Re: [Gluster-devel] 2.0.0rc4 (and rc5) locks up when used as root

[Gordan Bobic]

Anand Avati wrote:
> > It looks like it locks up when used as root (afr/replicate) at the point
> > where it initially starts up udev (not 100% sure where exactly yet, will
> > have to put some trace code in rc.sysinit).
> >
> > 2.0.0rc2 didn't have this problem.
>
> Can you try rc5? Though still it is still under QA, you might want to
> give it a try since some transport related code changes have gone it
> which might be the reason for your lockup.

The lock-up still occurs with rc5.

I've done some more digging, however. It appears to die at this pint in 
rc.sysinit, between debug 3 and debug 4:

#################
echo "debug 3"
# Load other user-defined modules
for file in /etc/sysconfig/modules/*.modules ; do
  [ -x $file ] && $file
done

# Load modules (for backward compatibility with VARs)
if [ -f /etc/rc.modules ]; then
        /etc/rc.modules
fi
echo "debug 4"
#################

There is no rc.modules file, so contents of that can be ruled out.

# ls -l /etc/sysconfig/modules/
total 8
-rwxr-xr-x 1 root root 100 May 25  2008 udev-stw.modules

# cat /etc/sysconfig/modules/udev-stw.modules
#!/bin/sh
for i in nvram floppy parport lp snd-powermac;do
        modprobe $i >/dev/null 2>&1
done

I have just rebuilt my initrds separately with rc2 and rc5. rc2 works 
fine, rc5 fails. No other changes to the system between the two attempts.

Oh, and the first access failure bug is still there.

I couldn't test for the memory leak in rc5 since I couldn't get it to 
boot due to the lock-up mentioned above.

I'll try disabling those modules listed above since I don't need them on 
this setup, but I can confirm that modprobe itself works fine. So it 
sounds like a problem/bug elsewhere. Possibly a buffer-overrun somewhere 
that gets triggered by rc4/rc5.

BTW, in case it's relevant, I'm using the fuse kernel module from 
2.6.24.7, rather than the one from the patched package, because the one 
in the kernel appears to be later. Can anyone confirm if there are any 
known problems with this? Is there any strong reason why I should use a 
different kernel module (e.g. one from the patched fuse 2.7.4 package)?

Gordan