From: Frederic Leroy Subject: Re: scp stalls mysteriously Date: Wed, 2 Dec 2009 16:44:03 +0100 Message-ID: <20091202154403.GB30730@sd-11162.dedibox.fr> References: <20091128010156.1219012a@houba> <20091130195004.38dbc319@houba> <20091130213727.2f4047d2@houba> <20091201211945.505d3c98@houba> <20091202085925.472136e2@houba> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Netdev , Asdo , David Miller , Eric Dumazet , Herbert Xu To: Ilpo =?iso-8859-1?Q?J=E4rvinen?= Return-path: Received: from sd-11162.dedibox.fr ([88.191.70.230]:54329 "EHLO sd-11162.dedibox.fr" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753644AbZLBPuX (ORCPT ); Wed, 2 Dec 2009 10:50:23 -0500 Content-Disposition: inline In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On Wed, Dec 02, 2009 at 02:59:20PM +0200, Ilpo J=E4rvinen wrote: > On Wed, 2 Dec 2009, Frederic Leroy wrote: >=20 > > Le Tue, 1 Dec 2009 22:27:01 +0200 (EET), > > "Ilpo J=E4rvinen" a =E9crit : > What is directly inferrable from the logs: Losses happen. RTO is trig= gered=20 > for the first time (for some reason also fast retransmit is missing a= nd it=20 > most likely it should have preceeded RTO), however, the retransmissio= n=20 > that RTO should make is never captured into the tcpdump. Then when RT= O=20 > triggers for the second time, the connection silently dies immediatel= y and=20 > consistently for some reason. ...From that point on resets are sent i= f the=20 > peer keeps trying. If you want, I can add printk when a rto is made, in order to compare w= ith tcpdump. > > > I wonder if you have had some working (past) kernel version? > > I tried a 2.6.31, and it seems to work. If it really work (connecti= on > > don't stall everytime at the moment), I'll try to make a "git bisec= t" > > asap. I can confirm now that 2.6.31 works well. I made a little change during the test :=20 Instead of launching scp on sg, I launched it from houba. The scp command return then to bash with an error, without stalling. Which appears to me logical because the connection is killed from houba= point of view. =20 > In one of the cases, also the sg end dies (the 4th case). I suppose t= hat=20 > was running earlier kernel already? This case was the one with tcp_frto=3D2 and tcp_timestamps=3D0 on houba= =2E I only rebooted sg two days ago. It still 2.6.32-rc5. Nothing had been = changed on this host during all tests. tcp_frto=3D2 tcp_timestamps=3D1 --=20 =46r=E9d=E9ric Leroy