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;
152 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
153 if (nr_pages > BIO_MAX_PAGES)
154 nr_pages = BIO_MAX_PAGES;
157 bio = bio_kmalloc(gfp_mask, nr_pages);
160 bio->bi_opf |= req_op(rq);
163 nr_pages = 1 << 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(rq->q->bounce_gfp | 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 (map_data && map_data->from_user)) {
211 ret = bio_copy_from_iter(bio, iter);
215 if (bmd->is_our_pages)
217 iov_iter_advance(iter, bio->bi_iter.bi_size);
220 bio->bi_private = bmd;
221 if (map_data && map_data->null_mapped)
222 bmd->is_null_mapped = true;
225 ret = blk_rq_append_bio(rq, &bounce_bio);
230 * We link the bounce buffer in and could have to traverse it later, so
231 * we have to get a ref to prevent it from being freed
244 static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
247 unsigned int max_sectors = queue_max_hw_sectors(rq->q);
248 struct bio *bio, *bounce_bio;
252 if (!iov_iter_count(iter))
255 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
258 bio->bi_opf |= req_op(rq);
260 while (iov_iter_count(iter)) {
263 size_t offs, added = 0;
266 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
267 if (unlikely(bytes <= 0)) {
268 ret = bytes ? bytes : -EFAULT;
272 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
274 if (unlikely(offs & queue_dma_alignment(rq->q))) {
278 for (j = 0; j < npages; j++) {
279 struct page *page = pages[j];
280 unsigned int n = PAGE_SIZE - offs;
281 bool same_page = false;
286 if (!bio_add_hw_page(rq->q, bio, page, n, offs,
287 max_sectors, &same_page)) {
297 iov_iter_advance(iter, added);
300 * release the pages we didn't map into the bio, if any
303 put_page(pages[j++]);
305 /* couldn't stuff something into bio? */
311 * Subtle: if we end up needing to bounce a bio, it would normally
312 * disappear when its bi_end_io is run. However, we need the original
313 * bio for the unmap, so grab an extra reference to it
318 ret = blk_rq_append_bio(rq, &bounce_bio);
323 * We link the bounce buffer in and could have to traverse it
324 * later, so we have to get a ref to prevent it from being freed
332 bio_release_pages(bio, false);
338 * bio_unmap_user - unmap a bio
339 * @bio: the bio being unmapped
341 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
344 * bio_unmap_user() may sleep.
346 static void bio_unmap_user(struct bio *bio)
348 bio_release_pages(bio, bio_data_dir(bio) == READ);
353 static void bio_invalidate_vmalloc_pages(struct bio *bio)
355 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
356 if (bio->bi_private && !op_is_write(bio_op(bio))) {
357 unsigned long i, len = 0;
359 for (i = 0; i < bio->bi_vcnt; i++)
360 len += bio->bi_io_vec[i].bv_len;
361 invalidate_kernel_vmap_range(bio->bi_private, len);
366 static void bio_map_kern_endio(struct bio *bio)
368 bio_invalidate_vmalloc_pages(bio);
373 * bio_map_kern - map kernel address into bio
374 * @q: the struct request_queue for the bio
375 * @data: pointer to buffer to map
376 * @len: length in bytes
377 * @gfp_mask: allocation flags for bio allocation
379 * Map the kernel address into a bio suitable for io to a block
380 * device. Returns an error pointer in case of error.
382 static struct bio *bio_map_kern(struct request_queue *q, void *data,
383 unsigned int len, gfp_t gfp_mask)
385 unsigned long kaddr = (unsigned long)data;
386 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
387 unsigned long start = kaddr >> PAGE_SHIFT;
388 const int nr_pages = end - start;
389 bool is_vmalloc = is_vmalloc_addr(data);
394 bio = bio_kmalloc(gfp_mask, nr_pages);
396 return ERR_PTR(-ENOMEM);
399 flush_kernel_vmap_range(data, len);
400 bio->bi_private = data;
403 offset = offset_in_page(kaddr);
404 for (i = 0; i < nr_pages; i++) {
405 unsigned int bytes = PAGE_SIZE - offset;
414 page = virt_to_page(data);
416 page = vmalloc_to_page(data);
417 if (bio_add_pc_page(q, bio, page, bytes,
419 /* we don't support partial mappings */
421 return ERR_PTR(-EINVAL);
429 bio->bi_end_io = bio_map_kern_endio;
433 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(p, page_address(bvec->bv_page), bvec->bv_len);
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(gfp_mask, nr_pages);
483 return ERR_PTR(-ENOMEM);
487 unsigned int bytes = PAGE_SIZE;
492 page = alloc_page(q->bounce_gfp | gfp_mask);
497 memcpy(page_address(page), p, bytes);
499 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
507 bio->bi_end_io = bio_copy_kern_endio_read;
508 bio->bi_private = data;
510 bio->bi_end_io = bio_copy_kern_endio;
518 return ERR_PTR(-ENOMEM);
522 * Append a bio to a passthrough request. Only works if the bio can be merged
523 * into the request based on the driver constraints.
