Barriers and journaling filesystems

I don't see the problem with the power failure scenario. If you encounter a barrier at an MD
device, I'd think you would essentially not submit IO to the underlying devices until such
time as all I/O's prior to the barrier had completed. In that scenario, at the first sign of a
power failure effecting *any* of the drives, I'd start reporting I/O errors and be free of any
guarantees whatsoever.