New criterion for Checksum

[Patrick O'Callaghan]

On Wed, 2012-02-01 at 16:33 +0000, Andre Robatino wrote:
> Frank Murphy <frankly3d <at> gmail.com> writes:
> 
> > > A built-in checksum is only useful for checking for natural corruption, not a
> > > deliberate fake (since in that case it's easy to change the checksum to the
> > > correct one for the fake). Even md5 is more than enough for this purpose.
> > >
> > 
> > So it's not error proof,
> > it can fail and still have a perfect disk, correct.
> 
> Yes, there can be a bug that causes the mediacheck to fail even though the disc
> is good (this actually happened recently during development). Conversely, it can
> pass even if the disc is fake.

mediacheck failing when the disk is good is a bug. mediacheck passing
when the disk is bad may or may not be a bug, due to how checksums work.
It's always possible (though *very* unlikely) to get bad data that gives
a good checksum. This is not a failing of MD5 but of all checksum
schemes.

poc