On 06/06/24 06:56AM, Christoph Hellwig wrote:
@@ -1301,15 +1301,49 @@ static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) for (left = size, i = 0; left > 0; left -= len, i++) { struct page *page = pages[i]; + struct folio *folio = page_folio(page); + + /* Calculate the offset of page in folio */ + folio_offset = (folio_page_idx(folio, page) << PAGE_SHIFT) + + offset; + + len = min_t(size_t, (folio_size(folio) - folio_offset), left); + + num_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); + + if (num_pages > 1) {I still hate having this logic in the block layer. Even if it is just a dumb wrapper with the same logic as here I'd much prefer to have a iov_iter_extract_folios, which can then shift down into a pin_user_folios_fast and into the MM slowly rather than adding this logic to the caller.
iov_iter_extract_folios will require allocation of a folio_vec array which then will be filled after processing the pages[] array. This will introduce extra allocation cost in the hot path. -- Kundan
