Bug#135504: Bug reproduced

[Phil Blundell <pb@nexus.co.uk>]

On Mon, 2002-03-04 at 11:12, Eduard Bloch wrote:
> #include <hallo.h>
> Eduard Bloch wrote on Mon Mar 04, 2002 um 11:26:36AM:
> 
> > > Kernel threads can become zombies in exactly the same way as any other
> > > process, ie if they die and their parent doesn't call wait.  As a
> > > temporary hack, try something like this:
> > 
> > Will do, and prepend a sync() in del_loop too.
> 
> Did not help. Since the 2.2.x kernel breaks too, the loop Zombies must
> be symptoms, but not the reason. The funny thing is:
> 
>  - it is mounted the first time, works, but on del_loop the ioctl fails
>  - disk is umounted though!

What error do you get from the del_loop ioctl?  That might be a clue.
Is the loop device itself properly unmounted at that point? 

p.