Subject: Re: Removing tmpfs' experimental status
To: Elad Efrat <>
From: Julio M. Merino Vidal <>
List: tech-kern
Date: 10/22/2006 18:28:41
On 10/22/06, Elad Efrat <> wrote:
> Julio M. Merino Vidal wrote:
> > On 10/22/06, Elad Efrat <> wrote:
> >> Julio M. Merino Vidal wrote:
> >>
> >> > - Enable TMPFS by default on all GENERIC kernels.
> >>
> >> Are you going to disable MFS by default too?
> >
> > No.
> Why not? I thought we don't want bloat in our kernel?

But we don't want to switch to TMPFS by default yet, hence MFS must
stay.  It is used by, e.g., sysinst if you choose so.

> >> > - Disable the tmpfs_vfstofh and tmpfs_fhtovfs vfsops so that tmpfs
> >> >  does not support exporting.  (I'm making them optional through a
> >> >  TMPFS_ENABLE_NFS macro so that people interested in fixing it can
> >> >  quickly enable the missing bits.  Of course the macro will go away
> >> >  when things work as expected.)
> >>
> >> Please don't do that.
> >
> > Why not?  It's just the same as an "#ifdef notyet" but with a more
> > meaningful name.  That macro will NOT go into any kernel
> > configuration, and it shouldn't be set there.  People interested in
> > fixing this stuff should '#define TMPFS_ENABLE_NFS' explicitly in the
> > .c file.
> Please check what happens with fileassoc(9) on tmpfs file-systems with
> this disabled before adding these options.

Oh, you were referring to disabling those specific vfsops... well,
it's the only way to disable NFS support for a given filesystem --
make one of those two operations null.  So we should probably have a
different way to specify this, such as a flag in the mount structure
or something.

Anyway, what happens for all those fs that don't implement these
operations?  There are several that currently do this.


Julio M. Merino Vidal <>
The Julipedia -