Subject: Re: Removing tmpfs' experimental status
To: Izumi Tsutsui <tsutsui@ceres.dti.ne.jp>
From: Julio M. Merino Vidal <jmmv84@gmail.com>
List: tech-kern
Date: 10/22/2006 21:15:17
On 10/22/06, Izumi Tsutsui <tsutsui@ceres.dti.ne.jp> wrote:
> jmmv84@gmail.com wrote:
>
> > I think it was martin@ who said that this would be
> > fine but that we first should disable obviously broken functionality
> > (read NFS support).  That seems wise advice so...
>
> vnconfig(8) against files on tmpfs fails silently:
> http://mail-index.netbsd.org/tech-kern/2006/08/03/0000.html
>
> Maybe it's better to return some error as well.

I've been looking at this and the problem comes from the lack of
VOP_STRATEGY and VOP_BMAP support in tmpfs, both of which are required
for vnd operation.

If these two operations could be set to NULL pointers instead of
genfs_eopnotsupp, vnd could be modified to detect that they are not
present and abort early.  But doing so will cause problems elsewhere
because these two operations are executed without previously checking
whether they are good or not.

So vnd silently allows setting up a new device for a file in tmpfs
but, later on, any operations issued on it fail because of the lack of
these two operations.

Any idea to workaround this other than implementing the two operations?

Thanks.

PS: Because... comparing the pointers to genfs_eopnotsupp could be UGLY, right?

-- 
Julio M. Merino Vidal <jmmv84@gmail.com>
The Julipedia - http://julipedia.blogspot.com/