Subject: Re: phasing out mfs; make init(8) use tmpfs?
To: Martin S. Weber <Ephaeton@gmx.net>
From: Julio M. Merino Vidal <jmmv@NetBSD.org>
Date: 02/17/2007 09:12:56
On Fri, Feb 16, 2007 at 10:59:30PM +0100, Martin S. Weber wrote:
> On Fri, Feb 16, 2007 at 02:12:19AM +0100, Juan RP wrote:
> > (...)
> > I can report that using null mounts is not a problem with tmpfs,
> > I've been using it for months with mklivecd... dunno about union
> > mounts.
> I've opened a PR for union mounts eons ago. It's still happily rotting open.
> Given it seems so hard to fix and the developer doesn't have time for
> it we probably should fall back to MFS and remove every trace of TMPFS
> from the CVSROOT...
So, just because of this, we should throw away all the work that went into
tmpfs and "return" to mfs, which is slower, is just a FFS hack to provide
a memory filesystem, and whose code is harder to understand?
I haven't looked at that issue yet, but I did address a lot of open PRs
not so long ago (at the end of past year). I wish I had more time to
effectively resolve the two remaining PRs (31944 and 35112) but I currently
As regards whiteouts, they don't seem too difficult to add. I'm not sure
yet but maybe all the code needs is a new type of tmpfs_node and a new
tmpfs_whiteout vnode operation. (Back when I started writing tmpfs, I
thought whiteouts were a FFS-specific thing and therefore simply ignored
them... Had I known they weren't, they'd probably work already ;-)
Julio M. Merino Vidal <jmmv@NetBSD.org>