NFS client bug, (Stale NFS file handle)

NFS client bug, (Stale NFS file handle)

[Brian Childs]

We beleive we're hitting an NFS client bug:

 Disto: SuSE-9.0 (64 bit)
  Arch: x86_64 SMP
Kernel: 2.4.21 (SuSE k_smp-2.4.21-201)

We're getting stale file handle errors for files in a directory two
levels below the mountpoint.

For example:

hosta$ cd /mountpoint/a/b
hosta$ ls -l
ls: A-spotpr: Stale NFS file handle
ls: B-spotpr: Stale NFS file handle
ls: C-spotpr: Stale NFS file handle
ls: D-spotpr: Stale NFS file handle
ls: E-spotpr: Stale NFS file handle
<snip>
total 0
hosta$ 

HOWEVER...

otherhost$ cd /mountpoint/a/b
otherhost$ ls -l
total 2352
-rw-rw-r--    1 db       data        82896 Mar 25 17:02 A-spotpr
-rw-rw-r--    1 db       data        57840 Mar 25 17:10 B-spotpr
-rw-rw-r--    1 db       data        82896 Mar 25 17:02 C-spotpr
-rw-rw-r--    1 db       data       213840 Mar 25 17:10 D-spotpr
-rw-rw-r--    1 db       data        82896 Mar 25 17:02 E-spotpr
<snip>
otherhost$

Where hosta & otherhost are the same type of machine running the same
OS, and they mount the directories at the same time.

I've never seen this before.  I'm used to seeing system-wide stale
handles for a mountpoint, but never for a specific file under a
mountpoint.

/mountpoint/a works without a problem, as does all files under
/mountpoint/a/c

I can't reliably reproduce this, but I know it happens after the
contents of /mountpoint/a/b are updated with an rsync.

Any help in diagnosing or fixing this will be greatly appreciated.

Brian