Avoiding natd overhead

[Brett Glass]

At 09:50 PM 10/21/2006, Julian Elischer wrote:

>one thing that you need to name sure of is that only the packets 
>that have potential of being on interest to natd are passed to natd.

I do. In fact, this is a capability I would lose if I used 
ipfilters or pf to do NAT, which is why I want to find a way to use 
a mechanism that's triggered by IPFW.

You were the person who invented "divert sockets," were you not? 
How hard would it be to create a mechanism (a sort of "kernel 
divert socket") so that kernel modules and/or netgraph nodes could 
do the same things which are now done by userland processes 
listening on divert sockets? This would boost the performance of 
any FreeBSD machine that did NAT (which many if not most do).

--Brett Glass