Re: NetworkManager creating IPv6 routes on the fly?

[Pavel Simerda <psimerda redhat com>]

----- Original Message -----
> From: "Dan Williams" <dcbw redhat com>
> To: "Bjørn Mork" <bjorn mork no>
> Cc: "Pavel Simerda" <psimerda redhat com>, networkmanager-list gnome org
> Sent: Thursday, February 13, 2014 7:05:59 PM
> Subject: Re: NetworkManager creating IPv6 routes on the fly?
>
> On Thu, 2014-02-13 at 11:05 +0100, Bjørn Mork wrote:
> > Pavel Simerda <psimerda redhat com> writes:
> >
> > > Those are device routes created to mimic the kernel-installed routes
> > > but with a proper metric to prioritize the selected interface. It is a
> > > hack in 0.9.8 (and prior) to get over kernel's crazy autoconfiguration
> > > feature.
> >
> > Trying to work around kernel features you don't like is guaranteed to
> > cause you and your users a lot of pain.

It certainly was causing a lot of pain and that's why I decided to get rid of kernel autoconf entirely some 
time ago and also why other developers agreed with me. They have recently updated this decision with a bit of 
more cooperation regarding temporary addresses and added kernel features for that.

> > This isn't supposed to be a war between kernel and userspace.

NetworkManager is often the main driver for kernel network configuration features. You are welcome to install 
and test a git snapshot of NetworkManager and/or participate in the development.

> We've actually been doing that, for example with the IPv6 tempaddr
> changes that recently went upstream.  We intend to keep doing that going
> forward too.  It's just that 0.9.8 was the point where we started doing
> that.
>
> That said, we haven't yet figured out how to solve the metric thing that
> Pavel is talking about here.

Yep. At least not some specific details like changing the metric without disturbing network communication, 
and also the specific way of choosing the metric. In my opinion, the kernel is not yet mature for this kind 
of operation.

Pavel