[PATCH v4] nvme: fix corruption for passthrough meta/data

[Kanchan Joshi]

On Thu, Oct 26, 2023 at 8:38 PM Keith Busch <kbusch at kernel.org> wrote:
>
> On Thu, Oct 26, 2023 at 08:03:30PM +0530, Kanchan Joshi wrote:
> > On Mon, Oct 16, 2023 at 11:16 AM Christoph Hellwig <hch at lst.de> wrote:
> > >
> > > On Mon, Oct 16, 2023 at 12:49:45AM +0530, Kanchan Joshi wrote:
> > > > OTOH, this patch implemented a software-only way out. There are some
> > > > checks, but someone (either SW or HW) has to do those to keep things
> > > > right.
> > >
> > > It only verifies it to the read/write family of commands by
> > > interpreting these commands.  It still leaves a wide hole for any
> > > other command.
> >
> > Can you please explain for what command do you see the hole? I am
> > trying to see if it is really impossible to fix this hole for good.
> >
> > We only need to check for specific io commands of the NVM/ZNS command
> > set that can do extra DMA.
>
> The spec defines a few commands that may use MPTR, but most of the
> possible opcodes that could use it are not defined in spec. You'd have
> to break forward compatibility through this interface for non-root users
> by limiting its use to only the known opcodes and reject everything
> else.

I see. Future commands. Looks like every solution involves non-root
users to face some form of inconvenience.