lib/iov_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <crypto/hash.h>
   3 #include <linux/export.h>
   4 #include <linux/bvec.h>
   5 #include <linux/fault-inject-usercopy.h>
   6 #include <linux/uio.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/highmem.h>
   9 #include <linux/slab.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/splice.h>
  12 #include <linux/compat.h>
  13 #include <net/checksum.h>
  14 #include <linux/scatterlist.h>
  15 #include <linux/instrumented.h>
  16
  17 #define PIPE_PARANOIA /* for now */
  18
  19 #define iterate_iovec(i, n, __v, __p, skip, STEP) {     \
  20         size_t left;                                    \
  21         size_t wanted = n;                              \
  22         __p = i->iov;                                   \
  23         __v.iov_len = min(n, __p->iov_len - skip);      \
  24         if (likely(__v.iov_len)) {                      \
  25                 __v.iov_base = __p->iov_base + skip;    \
  26                 left = (STEP);                          \
  27                 __v.iov_len -= left;                    \
  28                 skip += __v.iov_len;                    \
  29                 n -= __v.iov_len;                       \
  30         } else {                                        \
  31                 left = 0;                               \
  32         }                                               \
  33         while (unlikely(!left && n)) {                  \
  34                 __p++;                                  \
  35                 __v.iov_len = min(n, __p->iov_len);     \
  36                 if (unlikely(!__v.iov_len))             \
  37                         continue;                       \
  38                 __v.iov_base = __p->iov_base;           \
  39                 left = (STEP);                          \
  40                 __v.iov_len -= left;                    \
  41                 skip = __v.iov_len;                     \
  42                 n -= __v.iov_len;                       \
  43         }                                               \
  44         n = wanted - n;                                 \
  45 }
  46
  47 #define iterate_kvec(i, n, __v, __p, skip, STEP) {      \
  48         size_t wanted = n;                              \
  49         __p = i->kvec;                                  \
  50         __v.iov_len = min(n, __p->iov_len - skip);      \
  51         if (likely(__v.iov_len)) {                      \
  52                 __v.iov_base = __p->iov_base + skip;    \
  53                 (void)(STEP);                           \
  54                 skip += __v.iov_len;                    \
  55                 n -= __v.iov_len;                       \
  56         }                                               \
  57         while (unlikely(n)) {                           \
  58                 __p++;                                  \
  59                 __v.iov_len = min(n, __p->iov_len);     \
  60                 if (unlikely(!__v.iov_len))             \
  61                         continue;                       \
  62                 __v.iov_base = __p->iov_base;           \
  63                 (void)(STEP);                           \
  64                 skip = __v.iov_len;                     \
  65                 n -= __v.iov_len;                       \
  66         }                                               \
  67         n = wanted;                                     \
  68 }
  69
  70 #define iterate_bvec(i, n, __v, __bi, skip, STEP) {     \
  71         struct bvec_iter __start;                       \
  72         __start.bi_size = n;                            \
  73         __start.bi_bvec_done = skip;                    \
  74         __start.bi_idx = 0;                             \
  75         for_each_bvec(__v, i->bvec, __bi, __start) {    \
  76                 (void)(STEP);                           \
  77         }                                               \
  78 }
  79
  80 #define iterate_xarray(i, n, __v, skip, STEP) {         \
  81         struct page *head = NULL;                               \
  82         size_t wanted = n, seg, offset;                         \
  83         loff_t start = i->xarray_start + skip;                  \
  84         pgoff_t index = start >> PAGE_SHIFT;                    \
  85         int j;                                                  \
  86                                                                 \
  87         XA_STATE(xas, i->xarray, index);                        \
  88                                                                 \
  89         rcu_read_lock();                                                \
  90         xas_for_each(&xas, head, ULONG_MAX) {                           \
  91                 if (xas_retry(&xas, head))                              \
  92                         continue;                                       \
  93                 if (WARN_ON(xa_is_value(head)))                         \
  94                         break;                                          \
  95                 if (WARN_ON(PageHuge(head)))                            \
  96                         break;                                          \
  97                 for (j = (head->index < index) ? index - head->index : 0; \
  98                      j < thp_nr_pages(head); j++) {                     \
  99                         __v.bv_page = head + j;                         \
 100                         offset = (i->xarray_start + skip) & ~PAGE_MASK; \
 101                         seg = PAGE_SIZE - offset;                       \
 102                         __v.bv_offset = offset;                         \
 103                         __v.bv_len = min(n, seg);                       \
 104                         (void)(STEP);                                   \
 105                         n -= __v.bv_len;                                \
 106                         skip += __v.bv_len;                             \
 107                         if (n == 0)                                     \
 108                                 break;                                  \
 109                 }                                                       \
 110                 if (n == 0)                                             \
 111                         break;                                          \
 112         }                                                       \
 113         rcu_read_unlock();                                      \
 114         n = wanted - n;                                         \
 115 }
 116
 117 #define iterate_all_kinds(i, n, v, I, B, K, X) {                \
 118         if (likely(n)) {                                        \
 119                 size_t skip = i->iov_offset;                    \
 120                 if (likely(iter_is_iovec(i))) {                 \
 121                         const struct iovec *iov;                \
 122                         struct iovec v;                         \
 123                         iterate_iovec(i, n, v, iov, skip, (I))  \
 124                 } else if (iov_iter_is_bvec(i)) {               \
 125                         struct bio_vec v;                       \
 126                         struct bvec_iter __bi;                  \
 127                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 128                 } else if (iov_iter_is_kvec(i)) {               \
 129                         const struct kvec *kvec;                \
 130                         struct kvec v;                          \
 131                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 132                 } else if (iov_iter_is_xarray(i)) {             \
 133                         struct bio_vec v;                       \
 134                         iterate_xarray(i, n, v, skip, (X));     \
 135                 }                                               \
 136         }                                                       \
 137 }
 138
 139 #define iterate_and_advance(i, n, v, I, B, K, X) {              \
 140         if (unlikely(i->count < n))                             \
 141                 n = i->count;                                   \
 142         if (i->count) {                                         \
 143                 size_t skip = i->iov_offset;                    \
 144                 if (likely(iter_is_iovec(i))) {                 \
 145                         const struct iovec *iov;                \
 146                         struct iovec v;                         \
 147                         iterate_iovec(i, n, v, iov, skip, (I))  \
 148                         if (skip == iov->iov_len) {             \
 149                                 iov++;                          \
 150                                 skip = 0;                       \
 151                         }                                       \
 152                         i->nr_segs -= iov - i->iov;             \
 153                         i->iov = iov;                           \
 154                 } else if (iov_iter_is_bvec(i)) {               \
 155                         const struct bio_vec *bvec = i->bvec;   \
 156                         struct bio_vec v;                       \
 157                         struct bvec_iter __bi;                  \
 158                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 159                         i->bvec = __bvec_iter_bvec(i->bvec, __bi);      \
 160                         i->nr_segs -= i->bvec - bvec;           \
 161                         skip = __bi.bi_bvec_done;               \
 162                 } else if (iov_iter_is_kvec(i)) {               \
 163                         const struct kvec *kvec;                \
 164                         struct kvec v;                          \
 165                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 166                         if (skip == kvec->iov_len) {            \
 167                                 kvec++;                         \
 168                                 skip = 0;                       \
 169                         }                                       \
 170                         i->nr_segs -= kvec - i->kvec;           \
 171                         i->kvec = kvec;                         \
 172                 } else if (iov_iter_is_xarray(i)) {             \
 173                         struct bio_vec v;                       \
 174                         iterate_xarray(i, n, v, skip, (X))      \
 175                 }                                               \
 176                 i->count -= n;                                  \
 177                 i->iov_offset = skip;                           \
 178         }                                                       \
 179 }
 180
 181 static int copyout(void __user *to, const void *from, size_t n)
 182 {
 183         if (should_fail_usercopy())
 184                 return n;
 185         if (access_ok(to, n)) {
 186                 instrument_copy_to_user(to, from, n);
 187                 n = raw_copy_to_user(to, from, n);
 188         }
 189         return n;
 190 }
 191
 192 static int copyin(void *to, const void __user *from, size_t n)
 193 {
 194         if (should_fail_usercopy())
 195                 return n;
 196         if (access_ok(from, n)) {
 197                 instrument_copy_from_user(to, from, n);
 198                 n = raw_copy_from_user(to, from, n);
 199         }
 200         return n;
 201 }
 202
 203 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 204                          struct iov_iter *i)
 205 {
 206         size_t skip, copy, left, wanted;
 207         const struct iovec *iov;
 208         char __user *buf;
 209         void *kaddr, *from;
 210
 211         if (unlikely(bytes > i->count))
 212                 bytes = i->count;
 213
 214         if (unlikely(!bytes))
 215                 return 0;
 216
 217         might_fault();
 218         wanted = bytes;
 219         iov = i->iov;
 220         skip = i->iov_offset;
 221         buf = iov->iov_base + skip;
 222         copy = min(bytes, iov->iov_len - skip);
 223
 224         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 225                 kaddr = kmap_atomic(page);
 226                 from = kaddr + offset;
 227
 228                 /* first chunk, usually the only one */
 229                 left = copyout(buf, from, copy);
 230                 copy -= left;
 231                 skip += copy;
 232                 from += copy;
 233                 bytes -= copy;
 234
 235                 while (unlikely(!left && bytes)) {
 236                         iov++;
 237                         buf = iov->iov_base;
 238                         copy = min(bytes, iov->iov_len);
 239                         left = copyout(buf, from, copy);
 240                         copy -= left;
 241                         skip = copy;
 242                         from += copy;
 243                         bytes -= copy;
 244                 }
 245                 if (likely(!bytes)) {
 246                         kunmap_atomic(kaddr);
 247                         goto done;
 248                 }
 249                 offset = from - kaddr;
 250                 buf += copy;
 251                 kunmap_atomic(kaddr);
 252                 copy = min(bytes, iov->iov_len - skip);
 253         }
 254         /* Too bad - revert to non-atomic kmap */
 255
 256         kaddr = kmap(page);
 257         from = kaddr + offset;
 258         left = copyout(buf, from, copy);
 259         copy -= left;
 260         skip += copy;
 261         from += copy;
 262         bytes -= copy;
 263         while (unlikely(!left && bytes)) {
 264                 iov++;
 265                 buf = iov->iov_base;
 266                 copy = min(bytes, iov->iov_len);
 267                 left = copyout(buf, from, copy);
 268                 copy -= left;
 269                 skip = copy;
 270                 from += copy;
 271                 bytes -= copy;
 272         }
 273         kunmap(page);
 274
 275 done:
 276         if (skip == iov->iov_len) {
 277                 iov++;
 278                 skip = 0;
 279         }
 280         i->count -= wanted - bytes;
 281         i->nr_segs -= iov - i->iov;
 282         i->iov = iov;
 283         i->iov_offset = skip;
 284         return wanted - bytes;
 285 }
 286
 287 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 288                          struct iov_iter *i)
 289 {
 290         size_t skip, copy, left, wanted;
 291         const struct iovec *iov;
 292         char __user *buf;
 293         void *kaddr, *to;
 294
 295         if (unlikely(bytes > i->count))
 296                 bytes = i->count;
 297
 298         if (unlikely(!bytes))
 299                 return 0;
 300
 301         might_fault();
 302         wanted = bytes;
 303         iov = i->iov;
 304         skip = i->iov_offset;
 305         buf = iov->iov_base + skip;
 306         copy = min(bytes, iov->iov_len - skip);
 307
 308         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 309                 kaddr = kmap_atomic(page);
 310                 to = kaddr + offset;
 311
 312                 /* first chunk, usually the only one */
 313                 left = copyin(to, buf, copy);
 314                 copy -= left;
 315                 skip += copy;
 316                 to += copy;
 317                 bytes -= copy;
 318
 319                 while (unlikely(!left && bytes)) {
 320                         iov++;
 321                         buf = iov->iov_base;
 322                         copy = min(bytes, iov->iov_len);
 323                         left = copyin(to, buf, copy);
 324                         copy -= left;
 325                         skip = copy;
 326                         to += copy;
 327                         bytes -= copy;
 328                 }
 329                 if (likely(!bytes)) {
 330                         kunmap_atomic(kaddr);
 331                         goto done;
 332                 }
 333                 offset = to - kaddr;
 334                 buf += copy;
 335                 kunmap_atomic(kaddr);
 336                 copy = min(bytes, iov->iov_len - skip);
 337         }
 338         /* Too bad - revert to non-atomic kmap */
 339
 340         kaddr = kmap(page);
 341         to = kaddr + offset;
 342         left = copyin(to, buf, copy);
 343         copy -= left;
 344         skip += copy;
 345         to += copy;
 346         bytes -= copy;
 347         while (unlikely(!left && bytes)) {
 348                 iov++;
 349                 buf = iov->iov_base;
 350                 copy = min(bytes, iov->iov_len);
 351                 left = copyin(to, buf, copy);
 352                 copy -= left;
 353                 skip = copy;
 354                 to += copy;
 355                 bytes -= copy;
 356         }
 357         kunmap(page);
 358
 359 done:
 360         if (skip == iov->iov_len) {
 361                 iov++;
 362                 skip = 0;
 363         }
 364         i->count -= wanted - bytes;
 365         i->nr_segs -= iov - i->iov;
 366         i->iov = iov;
 367         i->iov_offset = skip;
 368         return wanted - bytes;
 369 }
 370
 371 #ifdef PIPE_PARANOIA
 372 static bool sanity(const struct iov_iter *i)
 373 {
 374         struct pipe_inode_info *pipe = i->pipe;
 375         unsigned int p_head = pipe->head;
 376         unsigned int p_tail = pipe->tail;
 377         unsigned int p_mask = pipe->ring_size - 1;
 378         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 379         unsigned int i_head = i->head;
 380         unsigned int idx;
 381
 382         if (i->iov_offset) {
 383                 struct pipe_buffer *p;
 384                 if (unlikely(p_occupancy == 0))
 385                         goto Bad;       // pipe must be non-empty
 386                 if (unlikely(i_head != p_head - 1))
 387                         goto Bad;       // must be at the last buffer...
 388
 389                 p = &pipe->bufs[i_head & p_mask];
 390                 if (unlikely(p->offset + p->len != i->iov_offset))
 391                         goto Bad;       // ... at the end of segment
 392         } else {
 393                 if (i_head != p_head)
 394                         goto Bad;       // must be right after the last buffer
 395         }
 396         return true;
 397 Bad:
 398         printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
 399         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 400                         p_head, p_tail, pipe->ring_size);
 401         for (idx = 0; idx < pipe->ring_size; idx++)
 402                 printk(KERN_ERR "[%p %p %d %d]\n",
 403                         pipe->bufs[idx].ops,
 404                         pipe->bufs[idx].page,
 405                         pipe->bufs[idx].offset,
 406                         pipe->bufs[idx].len);
 407         WARN_ON(1);
 408         return false;
 409 }
 410 #else
 411 #define sanity(i) true
 412 #endif
 413
 414 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 415                          struct iov_iter *i)
 416 {
 417         struct pipe_inode_info *pipe = i->pipe;
 418         struct pipe_buffer *buf;
 419         unsigned int p_tail = pipe->tail;
 420         unsigned int p_mask = pipe->ring_size - 1;
 421         unsigned int i_head = i->head;
 422         size_t off;
 423
 424         if (unlikely(bytes > i->count))
 425                 bytes = i->count;
 426
 427         if (unlikely(!bytes))
 428                 return 0;
 429
 430         if (!sanity(i))
 431                 return 0;
 432
 433         off = i->iov_offset;
 434         buf = &pipe->bufs[i_head & p_mask];
 435         if (off) {
 436                 if (offset == off && buf->page == page) {
 437                         /* merge with the last one */
 438                         buf->len += bytes;
 439                         i->iov_offset += bytes;
 440                         goto out;
 441                 }
 442                 i_head++;
 443                 buf = &pipe->bufs[i_head & p_mask];
 444         }
 445         if (pipe_full(i_head, p_tail, pipe->max_usage))
 446                 return 0;
 447
 448         buf->ops = &page_cache_pipe_buf_ops;
 449         get_page(page);
 450         buf->page = page;
 451         buf->offset = offset;
 452         buf->len = bytes;
 453
 454         pipe->head = i_head + 1;
 455         i->iov_offset = offset + bytes;
 456         i->head = i_head;
 457 out:
 458         i->count -= bytes;
 459         return bytes;
 460 }
 461
 462 /*
 463  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 464  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 465  *
 466  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 467  * because it is an invalid address).
 468  */
 469 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
 470 {
 471         size_t skip = i->iov_offset;
 472         const struct iovec *iov;
 473         int err;
 474         struct iovec v;
 475
 476         if (iter_is_iovec(i)) {
 477                 iterate_iovec(i, bytes, v, iov, skip, ({
 478                         err = fault_in_pages_readable(v.iov_base, v.iov_len);
 479                         if (unlikely(err))
 480                         return err;
 481                 0;}))
 482         }
 483         return 0;
 484 }
 485 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 486
 487 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 488                         const struct iovec *iov, unsigned long nr_segs,
 489                         size_t count)
 490 {
 491         WARN_ON(direction & ~(READ | WRITE));
 492         WARN_ON_ONCE(uaccess_kernel());
 493         *i = (struct iov_iter) {
 494                 .iter_type = ITER_IOVEC,
 495                 .data_source = direction,
 496                 .iov = iov,
 497                 .nr_segs = nr_segs,
 498                 .iov_offset = 0,
 499                 .count = count
 500         };
 501 }
 502 EXPORT_SYMBOL(iov_iter_init);
 503
 504 static inline bool allocated(struct pipe_buffer *buf)
 505 {
 506         return buf->ops == &default_pipe_buf_ops;
 507 }
 508
 509 static inline void data_start(const struct iov_iter *i,
 510                               unsigned int *iter_headp, size_t *offp)
 511 {
 512         unsigned int p_mask = i->pipe->ring_size - 1;
 513         unsigned int iter_head = i->head;
 514         size_t off = i->iov_offset;
 515
 516         if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
 517                     off == PAGE_SIZE)) {
 518                 iter_head++;
 519                 off = 0;
 520         }
 521         *iter_headp = iter_head;
 522         *offp = off;
 523 }
 524
 525 static size_t push_pipe(struct iov_iter *i, size_t size,
 526                         int *iter_headp, size_t *offp)
 527 {
 528         struct pipe_inode_info *pipe = i->pipe;
 529         unsigned int p_tail = pipe->tail;
 530         unsigned int p_mask = pipe->ring_size - 1;
 531         unsigned int iter_head;
 532         size_t off;
 533         ssize_t left;
 534
 535         if (unlikely(size > i->count))
 536                 size = i->count;
 537         if (unlikely(!size))
 538                 return 0;
 539
 540         left = size;
 541         data_start(i, &iter_head, &off);
 542         *iter_headp = iter_head;
 543         *offp = off;
 544         if (off) {
 545                 left -= PAGE_SIZE - off;
 546                 if (left <= 0) {
 547                         pipe->bufs[iter_head & p_mask].len += size;
 548                         return size;
 549                 }
 550                 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
 551                 iter_head++;
 552         }
 553         while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
 554                 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
 555                 struct page *page = alloc_page(GFP_USER);
 556                 if (!page)
 557                         break;
 558
 559                 buf->ops = &default_pipe_buf_ops;
 560                 buf->page = page;
 561                 buf->offset = 0;
 562                 buf->len = min_t(ssize_t, left, PAGE_SIZE);
 563                 left -= buf->len;
 564                 iter_head++;
 565                 pipe->head = iter_head;
 566
 567                 if (left == 0)
 568                         return size;
 569         }
 570         return size - left;
 571 }
 572
 573 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 574                                 struct iov_iter *i)
 575 {
 576         struct pipe_inode_info *pipe = i->pipe;
 577         unsigned int p_mask = pipe->ring_size - 1;
 578         unsigned int i_head;
 579         size_t n, off;
 580
 581         if (!sanity(i))
 582                 return 0;
 583
 584         bytes = n = push_pipe(i, bytes, &i_head, &off);
 585         if (unlikely(!n))
 586                 return 0;
 587         do {
 588                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 589                 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
 590                 i->head = i_head;
 591                 i->iov_offset = off + chunk;
 592                 n -= chunk;
 593                 addr += chunk;
 594                 off = 0;
 595                 i_head++;
 596         } while (n);
 597         i->count -= bytes;
 598         return bytes;
 599 }
 600
 601 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 602                               __wsum sum, size_t off)
 603 {
 604         __wsum next = csum_partial_copy_nocheck(from, to, len);
 605         return csum_block_add(sum, next, off);
 606 }
 607
 608 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 609                                          struct csum_state *csstate,
 610                                          struct iov_iter *i)
 611 {
 612         struct pipe_inode_info *pipe = i->pipe;
 613         unsigned int p_mask = pipe->ring_size - 1;
 614         __wsum sum = csstate->csum;
 615         size_t off = csstate->off;
 616         unsigned int i_head;
 617         size_t n, r;
 618
 619         if (!sanity(i))
 620                 return 0;
 621
 622         bytes = n = push_pipe(i, bytes, &i_head, &r);
 623         if (unlikely(!n))
 624                 return 0;
 625         do {
 626                 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
 627                 char *p = kmap_atomic(pipe->bufs[i_head & p_mask].page);
 628                 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
 629                 kunmap_atomic(p);
 630                 i->head = i_head;
 631                 i->iov_offset = r + chunk;
 632                 n -= chunk;
 633                 off += chunk;
 634                 addr += chunk;
 635                 r = 0;
 636                 i_head++;
 637         } while (n);
 638         i->count -= bytes;
 639         csstate->csum = sum;
 640         csstate->off = off;
 641         return bytes;
 642 }
 643
 644 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 645 {
 646         const char *from = addr;
 647         if (unlikely(iov_iter_is_pipe(i)))
 648                 return copy_pipe_to_iter(addr, bytes, i);
 649         if (iter_is_iovec(i))
 650                 might_fault();
 651         iterate_and_advance(i, bytes, v,
 652                 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 653                 memcpy_to_page(v.bv_page, v.bv_offset,
 654                                (from += v.bv_len) - v.bv_len, v.bv_len),
 655                 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 656                 memcpy_to_page(v.bv_page, v.bv_offset,
 657                                (from += v.bv_len) - v.bv_len, v.bv_len)
 658         )
 659
 660         return bytes;
 661 }
 662 EXPORT_SYMBOL(_copy_to_iter);
 663
 664 #ifdef CONFIG_ARCH_HAS_COPY_MC
 665 static int copyout_mc(void __user *to, const void *from, size_t n)
 666 {
 667         if (access_ok(to, n)) {
 668                 instrument_copy_to_user(to, from, n);
 669                 n = copy_mc_to_user((__force void *) to, from, n);
 670         }
 671         return n;
 672 }
 673
 674 static unsigned long copy_mc_to_page(struct page *page, size_t offset,
 675                 const char *from, size_t len)
 676 {
 677         unsigned long ret;
 678         char *to;
 679
 680         to = kmap_atomic(page);
 681         ret = copy_mc_to_kernel(to + offset, from, len);
 682         kunmap_atomic(to);
 683
 684         return ret;
 685 }
 686
 687 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 688                                 struct iov_iter *i)
 689 {
 690         struct pipe_inode_info *pipe = i->pipe;
 691         unsigned int p_mask = pipe->ring_size - 1;
 692         unsigned int i_head;
 693         size_t n, off, xfer = 0;
 694
 695         if (!sanity(i))
 696                 return 0;
 697
 698         bytes = n = push_pipe(i, bytes, &i_head, &off);
 699         if (unlikely(!n))
 700                 return 0;
 701         do {
 702                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 703                 unsigned long rem;
 704
 705                 rem = copy_mc_to_page(pipe->bufs[i_head & p_mask].page,
 706                                             off, addr, chunk);
 707                 i->head = i_head;
 708                 i->iov_offset = off + chunk - rem;
 709                 xfer += chunk - rem;
 710                 if (rem)
 711                         break;
 712                 n -= chunk;
 713                 addr += chunk;
 714                 off = 0;
 715                 i_head++;
 716         } while (n);
 717         i->count -= xfer;
 718         return xfer;
 719 }
 720
 721 /**
 722  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 723  * @addr: source kernel address
 724  * @bytes: total transfer length
 725  * @iter: destination iterator
 726  *
 727  * The pmem driver deploys this for the dax operation
 728  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 729  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 730  * successfully copied.
 731  *
 732  * The main differences between this and typical _copy_to_iter().
 733  *
 734  * * Typical tail/residue handling after a fault retries the copy
 735  *   byte-by-byte until the fault happens again. Re-triggering machine
 736  *   checks is potentially fatal so the implementation uses source
 737  *   alignment and poison alignment assumptions to avoid re-triggering
 738  *   hardware exceptions.
 739  *
 740  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 741  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 742  *   a short copy.
 743  */
 744 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 745 {
 746         const char *from = addr;
 747         unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
 748
 749         if (unlikely(iov_iter_is_pipe(i)))
 750                 return copy_mc_pipe_to_iter(addr, bytes, i);
 751         if (iter_is_iovec(i))
 752                 might_fault();
 753         iterate_and_advance(i, bytes, v,
 754                 copyout_mc(v.iov_base, (from += v.iov_len) - v.iov_len,
 755                            v.iov_len),
 756                 ({
 757                 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
 758                                       (from += v.bv_len) - v.bv_len, v.bv_len);
 759                 if (rem) {
 760                         curr_addr = (unsigned long) from;
 761                         bytes = curr_addr - s_addr - rem;
 762                         return bytes;
 763                 }
 764                 }),
 765                 ({
 766                 rem = copy_mc_to_kernel(v.iov_base, (from += v.iov_len)
 767                                         - v.iov_len, v.iov_len);
 768                 if (rem) {
 769                         curr_addr = (unsigned long) from;
 770                         bytes = curr_addr - s_addr - rem;
 771                         return bytes;
 772                 }
 773                 }),
 774                 ({
 775                 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
 776                                       (from += v.bv_len) - v.bv_len, v.bv_len);
 777                 if (rem) {
 778                         curr_addr = (unsigned long) from;
 779                         bytes = curr_addr - s_addr - rem;
 780                         rcu_read_unlock();
 781                         i->iov_offset += bytes;
 782                         i->count -= bytes;
 783                         return bytes;
 784                 }
 785                 })
 786         )
 787
 788         return bytes;
 789 }
 790 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 791 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 792
 793 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 794 {
 795         char *to = addr;
 796         if (unlikely(iov_iter_is_pipe(i))) {
 797                 WARN_ON(1);
 798                 return 0;
 799         }
 800         if (iter_is_iovec(i))
 801                 might_fault();
 802         iterate_and_advance(i, bytes, v,
 803                 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 804                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 805                                  v.bv_offset, v.bv_len),
 806                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 807                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 808                                  v.bv_offset, v.bv_len)
 809         )
 810
 811         return bytes;
 812 }
 813 EXPORT_SYMBOL(_copy_from_iter);
 814
 815 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 816 {
 817         char *to = addr;
 818         if (unlikely(iov_iter_is_pipe(i))) {
 819                 WARN_ON(1);
 820                 return 0;
 821         }
 822         iterate_and_advance(i, bytes, v,
 823                 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 824                                          v.iov_base, v.iov_len),
 825                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 826                                  v.bv_offset, v.bv_len),
 827                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 828                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 829                                  v.bv_offset, v.bv_len)
 830         )
 831
 832         return bytes;
 833 }
 834 EXPORT_SYMBOL(_copy_from_iter_nocache);
 835
 836 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 837 /**
 838  * _copy_from_iter_flushcache - write destination through cpu cache
 839  * @addr: destination kernel address
 840  * @bytes: total transfer length
 841  * @iter: source iterator
 842  *
 843  * The pmem driver arranges for filesystem-dax to use this facility via
 844  * dax_copy_from_iter() for ensuring that writes to persistent memory
 845  * are flushed through the CPU cache. It is differentiated from
 846  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 847  * all iterator types. The _copy_from_iter_nocache() only attempts to
 848  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 849  * instructions that strand dirty-data in the cache.
 850  */
 851 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 852 {
 853         char *to = addr;
 854         if (unlikely(iov_iter_is_pipe(i))) {
 855                 WARN_ON(1);
 856                 return 0;
 857         }
 858         iterate_and_advance(i, bytes, v,
 859                 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
 860                                          v.iov_base, v.iov_len),
 861                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 862                                  v.bv_offset, v.bv_len),
 863                 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
 864                         v.iov_len),
 865                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 866                                  v.bv_offset, v.bv_len)
 867         )
 868
 869         return bytes;
 870 }
 871 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 872 #endif
 873
 874 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 875 {
 876         struct page *head;
 877         size_t v = n + offset;
 878
 879         /*
 880          * The general case needs to access the page order in order
 881          * to compute the page size.
 882          * However, we mostly deal with order-0 pages and thus can
 883          * avoid a possible cache line miss for requests that fit all
 884          * page orders.
 885          */
 886         if (n <= v && v <= PAGE_SIZE)
 887                 return true;
 888
 889         head = compound_head(page);
 890         v += (page - head) << PAGE_SHIFT;
 891
 892         if (likely(n <= v && v <= (page_size(head))))
 893                 return true;
 894         WARN_ON(1);
 895         return false;
 896 }
 897
 898 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 899                          struct iov_iter *i)
 900 {
 901         if (likely(iter_is_iovec(i)))
 902                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 903         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 904                 void *kaddr = kmap_atomic(page);
 905                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 906                 kunmap_atomic(kaddr);
 907                 return wanted;
 908         }
 909         if (iov_iter_is_pipe(i))
 910                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 911         if (unlikely(iov_iter_is_discard(i))) {
 912                 if (unlikely(i->count < bytes))
 913                         bytes = i->count;
 914                 i->count -= bytes;
 915                 return bytes;
 916         }
 917         WARN_ON(1);
 918         return 0;
 919 }
 920
 921 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 922                          struct iov_iter *i)
 923 {
 924         size_t res = 0;
 925         if (unlikely(!page_copy_sane(page, offset, bytes)))
 926                 return 0;
 927         page += offset / PAGE_SIZE; // first subpage
 928         offset %= PAGE_SIZE;
 929         while (1) {
 930                 size_t n = __copy_page_to_iter(page, offset,
 931                                 min(bytes, (size_t)PAGE_SIZE - offset), i);
 932                 res += n;
 933                 bytes -= n;
 934                 if (!bytes || !n)
 935                         break;
 936                 offset += n;
 937                 if (offset == PAGE_SIZE) {
 938                         page++;
 939                         offset = 0;
 940                 }
 941         }
 942         return res;
 943 }
 944 EXPORT_SYMBOL(copy_page_to_iter);
 945
 946 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 947                          struct iov_iter *i)
 948 {
 949         if (unlikely(!page_copy_sane(page, offset, bytes)))
 950                 return 0;
 951         if (likely(iter_is_iovec(i)))
 952                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 953         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 954                 void *kaddr = kmap_atomic(page);
 955                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 956                 kunmap_atomic(kaddr);
 957                 return wanted;
 958         }
 959         WARN_ON(1);
 960         return 0;
 961 }
 962 EXPORT_SYMBOL(copy_page_from_iter);
 963
 964 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 965 {
 966         struct pipe_inode_info *pipe = i->pipe;
 967         unsigned int p_mask = pipe->ring_size - 1;
 968         unsigned int i_head;
 969         size_t n, off;
 970
 971         if (!sanity(i))
 972                 return 0;
 973
 974         bytes = n = push_pipe(i, bytes, &i_head, &off);
 975         if (unlikely(!n))
 976                 return 0;
 977
 978         do {
 979                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 980                 memzero_page(pipe->bufs[i_head & p_mask].page, off, chunk);
 981                 i->head = i_head;
 982                 i->iov_offset = off + chunk;
 983                 n -= chunk;
 984                 off = 0;
 985                 i_head++;
 986         } while (n);
 987         i->count -= bytes;
 988         return bytes;
 989 }
 990
 991 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 992 {
 993         if (unlikely(iov_iter_is_pipe(i)))
 994                 return pipe_zero(bytes, i);
