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 /* covers iovec and kvec alike */
  20 #define iterate_iovec(i, n, base, len, off, __p, STEP) {        \
  21         size_t off = 0;                                         \
  22         size_t skip = i->iov_offset;                            \
  23         do {                                                    \
  24                 len = min(n, __p->iov_len - skip);              \
  25                 if (likely(len)) {                              \
  26                         base = __p->iov_base + skip;            \
  27                         len -= (STEP);                          \
  28                         off += len;                             \
  29                         skip += len;                            \
  30                         n -= len;                               \
  31                         if (skip < __p->iov_len)                \
  32                                 break;                          \
  33                 }                                               \
  34                 __p++;                                          \
  35                 skip = 0;                                       \
  36         } while (n);                                            \
  37         i->iov_offset = skip;                                   \
  38         n = off;                                                \
  39 }
  40
  41 #define iterate_bvec(i, n, base, len, off, p, STEP) {           \
  42         size_t off = 0;                                         \
  43         unsigned skip = i->iov_offset;                          \
  44         while (n) {                                             \
  45                 unsigned offset = p->bv_offset + skip;          \
  46                 unsigned left;                                  \
  47                 void *kaddr = kmap_local_page(p->bv_page +      \
  48                                         offset / PAGE_SIZE);    \
  49                 base = kaddr + offset % PAGE_SIZE;              \
  50                 len = min(min(n, (size_t)(p->bv_len - skip)),   \
  51                      (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
  52                 left = (STEP);                                  \
  53                 kunmap_local(kaddr);                            \
  54                 len -= left;                                    \
  55                 off += len;                                     \
  56                 skip += len;                                    \
  57                 if (skip == p->bv_len) {                        \
  58                         skip = 0;                               \
  59                         p++;                                    \
  60                 }                                               \
  61                 n -= len;                                       \
  62                 if (left)                                       \
  63                         break;                                  \
  64         }                                                       \
  65         i->iov_offset = skip;                                   \
  66         n = off;                                                \
  67 }
  68
  69 #define iterate_xarray(i, n, base, len, __off, STEP) {          \
  70         __label__ __out;                                        \
  71         size_t __off = 0;                                       \
  72         struct folio *folio;                                    \
  73         loff_t start = i->xarray_start + i->iov_offset;         \
  74         pgoff_t index = start / PAGE_SIZE;                      \
  75         XA_STATE(xas, i->xarray, index);                        \
  76                                                                 \
  77         len = PAGE_SIZE - offset_in_page(start);                \
  78         rcu_read_lock();                                        \
  79         xas_for_each(&xas, folio, ULONG_MAX) {                  \
  80                 unsigned left;                                  \
  81                 size_t offset;                                  \
  82                 if (xas_retry(&xas, folio))                     \
  83                         continue;                               \
  84                 if (WARN_ON(xa_is_value(folio)))                \
  85                         break;                                  \
  86                 if (WARN_ON(folio_test_hugetlb(folio)))         \
  87                         break;                                  \
  88                 offset = offset_in_folio(folio, start + __off); \
  89                 while (offset < folio_size(folio)) {            \
  90                         base = kmap_local_folio(folio, offset); \
  91                         len = min(n, len);                      \
  92                         left = (STEP);                          \
  93                         kunmap_local(base);                     \
  94                         len -= left;                            \
  95                         __off += len;                           \
  96                         n -= len;                               \
  97                         if (left || n == 0)                     \
  98                                 goto __out;                     \
  99                         offset += len;                          \
 100                         len = PAGE_SIZE;                        \
 101                 }                                               \
 102         }                                                       \
 103 __out:                                                          \
 104         rcu_read_unlock();                                      \
 105         i->iov_offset += __off;                                 \
 106         n = __off;                                              \
 107 }
 108
 109 #define __iterate_and_advance(i, n, base, len, off, I, K) {     \
 110         if (unlikely(i->count < n))                             \
 111                 n = i->count;                                   \
 112         if (likely(n)) {                                        \
 113                 if (likely(iter_is_iovec(i))) {                 \
 114                         const struct iovec *iov = i->iov;       \
 115                         void __user *base;                      \
 116                         size_t len;                             \
 117                         iterate_iovec(i, n, base, len, off,     \
 118                                                 iov, (I))       \
 119                         i->nr_segs -= iov - i->iov;             \
 120                         i->iov = iov;                           \
 121                 } else if (iov_iter_is_bvec(i)) {               \
 122                         const struct bio_vec *bvec = i->bvec;   \
 123                         void *base;                             \
 124                         size_t len;                             \
 125                         iterate_bvec(i, n, base, len, off,      \
 126                                                 bvec, (K))      \
 127                         i->nr_segs -= bvec - i->bvec;           \
 128                         i->bvec = bvec;                         \
 129                 } else if (iov_iter_is_kvec(i)) {               \
 130                         const struct kvec *kvec = i->kvec;      \
 131                         void *base;                             \
 132                         size_t len;                             \
 133                         iterate_iovec(i, n, base, len, off,     \
 134                                                 kvec, (K))      \
 135                         i->nr_segs -= kvec - i->kvec;           \
 136                         i->kvec = kvec;                         \
