Ultrix vs SunOS NFS client

[Barry Margolin]

At our site we have Symbolics Lisp Machines running an NFS server
implemented by International Lisp Associates.  We also have Unix
machines running MtXinu 4.3bsd, SunOS 3.5 and 3.2, and Ultrix 2.0 and
2.4.  Everything works pretty well, except when the Ultrix machines
try to list Lisp Machine directories; in that case, Ultrix always
reports that the directory is completely empty.

If occam:> is mounted on /lispm/occam on Ultrix, when I try to list
/lispm/occam/barmar (my directory) the following RPC transactions
occur: GETATTR on the LispM root, LOOKUP for "barmar", and GETATTR on
the file handle returned by the LOOKUP.  But it never goes into a
READDIR loop, the way SunOS and MtXinu do.  (ILA's NFS includes an
excellent tracing feature, that displays every RPC datagram received
or sent.)  The fattr returned by the last GETATTR looks like:

type=2, mode=40777, nlink=1, uid=1201 (my uid), gid=10 (my gid),
size=4544, blocksize=4544, rdev=0, blocks=1, fsid=1024,
fileid=1377896943, atime/mtime/ctime=<now> (directories always show up
as newly modified, because we once thought that Unix was trying to
cache things).

I tried making blocks=2 (since Unix directories that only contain "."
and ".." take up 2 blocks), nlink=2 (since Unix directories always
have at least two links, "." and "../name"), and rdev=1 (who knows
why), but none of these things helped.  Does anyone have any ideas why
Ultrix would decide that the directory is empty just from the above
attributes?


Barry Margolin
Thinking Machines Corp.

barmar at think.com
{uunet,harvard}!think!barmar