Re: xfslogd-spinlock bug?

[David Chinner <dgc@sgi.com>]

On Tue, Dec 19, 2006 at 12:39:46AM +0100, Haar János wrote:
> From: "David Chinner" <dgc@sgi.com>
> > On Mon, Dec 18, 2006 at 09:17:50AM +0100, Haar János wrote:
> > > From: "David Chinner" <dgc@sgi.com>
> > > > Ok, I've never heard of a problem like this before and you are doing
> > > > something that very few ppl are doing (i.e. XFS on NBD). I'd start
> > > > Hence  I'd start by suspecting a bug in the NBD driver.
> > >
> > > Ok, if you have right, this also can be in context with the following
> issue:
> > >
> > > http://download.netcenter.hu/bughunt/20061217/messages.txt   (10KB)
> >
> > Which appears to be a crash in wake_up_process() when doing memory
> > reclaim (waking the xfsbufd).
> 
> Sorry, can you translate it to "poor mans language"? :-)
> This is a different bug?

Don't know - it's a different crash, but once again one that I've
never heard of occurring before.

> > Ok, I've found this pattern:
> >
> > #define POISON_FREE 0x6b
> >
> > Can you confirm that you are running with CONFIG_DEBUG_SLAB=y?
> 
> Yes, i build with this option enabled.
> Is this wrong?

No, but it does slow your machine down.

> > If so, we have a use after free occurring here and it would also
> > explain why no-one has reported it before.
> >
> > FWIW, can you turn on CONFIG_XFS_DEBUG=y and see if that triggers
> > a different bug check prior to the above dump?
> 
> [root@X64 linux-2.6.19]# make bzImage
> scripts/kconfig/conf -s arch/x86_64/Kconfig
> .config:7:warning: trying to assign nonexistent symbol XFS_DEBUG
> 
> I have missed something?

No - I forgot that config option doesn't exist in mainline XFS - it's
only in the dev tree.

FWIW, I've run XFSQA twice now on a scsi disk with slab debuggin turned
on and I haven't seen this problem. I'm not sure how to track down
the source of the problem without a test case, but as a quick test, can
you try the following patch?

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group


---
 fs/xfs/linux-2.6/xfs_buf.c |    4 ++++
 1 file changed, 4 insertions(+)

Index: 2.6.x-xfs-new/fs/xfs/linux-2.6/xfs_buf.c
===================================================================
--- 2.6.x-xfs-new.orig/fs/xfs/linux-2.6/xfs_buf.c	2006-12-19 12:22:54.000000000 +1100
+++ 2.6.x-xfs-new/fs/xfs/linux-2.6/xfs_buf.c	2006-12-19 13:48:36.937118569 +1100
@@ -942,11 +942,14 @@ xfs_buf_unlock(
 /*
  *	Pinning Buffer Storage in Memory
  *	Ensure that no attempt to force a buffer to disk will succeed.
+ *	Hold the buffer so we don't attempt to free it while it
+ *	is pinned.
  */
 void
 xfs_buf_pin(
 	xfs_buf_t		*bp)
 {
+	xfs_buf_hold(bp);
 	atomic_inc(&bp->b_pin_count);
 	XB_TRACE(bp, "pin", (long)bp->b_pin_count.counter);
 }
@@ -958,6 +961,7 @@ xfs_buf_unpin(
 	if (atomic_dec_and_test(&bp->b_pin_count))
 		wake_up_all(&bp->b_waiters);
 	XB_TRACE(bp, "unpin", (long)bp->b_pin_count.counter);
+	xfs_buf_rele(bp);
 }
 
 int
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/