 137                 } else if (iov_iter_is_xarray(i)) {             \
 138                         void *base;                             \
 139                         size_t len;                             \
 140                         iterate_xarray(i, n, base, len, off,    \
 141                                                         (K))    \
 142                 }                                               \
 143                 i->count -= n;                                  \
 144         }                                                       \
 145 }
 146 #define iterate_and_advance(i, n, base, len, off, I, K) \
 147         __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
 148
 149 static int copyout(void __user *to, const void *from, size_t n)
 150 {
 151         if (should_fail_usercopy())
 152                 return n;
 153         if (access_ok(to, n)) {
 154                 instrument_copy_to_user(to, from, n);
 155                 n = raw_copy_to_user(to, from, n);
 156         }
 157         return n;
 158 }
 159
 160 static int copyin(void *to, const void __user *from, size_t n)
 161 {
 162         if (should_fail_usercopy())
 163                 return n;
 164         if (access_ok(from, n)) {
 165                 instrument_copy_from_user(to, from, n);
 166                 n = raw_copy_from_user(to, from, n);
 167         }
 168         return n;
 169 }
 170
 171 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 172                          struct iov_iter *i)
 173 {
 174         size_t skip, copy, left, wanted;
 175         const struct iovec *iov;
 176         char __user *buf;
 177         void *kaddr, *from;
 178
 179         if (unlikely(bytes > i->count))
 180                 bytes = i->count;
 181
 182         if (unlikely(!bytes))
 183                 return 0;
 184
 185         might_fault();
 186         wanted = bytes;
 187         iov = i->iov;
 188         skip = i->iov_offset;
 189         buf = iov->iov_base + skip;
 190         copy = min(bytes, iov->iov_len - skip);
 191
 192         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_writeable(buf, copy)) {
 193                 kaddr = kmap_atomic(page);
 194                 from = kaddr + offset;
 195
 196                 /* first chunk, usually the only one */
 197                 left = copyout(buf, from, copy);
 198                 copy -= left;
 199                 skip += copy;
 200                 from += copy;
 201                 bytes -= copy;
 202
 203                 while (unlikely(!left && bytes)) {
 204                         iov++;
 205                         buf = iov->iov_base;
 206                         copy = min(bytes, iov->iov_len);
 207                         left = copyout(buf, from, copy);
 208                         copy -= left;
 209                         skip = copy;
 210                         from += copy;
 211                         bytes -= copy;
 212                 }
 213                 if (likely(!bytes)) {
 214                         kunmap_atomic(kaddr);
 215                         goto done;
 216                 }
 217                 offset = from - kaddr;
 218                 buf += copy;
 219                 kunmap_atomic(kaddr);
 220                 copy = min(bytes, iov->iov_len - skip);
 221         }
 222         /* Too bad - revert to non-atomic kmap */
 223
 224         kaddr = kmap(page);
 225         from = kaddr + offset;
 226         left = copyout(buf, from, copy);
 227         copy -= left;
 228         skip += copy;
 229         from += copy;
 230         bytes -= copy;
 231         while (unlikely(!left && bytes)) {
 232                 iov++;
 233                 buf = iov->iov_base;
 234                 copy = min(bytes, iov->iov_len);
 235                 left = copyout(buf, from, copy);
 236                 copy -= left;
 237                 skip = copy;
 238                 from += copy;
 239                 bytes -= copy;
 240         }
 241         kunmap(page);
 242
 243 done:
 244         if (skip == iov->iov_len) {
 245                 iov++;
 246                 skip = 0;
 247         }
 248         i->count -= wanted - bytes;
 249         i->nr_segs -= iov - i->iov;
 250         i->iov = iov;
 251         i->iov_offset = skip;
 252         return wanted - bytes;
 253 }
 254
 255 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 256                          struct iov_iter *i)
 257 {
 258         size_t skip, copy, left, wanted;
 259         const struct iovec *iov;
 260         char __user *buf;
 261         void *kaddr, *to;
 262
 263         if (unlikely(bytes > i->count))
 264                 bytes = i->count;
 265
 266         if (unlikely(!bytes))
 267                 return 0;
 268
 269         might_fault();
 270         wanted = bytes;
 271         iov = i->iov;
 272         skip = i->iov_offset;
 273         buf = iov->iov_base + skip;
 274         copy = min(bytes, iov->iov_len - skip);
 275
 276         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_readable(buf, copy)) {
 277                 kaddr = kmap_atomic(page);
 278                 to = kaddr + offset;
 279
 280                 /* first chunk, usually the only one */
 281                 left = copyin(to, buf, copy);
 282                 copy -= left;
 283                 skip += copy;
 284                 to += copy;
 285                 bytes -= copy;
 286
 287                 while (unlikely(!left && bytes)) {
 288                         iov++;
 289                         buf = iov->iov_base;
 290                         copy = min(bytes, iov->iov_len);
 291                         left = copyin(to, buf, copy);
 292                         copy -= left;
 293                         skip = copy;
 294                         to += copy;
 295                         bytes -= copy;
 296                 }
 297                 if (likely(!bytes)) {
 298                         kunmap_atomic(kaddr);
 299                         goto done;
 300                 }
 301                 offset = to - kaddr;
 302                 buf += copy;
 303                 kunmap_atomic(kaddr);
 304                 copy = min(bytes, iov->iov_len - skip);
 305         }
 306         /* Too bad - revert to non-atomic kmap */
 307
 308         kaddr = kmap(page);
 309         to = kaddr + offset;
 310         left = copyin(to, buf, copy);
 311         copy -= left;
 312         skip += copy;
 313         to += copy;
 314         bytes -= copy;
 315         while (unlikely(!left && bytes)) {
 316                 iov++;
 317                 buf = iov->iov_base;
 318                 copy = min(bytes, iov->iov_len);
 319                 left = copyin(to, buf, copy);
 320                 copy -= left;
 321                 skip = copy;
 322                 to += copy;
 323                 bytes -= copy;
 324         }
 325         kunmap(page);
 326
 327 done:
 328         if (skip == iov->iov_len) {
 329                 iov++;
 330                 skip = 0;
 331         }
 332         i->count -= wanted - bytes;
 333         i->nr_segs -= iov - i->iov;
 334         i->iov = iov;
 335         i->iov_offset = skip;
 336         return wanted - bytes;
 337 }
 338
 339 #ifdef PIPE_PARANOIA
 340 static bool sanity(const struct iov_iter *i)
 341 {
 342         struct pipe_inode_info *pipe = i->pipe;
 343         unsigned int p_head = pipe->head;
 344         unsigned int p_tail = pipe->tail;
 345         unsigned int p_mask = pipe->ring_size - 1;
 346         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 347         unsigned int i_head = i->head;
 348         unsigned int idx;
 349
 350         if (i->iov_offset) {
 351                 struct pipe_buffer *p;
 352                 if (unlikely(p_occupancy == 0))
 353                         goto Bad;       // pipe must be non-empty
 354                 if (unlikely(i_head != p_head - 1))
 355                         goto Bad;       // must be at the last buffer...
