1 // SPDX-License-Identifier: GPL-2.0
3 * Functions related to mapping data to requests
5 #include <linux/kernel.h>
6 #include <linux/sched/task_stack.h>
7 #include <linux/module.h>
9 #include <linux/blkdev.h>
10 #include <linux/uio.h>
15 bool is_our_pages : 1;
16 bool is_null_mapped : 1;
21 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
24 struct bio_map_data *bmd;
26 if (data->nr_segs > UIO_MAXIOV)
29 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask);
33 if (iter_is_iovec(data)) {
34 memcpy(bmd->iov, iter_iov(data), sizeof(struct iovec) * data->nr_segs);
35 bmd->iter.__iov = bmd->iov;
41 * bio_copy_from_iter - copy all pages from iov_iter to bio
42 * @bio: The &struct bio which describes the I/O as destination
43 * @iter: iov_iter as source
45 * Copy all pages from iov_iter to bio.
46 * Returns 0 on success, or error on failure.
48 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
51 struct bvec_iter_all iter_all;
53 bio_for_each_segment_all(bvec, bio, iter_all) {
56 ret = copy_page_from_iter(bvec->bv_page,
61 if (!iov_iter_count(iter))
64 if (ret < bvec->bv_len)
72 * bio_copy_to_iter - copy all pages from bio to iov_iter
73 * @bio: The &struct bio which describes the I/O as source
74 * @iter: iov_iter as destination
76 * Copy all pages from bio to iov_iter.
77 * Returns 0 on success, or error on failure.
79 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
82 struct bvec_iter_all iter_all;
84 bio_for_each_segment_all(bvec, bio, iter_all) {
87 ret = copy_page_to_iter(bvec->bv_page,
92 if (!iov_iter_count(&iter))
95 if (ret < bvec->bv_len)
103 * bio_uncopy_user - finish previously mapped bio
104 * @bio: bio being terminated
106 * Free pages allocated from bio_copy_user_iov() and write back data
107 * to user space in case of a read.
109 static int bio_uncopy_user(struct bio *bio)
111 struct bio_map_data *bmd = bio->bi_private;
114 if (!bmd->is_null_mapped) {
116 * if we're in a workqueue, the request is orphaned, so
117 * don't copy into a random user address space, just free
118 * and return -EINTR so user space doesn't expect any data.
122 else if (bio_data_dir(bio) == READ)
123 ret = bio_copy_to_iter(bio, bmd->iter);
124 if (bmd->is_our_pages)
131 static int bio_copy_user_iov(struct request *rq, struct rq_map_data *map_data,
132 struct iov_iter *iter, gfp_t gfp_mask)
134 struct bio_map_data *bmd;
139 unsigned int len = iter->count;
140 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
142 bmd = bio_alloc_map_data(iter, gfp_mask);
147 * We need to do a deep copy of the iov_iter including the iovecs.
148 * The caller provided iov might point to an on-stack or otherwise
151 bmd->is_our_pages = !map_data;
152 bmd->is_null_mapped = (map_data && map_data->null_mapped);
154 nr_pages = bio_max_segs(DIV_ROUND_UP(offset + len, PAGE_SIZE));
157 bio = bio_kmalloc(nr_pages, gfp_mask);
160 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, req_op(rq));
163 nr_pages = 1U << map_data->page_order;
164 i = map_data->offset / PAGE_SIZE;
167 unsigned int bytes = PAGE_SIZE;
175 if (i == map_data->nr_entries * nr_pages) {
180 page = map_data->pages[i / nr_pages];
181 page += (i % nr_pages);
185 page = alloc_page(GFP_NOIO | gfp_mask);
192 if (bio_add_pc_page(rq->q, bio, page, bytes, offset) < bytes) {
