Re: [PATCH v2] Implement IP_UNICAST_IF socket option.

[Eric Dumazet <eric.dumazet@gmail.com>]

Le jeudi 02 février 2012 à 15:29 -0700, Erich E. Hoover a écrit :
> On Thu, Feb 2, 2012 at 2:57 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > Le jeudi 02 février 2012 à 14:36 -0700, Erich E. Hoover a écrit :
> > > ...
> > > +             if (sk->sk_bound_dev_if && ifindex != sk->sk_bound_dev_if)
> > > +                     break;
> >
> > what happens if later sk_bound_dev_if is changed ? Should we redo the
> > above tests ?
> 
> Would it be better if ip_default_ifindex performed the same comparison
> and prioritized the sk->sk_bound_dev_if value?
> 

I guess you should mimic other OS behavior. Tests are probably needed.

> > > ...
> > > +     case IP_UNICAST_IF:
> > > +     {
> > > +             __be32 iface;
> > > +
> > > +             len = sizeof(__be32);
> > > +             iface = htonl(inet->outif_index);
> > > +             release_sock(sk);
> > > +
> > > +             if (put_user(len, optlen))
> > > +                     return -EFAULT;
> > > +             if (copy_to_user(optval, &iface, len))
> > > +                     return -EFAULT;
> > > +             return 0;
> >
> > wow.... thats not pretty...
> >
> >        what about :
> >
> >        case IP_UNICASE_IF:
> >                val = htonl(inet->outif_index);
> >                break;
> >
> 
> The return value needs to be 4 bytes even if sizeof(int) != 4.
> 

On all arches supported by linux, sizeof(int) == 4

Maybe avoid sparse warning since val is not __be32

	val = (__force int)htonl(inet->outif_index);
	break;

> > > ...
> > IPv6 not supported ?
> 
> Nope, at least according to the documentation the option is only for
> IPv4 (IPPROTO_IP).

Probably because of the mix on interface index and IPv4 addr on "other
OS".

But if we implement interface index only, it could be used on IPv6
too... I dont know...