Linux Ethernet Bridging - Feb 2015 - [Bridge] bride: IPv6 multicast snooping enhancements

Hi Vasily,

On Tue, Feb 10, 2015 at 11:44:29AM +0300, Vasily Averin
wrote:
> This patch prevent forwarding of ICMPv6 in bridges,
> so containers/VMs with virtual eth adapters connected in local bridge
cannot ping each other via ipv6 (but can do it via ipv4)
If a host wants to receive packets, then it needs to signalize
that via MLD. If your host does not do that, then it is expected
to not receive ICMPv6 echo requests to multicast addresses. An
exception is ff02::1, that should always work.
> 
> Could you please clarify, is it expected behavior?
> Do we need to enable multicast routing or multicast_snooping on all local
ports on such bridges to enable just ICMPv6?
Nope, you shouldn't. You'd need multicast listeners. You shouldn't
need to play with the bridge settings to fix protocols.
> I believe ICMPv6 is an exception and should not be filtered by multicast
spoofing.
Signaling multicast joins is mandatory by the IPv6 standard. If
your protocol/application does not do that, then it seems to me
that the application might be broken.


By the way, which kernel version(s) are you using?

Cheers, Linus

On 10.02.2015 14:44, Linus L?ssing wrote:
> Hi Vasily,
> 
> On Tue, Feb 10, 2015 at 11:44:29AM +0300, Vasily Averin wrote:
>> This patch prevent forwarding of ICMPv6 in bridges,
>> so containers/VMs with virtual eth adapters connected in local bridge
cannot ping each other via ipv6 (but can do it via ipv4)
> 
> If a host wants to receive packets, then it needs to signalize
> that via MLD. If your host does not do that, then it is expected
> to not receive ICMPv6 echo requests to multicast addresses. An
> exception is ff02::1, that should always work.
Thank you for explanation, seems now I understand finally how it should work.

I'm trying to fix ICMPv6 processing broken in OpenVZ after rebase to last
RHEL6u6 kernel.
After some unclear manipulation bridge begins to forward icmp6 NS (fe02::1) into
wrong port,
and at present I do not found the reason of this failure.

Thank you,
	Vasily Averin