updatedb triggers XFS internal error

Hello,
I'm having a problem with one of my linux servers. whenever updatedb is run, the following errors occur in the system log.

0x0: c9 00 5a f1 3a 7f 66 be a3 c1 d4 7f e8 1d 6b c9
Filesystem "md0": XFS internal error xfs_da_do_buf(2) at line 2271 of file fs/xfs/xfs_da_btree.c. Caller 0xffffffff8817957f

Call Trace:<ffffffff881793ef>{:xfs:xfs_da_do_buf+1513} <ffffffff8817957f>{:xfs:xfs_da_read_buf+36}
      <ffffffff8018bfca>{do_lookup+83} 
<ffffffff8817d46f>{:xfs:xfs_dir2_put_dirent64_direct+0}
      <ffffffff8817957f>{:xfs:xfs_da_read_buf+36} 
<ffffffff8817e64d>{:xfs:xfs_dir2_block_getdents+183}
      <ffffffff8817e64d>{:xfs:xfs_dir2_block_getdents+183}
      <ffffffff8817d46f>{:xfs:xfs_dir2_put_dirent64_direct+0}
      <ffffffff8018d173>{link_path_walk+196} 
<ffffffff8816c8db>{:xfs:xfs_bmap_last_offset+226}
      <ffffffff8817d46f>{:xfs:xfs_dir2_put_dirent64_direct+0}
      <ffffffff8817d0a7>{:xfs:xfs_dir2_getdents+222} 
<ffffffff881a740b>{:xfs:xfs_readdir+84}
      <ffffffff881b104d>{:xfs:linvfs_readdir+213} 
<ffffffff80190c9a>{filldir64+0}
      <ffffffff80190c9a>{filldir64+0} <ffffffff80190b4e>{vfs_readdir+154}
      <ffffffff80190dc8>{sys_getdents64+116} 

<ffffffff8010e89e>{tracesys+209}

0x0: 28 ab a5 ec 3e 42 55 1f 76 9e 01 72 72 ee bd f1
Filesystem "md0": XFS internal error xfs_da_do_buf(2) at line 2271 of file fs/xfs/xfs_da_btree.c. Caller 0xffffffff8817957f

Call Trace:<ffffffff881793ef>{:xfs:xfs_da_do_buf+1513} <ffffffff8817957f>{:xfs:xfs_da_read_buf+36}
      <ffffffff8015dbae>{find_or_create_page+30} 
<ffffffff8817957f>{:xfs:xfs_da_read_buf+36}
      <ffffffff88181686>{:xfs:xfs_dir2_leaf_getdents+1107}
      <ffffffff88181686>{:xfs:xfs_dir2_leaf_getdents+1107}
      <ffffffff8817d46f>{:xfs:xfs_dir2_put_dirent64_direct+0}
      <ffffffff8816c8db>{:xfs:xfs_bmap_last_offset+226} 
<ffffffff8817d46f>{:xfs:xfs_dir2_put_dirent64_direct+0}
      <ffffffff8817d0c3>{:xfs:xfs_dir2_getdents+250} 
<ffffffff881a740b>{:xfs:xfs_readdir+84}
      <ffffffff881b104d>{:xfs:linvfs_readdir+213} 
<ffffffff80190c9a>{filldir64+0}
      <ffffffff80190c9a>{filldir64+0} <ffffffff80190b4e>{vfs_readdir+154}
      <ffffffff80190dc8>{sys_getdents64+116} 

<ffffffff8010e89e>{tracesys+209}

Its a dual opteron system running Fedora Core 4 and running the fedora packaged 2.6.12-1.1456_FC4smp kernel. The filesystem in question is on a md stripe raid running across an Infortrend 1.4TB hardware SCSI raid.

The output from xfs_info is:

meta-data=/user_data isize=256 agcount=32, agsize=11180624 blks
= sectsz=512
data = bsize=4096 blocks=357779520, imaxpct=25
= sunit=16 swidth=32 blks, unwritten=1

naming   =version 2              bsize=4096
log      =internal               bsize=4096   blocks=32768, version=1

= sectsz=512 sunit=0 blks
realtime =none extsz=131072 blocks=0, rtextents=0

Any help will be gratefully received!

Cheers, Evan.

[email protected]
Linux Systems Administrator
Double Negative
tel: +44 (0)20 7534 4400 fax: +44 (0)20 7534 4452
77 shaftesbury avenue, w1d 5du, London

This indicates bad metadata magic read from disk. You'll probably want to run xfs_repair; you can run it with -n to see what it *would* do first, to get an idea of how drastic the repair might be. Repairing 1.4T won't probably be lots of fun, but you've got corruption in there somewhere...

-Eric

Thanks for your help.

Cheers, Evan.

[email protected]
Linux Systems Administrator
Double Negative
tel: +44 (0)20 7534 4400 fax: +44 (0)20 7534 4452
77 shaftesbury avenue, w1d 5du, London