On Tue, May 17, 2016 at 04:51:08PM +0200, Jiri Benc wrote:
> On Wed, 4 May 2016 16:36:31 +0900, Simon Horman wrote:
> > + /* Packets from user space for execution only have metadata key
> > + * attributes. OVS_KEY_ATTR_PACKET_ETHERTYPE is then used to specify
> > + * the starting layer of the packet. Packets with Ethernet headers
> > + * have this attribute set to 0
> > + */
> > + if (*attrs & (1ULL << OVS_KEY_ATTR_PACKET_ETHERTYPE)) {
> > + __be16 eth_type;
> > +
> > + if (is_mask) {
> > + /* Always exact match packet EtherType */
> > + eth_type = htons(0xffff);
> > + } else {
> > + eth_type =
> > nla_get_be16(a[OVS_KEY_ATTR_PACKET_ETHERTYPE]);
> > + is_layer3 = ((eth_type == htons(ETH_P_IP)) ||
> > + (eth_type == htons(ETH_P_IPV6)));
>
> Unknown types need to be rejected, not treated as layer2, otherwise we
> may run into problems later (with combination of this kernel + newer
> user space) when we add more types, such as ETH_P_NSH.
I believe this is fixed by the following patch.
Possibly it makes sense to squash that patch into and earlier patches.
