The process was stuck in a loop somewhere in Kernel-land I assume, as
I couldn't terminate it. Ended up changing into single user mode to
unmount and fsck the filesystem. Still had this damn rm running chewing
up all idle CPU time.
If you were able to unmount the filesystem while the rm was still
running, then things were pretty confused indeed. The real question is
wheather this was caused by (a) filesystem corruption, (b) a real kernel
bug / race condition in the filesystem code, (c) a kernel bug somewhere
else that trashed memory which the filesystem code depends on, or (d) a
memory corruption bug.
Unfortunately, it sounds like you weren't able to collect enough
information to figure out which one of these cases is correct.
- Ted