Re: [PATCH 0/5] block: another block copy offload

[Christoph Hellwig <hch@infradead.org>]

On Thu, May 22, 2025 at 08:52:19AM -0700, Bart Van Assche wrote:
> Before any copy offloading code is merged in the upstream kernel, a plan
> should be presented for how this partial implementation will evolve into
> a generic solution. As Hannes already pointed out, support for
> copying between block devices is missing. SCSI support is missing.
> Device mapper support is missing. What is the plan for evolving this
> patch series into a generic solution? Is it even possible to evolve this
> patch series into a generic solution?

While I agree with some of this, please keep cross-device copies
out of the picture here.  The cross-namepace copies and cross LUN
copies in SCSI are a horrible misfeature that is almost impossible
to implement in an I/O stack.  If someone ever manages to get them
to work they most certainly would have to use a very different
infrastructure outside the block layer.