block dangerous passthrough operation

[Keith Busch]

On Tue, Nov 22, 2022 at 07:47:43AM +0100, Christoph Hellwig wrote:
> On Mon, Nov 21, 2022 at 08:35:42AM -0700, Keith Busch wrote:
> > It's not like just anyone can do this kind of craziness. It's like a
> > soft equivalent of pulling a cable. There are usually access
> > restrictions for untrusted people.
> 
> Yes, there are.  But that doesn't change us having to deal with the
> fallout.

What exactly happened here? If I got a bug report that says "it broke
when I did this admin command", the answer is "don't do that. *plonk*".

Did you spend too much time on a bug report that didn't include repro
steps or something?

If you really need to do this, I think augmenting the
nvme_known_admin_effects() with a driver specific flag saying "Do Not
Execute" or something like that provides a more centralized location for
this type of check.