How can I simulate disk failure on 2.4?

I need some kind of way to fail a single disk partition (hdg8) while the
system is running, then reenable it. It doesn't have to be general-purpose and
I don't mind doing ugly hacks to the disk code to make it happen.

I was thinking of putting a flag in /proc somewhere, along with some code
in generic_make_request like this:

  if (proc_fail_the_disk && bh->b_rdev == the_disk)
    do_something_to_the_request;

What field could I change in the request that would make it fail without
causing too many side effects? (I suppose I'd learn a lot by just picking a likely field and whacking it, but I'm not sure I want to try and debug the mess.)

Is this even the right approach at all?

--
Chuck

Stephen Tweedie has a gizmo which does that.

http://people.redhat.com/sct/patches/testdrive/testdrive-1.1-for-2.4.19pre10.patch

The documentation is at google("tweedie testdrive") ;)