Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to mapping data to requests
4 */
5#include <linux/kernel.h>
6#include <linux/sched/task_stack.h>
7#include <linux/module.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/uio.h>
11
12#include "blk.h"
13
14struct bio_map_data {
15 bool is_our_pages : 1;
16 bool is_null_mapped : 1;
17 struct iov_iter iter;
18 struct iovec iov[];
19};
20
21static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
22 gfp_t gfp_mask)
23{
24 struct bio_map_data *bmd;
25
26 if (data->nr_segs > UIO_MAXIOV)
27 return NULL;
28
29 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask);
30 if (!bmd)
31 return NULL;
32 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
33 bmd->iter = *data;
34 bmd->iter.iov = bmd->iov;
35 return bmd;
36}
37
38/**
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
42 *
43 * Copy all pages from iov_iter to bio.
44 * Returns 0 on success, or error on failure.
45 */
46static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
47{
48 struct bio_vec *bvec;
49 struct bvec_iter_all iter_all;
50
51 bio_for_each_segment_all(bvec, bio, iter_all) {
52 ssize_t ret;
53
54 ret = copy_page_from_iter(bvec->bv_page,
55 bvec->bv_offset,
56 bvec->bv_len,
57 iter);
58
59 if (!iov_iter_count(iter))
60 break;
61
62 if (ret < bvec->bv_len)
63 return -EFAULT;
64 }
65
66 return 0;
67}
68
69/**
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
73 *
74 * Copy all pages from bio to iov_iter.
75 * Returns 0 on success, or error on failure.
76 */
77static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
78{
79 struct bio_vec *bvec;
80 struct bvec_iter_all iter_all;
81
82 bio_for_each_segment_all(bvec, bio, iter_all) {
83 ssize_t ret;
84
85 ret = copy_page_to_iter(bvec->bv_page,
86 bvec->bv_offset,
87 bvec->bv_len,
88 &iter);
89
90 if (!iov_iter_count(&iter))
91 break;
92
93 if (ret < bvec->bv_len)
94 return -EFAULT;
95 }
96
97 return 0;
98}
99
100/**
101 * bio_uncopy_user - finish previously mapped bio
102 * @bio: bio being terminated
103 *
104 * Free pages allocated from bio_copy_user_iov() and write back data
105 * to user space in case of a read.
106 */
107static int bio_uncopy_user(struct bio *bio)
108{
109 struct bio_map_data *bmd = bio->bi_private;
110 int ret = 0;
111
112 if (!bmd->is_null_mapped) {
113 /*
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.
117 */
118 if (!current->mm)
119 ret = -EINTR;
120 else if (bio_data_dir(bio) == READ)
121 ret = bio_copy_to_iter(bio, bmd->iter);
122 if (bmd->is_our_pages)
123 bio_free_pages(bio);
124 }
125 kfree(bmd);
126 return ret;
127}
128
129static int bio_copy_user_iov(struct request *rq, struct rq_map_data *map_data,
130 struct iov_iter *iter, gfp_t gfp_mask)
131{
132 struct bio_map_data *bmd;
133 struct page *page;
134 struct bio *bio;
135 int i = 0, ret;
136 int nr_pages;
137 unsigned int len = iter->count;
138 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
139
140 bmd = bio_alloc_map_data(iter, gfp_mask);
141 if (!bmd)
142 return -ENOMEM;
143
144 /*
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
147 * shortlived one.
