From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Denis V. Lunev" Subject: Re: [PATCH 1/3] [IPV6]: Event type in addrconf_ifdown is mis-used. Date: Sun, 23 Mar 2008 17:34:59 +0300 Message-ID: <1206282899.7528.18.camel@iris.sw.ru> References: <1205850925-11736-1-git-send-email-den@openvz.org> <20080322.173843.71057686.davem@davemloft.net> <1206259996.7528.12.camel@iris.sw.ru> <20080323.031724.208377895.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20080323.031724.208377895.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org To: David Miller Cc: yoshfuji@linux-ipv6.org, netdev@vger.kernel.org, kaber@trash.net, containers@lists.osdl.org, dlezcano@fr.ibm.com, benjamin.thery@bull.net List-Id: containers.vger.kernel.org On Sun, 2008-03-23 at 03:17 -0700, David Miller wrote: > From: "Denis V. Lunev" > Date: Sun, 23 Mar 2008 11:13:16 +0300 > > > First, this behaviour is broken for a namespace right now in the 2.6.26 > > tree. inet6_dev pointer will be NULL for a loopback inside the > > namespace. The case is simple. Just remove all INET6 addresses from a > > loopback device inside a VE. This will call > > inet6_addr_del > > addrconf_ifdown(dev, 1); > > if (dev == init_net.loopback_dev && how == 1) > > how = 0; > > the condition will be false and how will not be changed here. > > That's a bug. > > You can't mark any namespace's loopback device's inet6_dev as NULL > until you know that all routes, devices, and packets referring to such > devices and routes in that namespace are %100 gone and unreferenced. > > It is now obviously apparent that there are several severe errors > here. You are perfectly correct and the place in addrconf_cleanup is that place when we believe that we should destroy all the staff. You see, it is pretty useless to call addrconf_ifdown(dev, 2) after addrconf_dev(dev, 0) for a loopback in the current code! No new cleanups will be performed for 2, pls check :)