block/blk-map.c

   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
  14 struct 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
  21 static 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  */
  46 static 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  */
  77 static 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  */
 107 static 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
 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)
 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 = 1 << 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;
 224 cleanup:
 225         if (!map_data)
 226                 bio_free_pages(bio);
 227         bio_uninit(bio);
 228         kfree(bio);
 229 out_bmd:
 230         kfree(bmd);
 231         return ret;
 232 }
 233
 234 static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
 235                 gfp_t gfp_mask)
 236 {
 237         unsigned int max_sectors = queue_max_hw_sectors(rq->q);
 238         unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS);
 239         struct bio *bio;
 240         int ret;
 241         int j;
 242
 243         if (!iov_iter_count(iter))
 244                 return -EINVAL;
 245
 246         bio = bio_kmalloc(nr_vecs, gfp_mask);
 247         if (!bio)
 248                 return -ENOMEM;
 249         bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq));
 250
 251         while (iov_iter_count(iter)) {
 252                 struct page **pages;
 253                 ssize_t bytes;
 254                 size_t offs, added = 0;
 255                 int npages;
 256
 257                 bytes = iov_iter_get_pages_alloc2(iter, &pages, LONG_MAX, &offs);
 258                 if (unlikely(bytes <= 0)) {
 259                         ret = bytes ? bytes : -EFAULT;
 260                         goto out_unmap;
 261                 }
 262
 263                 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
 264
 265                 if (unlikely(offs & queue_dma_alignment(rq->q)))
 266                         j = 0;
 267                 else {
 268                         for (j = 0; j < npages; j++) {
 269                                 struct page *page = pages[j];
 270                                 unsigned int n = PAGE_SIZE - offs;
 271                                 bool same_page = false;
 272
 273                                 if (n > bytes)
 274                                         n = bytes;
 275
 276                                 if (!bio_add_hw_page(rq->q, bio, page, n, offs,
 277                                                      max_sectors, &same_page)) {
 278                                         if (same_page)
 279                                                 put_page(page);
 280                                         break;
 281                                 }
 282
 283                                 added += n;
 284                                 bytes -= n;
 285                                 offs = 0;
 286                         }
 287                 }
 288                 /*
 289                  * release the pages we didn't map into the bio, if any
 290                  */
 291                 while (j < npages)
 292                         put_page(pages[j++]);
 293                 kvfree(pages);
 294                 /* couldn't stuff something into bio? */
 295                 if (bytes) {
 296                         iov_iter_revert(iter, bytes);
 297                         break;
 298                 }
 299         }
 300
 301         ret = blk_rq_append_bio(rq, bio);
 302         if (ret)
 303                 goto out_unmap;
 304         return 0;
 305
 306  out_unmap:
 307         bio_release_pages(bio, false);
 308         bio_uninit(bio);
 309         kfree(bio);
 310         return ret;
 311 }
 312
 313 static void bio_invalidate_vmalloc_pages(struct bio *bio)
 314 {
 315 #ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
 316         if (bio->bi_private && !op_is_write(bio_op(bio))) {
 317                 unsigned long i, len = 0;
 318
 319                 for (i = 0; i < bio->bi_vcnt; i++)
 320                         len += bio->bi_io_vec[i].bv_len;
 321                 invalidate_kernel_vmap_range(bio->bi_private, len);
 322         }
 323 #endif
 324 }
 325
 326 static void bio_map_kern_endio(struct bio *bio)
 327 {
 328         bio_invalidate_vmalloc_pages(bio);
 329         bio_uninit(bio);
 330         kfree(bio);
 331 }
 332
 333 /**
 334  *      bio_map_kern    -       map kernel address into bio
 335  *      @q: the struct request_queue for the bio
 336  *      @data: pointer to buffer to map
 337  *      @len: length in bytes
 338  *      @gfp_mask: allocation flags for bio allocation
 339  *
 340  *      Map the kernel address into a bio suitable for io to a block
 341  *      device. Returns an error pointer in case of error.