148 */
149 bmd->is_our_pages = !map_data;
150 bmd->is_null_mapped = (map_data && map_data->null_mapped);
151
152 nr_pages = bio_max_segs(DIV_ROUND_UP(offset + len, PAGE_SIZE));
153
154 ret = -ENOMEM;
155 bio = bio_kmalloc(nr_pages, gfp_mask);
156 if (!bio)
157 goto out_bmd;
158 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, req_op(rq));
159
160 if (map_data) {
161 nr_pages = 1U << map_data->page_order;
162 i = map_data->offset / PAGE_SIZE;
163 }
164 while (len) {
165 unsigned int bytes = PAGE_SIZE;
166
167 bytes -= offset;
168
169 if (bytes > len)
170 bytes = len;
171
172 if (map_data) {
173 if (i == map_data->nr_entries * nr_pages) {
174 ret = -ENOMEM;
175 goto cleanup;
176 }
177
178 page = map_data->pages[i / nr_pages];
179 page += (i % nr_pages);
180
181 i++;
182 } else {
183 page = alloc_page(GFP_NOIO | gfp_mask);
184 if (!page) {
185 ret = -ENOMEM;
186 goto cleanup;
187 }
188 }
189
190 if (bio_add_pc_page(rq->q, bio, page, bytes, offset) < bytes) {
191 if (!map_data)
192 __free_page(page);
193 break;
194 }
195
196 len -= bytes;
197 offset = 0;
198 }
199
200 if (map_data)
201 map_data->offset += bio->bi_iter.bi_size;
202
203 /*
204 * success
205 */
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);
210 if (ret)
211 goto cleanup;
212 } else {
213 if (bmd->is_our_pages)
214 zero_fill_bio(bio);
215 iov_iter_advance(iter, bio->bi_iter.bi_size);
216 }
217
218 bio->bi_private = bmd;
219
220 ret = blk_rq_append_bio(rq, bio);
221 if (ret)
222 goto cleanup;
223 return 0;
224cleanup:
225 if (!map_data)
226 bio_free_pages(bio);
227 bio_uninit(bio);
228 kfree(bio);
229out_bmd:
230 kfree(bmd);
231 return ret;
232}
233
234static void blk_mq_map_bio_put(struct bio *bio)
235{
236 if (bio->bi_opf & REQ_ALLOC_CACHE) {
237 bio_put(bio);
238 } else {
239 bio_uninit(bio);
240 kfree(bio);
241 }
242}
243
244static struct bio *blk_rq_map_bio_alloc(struct request *rq,
245 unsigned int nr_vecs, gfp_t gfp_mask)
246{
247 struct bio *bio;
248
249 if (rq->cmd_flags & REQ_POLLED) {
250 blk_opf_t opf = rq->cmd_flags | REQ_ALLOC_CACHE;
251
252 bio = bio_alloc_bioset(NULL, nr_vecs, opf, gfp_mask,
253 &fs_bio_set);
254 if (!bio)
255 return NULL;
256 } else {
257 bio = bio_kmalloc(nr_vecs, gfp_mask);
258 if (!bio)
259 return NULL;
260 bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq));
261 }
262 return bio;
263}
264
265static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
266 gfp_t gfp_mask)
267{
268 unsigned int max_sectors = queue_max_hw_sectors(rq->q);
269 unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS);
270 unsigned int gup_flags = 0;
271 struct bio *bio;
272 int ret;
273 int j;
274
275 if (!iov_iter_count(iter))
276 return -EINVAL;
277
278 bio = blk_rq_map_bio_alloc(rq, nr_vecs, gfp_mask);
279 if (bio == NULL)
280 return -ENOMEM;
281
282 if (blk_queue_pci_p2pdma(rq->q))
283 gup_flags |= FOLL_PCI_P2PDMA;
284
285 while (iov_iter_count(iter)) {
286 struct page **pages, *stack_pages[UIO_FASTIOV];
287 ssize_t bytes;
288 size_t offs;
289 int npages;
290
291 if (nr_vecs <= ARRAY_SIZE(stack_pages)) {
292 pages = stack_pages;
293 bytes = iov_iter_get_pages(iter, pages, LONG_MAX,
294 nr_vecs, &offs, gup_flags);
295 } else {
296 bytes = iov_iter_get_pages_alloc(iter, &pages,
297 LONG_MAX, &offs, gup_flags);
298 }
299 if (unlikely(bytes <= 0)) {
300 ret = bytes ? bytes : -EFAULT;
301 goto out_unmap;
302 }
303
304 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
305
306 if (unlikely(offs & queue_dma_alignment(rq->q)))
307 j = 0;
308 else {
309 for (j = 0; j < npages; j++) {
310 struct page *page = pages[j];
311 unsigned int n = PAGE_SIZE - offs;
312 bool same_page = false;
