Re: [PATCH net-next 06/13] ipv6: Fix nexthdr for reinjection

[Shmulik Ladkani <shmulik.ladkani@gmail.com>]

Hi,

On Thu, 12 May 2016 14:45:36 -0700 Tom Herbert <tom@herbertland.com> wrote:
> On Thu, May 12, 2016 at 1:23 PM, Shmulik Ladkani
> <shmulik.ladkani@gmail.com> wrote:
> >> @@ -222,13 +222,14 @@ static int ip6_input_finish(struct net *net, struct sock *sk, struct sk_buff *sk
> >>        */
> >>
> >>       rcu_read_lock();
> >> -resubmit:
> >> +
> >>       idev = ip6_dst_idev(skb_dst(skb));
> >>       if (!pskb_pull(skb, skb_transport_offset(skb)))
> >>               goto discard;
> >>       nhoff = IP6CB(skb)->nhoff;
> >>       nexthdr = skb_network_header(skb)[nhoff];
> >>
> >> +resubmit:  
> >
> > This has already been attempted in 0243508edd "ipv6: Fix protocol
> > resubmission" and reverted in 1b0ccfe54a.
> >
> > It looks that in some genuine extension header handling cases of ipv6
> > (not related to encapsulation), the original resubmission code REALLY
> > requires one to re-read IP6CB(skb)->nhoff and refetch the nexthdr.
> >  
> Is there any reason why the EH handlers can't read the nexthdr and return that?

One additional thing:

Seems the

    if (!pskb_pull(skb, skb_transport_offset(skb)))

located at the original resubmit label was also necessary, as the EH
handlers may increment skb->transport_header (both ipv6_destopt_rcv and
ipv6_rthdr_rcv do so).

So if we'd like to read the nexthdr at the EH handlers we should repeat
the "skb pull; read nexthdr from skb_network_header(skb)[new nhoff];
return nexhdr;" prior each positive return from EH handlers.

Thanks
Shmulik