[ovs-dev] [PATCH 11/12] datapath: compat: Add support for IPv6 UDP tunnel segmentation.

[Jesse Gross]

On Wed, May 4, 2016 at 4:35 PM, Pravin B Shelar <pshelar at ovn.org> wrote:
> diff --git a/datapath/linux/compat/include/net/ipv6.h b/datapath/linux/compat/include/net/ipv6.h
> index 5cc5b6e..6faafb0 100644
> --- a/datapath/linux/compat/include/net/ipv6.h
> +++ b/datapath/linux/compat/include/net/ipv6.h
> @@ -81,4 +81,9 @@ static inline __be32 ip6_make_flowinfo(unsigned int tclass, __be32 flowlabel)
>         return htonl(tclass << IPV6_TCLASS_SHIFT) | flowlabel;
>  }
>
> +#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
> +#define ip6_local_out rpl_ip6_local_out
> +int rpl_ip6_local_out(struct sk_buff *skb);
> +#endif
> +
>  #endif

The IPv4 version is defined in gso.h and the implementations of both
functions are in gso.c. Should we move both definitions to one header
or another?

> +void ovs_udp_csum_gso(struct sk_buff *skb)
> +{
> +       int udp_offset = skb_transport_offset(skb);
> +
> +       ovs_udp_gso(skb);
> +
> +       if (!OVS_GSO_CB(skb)->ipv6) {
> +               struct iphdr *iph = ip_hdr(skb);
> +
> +               /* csum segment if tunnel sets skb with csum. The cleanest way
> +                * to do this just to set it up from scratch. */
> +               skb->ip_summed = CHECKSUM_NONE;
> +               udp_set_csum(false, skb, iph->saddr, iph->daddr,
> +                               skb->len - udp_offset);
> +#if IS_ENABLED(CONFIG_IPV6)
> +       } else {
> +               struct ipv6hdr *ip6h;
> +
> +               ip6h = ipv6_hdr(skb);
> +               udp6_set_csum(false, skb, &ip6h->saddr, &ip6h->daddr,
> +                             skb->len - udp_offset);

It seems like there is an inconsistency between IPv4 and IPv6 when it
comes to setting skb->ip_summed = CHECKSUM_NONE. Presumably, this
isn't really necessary since we are passing in 0 for feature flags to
skb_gso_segment().