313
314 if (n > bytes)
315 n = bytes;
316
317 if (!bio_add_hw_page(rq->q, bio, page, n, offs,
318 max_sectors, &same_page)) {
319 if (same_page)
320 put_page(page);
321 break;
322 }
323
324 bytes -= n;
325 offs = 0;
326 }
327 }
328 /*
329 * release the pages we didn't map into the bio, if any
330 */
331 while (j < npages)
332 put_page(pages[j++]);
333 if (pages != stack_pages)
334 kvfree(pages);
335 /* couldn't stuff something into bio? */
336 if (bytes) {
337 iov_iter_revert(iter, bytes);
338 break;
339 }
340 }
341
342 ret = blk_rq_append_bio(rq, bio);
343 if (ret)
344 goto out_unmap;
345 return 0;
346
347 out_unmap:
348 bio_release_pages(bio, false);
349 blk_mq_map_bio_put(bio);
350 return ret;
351}
352
353static void bio_invalidate_vmalloc_pages(struct bio *bio)
354{
355#ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
356 if (bio->bi_private && !op_is_write(bio_op(bio))) {
357 unsigned long i, len = 0;
358
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);
362 }
363#endif
364}
365
366static void bio_map_kern_endio(struct bio *bio)
367{
368 bio_invalidate_vmalloc_pages(bio);
369 bio_uninit(bio);
370 kfree(bio);
371}
372
373/**
374 * bio_map_kern - map kernel address into bio
375 * @q: the struct request_queue for the bio
376 * @data: pointer to buffer to map
377 * @len: length in bytes
378 * @gfp_mask: allocation flags for bio allocation
379 *
380 * Map the kernel address into a bio suitable for io to a block
381 * device. Returns an error pointer in case of error.
382 */
383static struct bio *bio_map_kern(struct request_queue *q, void *data,
384 unsigned int len, gfp_t gfp_mask)
385{
386 unsigned long kaddr = (unsigned long)data;
387 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
388 unsigned long start = kaddr >> PAGE_SHIFT;
389 const int nr_pages = end - start;
390 bool is_vmalloc = is_vmalloc_addr(data);
391 struct page *page;
392 int offset, i;
393 struct bio *bio;
394
395 bio = bio_kmalloc(nr_pages, gfp_mask);
396 if (!bio)
397 return ERR_PTR(-ENOMEM);
398 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
399
400 if (is_vmalloc) {
401 flush_kernel_vmap_range(data, len);
402 bio->bi_private = data;
403 }
404
405 offset = offset_in_page(kaddr);
406 for (i = 0; i < nr_pages; i++) {
407 unsigned int bytes = PAGE_SIZE - offset;
408
409 if (len <= 0)
410 break;
411
412 if (bytes > len)
413 bytes = len;
414
415 if (!is_vmalloc)
416 page = virt_to_page(data);
417 else
418 page = vmalloc_to_page(data);
419 if (bio_add_pc_page(q, bio, page, bytes,
420 offset) < bytes) {
421 /* we don't support partial mappings */
422 bio_uninit(bio);
423 kfree(bio);
424 return ERR_PTR(-EINVAL);
425 }
426
427 data += bytes;
428 len -= bytes;
429 offset = 0;
430 }
431
432 bio->bi_end_io = bio_map_kern_endio;
433 return bio;
434}
435
436static void bio_copy_kern_endio(struct bio *bio)
437{
438 bio_free_pages(bio);
439 bio_uninit(bio);
440 kfree(bio);
441}
442
443static void bio_copy_kern_endio_read(struct bio *bio)
444{
445 char *p = bio->bi_private;
446 struct bio_vec *bvec;
447 struct bvec_iter_all iter_all;
448
449 bio_for_each_segment_all(bvec, bio, iter_all) {
450 memcpy_from_bvec(p, bvec);
451 p += bvec->bv_len;
452 }
453
454 bio_copy_kern_endio(bio);
455}
456
457/**
458 * bio_copy_kern - copy kernel address into bio
459 * @q: the struct request_queue for the bio
460 * @data: pointer to buffer to copy
461 * @len: length in bytes
462 * @gfp_mask: allocation flags for bio and page allocation
463 * @reading: data direction is READ
464 *
465 * copy the kernel address into a bio suitable for io to a block
466 * device. Returns an error pointer in case of error.