 356
 357                 p = &pipe->bufs[i_head & p_mask];
 358                 if (unlikely(p->offset + p->len != i->iov_offset))
 359                         goto Bad;       // ... at the end of segment
 360         } else {
 361                 if (i_head != p_head)
 362                         goto Bad;       // must be right after the last buffer
 363         }
 364         return true;
 365 Bad:
 366         printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
 367         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 368                         p_head, p_tail, pipe->ring_size);
 369         for (idx = 0; idx < pipe->ring_size; idx++)
 370                 printk(KERN_ERR "[%p %p %d %d]\n",
 371                         pipe->bufs[idx].ops,
 372                         pipe->bufs[idx].page,
 373                         pipe->bufs[idx].offset,
 374                         pipe->bufs[idx].len);
 375         WARN_ON(1);
 376         return false;
 377 }
 378 #else
 379 #define sanity(i) true
 380 #endif
 381
 382 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 383                          struct iov_iter *i)
 384 {
 385         struct pipe_inode_info *pipe = i->pipe;
 386         struct pipe_buffer *buf;
 387         unsigned int p_tail = pipe->tail;
 388         unsigned int p_mask = pipe->ring_size - 1;
 389         unsigned int i_head = i->head;
 390         size_t off;
 391
 392         if (unlikely(bytes > i->count))
 393                 bytes = i->count;
 394
 395         if (unlikely(!bytes))
 396                 return 0;
 397
 398         if (!sanity(i))
 399                 return 0;
 400
 401         off = i->iov_offset;
 402         buf = &pipe->bufs[i_head & p_mask];
 403         if (off) {
 404                 if (offset == off && buf->page == page) {
 405                         /* merge with the last one */
 406                         buf->len += bytes;
 407                         i->iov_offset += bytes;
 408                         goto out;
 409                 }
 410                 i_head++;
 411                 buf = &pipe->bufs[i_head & p_mask];
 412         }
 413         if (pipe_full(i_head, p_tail, pipe->max_usage))
 414                 return 0;
 415
 416         buf->ops = &page_cache_pipe_buf_ops;
 417         buf->flags = 0;
 418         get_page(page);
 419         buf->page = page;
 420         buf->offset = offset;
 421         buf->len = bytes;
 422
 423         pipe->head = i_head + 1;
 424         i->iov_offset = offset + bytes;
 425         i->head = i_head;
 426 out:
 427         i->count -= bytes;
 428         return bytes;
 429 }
 430
 431 /*
 432  * fault_in_iov_iter_readable - fault in iov iterator for reading
 433  * @i: iterator
 434  * @size: maximum length
 435  *
 436  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 437  * @size.  For each iovec, fault in each page that constitutes the iovec.
 438  *
 439  * Returns the number of bytes not faulted in (like copy_to_user() and
 440  * copy_from_user()).
 441  *
 442  * Always returns 0 for non-userspace iterators.
 443  */
 444 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
 445 {
 446         if (iter_is_iovec(i)) {
 447                 size_t count = min(size, iov_iter_count(i));
 448                 const struct iovec *p;
 449                 size_t skip;
 450
 451                 size -= count;
 452                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 453                         size_t len = min(count, p->iov_len - skip);
 454                         size_t ret;
 455
 456                         if (unlikely(!len))
 457                                 continue;
 458                         ret = fault_in_readable(p->iov_base + skip, len);
 459                         count -= len - ret;
 460                         if (ret)
 461                                 break;
 462                 }
 463                 return count + size;
 464         }
 465         return 0;
 466 }
 467 EXPORT_SYMBOL(fault_in_iov_iter_readable);
 468
 469 /*
 470  * fault_in_iov_iter_writeable - fault in iov iterator for writing
 471  * @i: iterator
 472  * @size: maximum length
 473  *
 474  * Faults in the iterator using get_user_pages(), i.e., without triggering
 475  * hardware page faults.  This is primarily useful when we already know that
 476  * some or all of the pages in @i aren't in memory.
 477  *
 478  * Returns the number of bytes not faulted in, like copy_to_user() and
 479  * copy_from_user().
 480  *
 481  * Always returns 0 for non-user-space iterators.
 482  */
 483 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
 484 {
 485         if (iter_is_iovec(i)) {
 486                 size_t count = min(size, iov_iter_count(i));
 487                 const struct iovec *p;
 488                 size_t skip;
 489
 490                 size -= count;
 491                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 492                         size_t len = min(count, p->iov_len - skip);
 493                         size_t ret;
 494
 495                         if (unlikely(!len))
 496                                 continue;
 497                         ret = fault_in_safe_writeable(p->iov_base + skip, len);
 498                         count -= len - ret;
 499                         if (ret)
 500                                 break;
 501                 }
 502                 return count + size;
 503         }
 504         return 0;
 505 }
 506 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
 507
 508 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 509                         const struct iovec *iov, unsigned long nr_segs,
 510                         size_t count)
 511 {
 512         WARN_ON(direction & ~(READ | WRITE));
 513         *i = (struct iov_iter) {
 514                 .iter_type = ITER_IOVEC,
 515                 .nofault = false,
 516                 .data_source = direction,
 517                 .iov = iov,
 518                 .nr_segs = nr_segs,
 519                 .iov_offset = 0,
 520                 .count = count
 521         };
 522 }
 523 EXPORT_SYMBOL(iov_iter_init);
 524
 525 static inline bool allocated(struct pipe_buffer *buf)
 526 {
 527         return buf->ops == &default_pipe_buf_ops;
 528 }
 529
 530 static inline void data_start(const struct iov_iter *i,
 531                               unsigned int *iter_headp, size_t *offp)
 532 {
 533         unsigned int p_mask = i->pipe->ring_size - 1;
 534         unsigned int iter_head = i->head;
 535         size_t off = i->iov_offset;
 536
 537         if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
 538                     off == PAGE_SIZE)) {
 539                 iter_head++;
 540                 off = 0;
 541         }
 542         *iter_headp = iter_head;
 543         *offp = off;
 544 }
 545
 546 static size_t push_pipe(struct iov_iter *i, size_t size,
 547                         int *iter_headp, size_t *offp)
 548 {
 549         struct pipe_inode_info *pipe = i->pipe;
 550         unsigned int p_tail = pipe->tail;
 551         unsigned int p_mask = pipe->ring_size - 1;
 552         unsigned int iter_head;
 553         size_t off;
 554         ssize_t left;
 555
 556         if (unlikely(size > i->count))
 557                 size = i->count;
 558         if (unlikely(!size))
 559                 return 0;
 560
 561         left = size;
 562         data_start(i, &iter_head, &off);
 563         *iter_headp = iter_head;
 564         *offp = off;
 565         if (off) {
 566                 left -= PAGE_SIZE - off;
 567                 if (left <= 0) {
 568                         pipe->bufs[iter_head & p_mask].len += size;
 569                         return size;
 570                 }
 571                 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
 572                 iter_head++;
 573         }
 574         while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
 575                 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
 576                 struct page *page = alloc_page(GFP_USER);
 577                 if (!page)
 578                         break;
 579
