
daemon@ATHENA.MIT.EDU (Alan Cox)
Thu Jul 20 17:55:40 1995
From: iialan@iifeak.swan.ac.uk (Alan Cox)
To: swen@uni-paderborn.de (Swen Thuemmler)
Date: Thu, 20 Jul 1995 09:14:11 +0100 (BST)
Cc: root@thunder.swansea.linux.org.uk, linux-net3@www.linux.org.uk,
linux-net@vger.rutgers.edu
In-Reply-To: <Pine.SOL.3.91.950720095854.28127C-100000@gingganz> from "Swen Thuemmler" at Jul 20, 95 10:06:10 am
> > These patches when coupled with the mount side patches (appended) allow NFS with
> > multihoming, like 4.2/4.3BSD as opposed to the normal Linux/4.4BSD behaviour. I've
> > used Jon Peatfields patches of the two sets I got as his are backward compatible.
> >
> Great. At last, the patches are in. Will they go into 1.2.12?
Up to Linus, I see no problem in this.
> One possible addition:
>
> --- sock.c.orig Thu Jul 20 09:53:37 1995
> +++ sock.c Thu Jul 20 09:58:06 1995
> @@ -209,7 +209,7 @@
> *
> */
> result=sock->ops->recvfrom(sock, (void *)start,
> - size + 1024, 1, 0, NULL,
> + size + 1024, 1, 0, &(server->toaddr),
> /* Here is NFS_SLACK_SPACE..., hack */
> &addrlen);
> if (result < 0) {
>
> This will fill server-toaddr with the address of the interface we're
> receiving the packet from, and will send the next request there. The
> question is: is this sensible? Will this be slower (due to filling in the
> address) or possibly faster (less routing on the other end required).
It will be slower on 1.3.x because the keeping the same address will let
the kernel cache the route and the ethernet header for that packet, and that
does make a performance difference. On 1.2 it won't make any odds.
Alan