467 */
468static struct bio *bio_copy_kern(struct request_queue *q, void *data,
469 unsigned int len, gfp_t gfp_mask, int reading)
470{
471 unsigned long kaddr = (unsigned long)data;
472 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
473 unsigned long start = kaddr >> PAGE_SHIFT;
474 struct bio *bio;
475 void *p = data;
476 int nr_pages = 0;
477
478 /*
479 * Overflow, abort
480 */
481 if (end < start)
482 return ERR_PTR(-EINVAL);
483
484 nr_pages = end - start;
485 bio = bio_kmalloc(nr_pages, gfp_mask);
486 if (!bio)
487 return ERR_PTR(-ENOMEM);
488 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
489
490 while (len) {
491 struct page *page;
492 unsigned int bytes = PAGE_SIZE;
493
494 if (bytes > len)
495 bytes = len;
496
497 page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask);
498 if (!page)
499 goto cleanup;
500
501 if (!reading)
502 memcpy(page_address(page), p, bytes);
503
504 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
505 break;
506
507 len -= bytes;
508 p += bytes;
509 }
510
511 if (reading) {
512 bio->bi_end_io = bio_copy_kern_endio_read;
513 bio->bi_private = data;
514 } else {
515 bio->bi_end_io = bio_copy_kern_endio;
516 }
517
518 return bio;
519
520cleanup:
521 bio_free_pages(bio);
522 bio_uninit(bio);
523 kfree(bio);
524 return ERR_PTR(-ENOMEM);
525}
526
527/*
528 * Append a bio to a passthrough request. Only works if the bio can be merged
529 * into the request based on the driver constraints.
530 */
531int blk_rq_append_bio(struct request *rq, struct bio *bio)
532{
533 struct bvec_iter iter;
534 struct bio_vec bv;
535 unsigned int nr_segs = 0;
536
537 bio_for_each_bvec(bv, bio, iter)
538 nr_segs++;
539
540 if (!rq->bio) {
541 blk_rq_bio_prep(rq, bio, nr_segs);
542 } else {
543 if (!ll_back_merge_fn(rq, bio, nr_segs))
544 return -EINVAL;
545 rq->biotail->bi_next = bio;
546 rq->biotail = bio;
547 rq->__data_len += (bio)->bi_iter.bi_size;
548 bio_crypt_free_ctx(bio);
549 }
550
551 return 0;
552}
553EXPORT_SYMBOL(blk_rq_append_bio);
554
555/* Prepare bio for passthrough IO given ITER_BVEC iter */
556static int blk_rq_map_user_bvec(struct request *rq, const struct iov_iter *iter)
557{
558 struct request_queue *q = rq->q;
559 size_t nr_iter = iov_iter_count(iter);
560 size_t nr_segs = iter->nr_segs;
561 struct bio_vec *bvecs, *bvprvp = NULL;
562 const struct queue_limits *lim = &q->limits;
563 unsigned int nsegs = 0, bytes = 0;
564 struct bio *bio;
565 size_t i;
566
567 if (!nr_iter || (nr_iter >> SECTOR_SHIFT) > queue_max_hw_sectors(q))
568 return -EINVAL;
569 if (nr_segs > queue_max_segments(q))
570 return -EINVAL;
571
572 /* no iovecs to alloc, as we already have a BVEC iterator */
573 bio = blk_rq_map_bio_alloc(rq, 0, GFP_KERNEL);
574 if (bio == NULL)
575 return -ENOMEM;
576
577 bio_iov_bvec_set(bio, (struct iov_iter *)iter);
578 blk_rq_bio_prep(rq, bio, nr_segs);
579
580 /* loop to perform a bunch of sanity checks */
581 bvecs = (struct bio_vec *)iter->bvec;
582 for (i = 0; i < nr_segs; i++) {
583 struct bio_vec *bv = &bvecs[i];
584
585 /*
586 * If the queue doesn't support SG gaps and adding this
587 * offset would create a gap, fallback to copy.
