Bug#295657: EXT3 on RAID problems in all 2.6-smp
On Thu, 17 Feb 2005, Jeroen van Disseldorp wrote:
> 2 months ago I bought 4 new hard drives, which I configured as follows:
>
> -----------------------------------------------------------------
> | Linux | Swap | RAIDSET-1 disk 1 |
> -----------------------------------------------------------------
> | RAIDSET-0 disk 1 | RAIDSET-1 disk 2 |
> -----------------------------------------------------------------
> | RAIDSET-0 disk 2 | RAIDSET-1 disk 3 |
> -----------------------------------------------------------------
> | RAIDSET-0 disk 3 | RAIDSET-1 disk 4 |
> -----------------------------------------------------------------
>
> RAIDSET-0 is 20G, RAIDSET-1 is 709G.
>
> On top of both raid devices I use the device mapper to encrypt those
> disks, so data travels to the disk through the following layers:
>
> vfs -> ext3 -> device mapper -> raid -> physical disks
>
> Since the beginning I have been having problems with heavy disk usage
> on RAIDSET-1. It seems that under normal circumstances the system is
> stable, but under heavy load the following error occurs:
>
> Feb 12 06:27:25 einstein kernel: EXT3-fs error (device dm-2): ext3_readdir: bad entry in directory #84509561: rec_len %% 4 != 0 - offset=0, inode=1629516644, rec_len=29795, name_len=117
> Feb 12 06:27:25 einstein kernel: Aborting journal on device dm-2.
> Feb 12 06:27:25 einstein kernel: ext3_abort called.
> Feb 12 06:27:25 einstein kernel: EXT3-fs abort (device dm-2): ext3_journal_start: Detected aborted journal
> Feb 12 06:27:25 einstein kernel: Remounting filesystem read-only
>
> As I do some spam-learning and backups at night, requiring some heavy
> disk use, this happens just about every day. Running fsck after the
> error is useless, as it finds no errors at all. So there is little
> else to do then remount and hope for the best for another day.
>
> My system is a dual Pentium II 400MHz with 384MB memory. All four
> disks are Hitachi/IBM Deskstar 250G. The problem occurs with every
> 2.6 kernel I have tried so far, starting from 2.6.7.
>
> Typical thing is that it only seems to occur on -smp kernels. I've
> been running some UP-kernels as well and (although slower) they do
> not thrash the ext3 journal. Could this be an SMP race?
>
> More system information can be found in the attached kern.log.
>
can you try to reproduce it with more recent 2.6.11 from unstable?
--
maks
Reply to: