From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: Network hangs with 2.6.30.5 Date: Tue, 01 Sep 2009 12:47:46 +0200 Message-ID: <4A9CFBD2.2050206@gmail.com> References: <7D2F0769-2994-4BB8-B107-DEF2B1346B3A@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: netdev@vger.kernel.org To: Clifford Heath Return-path: Received: from gw1.cosmosbay.com ([212.99.114.194]:39900 "EHLO gw1.cosmosbay.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753197AbZIAKrq (ORCPT ); Tue, 1 Sep 2009 06:47:46 -0400 In-Reply-To: <7D2F0769-2994-4BB8-B107-DEF2B1346B3A@gmail.com> Sender: netdev-owner@vger.kernel.org List-ID: Clifford Heath a =E9crit : > I sent this email last Friday, but received no response. >=20 > As far as I can see, some recent work in the stable > Linux kernel has broken the TCP stack, at least on my > (pretty common) hardware. Can anyone confirm that > they've seen and perhaps fixed similar symptoms, or > at least tell me what else I need to do to help them > identify the problem? >=20 > I recently upgraded my Debian system (a Dell Optiplex GX270) from a > 2.6.16.11 kernel to a 2.6.30.5 one (current stable). My networking is > now misbehaving. If I could revert to an earlier kernel, I would (and > did, it worked), but now I can't because of the glibc version change; > the old kernel panics on startup. This leaves me with a *broken > computer* which I cannot use for my regular work, and cannot afford t= o > completely re-install. >=20 > POP, IMAP, and NNTP connect ok (multiple packets each way, viewed usi= ng > a logging proxy; no tcpdump but I can get one for you) but as soon as= a > message should start coming down, the connection hangs and then times > out. I think it's the first large packet that causes this, and I saw > that the net team have been working on some features to increase > throughput through read aggregation or something... Whatever it is, i= t's > clearly broken (on my hardware at least). >=20 > My kernel .config file and the output of "lspci" are included below. > Thanks for any help you can give. Let me know if you need more inform= ation. >=20 You could provide a tcpdump for example, and tell us which way is broke= n=20 (your machine receiving a big packet, or sending a big packet) You might try to change /proc/sys/net/ipv4/tcp_ecn to 0 echo 0 >/proc/sys/net/ipv4/tcp_ecn You might try to change device settings ethtool -K eth0 sg off and various settings as well (tso, gso ...)