> On Nov 24, 2016, at 8:10 AM, Jiri Benc <jb...@redhat.com> wrote:
>
> On Tue, 22 Nov 2016 20:09:34 -0800, Jarno Rajahalme wrote:
>> Do not set skb->protocol to be the ethertype of the L3 header, unless
>> the packet only has the L3 header. For a non-hardware offloaded VLAN
>> frame skb->protocol needs to be one of the VLAN ethertypes.
>>
>> Any VLAN offloading is undone on the OVS netlink interface. Due to
>> this all VLAN packets sent to openvswitch module from userspace are
>> non-offloaded.
>
> This is exactly why I wanted to always accelerate the vlan tag, the
> same way it is done in other parts of the networking stack: to prevent
> all those weird corner cases.
>
> Looks to me this is the only real way forward.
>
I’m not sure what you suggest here. Obviously the kernel ABI can not be changed
as existing userspace code expects upcalled packets to be non-accelerated.
Also, if userspace pushes vlan headers, the packet will actually have them.
> This patch is wrong, it would leave skb->protocol as ETH_P_TEB for L2
> frames received via ARPHRD_NONE interface.
>
Would this incremental fix this:
diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c
index 9be9fda..37f1bb9 100644
--- a/net/openvswitch/flow.c
+++ b/net/openvswitch/flow.c
@@ -354,6 +354,8 @@ static int parse_vlan(struct sk_buff *skb, struct
sw_flow_key *key)
res = parse_vlan_tag(skb, &key->eth.vlan);
if (res <= 0)
return res;
+ if (skb->protocol == htons(ETH_P_TEB))
+ skb->protocol = key->eth.vlan.tpid;
}
/* Parse inner vlan tag. */
Jarno