 580                 buf->ops = &default_pipe_buf_ops;
 581                 buf->flags = 0;
 582                 buf->page = page;
 583                 buf->offset = 0;
 584                 buf->len = min_t(ssize_t, left, PAGE_SIZE);
 585                 left -= buf->len;
 586                 iter_head++;
 587                 pipe->head = iter_head;
 588
 589                 if (left == 0)
 590                         return size;
 591         }
 592         return size - left;
 593 }
 594
 595 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 596                                 struct iov_iter *i)
 597 {
 598         struct pipe_inode_info *pipe = i->pipe;
 599         unsigned int p_mask = pipe->ring_size - 1;
 600         unsigned int i_head;
 601         size_t n, off;
 602
 603         if (!sanity(i))
 604                 return 0;
 605
 606         bytes = n = push_pipe(i, bytes, &i_head, &off);
 607         if (unlikely(!n))
 608                 return 0;
 609         do {
 610                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 611                 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
 612                 i->head = i_head;
 613                 i->iov_offset = off + chunk;
 614                 n -= chunk;
 615                 addr += chunk;
 616                 off = 0;
 617                 i_head++;
 618         } while (n);
 619         i->count -= bytes;
 620         return bytes;
 621 }
 622
 623 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 624                               __wsum sum, size_t off)
 625 {
 626         __wsum next = csum_partial_copy_nocheck(from, to, len);
 627         return csum_block_add(sum, next, off);
 628 }
 629
 630 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 631                                          struct iov_iter *i, __wsum *sump)
 632 {
 633         struct pipe_inode_info *pipe = i->pipe;
 634         unsigned int p_mask = pipe->ring_size - 1;
 635         __wsum sum = *sump;
 636         size_t off = 0;
 637         unsigned int i_head;
 638         size_t r;
 639
 640         if (!sanity(i))
 641                 return 0;
 642
 643         bytes = push_pipe(i, bytes, &i_head, &r);
 644         while (bytes) {
 645                 size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r);
 646                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 647                 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
 648                 kunmap_local(p);
 649                 i->head = i_head;
 650                 i->iov_offset = r + chunk;
 651                 bytes -= chunk;
 652                 off += chunk;
 653                 r = 0;
 654                 i_head++;
 655         }
 656         *sump = sum;
 657         i->count -= off;
 658         return off;
 659 }
 660
 661 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 662 {
 663         if (unlikely(iov_iter_is_pipe(i)))
 664                 return copy_pipe_to_iter(addr, bytes, i);
 665         if (iter_is_iovec(i))
 666                 might_fault();
 667         iterate_and_advance(i, bytes, base, len, off,
 668                 copyout(base, addr + off, len),
 669                 memcpy(base, addr + off, len)
 670         )
 671
 672         return bytes;
 673 }
 674 EXPORT_SYMBOL(_copy_to_iter);
 675
 676 #ifdef CONFIG_ARCH_HAS_COPY_MC
 677 static int copyout_mc(void __user *to, const void *from, size_t n)
 678 {
 679         if (access_ok(to, n)) {
 680                 instrument_copy_to_user(to, from, n);
 681                 n = copy_mc_to_user((__force void *) to, from, n);
 682         }
 683         return n;
 684 }
 685
 686 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 687                                 struct iov_iter *i)
 688 {
 689         struct pipe_inode_info *pipe = i->pipe;
 690         unsigned int p_mask = pipe->ring_size - 1;
 691         unsigned int i_head;
 692         size_t n, off, xfer = 0;
 693
 694         if (!sanity(i))
 695                 return 0;
 696
 697         n = push_pipe(i, bytes, &i_head, &off);
 698         while (n) {
 699                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 700                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 701                 unsigned long rem;
 702                 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
 703                 chunk -= rem;
 704                 kunmap_local(p);
 705                 i->head = i_head;
 706                 i->iov_offset = off + chunk;
 707                 xfer += chunk;
 708                 if (rem)
 709                         break;
 710                 n -= chunk;
 711                 off = 0;
 712                 i_head++;
 713         }
 714         i->count -= xfer;
 715         return xfer;
 716 }
 717
 718 /**
 719  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 720  * @addr: source kernel address
 721  * @bytes: total transfer length
 722  * @i: destination iterator
 723  *
 724  * The pmem driver deploys this for the dax operation
 725  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 726  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 727  * successfully copied.
 728  *
 729  * The main differences between this and typical _copy_to_iter().
 730  *
 731  * * Typical tail/residue handling after a fault retries the copy
 732  *   byte-by-byte until the fault happens again. Re-triggering machine
 733  *   checks is potentially fatal so the implementation uses source
 734  *   alignment and poison alignment assumptions to avoid re-triggering
 735  *   hardware exceptions.
 736  *
 737  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 738  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 739  *   a short copy.
 740  *
 741  * Return: number of bytes copied (may be %0)
 742  */
 743 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 744 {
 745         if (unlikely(iov_iter_is_pipe(i)))
 746                 return copy_mc_pipe_to_iter(addr, bytes, i);
 747         if (iter_is_iovec(i))
 748                 might_fault();
 749         __iterate_and_advance(i, bytes, base, len, off,
 750                 copyout_mc(base, addr + off, len),
 751                 copy_mc_to_kernel(base, addr + off, len)
 752         )
 753
 754         return bytes;
 755 }
 756 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 757 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 758
 759 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 760 {
 761         if (unlikely(iov_iter_is_pipe(i))) {
 762                 WARN_ON(1);
 763                 return 0;
 764         }
 765         if (iter_is_iovec(i))
 766                 might_fault();
 767         iterate_and_advance(i, bytes, base, len, off,
 768                 copyin(addr + off, base, len),
 769                 memcpy(addr + off, base, len)
 770         )
 771
 772         return bytes;
 773 }
 774 EXPORT_SYMBOL(_copy_from_iter);
 775
 776 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 777 {
 778         if (unlikely(iov_iter_is_pipe(i))) {
 779                 WARN_ON(1);
 780                 return 0;
 781         }
 782         iterate_and_advance(i, bytes, base, len, off,
 783                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 784                 memcpy(addr + off, base, len)
 785         )
 786
 787         return bytes;
 788 }
 789 EXPORT_SYMBOL(_copy_from_iter_nocache);
 790
 791 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 792 /**
 793  * _copy_from_iter_flushcache - write destination through cpu cache
 794  * @addr: destination kernel address
 795  * @bytes: total transfer length
 796  * @i: source iterator
 797  *
 798  * The pmem driver arranges for filesystem-dax to use this facility via
 799  * dax_copy_from_iter() for ensuring that writes to persistent memory
 800  * are flushed through the CPU cache. It is differentiated from
 801  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 802  * all iterator types. The _copy_from_iter_nocache() only attempts to
 803  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 804  * instructions that strand dirty-data in the cache.
 805  *
 806  * Return: number of bytes copied (may be %0)
 807  */
 808 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 809 {
 810         if (unlikely(iov_iter_is_pipe(i))) {
 811                 WARN_ON(1);
 812                 return 0;
 813         }
 814         iterate_and_advance(i, bytes, base, len, off,
 815                 __copy_from_user_flushcache(addr + off, base, len),
 816                 memcpy_flushcache(addr + off, base, len)
 817         )
 818
 819         return bytes;
 820 }
 821 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 822 #endif
 823
 824 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 825 {
 826         struct page *head;
 827         size_t v = n + offset;
 828
 829         /*
 830          * The general case needs to access the page order in order
 831          * to compute the page size.
 832          * However, we mostly deal with order-0 pages and thus can
 833          * avoid a possible cache line miss for requests that fit all
 834          * page orders.
 835          */
 836         if (n <= v && v <= PAGE_SIZE)
 837                 return true;
 838
 839         head = compound_head(page);
 840         v += (page - head) << PAGE_SHIFT;
 841
 842         if (likely(n <= v && v <= (page_size(head))))
 843                 return true;
 844         WARN_ON(1);
 845         return false;
 846 }
 847
 848 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 849                          struct iov_iter *i)
 850 {
 851         if (likely(iter_is_iovec(i)))
 852                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 853         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 854                 void *kaddr = kmap_local_page(page);
 855                 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
 856                 kunmap_local(kaddr);
 857                 return wanted;
 858         }
 859         if (iov_iter_is_pipe(i))
 860                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 861         if (unlikely(iov_iter_is_discard(i))) {
 862                 if (unlikely(i->count < bytes))
 863                         bytes = i->count;
 864                 i->count -= bytes;
 865                 return bytes;
 866         }
 867         WARN_ON(1);
 868         return 0;
 869 }
 870
 871 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 872                          struct iov_iter *i)
 873 {
 874         size_t res = 0;
 875         if (unlikely(!page_copy_sane(page, offset, bytes)))
 876                 return 0;
 877         page += offset / PAGE_SIZE; // first subpage
 878         offset %= PAGE_SIZE;
 879         while (1) {
 880                 size_t n = __copy_page_to_iter(page, offset,
 881                                 min(bytes, (size_t)PAGE_SIZE - offset), i);
 882                 res += n;
 883                 bytes -= n;
 884                 if (!bytes || !n)
 885                         break;
 886                 offset += n;
 887                 if (offset == PAGE_SIZE) {
 888                         page++;
 889                         offset = 0;
 890                 }
 891         }
 892         return res;
 893 }
 894 EXPORT_SYMBOL(copy_page_to_iter);
 895
 896 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 897                          struct iov_iter *i)
 898 {
 899         if (unlikely(!page_copy_sane(page, offset, bytes)))
 900                 return 0;
 901         if (likely(iter_is_iovec(i)))
 902                 return copy_page_from_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_local_page(page);
 905                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 906                 kunmap_local(kaddr);
 907                 return wanted;
 908         }
 909         WARN_ON(1);
 910         return 0;
 911 }
 912 EXPORT_SYMBOL(copy_page_from_iter);
 913
 914 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 915 {
 916         struct pipe_inode_info *pipe = i->pipe;
 917         unsigned int p_mask = pipe->ring_size - 1;
 918         unsigned int i_head;
 919         size_t n, off;
 920
 921         if (!sanity(i))
 922                 return 0;
 923
 924         bytes = n = push_pipe(i, bytes, &i_head, &off);
 925         if (unlikely(!n))
 926                 return 0;
 927
 928         do {
 929                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 930                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 931                 memset(p + off, 0, chunk);
 932                 kunmap_local(p);
 933                 i->head = i_head;
 934                 i->iov_offset = off + chunk;
 935                 n -= chunk;
 936                 off = 0;
 937                 i_head++;
 938         } while (n);
 939         i->count -= bytes;
 940         return bytes;
 941 }
 942
 943 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 944 {
 945         if (unlikely(iov_iter_is_pipe(i)))
 946                 return pipe_zero(bytes, i);
 947         iterate_and_advance(i, bytes, base, len, count,
 948                 clear_user(base, len),
 949                 memset(base, 0, len)
 950         )
 951
 952         return bytes;
 953 }
 954 EXPORT_SYMBOL(iov_iter_zero);
 955
 956 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 957                                   struct iov_iter *i)
 958 {
 959         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 960         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 961                 kunmap_atomic(kaddr);
 962                 return 0;
 963         }
 964         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 965                 kunmap_atomic(kaddr);
 966                 WARN_ON(1);
 967                 return 0;
 968         }
 969         iterate_and_advance(i, bytes, base, len, off,
 970                 copyin(p + off, base, len),
 971                 memcpy(p + off, base, len)
 972         )
 973         kunmap_atomic(kaddr);
 974         return bytes;
 975 }
 976 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 977
 978 static inline void pipe_truncate(struct iov_iter *i)
 979 {
 980         struct pipe_inode_info *pipe = i->pipe;
 981         unsigned int p_tail = pipe->tail;
 982         unsigned int p_head = pipe->head;
 983         unsigned int p_mask = pipe->ring_size - 1;
 984
 985         if (!pipe_empty(p_head, p_tail)) {
 986                 struct pipe_buffer *buf;
 987                 unsigned int i_head = i->head;
 988                 size_t off = i->iov_offset;
 989
 990                 if (off) {
 991                         buf = &pipe->bufs[i_head & p_mask];
 992                         buf->len = off - buf->offset;
 993                         i_head++;
 994                 }
 995                 while (p_head != i_head) {
 996                         p_head--;
 997                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
 998                 }
 999
1000                 pipe->head = p_head;
1001         }
1002 }
1003
1004 static void pipe_advance(struct iov_iter *i, size_t size)
1005 {
1006         struct pipe_inode_info *pipe = i->pipe;
1007         if (size) {
1008                 struct pipe_buffer *buf;
1009                 unsigned int p_mask = pipe->ring_size - 1;
1010                 unsigned int i_head = i->head;
1011                 size_t off = i->iov_offset, left = size;
1012
1013                 if (off) /* make it relative to the beginning of buffer */
1014                         left += off - pipe->bufs[i_head & p_mask].offset;
1015                 while (1) {
1016                         buf = &pipe->bufs[i_head & p_mask];
1017                         if (left <= buf->len)
1018                                 break;
1019                         left -= buf->len;
