From mboxrd@z Thu Jan 1 00:00:00 1970 From: Hannes Frederic Sowa Subject: Re: IPV6 routing problem Date: Wed, 22 Jan 2014 08:07:59 +0100 Message-ID: <20140122070759.GB28225@order.stressinduktion.org> References: Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Cc: netdev@vger.kernel.org To: Sharat Masetty Return-path: Received: from order.stressinduktion.org ([87.106.68.36]:57130 "EHLO order.stressinduktion.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750853AbaAVHIO (ORCPT ); Wed, 22 Jan 2014 02:08:14 -0500 Content-Disposition: inline In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: Hi! On Tue, Jan 21, 2014 at 06:41:58PM -0700, Sharat Masetty wrote: > I have an IPV6 routing problem that has only surfaced on a 3.10 > kernel version. This problem is not seen on 3.4 kernel. I will keep > the problem statement as brief as possible. Could you do me a favor and test this on a recent 3.13 kernel? Thanks! Please also state the specific kernel version (I guess you use one). > I have two interfaces say eth0 and eth1. I have a host route to a > destination over eth1, but this route has a global gateway > address(via) of the router on the link. When this global gateway is > present in the route, the kernel does not seem to pick up the route, > but falls back to the default route which is eth0. When this gateway > is removed from the route or if the gateway is changed to a link local > address of the router, instead of the global address, then routing > seems to work as expected and kernel picks up interface eth1. > > == does not work == > via dev eth1 metric 1024 > dev eth1 metric 1024 Do you have CONFIG_IPV6_ROUTER_PREF activated? > > I checked the git commits diff between 3.4 and 3.10 for ipv6/route.c > and ipv6/ip6_fib.c, but looks like they are close to an year apart > with lots of changes. Hence I wanted to reach out to you to see if any > recent changes in IPV6 routing is causing this difference in behavior. It should be a problem in ipv6/route.c, ip6_fib.c does not really care about interfaces that much. Thanks, Hannes