| > Jan 6 22:18:42 jaguar kernel: EXT3-fs error (device ide0(3,3)):
| > ext3_readdir: bad entry in directory #147553: rec_len %% 4 != 0 -
| > offset=0, inode=1651076143, rec_len=19527, name_len=85
|
| Hmm, this looks like data being written into a directory block:
|
| 1651076143 = 0x62696c2f = "/lib"
| 19527 = 0x4c47 = GL
| 85 = 0x55 = U
We see this regularly with our 2.2.19-4.1mdk (Mandrake 7.2) kernel,
ext2, with the 3ware Escalade 4-port IDE RAID cards (SCSI emulation)
and four IBM DeskStar disks.
Two different machines - identical configuration - same problem.
Directories get overwritten and if you translate the huge inode number,
you get ASCII text.
The main server serves a bunch of Windows machines via Samba. Large files
(300MB to 800MB) get tossed around with some regularity. The second
server receives a backup from the first server every 20 minutes via
rsync from the first server.
We saw the same corruption when using the Mandrake 2.2.17 kernel.
We upgraded the second server to 60GB IBM DeskStar disks, copied over
the entire file system from the four 30GB disks (via dump/restore) and
a few days later the new disks had corruption. Maybe it was "latent"
corruption from the previous 30GB disks; or, maybe it isn't the disk
drives that are the problem here. I'm suspecting something in the SCSI
disk drivers, since a Google search for this type of corruption always
fingers SCSI disks.
In any case, the corruption isn't an ext3 issue since we don't run ext3
(yet). As soon as the 2.4.x kernels settle down for a few months,
we'll move there and run ext3.
--
-IAN! Ian! D. Allen Ottawa, Ontario, Canada idallen@ncf.ca
Home Page on the Ottawa FreeNet: http://www.ncf.ca/~aa610/
College professor at: http://www.algonquincollege.com/~alleni/
Board Member, TeleCommunities CANADA http://www.tc.ca/
