On Wed, Jul 13, 2022 at 01:17:56PM +0100, Matthew Wilcox wrote:

The firmware needs to prove to me that it *did something*.  That it
actually read those bytes that it claims to have verified.  The simplest
way to do so is to calculate a hash over the blocks which were read
(maybe the host needs to provide a nonce as part of the VERIFY command
so the drive can't "remember" the checksum).

From a protocol perspective, NVMe's verify command currently leaves 8-bytes in
the response unused, so this could be a reasonable way to return something like
a crc64 of the verified blocks. The verify command starts at an arbitrary block
and spans an arbitrary number of them, so I think a device "remembering" all
possible combinations as a way to cheat may be harder than just doing the work
:).

But this is just theoretical; it'd be some time before we'd see devices
supporting such a scheme, assuming there's support from the standards
committees.