525 int blk_rq_append_bio(struct request *rq, struct bio **bio)
527 struct bio *orig_bio = *bio;
528 struct bvec_iter iter;
530 unsigned int nr_segs = 0;
532 blk_queue_bounce(rq->q, bio);
534 bio_for_each_bvec(bv, *bio, iter)
538 blk_rq_bio_prep(rq, *bio, nr_segs);
540 if (!ll_back_merge_fn(rq, *bio, nr_segs)) {
541 if (orig_bio != *bio) {
548 rq->biotail->bi_next = *bio;
550 rq->__data_len += (*bio)->bi_iter.bi_size;
551 bio_crypt_free_ctx(*bio);
556 EXPORT_SYMBOL(blk_rq_append_bio);
559 * blk_rq_map_user_iov - map user data to a request, for passthrough requests
560 * @q: request queue where request should be inserted
561 * @rq: request to map data to
562 * @map_data: pointer to the rq_map_data holding pages (if necessary)
563 * @iter: iovec iterator
564 * @gfp_mask: memory allocation flags
567 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
568 * a kernel bounce buffer is used.
570 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
571 * still in process context.
573 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
574 * before being submitted to the device, as pages mapped may be out of
575 * reach. It's the callers responsibility to make sure this happens. The
576 * original bio must be passed back in to blk_rq_unmap_user() for proper
579 int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
580 struct rq_map_data *map_data,
581 const struct iov_iter *iter, gfp_t gfp_mask)
584 unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
585 struct bio *bio = NULL;
589 if (!iter_is_iovec(iter))
594 else if (iov_iter_alignment(iter) & align)
596 else if (queue_virt_boundary(q))
597 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
602 ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
604 ret = bio_map_user_iov(rq, &i, gfp_mask);
609 } while (iov_iter_count(&i));
614 blk_rq_unmap_user(bio);
619 EXPORT_SYMBOL(blk_rq_map_user_iov);
621 int blk_rq_map_user(struct request_queue *q, struct request *rq,
622 struct rq_map_data *map_data, void __user *ubuf,
623 unsigned long len, gfp_t gfp_mask)
627 int ret = import_single_range(rq_data_dir(rq), ubuf, len, &iov, &i);
629 if (unlikely(ret < 0))
632 return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
634 EXPORT_SYMBOL(blk_rq_map_user);
637 * blk_rq_unmap_user - unmap a request with user data
638 * @bio: start of bio list
641 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
642 * supply the original rq->bio from the blk_rq_map_user() return, since
643 * the I/O completion may have changed rq->bio.
645 int blk_rq_unmap_user(struct bio *bio)
647 struct bio *mapped_bio;
652 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
653 mapped_bio = bio->bi_private;
655 if (bio->bi_private) {
656 ret2 = bio_uncopy_user(mapped_bio);
660 bio_unmap_user(mapped_bio);
670 EXPORT_SYMBOL(blk_rq_unmap_user);
673 * blk_rq_map_kern - map kernel data to a request, for passthrough requests
674 * @q: request queue where request should be inserted
675 * @rq: request to fill
676 * @kbuf: the kernel buffer
677 * @len: length of user data
678 * @gfp_mask: memory allocation flags
681 * Data will be mapped directly if possible. Otherwise a bounce
682 * buffer is used. Can be called multiple times to append multiple
685 int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
686 unsigned int len, gfp_t gfp_mask)
688 int reading = rq_data_dir(rq) == READ;
689 unsigned long addr = (unsigned long) kbuf;
690 struct bio *bio, *orig_bio;
693 if (len > (queue_max_hw_sectors(q) << 9))
698 if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf))
699 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
701 bio = bio_map_kern(q, kbuf, len, gfp_mask);
706 bio->bi_opf &= ~REQ_OP_MASK;
707 bio->bi_opf |= req_op(rq);
710 ret = blk_rq_append_bio(rq, &bio);
712 /* request is too big */
719 EXPORT_SYMBOL(blk_rq_map_kern);