On Tue, Dec 19, 2017 at 08:39:29PM +0100, Jiri Benc wrote:
> On Tue, 19 Dec 2017 13:57:53 -0500, Eric Garver wrote:
> > --- a/net/openvswitch/flow.c
> > +++ b/net/openvswitch/flow.c
> > @@ -559,8 +559,9 @@ static int parse_nsh(struct sk_buff *skb, struct
> > sw_flow_key *key)
> > * of a correct length, otherwise the same as skb->network_header.
> > * For other key->eth.type values it is left untouched.
> > *
> > - * - skb->protocol: the type of the data starting at
> > skb->network_header.
> > - * Equals to key->eth.type.
> > + * - skb->protocol: For Ethernet, the ethertype or VLAN TPID.
> > + * For non-Ethernet, the type of the data starting at
> > skb->network_header
> > + * (also equal to key->eth.type).
> > */
> > static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
> > {
> > @@ -579,6 +580,7 @@ static int key_extract(struct sk_buff *skb, struct
> > sw_flow_key *key)
> > return -EINVAL;
> >
> > skb_reset_network_header(skb);
> > + key->eth.type = skb->protocol;
> > } else {
> > eth = eth_hdr(skb);
> > ether_addr_copy(key->eth.src, eth->h_source);
> > @@ -592,15 +594,14 @@ static int key_extract(struct sk_buff *skb, struct
> > sw_flow_key *key)
> > if (unlikely(parse_vlan(skb, key)))
> > return -ENOMEM;
> >
> > - skb->protocol = parse_ethertype(skb);
> > - if (unlikely(skb->protocol == htons(0)))
> > + key->eth.type = parse_ethertype(skb);
> > + if (unlikely(key->eth.type == htons(0)))
> > return -ENOMEM;
> >
> > skb_reset_network_header(skb);
> > __skb_push(skb, skb->data - skb_mac_header(skb));
> > }
> > skb_reset_mac_len(skb);
> > - key->eth.type = skb->protocol;
> >
> > /* Network layer. */
> > if (key->eth.type == htons(ETH_P_IP)) {
>
> Unfortunately, this does not work. key_extract must set skb->protocol
> even for Ethernet frames that come from a mixed L2/L3 tunnel. Such
> packets will have key->mac_proto set to MAC_PROTO_ETHERNET and
> skb->protocol set to ETH_P_TEB (see key_extract_mac_proto). In
> key_extract, skb->protocol has to be correctly set to the dissected
> value.
AFAICS, it's always overridden to ETH_P_TEB on output by
ovs_vport_send() and that's the sole reason it works today.
For dissecting, the L2 case is currently setting skb->protocol to the
real ethertype (e.g. 0x800) not ETH_P_TEB. For RX from tunnel case it'll
indeed be ETH_P_TEB.
>
> Which means that we have to check for the existence of inner vlan tag
> (by checking key->eth.cvlan.tci or, perhaps better, by returning it
> from parse_vlan) and set skb->protocol accordingly.
>
> Jiri
