Shades of mdraid ...

It seems most systems have difficulty passing an error back up the chain when there's a problem short of an out-and-out failure.

Raid, for example, could check majority voting if there's a mirror, or parity for raids 5 and 6. But at present, on a read, it just assumes everything is hunky-dory if the data blocks read okay. There is no mechanism for checking and indicating a problem. Okay, with a two-disk mirror or raid 5, there's nothing that can be done to fix an integrity error, but it would be nice to know it's happened. And if you know a disk is flaky it would be nice to be able to tell any fix utility to assume that's the problem. At present, the fix utilities assume by default it's the parity that's dud, so even with raid-6 an ill-informed "fix" will simply trash your data if that assumption isn't true.

Cheers,
Wol