 995         iterate_and_advance(i, bytes, v,
 996                 clear_user(v.iov_base, v.iov_len),
 997                 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
 998                 memset(v.iov_base, 0, v.iov_len),
 999                 memzero_page(v.bv_page, v.bv_offset, v.bv_len)
1000         )
1001
1002         return bytes;
1003 }
1004 EXPORT_SYMBOL(iov_iter_zero);
1005
1006 size_t iov_iter_copy_from_user_atomic(struct page *page,
1007                 struct iov_iter *i, unsigned long offset, size_t bytes)
1008 {
1009         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
1010         if (unlikely(!page_copy_sane(page, offset, bytes))) {
1011                 kunmap_atomic(kaddr);
1012                 return 0;
1013         }
1014         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1015                 kunmap_atomic(kaddr);
1016                 WARN_ON(1);
1017                 return 0;
1018         }
1019         iterate_all_kinds(i, bytes, v,
1020                 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1021                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1022                                  v.bv_offset, v.bv_len),
1023                 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1024                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1025                                  v.bv_offset, v.bv_len)
1026         )
1027         kunmap_atomic(kaddr);
1028         return bytes;
1029 }
1030 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1031
1032 static inline void pipe_truncate(struct iov_iter *i)
1033 {
1034         struct pipe_inode_info *pipe = i->pipe;
1035         unsigned int p_tail = pipe->tail;
1036         unsigned int p_head = pipe->head;
1037         unsigned int p_mask = pipe->ring_size - 1;
1038
1039         if (!pipe_empty(p_head, p_tail)) {
1040                 struct pipe_buffer *buf;
1041                 unsigned int i_head = i->head;
1042                 size_t off = i->iov_offset;
1043
1044                 if (off) {
1045                         buf = &pipe->bufs[i_head & p_mask];
1046                         buf->len = off - buf->offset;
1047                         i_head++;
1048                 }
1049                 while (p_head != i_head) {
1050                         p_head--;
1051                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1052                 }
1053
1054                 pipe->head = p_head;
1055         }
1056 }
1057
1058 static void pipe_advance(struct iov_iter *i, size_t size)
1059 {
1060         struct pipe_inode_info *pipe = i->pipe;
1061         if (size) {
1062                 struct pipe_buffer *buf;
1063                 unsigned int p_mask = pipe->ring_size - 1;
1064                 unsigned int i_head = i->head;
1065                 size_t off = i->iov_offset, left = size;
1066
1067                 if (off) /* make it relative to the beginning of buffer */
1068                         left += off - pipe->bufs[i_head & p_mask].offset;
1069                 while (1) {
1070                         buf = &pipe->bufs[i_head & p_mask];
1071                         if (left <= buf->len)
1072                                 break;
1073                         left -= buf->len;
1074                         i_head++;
1075                 }
1076                 i->head = i_head;
1077                 i->iov_offset = buf->offset + left;
1078         }
1079         i->count -= size;
1080         /* ... and discard everything past that point */
1081         pipe_truncate(i);
1082 }
1083
1084 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1085 {
1086         struct bvec_iter bi;
1087
1088         bi.bi_size = i->count;
1089         bi.bi_bvec_done = i->iov_offset;
1090         bi.bi_idx = 0;
1091         bvec_iter_advance(i->bvec, &bi, size);
1092
1093         i->bvec += bi.bi_idx;
1094         i->nr_segs -= bi.bi_idx;
1095         i->count = bi.bi_size;
1096         i->iov_offset = bi.bi_bvec_done;
1097 }
1098
1099 void iov_iter_advance(struct iov_iter *i, size_t size)
1100 {
1101         if (unlikely(i->count < size))
1102                 size = i->count;
1103         if (unlikely(iov_iter_is_pipe(i))) {
1104                 pipe_advance(i, size);
1105                 return;
1106         }
1107         if (unlikely(iov_iter_is_discard(i))) {
1108                 i->count -= size;
1109                 return;
1110         }
1111         if (unlikely(iov_iter_is_xarray(i))) {
1112                 i->iov_offset += size;
1113                 i->count -= size;
1114                 return;
1115         }
1116         if (iov_iter_is_bvec(i)) {
1117                 iov_iter_bvec_advance(i, size);
1118                 return;
1119         }
1120         iterate_and_advance(i, size, v, 0, 0, 0, 0)
1121 }
1122 EXPORT_SYMBOL(iov_iter_advance);
1123
1124 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1125 {
1126         if (!unroll)
1127                 return;
1128         if (WARN_ON(unroll > MAX_RW_COUNT))
1129                 return;
1130         i->count += unroll;
1131         if (unlikely(iov_iter_is_pipe(i))) {
1132                 struct pipe_inode_info *pipe = i->pipe;
1133                 unsigned int p_mask = pipe->ring_size - 1;
1134                 unsigned int i_head = i->head;
1135                 size_t off = i->iov_offset;
1136                 while (1) {
1137                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1138                         size_t n = off - b->offset;
1139                         if (unroll < n) {
1140                                 off -= unroll;
1141                                 break;
1142                         }
1143                         unroll -= n;
1144                         if (!unroll && i_head == i->start_head) {
1145                                 off = 0;
1146                                 break;
1147                         }
1148                         i_head--;
1149                         b = &pipe->bufs[i_head & p_mask];
1150                         off = b->offset + b->len;
1151                 }
1152                 i->iov_offset = off;
1153                 i->head = i_head;
1154                 pipe_truncate(i);
1155                 return;
1156         }
1157         if (unlikely(iov_iter_is_discard(i)))
1158                 return;
1159         if (unroll <= i->iov_offset) {
1160                 i->iov_offset -= unroll;
1161                 return;
1162         }
1163         unroll -= i->iov_offset;
1164         if (iov_iter_is_xarray(i)) {
1165                 BUG(); /* We should never go beyond the start of the specified
1166                         * range since we might then be straying into pages that
1167                         * aren't pinned.
1168                         */
1169         } else if (iov_iter_is_bvec(i)) {
1170                 const struct bio_vec *bvec = i->bvec;
1171                 while (1) {
1172                         size_t n = (--bvec)->bv_len;
1173                         i->nr_segs++;
1174                         if (unroll <= n) {
1175                                 i->bvec = bvec;
1176                                 i->iov_offset = n - unroll;
1177                                 return;
1178                         }
1179                         unroll -= n;
1180                 }
1181         } else { /* same logics for iovec and kvec */
1182                 const struct iovec *iov = i->iov;
1183                 while (1) {
1184                         size_t n = (--iov)->iov_len;
1185                         i->nr_segs++;
1186                         if (unroll <= n) {
1187                                 i->iov = iov;
1188                                 i->iov_offset = n - unroll;
1189                                 return;
1190                         }
1191                         unroll -= n;
1192                 }
1193         }
1194 }
1195 EXPORT_SYMBOL(iov_iter_revert);
1196
1197 /*
1198  * Return the count of just the current iov_iter segment.
1199  */
1200 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1201 {
1202         if (i->nr_segs > 1) {
1203                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1204                         return min(i->count, i->iov->iov_len - i->iov_offset);
1205                 if (iov_iter_is_bvec(i))
1206                         return min(i->count, i->bvec->bv_len - i->iov_offset);
1207         }
1208         return i->count;
1209 }
1210 EXPORT_SYMBOL(iov_iter_single_seg_count);
1211
1212 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1213                         const struct kvec *kvec, unsigned long nr_segs,
1214                         size_t count)
1215 {
1216         WARN_ON(direction & ~(READ | WRITE));
1217         *i = (struct iov_iter){
1218                 .iter_type = ITER_KVEC,
1219                 .data_source = direction,
1220                 .kvec = kvec,
1221                 .nr_segs = nr_segs,
1222                 .iov_offset = 0,
1223                 .count = count
1224         };
1225 }
1226 EXPORT_SYMBOL(iov_iter_kvec);
1227
1228 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1229                         const struct bio_vec *bvec, unsigned long nr_segs,
1230                         size_t count)
1231 {
1232         WARN_ON(direction & ~(READ | WRITE));
1233         *i = (struct iov_iter){
1234                 .iter_type = ITER_BVEC,
1235                 .data_source = direction,
1236                 .bvec = bvec,
1237                 .nr_segs = nr_segs,
1238                 .iov_offset = 0,
1239                 .count = count
1240         };
1241 }
1242 EXPORT_SYMBOL(iov_iter_bvec);
1243
1244 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1245                         struct pipe_inode_info *pipe,
1246                         size_t count)
1247 {
1248         BUG_ON(direction != READ);
1249         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1250         *i = (struct iov_iter){
1251                 .iter_type = ITER_PIPE,
1252                 .data_source = false,
1253                 .pipe = pipe,
1254                 .head = pipe->head,
1255                 .start_head = pipe->head,
1256                 .iov_offset = 0,
1257                 .count = count
1258         };
1259 }
1260 EXPORT_SYMBOL(iov_iter_pipe);
1261
1262 /**
1263  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1264  * @i: The iterator to initialise.