1020                         i_head++;
1021                 }
1022                 i->head = i_head;
1023                 i->iov_offset = buf->offset + left;
1024         }
1025         i->count -= size;
1026         /* ... and discard everything past that point */
1027         pipe_truncate(i);
1028 }
1029
1030 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1031 {
1032         struct bvec_iter bi;
1033
1034         bi.bi_size = i->count;
1035         bi.bi_bvec_done = i->iov_offset;
1036         bi.bi_idx = 0;
1037         bvec_iter_advance(i->bvec, &bi, size);
1038
1039         i->bvec += bi.bi_idx;
1040         i->nr_segs -= bi.bi_idx;
1041         i->count = bi.bi_size;
1042         i->iov_offset = bi.bi_bvec_done;
1043 }
1044
1045 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
1046 {
1047         const struct iovec *iov, *end;
1048
1049         if (!i->count)
1050                 return;
1051         i->count -= size;
1052
1053         size += i->iov_offset; // from beginning of current segment
1054         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1055                 if (likely(size < iov->iov_len))
1056                         break;
1057                 size -= iov->iov_len;
1058         }
1059         i->iov_offset = size;
1060         i->nr_segs -= iov - i->iov;
1061         i->iov = iov;
1062 }
1063
1064 void iov_iter_advance(struct iov_iter *i, size_t size)
1065 {
1066         if (unlikely(i->count < size))
1067                 size = i->count;
1068         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1069                 /* iovec and kvec have identical layouts */
1070                 iov_iter_iovec_advance(i, size);
1071         } else if (iov_iter_is_bvec(i)) {
1072                 iov_iter_bvec_advance(i, size);
1073         } else if (iov_iter_is_pipe(i)) {
1074                 pipe_advance(i, size);
1075         } else if (unlikely(iov_iter_is_xarray(i))) {
1076                 i->iov_offset += size;
1077                 i->count -= size;
1078         } else if (iov_iter_is_discard(i)) {
1079                 i->count -= size;
1080         }
1081 }
1082 EXPORT_SYMBOL(iov_iter_advance);
1083
1084 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1085 {
1086         if (!unroll)
1087                 return;
1088         if (WARN_ON(unroll > MAX_RW_COUNT))
1089                 return;
1090         i->count += unroll;
1091         if (unlikely(iov_iter_is_pipe(i))) {
1092                 struct pipe_inode_info *pipe = i->pipe;
1093                 unsigned int p_mask = pipe->ring_size - 1;
1094                 unsigned int i_head = i->head;
1095                 size_t off = i->iov_offset;
1096                 while (1) {
1097                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1098                         size_t n = off - b->offset;
1099                         if (unroll < n) {
1100                                 off -= unroll;
1101                                 break;
1102                         }
1103                         unroll -= n;
1104                         if (!unroll && i_head == i->start_head) {
1105                                 off = 0;
1106                                 break;
1107                         }
1108                         i_head--;
1109                         b = &pipe->bufs[i_head & p_mask];
1110                         off = b->offset + b->len;
1111                 }
1112                 i->iov_offset = off;
1113                 i->head = i_head;
1114                 pipe_truncate(i);
1115                 return;
1116         }
1117         if (unlikely(iov_iter_is_discard(i)))
1118                 return;
1119         if (unroll <= i->iov_offset) {
1120                 i->iov_offset -= unroll;
1121                 return;
1122         }
1123         unroll -= i->iov_offset;
1124         if (iov_iter_is_xarray(i)) {
1125                 BUG(); /* We should never go beyond the start of the specified
1126                         * range since we might then be straying into pages that
1127                         * aren't pinned.
1128                         */
1129         } else if (iov_iter_is_bvec(i)) {
1130                 const struct bio_vec *bvec = i->bvec;
1131                 while (1) {
1132                         size_t n = (--bvec)->bv_len;
1133                         i->nr_segs++;
1134                         if (unroll <= n) {
1135                                 i->bvec = bvec;
1136                                 i->iov_offset = n - unroll;
1137                                 return;
1138                         }
1139                         unroll -= n;
1140                 }
1141         } else { /* same logics for iovec and kvec */
1142                 const struct iovec *iov = i->iov;
1143                 while (1) {
1144                         size_t n = (--iov)->iov_len;
1145                         i->nr_segs++;
1146                         if (unroll <= n) {
1147                                 i->iov = iov;
1148                                 i->iov_offset = n - unroll;
1149                                 return;
1150                         }
1151                         unroll -= n;
1152                 }
1153         }
1154 }
1155 EXPORT_SYMBOL(iov_iter_revert);
1156
1157 /*
1158  * Return the count of just the current iov_iter segment.
1159  */
1160 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1161 {
1162         if (i->nr_segs > 1) {
1163                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1164                         return min(i->count, i->iov->iov_len - i->iov_offset);
1165                 if (iov_iter_is_bvec(i))
1166                         return min(i->count, i->bvec->bv_len - i->iov_offset);
1167         }
1168         return i->count;
1169 }
1170 EXPORT_SYMBOL(iov_iter_single_seg_count);
1171
1172 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1173                         const struct kvec *kvec, unsigned long nr_segs,
1174                         size_t count)
1175 {
1176         WARN_ON(direction & ~(READ | WRITE));
1177         *i = (struct iov_iter){
1178                 .iter_type = ITER_KVEC,
1179                 .data_source = direction,
1180                 .kvec = kvec,
1181                 .nr_segs = nr_segs,
1182                 .iov_offset = 0,
1183                 .count = count
1184         };
1185 }
1186 EXPORT_SYMBOL(iov_iter_kvec);
1187
1188 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1189                         const struct bio_vec *bvec, unsigned long nr_segs,
1190                         size_t count)
1191 {
1192         WARN_ON(direction & ~(READ | WRITE));
1193         *i = (struct iov_iter){
1194                 .iter_type = ITER_BVEC,
1195                 .data_source = direction,
1196                 .bvec = bvec,
1197                 .nr_segs = nr_segs,
1198                 .iov_offset = 0,
1199                 .count = count
1200         };
1201 }
1202 EXPORT_SYMBOL(iov_iter_bvec);
1203
1204 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1205                         struct pipe_inode_info *pipe,
1206                         size_t count)
1207 {
1208         BUG_ON(direction != READ);
1209         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1210         *i = (struct iov_iter){
1211                 .iter_type = ITER_PIPE,
1212                 .data_source = false,
1213                 .pipe = pipe,
1214                 .head = pipe->head,
1215                 .start_head = pipe->head,
1216                 .iov_offset = 0,
1217                 .count = count
1218         };
1219 }
1220 EXPORT_SYMBOL(iov_iter_pipe);
1221
1222 /**
1223  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1224  * @i: The iterator to initialise.
