process shared robust mutexes for tdb

[Rusty Russell]

Volker Lendecke <Volker.Lendecke at SerNet.DE> writes:
> On Tue, Mar 26, 2013 at 03:21:42PM +1030, Rusty Russell wrote:
>> I think we can do better: why not always hold the ACTIVE_LOCK (fcntl
>> lock) if TDB_MUTEX_LOCKING?  If they *don't* specify CLEAR_IF_FIRST, and
>> we get a write lock, then reset the mutexes (in theory we could resize
>> them at this point if we needed to, too).
>
> I don't think that's safe. You can't hold the ACTIVE_LOCK
> from all openers, we did that initially and killed the
> linked list flock implementation everywhere.

Ah, I forgot that we are still using fcntl locks for the per-record
locks.  Otherwise it's only slow down open, which is probably
acceptable.

I'll have to think harder on this one.  Or perhaps the answer is that
when I steal your code for ntdb, we can make it general (no per-record
locks there!) :)

Thanks,
Rusty.