1265  * @direction: The direction of the transfer.
1266  * @xarray: The xarray to access.
1267  * @start: The start file position.
1268  * @count: The size of the I/O buffer in bytes.
1269  *
1270  * Set up an I/O iterator to either draw data out of the pages attached to an
1271  * inode or to inject data into those pages.  The pages *must* be prevented
1272  * from evaporation, either by taking a ref on them or locking them by the
1273  * caller.
1274  */
1275 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1276                      struct xarray *xarray, loff_t start, size_t count)
1277 {
1278         BUG_ON(direction & ~1);
1279         *i = (struct iov_iter) {
1280                 .iter_type = ITER_XARRAY,
1281                 .data_source = direction,
1282                 .xarray = xarray,
1283                 .xarray_start = start,
1284                 .count = count,
1285                 .iov_offset = 0
1286         };
1287 }
1288 EXPORT_SYMBOL(iov_iter_xarray);
1289
1290 /**
1291  * iov_iter_discard - Initialise an I/O iterator that discards data
1292  * @i: The iterator to initialise.
1293  * @direction: The direction of the transfer.
1294  * @count: The size of the I/O buffer in bytes.
1295  *
1296  * Set up an I/O iterator that just discards everything that's written to it.
1297  * It's only available as a READ iterator.
1298  */
1299 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1300 {
1301         BUG_ON(direction != READ);
1302         *i = (struct iov_iter){
1303                 .iter_type = ITER_DISCARD,
1304                 .data_source = false,
1305                 .count = count,
1306                 .iov_offset = 0
1307         };
1308 }
1309 EXPORT_SYMBOL(iov_iter_discard);
1310
1311 unsigned long iov_iter_alignment(const struct iov_iter *i)
1312 {
1313         unsigned long res = 0;
1314         size_t size = i->count;
1315
1316         if (unlikely(iov_iter_is_pipe(i))) {
1317                 unsigned int p_mask = i->pipe->ring_size - 1;
1318
1319                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1320                         return size | i->iov_offset;
1321                 return size;
1322         }
1323         if (unlikely(iov_iter_is_xarray(i)))
1324                 return (i->xarray_start + i->iov_offset) | i->count;
1325         iterate_all_kinds(i, size, v,
1326                 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1327                 res |= v.bv_offset | v.bv_len,
1328                 res |= (unsigned long)v.iov_base | v.iov_len,
1329                 res |= v.bv_offset | v.bv_len
1330         )
1331         return res;
1332 }
1333 EXPORT_SYMBOL(iov_iter_alignment);
1334
1335 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1336 {
1337         unsigned long res = 0;
1338         size_t size = i->count;
1339
1340         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1341                 WARN_ON(1);
1342                 return ~0U;
1343         }
1344
1345         iterate_all_kinds(i, size, v,
1346                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1347                         (size != v.iov_len ? size : 0), 0),
1348                 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1349                         (size != v.bv_len ? size : 0)),
1350                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1351                         (size != v.iov_len ? size : 0)),
1352                 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1353                         (size != v.bv_len ? size : 0))
1354                 );
1355         return res;
1356 }
1357 EXPORT_SYMBOL(iov_iter_gap_alignment);
1358
1359 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1360                                 size_t maxsize,
1361                                 struct page **pages,
1362                                 int iter_head,
1363                                 size_t *start)
1364 {
1365         struct pipe_inode_info *pipe = i->pipe;
1366         unsigned int p_mask = pipe->ring_size - 1;
1367         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1368         if (!n)
1369                 return -EFAULT;
1370
1371         maxsize = n;
1372         n += *start;
1373         while (n > 0) {
1374                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1375                 iter_head++;
1376                 n -= PAGE_SIZE;
1377         }
1378
1379         return maxsize;
1380 }
1381
1382 static ssize_t pipe_get_pages(struct iov_iter *i,
1383                    struct page **pages, size_t maxsize, unsigned maxpages,
1384                    size_t *start)
1385 {
1386         unsigned int iter_head, npages;
1387         size_t capacity;
1388
1389         if (!maxsize)
1390                 return 0;
1391
1392         if (!sanity(i))
1393                 return -EFAULT;
1394
1395         data_start(i, &iter_head, start);
1396         /* Amount of free space: some of this one + all after this one */
1397         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1398         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1399
1400         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1401 }
1402
1403 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1404                                           pgoff_t index, unsigned int nr_pages)
1405 {
1406         XA_STATE(xas, xa, index);
1407         struct page *page;
1408         unsigned int ret = 0;
1409
1410         rcu_read_lock();
1411         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1412                 if (xas_retry(&xas, page))
1413                         continue;
1414
1415                 /* Has the page moved or been split? */
1416                 if (unlikely(page != xas_reload(&xas))) {
1417                         xas_reset(&xas);
1418                         continue;
1419                 }
1420
1421                 pages[ret] = find_subpage(page, xas.xa_index);
1422                 get_page(pages[ret]);
1423                 if (++ret == nr_pages)
1424                         break;
1425         }
1426         rcu_read_unlock();
1427         return ret;
1428 }
1429
1430 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1431                                      struct page **pages, size_t maxsize,
1432                                      unsigned maxpages, size_t *_start_offset)
1433 {
1434         unsigned nr, offset;
1435         pgoff_t index, count;
1436         size_t size = maxsize, actual;
1437         loff_t pos;
1438
1439         if (!size || !maxpages)
1440                 return 0;
1441
1442         pos = i->xarray_start + i->iov_offset;
1443         index = pos >> PAGE_SHIFT;
1444         offset = pos & ~PAGE_MASK;
1445         *_start_offset = offset;
1446
1447         count = 1;
1448         if (size > PAGE_SIZE - offset) {
1449                 size -= PAGE_SIZE - offset;
1450                 count += size >> PAGE_SHIFT;
1451                 size &= ~PAGE_MASK;
1452                 if (size)
1453                         count++;
1454         }
1455
1456         if (count > maxpages)
1457                 count = maxpages;
1458
1459         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1460         if (nr == 0)
1461                 return 0;
1462
1463         actual = PAGE_SIZE * nr;
1464         actual -= offset;
1465         if (nr == count && size > 0) {
1466                 unsigned last_offset = (nr > 1) ? 0 : offset;
1467                 actual -= PAGE_SIZE - (last_offset + size);
1468         }
1469         return actual;
1470 }
1471
1472 ssize_t iov_iter_get_pages(struct iov_iter *i,
1473                    struct page **pages, size_t maxsize, unsigned maxpages,
1474                    size_t *start)
1475 {
1476         if (maxsize > i->count)
1477                 maxsize = i->count;
1478
1479         if (unlikely(iov_iter_is_pipe(i)))
1480                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1481         if (unlikely(iov_iter_is_xarray(i)))
1482                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1483         if (unlikely(iov_iter_is_discard(i)))
1484                 return -EFAULT;
1485
1486         iterate_all_kinds(i, maxsize, v, ({
1487                 unsigned long addr = (unsigned long)v.iov_base;
1488                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1489                 int n;
1490                 int res;
1491
1492                 if (len > maxpages * PAGE_SIZE)
1493                         len = maxpages * PAGE_SIZE;
1494                 addr &= ~(PAGE_SIZE - 1);
1495                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1496                 res = get_user_pages_fast(addr, n,
1497                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0,
1498                                 pages);
1499                 if (unlikely(res < 0))
1500                         return res;
1501                 return (res == n ? len : res * PAGE_SIZE) - *start;
1502         0;}),({
1503                 /* can't be more than PAGE_SIZE */
1504                 *start = v.bv_offset;
1505                 get_page(*pages = v.bv_page);
1506                 return v.bv_len;
1507         }),({
1508                 return -EFAULT;
1509         }),
1510         0
1511         )
1512         return 0;
1513 }
1514 EXPORT_SYMBOL(iov_iter_get_pages);
1515
1516 static struct page **get_pages_array(size_t n)
1517 {
1518         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1519 }
1520