1225  * @direction: The direction of the transfer.
1226  * @xarray: The xarray to access.
1227  * @start: The start file position.
1228  * @count: The size of the I/O buffer in bytes.
1229  *
1230  * Set up an I/O iterator to either draw data out of the pages attached to an
1231  * inode or to inject data into those pages.  The pages *must* be prevented
1232  * from evaporation, either by taking a ref on them or locking them by the
1233  * caller.
1234  */
1235 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1236                      struct xarray *xarray, loff_t start, size_t count)
1237 {
1238         BUG_ON(direction & ~1);
1239         *i = (struct iov_iter) {
1240                 .iter_type = ITER_XARRAY,
1241                 .data_source = direction,
1242                 .xarray = xarray,
1243                 .xarray_start = start,
1244                 .count = count,
1245                 .iov_offset = 0
1246         };
1247 }
1248 EXPORT_SYMBOL(iov_iter_xarray);
1249
1250 /**
1251  * iov_iter_discard - Initialise an I/O iterator that discards data
1252  * @i: The iterator to initialise.
1253  * @direction: The direction of the transfer.
1254  * @count: The size of the I/O buffer in bytes.
1255  *
1256  * Set up an I/O iterator that just discards everything that's written to it.
1257  * It's only available as a READ iterator.
1258  */
1259 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1260 {
1261         BUG_ON(direction != READ);
1262         *i = (struct iov_iter){
1263                 .iter_type = ITER_DISCARD,
1264                 .data_source = false,
1265                 .count = count,
1266                 .iov_offset = 0
1267         };
1268 }
1269 EXPORT_SYMBOL(iov_iter_discard);
1270
1271 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1272 {
1273         unsigned long res = 0;
1274         size_t size = i->count;
1275         size_t skip = i->iov_offset;
1276         unsigned k;
1277
1278         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1279                 size_t len = i->iov[k].iov_len - skip;
1280                 if (len) {
1281                         res |= (unsigned long)i->iov[k].iov_base + skip;
1282                         if (len > size)
1283                                 len = size;
1284                         res |= len;
1285                         size -= len;
1286                         if (!size)
1287                                 break;
1288                 }
1289         }
1290         return res;
1291 }
1292
1293 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1294 {
1295         unsigned res = 0;
1296         size_t size = i->count;
1297         unsigned skip = i->iov_offset;
1298         unsigned k;
1299
1300         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1301                 size_t len = i->bvec[k].bv_len - skip;
1302                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1303                 if (len > size)
1304                         len = size;
1305                 res |= len;
1306                 size -= len;
1307                 if (!size)
1308                         break;
1309         }
1310         return res;
1311 }
1312
1313 unsigned long iov_iter_alignment(const struct iov_iter *i)
1314 {
1315         /* iovec and kvec have identical layouts */
1316         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1317                 return iov_iter_alignment_iovec(i);
1318
1319         if (iov_iter_is_bvec(i))
1320                 return iov_iter_alignment_bvec(i);
1321
1322         if (iov_iter_is_pipe(i)) {
1323                 unsigned int p_mask = i->pipe->ring_size - 1;
1324                 size_t size = i->count;
1325
1326                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1327                         return size | i->iov_offset;
1328                 return size;
1329         }
1330
1331         if (iov_iter_is_xarray(i))
1332                 return (i->xarray_start + i->iov_offset) | i->count;
1333
1334         return 0;
1335 }
1336 EXPORT_SYMBOL(iov_iter_alignment);
1337
1338 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1339 {
1340         unsigned long res = 0;
1341         unsigned long v = 0;
1342         size_t size = i->count;
1343         unsigned k;
1344
1345         if (WARN_ON(!iter_is_iovec(i)))
1346                 return ~0U;
1347
1348         for (k = 0; k < i->nr_segs; k++) {
1349                 if (i->iov[k].iov_len) {
1350                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1351                         if (v) // if not the first one
1352                                 res |= base | v; // this start | previous end
1353                         v = base + i->iov[k].iov_len;
1354                         if (size <= i->iov[k].iov_len)
1355                                 break;
1356                         size -= i->iov[k].iov_len;
1357                 }
1358         }
1359         return res;
1360 }
1361 EXPORT_SYMBOL(iov_iter_gap_alignment);
1362
1363 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1364                                 size_t maxsize,
1365                                 struct page **pages,
1366                                 int iter_head,
1367                                 size_t *start)
1368 {
1369         struct pipe_inode_info *pipe = i->pipe;
1370         unsigned int p_mask = pipe->ring_size - 1;
1371         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1372         if (!n)
1373                 return -EFAULT;
1374
1375         maxsize = n;
1376         n += *start;
1377         while (n > 0) {
1378                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1379                 iter_head++;
1380                 n -= PAGE_SIZE;
1381         }
1382
1383         return maxsize;
1384 }
1385
1386 static ssize_t pipe_get_pages(struct iov_iter *i,
1387                    struct page **pages, size_t maxsize, unsigned maxpages,
1388                    size_t *start)
1389 {
1390         unsigned int iter_head, npages;
1391         size_t capacity;
1392
1393         if (!sanity(i))
1394                 return -EFAULT;
1395
1396         data_start(i, &iter_head, start);
1397         /* Amount of free space: some of this one + all after this one */
1398         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1399         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1400
1401         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1402 }
1403
1404 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1405                                           pgoff_t index, unsigned int nr_pages)
1406 {
1407         XA_STATE(xas, xa, index);
1408         struct page *page;
1409         unsigned int ret = 0;
1410
1411         rcu_read_lock();
1412         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1413                 if (xas_retry(&xas, page))
1414                         continue;
1415
1416                 /* Has the page moved or been split? */
1417                 if (unlikely(page != xas_reload(&xas))) {
1418                         xas_reset(&xas);
1419                         continue;
1420                 }
1421
