On Thu, 11 Feb 1999, Bernd Salbrechter wrote:

> Dr. Bill Studenmund wrote:
> 
> > I've tested compiling on an i386, and have activly used ffs & nfs
> > filesystems. Can others please try all our other filesystems?
> 
> I using the union-fs on my private NetBSD-boxes (a i486 tower and a i586
> Laptop), but not very havily.

Just to make sure, you've tried this with the patches I sent out earlier
this week? _They_ are what I need tested. :-)

> The only trouble i steped in, was mounting them using manual mount and
> parameters from the fstab. In the case of thre stacked filesystems the
> mountpoint of the second gets the device of the third and if i gave
> this name it badly mounts he seconde two times. That looks more like a
> problem in mount and not the fs. A defined prefered type of the given
> name to mount, would be great for stacked file systems.

True.

> For this application the way the union-fs handles removes, are not realy
> what I like. In this case I will prefere to fall back to the original
> version, if I remove a modified version, but the file disaperes. 

That was actually a feature of unionfs. It's creating whiteouts in the
upper layer. They make it so that you have the full semantics of a
writeable fs even though your underlying one might be r/o.

If you do an ls -W, you will see the whiteout in the upper layer. An ls
-Wl will show the whiteouts with a "w" in the left column (which is
normally '-', and 'd' for directories, etc). 

rm -W will delete a whiteout. So rm <name>; rm -W <name> will give you the
unmodified file <name>.

Take care,

Bill