[Users] Problem with bonding, vlan, bridge, veth

[Kirill Korotaev]

Kristian,

thanks for sharing this info. 
However, since it looks like your problem is related to bonding and bridges
(not OpenVZ itself) I think you would be able to get quicker/better reply from
netdev at vger@kernel.org mailing list. Please, keep this mail list on CC.

Thanks,
Kirill

> On Friday den 10. November 2006 11:12, Kristian F. Høgh wrote:
> 
>>Hi list,
> 
> Hi list, will reply myelf :-)
> 
> 
>>I have a bonding/vlan/bridge/veth problem.
>>Sometimes a bridge think a veth device move to another port.
>>If I remove a physical interface from bond, the bridge behaves normally.
>>
>>Kernel 2.6.16 + openvz test020
>>VE0 Ubuntu dapper/6.06LTS, IP 172.31.1.26 on VLAN 254
>>VE1028 Debian stable/sarge/3.1, IP 10.1.28.12 on VLAN 28
>>
>>I have a server (vs5, VE0) using eth0 and eth1 in a bonding interface
>>bond0. bond0 is on tagged vlan.
>>I create a vlan device vlan254 on vlan 254. This is VE0 IP.
>>For each VE (XX) I do
>>  create a vlan device vlanXX on vlan XX.
>>  create a bridge bvXX and add vlanXX to it.
>>  create a VE (VE10XX) using veth.
>>  VETH="ve10XX.0,aa:00:04:56:YY:ZZ,eth0,aa:00:04:57:YY:ZZ"
>>  add ve10XX.0 to the bridge.
>>  YY and ZZ are calculated from VEID number (VLAN + 1000)
>>
>>      eth0     eth1
>>         \     /
>>          bond0
>>         /     \                  veth
>>  vlan254       vlanXX    ve10XX.0 -- eth0 (ve10XX)
>>    VE0              \    /
>>                      bvXX (bridge)
>>
> 
> 
> The drawing above is correct, but the part not drawed
> is the important one.
> 
> eth0 and eth1 are each connected to a switch.
> These are connected by trunk ports 1 and 2.
> The bond interface (eth0 + eth1) is in active/backup mode.
> 
> When I ping 10.1.28.101 in vlan28 from ve1028 (10.1.28.12),
> it sends the following arp request:
> aa:00:04:57:04:04 > ff:ff:ff:ff:ff:ff arp who-has 10.1.28.101 tell 10.1.28.12
> 
> The request will go from eth0 (VE1028) to ve1028.0 -> bv28 -> vlan28 -> 
> bond0 -> eth0 -> SW1port16 -> SW1 ALL ports but 16 -> including SW2port1/2 -> 
> SW2 ALL ports but 1/2 -> including target and eth1 -> bond0 -> vlan28 -> 
> bv28 -> ve1028.0 -> eth0
> 
> The target 10.28.1.101, receives the request through SW2 port 6.
> The switches/bridges gets updated as follows:
>  bv28 know aa:00:04:57:04:04 is at port 2 (ve1028.0)
>  SW1 know aa:00:04:57:04:04 is at port 16
>  SW2 know aa:00:04:57:04:04 is at port 1/2
>  bv28 know aa:00:04:57:04:04 is at port 1 (vlan28)
> Note bv28 gets updated twice.
> 
> The target replies:
> 00:03:fa:0f:a3:a7 > aa:00:04:57:04:04 arp reply 10.1.28.101 is-at ...:0f:a3:a7
> 
> The arp reply will go from SW2port6 -> SW2port1/2 ->  SW1port1/2 ->
> SW1port16 -> eth0 -> bond0 -> vlan28 -> bv28 -> NULL
> As bv28 received the arp request from "aa:00:04:57:04:04" on port 1 (vlan28)
> it will not forward the arp reply to port 2 (ve1028.0), therefore eth0 in 
> VE1028 never receives the arp reply... No communication.
> 
> So the problem is bridging over bonding.
> The backup interface receives broadcast frames and forwards them to the bridge 
> which updates its mac table.
> 
> I will test the following.
> 
> 
>        SW1 ----- SW2         
>          |        |
>        eth0     eth1
>          |        |
>      eth0.XX  eth1.XX        vlan
>           \      /
>             bvXX             bridge
>              |
>           ve10XX.0            \
>              |                 veth
>             eth0 (ve10XX)     /
> 
> I just have to make sure to use spanning tree.
> The linux box should be in blocking mode.
> 
> Comments?
> 
> Regards,
> Kristian.
> 
> _______________________________________________
> Users mailing list
> Users at openvz.org
> https://openvz.org/mailman/listinfo/users
>