Unkillables

[Jeremy Allison]

On Thu, Nov 29, 2001 at 04:25:15PM -0600, Chris Tooley wrote:
> There is a real need for a kernel level tool to perform an actual kill
> on a process.  Not sending a SIGKILL to the process but actually ripping
> the process out of the scheduler deallocating the memory and closing
> open files.  For this exact reason.  This just doesn't seem to be
> available.  And for all the zombie processes, throughout the history of
> computing I'm amazed that there doesn't seem to be anything, no matter
> how difficult to impliment, to do this.
> 
> Ahhh, now I feel better.  I've been bitten by zombie's comsuming a
> server twice this week and I'm a little off balance about it. :)

I used to feel like that, then I wandered around in kernel
source code, and became "enlightened" :-) :-).

You just can't do that. It would make the kernel unstable.
Such things are the sign of a kernel bug, and it's better
to fix it than to hack in code to destabilise a running kernel.

It's the same problem as the "why can't I just kill/suspend
my pthread, I know it's gone wrong, why can't I just make it
die...." :-). The same reasons apply (read comp.programming.threads
and read *anything* by Dave Buttenhof (sp?) to achieve
enlightenment :-).

Jeremy.