Re: [PATCH v3 2/5] RDMA/core: use IOVA-based DMA mapping for bvec RDMA operations

[Christoph Hellwig <hch@lst.de>]

On Fri, Jan 23, 2026 at 10:04:07AM -0500, Chuck Lever wrote:
> On Fri, Jan 23, 2026, at 1:28 AM, Christoph Hellwig wrote:
> >> +	/* Link all bvecs into the IOVA space */
> >> +	link_iter = *iter;
> >> +	while (link_iter.bi_size) {
> >> +		struct bio_vec bv = mp_bvec_iter_bvec(bvec, link_iter);
> >> +
> >> +		ret = dma_iova_link(dma_dev, &ctx->iova.state, bvec_phys(&bv),
> >> +				    mapped_len, bv.bv_len, dir, 0);
> >> +		if (ret)
> >> +			goto out_destroy;
> >> +
> >> +		mapped_len += bv.bv_len;
> >> +		bvec_iter_advance(bvec, &link_iter, bv.bv_len);
> >> +	}
> >
> > Why is this using a local link_iter?  We're not using iter later.
> 
> I think we don't want to leak a partially-updated iter if the
> API call returns an error.

That's how all the block layer bvec_iter-based API work.  The
functions consume the iter.  If a caller needs to save it for
some reason, it stashes away a copy.