203 map_data->offset += bio->bi_iter.bi_size;
208 if (iov_iter_rw(iter) == WRITE &&
209 (!map_data || !map_data->null_mapped)) {
210 ret = bio_copy_from_iter(bio, iter);
213 } else if (map_data && map_data->from_user) {
214 struct iov_iter iter2 = *iter;
216 /* This is the copy-in part of SG_DXFER_TO_FROM_DEV. */
217 iter2.data_source = ITER_SOURCE;
218 ret = bio_copy_from_iter(bio, &iter2);
222 if (bmd->is_our_pages)
224 iov_iter_advance(iter, bio->bi_iter.bi_size);
227 bio->bi_private = bmd;
229 ret = blk_rq_append_bio(rq, bio);
243 static void blk_mq_map_bio_put(struct bio *bio)
245 if (bio->bi_opf & REQ_ALLOC_CACHE) {
253 static struct bio *blk_rq_map_bio_alloc(struct request *rq,
254 unsigned int nr_vecs, gfp_t gfp_mask)
258 if (rq->cmd_flags & REQ_ALLOC_CACHE && (nr_vecs <= BIO_INLINE_VECS)) {
259 bio = bio_alloc_bioset(NULL, nr_vecs, rq->cmd_flags, gfp_mask,
264 bio = bio_kmalloc(nr_vecs, gfp_mask);
267 bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq));
272 static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
275 iov_iter_extraction_t extraction_flags = 0;
276 unsigned int max_sectors = queue_max_hw_sectors(rq->q);
277 unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS);
282 if (!iov_iter_count(iter))
285 bio = blk_rq_map_bio_alloc(rq, nr_vecs, gfp_mask);
289 if (blk_queue_pci_p2pdma(rq->q))
290 extraction_flags |= ITER_ALLOW_P2PDMA;
291 if (iov_iter_extract_will_pin(iter))
292 bio_set_flag(bio, BIO_PAGE_PINNED);
294 while (iov_iter_count(iter)) {
295 struct page *stack_pages[UIO_FASTIOV];
296 struct page **pages = stack_pages;
301 if (nr_vecs > ARRAY_SIZE(stack_pages))
304 bytes = iov_iter_extract_pages(iter, &pages, LONG_MAX,
305 nr_vecs, extraction_flags, &offs);
306 if (unlikely(bytes <= 0)) {
307 ret = bytes ? bytes : -EFAULT;
311 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
313 if (unlikely(offs & queue_dma_alignment(rq->q)))
316 for (j = 0; j < npages; j++) {
317 struct page *page = pages[j];
318 unsigned int n = PAGE_SIZE - offs;
319 bool same_page = false;
324 if (!bio_add_hw_page(rq->q, bio, page, n, offs,
325 max_sectors, &same_page))
329 bio_release_page(bio, page);
335 * release the pages we didn't map into the bio, if any
338 bio_release_page(bio, pages[j++]);
339 if (pages != stack_pages)
341 /* couldn't stuff something into bio? */
343 iov_iter_revert(iter, bytes);
348 ret = blk_rq_append_bio(rq, bio);
354 bio_release_pages(bio, false);
355 blk_mq_map_bio_put(bio);
359 static void bio_invalidate_vmalloc_pages(struct bio *bio)
361 #ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
362 if (bio->bi_private && !op_is_write(bio_op(bio))) {
363 unsigned long i, len = 0;
365 for (i = 0; i < bio->bi_vcnt; i++)
366 len += bio->bi_io_vec[i].bv_len;
367 invalidate_kernel_vmap_range(bio->bi_private, len);
372 static void bio_map_kern_endio(struct bio *bio)
374 bio_invalidate_vmalloc_pages(bio);
380 * bio_map_kern - map kernel address into bio
381 * @q: the struct request_queue for the bio
382 * @data: pointer to buffer to map
383 * @len: length in bytes
384 * @gfp_mask: allocation flags for bio allocation
386 * Map the kernel address into a bio suitable for io to a block
387 * device. Returns an error pointer in case of error.
