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);
32 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
34 bmd->iter.iov = bmd->iov;
39 * bio_copy_from_iter - copy all pages from iov_iter to bio
40 * @bio: The &struct bio which describes the I/O as destination
41 * @iter: iov_iter as source
43 * Copy all pages from iov_iter to bio.
44 * Returns 0 on success, or error on failure.
46 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
49 struct bvec_iter_all iter_all;
51 bio_for_each_segment_all(bvec, bio, iter_all) {
54 ret = copy_page_from_iter(bvec->bv_page,
59 if (!iov_iter_count(iter))
62 if (ret < bvec->bv_len)
70 * bio_copy_to_iter - copy all pages from bio to iov_iter
71 * @bio: The &struct bio which describes the I/O as source
72 * @iter: iov_iter as destination
74 * Copy all pages from bio to iov_iter.
75 * Returns 0 on success, or error on failure.
77 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
80 struct bvec_iter_all iter_all;
82 bio_for_each_segment_all(bvec, bio, iter_all) {
85 ret = copy_page_to_iter(bvec->bv_page,
90 if (!iov_iter_count(&iter))
93 if (ret < bvec->bv_len)
101 * bio_uncopy_user - finish previously mapped bio
102 * @bio: bio being terminated
104 * Free pages allocated from bio_copy_user_iov() and write back data
105 * to user space in case of a read.
107 static int bio_uncopy_user(struct bio *bio)
109 struct bio_map_data *bmd = bio->bi_private;
112 if (!bmd->is_null_mapped) {
114 * if we're in a workqueue, the request is orphaned, so
115 * don't copy into a random user address space, just free
116 * and return -EINTR so user space doesn't expect any data.
120 else if (bio_data_dir(bio) == READ)
121 ret = bio_copy_to_iter(bio, bmd->iter);
122 if (bmd->is_our_pages)
129 static int bio_copy_user_iov(struct request *rq, struct rq_map_data *map_data,
130 struct iov_iter *iter, gfp_t gfp_mask)
132 struct bio_map_data *bmd;
137 unsigned int len = iter->count;
138 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
140 bmd = bio_alloc_map_data(iter, gfp_mask);
145 * We need to do a deep copy of the iov_iter including the iovecs.
146 * The caller provided iov might point to an on-stack or otherwise
149 bmd->is_our_pages = !map_data;
150 bmd->is_null_mapped = (map_data && map_data->null_mapped);
152 nr_pages = bio_max_segs(DIV_ROUND_UP(offset + len, PAGE_SIZE));
155 bio = bio_kmalloc(nr_pages, gfp_mask);
158 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, req_op(rq));
161 nr_pages = 1U << map_data->page_order;
162 i = map_data->offset / PAGE_SIZE;
165 unsigned int bytes = PAGE_SIZE;
173 if (i == map_data->nr_entries * nr_pages) {
178 page = map_data->pages[i / nr_pages];
179 page += (i % nr_pages);
183 page = alloc_page(GFP_NOIO | gfp_mask);
190 if (bio_add_pc_page(rq->q, bio, page, bytes, offset) < bytes) {
201 map_data->offset += bio->bi_iter.bi_size;
206 if ((iov_iter_rw(iter) == WRITE &&
207 (!map_data || !map_data->null_mapped)) ||
208 (map_data && map_data->from_user)) {
209 ret = bio_copy_from_iter(bio, iter);
213 if (bmd->is_our_pages)
215 iov_iter_advance(iter, bio->bi_iter.bi_size);
218 bio->bi_private = bmd;
220 ret = blk_rq_append_bio(rq, bio);
234 static void blk_mq_map_bio_put(struct bio *bio)
236 if (bio->bi_opf & REQ_ALLOC_CACHE) {
244 static struct bio *blk_rq_map_bio_alloc(struct request *rq,
245 unsigned int nr_vecs, gfp_t gfp_mask)
249 if (rq->cmd_flags & REQ_POLLED) {
250 blk_opf_t opf = rq->cmd_flags | REQ_ALLOC_CACHE;
252 bio = bio_alloc_bioset(NULL, nr_vecs, opf, gfp_mask,
257 bio = bio_kmalloc(nr_vecs, gfp_mask);
260 bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq));
