Re: [PATCHv2 2/7] block: align the bio after building it

[Christoph Hellwig <hch@infradead.org>]

On Tue, Aug 05, 2025 at 07:11:18AM -0700, Keith Busch wrote:
> +static inline void bio_revert(struct bio *bio, unsigned int nbytes)

Can you add a little comment explaining what this code does?  The name
suggast it is similar to iov_iter_revert, but I'm not sure how similar
it is intended to be.  The direct poking into bi_io_vec suggest it
can only be used by the I/O submitter, and the use of bio_release_page
suggests it is closely tied to to bio_iov_iter_get_pages despite the
very generic name.

> +static int bio_align_to_lbs(struct bio *bio, struct iov_iter *iter)
> +{
> +	struct block_device *bdev = bio->bi_bdev;
> +	size_t nbytes;
> +
> +	if (!bdev)
> +		return 0;

So this is something horribly Kent put in where he failed to deal with
review feedback and just fed his stuff to Linus, and I think we need
to fix it when touching this code again.  Assuming we want to support
bio_iov_iter_get_pages on bios without bi_bdev set we simplify can't
rely in looking at bdev_logical_block_size here, but instead need to
pass it explicitly.  Which honestly doesn't sound to bad, just add an
explicit argument for the required alignment to bio_iov_iter_get_pages
instead of trying to derive it.  Which is actually going to be useful
to reduce duplicate checks for file systems the require > LBA size
alignment as well.

> -	return bio->bi_vcnt ? 0 : ret;
> +	if (bio->bi_vcnt)
> +		return bio_align_to_lbs(bio, iter);
> +	return ret;

Nit, this would flow a little more easily by moving the error /
exceptional case into the branch, i.e.

	if (!bio->bi_vcnt)
		return ret;
	return bio_align_to_lbs(bio, iter);