Re: Veth problems with bridge

[Patrick McHardy <kaber@trash.net>]

Bernhard Miklautz wrote:
> Hi Patrick,
> 
> Patrick McHardy wrote:
>>>>> I also tried the whole setup without using veth; the IP directly bound
>>>>> to br0, as well as without the bridge at all. No problems with that.
>>>>> So there might be some problems with veth?
>>>> Does "echo 0 > /proc/sys/net/bridge/bridge-nf-call-iptables" fix it?
>>> On my hardware machine this seems to fix the problem :). But why does
>>> bridge-nf-call-iptables influent source nat on an other interface? -
>>> Shouldn't the source address always be translated when an output
>>> interface is set (iptables -A POSTROUTING -o eth3 -t nat -j MASQUERADE)?
>> The bridging code passes packets through IPv4 netfilter and
>> connection tracking, so when they hit your MASQUERADE rule,
>> the NAT mappings have already been set up.
> 
> Remember my setup veth0 and eth1 bridged together to br0, eth3 is the
> outgoing interface.
> 
> Cases:
> 
> 1) The ip address set on the bridge and no ip address on veth1 works
>    fine regardless whether bridge-nf-call-iptables is set or unset.
> 
> 2) The ip set on veth1 and no ip on the bridge the
>    MASQUERADE rule is only hit when bridge-nf-call-iptables is unset.
> 
> If I understood you correctly then the netfilters (nat/postrouting)
> would only be applied once in the latter case when
> bridge-nf-call-iptables is enabled.

No, they will be applied twice, but NAT mappings are only set up
on the first packet, so when the eth3 rule is hit, its too late.

> But if veth should behave like a "regular" interface shouldn't the
> netfilter rules be applied twice? - First when the packets enter the
> bridge on eth0 and leave it on veth0, and secondly when they enter veth1
> and and leave it at the final outgoing interface.

They are (see above). But NAT is a special case and would need
namespace-aware connection tracking and both veths living in
different namespaces for the scenario you describe (or disabled
IPv4 netfilter for bridging).