588 */
589 if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv->bv_offset)) {
590 blk_mq_map_bio_put(bio);
591 return -EREMOTEIO;
592 }
593 /* check full condition */
594 if (nsegs >= nr_segs || bytes > UINT_MAX - bv->bv_len)
595 goto put_bio;
596 if (bytes + bv->bv_len > nr_iter)
597 goto put_bio;
598 if (bv->bv_offset + bv->bv_len > PAGE_SIZE)
599 goto put_bio;
600
601 nsegs++;
602 bytes += bv->bv_len;
603 bvprvp = bv;
604 }
605 return 0;
606put_bio:
607 blk_mq_map_bio_put(bio);
608 return -EINVAL;
609}
610
611/**
612 * blk_rq_map_user_iov - map user data to a request, for passthrough requests
613 * @q: request queue where request should be inserted
614 * @rq: request to map data to
615 * @map_data: pointer to the rq_map_data holding pages (if necessary)
616 * @iter: iovec iterator
617 * @gfp_mask: memory allocation flags
618 *
619 * Description:
620 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
621 * a kernel bounce buffer is used.
622 *
623 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
624 * still in process context.
625 */
626int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
627 struct rq_map_data *map_data,
628 const struct iov_iter *iter, gfp_t gfp_mask)
629{
630 bool copy = false, map_bvec = false;
631 unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
632 struct bio *bio = NULL;
633 struct iov_iter i;
634 int ret = -EINVAL;
635
636 if (map_data)
637 copy = true;
638 else if (blk_queue_may_bounce(q))
639 copy = true;
640 else if (iov_iter_alignment(iter) & align)
641 copy = true;
642 else if (iov_iter_is_bvec(iter))
643 map_bvec = true;
644 else if (!iter_is_iovec(iter))
645 copy = true;
646 else if (queue_virt_boundary(q))
647 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
648
649 if (map_bvec) {
650 ret = blk_rq_map_user_bvec(rq, iter);
651 if (!ret)
652 return 0;
653 if (ret != -EREMOTEIO)
654 goto fail;
655 /* fall back to copying the data on limits mismatches */
656 copy = true;
657 }
658
659 i = *iter;
660 do {
661 if (copy)
662 ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
663 else
664 ret = bio_map_user_iov(rq, &i, gfp_mask);
665 if (ret)
666 goto unmap_rq;
667 if (!bio)
668 bio = rq->bio;
669 } while (iov_iter_count(&i));
670
671 return 0;
672
673unmap_rq:
674 blk_rq_unmap_user(bio);
675fail:
676 rq->bio = NULL;
677 return ret;
678}
679EXPORT_SYMBOL(blk_rq_map_user_iov);
680
681int blk_rq_map_user(struct request_queue *q, struct request *rq,
682 struct rq_map_data *map_data, void __user *ubuf,
683 unsigned long len, gfp_t gfp_mask)
684{
685 struct iovec iov;
686 struct iov_iter i;
687 int ret = import_single_range(rq_data_dir(rq), ubuf, len, &iov, &i);
688
689 if (unlikely(ret < 0))
690 return ret;
691
692 return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
693}
694EXPORT_SYMBOL(blk_rq_map_user);
695
696int blk_rq_map_user_io(struct request *req, struct rq_map_data *map_data,
697 void __user *ubuf, unsigned long buf_len, gfp_t gfp_mask,
698 bool vec, int iov_count, bool check_iter_count, int rw)
699{
700 int ret = 0;
701
702 if (vec) {
703 struct iovec fast_iov[UIO_FASTIOV];
704 struct iovec *iov = fast_iov;
705 struct iov_iter iter;
706
707 ret = import_iovec(rw, ubuf, iov_count ? iov_count : buf_len,
708 UIO_FASTIOV, &iov, &iter);
709 if (ret < 0)
710 return ret;
711
712 if (iov_count) {
713 /* SG_IO howto says that the shorter of the two wins */
714 iov_iter_truncate(&iter, buf_len);
715 if (check_iter_count && !iov_iter_count(&iter)) {
716 kfree(iov);
717 return -EINVAL;
718 }
719 }
720
721 ret = blk_rq_map_user_iov(req->q, req, map_data, &iter,
722 gfp_mask);
723 kfree(iov);
724 } else if (buf_len) {
725 ret = blk_rq_map_user(req->q, req, map_data, ubuf, buf_len,
726 gfp_mask);
727 }
728 return ret;
729}
730EXPORT_SYMBOL(blk_rq_map_user_io);
731
732/**
733 * blk_rq_unmap_user - unmap a request with user data
734 * @bio: start of bio list
735 *
736 * Description:
737 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
738 * supply the original rq->bio from the blk_rq_map_user() return, since
739 * the I/O completion may have changed rq->bio.
