Subject: Re: the size of an argument for set/getsockopt(2)
To: None <tech-kern@netbsd.org>
From: Hitoshi Asaeda <Hitoshi.Asaeda@sophia.inria.fr>
List: tech-kern
Date: 05/29/2002 10:06:02
> >>Current NetBSD kernel (1.5 or later) seems the argument of
> >>set/getsockopt() is stored in a mbuf. This implies we cannot input/get
> >>larger storage than MLEN to/from it. (I think this is only inherited
> >>some historical rules.)
> >	of course we can use external mbufs, if needed...

I'm talking about general topics, not only for my personal kernel
change.

> 	we may need to change some of uicp_socket/syscalls.c, but anyway,
> 	we need no major change here.  just support bigger mbufs, that's all.

I felt the change of sys_setsockopt() in uipc_syscalls.c as well as
sosetopt() in uipc_socket.c affected many parts.
I don't prefer making users change mbuf size using sysctl after my
code is comitted.
Also, in fact, changing the size of mbuf is not a solution, since when
we encounter "shorter storage for setsockopt()" after we change it, we
again expand the size?

So, I vote "change the storage policy, not using mbuf".
Is it possible in *near* future? This is my question.
--
Hitoshi Asaeda