389 static struct bio *bio_map_kern(struct request_queue *q, void *data,
390 unsigned int len, gfp_t gfp_mask)
392 unsigned long kaddr = (unsigned long)data;
393 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
394 unsigned long start = kaddr >> PAGE_SHIFT;
395 const int nr_pages = end - start;
396 bool is_vmalloc = is_vmalloc_addr(data);
401 bio = bio_kmalloc(nr_pages, gfp_mask);
403 return ERR_PTR(-ENOMEM);
404 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
407 flush_kernel_vmap_range(data, len);
408 bio->bi_private = data;
411 offset = offset_in_page(kaddr);
412 for (i = 0; i < nr_pages; i++) {
413 unsigned int bytes = PAGE_SIZE - offset;
422 page = virt_to_page(data);
424 page = vmalloc_to_page(data);
425 if (bio_add_pc_page(q, bio, page, bytes,
427 /* we don't support partial mappings */
430 return ERR_PTR(-EINVAL);
438 bio->bi_end_io = bio_map_kern_endio;
442 static void bio_copy_kern_endio(struct bio *bio)
449 static void bio_copy_kern_endio_read(struct bio *bio)
451 char *p = bio->bi_private;
452 struct bio_vec *bvec;
453 struct bvec_iter_all iter_all;
455 bio_for_each_segment_all(bvec, bio, iter_all) {
456 memcpy_from_bvec(p, bvec);
460 bio_copy_kern_endio(bio);
464 * bio_copy_kern - copy kernel address into bio
465 * @q: the struct request_queue for the bio
466 * @data: pointer to buffer to copy
467 * @len: length in bytes
468 * @gfp_mask: allocation flags for bio and page allocation
469 * @reading: data direction is READ
471 * copy the kernel address into a bio suitable for io to a block
472 * device. Returns an error pointer in case of error.
474 static struct bio *bio_copy_kern(struct request_queue *q, void *data,
475 unsigned int len, gfp_t gfp_mask, int reading)
477 unsigned long kaddr = (unsigned long)data;
478 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
479 unsigned long start = kaddr >> PAGE_SHIFT;
488 return ERR_PTR(-EINVAL);
490 nr_pages = end - start;
491 bio = bio_kmalloc(nr_pages, gfp_mask);
493 return ERR_PTR(-ENOMEM);
494 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
498 unsigned int bytes = PAGE_SIZE;
503 page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask);
508 memcpy(page_address(page), p, bytes);
510 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
518 bio->bi_end_io = bio_copy_kern_endio_read;
519 bio->bi_private = data;
521 bio->bi_end_io = bio_copy_kern_endio;
530 return ERR_PTR(-ENOMEM);
534 * Append a bio to a passthrough request. Only works if the bio can be merged
535 * into the request based on the driver constraints.
537 int blk_rq_append_bio(struct request *rq, struct bio *bio)
539 struct bvec_iter iter;
541 unsigned int nr_segs = 0;
543 bio_for_each_bvec(bv, bio, iter)
547 blk_rq_bio_prep(rq, bio, nr_segs);
549 if (!ll_back_merge_fn(rq, bio, nr_segs))
551 rq->biotail->bi_next = bio;
553 rq->__data_len += (bio)->bi_iter.bi_size;
554 bio_crypt_free_ctx(bio);
559 EXPORT_SYMBOL(blk_rq_append_bio);
561 /* Prepare bio for passthrough IO given ITER_BVEC iter */
562 static int blk_rq_map_user_bvec(struct request *rq, const struct iov_iter *iter)
564 struct request_queue *q = rq->q;
565 size_t nr_iter = iov_iter_count(iter);
566 size_t nr_segs = iter->nr_segs;
567 struct bio_vec *bvecs, *bvprvp = NULL;
568 const struct queue_limits *lim = &q->limits;
569 unsigned int nsegs = 0, bytes = 0;
573 if (!nr_iter || (nr_iter >> SECTOR_SHIFT) > queue_max_hw_sectors(q))
575 if (nr_segs > queue_max_segments(q))
578 /* no iovecs to alloc, as we already have a BVEC iterator */
579 bio = blk_rq_map_bio_alloc(rq, 0, GFP_KERNEL);
583 bio_iov_bvec_set(bio, (struct iov_iter *)iter);
584 blk_rq_bio_prep(rq, bio, nr_segs);
586 /* loop to perform a bunch of sanity checks */
587 bvecs = (struct bio_vec *)iter->bvec;
588 for (i = 0; i < nr_segs; i++) {
589 struct bio_vec *bv = &bvecs[i];
592 * If the queue doesn't support SG gaps and adding this
593 * offset would create a gap, fallback to copy.
595 if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv->bv_offset)) {
596 blk_mq_map_bio_put(bio);
599 /* check full condition */
600 if (nsegs >= nr_segs || bytes > UINT_MAX - bv->bv_len)
602 if (bytes + bv->bv_len > nr_iter)
604 if (bv->bv_offset + bv->bv_len > PAGE_SIZE)
613 blk_mq_map_bio_put(bio);
618 * blk_rq_map_user_iov - map user data to a request, for passthrough requests
619 * @q: request queue where request should be inserted
620 * @rq: request to map data to
621 * @map_data: pointer to the rq_map_data holding pages (if necessary)
622 * @iter: iovec iterator
623 * @gfp_mask: memory allocation flags
626 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
627 * a kernel bounce buffer is used.
