[Haifux] Kernel oops, so what?
Kohn Emil Dan
emild at cs.technion.ac.il
Sat Jan 16 03:27:06 MSK 2010
On Fri, 15 Jan 2010, Eli Billauer wrote:
> Thanks for your answers (on this one and on my other issues).
> I realize that the oops is still an oops, only nowadays nobody want to
> stop the whole show, just because some kernel code misbehaved. If the
> general idea is that the worst thing a kernel can do is to crash,
No, it isn't. Crashing is one of the best of the options in some
situations. Imagine a bug in the filesystem that writes a zero byte at
random places on the filesystem or even funnier, on the neighboring
partitions where you have installed another operating system. (yes
I made one such bug once but I did not have another
OS installed on other partitions;-). This won't crash the kernel
too fast. But the consequences are not funny at all.
> why crash now?
Because the consequences of a faulty kernel can be very dire. The kernel
controls the hardware. A faulty kernel can decide to
flash the firmware on your DVD drive. Or the BIOS on your motherboard.
It's not funny to recover from these.
The kernel is assumed to be a trusted component. When this assumption is
gone, all your assumptions on security are gone. You are basically back to
Windows 3.1 when every process can crash the other one at will. Therefore
as soon as the kernel finds itself in an inconsistent state from which it
cannot recover, the best option is to crash.
> Kill the offender, hope it didn't have time to screw things
> up too much, and go on as if nothing happened.
> Sooner or later the bug will be fixed, user upgrades kernel, the user
> won't notice anything. Why make the user's experience miserable with
> ugly crashes?
Well, some people believe that a corrupted filesystem in an irreparable
state or a dead DVD-ROM drive is a much worse outcome than a kernel crash.
Some security-conscious people (no, I am not one of them) will argue that
a cryptography grade random number generator that no longer generates
random numbers due to a bug in a driver is a very bad thing.
> As for my own computer's welfare, this seems to happen once per reboot,
> with no stability issues at all. I'll see if I can find a way to avoid
> this, but my gut feeling tells me that the answer is a kernel patch,
> which is already waiting in the kernel's digestive tract.
Sure, as long as it is a personal computer with little important data on
it, this is almost OK. However some people want to use linux for more
serious applications. I doubt that such kernel behavior is acceptable.
> Web: http://www.billauer.co.il
> Haifux mailing list
> Haifux at haifux.org
More information about the Haifux