[net-next] vxlan: set mac_header correctly in GPE mode

For VXLAN-GPE, the interface is ARPHRD_NONE, thus we need to reset
mac_header after pulling the outer header.

Fixes: e1e5314de08b ("vxlan: implement GPE")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
---
 drivers/net/vxlan.c | 2 ++
 1 file changed, 2 insertions(+)

Hi Jiri,

On Wed, 11 May 2016 15:55:06 +0200 Jiri Benc <jbenc@redhat.com> wrote:
> For VXLAN-GPE, the interface is ARPHRD_NONE, thus we need to reset
> mac_header after pulling the outer header.
> 
> Fixes: e1e5314de08b ("vxlan: implement GPE")
> Signed-off-by: Jiri Benc <jbenc@redhat.com>
> ---
>  drivers/net/vxlan.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
> index 2f29d20aa08f..e030a804b772 100644
> --- a/drivers/net/vxlan.c
> +++ b/drivers/net/vxlan.c
> @@ -1338,6 +1338,8 @@ static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
>  	if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
>  				   !net_eq(vxlan->net, dev_net(vxlan->dev))))
>  			goto drop;
> +	if (raw_proto)
> +		skb_reset_mac_header(skb);

Would it make sense to perform this within __iptunnel_pull_header (in
case raw_proto is true) for all __iptunnel_pull_header callers?

If not (meaning vxlan specific), in next few lines we have:

	if (!raw_proto) {
		if (!vxlan_set_mac(vxlan, vs, skb))
			goto drop;
	} else {
		skb->dev = vxlan->dev;
		skb->pkt_type = PACKET_HOST;
	}

Seems more appropriate to place the missing 'skb_reset_mac_header'
within the "else" part, which logically holds all "things to initialize
in the skb if raw_proto is true, thus havn't called vxlan_set_mac".

Regards,
Shmulik

On Thu, 12 May 2016 09:47:58 +0300, Shmulik Ladkani wrote:
> Would it make sense to perform this within __iptunnel_pull_header (in
> case raw_proto is true) for all __iptunnel_pull_header callers?

raw_proto just denotes that inner_proto of ETH_P_TEB should not be
treated specially. Guessing any other special meaning based on this
flag would be unexpected and confusing. Specifically,
__iptunnel_pull_header doesn't have and should not have any notion of
the device type; although all current users of raw_proto == true are of
ARPHRD_NONE type, it's not a requirement.

> If not (meaning vxlan specific), in next few lines we have:
> 
> 	if (!raw_proto) {
> 		if (!vxlan_set_mac(vxlan, vs, skb))
> 			goto drop;
> 	} else {
> 		skb->dev = vxlan->dev;
> 		skb->pkt_type = PACKET_HOST;
> 	}
> 
> Seems more appropriate to place the missing 'skb_reset_mac_header'
> within the "else" part, which logically holds all "things to initialize
> in the skb if raw_proto is true, thus havn't called vxlan_set_mac".

I agree. I'll respin the patch.

Thanks!

 Jiri