Message-ID: <20150127160915.GD13164@gospo.home.greyhouse.net>
Date:	Tue, 27 Jan 2015 11:09:15 -0500
From:	Andy Gospodarek <gospo@...ulusnetworks.com>
To:	Hannes Frederic Sowa <hannes@...essinduktion.org>
Cc:	Stephen Hemminger <stephen@...workplumber.org>,
	David Ahern <dsahern@...il.com>, netdev@...r.kernel.org
Subject: Re: [RFC PATCH] net: ipv6: Make address flushing on ifdown optional

On Tue, Jan 27, 2015 at 10:44:21AM +0100, Hannes Frederic Sowa wrote:
> Hi,
> 
> On Mo, 2015-01-26 at 23:56 -0500, Andy Gospodarek wrote:
> > On Fri, Jan 23, 2015 at 01:22:17PM +0100, Hannes Frederic Sowa wrote:
> > > On Do, 2015-01-22 at 22:40 -0800, Stephen Hemminger wrote:
> > > > On Wed, 14 Jan 2015 12:17:19 -0700
> > > > David Ahern <dsahern@...il.com> wrote:
> > > > 
> > > > > Currently, ipv6 addresses are flushed when the interface is configured down:
> > > > > 
> > > > > [root@f20 ~]# ip -6 addr add dev eth1 2000:11:1:1::1/64
> > > > > [root@f20 ~]# ip addr show dev eth1
> > > > > 3: eth1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default qlen 1000
> > > > >     link/ether 02:04:11:22:33:01 brd ff:ff:ff:ff:ff:ff
> > > > >     inet6 2000:11:1:1::1/64 scope global tentative
> > > > >        valid_lft forever preferred_lft forever
> > > > > [root@f20 ~]# ip link set dev eth1 up
> > > > > [root@f20 ~]# ip link set dev eth1 down
> > > > > [root@f20 ~]# ip addr show dev eth1
> > > > > 3: eth1: <BROADCAST,MULTICAST> mtu 1500 qdisc pfifo_fast state DOWN group default qlen 1000
> > > > >     link/ether 02:04:11:22:33:01 brd ff:ff:ff:ff:ff:ff
> > > > > 
> > > > > Add a new sysctl to make this behavior optional. Setting defaults to flush
> > > > > addresses to maintain backwards compatibility. When reset flushing is bypassed:
> > > > > 
> > > > > [root@f20 ~]# echo 0 > /proc/sys/net/ipv6/conf/eth1/flush_addr_on_down
> > > > > [root@f20 ~]# ip -6 addr add dev eth1 2000:11:1:1::1/64
> > > > > [root@f20 ~]# ip addr show dev eth1
> > > > > 3: eth1: <BROADCAST,MULTICAST> mtu 1500 qdisc pfifo_fast state DOWN group default qlen 1000
> > > > >     link/ether 02:04:11:22:33:01 brd ff:ff:ff:ff:ff:ff
> > > > >     inet6 2000:11:1:1::1/64 scope global tentative
> > > > >        valid_lft forever preferred_lft forever
> > > > > [root@f20 ~]#  ip link set dev eth1 up
> > > > > [root@f20 ~]#  ip link set dev eth1 down
> > > > > [root@f20 ~]# ip addr show dev eth1
> > > > > 3: eth1: <BROADCAST,MULTICAST> mtu 1500 qdisc pfifo_fast state DOWN group default qlen 1000
> > > > >     link/ether 02:04:11:22:33:01 brd ff:ff:ff:ff:ff:ff
> > > > >     inet6 2000:11:1:1::1/64 scope global
> > > > >        valid_lft forever preferred_lft forever
> > > > >     inet6 fe80::4:11ff:fe22:3301/64 scope link
> > > > >        valid_lft forever preferred_lft forever
> > > > > 
> > > > > Suggested-by: Hannes Frederic Sowa <hannes@...hat.com>
> > > > > Signed-off-by: David Ahern <dsahern@...il.com>
> > > > > Cc: Hannes Frederic Sowa <hannes@...hat.com>
> > > > 
> > > > Would this break existing application expecting a particular semantic
> > > > by listening to netlink?  What happens to packets received with the static
> > > > address when interface is down? With IPv4 Linux is mostly a weak host
> > > > model, and IPv6 somewhere in between.
> > > 
> > > IPv6 is mostly a weak end model, too, but IFA_LINK addresses are used
> > > much more. So yes, it is somewhere in between.
> > > 
> > > Addresses bound to interfaces which are currently down will work with
> > > IPv6 (in contrast to IPv4).
> > > 
> > > > For vendors that control the application stack or have limited number
> > > > of services this would work fine, but what about RHEL?
> > > 
> > > The new model is only enabled if the sysctl is set. I don't expect a lot
> > > of vendors or distributions switching anytime soon.
> > 
> > You are probably correct that many will not switch, but this sysctl
> > gives those who want to switch a nice option without having to carry
> > extra kernel patches.  I like it. 
> 
> I don't see any problem with the patch in general but DAD should
> definitely happen on reenabling interfaces. Current behavior does not
> seem fine to me. It's ok if this patch doesn't change this behavior now
> but a follow-up one will be needed then.
> 
> David, will you look after the DAD logic and do you plan a follow-up
> patch?
> 
> > I have been pondering a few different changes to interface address and
> > route behavior on both interface and link (gasp!) down and would like to
> > use sysctls to make those options available to those who are interested
> > without changing the current model.
> 
> Can you be more specific? In the last year we added some per interface
> flags to e.g. handle the creation of on-link routes in case of an
> address gets added to the interface. Maybe this is the better approach,
> because quite a lot of stuff happens interface local in IPv6.

Unfortunately some of the work I've been doing is on a 3.2 long-term
kernel, but I've also been testing some of my hacks^Wpatches on the
latest tree as well to make control/mark FIB entries based on
link-status (carrier not I would call 'admin state' set via 'ip link
set').

Overloading the use of the onlink could be a good option -- I'll review
your patches and the functionality a bit more closely to see if I can
merge my changes with those and clean everything up.

I know that NM, netplugd, ifplugd, networkd, etc. could be used to
handle issue like this, but an in-kernel solution has a much better
chance of being reliable when you are dealing with ~100 network
interfaces and do not want to black-hole traffic when link is down to a
router in your datacenter and there is another possible route to get to
the network that was previously directly connected.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists