On Fri, 1 Feb 2013, Lukáš Czerner wrote:
> Date: Fri, 1 Feb 2013 11:58:04 +0100 (CET)
> From: Lukáš Czerner <lczerner@xxxxxxxxxx>
> To: Dave Chinner <david@xxxxxxxxxxxxx>
> Cc: Lukáš Czerner <lczerner@xxxxxxxxxx>, xfs@xxxxxxxxxxx
> Subject: Re: Xfs fails in xfstests 013
>
> On Fri, 1 Feb 2013, Dave Chinner wrote:
>
> > Date: Fri, 1 Feb 2013 11:29:27 +1100
> > From: Dave Chinner <david@xxxxxxxxxxxxx>
> > To: Lukáš Czerner <lczerner@xxxxxxxxxx>
> > Cc: xfs@xxxxxxxxxxx
> > Subject: Re: Xfs fails in xfstests 013
> >
> > On Thu, Jan 31, 2013 at 09:19:23AM +0100, Lukáš Czerner wrote:
> > > Hi.
> > >
> > > I've just run xfstest 013 on xfs and it fails with this backtrace
> > >
> > >
> > > Jan 31 03:09:07 rhel6_vm1 kernel: ffff88020b6d1000: 78 78 78 78 78 78 78 78 78 78 2f 78 78 78 78 78 xxxxxxxxxx/xxxxx
> > > Jan 31 03:09:07 rhel6_vm1 kernel: XFS (sda): Internal error xfs_bmbt_verify at line 747 of file fs/xfs/xfs_bmap_btree.c. Caller 0xffffffffa025486e
> >
> > Same problem Cai Qian just reported - a remote symlink with a bmbt
> > verifier attached to it.
> >
> > > I can reproduce it every time with xfstest 013 on 3.8.0-rc5.
> >
> > And the configuration you are testing?
>
> The system is x86_64 KVM guest and I am using the following config
>
> export TEST_DEV=/dev/sda
> export TEST_DIR=/mnt/test
> export SCRATCH_DEV=/dev/sdb
> export SCRATCH_MNT=/mnt/test1
> export MKFS_OPTIONS="-f "
> export FSTYP=xfs
>
> Where /dev/sda and /dev/sdb are 50GB in size and those are exported
> from the host linear logical volumes.
>
> >
> > I'd suggest that you add an:
> >
> > ASSERT(bp->b_ops == NULL);
> >
> > into xfs_symlink() after the xfs_trans_get_buf() call in the remote
> > symlink crate loop, because the problem occurring implies that the
> > code is getting a new buffer with a stale ops structure on it
> > (though I can't see how that is possible right now).
> >
> > You should probably put the same assert into
> > xfs_inactive_symlink_rmt() between the get buf and the
> > xfs_trans_binval() call, and into xfs_readlink_bmap() after the
> > contents of a symlink are read from disk.
>
> Ok, I'll try that and let you know.
This is what I've done:
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index d95f565..22590ef 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -89,6 +89,7 @@ xfs_readlink_bmap(
xfs_buf_relse(bp);
goto out;
}
+ ASSERT(bp->b_ops == NULL);
if (pathlen < byte_cnt)
byte_cnt = pathlen;
pathlen -= byte_cnt;
@@ -313,6 +314,7 @@ xfs_inactive_symlink_rmt(
error = ENOMEM;
goto error1;
}
+ ASSERT(bp->b_ops == NULL);
xfs_trans_binval(tp, bp);
}
/*
@@ -1535,6 +1537,7 @@ xfs_symlink(
error = ENOMEM;
goto error2;
}
+ ASSERT(bp->b_ops == NULL);
if (pathlen < byte_cnt) {
byte_cnt = pathlen;
}
and I have not hit the assert. However I think that I know what's
going on. The verifier is obviously not attached when the symlink is
created, however it is attached when security attributes are created
for the symlink by xfs_init_security().
Thanks!
-Lukas_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs
