Re: [PATCH] block: support zone append bvecs

[Christoph Hellwig <hch@lst.de>]

On Wed, Mar 24, 2021 at 07:07:27AM +0000, Johannes Thumshirn wrote:
> >>  	if (iov_iter_is_bvec(iter)) {
> >> -		if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
> >> -			return -EINVAL;
> >> +		if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
> >> +			struct request_queue *q = bio->bi_bdev->bd_disk->queue;
> >> +			unsigned int max_append =
> >> +				queue_max_zone_append_sectors(q) << 9;
> >> +
> >> +			if (WARN_ON_ONCE(iter->count > max_append))
> >> +				return -EINVAL;
> >> +		}
> > 
> > That is not correct.  bio_iov_iter_get_pages just fills the bio as far
> > as we can, and then returns 0 for the next call to continue.  Basically
> > what you want here is a partial version of bio_iov_bvec_set.
> > 
> 
> Isn't that what I did? The above is checking if we have REQ_OP_ZONE_APPEND and
> then returns EINVAL if iter->count is bigger than queue_max_zone_append_sectors().
> If the check doesn't fail, its going to call bio_iov_bvec_set().

And that is the problem.  It should not fail, the payload is decoupled
from the max_append size.

Doing the proper thing is not too hard as described above - make sure
the bi_iter points to only the chunk of iter passed in that fits, and
only advance the passed in iter by that amount.