Comment # 5 on bug 1165535 from Hannes Reinecke

I wouldn't discount the possibility that we're not properly for memory
allocation problems, or that the code paths catching these issues are
improperly done.
IE if there's a memory allocation error (in, say, dm-crypt), and dm-crypt
improperly handles the failed allocation in that it catches -ENOMEM yet
continues with operation we might end up with such a situation.

But as you figured, triggering this is extremely hard.
But maybe worthwhile a testcase, and maybe our eager QA team can spin off
something here.
Sebastian? Do you think it would be possible to devise a testcase for testing
this situation?