 342  */
 343 static struct bio *bio_map_kern(struct request_queue *q, void *data,
 344                 unsigned int len, gfp_t gfp_mask)
 345 {
 346         unsigned long kaddr = (unsigned long)data;
 347         unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 348         unsigned long start = kaddr >> PAGE_SHIFT;
 349         const int nr_pages = end - start;
 350         bool is_vmalloc = is_vmalloc_addr(data);
 351         struct page *page;
 352         int offset, i;
 353         struct bio *bio;
 354
 355         bio = bio_kmalloc(nr_pages, gfp_mask);
 356         if (!bio)
 357                 return ERR_PTR(-ENOMEM);
 358         bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
 359
 360         if (is_vmalloc) {
 361                 flush_kernel_vmap_range(data, len);
 362                 bio->bi_private = data;
 363         }
 364
 365         offset = offset_in_page(kaddr);
 366         for (i = 0; i < nr_pages; i++) {
 367                 unsigned int bytes = PAGE_SIZE - offset;
 368
 369                 if (len <= 0)
 370                         break;
 371
 372                 if (bytes > len)
 373                         bytes = len;
 374
 375                 if (!is_vmalloc)
 376                         page = virt_to_page(data);
 377                 else
 378                         page = vmalloc_to_page(data);
 379                 if (bio_add_pc_page(q, bio, page, bytes,
 380                                     offset) < bytes) {
 381                         /* we don't support partial mappings */
 382                         bio_uninit(bio);
 383                         kfree(bio);
 384                         return ERR_PTR(-EINVAL);
 385                 }
 386
 387                 data += bytes;
 388                 len -= bytes;
 389                 offset = 0;
 390         }
 391
 392         bio->bi_end_io = bio_map_kern_endio;
 393         return bio;
 394 }
 395
 396 static void bio_copy_kern_endio(struct bio *bio)
 397 {
 398         bio_free_pages(bio);
 399         bio_uninit(bio);
 400         kfree(bio);
 401 }
 402
 403 static void bio_copy_kern_endio_read(struct bio *bio)
 404 {
 405         char *p = bio->bi_private;
 406         struct bio_vec *bvec;
 407         struct bvec_iter_all iter_all;
 408
 409         bio_for_each_segment_all(bvec, bio, iter_all) {
 410                 memcpy_from_bvec(p, bvec);
 411                 p += bvec->bv_len;
 412         }
 413
 414         bio_copy_kern_endio(bio);
 415 }
 416
 417 /**
 418  *      bio_copy_kern   -       copy kernel address into bio
 419  *      @q: the struct request_queue for the bio
 420  *      @data: pointer to buffer to copy
 421  *      @len: length in bytes
 422  *      @gfp_mask: allocation flags for bio and page allocation
 423  *      @reading: data direction is READ
 424  *
 425  *      copy the kernel address into a bio suitable for io to a block
 426  *      device. Returns an error pointer in case of error.
 427  */
 428 static struct bio *bio_copy_kern(struct request_queue *q, void *data,
 429                 unsigned int len, gfp_t gfp_mask, int reading)
 430 {
 431         unsigned long kaddr = (unsigned long)data;
 432         unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 433         unsigned long start = kaddr >> PAGE_SHIFT;
 434         struct bio *bio;
 435         void *p = data;
 436         int nr_pages = 0;
 437
 438         /*
 439          * Overflow, abort
 440          */
 441         if (end < start)
 442                 return ERR_PTR(-EINVAL);
 443
 444         nr_pages = end - start;
 445         bio = bio_kmalloc(nr_pages, gfp_mask);
 446         if (!bio)
 447                 return ERR_PTR(-ENOMEM);
 448         bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0);
 449
 450         while (len) {
 451                 struct page *page;
 452                 unsigned int bytes = PAGE_SIZE;
 453
 454                 if (bytes > len)
 455                         bytes = len;
 456
 457                 page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask);
 458                 if (!page)
 459                         goto cleanup;
 460
 461                 if (!reading)
 462                         memcpy(page_address(page), p, bytes);
 463
 464                 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
 465                         break;
 466
 467                 len -= bytes;
 468                 p += bytes;
 469         }
 470
 471         if (reading) {
 472                 bio->bi_end_io = bio_copy_kern_endio_read;
 473                 bio->bi_private = data;
 474         } else {
 475                 bio->bi_end_io = bio_copy_kern_endio;
 476         }
 477
 478         return bio;
 479
 480 cleanup:
 481         bio_free_pages(bio);
 482         bio_uninit(bio);
 483         kfree(bio);
 484         return ERR_PTR(-ENOMEM);
 485 }
 486
 487 /*
 488  * Append a bio to a passthrough request.  Only works if the bio can be merged
 489  * into the request based on the driver constraints.
 490  */
 491 int blk_rq_append_bio(struct request *rq, struct bio *bio)
 492 {
 493         struct bvec_iter iter;
 494         struct bio_vec bv;
 495         unsigned int nr_segs = 0;
 496
 497         bio_for_each_bvec(bv, bio, iter)
 498                 nr_segs++;
 499
 500         if (!rq->bio) {
 501                 blk_rq_bio_prep(rq, bio, nr_segs);
 502         } else {
 503                 if (!ll_back_merge_fn(rq, bio, nr_segs))
 504                         return -EINVAL;
 505                 rq->biotail->bi_next = bio;
 506                 rq->biotail = bio;
 507                 rq->__data_len += (bio)->bi_iter.bi_size;
 508                 bio_crypt_free_ctx(bio);
 509         }
 510
 511         return 0;
 512 }
 513 EXPORT_SYMBOL(blk_rq_append_bio);
 514
 515 /**
 516  * blk_rq_map_user_iov - map user data to a request, for passthrough requests
 517  * @q:          request queue where request should be inserted
 518  * @rq:         request to map data to
 519  * @map_data:   pointer to the rq_map_data holding pages (if necessary)
 520  * @iter:       iovec iterator
 521  * @gfp_mask:   memory allocation flags
 522  *
 523  * Description:
 524  *    Data will be mapped directly for zero copy I/O, if possible. Otherwise
 525  *    a kernel bounce buffer is used.
 526  *
 527  *    A matching blk_rq_unmap_user() must be issued at the end of I/O, while
 528  *    still in process context.
 529  */
 530 int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
 531                         struct rq_map_data *map_data,
 532                         const struct iov_iter *iter, gfp_t gfp_mask)
 533 {
 534         bool copy = false;
 535         unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
 536         struct bio *bio = NULL;
 537         struct iov_iter i;
 538         int ret = -EINVAL;
 539
 540         if (!iter_is_iovec(iter))
 541                 goto fail;
 542
 543         if (map_data)
 544                 copy = true;
 545         else if (blk_queue_may_bounce(q))
 546                 copy = true;
 547         else if (iov_iter_alignment(iter) & align)
 548                 copy = true;
 549         else if (queue_virt_boundary(q))
 550                 copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
 551
 552         i = *iter;
 553         do {
 554                 if (copy)
 555                         ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
 556                 else
 557                         ret = bio_map_user_iov(rq, &i, gfp_mask);
 558                 if (ret)
 559                         goto unmap_rq;
 560                 if (!bio)
 561                         bio = rq->bio;
 562         } while (iov_iter_count(&i));
 563
 564         return 0;
 565
 566 unmap_rq:
 567         blk_rq_unmap_user(bio);
 568 fail:
 569         rq->bio = NULL;
 570         return ret;
 571 }
 572 EXPORT_SYMBOL(blk_rq_map_user_iov);
 573
 574 int blk_rq_map_user(struct request_queue *q, struct request *rq,
 575                     struct rq_map_data *map_data, void __user *ubuf,
 576                     unsigned long len, gfp_t gfp_mask)
 577 {
 578         struct iovec iov;
 579         struct iov_iter i;
 580         int ret = import_single_range(rq_data_dir(rq), ubuf, len, &iov, &i);
 581
 582         if (unlikely(ret < 0))
 583                 return ret;
 584
 585         return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
 586 }
 587 EXPORT_SYMBOL(blk_rq_map_user);
 588
 589 /**
 590  * blk_rq_unmap_user - unmap a request with user data
 591  * @bio:               start of bio list
 592  *
 593  * Description:
 594  *    Unmap a rq previously mapped by blk_rq_map_user(). The caller must
 595  *    supply the original rq->bio from the blk_rq_map_user() return, since
 596  *    the I/O completion may have changed rq->bio.
 597  */
 598 int blk_rq_unmap_user(struct bio *bio)
 599 {
 600         struct bio *next_bio;
 601         int ret = 0, ret2;
 602
 603         while (bio) {
 604                 if (bio->bi_private) {
 605                         ret2 = bio_uncopy_user(bio);
 606                         if (ret2 && !ret)
 607                                 ret = ret2;
 608                 } else {
 609                         bio_release_pages(bio, bio_data_dir(bio) == READ);
 610                 }
 611
 612                 next_bio = bio;
 613                 bio = bio->bi_next;
 614                 bio_uninit(next_bio);
 615                 kfree(next_bio);
 616         }
 617
 618         return ret;
 619 }
 620 EXPORT_SYMBOL(blk_rq_unmap_user);
 621
 622 /**
 623  * blk_rq_map_kern - map kernel data to a request, for passthrough requests
 624  * @q:          request queue where request should be inserted
 625  * @rq:         request to fill
 626  * @kbuf:       the kernel buffer
 627  * @len:        length of user data
 628  * @gfp_mask:   memory allocation flags
 629  *
 630  * Description:
 631  *    Data will be mapped directly if possible. Otherwise a bounce
 632  *    buffer is used. Can be called multiple times to append multiple
 633  *    buffers.
 634  */
 635 int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
 636                     unsigned int len, gfp_t gfp_mask)
 637 {
 638         int reading = rq_data_dir(rq) == READ;
 639         unsigned long addr = (unsigned long) kbuf;
 640         struct bio *bio;
 641         int ret;
 642
 643         if (len > (queue_max_hw_sectors(q) << 9))
 644                 return -EINVAL;
 645         if (!len || !kbuf)
 646                 return -EINVAL;
 647
 648         if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||
 649             blk_queue_may_bounce(q))
 650                 bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
 651         else
 652                 bio = bio_map_kern(q, kbuf, len, gfp_mask);
 653
 654         if (IS_ERR(bio))
 655                 return PTR_ERR(bio);
 656
 657         bio->bi_opf &= ~REQ_OP_MASK;
 658         bio->bi_opf |= req_op(rq);
 659
 660         ret = blk_rq_append_bio(rq, bio);
 661         if (unlikely(ret)) {
 662                 bio_uninit(bio);
 663                 kfree(bio);
 664         }
 665         return ret;
 666 }
 667 EXPORT_SYMBOL(blk_rq_map_kern);