740 */
741int blk_rq_unmap_user(struct bio *bio)
742{
743 struct bio *next_bio;
744 int ret = 0, ret2;
745
746 while (bio) {
747 if (bio->bi_private) {
748 ret2 = bio_uncopy_user(bio);
749 if (ret2 && !ret)
750 ret = ret2;
751 } else {
752 bio_release_pages(bio, bio_data_dir(bio) == READ);
753 }
754
755 next_bio = bio;
756 bio = bio->bi_next;
757 blk_mq_map_bio_put(next_bio);
758 }
759
760 return ret;
761}
762EXPORT_SYMBOL(blk_rq_unmap_user);
763
764/**
765 * blk_rq_map_kern - map kernel data to a request, for passthrough requests
766 * @q: request queue where request should be inserted
767 * @rq: request to fill
768 * @kbuf: the kernel buffer
769 * @len: length of user data
770 * @gfp_mask: memory allocation flags
771 *
772 * Description:
773 * Data will be mapped directly if possible. Otherwise a bounce
774 * buffer is used. Can be called multiple times to append multiple
775 * buffers.
776 */
777int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
778 unsigned int len, gfp_t gfp_mask)
779{
780 int reading = rq_data_dir(rq) == READ;
781 unsigned long addr = (unsigned long) kbuf;
782 struct bio *bio;
783 int ret;
784
785 if (len > (queue_max_hw_sectors(q) << 9))
786 return -EINVAL;
787 if (!len || !kbuf)
788 return -EINVAL;
789
790 if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||
791 blk_queue_may_bounce(q))
792 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
793 else
794 bio = bio_map_kern(q, kbuf, len, gfp_mask);
795
796 if (IS_ERR(bio))
797 return PTR_ERR(bio);
798
799 bio->bi_opf &= ~REQ_OP_MASK;
800 bio->bi_opf |= req_op(rq);
801
802 ret = blk_rq_append_bio(rq, bio);
803 if (unlikely(ret)) {
804 bio_uninit(bio);
805 kfree(bio);
806 }
807 return ret;
808}
809EXPORT_SYMBOL(blk_rq_map_kern);
1/*
2 * Functions related to mapping data to requests
3 */
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/bio.h>
7#include <linux/blkdev.h>
8#include <scsi/sg.h> /* for struct sg_iovec */
9
10#include "blk.h"
11
12int blk_rq_append_bio(struct request_queue *q, struct request *rq,
13 struct bio *bio)
14{
15 if (!rq->bio)
16 blk_rq_bio_prep(q, rq, bio);
17 else if (!ll_back_merge_fn(q, rq, bio))
18 return -EINVAL;
19 else {
20 rq->biotail->bi_next = bio;
21 rq->biotail = bio;
22
23 rq->__data_len += bio->bi_iter.bi_size;
24 }
25 return 0;
26}
27
28static int __blk_rq_unmap_user(struct bio *bio)
29{
30 int ret = 0;
31
32 if (bio) {
33 if (bio_flagged(bio, BIO_USER_MAPPED))
34 bio_unmap_user(bio);
35 else
36 ret = bio_uncopy_user(bio);
37 }
38
39 return ret;
40}
41
42static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
43 struct rq_map_data *map_data, void __user *ubuf,
44 unsigned int len, gfp_t gfp_mask)
45{
46 unsigned long uaddr;
47 struct bio *bio, *orig_bio;
48 int reading, ret;
49
50 reading = rq_data_dir(rq) == READ;
51
52 /*
53 * if alignment requirement is satisfied, map in user pages for
54 * direct dma. else, set up kernel bounce buffers
55 */
56 uaddr = (unsigned long) ubuf;
57 if (blk_rq_aligned(q, uaddr, len) && !map_data)
58 bio = bio_map_user(q, NULL, uaddr, len, reading, gfp_mask);
59 else
60 bio = bio_copy_user(q, map_data, uaddr, len, reading, gfp_mask);
61
62 if (IS_ERR(bio))
63 return PTR_ERR(bio);
64
65 if (map_data && map_data->null_mapped)
66 bio->bi_flags |= (1 << BIO_NULL_MAPPED);
67
68 orig_bio = bio;
69 blk_queue_bounce(q, &bio);
70
71 /*
72 * We link the bounce buffer in and could have to traverse it
73 * later so we have to get a ref to prevent it from being freed
74 */
75 bio_get(bio);
76
77 ret = blk_rq_append_bio(q, rq, bio);
78 if (!ret)
79 return bio->bi_iter.bi_size;
80
81 /* if it was boucned we must call the end io function */
82 bio_endio(bio, 0);
83 __blk_rq_unmap_user(orig_bio);
84 bio_put(bio);
85 return ret;
86}
87
88/**
89 * blk_rq_map_user - map user data to a request, for REQ_TYPE_BLOCK_PC usage
90 * @q: request queue where request should be inserted
91 * @rq: request structure to fill
92 * @map_data: pointer to the rq_map_data holding pages (if necessary)
93 * @ubuf: the user buffer
94 * @len: length of user data
95 * @gfp_mask: memory allocation flags
96 *
97 * Description:
98 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
99 * a kernel bounce buffer is used.
