[RFC 5/5] nvme: wire-up support for async-passthru on char-device.

[Kanchan Joshi]

On Tue, Apr 5, 2022 at 11:32 AM Christoph Hellwig <hch at lst.de> wrote:
>
> On Mon, Apr 04, 2022 at 07:55:05PM +0530, Kanchan Joshi wrote:
> > > Something like this (untested) patch should help to separate
> > > the much better:
> >
> > It does, thanks. But the only thing is - it would be good to support
> > vectored-passthru too (i.e. NVME_IOCTL_IO64_CMD_VEC) for this path.
> > For the new opcode "NVME_URING_CMD_IO" , either we can change the
> > cmd-structure or flag-based handling so that vectored-io is supported.
> > Or we introduce NVME_URING_CMD_IO_VEC also for that.
> > Which one do you prefer?
>
> I agree vectored I/O support is useful.
>
> Do we even need to support the non-vectored case?
Would be good to have, I suppose.
Helps keeping it simple when user-space wants to use a single-buffer
(otherwise it must carry psuedo iovec for that too).

> Also I think we'll want admin command passthrough on /dev/nvmeX as
> well, but I'm fine solving the other items first.
>
> > > +static int nvme_ioctl_finish_metadata(struct bio *bio, int ret,
> > > +               void __user *meta_ubuf)
> > > +{
> > > +       struct bio_integrity_payload *bip = bio_integrity(bio);
> > > +
> > > +       if (bip) {
> > > +               void *meta = bvec_virt(bip->bip_vec);
> > > +
> > > +               if (!ret && bio_op(bio) == REQ_OP_DRV_IN &&
> > > +                   copy_to_user(meta_ubuf, meta, bip->bip_vec->bv_len))
> > > +                       ret = -EFAULT;
> >
> > Maybe it is better to move the check "bio_op(bio) != REQ_OP_DRV_IN" outside.
> > Because this can be common, and for that we can avoid entering into
> > the function call itself (i.e. nvme_ioctl_finish_metadata).
>
> Function calls are pretty cheap, but I'll see what we can do.  I'll try
> to come up with a prep series to refactor the passthrough support for
> easier adding of the io_uring in the next days.

In that case we will base the newer version on its top.