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)
130 static int bio_copy_user_iov(struct request *rq, struct rq_map_data *map_data,
131 struct iov_iter *iter, gfp_t gfp_mask)
133 struct bio_map_data *bmd;
135 struct bio *bio, *bounce_bio;
138 unsigned int len = iter->count;
139 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
141 bmd = bio_alloc_map_data(iter, gfp_mask);
146 * We need to do a deep copy of the iov_iter including the iovecs.
147 * The caller provided iov might point to an on-stack or otherwise
150 bmd->is_our_pages = !map_data;
151 bmd->is_null_mapped = (map_data && map_data->null_mapped);
153 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
154 if (nr_pages > BIO_MAX_PAGES)
155 nr_pages = BIO_MAX_PAGES;
158 bio = bio_kmalloc(gfp_mask, nr_pages);
161 bio->bi_opf |= req_op(rq);
164 nr_pages = 1 << map_data->page_order;
165 i = map_data->offset / PAGE_SIZE;
168 unsigned int bytes = PAGE_SIZE;
176 if (i == map_data->nr_entries * nr_pages) {
181 page = map_data->pages[i / nr_pages];
182 page += (i % nr_pages);
186 page = alloc_page(rq->q->bounce_gfp | gfp_mask);
193 if (bio_add_pc_page(rq->q, bio, page, bytes, offset) < bytes) {
204 map_data->offset += bio->bi_iter.bi_size;
209 if ((iov_iter_rw(iter) == WRITE &&
210 (!map_data || !map_data->null_mapped)) ||
211 (map_data && map_data->from_user)) {
212 ret = bio_copy_from_iter(bio, iter);
216 if (bmd->is_our_pages)
218 iov_iter_advance(iter, bio->bi_iter.bi_size);
221 bio->bi_private = bmd;
224 ret = blk_rq_append_bio(rq, &bounce_bio);
229 * We link the bounce buffer in and could have to traverse it later, so
230 * we have to get a ref to prevent it from being freed
243 static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
246 unsigned int max_sectors = queue_max_hw_sectors(rq->q);
247 struct bio *bio, *bounce_bio;
251 if (!iov_iter_count(iter))
254 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
257 bio->bi_opf |= req_op(rq);
259 while (iov_iter_count(iter)) {
262 size_t offs, added = 0;
265 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
266 if (unlikely(bytes <= 0)) {
267 ret = bytes ? bytes : -EFAULT;
271 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
273 if (unlikely(offs & queue_dma_alignment(rq->q))) {
277 for (j = 0; j < npages; j++) {
278 struct page *page = pages[j];
279 unsigned int n = PAGE_SIZE - offs;
280 bool same_page = false;
285 if (!bio_add_hw_page(rq->q, bio, page, n, offs,
286 max_sectors, &same_page)) {
296 iov_iter_advance(iter, added);
299 * release the pages we didn't map into the bio, if any
302 put_page(pages[j++]);
304 /* couldn't stuff something into bio? */
310 * Subtle: if we end up needing to bounce a bio, it would normally
311 * disappear when its bi_end_io is run. However, we need the original
312 * bio for the unmap, so grab an extra reference to it
317 ret = blk_rq_append_bio(rq, &bounce_bio);
322 * We link the bounce buffer in and could have to traverse it
323 * later, so we have to get a ref to prevent it from being freed
331 bio_release_pages(bio, false);
337 * bio_unmap_user - unmap a bio
338 * @bio: the bio being unmapped
340 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
343 * bio_unmap_user() may sleep.
345 static void bio_unmap_user(struct bio *bio)
347 bio_release_pages(bio, bio_data_dir(bio) == READ);
352 static void bio_invalidate_vmalloc_pages(struct bio *bio)
354 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
355 if (bio->bi_private && !op_is_write(bio_op(bio))) {
356 unsigned long i, len = 0;
358 for (i = 0; i < bio->bi_vcnt; i++)
359 len += bio->bi_io_vec[i].bv_len;
360 invalidate_kernel_vmap_range(bio->bi_private, len);
365 static void bio_map_kern_endio(struct bio *bio)
367 bio_invalidate_vmalloc_pages(bio);
372 * bio_map_kern - map kernel address into bio
373 * @q: the struct request_queue for the bio
374 * @data: pointer to buffer to map
375 * @len: length in bytes
376 * @gfp_mask: allocation flags for bio allocation
378 * Map the kernel address into a bio suitable for io to a block
379 * device. Returns an error pointer in case of error.
381 static struct bio *bio_map_kern(struct request_queue *q, void *data,
382 unsigned int len, gfp_t gfp_mask)
384 unsigned long kaddr = (unsigned long)data;
385 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
386 unsigned long start = kaddr >> PAGE_SHIFT;
387 const int nr_pages = end - start;
388 bool is_vmalloc = is_vmalloc_addr(data);
393 bio = bio_kmalloc(gfp_mask, nr_pages);
395 return ERR_PTR(-ENOMEM);
398 flush_kernel_vmap_range(data, len);
399 bio->bi_private = data;
402 offset = offset_in_page(kaddr);
403 for (i = 0; i < nr_pages; i++) {
404 unsigned int bytes = PAGE_SIZE - offset;
413 page = virt_to_page(data);
415 page = vmalloc_to_page(data);
416 if (bio_add_pc_page(q, bio, page, bytes,
418 /* 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)
438 static void bio_copy_kern_endio_read(struct bio *bio)
440 char *p = bio->bi_private;
441 struct bio_vec *bvec;
442 struct bvec_iter_all iter_all;
444 bio_for_each_segment_all(bvec, bio, iter_all) {
445 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
449 bio_copy_kern_endio(bio);
453 * bio_copy_kern - copy kernel address into bio
454 * @q: the struct request_queue for the bio
455 * @data: pointer to buffer to copy
456 * @len: length in bytes
457 * @gfp_mask: allocation flags for bio and page allocation
458 * @reading: data direction is READ
460 * copy the kernel address into a bio suitable for io to a block
461 * device. Returns an error pointer in case of error.
