Linux Ethernet Bridging - Sep 2009 - [Bridge] Bridge and VLANs. Strange behaviour.

Hello list,

      I am trying to understand how the linux-based bridges work but I am
watching very strange things.

      For example, if I have this configuration for the bridge BR0 with 2
physical interfaces eth0 and eth1:

[image: trying1.jpg]

     If PC1 sends an ARP REQUEST, PC2 receives this ARP REQUEST but they are
in different VLANs!
     Why the VLAN 20 can receive ARP REQUEST messages from the VLAN 10 and
why the VLAN 10 can receive ARP REPLY messages from VLAN 20 if they are
different VLANs? This is crazy for me.
     I think linux-based bridges don't work very well with VLANs (probably I
am wrong because I am just a beginner)
     Can someone help me to understand that? Your help would be very
appreciated. :-)

Thanks in advance and sorry for my English.

-- 
Gustavo Mart?n Morcuende
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.linux-foundation.org/pipermail/bridge/attachments/20090904/66beace7/attachment-0001.htm
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/jpeg
Size: 21465 bytes
Desc: not available
Url :
http://lists.linux-foundation.org/pipermail/bridge/attachments/20090904/66beace7/attachment-0001.jpeg

Gustavo Martin wrote:
> Hello list,
> 
>       I am trying to understand how the linux-based bridges work but I 
> am watching very strange things.
> 
>       For example, if I have this configuration for the bridge BR0 with 
> 2 physical interfaces eth0 and eth1:
> 
> trying1.jpg
> 
>      If PC1 sends an ARP REQUEST, PC2 receives this ARP REQUEST but they 
> are in different VLANs!
>      Why the VLAN 20 can receive ARP REQUEST messages from the VLAN 10 
> and why the VLAN 10 can receive ARP REPLY messages from VLAN 20 if they 
> are different VLANs? This is crazy for me.
>      I think linux-based bridges don't work very well with VLANs 
> (probably I am wrong because I am just a beginner)
>      Can someone help me to understand that? Your help would be very 
> appreciated. :-)
The behavior you describe is normal and exactly the one you asked for.

You have put eth0.10 and eth1.20 in the same bridge, so you asked to bridge 
between eth0.10 and eth1.20.

When a 802.1Q tagged packet is received on eth0, the kernel removes the 802.1Q 
header, having VLAN ID=10, then deliver the packet to eth0.10. Then, the kernel 
deliver the packet to the bridge, because eth0.10 is a member of the bridge.

At this time, the bridge code has no knowledge of the original VLAN ID because 
this VLAN ID was removed before the bridge code has an opportunity to look at
it.

The bridge code send the un tagged packet to the right output port if it can 
find a suitable destination mac address entry in its mac_to_port table. Else, it
simply send the untagged packet to all ports except the one where the packet was
received on.

So the untagged packet is sent to eth1.20. Then, the kernel add a 802.1Q header 
to the packet, with VLAN ID=20, then deliver the packet to eth1, to be sent on 
the wire.

And you end up with the behavior you describe.

If you want to bridge all ports in VLAN 10 together, all ports in VLAN 20 
together, and route between VLAN 10 and VLAN 20, you need to setup two different
bridges :

br10 : eth0.10 + eth1.10 + ... + ethN.10 and possibly some non-802.1Q ports.
br20 : eth0.20 + eth1.20 + ... + ethN.20 and possibly some non-802.1Q ports.

Then setup an IP address in the right subnet on br10 (192.168.1.1) and br20 
(172.16.1.1).

Then ensure that routing is enabled :

echo 1 > /proc/sys/net/ipv4/ip_forward

Feel free to ask, if all this is not clear enough.

	Nicolas.
> 
> Thanks in advance and sorry for my English.
> 
> -- 
> Gustavo Mart?n Morcuende