2003-01-17 05:49:41

by Linux Geek

[permalink] [raw]
Subject: [DUMB]: Is kernel oops always a kernel bug ???

Hi all ,

If a kernel oops, then is the problem always with the kernel ? Can't a
user proggie ( running as root ) do something insane and make the
kernel oops ?

TIA


2003-01-17 06:06:17

by jurriaan

[permalink] [raw]
Subject: Re: [DUMB]: Is kernel oops always a kernel bug ???

From: Linux Geek <[email protected]>
Date: Fri, Jan 17, 2003 at 11:33:46AM +0530
> Hi all ,
>
> If a kernel oops, then is the problem always with the kernel ? Can't a
> user proggie ( running as root ) do something insane and make the
> kernel oops ?
>
That shouldn't happen - it's called DOS (Denial-Of-Service) if a user
program can make the kernel oops. Some of these probably still exist in
the linux kernel, nothing is perfect :-)

But another option exists: hardware failures. If your memory decides to
take a break, you will get an oops also.

Good luck,
Jurriaan
--
Holed up in your little room, we talk for hour on empty hour,
pacing up and down between the walls that we have built ourselves.
New Model Army - Long Goodbye
GNU/Linux 2.5.58 SMP/ReiserFS 2x2752 bogomips load av: 0.16 0.61 0.61

2003-01-17 06:09:48

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: [DUMB]: Is kernel oops always a kernel bug ???

On Fri, 17 Jan 2003 11:33:46 +0530, Linux Geek <[email protected]> said:
> If a kernel oops, then is the problem always with the kernel ? Can't a
> user proggie ( running as root ) do something insane and make the
> kernel oops ?

At least in theory, there should be *nothing* that can happen in user space
that will kill the kernel. However, in practice, if a program is running
as root, it can definitely blortch things up. This is mostly due to the
assumption that the root user has a clue, and that if they are (for instance)
opening /proc/kcore for writing, that they know what they're doing. Similarly,
if a program opens /dev/hda1 for writing and scribbles over the superblock,
it may be a bit difficult to mount the filesystem.

So in general, yes, the root user can screw things up in the kernel. On the
other hand, the root user can also 'rm -rf /' which doesn't require any
extraordinary kernel assistance, just the unlink() system call. The only
solution here is to not give root to clueless bozos, and to take regular
backups. ;)
--
Valdis Kletnieks
Computer Systems Senior Engineer
Virginia Tech


Attachments:
(No filename) (226.00 B)