463 static struct bio *bio_copy_kern(struct request_queue *q, void *data,
464 unsigned int len, gfp_t gfp_mask, int reading)
466 unsigned long kaddr = (unsigned long)data;
467 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
468 unsigned long start = kaddr >> PAGE_SHIFT;
477 return ERR_PTR(-EINVAL);
479 nr_pages = end - start;
480 bio = bio_kmalloc(gfp_mask, nr_pages);
482 return ERR_PTR(-ENOMEM);
486 unsigned int bytes = PAGE_SIZE;
491 page = alloc_page(q->bounce_gfp | gfp_mask);
496 memcpy(page_address(page), p, bytes);
498 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
506 bio->bi_end_io = bio_copy_kern_endio_read;
507 bio->bi_private = data;
509 bio->bi_end_io = bio_copy_kern_endio;
517 return ERR_PTR(-ENOMEM);
521 * Append a bio to a passthrough request. Only works if the bio can be merged
522 * into the request based on the driver constraints.
524 int blk_rq_append_bio(struct request *rq, struct bio **bio)
526 struct bio *orig_bio = *bio;
527 struct bvec_iter iter;
529 unsigned int nr_segs = 0;
531 blk_queue_bounce(rq->q, bio);
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)) {
540 if (orig_bio != *bio) {
547 rq->biotail->bi_next = *bio;
549 rq->__data_len += (*bio)->bi_iter.bi_size;
550 bio_crypt_free_ctx(*bio);
555 EXPORT_SYMBOL(blk_rq_append_bio);
558 * blk_rq_map_user_iov - map user data to a request, for passthrough requests
559 * @q: request queue where request should be inserted
560 * @rq: request to map data to
561 * @map_data: pointer to the rq_map_data holding pages (if necessary)
562 * @iter: iovec iterator
563 * @gfp_mask: memory allocation flags
566 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
567 * a kernel bounce buffer is used.
569 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
570 * still in process context.
572 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
573 * before being submitted to the device, as pages mapped may be out of
574 * reach. It's the callers responsibility to make sure this happens. The
575 * original bio must be passed back in to blk_rq_unmap_user() for proper
578 int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
579 struct rq_map_data *map_data,
580 const struct iov_iter *iter, gfp_t gfp_mask)
583 unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
584 struct bio *bio = NULL;
588 if (!iter_is_iovec(iter))
593 else if (iov_iter_alignment(iter) & align)
595 else if (queue_virt_boundary(q))
596 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
601 ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
603 ret = bio_map_user_iov(rq, &i, gfp_mask);
608 } while (iov_iter_count(&i));
613 blk_rq_unmap_user(bio);
618 EXPORT_SYMBOL(blk_rq_map_user_iov);
620 int blk_rq_map_user(struct request_queue *q, struct request *rq,
621 struct rq_map_data *map_data, void __user *ubuf,
622 unsigned long len, gfp_t gfp_mask)
626 int ret = import_single_range(rq_data_dir(rq), ubuf, len, &iov, &i);
628 if (unlikely(ret < 0))
631 return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
633 EXPORT_SYMBOL(blk_rq_map_user);
636 * blk_rq_unmap_user - unmap a request with user data
637 * @bio: start of bio list
640 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
641 * supply the original rq->bio from the blk_rq_map_user() return, since
642 * the I/O completion may have changed rq->bio.
644 int blk_rq_unmap_user(struct bio *bio)
646 struct bio *mapped_bio;
651 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
652 mapped_bio = bio->bi_private;
654 if (bio->bi_private) {
655 ret2 = bio_uncopy_user(mapped_bio);
659 bio_unmap_user(mapped_bio);
669 EXPORT_SYMBOL(blk_rq_unmap_user);
672 * blk_rq_map_kern - map kernel data to a request, for passthrough requests
673 * @q: request queue where request should be inserted
674 * @rq: request to fill
675 * @kbuf: the kernel buffer
676 * @len: length of user data
677 * @gfp_mask: memory allocation flags
680 * Data will be mapped directly if possible. Otherwise a bounce
681 * buffer is used. Can be called multiple times to append multiple
684 int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
685 unsigned int len, gfp_t gfp_mask)
687 int reading = rq_data_dir(rq) == READ;
688 unsigned long addr = (unsigned long) kbuf;
689 struct bio *bio, *orig_bio;
692 if (len > (queue_max_hw_sectors(q) << 9))
697 if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf))
698 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
700 bio = bio_map_kern(q, kbuf, len, gfp_mask);
705 bio->bi_opf &= ~REQ_OP_MASK;
706 bio->bi_opf |= req_op(rq);
709 ret = blk_rq_append_bio(rq, &bio);
711 /* request is too big */
718 EXPORT_SYMBOL(blk_rq_map_kern);