Re: panic: System call lstat returning with 1 locks held

On 1/24/08, Attilio Rao <attilio_at_freebsd.org> wrote:
> 2008/1/24, Yar Tikhiy <yar_at_freebsd.org>:
> > On Tue, Jan 15, 2008 at 04:39:24PM +0200, Kostik Belousov wrote:
> > >
> > > I think this could be related to the recent vn_lock()/VOP_LOCK() KPI changes.
> > > Please, add DEBUG_VFS_LOCKS to the kernel config, and do the
> > >       show lockedvnods
> > > from the ddb prompt when the panic occurs. The witness does not track
> > > the lockmgr locks.
> >
> > I think I'm seeing the same panic on UFS.  It's rather nasty: I
> > cannot rebuild CURRENT natively due to it so I have to build it
> > under 6-STABLE.  My favourite way to trigger the panic reliably is
> > running `make install' in a simple port directory, e.g., portmaster,
> > but my system also panics during daily scripts run and, as already
> > said, if trying to build world.
>
> Yar,
> as it seems reproducible for you, can you please add this patch to the tree:
> http://www.freebsd.org/~attilio/debug_tdlocks.diff
>
> compile your kernel with:
> options KTR
> options KTR_COMPILE=(KTR_SPARE2)
> options KTR_MASK=(KTR_SPARE2)
> options KTR_ENTRIES=32768
>
> and once kernel panics, at ddb prompts do:
> > show ktr
>

I added the above options to my kernel, and performed a scripted textdump.

/sbin/ddb script lockinfo="show locks; show alllocks; show lockedvnods"
/sbin/ddb script kdb.enter.panic="textdump set; capture on; show ktr ;
run lockinfo ; show pcpu; bt; ps; alltrace; capture off; call doadump;
reset"

After the kernel paniced, the kdb.enter.panic script ran and created a
textdump.  When I extracted the 2.7M ddb.txt file, it didn't show any
calls to lockmgr_disown in the ktr trace.

Let me know if there is anything else that I can do.

To get this dump, DB_CAPTURE_MAXBUFSIZE (sys/ddb/db_capture.c) needed
to be increased from its default of 512K to 5M, and then setting the
debug.ddb.capure.bufsize to 5M after rebooting with the new kernel.

See PR 119993 (http://www.freebsd.org/cgi/query-pr.cgi?pr=119993)
which adds two new kernel options to allow the capture buffer size to
be changed at compile time.

Scot