Re: openldap server + kse = bewm

On Fri, 20 Feb 2004, Daniel Eischen wrote:

> On Thu, 19 Feb 2004, Doug White wrote:
>
> > hey folks,
> >
> > Looks like the OpenLDAP server, slapd, and KSE don't get along too well.
> >
> > I can reliably segfault slapd by doing a few requests of it on a -CURRENT
> > machine built this morning PST.  TLS seems to accelerate things, but it
>
> What is TLS?

In this case, Transport Layer Security, not Thread Local Storage. :)

> > This is 100% reproducible, although initially it was croaking in
> > pthread_testcancel() instead of a kse function. This leads me to suspect
> > strange mutex corruption, but I'd like someone who understands kse to at
> > least spot-check.

I forgot to mention "croak" means "segfault".

> Usually, this is from something that is using %gs and stomping
> on our LDTs.  Any warnings from the kernel about static LDT
> allocations?

NNo such message that I saw.  I'm running X, but not the nvidia driver
(the machine has a cirrus logic chip onboard).

> > I thought at first it might be some strange interaction between berkeley
> > db 4.2's special assembly mutexes and kse, but I rebuilt db42 with pthread
> > mutexes and rebuilt openldap to use DB_PRIVATE so the db would mount, but
> > no change in status.
> >
> > Here's the trace from gdb:
>
> This doesn't show much to me...

As to the libc thing, I'm assuming its getting picked up as a dependency
on something else.  I'll check the build to make sure it isn't getting
requested explicitly.

Since the libc pthread_mutex_lock() immediately calls the libpthread one,
I assumed it was some sort of stub function.

-- 
Doug White                    |  FreeBSD: The Power to Serve
dwhite_at_gumbysoft.com          |  www.FreeBSD.org