From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Wed, 17 Jan 2001 12:36:18 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Wed, 17 Jan 2001 12:36:07 -0500 Received: from Cantor.suse.de ([194.112.123.193]:2316 "HELO Cantor.suse.de") by vger.kernel.org with SMTP id ; Wed, 17 Jan 2001 12:36:04 -0500 Date: Wed, 17 Jan 2001 18:35:47 +0100 From: Andi Kleen To: Tony Gale Cc: Jussi Hamalainen , linux-kernel@vger.kernel.org Subject: Re: IP defrag (was RE: ipchains blocking port 65535) Message-ID: <20010117183547.A2528@gruyere.muc.suse.de> In-Reply-To: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2.5i In-Reply-To: ; from gale@syntax.dera.gov.uk on Wed, Jan 17, 2001 at 05:15:54PM -0000 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Jan 17, 2001 at 05:15:54PM -0000, Tony Gale wrote: > > On 17-Jan-2001 Jussi Hamalainen wrote: > > On Wed, 17 Jan 2001, Tony Gale wrote: > > > >> It looks like this is due to the odd way in which ipchains handles > >> fragments. Try: > >> > >> echo 1 > /proc/sys/net/ipv4/ip_always_defrag > > > > Thanks, this seems to do the trick. Does this oddity still exist > > in 2.4? > > > > Well, I haven't found it, but there is > /proc/sys/net/ipv4/ipfrag_high_thresh > /proc/sys/net/ipv4/ipfrag_low_thresh > /proc/sys/net/ipv4/ipfrag_time > > Perhaps 2.4 always defrags packets by default. Anyone confirm? This > is pretty much needed for any kind of firewall/masquerading system. Connection tracking always defrags as needed. masquerading/NAT/iptables with connection tracking uses that. This means that if any of these are enabled and your machine acts as a router lots of CPU could get burned in defragmentation, and packets will not forwarded until all fragments arrived. All very nasty, but unfortunately there is no alternative. -Andi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/