It took some further time to understand what is closer to the source
of the problem. Previously I had just bisected backwards until ptrace
started working again because I knew it had broken between 2.6.14 and
2.6.21. The test case provided in the patch I submitted either always
fails or always succeeds. I had a particular machine and file system
that it always failed on. I reduced the configuration to UP i386 with
a file system that was using full kernel auditing.
It turns out that it is the _TIF_SYSCALL_AUDIT interaction in entry.S
is more likely the culprit here. This flag was getting turned on as a
result of using kernel/user space auditing. I found that you can turn
off CONFIG_AUDIT and use the patch below to "simulate" the same
circumstance. Then you should be able to observe the same failure I
saw directly with a vanilla 2.6.21 i386 kernel.
diff --git a/arch/i386/kernel/entry.S b/arch/i386/kernel/entry.S
diff --git a/kernel/fork.c b/kernel/fork.c
index 6af959c..fb47ab9 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1154,6 +1154,9 @@ static struct task_struct *copy_process(unsigned long clone_flags,
#ifdef TIF_SYSCALL_EMU
clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
#endif
+ /* HACK to always turn on syscall auditing */
+ set_tsk_thread_flag(p, TIF_SYSCALL_AUDIT);
+ /* end HACK to simulate auditing */
/* Our parent execution domain becomes current domain
These must match for thread signalling to apply */
Let me know if you need further details, and it certainly means some
further testing is in order against your newer patch. I am also
interested in what test cases fail that you mentioned in your original
e-mail on this topic.
Thanks,
Jason.
--