265 static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
268 unsigned int max_sectors = queue_max_hw_sectors(rq->q);
269 unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS);
274 if (!iov_iter_count(iter))
277 bio = blk_rq_map_bio_alloc(rq, nr_vecs, gfp_mask);
281 while (iov_iter_count(iter)) {
282 struct page **pages, *stack_pages[UIO_FASTIOV];
287 if (nr_vecs <= ARRAY_SIZE(stack_pages)) {
289 bytes = iov_iter_get_pages2(iter, pages, LONG_MAX,
292 bytes = iov_iter_get_pages_alloc2(iter, &pages,
295 if (unlikely(bytes <= 0)) {
296 ret = bytes ? bytes : -EFAULT;
300 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
302 if (unlikely(offs & queue_dma_alignment(rq->q)))
305 for (j = 0; j < npages; j++) {
306 struct page *page = pages[j];
307 unsigned int n = PAGE_SIZE - offs;
308 bool same_page = false;
313 if (!bio_add_hw_page(rq->q, bio, page, n, offs,
314 max_sectors, &same_page)) {
325 * release the pages we didn't map into the bio, if any
328 put_page(pages[j++]);
329 if (pages != stack_pages)
331 /* couldn't stuff something into bio? */
333 iov_iter_revert(iter, bytes);
338 ret = blk_rq_append_bio(rq, bio);
344 bio_release_pages(bio, false);
345 blk_mq_map_bio_put(bio);
349 static void bio_invalidate_vmalloc_pages(struct bio *bio)
351 #ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
352 if (bio->bi_private && !op_is_write(bio_op(bio))) {
353 unsigned long i, len = 0;
355 for (i = 0; i < bio->bi_vcnt; i++)
356 len += bio->bi_io_vec[i].bv_len;
357 invalidate_kernel_vmap_range(bio->bi_private, len);
362 static void bio_map_kern_endio(struct bio *bio)
364 bio_invalidate_vmalloc_pages(bio);
370 * bio_map_kern - map kernel address into bio
371 * @q: the struct request_queue for the bio
372 * @data: pointer to buffer to map
373 * @len: length in bytes
374 * @gfp_mask: allocation flags for bio allocation
376 * Map the kernel address into a bio suitable for io to a block
377 * device. Returns an error pointer in case of error.
379 static struct bio *bio_map_kern(struct request_queue *q, void *data,
380 unsigned int len, gfp_t gfp_mask)
382 unsigned long kaddr = (unsigned long)data;
383 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
384 unsigned long start = kaddr >> PAGE_SHIFT;
385 const int nr_pages = end - start;
386 bool is_vmalloc = is_vmalloc_addr(data);
391 bio = bio_kmalloc(nr_pages, gfp_mask);
393 return ERR_PTR(-ENOMEM);
394 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
397 flush_kernel_vmap_range(data, len);
398 bio->bi_private = data;
401 offset = offset_in_page(kaddr);
402 for (i = 0; i < nr_pages; i++) {
403 unsigned int bytes = PAGE_SIZE - offset;
412 page = virt_to_page(data);
414 page = vmalloc_to_page(data);
415 if (bio_add_pc_page(q, bio, page, bytes,
417 /* we don't support partial mappings */
420 return ERR_PTR(-EINVAL);
428 bio->bi_end_io = bio_map_kern_endio;
432 static void bio_copy_kern_endio(struct bio *bio)
439 static void bio_copy_kern_endio_read(struct bio *bio)
441 char *p = bio->bi_private;
442 struct bio_vec *bvec;
443 struct bvec_iter_all iter_all;
445 bio_for_each_segment_all(bvec, bio, iter_all) {
446 memcpy_from_bvec(p, bvec);
450 bio_copy_kern_endio(bio);
454 * bio_copy_kern - copy kernel address into bio
455 * @q: the struct request_queue for the bio
456 * @data: pointer to buffer to copy
457 * @len: length in bytes
458 * @gfp_mask: allocation flags for bio and page allocation
459 * @reading: data direction is READ
461 * copy the kernel address into a bio suitable for io to a block
462 * device. Returns an error pointer in case of error.
