Bridge: request for the via_phys_dev feature discussion

Bridge: request for the via_phys_dev feature discussion

[Pavel Emelyanov]

Hi, Stephen.

Some time ago Cyrill Corcunov sent a patches, that added to bridges the 
ability to send/receive packets via ethernet device in it rather than via 
the bridge device itself. Here's the link on the thread:
http://www.mail-archive.com/bridge@lists.linux-foundation.org/msg01108.html

That time you told, that the problem we were trying to solve with the 
patches could be solved by other means. If you don't mind can we discus
this in more details?

In simple situation containers users want to bridge their containers through
veth device with the host eth0 adapter. But after one add his eth0 to bridge
he need to
a) reconfigure everything including IP addresses, routing tables and
   netfilter rules
b) recreate connections that were bound to eth0

That's OK if one setup the box from the scratch and adds eth0 to bridge from
the very beginning, but for those who start using containers on pre-configured
boxes or for those, who decided to switch to bridge+veth from some other 
virtual device (e.g. macvlan or venet device in OpenVZ) this becomes real pain.

I don't insist you accept the patches Cyrill proposed, I don't even insist
we rework them keeping the idea intact. I just want to know your opinion about
how to solve the above problem better.

Thanks,
Pavel

Re: Bridge: request for the via_phys_dev feature discussion

[David Miller]

From: Pavel Emelyanov <xemul@openvz.org>
Date: Thu, 18 Feb 2010 15:21:12 +0300

> In simple situation containers users want to bridge their containers through
> veth device with the host eth0 adapter. But after one add his eth0 to bridge
> he need to
> a) reconfigure everything including IP addresses, routing tables and
>    netfilter rules
> b) recreate connections that were bound to eth0
> 
> That's OK if one setup the box from the scratch and adds eth0 to bridge from
> the very beginning, but for those who start using containers on pre-configured
> boxes or for those, who decided to switch to bridge+veth from some other 
> virtual device (e.g. macvlan or venet device in OpenVZ) this becomes real pain.
> 
> I don't insist you accept the patches Cyrill proposed, I don't even insist
> we rework them keeping the idea intact. I just want to know your opinion about
> how to solve the above problem better.

Stephen, please work with Pavel on finding a solution for this issue
that works for you.

Thanks.