On 2021/11/24 14:09, hch@infradead.org wrote:

On Wed, Nov 24, 2021 at 07:07:18AM +0800, Qu Wenruo wrote:

quoted

In that case, the missing piece seems to be a way to convert a splitted
plain bio into a REQ_OP_ZONE_APPEND bio.

Can this be done without slow bvec copying?

Yes.  I have a WIP stacking driver that converts writes to zone appends
and it does just that:

	sector_t orig_sector = bio->bi_iter.bi_sector;
	unsigned int bio_flags = bio->bi_opf & ~REQ_OP_MASK;

	...

	clone = bio_clone_fast(bio, GFP_NOIO, &bdev->write_bio_set);

	...

	clone->bi_opf = REQ_OP_ZONE_APPEND | REQ_NOMERGE | bio_flags;

Just so simple? Then that's super awesome.

But I'm a little concerned about the bio_add_hw_page() call in
bio_add_zoned_append().

It's not exactly the same as bio_add_page().

Does it mean as long as our splitted bio doesn't exceed zone limit, we
can do the convert without any further problem?

Thanks,
Qu

	bio_set_dev(clone, dev->lower_bdev);
	clone->bi_iter.bi_sector = zone_sector;
	trace_block_bio_remap(clone, disk_devt(disk), orig_sector);

quoted

Thanks,
Qu

---end quoted text---