From mboxrd@z Thu Jan 1 00:00:00 1970 From: Patrick McHardy Subject: Re: pmtu discovery on sa esp Date: Wed, 13 Jun 2007 14:50:34 +0200 Message-ID: <466FE81A.80907@trash.net> References: <466E92E4.4020008@trash.net> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-15 Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org To: Marco Berizzi Return-path: Received: from stinky.trash.net ([213.144.137.162]:63045 "EHLO stinky.trash.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756254AbXFMMxv (ORCPT ); Wed, 13 Jun 2007 08:53:51 -0400 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Marco Berizzi wrote: > Patrick McHardy wrote: > >>We have some MTU opimiztations in 2.6.22-rc that might be related. >>Please check with tcpdump what exactly is happening and whether >>the 2.6.22-rc box is sending too large packets. > > > I have done a tcpdump capture on the external > interface but I don't see anything strange. Try dumping on loopback as well. > (I can send to you the capture if you want/need) > I have noticed that the mtu on the aes tunnels > now is equal to 1450 byte (with 2.6.21 it was > 1428). Let me explain: > > linux 2.6.22-rc4 ->>-AES tunnel ->>- linux 2.6.21 mtu=1450 > linux 2.6.21 ->>-AES tunnel ->>- linux 2.6.22-rc4 mtu=1428 > > Now as a collateral effects all the windoze boxes > aren't able to exchange large packets: I must > upgrade all ipsec gateway to 2.6.22-rc4 (or > downgrade this box to 2.6.21 again). Hints? The question is whether 1450 is correct. Could you send me the output of "ip x s" (obfuscate keys if you want) and "ip x p"? What is the MTU of the underlying device? Do the encapsulated packets still fit? BTW, are you just using pluto or the entire openswan patch?