629 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
630 * still in process context.
632 int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
633 struct rq_map_data *map_data,
634 const struct iov_iter *iter, gfp_t gfp_mask)
636 bool copy = false, map_bvec = false;
637 unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
638 struct bio *bio = NULL;
644 else if (blk_queue_may_bounce(q))
646 else if (iov_iter_alignment(iter) & align)
648 else if (iov_iter_is_bvec(iter))
650 else if (!user_backed_iter(iter))
652 else if (queue_virt_boundary(q))
653 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
656 ret = blk_rq_map_user_bvec(rq, iter);
659 if (ret != -EREMOTEIO)
661 /* fall back to copying the data on limits mismatches */
668 ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
670 ret = bio_map_user_iov(rq, &i, gfp_mask);
675 } while (iov_iter_count(&i));
680 blk_rq_unmap_user(bio);
685 EXPORT_SYMBOL(blk_rq_map_user_iov);
687 int blk_rq_map_user(struct request_queue *q, struct request *rq,
688 struct rq_map_data *map_data, void __user *ubuf,
689 unsigned long len, gfp_t gfp_mask)
692 int ret = import_ubuf(rq_data_dir(rq), ubuf, len, &i);
694 if (unlikely(ret < 0))
697 return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
699 EXPORT_SYMBOL(blk_rq_map_user);
701 int blk_rq_map_user_io(struct request *req, struct rq_map_data *map_data,
702 void __user *ubuf, unsigned long buf_len, gfp_t gfp_mask,
703 bool vec, int iov_count, bool check_iter_count, int rw)
708 struct iovec fast_iov[UIO_FASTIOV];
709 struct iovec *iov = fast_iov;
710 struct iov_iter iter;
712 ret = import_iovec(rw, ubuf, iov_count ? iov_count : buf_len,
713 UIO_FASTIOV, &iov, &iter);
718 /* SG_IO howto says that the shorter of the two wins */
719 iov_iter_truncate(&iter, buf_len);
720 if (check_iter_count && !iov_iter_count(&iter)) {
726 ret = blk_rq_map_user_iov(req->q, req, map_data, &iter,
729 } else if (buf_len) {
730 ret = blk_rq_map_user(req->q, req, map_data, ubuf, buf_len,
735 EXPORT_SYMBOL(blk_rq_map_user_io);
738 * blk_rq_unmap_user - unmap a request with user data
739 * @bio: start of bio list
742 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
743 * supply the original rq->bio from the blk_rq_map_user() return, since
744 * the I/O completion may have changed rq->bio.
746 int blk_rq_unmap_user(struct bio *bio)
748 struct bio *next_bio;
752 if (bio->bi_private) {
753 ret2 = bio_uncopy_user(bio);
757 bio_release_pages(bio, bio_data_dir(bio) == READ);
762 blk_mq_map_bio_put(next_bio);
767 EXPORT_SYMBOL(blk_rq_unmap_user);
770 * blk_rq_map_kern - map kernel data to a request, for passthrough requests
771 * @q: request queue where request should be inserted
772 * @rq: request to fill
773 * @kbuf: the kernel buffer
774 * @len: length of user data
775 * @gfp_mask: memory allocation flags
778 * Data will be mapped directly if possible. Otherwise a bounce
779 * buffer is used. Can be called multiple times to append multiple
782 int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
783 unsigned int len, gfp_t gfp_mask)
785 int reading = rq_data_dir(rq) == READ;
786 unsigned long addr = (unsigned long) kbuf;
790 if (len > (queue_max_hw_sectors(q) << 9))
795 if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||
796 blk_queue_may_bounce(q))
797 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
799 bio = bio_map_kern(q, kbuf, len, gfp_mask);
804 bio->bi_opf &= ~REQ_OP_MASK;
805 bio->bi_opf |= req_op(rq);
807 ret = blk_rq_append_bio(rq, bio);
814 EXPORT_SYMBOL(blk_rq_map_kern);