netdev
[Top] [All Lists]

Re: [patch 4/10] s390: network driver.

To: Jeff Garzik <jgarzik@xxxxxxxxx>
Subject: Re: [patch 4/10] s390: network driver.
From: Paul Jakma <paul@xxxxxxxx>
Date: Sun, 21 Nov 2004 08:16:34 +0000 (GMT)
Cc: Thomas Spatzier <thomas.spatzier@xxxxxxxxxx>, linux-kernel@xxxxxxxxxxxxxxx, Netdev <netdev@xxxxxxxxxxx>
In-reply-to: <Pine.LNX.4.61.0411150735070.10262@xxxxxxxxxxxxxxxxxx>
Mail-followup-to: paul@xxxxxxxxxxxxxxxxxx
References: <OF88EC0E9F.DE8FC278-ONC1256F4A.0038D5C0-C1256F4A.00398E11@xxxxxxxxxx> <4196B4E9.40502@xxxxxxxxx> <Pine.LNX.4.61.0411150735070.10262@xxxxxxxxxxxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
On Mon, 15 Nov 2004, Paul Jakma wrote:

non-raw/header-included sockets, eg BGP tcp sockets, a user like GNU Zebra / Quagga would much prefer packets to be dropped.

Ur... not for TCP.. obviously.

Anyway, is there any advice on how applications that use a single socket for raw/udp should deal with this new behaviour? All of the link-orientated routing protocol daemons in Quagga/GNU Zebra are going to break on Linux with this new behaviour.

Should such applications be changed to open a seperate socket per interface? Or could we have a SO_DROP_DONT_QUEUE sockopt to allow a privileged application to retain the previous behaviour, or some way to flush the queue for a socket?

Using a socket per interface wont address problem of sending quite stale packets when a link comes back after a long time down, AUI. (not a huge problem - but not nice).

Jeff???

regards,
--
Paul Jakma      paul@xxxxxxxx   paul@xxxxxxxxx  Key ID: 64A2FF6A
Fortune:
Be incomprehensible.  If they can't understand, they can't disagree.

<Prev in Thread] Current Thread [Next in Thread>