Created attachment 25676 [details]
The kernel oops

Obvious questions: is it repeatable and if so, is the script public ?

It is not repeatable.  The script is very similar to lndir from X11R6. When
it happened, it was linking a HUGE source directory to another, like

# lndir.pl src tgt

where `src' was a HUGE tree and `tgt' was a newly created dir.

Which kernel?  You mention 2.4-0.9.3.1, but the beta3 kernel (which has the same
ext3 version) uses modular ext3 and your oops seems to have ext3 built-in.

Thanks.

I  rebuilt the kernel with the builtin ext3.

Hopefully fixed in ext3 cvs head.  We'll pull that into the kernel srpm and push
a build through shortly.

I was told it was fixed a couple of days ago in 0.9.5. Here is
the relevant change:

Index: transaction.c
===================================================================
RCS file: /cvsroot/gkernel/ext3/fs/jbd/transaction.c,v
retrieving revision 1.62
retrieving revision 1.62.2.1
diff -u -r1.62 -r1.62.2.1
--- transaction.c       2001/07/28 06:01:33     1.62
+++ transaction.c       2001/07/30 02:29:25     1.62.2.1
@@ -552,6 +561,9 @@
                        JBUFFER_TRACE(jh, "stealing from checkpoint mode");
                        J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
                        J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
+
+                       J_ASSERT(handle->h_buffer_credits > 0);
+                       handle->h_buffer_credits--;

                        /* This will clear BH_Dirty and set BH_JBDDirty. */
                        JBUFFER_TRACE(jh, "file as BJ_Reserved");

This defect is considered SHOULD-FIX for Fairfax.

Fixed in recent kernel builds: roswell has the patch.