is_locked optimization

[tridge at samba.org]

Jeremy,

 > > hmm, very good point! The flush on oplock break happens with the
 > > oplock still held. The optimisation would need to handle that.
 > 
 > No it doesn't. We're depending on the client flushing the
 > held byte range locks to the server at break time. Windows
 > clients do this - I've seen it.

the flush happens with the oplock still held. The client logically
holds the oplock until it either acknowledges the oplock break from
the server, or it closes the file. I'm not sure if this harms your
planned optimisation, but it is at least worth thinking about.

 > I'm not asking you to add it to Samba4 :-). I can micro benchmark
 > it later.

I'll be very interested to see if it helps. Maybe you could extend the
loadfile in current versions of NBENCH to add more byte range locking?

Cheers, Tridge