[Unionfs] Basic testing with NFS exports

[Jesse I Pollard]

Most of this is already known, but I went ahead and did a bit of regression
testing between various linux kernels and unionfs kits.

I have found:

linux           unionfs         notes
==============  =============   
================================================
2.6.23-rc3-mm1  (included)      works locally, does not support NFS export
2.6.20.16       2.1.2+ODF       works locally, does not support NFS export
2.6.18-8.1.8    NA              vendor base kernel starting point.
2.6.18.8        1.4             NFS permission denied...
2.6.17.14       1.3             NFS permission denied...
2.6.16.53       1.2             can NFS mount
                                used fsid=10,crossmnt,nohide,no_root_squash
                                Input/output errors on large directories
                                causes NULL pointer errors in NFS 
server, and
                                        NFS hangs in clients
                                was not using nfsro union mount
                                touch file causes copy up on first try.
2.6.15.7        1.1.5           can NFS mount
                                used fsid=10,nohide,no_root_squash
                                causes NULL pointer errors in NFS 
server, and
                                        NFS hangs in clients
                                "touch file" causes permission denied the
                                        first time, works second (file has
                                        migrated)
                                using nfsro union mount

The NULL pointer errors may have been patched in more current kernels 
 >2.6.22
Unfortunately, the patch has too many other patches intervening to easily
backport (at least by me).

I checked 2.6.20.16/2.1.2+ODF just out of curiosity. The ODF looks very 
useful
for NFS persistence, but none of the unionfs 2.x versions support NFS.

unionfs 1.3/1.4 might be NFS mountable, but may require additional 
options. I
didn't see any that immediately seemed relevant though. I will test any if
someone has suggestions.

As long as I have a test VM system I will be willing to perform some 
testing.

Currently, I have access to a 12 TB NetApp fileserver for use as a read-only
data source. We are PLANNING on having a 16-20 TB server access via an SGI
CXFS client on a Red Hat release (an old one - RHEL 3, acting as a NFS
server since the CXFS client isn't supported on anything newer) as a 
read/write
data sink shortly.

If things appear to be working between my workstation (as a NFS client), 
then
testing some Solaris NFS clients is also possible.

I'm also open to suggestions for testing procedures too.

The test environment should last at least 30 days, and may last until 
the end
of the year.