[PATCH 5/6] nvme: also return I/O command effects from nvme_command_effects

[Christoph Hellwig]

On Thu, Dec 22, 2022 at 08:38:08AM -0700, Keith Busch wrote:
> On Thu, Dec 22, 2022 at 09:57:37AM +0100, Christoph Hellwig wrote:
> > On Wed, Dec 21, 2022 at 02:42:47PM -0700, Keith Busch wrote:
> > > Just noticed we need to mask out the CSE value for IO commands in case
> > > some crazy controller sets it. If the driver were to see that value as
> > > non-zero, it will freeze the IO queue the ioctl requires to to submit
> > > its own IO request.
> > 
> > I'd rather reject passthrough of I/O commands with that effect
> > entirely, does that sound reasonable?
> 
> Sounds good to me.

I ended up to actually implement your suggestion of clearing it.
This does not block command execution, and we already warn about
unusual effects.