oplock_break problem (PR#20285)

[tridge at linuxcare.com]

> There is a also the possibility it is a problem in the underlying
> TCP timing between Solaris etc. and Windows clients that causes
> clients to not respond to the oplock break message, and that changes
> in Samba 2.0.x cause this timing problem to happen more than
> 1.9.18 does.

I know we have had that theory for a while but I'm not totally
convinced. In particular the sniff from Giulio revealed a bad bug in
our oplock resend handling. Unfortunately that bug doesn't explain the
main problem but it does show that despite lots of time spent on this
code we still have significant bugs in it.

What we need now is a more complete sniffer capture showing a complete
session from initial TCP connection to the oplock break timeout. My
pet theory at the moment is that the problem actually occurs a long
way back in the SMB session and the timeout is a symptom of a much
earlier problem. I want to analyse a session right the way through
looking for something unusual. I've asked Giulio if he can help.

Cheers, Tridge