I applied your patch against 2.1.63, and it does not affect the problem
with reads failing after file re-open. I went so far as to set the delay
to 1 * HZ - no luck. To make matters worse, I started seeing problems
with file _creation_ (as reported by others) when mounting from a third
machine.
Please correct me, but I don't understand how you have concluded that the
problem lies on the server end. We've already seen that the first read
call never makes it to the smbfs filesystem driver, so how is a
Windows-based latency to blame? I would think that it is the Linux kernel
which has not "caught up" with the fact that the file is completely
written and closed. Can you briefly outline the chain of events?
Steve