100 *
101 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
102 * still in process context.
103 *
104 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
105 * before being submitted to the device, as pages mapped may be out of
106 * reach. It's the callers responsibility to make sure this happens. The
107 * original bio must be passed back in to blk_rq_unmap_user() for proper
108 * unmapping.
109 */
110int blk_rq_map_user(struct request_queue *q, struct request *rq,
111 struct rq_map_data *map_data, void __user *ubuf,
112 unsigned long len, gfp_t gfp_mask)
113{
114 unsigned long bytes_read = 0;
115 struct bio *bio = NULL;
116 int ret;
117
118 if (len > (queue_max_hw_sectors(q) << 9))
119 return -EINVAL;
120 if (!len)
121 return -EINVAL;
122
123 if (!ubuf && (!map_data || !map_data->null_mapped))
124 return -EINVAL;
125
126 while (bytes_read != len) {
127 unsigned long map_len, end, start;
128
129 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
130 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
131 >> PAGE_SHIFT;
132 start = (unsigned long)ubuf >> PAGE_SHIFT;
133
134 /*
135 * A bad offset could cause us to require BIO_MAX_PAGES + 1
136 * pages. If this happens we just lower the requested
137 * mapping len by a page so that we can fit
138 */
139 if (end - start > BIO_MAX_PAGES)
140 map_len -= PAGE_SIZE;
141
142 ret = __blk_rq_map_user(q, rq, map_data, ubuf, map_len,
143 gfp_mask);
144 if (ret < 0)
145 goto unmap_rq;
146 if (!bio)
147 bio = rq->bio;
148 bytes_read += ret;
149 ubuf += ret;
150
151 if (map_data)
152 map_data->offset += ret;
153 }
154
155 if (!bio_flagged(bio, BIO_USER_MAPPED))
156 rq->cmd_flags |= REQ_COPY_USER;
157
158 rq->buffer = NULL;
159 return 0;
160unmap_rq:
161 blk_rq_unmap_user(bio);
162 rq->bio = NULL;
163 return ret;
164}
165EXPORT_SYMBOL(blk_rq_map_user);
166
167/**
168 * blk_rq_map_user_iov - map user data to a request, for REQ_TYPE_BLOCK_PC usage
169 * @q: request queue where request should be inserted
170 * @rq: request to map data to
171 * @map_data: pointer to the rq_map_data holding pages (if necessary)
172 * @iov: pointer to the iovec
173 * @iov_count: number of elements in the iovec
174 * @len: I/O byte count
175 * @gfp_mask: memory allocation flags
176 *
177 * Description:
178 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
179 * a kernel bounce buffer is used.
180 *
181 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
182 * still in process context.
183 *
184 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
185 * before being submitted to the device, as pages mapped may be out of
186 * reach. It's the callers responsibility to make sure this happens. The
187 * original bio must be passed back in to blk_rq_unmap_user() for proper
188 * unmapping.
