Ext3 users - Feb 2001 - Crash-report; 2.2.19pre14-ext3-0.0.6b

Our main NFS-server, running Debian Potato, died this morning whilst
under quite heavy load - due to a runaway perl-script forking of some
200 instances of "/bin/cp" (the load was ~ 50). The server is running
2.2.19pre14 with ext3-0.0.6b. The server is running kernel-nfs.

Assertion failure in journal_dirty_metadata() at transaction.c line 796:
"bh->b_next_transaction == transaction"
Unable to handle kernel NULL pointer dereference at virtual address 00000000
current->tss.cr3 = 07ad0000, %cr3 = 07ad0000
*pde = 00000000
Oops: 0002
CPU:    1
EIP:    0010:[<c0154731>]
EFLAGS: 00010282
eax: 00000074   ebx: cffe3c80   ecx: ca5b6000   edx: 0000003a
esi: cb34eec0   edi: c40a5640   ebp: c8f697e0   esp: c5125f14
ds: 0018   es: 0018   ss: 0018

Whilst booting after the crash the following output was recorded while
mounting the ext3 filsystems;

JFS DEBUG: (recovery.c, 255): journal_recover: JFS: recovery, exit status 0,
recovered transactions 13283215 to 13283259
JFS DEBUG: (recovery.c, 257): journal_recover: JFS: Replayed 477 and revoked 0/0
blocks
EXT3-fs: 30:05: 3 orphan inodes deleted
EXT3-fs: recovery complete.
EXT3-fs: mounted filesystem with journal data mode.
EXT3-fs: mounted filesystem with journal data mode.
JFS DEBUG: (recovery.c, 255): journal_recover: JFS: recovery, exit status 0,
recovered transactions 2626369 to 2626370
JFS DEBUG: (recovery.c, 257): journal_recover: JFS: Replayed 2767 and revoked
0/0 blocks
EXT3-fs: 30:06: 7 orphan inodes deleted
EXT3-fs: recovery complete.
EXT3-fs: mounted filesystem with journal data mode.
JFS DEBUG: (recovery.c, 255): journal_recover: JFS: recovery, exit status 0,
recovered transactions 185215 to 185270
JFS DEBUG: (recovery.c, 257): journal_recover: JFS: Replayed 4887 and revoked
0/0 blocks
EXT3-fs: 30:0d: 1 orphan inode deleted
EXT3-fs: recovery complete.
EXT3-fs: mounted filesystem with journal data mode.
EXT3-fs: recovery complete.
EXT3-fs: mounted filesystem with journal data mode.
EXT3-fs: recovery complete.
EXT3-fs: mounted filesystem with journal data mode.


We will be happy to assist in futher investigation. 

-- 
Per Andreas Buer

Hi,

On Wed, Feb 28, 2001 at 11:11:12AM +0100, Per Andreas Buer
wrote:
> Our main NFS-server, running Debian Potato, died this morning whilst
> under quite heavy load - due to a runaway perl-script forking of some
> 200 instances of "/bin/cp" (the load was ~ 50). The server is
running
> 2.2.19pre14 with ext3-0.0.6b. The server is running kernel-nfs.
> 
> Assertion failure in journal_dirty_metadata() at transaction.c line 796:
"bh->b_next_transaction == transaction"
> Unable to handle kernel NULL pointer dereference at virtual address
00000000
> current->tss.cr3 = 07ad0000, %cr3 = 07ad0000
> *pde = 00000000
> Oops: 0002
> CPU:    1
> EIP:    0010:[<c0154731>]
> EFLAGS: 00010282
> eax: 00000074   ebx: cffe3c80   ecx: ca5b6000   edx: 0000003a
> esi: cb34eec0   edi: c40a5640   ebp: c8f697e0   esp: c5125f14
> ds: 0018   es: 0018   ss: 0018
This most likely indicates that somewhere, the VM tried to recurse
into ext3 while a transaction on a different partition was already in
progress.  Do you still have the backtrace so I can see where this
happened?

I should be able to reproduce this, I'll let you know how it goes.

Cheers,
 Stephen