Re: panic: ffs_blkfree: freeing free block

On Wednesday 17 October 2007, Eric Anderson wrote:
> Kostik Belousov wrote:
> > On Tue, Oct 16, 2007 at 12:20:59PM +0200, Arjan van Leeuwen wrote:
> >> 2007/10/16, Eric Anderson <anderson_at_freebsd.org>:
> >>> Pieter de Goeje wrote:
> >>>> On Sunday 14 October 2007, Eric Anderson wrote:
> >>>>> Pieter de Goeje wrote:
> >>>>>> On Saturday 13 October 2007, Rolf Witt wrote:
> >>>>>>> Backtrace from crash tonight (after portsnap cron):
> >>>>>>
> >>>>>> <snip>
> >>>>>>
> >>>>>>> # uname -a
> >>>>>>> FreeBSD peanuts.homenet.local 7.0-PRERELEASE FreeBSD 7.0-PRERELEASE
> >>>>>>> #121: Thu Oct 11 16:29:05 CEST 2007
> >>>>>>> rowi_at_peanuts.homenet.local:/usr/obj/usr/src/sys/PEANUTS  i386
> >>>>>>>
> >>>>>>> Sources from Builddate.
> >>>>>>>
> >>>>>>> Rowi
> >>>>>>
> >>>>>> Eric sent me this patch:
> >>>>>> http://www.googlebit.com/freebsd/patches/ffs_softdep.c-patch
> >>>>>> which seems to be working great so far. I am still testing it, but I
> >>>>>> think it fixed the problem.
> >>>>>> To apply, cd /usr/src; patch < /path/to/patch and then rebuild the
> >>>>>> kernel.
> >>>>>>
> >>>>>> Cheers,
> >>>>>> Pieter de Goeje
> >>>>>
> >>>>> It doesn't actually 'fix' the problem, but I think it helps identify
> >>>
> >>> it.
> >>>
> >>>>> I'm not 100% certain if this is the right fix our not, but so far
> >>>>> feedback has been good when running with this patch.
> >>>>>
> >>>>> Can somebody confirm that this patch is ok?
> >>>>>
> >>>>> Eric
> >>>>
> >>>> Can you elaborate on what this patch exactly does / what the problem
> >>>> is?
> >>>>
> >>>> Pieter
> >>>
> >>> I think the problem is that blocks are being put in the worklist twice,
> >>> but I'm not certain why yet.  The patch reduces the chance of this
> >>> happening by more often removing the block from the worklist instead of
> >>> leaving it on.  I think actually the patch is hiding the real issue
> >>> more than anything, which is why I said it isn't fixing the problem,
> >>> but because of the reports I've seen, I think we're sniffing in the
> >>> right area.  I'll be looking more into this soon, when I get another
> >>> few minutes of free time and my debugger..
> >>
> >> I actually seem to be getting the panic more often with the patch (i.e.
> >> twice in just 30 minutes with the patch, about twice per day without the
> >> patch), while doing the same things as usual. The panic usually happens
> >> when I'm compiling or linking. Is that strange?
> >
> > Patch is wrong. Actually, it does put the dirrem to the proccessing twice
> > when xp != NULL.
>
> Yes, exactly why I said 'it doesn't fix the problem'.  In fact, I
> actually did not post it to the list, and didn't intend for it to be
> posted.  I was hoping for feedback for further examination.

I had (and still have) trouble reproducing the panic with the patch. I tried 
to write a program which would trigger the panic, but I failed (hence my 
question on what problem the patch masks so my test tool could focus on the 
problem area more). I ended up trying to write something similar to portsnap 
followed by portsdb -u. Now that portsnap works again for -CURRENT, I am able 
to reproduce the panic much more reliably.

I posted the patch to current_at_ because of my feedback being useless for the 
aforementioned reasons and I hoped someone else had a better testcase.

Regards,
Pieter de Goeje
>
> Kostik - do you have any ideas on what is going on here?  Did you see
> the message in another thread that looked similar (it was regarding
> gjournal)?
>
> Eric
>