[Date Prev][Date Next]
Re: SSS/VLV implementation
- To: openldap-devel@OpenLDAP.org
- Subject: Re: SSS/VLV implementation
- From: Bertrand Croq <email@example.com>
- Date: Wed, 1 Dec 2004 10:33:26 +0100
- Content-disposition: inline
- In-reply-to: <41ACA77D.firstname.lastname@example.org>
- References: <20041130112356.GA8499@raoul.bruz.freeskop.net> <41ACA77D.email@example.com>
- User-agent: Mutt/1.5.6+20040907i
On Tue, Nov 30, 2004 at 09:01:49AM -0800, Howard Chu wrote:
> You might want to read this thread from last week:
> You really should be adding this feature via an overlay and not touching
> the back-bdb code at all.
> As for the reader locks, there is a BDB locker ID associated with each
> thread. There is also a lock structure associated with each operation.
> Very likely you're seeing "lock is no longer valid" because the lock
> structure has been overwritten by an access to a subsequent entry. This
> is a bit tricky and unfortunately not handled very cleanly right now.
> Ideally, the details of back-bdb's internal locks should be invisible to
> the rest of the server but that's not how things work presently. You
> should take a look at how backend.c: backend_group() handles the
> op->o_private structure for an idea.
> The normal -devel advice also applies - make sure you're working with
> CVS HEAD when writing new features. You should also look at
> overlays/syncprov.c or overlays/ppolicy.c (in CVS HEAD) for examples of
> how to implement new controls as part of an overlay.
Ok, I'll try reimplementing it using an overlay, based on CVS HEAD
instead of Debian latest source package... I haven't writtent a lot (I
have mostly spent time trying to understand Outlook and OpenLDAP's code
architecture) so it shouldn't be too long.
Bertrand Croq - FreesKop - Avenue Robert Schuman - 35170 BRUZ
http://www.freeskop.com/ - Tel: 02 99 05 04 56 - Fax: 02 99 05 96 40