From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: [PATCH] tcp FRTO: in-order-only "TCP proxy" fragility workaround Date: Fri, 22 Aug 2008 14:37:09 -0700 (PDT) Message-ID: <20080822.143709.65615512.davem@davemloft.net> References: <20080819213417.45133573@tux> <20080822183224.2d52f16c@tux> Mime-Version: 1.0 Content-Type: Text/Plain; charset=iso-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: ilpo.jarvinen@helsinki.fi, thomas.jarosch@intra2net.com, billfink@mindspring.com, netdev@vger.kernel.org, kaber@trash.net, netfilter-devel@vger.kernel.org, kadlec@blackhole.kfki.hu To: fragabr@gmail.com Return-path: Received: from 74-93-104-97-Washington.hfc.comcastbusiness.net ([74.93.104.97]:51722 "EHLO sunset.davemloft.net" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1757792AbYHVVhL convert rfc822-to-8bit (ORCPT ); Fri, 22 Aug 2008 17:37:11 -0400 In-Reply-To: <20080822183224.2d52f16c@tux> Sender: netdev-owner@vger.kernel.org List-ID: =46rom: D=E2niel Fraga Date: Fri, 22 Aug 2008 18:32:24 -0300 > On Wed, 20 Aug 2008 15:37:13 +0300 (EEST) > "Ilpo J=E4rvinen" wrote: >=20 > > Based on irc discussion with davem, there is a htb bug which can ca= use=20 > > corruption of the retransmitted TCP packets (and then a discard due= to=20 > > checksum mismatch). That would also explain the strange headers I n= oticed=20 > > earlier. There's a patch below (should apply to 2.6.26), please put= it at=20 > > least on the host(s) which use htb (I don't know if both server and= the=20 > > client do use wondershaper script or just the client). An different= =20 > > failure symptoms (one could be somehow frto related as FRTO is used= while=20 > > retransmitting) are also quite well explainable. > >=20 > > But FRTO is mostly not a suspect based on the tcpdump you provided = (no=20 > > FRTO workaround would help in that). >=20 > Ilpo, I have good news. I decided to disable completely HTB and > the problem seems to have gone. And frto is enabled, of course. So th= e > problem was with HTB, not frto as I thought. >=20 > The HTB patches you provided are going to be included in the > next 2.6.27 kernel, right? Yes, but it's important that you verify that the patch makes the problem go away when HTB is enabled. Please make this test if you can.