794 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * Functions related to mapping data to requests4 */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 bmd->iter = *data;33 if (iter_is_iovec(data)) {34 memcpy(bmd->iov, iter_iov(data), sizeof(struct iovec) * data->nr_segs);35 bmd->iter.__iov = bmd->iov;36 }37 return bmd;38}39 40/**41 * bio_copy_from_iter - copy all pages from iov_iter to bio42 * @bio: The &struct bio which describes the I/O as destination43 * @iter: iov_iter as source44 *45 * Copy all pages from iov_iter to bio.46 * Returns 0 on success, or error on failure.47 */48static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)49{50 struct bio_vec *bvec;51 struct bvec_iter_all iter_all;52 53 bio_for_each_segment_all(bvec, bio, iter_all) {54 ssize_t ret;55 56 ret = copy_page_from_iter(bvec->bv_page,57 bvec->bv_offset,58 bvec->bv_len,59 iter);60 61 if (!iov_iter_count(iter))62 break;63 64 if (ret < bvec->bv_len)65 return -EFAULT;66 }67 68 return 0;69}70 71/**72 * bio_copy_to_iter - copy all pages from bio to iov_iter73 * @bio: The &struct bio which describes the I/O as source74 * @iter: iov_iter as destination75 *76 * Copy all pages from bio to iov_iter.77 * Returns 0 on success, or error on failure.78 */79static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)80{81 struct bio_vec *bvec;82 struct bvec_iter_all iter_all;83 84 bio_for_each_segment_all(bvec, bio, iter_all) {85 ssize_t ret;86 87 ret = copy_page_to_iter(bvec->bv_page,88 bvec->bv_offset,89 bvec->bv_len,90 &iter);91 92 if (!iov_iter_count(&iter))93 break;94 95 if (ret < bvec->bv_len)96 return -EFAULT;97 }98 99 return 0;100}101 102/**103 * bio_uncopy_user - finish previously mapped bio104 * @bio: bio being terminated105 *106 * Free pages allocated from bio_copy_user_iov() and write back data107 * to user space in case of a read.108 */109static int bio_uncopy_user(struct bio *bio)110{111 struct bio_map_data *bmd = bio->bi_private;112 int ret = 0;113 114 if (!bmd->is_null_mapped) {115 /*116 * if we're in a workqueue, the request is orphaned, so117 * don't copy into a random user address space, just free118 * and return -EINTR so user space doesn't expect any data.119 */120 if (!current->mm)121 ret = -EINTR;122 else if (bio_data_dir(bio) == READ)123 ret = bio_copy_to_iter(bio, bmd->iter);124 if (bmd->is_our_pages)125 bio_free_pages(bio);126 }127 kfree(bmd);128 return ret;129}130 131static int bio_copy_user_iov(struct request *rq, struct rq_map_data *map_data,132 struct iov_iter *iter, gfp_t gfp_mask)133{134 struct bio_map_data *bmd;135 struct page *page;136 struct bio *bio;137 int i = 0, ret;138 int nr_pages;139 unsigned int len = iter->count;140 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;141 142 bmd = bio_alloc_map_data(iter, gfp_mask);143 if (!bmd)144 return -ENOMEM;145 146 /*147 * We need to do a deep copy of the iov_iter including the iovecs.148 * The caller provided iov might point to an on-stack or otherwise149 * shortlived one.150 */151 bmd->is_our_pages = !map_data;152 bmd->is_null_mapped = (map_data && map_data->null_mapped);153 154 nr_pages = bio_max_segs(DIV_ROUND_UP(offset + len, PAGE_SIZE));155 156 ret = -ENOMEM;157 bio = bio_kmalloc(nr_pages, gfp_mask);158 if (!bio)159 goto out_bmd;160 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, req_op(rq));161 162 if (map_data) {163 nr_pages = 1U << map_data->page_order;164 i = map_data->offset / PAGE_SIZE;165 }166 while (len) {167 unsigned int bytes = PAGE_SIZE;168 169 bytes -= offset;170 171 if (bytes > len)172 bytes = len;173 174 if (map_data) {175 if (i == map_data->nr_entries * nr_pages) {176 ret = -ENOMEM;177 goto cleanup;178 }179 180 page = map_data->pages[i / nr_pages];181 page += (i % nr_pages);182 183 i++;184 } else {185 page = alloc_page(GFP_NOIO | gfp_mask);186 if (!page) {187 ret = -ENOMEM;188 goto cleanup;189 }190 }191 192 if (bio_add_pc_page(rq->q, bio, page, bytes, offset) < bytes) {193 if (!map_data)194 __free_page(page);195 break;196 }197 198 len -= bytes;199 offset = 0;200 }201 202 if (map_data)203 map_data->offset += bio->bi_iter.bi_size;204 205 /*206 * success207 */208 if (iov_iter_rw(iter) == WRITE &&209 (!map_data || !map_data->null_mapped)) {210 ret = bio_copy_from_iter(bio, iter);211 if (ret)212 goto cleanup;213 } else if (map_data && map_data->from_user) {214 struct iov_iter iter2 = *iter;215 216 /* This is the copy-in part of SG_DXFER_TO_FROM_DEV. */217 iter2.data_source = ITER_SOURCE;218 ret = bio_copy_from_iter(bio, &iter2);219 if (ret)220 goto cleanup;221 } else {222 if (bmd->is_our_pages)223 zero_fill_bio(bio);224 iov_iter_advance(iter, bio->bi_iter.bi_size);225 }226 227 bio->bi_private = bmd;228 229 ret = blk_rq_append_bio(rq, bio);230 if (ret)231 goto cleanup;232 return 0;233cleanup:234 if (!map_data)235 bio_free_pages(bio);236 bio_uninit(bio);237 kfree(bio);238out_bmd:239 kfree(bmd);240 return ret;241}242 243static void blk_mq_map_bio_put(struct bio *bio)244{245 if (bio->bi_opf & REQ_ALLOC_CACHE) {246 bio_put(bio);247 } else {248 bio_uninit(bio);249 kfree(bio);250 }251}252 253static struct bio *blk_rq_map_bio_alloc(struct request *rq,254 unsigned int nr_vecs, gfp_t gfp_mask)255{256 struct bio *bio;257 258 if (rq->cmd_flags & REQ_ALLOC_CACHE && (nr_vecs <= BIO_INLINE_VECS)) {259 bio = bio_alloc_bioset(NULL, nr_vecs, rq->cmd_flags, gfp_mask,260 &fs_bio_set);261 if (!bio)262 return NULL;263 } else {264 bio = bio_kmalloc(nr_vecs, gfp_mask);265 if (!bio)266 return NULL;267 bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq));268 }269 return bio;270}271 272static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,273 gfp_t gfp_mask)274{275 iov_iter_extraction_t extraction_flags = 0;276 unsigned int max_sectors = queue_max_hw_sectors(rq->q);277 unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS);278 struct bio *bio;279 int ret;280 int j;281 282 if (!iov_iter_count(iter))283 return -EINVAL;284 285 bio = blk_rq_map_bio_alloc(rq, nr_vecs, gfp_mask);286 if (bio == NULL)287 return -ENOMEM;288 289 if (blk_queue_pci_p2pdma(rq->q))290 extraction_flags |= ITER_ALLOW_P2PDMA;291 if (iov_iter_extract_will_pin(iter))292 bio_set_flag(bio, BIO_PAGE_PINNED);293 294 while (iov_iter_count(iter)) {295 struct page *stack_pages[UIO_FASTIOV];296 struct page **pages = stack_pages;297 ssize_t bytes;298 size_t offs;299 int npages;300 301 if (nr_vecs > ARRAY_SIZE(stack_pages))302 pages = NULL;303 304 bytes = iov_iter_extract_pages(iter, &pages, LONG_MAX,305 nr_vecs, extraction_flags, &offs);306 if (unlikely(bytes <= 0)) {307 ret = bytes ? bytes : -EFAULT;308 goto out_unmap;309 }310 311 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);312 313 if (unlikely(offs & queue_dma_alignment(rq->q)))314 j = 0;315 else {316 for (j = 0; j < npages; j++) {317 struct page *page = pages[j];318 unsigned int n = PAGE_SIZE - offs;319 bool same_page = false;320 321 if (n > bytes)322 n = bytes;323 324 if (!bio_add_hw_page(rq->q, bio, page, n, offs,325 max_sectors, &same_page))326 break;327 328 if (same_page)329 bio_release_page(bio, page);330 bytes -= n;331 offs = 0;332 }333 }334 /*335 * release the pages we didn't map into the bio, if any336 */337 while (j < npages)338 bio_release_page(bio, pages[j++]);339 if (pages != stack_pages)340 kvfree(pages);341 /* couldn't stuff something into bio? */342 if (bytes) {343 iov_iter_revert(iter, bytes);344 break;345 }346 }347 348 ret = blk_rq_append_bio(rq, bio);349 if (ret)350 goto out_unmap;351 