Re: [PATCH 1/2] bridge: Check if vlan filtering is enabled only once.

[Vlad Yasevich <vyasevic@redhat.com>]

On 09/14/2014 10:43 AM, Toshiaki Makita wrote:
> (14/09/13 (土) 5:26), Vladislav Yasevich wrote:
>> The bridge code checks if vlan filtering is enabled on both
>> ingress and egress.   When the state flip happens, it
>> is possible for the bridge to currently be forwarding packets
>> and forwarding behavior becomes non-deterministic.  Bridge
>> may drop packets on some interfaces, but not others.
>>
>> This patch solves this by caching the filtered state of the
>> packet into skb_cb on ingress.  The skb_cb is guaranteed to
>> not be over-written between the time packet entres bridge
>> forwarding path and the time it leaves it.  On egress, we
>> can then check the cached state to see if we need to
>> apply filtering information.
> ...
>> @@ -270,7 +275,8 @@ bool br_should_learn(struct net_bridge_port *p, struct sk_buff *skb, u16 *vid)
>>   	struct net_bridge *br = p->br;
>>   	struct net_port_vlans *v;
>>   
>> -	if (!br->vlan_enabled)
>> +	/* If filtering was disabled at input, let it pass. */
>> +	if (!BR_INPUT_SKB_CB(skb)->vlan_filtered)
>>   		return true;
>>   
>>   	v = rcu_dereference(p->vlan_info);
>>
> I'm afraid br_should_learn() is not called after calling
> br_allowed_ingress(), so vlan_filtered doesn't seem to be initialized at
> this point.
> 

You are right.  This the local input path so it can still use vlan_enabled. I'll resubmit.

Thanks
-vlad
> Thanks,
> Toshiaki Makita
>