2.4.20: ext3: Assertion failure in journal_forget() - filesystem destroyed

This started to happen during larger (10MB-420MB) rsync based writes to a striped ext3 partition (/dev/md11) residing on 4 scsi disks which is mounted with defaults, i.e. data=ordered (rsync over 100Mbps link):

Dec  1 12:25:43 pollux kernel: EXT3-fs error (device md(9,11)):
ext3_new_block:
Allocating block in system zone - block = 114696
Dec  1 12:25:43 pollux kernel: EXT3-fs error (device md(9,11)):
ext3_new_block:
Allocating block in system zone - block = 114697
Dec  1 12:25:43 pollux kernel: EXT3-fs error (device md(9,11)):
ext3_new_block:
Allocating block in system zone - block = 114700
Dec  1 12:25:43 pollux kernel: EXT3-fs error (device md(9,11)):
ext3_new_block:
Allocating block in system zone - block = 114701
Dec  1 12:25:43 pollux kernel: EXT3-fs error (device md(9,11)):
ext3_new_block:
Allocating block in system zone - block = 114702
Dec  1 12:25:43 pollux kernel: EXT3-fs error (device md(9,11)):

ext3_new_block:
Allocating block in system zone - block = 114706

<snip>

Dec  1 22:17:55 pollux kernel: EXT3-fs error (device md(9,11)):
ext3_free_blocks
: Freeing blocks in system zones - Block = 573501, count = 2
Dec  1 22:17:55 pollux kernel: EXT3-fs error (device md(9,11)):
ext3_free_blocks
: Freeing blocks in system zones - Block = 573552, count = 14
Dec  1 22:17:55 pollux kernel: Assertion failure in journal_forget() at

transaction.c:1225: "!jh->b_committed_data"

Trying to access the partition resulted in processes hanging in D state:

5336 pts/0 D 0:00 ls -a -N --color=tty -T 0 -l /mnt/data8

e2fstools version is 1.32 and the partition was created with this version using 'mke2fs -j -b 2048 -i 4096 -R stride=16 /dev/md11'.

An earlier dump of the partition data using tune2fs -l gave:

tune2fs 1.32 (09-Nov-2002)

Filesystem volume name:   <none>
Last mounted on:          <not available>
Filesystem UUID:          7c8d7827-4b25-40ab-a3b8-1c4c6e286868
Filesystem magic number:  0xEF53
Filesystem revision #:    1 (dynamic)
Filesystem features:      has_journal filetype needs_recovery sparse_super
Default mount options:    (none)
Filesystem state:         clean with errors
Errors behavior:          Continue
Filesystem OS type:       Linux
Inode count:              2621440
Block count:              5236992
Reserved block count:     261849
Free blocks:              4855697
Free inodes:              2621416
First block:              0
Block size:               2048
Fragment size:            2048
Blocks per group:         16384
Fragments per group:      16384
Inodes per group:         8192
Inode blocks per group:   512
Last mount time:          Sat Nov 30 11:23:59 2002
Last write time:          Sun Dec  1 14:09:55 2002
Mount count:              2
Maximum mount count:      -1
Last checked:             Fri Dec  1 19:18:16 2000
Check interval:           0 (<none>)
Reserved blocks uid:      0 (user root)
Reserved blocks gid:      0 (group root)
First inode:              11
Inode size:               128
Journal UUID:             <none>
Journal inode:            8
Journal device:           0x0000
First orphan inode:       0

Trying 'e2fsck -y /dev/md11' after a reboot showed so many errors and continued to run for minutes that I aborted e2fsck and do assume that the file system was completely destroyed.

After recreation of the filesystem on /dev/md11 a rsync run completed without errors.

As a side note: the system having the rsync sources has an identical formatted partition (the systems are hardware twins) and doesn't show any errors.

Some final information about the raid configuration of /dev/md11:

raiddev /dev/md11

         raid-level              0
         nr-raid-disks           4
         nr-spare-disks          0
         chunk-size              32
         persistent-superblock   1
         device                  /dev/sda13
         raid-disk               0
         device                  /dev/sdb13
         raid-disk               1
         device                  /dev/sdc15
         raid-disk               2
         device                  /dev/sdd15
         raid-disk               3

Andreas Steinmetz
D.O.M. Datenverarbeitung GmbH