Subject: Re: phasing out mfs; make init(8) use tmpfs?
To: Izumi Tsutsui <tsutsui@ceres.dti.ne.jp>
From: Eric Haszlakiewicz <erh@swapsimple.com>
List: tech-kern
Date: 02/16/2007 13:56:58
On Sat, Feb 17, 2007 at 03:39:29AM +0900, Izumi Tsutsui wrote:
> erh@swapsimple.com wrote:
> 
> > So why is that any different than mfs not being in the kernel?
> > i.e. why do we need to worry about this error when using tmpfs but
> >  not when using mfs?
> 
> As I wrote, MFS could be smaller than TMPFS?
> (i.e. does it fit in i386 INSTALL floppies?)

mfs is definitely smaller:
With MFS:
   text    data     bss     dec     hex filename
5965061  120964  251504 6337529  60b3f9 netbsd

With TMPFS:
   text    data     bss     dec     hex filename
5997217  121060  251504 6369781  6131f5 netbsd

A difference of 2252 bytes.  I don't know if that's enough to cause a
problem with the install media.

> Anyway some ports (hpcsh for example) don't have even MFS and
> IIRC init(8) fails to create /dev silently in such case.
So tmpfs failing is not any worse.

> What's the problem on trying both?

It makes init bigger. (only a little)
It puts a stumbling block in the way of phasing out mfs entirely.

Keeping it as a compatibility feature is fine, but if we're really
getting rid of mfs we should define a timeframe for when stuff like
this goes away.  e.g. "init will stop supporting mount_mfs in NetBSD 5"

eric