Re: [PATCH 07/14] iomap: simplify iomap_dio_bio_iter

[Christoph Hellwig <hch@lst.de>]

On Wed, Jan 14, 2026 at 02:51:23PM -0800, Darrick J. Wong wrote:
> On Wed, Jan 14, 2026 at 08:41:05AM +0100, Christoph Hellwig wrote:
> > Use iov_iter_count to check if we need to continue as that just reads
> > a field in the iov_iter, and only use bio_iov_vecs_to_alloc to calculate
> > the actual number of vectors to allocate for the bio.
> > 
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> 
> Huh, interesting.  bio_iov_vecs_to_alloc returns 0 if the iov_iter is of
> type bvec, so I guess we'd only run the loop body once before, and with
> zero pages?  Hrmm, that doesn't seem right.
> 
> Does that mean that we could always construct a bio for the entire bvec?
> Or does that just mean that directio doesn't get called with a bvec
> iterator?
> 
> Or, basic question: what the heck is a bvec?  A bio_vec?  So perhaps
> iomap_dio_bio_iter can't be called with a bio_vec because we're
> constructing a bio, not dealing with an existing one?
> 
> <have the cold meds kicked in yet?>

bio_iov_iter_get_pages reused the passed in bio_vec (yes, _bvec stands
for an array of bio_vecs) for iov_iter_is_bvec iters.  bios haver no
real size limit, although lower layers can't deal with more than
BIO_MAX_VECS vectors in a bunch of places.

So when bio_iov_iter_get_pages calls bio_iov_bvec_set for bvec iters,
there is an implicit assumption that it has less than BIO_MAX_VECS
vectors.  We should probably make that explicit, but I'll keep that
separate from this series.