From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: [PATCH net-next] tcp: tcp_transmit_skb() optimizations Date: Sat, 12 Oct 2013 01:06:05 -0400 (EDT) Message-ID: <20131012.010605.1320162777840311015.davem@davemloft.net> References: <20131011.174858.1461090000704613432.davem@davemloft.net> <1381531436.4971.114.camel@edumazet-glaptop.roam.corp.google.com> <1381540819.4971.117.camel@edumazet-glaptop.roam.corp.google.com> Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, fitz@cs.yale.edu, ycheng@google.com, ncardwell@google.com To: eric.dumazet@gmail.com Return-path: Received: from shards.monkeyblade.net ([149.20.54.216]:59727 "EHLO shards.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750766Ab3JLFGI (ORCPT ); Sat, 12 Oct 2013 01:06:08 -0400 In-Reply-To: <1381540819.4971.117.camel@edumazet-glaptop.roam.corp.google.com> Sender: netdev-owner@vger.kernel.org List-ID: From: Eric Dumazet Date: Fri, 11 Oct 2013 18:20:19 -0700 > On Fri, 2013-10-11 at 15:43 -0700, Eric Dumazet wrote: >> On Fri, 2013-10-11 at 17:48 -0400, David Miller wrote: >> >> > This patch is correct, so I've applied it, but it points out a bug. >> > >> > The __tcp_retransmit_skb() code that does a __pskb_copy() to handle >> > NET_IP_ALIGN violations and skb_headroom() overflows is buggy because >> > it needs to store a congestion control timestamp in the original 'skb' >> > since that's what we'll look at in the retransmit queue. >> >> Yes, I saw that, indeed. >> >> I added it as low priority bug for the moment, as the default congestion >> module do not really care, and this case is really unlikely ;) >> > > Ah, I remember now that the conclusion was : > the timestamp is not taken into account for retransmits. > > ( FLAG_RETRANS_DATA_ACKED ) Great, thanks for the clarification.