Re: [RFC] SPD basic actions per netdev

["Timo Teräs" <timo.teras@iki.fi>]

jamal wrote:
> On Thu, 2010-04-01 at 14:47 +0300, Timo Teräs wrote:
> 
>> The thing is that currently FWD 'dev blah' matches the interface
>> to which the packet is being forwarded to. Someone might be using
>> this feature already.
> 
> So this is the part i am missing i think. If i look at:
> 
> int ip_forward(struct sk_buff *skb)
> {
> .....
>         if (!xfrm4_policy_check(NULL, XFRM_POLICY_FWD, skb))
>                 goto drop;
> ....
> ........later forwarding happens here ...
>         if (!xfrm4_route_forward(skb))
>                 goto drop;
> ...
> }
> 
> On entry we have a legit skb->skb_iif.
> The validity check is before forwarding decision (where the interface
> the packet is being forwarded to is recognized).

On entry to ip_forward the routing decision has already been made.
Both oif and iif are valid on entry.

Currently policy_check() uses oif for SPD matching.

Do note that xfrm4_route_forward() is a no-op if there's no matching
policy. It has nothing to do with routing decision, it's purpose
is to wrap the dst_entry with xfrm_dst if the flow matches a valid
SPD.

>> Your patch changes semantics on how FWD policies are matched.
> 
> I agree if what you say earlier is true.