464 static struct bio *bio_copy_kern(struct request_queue *q, void *data,
465 unsigned int len, gfp_t gfp_mask, int reading)
467 unsigned long kaddr = (unsigned long)data;
468 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
469 unsigned long start = kaddr >> PAGE_SHIFT;
478 return ERR_PTR(-EINVAL);
480 nr_pages = end - start;
481 bio = bio_kmalloc(nr_pages, gfp_mask);
483 return ERR_PTR(-ENOMEM);
484 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
488 unsigned int bytes = PAGE_SIZE;
493 page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask);
498 memcpy(page_address(page), p, bytes);
500 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
508 bio->bi_end_io = bio_copy_kern_endio_read;
509 bio->bi_private = data;
511 bio->bi_end_io = bio_copy_kern_endio;
520 return ERR_PTR(-ENOMEM);
524 * Append a bio to a passthrough request. Only works if the bio can be merged
525 * into the request based on the driver constraints.
527 int blk_rq_append_bio(struct request *rq, struct bio *bio)
529 struct bvec_iter iter;
531 unsigned int nr_segs = 0;
533 bio_for_each_bvec(bv, bio, iter)
537 blk_rq_bio_prep(rq, bio, nr_segs);
539 if (!ll_back_merge_fn(rq, bio, nr_segs))
541 rq->biotail->bi_next = bio;
543 rq->__data_len += (bio)->bi_iter.bi_size;
544 bio_crypt_free_ctx(bio);
549 EXPORT_SYMBOL(blk_rq_append_bio);
551 /* Prepare bio for passthrough IO given ITER_BVEC iter */
552 static int blk_rq_map_user_bvec(struct request *rq, const struct iov_iter *iter)
554 struct request_queue *q = rq->q;
555 size_t nr_iter = iov_iter_count(iter);
556 size_t nr_segs = iter->nr_segs;
557 struct bio_vec *bvecs, *bvprvp = NULL;
558 struct queue_limits *lim = &q->limits;
559 unsigned int nsegs = 0, bytes = 0;
563 if (!nr_iter || (nr_iter >> SECTOR_SHIFT) > queue_max_hw_sectors(q))
565 if (nr_segs > queue_max_segments(q))
568 /* no iovecs to alloc, as we already have a BVEC iterator */
569 bio = blk_rq_map_bio_alloc(rq, 0, GFP_KERNEL);
573 bio_iov_bvec_set(bio, (struct iov_iter *)iter);
574 blk_rq_bio_prep(rq, bio, nr_segs);
576 /* loop to perform a bunch of sanity checks */
577 bvecs = (struct bio_vec *)iter->bvec;
578 for (i = 0; i < nr_segs; i++) {
579 struct bio_vec *bv = &bvecs[i];
582 * If the queue doesn't support SG gaps and adding this
583 * offset would create a gap, fallback to copy.
585 if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv->bv_offset)) {
586 blk_mq_map_bio_put(bio);
589 /* check full condition */
590 if (nsegs >= nr_segs || bytes > UINT_MAX - bv->bv_len)
592 if (bytes + bv->bv_len > nr_iter)
594 if (bv->bv_offset + bv->bv_len > PAGE_SIZE)
603 blk_mq_map_bio_put(bio);
608 * blk_rq_map_user_iov - map user data to a request, for passthrough requests
609 * @q: request queue where request should be inserted
610 * @rq: request to map data to
611 * @map_data: pointer to the rq_map_data holding pages (if necessary)
612 * @iter: iovec iterator
613 * @gfp_mask: memory allocation flags
616 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
617 * a kernel bounce buffer is used.
