Re: [PATCH v2 6/7] block: unpin all pages of a bvec in bio_iov_iter_align_down()

On Fri, Aug 28, 2026 at 05:36:24PM +0300, Tal Zussman wrote:

quoted

Both issues occur only with a logical block size above PAGE_SIZE and a
large folio backing the user buffer. On a device with a 64K logical
block size, an O_DIRECT pwritev() from a hugetlb mapping that ends 16K
past a block boundary leaks one huge page per call, whether the
remainder is its own bvec or the tail of a larger one.

Do you have a reproducer for that?

quoted

@@ -1213,14 +1219,24 @@ static int bio_iov_iter_align_down(struct bio *bio, struct iov_iter *iter,
 	bio->bi_iter.bi_size -= nbytes;
 	while (nbytes >= bv->bv_len) {
 		if (bio_flagged(bio, BIO_PAGE_PINNED))
-			unpin_user_page(bv->bv_page);
+			unpin_user_folio(bvec_folio(bv),
+					 bvec_nr_pages(bv));

And right after sending this, I noticed bvec_unpin()...

I think a cleaner solution is to move bvec_unpin() up, make bvec_nr_pages()
handle bv_offset > PAGE_SIZE, use bvec_nr_pages() in bvec_unpin(), and then
use bvec_unpin() here.

Yes, that sounds better.