return 0;352 353 out_unmap:354 bio_release_pages(bio, false);355 blk_mq_map_bio_put(bio);356 return ret;357}358 359static void bio_invalidate_vmalloc_pages(struct bio *bio)360{361#ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE362 if (bio->bi_private && !op_is_write(bio_op(bio))) {363 unsigned long i, len = 0;364 365 for (i = 0; i < bio->bi_vcnt; i++)366 len += bio->bi_io_vec[i].bv_len;367 invalidate_kernel_vmap_range(bio->bi_private, len);368 }369#endif370}371 372static void bio_map_kern_endio(struct bio *bio)373{374 bio_invalidate_vmalloc_pages(bio);375 bio_uninit(bio);376 kfree(bio);377}378 379/**380 * bio_map_kern - map kernel address into bio381 * @q: the struct request_queue for the bio382 * @data: pointer to buffer to map383 * @len: length in bytes384 * @gfp_mask: allocation flags for bio allocation385 *386 * Map the kernel address into a bio suitable for io to a block387 * device. Returns an error pointer in case of error.388 */389static struct bio *bio_map_kern(struct request_queue *q, void *data,390 unsigned int len, gfp_t gfp_mask)391{392 unsigned long kaddr = (unsigned long)data;393 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;394 unsigned long start = kaddr >> PAGE_SHIFT;395 const int nr_pages = end - start;396 bool is_vmalloc = is_vmalloc_addr(data);397 struct page *page;398 int offset, i;399 struct bio *bio;400 401 bio = bio_kmalloc(nr_pages, gfp_mask);402 if (!bio)403 return ERR_PTR(-ENOMEM);404 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);405 406 if (is_vmalloc) {407 flush_kernel_vmap_range(data, len);408 bio->bi_private = data;409 }410 411 offset = offset_in_page(kaddr);412 for (i = 0; i < nr_pages; i++) {413 unsigned int bytes = PAGE_SIZE - offset;414 415 if (len <= 0)416 break;417 418 if (bytes > len)419 bytes = len;420 421 if (!is_vmalloc)422 page = virt_to_page(data);423 else424 page = vmalloc_to_page(data);425 if (bio_add_pc_page(q, bio, page, bytes,426 offset) < bytes) {427 /* we don't support partial mappings */428 bio_uninit(bio);429 kfree(bio);430 return ERR_PTR(-EINVAL);431 }432 433 data += bytes;434 len -= bytes;435 offset = 0;436 }437 438 bio->bi_end_io = bio_map_kern_endio;439 return bio;440}441 442static void bio_copy_kern_endio(struct bio *bio)443{444 bio_free_pages(bio);445 bio_uninit(bio);446 kfree(bio);447}448 449static void bio_copy_kern_endio_read(struct bio *bio)450{451 char *p = bio->bi_private;452 struct bio_vec *bvec;453 struct bvec_iter_all iter_all;454 455 bio_for_each_segment_all(bvec, bio, iter_all) {456 memcpy_from_bvec(p, bvec);457 p += bvec->bv_len;458 }459 460 bio_copy_kern_endio(bio);461}462 463/**464 * bio_copy_kern - copy kernel address into bio465 * @q: the struct request_queue for the bio466 * @data: pointer to buffer to copy467 * @len: length in bytes468 * @gfp_mask: allocation flags for bio and page allocation469 * @reading: data direction is READ470 *471 * copy the kernel address into a bio suitable for io to a block472 * device. Returns an error pointer in case of error.473 */474static struct bio *bio_copy_kern(struct request_queue *q, void *data,475 unsigned int len, gfp_t gfp_mask, int reading)476{477 unsigned long kaddr = (unsigned long)data;478 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;479 unsigned long start = kaddr >> PAGE_SHIFT;480 struct bio *bio;481 void *p = data;482 int nr_pages = 0;483 484 /*485 * Overflow, abort486 */487 if (end < start)488 return ERR_PTR(-EINVAL);489 490 nr_pages = end - start;491 bio = bio_kmalloc(nr_pages, gfp_mask);492 if (!bio)493 return ERR_PTR(-ENOMEM);494 bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);495 496 while (len) {497 struct page *page;498 unsigned int bytes = PAGE_SIZE;499 500 if (bytes > len)501 