Hi,
Le dimanche 07 janvier 2007 à 15:26 +0100, Harald Welte a écrit :
> On Fri, Dec 29, 2006 at 12:40:47AM +0100, Eric Leblond wrote:
> > It will add 4*IFNAMSIZ = 64 octets to each nfnetlink_queue message but
> > this is not impressive as a part of the packet payload is usually sent.
> 
> No, please don't do that.  
> Also, I _really_ want to get rid of interface names throughout
> netfilter/iptables at some point in the future.  all 'next generation'
> subsystems (such as all nfnetlink based services) should just deal with
> ifindexes.
> 
> Please put some functions into libnfnetlink, even if it doesn't natively
> belong there.   If there's more netlink unification, we can always make
> those just wrappers for the 'real new unified' ifindex/name resolving
> functions.

I've thought at this solution before getting interessed by pablo's
suggestion. But I've got a conception problem.

It is not correct to do a dump (and send a netlink message) for each
ifindex resolution. Thus, libnfnetlink needs to listen to netlink
interfaces message. A permanent "task" is thus needed to have the job
done : It can be a dedicated thread or a carefully hidden select.

In fact, if we omit the thread approach which is somehow intrusive, I
don't see a way to do this via a simple call to added functions.

Any idea welcome.

> [yes, I'm back, more or less]

Great news !

BR,
-- 
Eric Leblond <eric@inl.fr>
INL