> Ulrich Windl wrote:
> > I had some kernel protection faults that caused a program (xosview)
> > to abort with SIGSEGV. During shutdown I noticed that /proc couldn't
> > be unmounted (/proc: Filesystem busy). Now I wonder if that's just a
> > very special case or whether the inode of an aborted process is still
> > busy somehow.
> A kernel oops or other bug may cause an inode not to be released or
> other such problems. But nothing from user space should ever cause a
> resource leak; otherwise there would be no way to guarantee the
> stability of the system over time.
> Even one leaked inode could be repeated until the system came down.
> So when you say "kernel protection faults", was this an oops or other
> kernel bug?
The problem was: The new version of xosview comes with a kernel module
named "memstat" which allows for more detailed memory monitoring. After
upgrading to 2.0.32 it's needed to recompile this module because a new
field (condemned) in the inode structure has been added. If you load the
old module without recompiling you will get the kernel oops described by