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 bio_map_put(struct bio *bio)
236 if (bio->bi_opf & REQ_ALLOC_CACHE) {
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 unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS);
253 if (!iov_iter_count(iter))
256 if (rq->cmd_flags & REQ_POLLED) {
257 blk_opf_t opf = rq->cmd_flags | REQ_ALLOC_CACHE;
259 bio = bio_alloc_bioset(NULL, nr_vecs, opf, gfp_mask,
264 bio = bio_kmalloc(nr_vecs, gfp_mask);
267 bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq));
270 while (iov_iter_count(iter)) {
271 struct page **pages, *stack_pages[UIO_FASTIOV];
276 if (nr_vecs <= ARRAY_SIZE(stack_pages)) {
278 bytes = iov_iter_get_pages2(iter, pages, LONG_MAX,
281 bytes = iov_iter_get_pages_alloc2(iter, &pages,
284 if (unlikely(bytes <= 0)) {
285 ret = bytes ? bytes : -EFAULT;
289 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
291 if (unlikely(offs & queue_dma_alignment(rq->q)))
294 for (j = 0; j < npages; j++) {
295 struct page *page = pages[j];
296 unsigned int n = PAGE_SIZE - offs;
297 bool same_page = false;
302 if (!bio_add_hw_page(rq->q, bio, page, n, offs,
303 max_sectors, &same_page)) {
314 * release the pages we didn't map into the bio, if any
317 put_page(pages[j++]);
318 if (pages != stack_pages)
320 /* couldn't stuff something into bio? */
322 iov_iter_revert(iter, bytes);
327 ret = blk_rq_append_bio(rq, bio);
333 bio_release_pages(bio, false);
338 static void bio_invalidate_vmalloc_pages(struct bio *bio)
340 #ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
341 if (bio->bi_private && !op_is_write(bio_op(bio))) {
342 unsigned long i, len = 0;
344 for (i = 0; i < bio->bi_vcnt; i++)
345 len += bio->bi_io_vec[i].bv_len;
346 invalidate_kernel_vmap_range(bio->bi_private, len);
351 static void bio_map_kern_endio(struct bio *bio)
353 bio_invalidate_vmalloc_pages(bio);
359 * bio_map_kern - map kernel address into bio
360 * @q: the struct request_queue for the bio
361 * @data: pointer to buffer to map
362 * @len: length in bytes
363 * @gfp_mask: allocation flags for bio allocation
365 * Map the kernel address into a bio suitable for io to a block
366 * device. Returns an error pointer in case of error.
368 static struct bio *bio_map_kern(struct request_queue *q, void *data,
369 unsigned int len, gfp_t gfp_mask)
371 unsigned long kaddr = (unsigned long)data;
372 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
373 unsigned long start = kaddr >> PAGE_SHIFT;
374 const int nr_pages = end - start;
375 bool is_vmalloc = is_vmalloc_addr(data);
380 bio = bio_kmalloc(nr_pages, gfp_mask);
382 return ERR_PTR(-ENOMEM);
383 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
386 flush_kernel_vmap_range(data, len);
387 bio->bi_private = data;
390 offset = offset_in_page(kaddr);
391 for (i = 0; i < nr_pages; i++) {
392 unsigned int bytes = PAGE_SIZE - offset;
401 page = virt_to_page(data);
403 page = vmalloc_to_page(data);
404 if (bio_add_pc_page(q, bio, page, bytes,
406 /* we don't support partial mappings */
409 return ERR_PTR(-EINVAL);
417 bio->bi_end_io = bio_map_kern_endio;
421 static void bio_copy_kern_endio(struct bio *bio)
428 static void bio_copy_kern_endio_read(struct bio *bio)
430 char *p = bio->bi_private;
431 struct bio_vec *bvec;
432 struct bvec_iter_all iter_all;
434 bio_for_each_segment_all(bvec, bio, iter_all) {
435 memcpy_from_bvec(p, bvec);
439 bio_copy_kern_endio(bio);
443 * bio_copy_kern - copy kernel address into bio
444 * @q: the struct request_queue for the bio
445 * @data: pointer to buffer to copy
446 * @len: length in bytes
447 * @gfp_mask: allocation flags for bio and page allocation
448 * @reading: data direction is READ
450 * copy the kernel address into a bio suitable for io to a block
451 * device. Returns an error pointer in case of error.
