On Wed Dec 12 2007 at 10:44:40 +0200, Alan Barrett wrote:
> On Tue, 11 Dec 2007, Bill Stouder-Studenmund wrote:
> > Note that -o union != mount -t unionfs. Also, my understanding was
> > that unionfs created the whiteouts, so that even if tmpfs supported
> > them, you wouldn't see a difference in how /dev is handled.
> 
> I know about the difference between mount -o union and mount -t unionfs,
> but I thought that both were supposed to support whiteouts.  I see now
> that in fact "mount_mfs -o union" and "mount_tmpfs -o unoin" both have
> the same behaviour here: deleting a file that exists in the lower layer
> really deletes it, it doesn't make a whiteout in the upper layer.

In -o union the file systems are used directly, not through a virtual
layered file system (unionfs).  It is implemented as a fork in lookup.
Therefore you can't access the lower layer at all if the name happens
to be in the upper layer (or maybe we should use "layer" instead).
You can use it only to unify namespaces, not e.g. make a cd writable.

-- 
Antti Kantee <pooka@iki.fi>                     Of course he runs NetBSD
http://www.iki.fi/pooka/                          http://www.NetBSD.org/
    "la qualité la plus indispensable du cuisinier est l'exactitude"