1521 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1522                    struct page ***pages, size_t maxsize,
1523                    size_t *start)
1524 {
1525         struct page **p;
1526         unsigned int iter_head, npages;
1527         ssize_t n;
1528
1529         if (!maxsize)
1530                 return 0;
1531
1532         if (!sanity(i))
1533                 return -EFAULT;
1534
1535         data_start(i, &iter_head, start);
1536         /* Amount of free space: some of this one + all after this one */
1537         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1538         n = npages * PAGE_SIZE - *start;
1539         if (maxsize > n)
1540                 maxsize = n;
1541         else
1542                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1543         p = get_pages_array(npages);
1544         if (!p)
1545                 return -ENOMEM;
1546         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1547         if (n > 0)
1548                 *pages = p;
1549         else
1550                 kvfree(p);
1551         return n;
1552 }
1553
1554 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1555                                            struct page ***pages, size_t maxsize,
1556                                            size_t *_start_offset)
1557 {
1558         struct page **p;
1559         unsigned nr, offset;
1560         pgoff_t index, count;
1561         size_t size = maxsize, actual;
1562         loff_t pos;
1563
1564         if (!size)
1565                 return 0;
1566
1567         pos = i->xarray_start + i->iov_offset;
1568         index = pos >> PAGE_SHIFT;
1569         offset = pos & ~PAGE_MASK;
1570         *_start_offset = offset;
1571
1572         count = 1;
1573         if (size > PAGE_SIZE - offset) {
1574                 size -= PAGE_SIZE - offset;
1575                 count += size >> PAGE_SHIFT;
1576                 size &= ~PAGE_MASK;
1577                 if (size)
1578                         count++;
1579         }
1580
1581         p = get_pages_array(count);
1582         if (!p)
1583                 return -ENOMEM;
1584         *pages = p;
1585
1586         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1587         if (nr == 0)
1588                 return 0;
1589
1590         actual = PAGE_SIZE * nr;
1591         actual -= offset;
1592         if (nr == count && size > 0) {
1593                 unsigned last_offset = (nr > 1) ? 0 : offset;
1594                 actual -= PAGE_SIZE - (last_offset + size);
1595         }
1596         return actual;
1597 }
1598
1599 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1600                    struct page ***pages, size_t maxsize,
1601                    size_t *start)
1602 {
1603         struct page **p;
1604
1605         if (maxsize > i->count)
1606                 maxsize = i->count;
1607
1608         if (unlikely(iov_iter_is_pipe(i)))
1609                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1610         if (unlikely(iov_iter_is_xarray(i)))
1611                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1612         if (unlikely(iov_iter_is_discard(i)))
1613                 return -EFAULT;
1614
1615         iterate_all_kinds(i, maxsize, v, ({
1616                 unsigned long addr = (unsigned long)v.iov_base;
1617                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1618                 int n;
1619                 int res;
1620
1621                 addr &= ~(PAGE_SIZE - 1);
1622                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1623                 p = get_pages_array(n);
1624                 if (!p)
1625                         return -ENOMEM;
1626                 res = get_user_pages_fast(addr, n,
1627                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0, p);
1628                 if (unlikely(res < 0)) {
1629                         kvfree(p);
1630                         return res;
1631                 }
1632                 *pages = p;
1633                 return (res == n ? len : res * PAGE_SIZE) - *start;
1634         0;}),({
1635                 /* can't be more than PAGE_SIZE */
1636                 *start = v.bv_offset;
1637                 *pages = p = get_pages_array(1);
1638                 if (!p)
1639                         return -ENOMEM;
1640                 get_page(*p = v.bv_page);
1641                 return v.bv_len;
1642         }),({
1643                 return -EFAULT;
1644         }), 0
1645         )
1646         return 0;
1647 }
1648 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1649
1650 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1651                                struct iov_iter *i)
1652 {
1653         char *to = addr;
1654         __wsum sum, next;
1655         size_t off = 0;
1656         sum = *csum;
1657         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1658                 WARN_ON(1);
1659                 return 0;
1660         }
1661         iterate_and_advance(i, bytes, v, ({
1662                 next = csum_and_copy_from_user(v.iov_base,
1663                                                (to += v.iov_len) - v.iov_len,
1664                                                v.iov_len);
1665                 if (next) {
1666                         sum = csum_block_add(sum, next, off);
1667                         off += v.iov_len;
1668                 }
1669                 next ? 0 : v.iov_len;
1670         }), ({
1671                 char *p = kmap_atomic(v.bv_page);
1672                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1673                                       p + v.bv_offset, v.bv_len,
1674                                       sum, off);
1675                 kunmap_atomic(p);
1676                 off += v.bv_len;
1677         }),({
1678                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1679                                       v.iov_base, v.iov_len,
1680                                       sum, off);
1681                 off += v.iov_len;
1682         }), ({
1683                 char *p = kmap_atomic(v.bv_page);
1684                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1685                                       p + v.bv_offset, v.bv_len,
1686                                       sum, off);
1687                 kunmap_atomic(p);
1688                 off += v.bv_len;
1689         })
1690         )
1691         *csum = sum;
1692         return bytes;
1693 }
1694 EXPORT_SYMBOL(csum_and_copy_from_iter);
1695
1696 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1697                              struct iov_iter *i)
1698 {
1699         struct csum_state *csstate = _csstate;
1700         const char *from = addr;
1701         __wsum sum, next;
1702         size_t off;
1703
1704         if (unlikely(iov_iter_is_pipe(i)))
1705                 return csum_and_copy_to_pipe_iter(addr, bytes, _csstate, i);
1706
1707         sum = csstate->csum;
1708         off = csstate->off;
1709         if (unlikely(iov_iter_is_discard(i))) {
1710                 WARN_ON(1);     /* for now */
1711                 return 0;
1712         }
1713         iterate_and_advance(i, bytes, v, ({
1714                 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1715                                              v.iov_base,
1716                                              v.iov_len);
1717                 if (next) {
1718                         sum = csum_block_add(sum, next, off);
1719                         off += v.iov_len;
1720                 }
1721                 next ? 0 : v.iov_len;
1722         }), ({
1723                 char *p = kmap_atomic(v.bv_page);
1724                 sum = csum_and_memcpy(p + v.bv_offset,
1725                                       (from += v.bv_len) - v.bv_len,
1726                                       v.bv_len, sum, off);
1727                 kunmap_atomic(p);
1728                 off += v.bv_len;
1729         }),({
1730                 sum = csum_and_memcpy(v.iov_base,
1731                                      (from += v.iov_len) - v.iov_len,
1732                                      v.iov_len, sum, off);
1733                 off += v.iov_len;
1734         }), ({
1735                 char *p = kmap_atomic(v.bv_page);
1736                 sum = csum_and_memcpy(p + v.bv_offset,
1737                                       (from += v.bv_len) - v.bv_len,
1738                                       v.bv_len, sum, off);
1739                 kunmap_atomic(p);
1740                 off += v.bv_len;
1741         })
1742         )
1743         csstate->csum = sum;
1744         csstate->off = off;
1745         return bytes;
1746 }
1747 EXPORT_SYMBOL(csum_and_copy_to_iter);
1748
1749 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1750                 struct iov_iter *i)
1751 {
1752 #ifdef CONFIG_CRYPTO_HASH
1753         struct ahash_request *hash = hashp;
1754         struct scatterlist sg;
1755         size_t copied;
1756
1757         copied = copy_to_iter(addr, bytes, i);
1758         sg_init_one(&sg, addr, copied);
1759         ahash_request_set_crypt(hash, &sg, NULL, copied);
1760         crypto_ahash_update(hash);
1761         return copied;
1762 #else
1763         return 0;
1764 #endif
1765 }
1766 EXPORT_SYMBOL(hash_and_copy_to_iter);
1767
1768 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1769 {
1770         size_t size = i->count;
1771         int npages = 0;
1772
1773         if (!size)
1774                 return 0;
1775         if (unlikely(iov_iter_is_discard(i)))