619 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
620 * still in process context.
622 int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
623 struct rq_map_data *map_data,
624 const struct iov_iter *iter, gfp_t gfp_mask)
626 bool copy = false, map_bvec = false;
627 unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
628 struct bio *bio = NULL;
634 else if (blk_queue_may_bounce(q))
636 else if (iov_iter_alignment(iter) & align)
638 else if (iov_iter_is_bvec(iter))
640 else if (!iter_is_iovec(iter))
642 else if (queue_virt_boundary(q))
643 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
646 ret = blk_rq_map_user_bvec(rq, iter);
649 if (ret != -EREMOTEIO)
651 /* fall back to copying the data on limits mismatches */
658 ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
660 ret = bio_map_user_iov(rq, &i, gfp_mask);
665 } while (iov_iter_count(&i));
670 blk_rq_unmap_user(bio);
675 EXPORT_SYMBOL(blk_rq_map_user_iov);
677 int blk_rq_map_user(struct request_queue *q, struct request *rq,
678 struct rq_map_data *map_data, void __user *ubuf,
679 unsigned long len, gfp_t gfp_mask)
683 int ret = import_single_range(rq_data_dir(rq), ubuf, len, &iov, &i);
685 if (unlikely(ret < 0))
688 return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
690 EXPORT_SYMBOL(blk_rq_map_user);
692 int blk_rq_map_user_io(struct request *req, struct rq_map_data *map_data,
693 void __user *ubuf, unsigned long buf_len, gfp_t gfp_mask,
694 bool vec, int iov_count, bool check_iter_count, int rw)
699 struct iovec fast_iov[UIO_FASTIOV];
700 struct iovec *iov = fast_iov;
701 struct iov_iter iter;
703 ret = import_iovec(rw, ubuf, iov_count ? iov_count : buf_len,
704 UIO_FASTIOV, &iov, &iter);
709 /* SG_IO howto says that the shorter of the two wins */
710 iov_iter_truncate(&iter, buf_len);
711 if (check_iter_count && !iov_iter_count(&iter)) {
717 ret = blk_rq_map_user_iov(req->q, req, map_data, &iter,
720 } else if (buf_len) {
721 ret = blk_rq_map_user(req->q, req, map_data, ubuf, buf_len,
726 EXPORT_SYMBOL(blk_rq_map_user_io);
729 * blk_rq_unmap_user - unmap a request with user data
730 * @bio: start of bio list
733 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
734 * supply the original rq->bio from the blk_rq_map_user() return, since
735 * the I/O completion may have changed rq->bio.
737 int blk_rq_unmap_user(struct bio *bio)
739 struct bio *next_bio;
743 if (bio->bi_private) {
744 ret2 = bio_uncopy_user(bio);
748 bio_release_pages(bio, bio_data_dir(bio) == READ);
753 blk_mq_map_bio_put(next_bio);
758 EXPORT_SYMBOL(blk_rq_unmap_user);
761 * blk_rq_map_kern - map kernel data to a request, for passthrough requests
762 * @q: request queue where request should be inserted
763 * @rq: request to fill
764 * @kbuf: the kernel buffer
765 * @len: length of user data
766 * @gfp_mask: memory allocation flags
769 * Data will be mapped directly if possible. Otherwise a bounce
770 * buffer is used. Can be called multiple times to append multiple
773 int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
774 unsigned int len, gfp_t gfp_mask)
776 int reading = rq_data_dir(rq) == READ;
777 unsigned long addr = (unsigned long) kbuf;
781 if (len > (queue_max_hw_sectors(q) << 9))
786 if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||
787 blk_queue_may_bounce(q))
788 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
790 bio = bio_map_kern(q, kbuf, len, gfp_mask);
795 bio->bi_opf &= ~REQ_OP_MASK;
796 bio->bi_opf |= req_op(rq);
798 ret = blk_rq_append_bio(rq, bio);
805 EXPORT_SYMBOL(blk_rq_map_kern);