Re: [Patch net-next] vxlan: revert "vxlan: Bypass encapsulation if the destination is local"

[Sridhar Samudrala <sri@us.ibm.com>]

On 4/12/2013 12:19 PM, David Miller wrote:
> From: Sridhar Samudrala <sri@us.ibm.com>
> Date: Thu, 11 Apr 2013 16:59:59 -0700
>
>> @@ -1013,7 +1013,7 @@ static netdev_tx_t vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
>>   	}
>>   
>>   	/* Bypass encapsulation if the destination is local */
>> -	if (rt->rt_flags & RTCF_LOCAL) {
>> +	if (rt->dst.dev->flags & IFF_LOOPBACK) {
>>   		struct vxlan_dev *dst_vxlan;
> This looks terrible, and ad-hoc.  You need to find a cleaner
> way to express exactly what you're trying to do otherwise I'm
> reverting your change.
The idea is to bypass encap if the destination vxlan endpoint is on the same
host. To do this, i thought it is good enough to check if the route to reach
the destination ip is a local route.(RTCF_LOCAL is set).
This works fine for all the cases where destination ip is assigned to a
normal ethernet device. rt->dst.dev points to loopback device.

In case of veth(veth0,veth1), where the peer(veth1) and the destination 
vxlan
endpoint are on a different netns, ip_route_output for veth1's ipis 
returning
a route entry that has RTCF_LOCAL set and rt->dst.dev pointing to veth0.
Is it a bug that RTCF_LOCAL is set here when veth0's peer is moved to a
different netns?
We don't want to bypass encap in this scenario.

To address this behavior seen with veth, i had to change the if 
condition to
check for rt->dst.dev->flags rather than rt->rt_flags.

Thanks
Sridhar