On Mon, Jul 26, 2004 at 07:30:24PM +0000, Thorsten Glaser wrote:
quoted text
> Definitively not OpenBSD - they do not support IPv4-mapped
> addresses (NetBSD does, but only after enabling a certain
> sysctl).

Why do you think he has a need for those addresses?

quoted text
> IPv4-mapped addresses allow you to listen on :: port 100 on
> one socket (AF_INET6, TCP) and get connections from IPv4-only
> hosts via ::FFFF:a.b.c.d addresses. (Same goes for IPX and some
> weird antique other protocol, but no mapping scheme has been
> specified yet for these AFAIK.)

Yes - mapped addresses - as thought of in the RFC's - are actually nice
but since some translation mechanisms use them on the wire they suck.
Itojun has stated that more than once...

quoted text
> If you take a certain networking d=E6mon, for example DJB tinydns,
> and want to make it work with IPv6, you either have to start two
> different instances, one listening on 0.0.0.0 port 53 UDP AF_INET
> and one on :: port 53 UDP AF_INET6, or convert it into a complicated
> application using select(). itojun-sensei still hasn't persuaded
> us to do the latter, and having to do the former sucks (code
> duplication, for example), especially in ports.

Well using select() and one more socket isn't that comlicated is it?

quoted text
> Of course, there's always the "security" argument, but I don't
> really believe in that - except for the "code complexity in kernel"
> part of it.

And if it is only the last part you believe in it is IMHO just enough to
get rid of that code in the Kernel. If he wants a solid and secure stack
OpenBSD might be an option.

    Mischa

--=20
"Mit seinem Geld begnuegt sich keiner, mit seinem Verstand jeder."
 (Quelle: Unbekannt)