189 */
190int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
191 struct rq_map_data *map_data, const struct sg_iovec *iov,
192 int iov_count, unsigned int len, gfp_t gfp_mask)
193{
194 struct bio *bio;
195 int i, read = rq_data_dir(rq) == READ;
196 int unaligned = 0;
197
198 if (!iov || iov_count <= 0)
199 return -EINVAL;
200
201 for (i = 0; i < iov_count; i++) {
202 unsigned long uaddr = (unsigned long)iov[i].iov_base;
203
204 if (!iov[i].iov_len)
205 return -EINVAL;
206
207 /*
208 * Keep going so we check length of all segments
209 */
210 if (uaddr & queue_dma_alignment(q))
211 unaligned = 1;
212 }
213
214 if (unaligned || (q->dma_pad_mask & len) || map_data)
215 bio = bio_copy_user_iov(q, map_data, iov, iov_count, read,
216 gfp_mask);
217 else
218 bio = bio_map_user_iov(q, NULL, iov, iov_count, read, gfp_mask);
219
220 if (IS_ERR(bio))
221 return PTR_ERR(bio);
222
223 if (bio->bi_iter.bi_size != len) {
224 /*
225 * Grab an extra reference to this bio, as bio_unmap_user()
226 * expects to be able to drop it twice as it happens on the
227 * normal IO completion path
228 */
229 bio_get(bio);
230 bio_endio(bio, 0);
231 __blk_rq_unmap_user(bio);
232 return -EINVAL;
233 }
234
235 if (!bio_flagged(bio, BIO_USER_MAPPED))
236 rq->cmd_flags |= REQ_COPY_USER;
237
238 blk_queue_bounce(q, &bio);
239 bio_get(bio);
240 blk_rq_bio_prep(q, rq, bio);
241 rq->buffer = NULL;
242 return 0;
243}
244EXPORT_SYMBOL(blk_rq_map_user_iov);
245
246/**
247 * blk_rq_unmap_user - unmap a request with user data
248 * @bio: start of bio list
249 *
250 * Description:
251 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
252 * supply the original rq->bio from the blk_rq_map_user() return, since
253 * the I/O completion may have changed rq->bio.
254 */
255int blk_rq_unmap_user(struct bio *bio)
256{
257 struct bio *mapped_bio;
258 int ret = 0, ret2;
259
260 while (bio) {
261 mapped_bio = bio;
262 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
263 mapped_bio = bio->bi_private;
264
265 ret2 = __blk_rq_unmap_user(mapped_bio);
266 if (ret2 && !ret)
267 ret = ret2;
268
269 mapped_bio = bio;
270 bio = bio->bi_next;
271 bio_put(mapped_bio);
272 }
273
274 return ret;
275}
276EXPORT_SYMBOL(blk_rq_unmap_user);
277
278/**
279 * blk_rq_map_kern - map kernel data to a request, for REQ_TYPE_BLOCK_PC usage
280 * @q: request queue where request should be inserted
281 * @rq: request to fill
282 * @kbuf: the kernel buffer
283 * @len: length of user data
284 * @gfp_mask: memory allocation flags
285 *
286 * Description:
287 * Data will be mapped directly if possible. Otherwise a bounce
288 * buffer is used. Can be called multiple times to append multiple
289 * buffers.
290 */
291int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
292 unsigned int len, gfp_t gfp_mask)
293{
294 int reading = rq_data_dir(rq) == READ;
295 unsigned long addr = (unsigned long) kbuf;
296 int do_copy = 0;
297 struct bio *bio;
298 int ret;
299
300 if (len > (queue_max_hw_sectors(q) << 9))
301 return -EINVAL;
302 if (!len || !kbuf)
303 return -EINVAL;
304
305 do_copy = !blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf);
306 if (do_copy)
307 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
308 else
309 bio = bio_map_kern(q, kbuf, len, gfp_mask);
310
311 if (IS_ERR(bio))
312 return PTR_ERR(bio);
313
314 if (!reading)
315 bio->bi_rw |= REQ_WRITE;
316
317 if (do_copy)
318 rq->cmd_flags |= REQ_COPY_USER;
319
320 ret = blk_rq_append_bio(q, rq, bio);
321 if (unlikely(ret)) {
322 /* request is too big */
323 bio_put(bio);
324 return ret;
325 }
326
327 blk_queue_bounce(q, &rq->bio);
328 rq->buffer = NULL;
329 return 0;
330}
331EXPORT_SYMBOL(blk_rq_map_kern);