close() hangs on file in NFS-mounted dir using 'ar'

[Erik Williamson <erik cpsc ucalgary ca>]

Hi All - 

This is a killer.  I've recently installed a few 7.3 boxes here
(completely up to date with patches), all with NFS-mounted home
directories (off of a Solaris 8 box).

If I run 'ar rc outlib.a somelib.o' where outlib.a is a file in an
NFS-mounted dir, the process hangs forever.  ('ps ax' shows the
process's state as 'D', a.k.a 'Uninterruptible sleep')

But Get this!  Magically enough, if I run 'top' in another terminal, the
process finishes normally - sometimes.  sometimes 'top' hangs too, but
only if 'ar' is hung first.  Want to know how I un-hang them?  I use
ssh2  to connect to the box, and when the password prompt comes up, ar
and top magically work like nothing has happened.

I've used strace, and found that it is hanging on the close() where ar
is closing the tempfile which later is moved to become outlib.a
mentioned earlier.

I can supply all files, strace outputs, or *anything* that anyone needs
to help me with this.  Other people here can't seem to figure it out
either, and we're going nuts!  I want to roll out 7.3 here, but not
until this is solved...

BTW, When I make the output file on a local filesystem, it works just
ducky.

Cheers, and thanks for any help!
Erik.

-- 
e r i k   w i l l i a m s o n                     erik cpsc ucalgary ca
 system admin . department of computer science . university of calgary