[PATCH] nvmet: fix KATO offset in Set Features

[Daniel Verkamp]

On 12/09/2016 11:28 AM, Christoph Hellwig wrote:
> On Wed, Dec 07, 2016 at 12:09:40PM -0700, Daniel Verkamp wrote:
>> Should I remove the nvmet_set_result() or fix it to at least return
>> the correct units?
> 
> Please remove it, as the spec does not define the result.  Otherwise
> the patch looks fine to me.

I sent a new version of the patch that just fixes the unused variable.
I'll send another one to remove the result.  I was going to remove it
in this patch, but I couldn't find out where (if anywhere) the result
is initialized to 0 in the common command handling.  Is it OK to just
remove the call to nvmet_set_result()? Is the whole request zeroed
out somewhere earlier? Or should I just change it to call
nvmet_set_result(req, 0) for clarity?

> Thanks for doing this work! Btw, did you catch this using a test
> suite?

I caught this when testing the SPDK NVMe over Fabrics host code, which
was using the Set Features command to set up KATO.  However, I later
realized that since we already set KATO during Connect, it's pretty
useless to configure it again with a Set Features - so I removed the
code that was hitting this bug.

A NVMe over Fabrics test suite would be a great idea, though...

Thanks,
-- Daniel