453 static struct bio *bio_copy_kern(struct request_queue *q, void *data,
454 unsigned int len, gfp_t gfp_mask, int reading)
456 unsigned long kaddr = (unsigned long)data;
457 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
458 unsigned long start = kaddr >> PAGE_SHIFT;
467 return ERR_PTR(-EINVAL);
469 nr_pages = end - start;
470 bio = bio_kmalloc(nr_pages, gfp_mask);
472 return ERR_PTR(-ENOMEM);
473 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
477 unsigned int bytes = PAGE_SIZE;
482 page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask);
487 memcpy(page_address(page), p, bytes);
489 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
497 bio->bi_end_io = bio_copy_kern_endio_read;
498 bio->bi_private = data;
500 bio->bi_end_io = bio_copy_kern_endio;
509 return ERR_PTR(-ENOMEM);
513 * Append a bio to a passthrough request. Only works if the bio can be merged
514 * into the request based on the driver constraints.
516 int blk_rq_append_bio(struct request *rq, struct bio *bio)
518 struct bvec_iter iter;
520 unsigned int nr_segs = 0;
522 bio_for_each_bvec(bv, bio, iter)
526 blk_rq_bio_prep(rq, bio, nr_segs);
528 if (!ll_back_merge_fn(rq, bio, nr_segs))
530 rq->biotail->bi_next = bio;
532 rq->__data_len += (bio)->bi_iter.bi_size;
533 bio_crypt_free_ctx(bio);
538 EXPORT_SYMBOL(blk_rq_append_bio);
541 * blk_rq_map_user_iov - map user data to a request, for passthrough requests
542 * @q: request queue where request should be inserted
543 * @rq: request to map data to
544 * @map_data: pointer to the rq_map_data holding pages (if necessary)
545 * @iter: iovec iterator
546 * @gfp_mask: memory allocation flags
549 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
550 * a kernel bounce buffer is used.
552 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
553 * still in process context.
555 int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
556 struct rq_map_data *map_data,
557 const struct iov_iter *iter, gfp_t gfp_mask)
560 unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
561 struct bio *bio = NULL;
565 if (!iter_is_iovec(iter))
570 else if (blk_queue_may_bounce(q))
572 else if (iov_iter_alignment(iter) & align)
574 else if (queue_virt_boundary(q))
575 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
580 ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
582 ret = bio_map_user_iov(rq, &i, gfp_mask);
587 } while (iov_iter_count(&i));
592 blk_rq_unmap_user(bio);
597 EXPORT_SYMBOL(blk_rq_map_user_iov);
599 int blk_rq_map_user(struct request_queue *q, struct request *rq,
600 struct rq_map_data *map_data, void __user *ubuf,
601 unsigned long len, gfp_t gfp_mask)
605 int ret = import_single_range(rq_data_dir(rq), ubuf, len, &iov, &i);
607 if (unlikely(ret < 0))
610 return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
612 EXPORT_SYMBOL(blk_rq_map_user);
615 * blk_rq_unmap_user - unmap a request with user data
616 * @bio: start of bio list
619 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
620 * supply the original rq->bio from the blk_rq_map_user() return, since
621 * the I/O completion may have changed rq->bio.
623 int blk_rq_unmap_user(struct bio *bio)
625 struct bio *next_bio;
629 if (bio->bi_private) {
630 ret2 = bio_uncopy_user(bio);
634 bio_release_pages(bio, bio_data_dir(bio) == READ);
639 bio_map_put(next_bio);
644 EXPORT_SYMBOL(blk_rq_unmap_user);
647 * blk_rq_map_kern - map kernel data to a request, for passthrough requests
648 * @q: request queue where request should be inserted
649 * @rq: request to fill
650 * @kbuf: the kernel buffer
651 * @len: length of user data
652 * @gfp_mask: memory allocation flags
655 * Data will be mapped directly if possible. Otherwise a bounce
656 * buffer is used. Can be called multiple times to append multiple
659 int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
660 unsigned int len, gfp_t gfp_mask)
662 int reading = rq_data_dir(rq) == READ;
663 unsigned long addr = (unsigned long) kbuf;
667 if (len > (queue_max_hw_sectors(q) << 9))
672 if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||
673 blk_queue_may_bounce(q))
674 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
676 bio = bio_map_kern(q, kbuf, len, gfp_mask);
681 bio->bi_opf &= ~REQ_OP_MASK;
682 bio->bi_opf |= req_op(rq);
684 ret = blk_rq_append_bio(rq, bio);
691 EXPORT_SYMBOL(blk_rq_map_kern);