1776                 return 0;
1777
1778         if (unlikely(iov_iter_is_pipe(i))) {
1779                 struct pipe_inode_info *pipe = i->pipe;
1780                 unsigned int iter_head;
1781                 size_t off;
1782
1783                 if (!sanity(i))
1784                         return 0;
1785
1786                 data_start(i, &iter_head, &off);
1787                 /* some of this one + all after this one */
1788                 npages = pipe_space_for_user(iter_head, pipe->tail, pipe);
1789                 if (npages >= maxpages)
1790                         return maxpages;
1791         } else if (unlikely(iov_iter_is_xarray(i))) {
1792                 unsigned offset;
1793
1794                 offset = (i->xarray_start + i->iov_offset) & ~PAGE_MASK;
1795
1796                 npages = 1;
1797                 if (size > PAGE_SIZE - offset) {
1798                         size -= PAGE_SIZE - offset;
1799                         npages += size >> PAGE_SHIFT;
1800                         size &= ~PAGE_MASK;
1801                         if (size)
1802                                 npages++;
1803                 }
1804                 if (npages >= maxpages)
1805                         return maxpages;
1806         } else iterate_all_kinds(i, size, v, ({
1807                 unsigned long p = (unsigned long)v.iov_base;
1808                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1809                         - p / PAGE_SIZE;
1810                 if (npages >= maxpages)
1811                         return maxpages;
1812         0;}),({
1813                 npages++;
1814                 if (npages >= maxpages)
1815                         return maxpages;
1816         }),({
1817                 unsigned long p = (unsigned long)v.iov_base;
1818                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1819                         - p / PAGE_SIZE;
1820                 if (npages >= maxpages)
1821                         return maxpages;
1822         }),
1823         0
1824         )
1825         return npages;
1826 }
1827 EXPORT_SYMBOL(iov_iter_npages);
1828
1829 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1830 {
1831         *new = *old;
1832         if (unlikely(iov_iter_is_pipe(new))) {
1833                 WARN_ON(1);
1834                 return NULL;
1835         }
1836         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1837                 return NULL;
1838         if (iov_iter_is_bvec(new))
1839                 return new->bvec = kmemdup(new->bvec,
1840                                     new->nr_segs * sizeof(struct bio_vec),
1841                                     flags);
1842         else
1843                 /* iovec and kvec have identical layout */
1844                 return new->iov = kmemdup(new->iov,
1845                                    new->nr_segs * sizeof(struct iovec),
1846                                    flags);
1847 }
1848 EXPORT_SYMBOL(dup_iter);
1849
1850 static int copy_compat_iovec_from_user(struct iovec *iov,
1851                 const struct iovec __user *uvec, unsigned long nr_segs)
1852 {
1853         const struct compat_iovec __user *uiov =
1854                 (const struct compat_iovec __user *)uvec;
1855         int ret = -EFAULT, i;
1856
1857         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1858                 return -EFAULT;
1859
1860         for (i = 0; i < nr_segs; i++) {
1861                 compat_uptr_t buf;
1862                 compat_ssize_t len;
1863
1864                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1865                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1866
1867                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1868                 if (len < 0) {
1869                         ret = -EINVAL;
1870                         goto uaccess_end;
1871                 }
1872                 iov[i].iov_base = compat_ptr(buf);
1873                 iov[i].iov_len = len;
1874         }
1875
1876         ret = 0;
1877 uaccess_end:
1878         user_access_end();
1879         return ret;
1880 }
1881
1882 static int copy_iovec_from_user(struct iovec *iov,
1883                 const struct iovec __user *uvec, unsigned long nr_segs)
1884 {
1885         unsigned long seg;
1886
1887         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1888                 return -EFAULT;
1889         for (seg = 0; seg < nr_segs; seg++) {
1890                 if ((ssize_t)iov[seg].iov_len < 0)
1891                         return -EINVAL;
1892         }
1893
1894         return 0;
1895 }
1896
1897 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1898                 unsigned long nr_segs, unsigned long fast_segs,
1899                 struct iovec *fast_iov, bool compat)
1900 {
1901         struct iovec *iov = fast_iov;
1902         int ret;
1903
1904         /*
1905          * SuS says "The readv() function *may* fail if the iovcnt argument was
1906          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1907          * traditionally returned zero for zero segments, so...
1908          */
1909         if (nr_segs == 0)
1910                 return iov;
1911         if (nr_segs > UIO_MAXIOV)
1912                 return ERR_PTR(-EINVAL);
1913         if (nr_segs > fast_segs) {
1914                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1915                 if (!iov)
1916                         return ERR_PTR(-ENOMEM);
1917         }
1918
1919         if (compat)
1920                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1921         else
1922                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1923         if (ret) {
1924                 if (iov != fast_iov)
1925                         kfree(iov);
1926                 return ERR_PTR(ret);
1927         }
1928
1929         return iov;
1930 }
1931
1932 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1933                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1934                  struct iov_iter *i, bool compat)
1935 {
1936         ssize_t total_len = 0;
1937         unsigned long seg;
1938         struct iovec *iov;
1939
1940         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1941         if (IS_ERR(iov)) {
1942                 *iovp = NULL;
1943                 return PTR_ERR(iov);
1944         }
1945
1946         /*
1947          * According to the Single Unix Specification we should return EINVAL if
1948          * an element length is < 0 when cast to ssize_t or if the total length
1949          * would overflow the ssize_t return value of the system call.
1950          *
1951          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1952          * overflow case.
1953          */
1954         for (seg = 0; seg < nr_segs; seg++) {
1955                 ssize_t len = (ssize_t)iov[seg].iov_len;
1956
1957                 if (!access_ok(iov[seg].iov_base, len)) {
1958                         if (iov != *iovp)
1959                                 kfree(iov);
1960                         *iovp = NULL;
1961                         return -EFAULT;
1962                 }
1963
1964                 if (len > MAX_RW_COUNT - total_len) {
1965                         len = MAX_RW_COUNT - total_len;
1966                         iov[seg].iov_len = len;
1967                 }
1968                 total_len += len;
1969         }
1970
1971         iov_iter_init(i, type, iov, nr_segs, total_len);
1972         if (iov == *iovp)
1973                 *iovp = NULL;
1974         else
1975                 *iovp = iov;
1976         return total_len;
1977 }
1978
1979 /**
1980  * import_iovec() - Copy an array of &struct iovec from userspace
1981  *     into the kernel, check that it is valid, and initialize a new
1982  *     &struct iov_iter iterator to access it.
1983  *
1984  * @type: One of %READ or %WRITE.
1985  * @uvec: Pointer to the userspace array.
1986  * @nr_segs: Number of elements in userspace array.
1987  * @fast_segs: Number of elements in @iov.
1988  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1989  *     on-stack) kernel array.
1990  * @i: Pointer to iterator that will be initialized on success.
1991  *
1992  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1993  * then this function places %NULL in *@iov on return. Otherwise, a new
1994  * array will be allocated and the result placed in *@iov. This means that
1995  * the caller may call kfree() on *@iov regardless of whether the small
1996  * on-stack array was used or not (and regardless of whether this function
1997  * returns an error or not).
1998  *
1999  * Return: Negative error code on error, bytes imported on success
2000  */
2001 ssize_t import_iovec(int type, const struct iovec __user *uvec,
2002                  unsigned nr_segs, unsigned fast_segs,
2003                  struct iovec **iovp, struct iov_iter *i)
2004 {
2005         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
2006                               in_compat_syscall());
2007 }
2008 EXPORT_SYMBOL(import_iovec);
2009
2010 int import_single_range(int rw, void __user *buf, size_t len,
2011                  struct iovec *iov, struct iov_iter *i)
2012 {
2013         if (len > MAX_RW_COUNT)
2014                 len = MAX_RW_COUNT;
2015         if (unlikely(!access_ok(buf, len)))
2016                 return -EFAULT;
2017
2018         iov->iov_base = buf;
2019         iov->iov_len = len;
2020         iov_iter_init(i, rw, iov, 1, len);
2021         return 0;
2022 }
2023 EXPORT_SYMBOL(import_single_range);