Re: [Jfs-discussion] Re: Question about file system failure

[Chris Penney <cpenney@gmail.com>]

On 6/27/05, Dave Kleikamp <shaggy@austin.ibm.com> wrote:
> On Mon, 2005-06-27 at 09:41 -0500, Dave Kleikamp wrote:
> > On Mon, 2005-06-27 at 10:10 -0400, Chris Penney wrote:
> > > I had an NFS file server using JFS fail this weekend.  A reboot, which
> > > made fsck do a full check, seems to have cleared everything up.  The
> > > initial errors I got were:
> > >
> > > Jun 25 09:27:04 nicfs2 kernel: Incorrect number of segments after building list
> > > Jun 25 09:27:04 nicfs2 kernel: counted 16, received 15
> > > Jun 25 09:27:04 nicfs2 kernel: req nr_sec 320, cur_nr_sec 8
> >
> > These are coming from scsi_init_io() in drivers/scsi/scsi_lib.c.  I
> > don't know what it means, but I'm inclined to think that it indicates a
> > software bug rather than a hardware error.
> >
> > > Jun 25 09:27:04 nicfs2 kernel: device-mapper: dm-multipath: Failing path 8:96.
> > > Jun 25 09:27:04 nicfs2 kernel: cfq: depth 4 reached, tagging now on
> > > Jun 25 09:27:04 nicfs2 kernel: end_request: I/O error, dev sdc, sector
> > > 1592060824
> > > Jun 25 09:27:04 nicfs2 kernel: device-mapper: dm-multipath: Failing path 8:32.
> > > Jun 25 09:27:04 nicfs2 kernel: end_request: I/O error, dev sdc, sector
> > > 1592062936
> >
> > I'm not sure if dm-multipath may be responsible.
> >
> > > Following that was a flurry of JFS errors.  I assume these messages
> > > have nothing at all to do with JFS, but I wanted to make certain.
> >
> > I don't think that JFS is the cause.
> >
> > > I can't turn up much googling that error.  If anyone has any idea what
> > > caused that I'd love to hear it.
> >
> > I'm copying this to linux-kernel in the hopes that someone there will be
> > able to help.  It would be useful to know what kernel you are running.
> 
> Well, I meant to cc linux-kernel.  :-)
> 

I'm running:
  * SMP Pentium 4 w/ hyperthreading enabled (IBM x345)
  * Dual QLogic 2340 HBAs
  * STK D280 Disk Array (nothing in logs)
  * Kernel 2.6.11.5 (built around March 24th)
  * 2.6.11-rc3-udm2 patch
  * linux-2.6.11-NFS_ALL patch (this was critical for me)

I'm using DM to make four multipath devices (each device is a 1TB lun)
and then one stripe made from the four multipaths.  Because the array
is unsupported by multipathd (at least in the code I have) it can only
failover once.  We then manually fail it back to primary if that
happens (which hasn't happened outside of testing).


> > > One last question, for an NFS server is it better to mount the volume
> > > with errors=panic?  It seems like that would keep I/Os from failing
> > > due to it being a read-only file system on error.  In this case it
> > > would seem like a panic + boot would have let a lot of processes (this
> > > is used in a batch environment) resume.
> >
> > Seems reasonable, but I'll let others comment.
> >
> > >    Chris
> >
> > Thanks,
> > Shaggy
> --
> David Kleikamp
> IBM Linux Technology Center
> 
> 

Thanks,

   Chris
-
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/