bytes = len;502 503 page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask);504 if (!page)505 goto cleanup;506 507 if (!reading)508 memcpy(page_address(page), p, bytes);509 510 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)511 break;512 513 len -= bytes;514 p += bytes;515 }516 517 if (reading) {518 bio->bi_end_io = bio_copy_kern_endio_read;519 bio->bi_private = data;520 } else {521 bio->bi_end_io = bio_copy_kern_endio;522 }523 524 return bio;525 526cleanup:527 bio_free_pages(bio);528 bio_uninit(bio);529 kfree(bio);530 return ERR_PTR(-ENOMEM);531}532 533/*534 * Append a bio to a passthrough request. Only works if the bio can be merged535 * into the request based on the driver constraints.536 */537int blk_rq_append_bio(struct request *rq, struct bio *bio)538{539 struct bvec_iter iter;540 struct bio_vec bv;541 unsigned int nr_segs = 0;542 543 bio_for_each_bvec(bv, bio, iter)544 nr_segs++;545 546 if (!rq->bio) {547 blk_rq_bio_prep(rq, bio, nr_segs);548 } else {549 if (!ll_back_merge_fn(rq, bio, nr_segs))550 return -EINVAL;551 rq->biotail->bi_next = bio;552 rq->biotail = bio;553 rq->__data_len += (bio)->bi_iter.bi_size;554 bio_crypt_free_ctx(bio);555 }556 557 return 0;558}559EXPORT_SYMBOL(blk_rq_append_bio);560 561/* Prepare bio for passthrough IO given ITER_BVEC iter */562static int blk_rq_map_user_bvec(struct request *rq, const struct iov_iter *iter)563{564 const struct queue_limits *lim = &rq->q->limits;565 unsigned int max_bytes = lim->max_hw_sectors << SECTOR_SHIFT;566 unsigned int nsegs;567 struct bio *bio;568 int ret;569 570 if (!iov_iter_count(iter) || iov_iter_count(iter) > max_bytes)571 return -EINVAL;572 573 /* reuse the bvecs from the iterator instead of allocating new ones */574 bio = blk_rq_map_bio_alloc(rq, 0, GFP_KERNEL);575 if (!bio)576 return -ENOMEM;577 bio_iov_bvec_set(bio, (struct iov_iter *)iter);578 579 /* check that the data layout matches the hardware restrictions */580 ret = bio_split_rw_at(bio, lim, &nsegs, max_bytes);581 if (ret) {582 /* if we would have to split the bio, copy instead */583 if (ret > 0)584 ret = -EREMOTEIO;585 blk_mq_map_bio_put(bio);586 return ret;587 }588 589 blk_rq_bio_prep(rq, bio, nsegs);590 return 0;591}592 593/**594 * blk_rq_map_user_iov - map user data to a request, for passthrough requests595 * @q: request queue where request should be inserted596 * @rq: request to map data to597 * @map_data: pointer to the rq_map_data holding pages (if necessary)598 * @iter: iovec iterator599 * @gfp_mask: memory allocation flags600 *601 * Description:602 * Data will be mapped directly for zero copy I/O, if possible. Otherwise603 * a kernel bounce buffer is used.604 *605 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while606 * still in process context.607 */608int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,609 struct rq_map_data *map_data,610 const struct iov_iter *iter, gfp_t gfp_mask)611{612 bool copy = false, map_bvec = false;613 unsigned long align = blk_lim_dma_alignment_and_pad(&q->limits);614 struct bio *bio = NULL;615 struct iov_iter i;616 int ret = -EINVAL;617 618 if (map_data)619 copy = true;620 else if (blk_queue_may_bounce(q))621 copy = true;622 else if (iov_iter_alignment(iter) & align)623 copy = true;624 else if (iov_iter_is_bvec(iter))625 map_bvec = true;626 else if (!user_backed_iter(iter))627 copy = true;628 else if (queue_virt_boundary(q))629 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);630 631 if (map_bvec) {632 ret = blk_rq_map_user_bvec(rq, iter);633 if (!ret)634 return 0;635 if (ret != -EREMOTEIO)636 goto fail;637 /* fall back to copying the data on limits mismatches */638 copy = true;639 }640 641 i = *iter;642 do {643 if (copy)644 ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);645 else646 ret = bio_map_user_iov(rq, &i, gfp_mask);647 if (ret)648 goto unmap_rq;649 if (!bio)650 bio = rq->bio;651 } while (iov_iter_count(&i));652 653 return 0;654 655unmap_rq:656 blk_rq_unmap_user(bio);657fail:658 rq->bio = NULL;659 return ret;660}661EXPORT_SYMBOL(blk_rq_map_user_iov);662 663int blk_rq_map_user(struct request_queue *q, struct request *rq,664 struct rq_map_data *map_data, void __user *ubuf,665 unsigned long len, gfp_t gfp_mask)666{667 struct iov_iter i;668 int ret = import_ubuf(rq_data_dir(rq), ubuf, len, &i);669 670 if (unlikely(ret < 0))671 return ret;672 673 return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);674}675EXPORT_SYMBOL(blk_rq_map_user);676 677int blk_rq_map_user_io(struct request *req, struct rq_map_data *map_data,678 void __user *ubuf, unsigned long buf_len, gfp_t gfp_mask,679 bool vec, int iov_count, bool check_iter_count, int rw)680{681 int ret = 0;682 683 if (vec) {684 struct iovec fast_iov[UIO_FASTIOV];685 struct iovec *iov = fast_iov;686 struct iov_iter iter;687 688 ret = import_iovec(rw, ubuf, iov_count ? iov_count : buf_len,689 UIO_FASTIOV, &iov, &iter);690 if (ret < 0)691 return ret;692 693 if (iov_count) {694 /* SG_IO howto says that the shorter of the two wins */695 iov_iter_truncate(&iter, buf_len);696 if (check_iter_count && !iov_iter_count(&iter)) {697 kfree(iov);698 return -EINVAL;699 }700 }701 702 ret = blk_rq_map_user_iov(req->q, req, map_data, &iter,703 gfp_mask);704 kfree(iov);705 } else if (buf_len) {706 ret = blk_rq_map_user(req->q, req, map_data, ubuf, buf_len,707 gfp_mask);708 }709 return ret;710}711EXPORT_SYMBOL(blk_rq_map_user_io);712 713/**714 * blk_rq_unmap_user - unmap a request with user data715 * @bio: start of bio list716 *717 * Description:718 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must719 * supply the original rq->bio from the blk_rq_map_user() return, since720 * the I/O completion may have changed rq->bio.721 */722int blk_rq_unmap_user(struct bio *bio)723{724 struct bio *next_bio;725 int ret = 0, ret2;726 727 while (bio) {728 if (bio->bi_private) {729 ret2 = bio_uncopy_user(bio);730 if (ret2 && !ret)731 ret = ret2;732 } else {733 bio_release_pages(bio, bio_data_dir(bio) == READ);734 }735 736 if (bio_integrity(bio))737 bio_integrity_unmap_user(bio);738 739 next_bio = bio;740 bio = bio->bi_next;741 blk_mq_map_bio_put(next_bio);742 }743 744 return ret;745}746EXPORT_SYMBOL(blk_rq_unmap_user);747 748/**749 * blk_rq_map_kern - map kernel data to a request, for passthrough requests750 * @q: request queue where request should be inserted751 * @rq: request to fill752 * @kbuf: the kernel buffer753 * @len: length of user data754 * @gfp_mask: memory allocation flags755 *756 * Description:757 * Data will be mapped directly if possible. Otherwise a bounce758 * buffer is used. Can be called multiple times to append multiple759 * buffers.760 */761int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,762 unsigned int len, gfp_t gfp_mask)763{764 int reading = rq_data_dir(rq) == READ;765 unsigned long addr = (unsigned long) kbuf;766 struct bio *bio;767 int ret;768 769 if (len > (queue_max_hw_sectors(q) << 9))770 return -EINVAL;771 if (!len || !kbuf)772 return -EINVAL;773 774 if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||775 blk_queue_may_bounce(q))776 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);777 else778 bio = bio_map_kern(q, kbuf, len, gfp_mask);779 780 if (IS_ERR(bio))781 return PTR_ERR(bio);782 783 bio->bi_opf &= ~REQ_OP_MASK;784 bio->bi_opf |= req_op(rq);785 786 ret = blk_rq_append_bio(rq, bio);787 if (unlikely(ret)) {788 bio_uninit(bio);789 kfree(bio);790 }791 return ret;792}793EXPORT_SYMBOL(blk_rq_map_kern);794