1422                 pages[ret] = find_subpage(page, xas.xa_index);
1423                 get_page(pages[ret]);
1424                 if (++ret == nr_pages)
1425                         break;
1426         }
1427         rcu_read_unlock();
1428         return ret;
1429 }
1430
1431 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1432                                      struct page **pages, size_t maxsize,
1433                                      unsigned maxpages, size_t *_start_offset)
1434 {
1435         unsigned nr, offset;
1436         pgoff_t index, count;
1437         size_t size = maxsize;
1438         loff_t pos;
1439
1440         if (!size || !maxpages)
1441                 return 0;
1442
1443         pos = i->xarray_start + i->iov_offset;
1444         index = pos >> PAGE_SHIFT;
1445         offset = pos & ~PAGE_MASK;
1446         *_start_offset = offset;
1447
1448         count = 1;
1449         if (size > PAGE_SIZE - offset) {
1450                 size -= PAGE_SIZE - offset;
1451                 count += size >> PAGE_SHIFT;
1452                 size &= ~PAGE_MASK;
1453                 if (size)
1454                         count++;
1455         }
1456
1457         if (count > maxpages)
1458                 count = maxpages;
1459
1460         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1461         if (nr == 0)
1462                 return 0;
1463
1464         return min(nr * PAGE_SIZE - offset, maxsize);
1465 }
1466
1467 /* must be done on non-empty ITER_IOVEC one */
1468 static unsigned long first_iovec_segment(const struct iov_iter *i,
1469                                          size_t *size, size_t *start,
1470                                          size_t maxsize, unsigned maxpages)
1471 {
1472         size_t skip;
1473         long k;
1474
1475         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1476                 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1477                 size_t len = i->iov[k].iov_len - skip;
1478
1479                 if (unlikely(!len))
1480                         continue;
1481                 if (len > maxsize)
1482                         len = maxsize;
1483                 len += (*start = addr % PAGE_SIZE);
1484                 if (len > maxpages * PAGE_SIZE)
1485                         len = maxpages * PAGE_SIZE;
1486                 *size = len;
1487                 return addr & PAGE_MASK;
1488         }
1489         BUG(); // if it had been empty, we wouldn't get called
1490 }
1491
1492 /* must be done on non-empty ITER_BVEC one */
1493 static struct page *first_bvec_segment(const struct iov_iter *i,
1494                                        size_t *size, size_t *start,
1495                                        size_t maxsize, unsigned maxpages)
1496 {
1497         struct page *page;
1498         size_t skip = i->iov_offset, len;
1499
1500         len = i->bvec->bv_len - skip;
1501         if (len > maxsize)
1502                 len = maxsize;
1503         skip += i->bvec->bv_offset;
1504         page = i->bvec->bv_page + skip / PAGE_SIZE;
1505         len += (*start = skip % PAGE_SIZE);
1506         if (len > maxpages * PAGE_SIZE)
1507                 len = maxpages * PAGE_SIZE;
1508         *size = len;
1509         return page;
1510 }
1511
1512 ssize_t iov_iter_get_pages(struct iov_iter *i,
1513                    struct page **pages, size_t maxsize, unsigned maxpages,
1514                    size_t *start)
1515 {
1516         size_t len;
1517         int n, res;
1518
1519         if (maxsize > i->count)
1520                 maxsize = i->count;
1521         if (!maxsize)
1522                 return 0;
1523
1524         if (likely(iter_is_iovec(i))) {
1525                 unsigned int gup_flags = 0;
1526                 unsigned long addr;
1527
1528                 if (iov_iter_rw(i) != WRITE)
1529                         gup_flags |= FOLL_WRITE;
1530                 if (i->nofault)
1531                         gup_flags |= FOLL_NOFAULT;
1532
1533                 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1534                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1535                 res = get_user_pages_fast(addr, n, gup_flags, pages);
1536                 if (unlikely(res <= 0))
1537                         return res;
1538                 return (res == n ? len : res * PAGE_SIZE) - *start;
1539         }
1540         if (iov_iter_is_bvec(i)) {
1541                 struct page *page;
1542
1543                 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1544                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1545                 while (n--)
1546                         get_page(*pages++ = page++);
1547                 return len - *start;
1548         }
1549         if (iov_iter_is_pipe(i))
1550                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1551         if (iov_iter_is_xarray(i))
1552                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1553         return -EFAULT;
1554 }
1555 EXPORT_SYMBOL(iov_iter_get_pages);
1556
1557 static struct page **get_pages_array(size_t n)
1558 {
1559         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1560 }
1561
1562 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1563                    struct page ***pages, size_t maxsize,
1564                    size_t *start)
1565 {
1566         struct page **p;
1567         unsigned int iter_head, npages;
1568         ssize_t n;
1569
1570         if (!sanity(i))
1571                 return -EFAULT;
1572
1573         data_start(i, &iter_head, start);
1574         /* Amount of free space: some of this one + all after this one */
1575         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1576         n = npages * PAGE_SIZE - *start;
1577         if (maxsize > n)
1578                 maxsize = n;
1579         else
1580                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1581         p = get_pages_array(npages);
1582         if (!p)
1583                 return -ENOMEM;
1584         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1585         if (n > 0)
1586                 *pages = p;
1587         else
1588                 kvfree(p);
1589         return n;
1590 }
1591
1592 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1593                                            struct page ***pages, size_t maxsize,
1594                                            size_t *_start_offset)
1595 {
1596         struct page **p;
1597         unsigned nr, offset;
1598         pgoff_t index, count;
1599         size_t size = maxsize;
1600         loff_t pos;
1601
1602         if (!size)
1603                 return 0;
1604
1605         pos = i->xarray_start + i->iov_offset;
1606         index = pos >> PAGE_SHIFT;
1607         offset = pos & ~PAGE_MASK;
1608         *_start_offset = offset;
1609
1610         count = 1;
1611         if (size > PAGE_SIZE - offset) {
1612                 size -= PAGE_SIZE - offset;
1613                 count += size >> PAGE_SHIFT;
1614                 size &= ~PAGE_MASK;
1615                 if (size)
1616                         count++;
1617         }
1618
1619         p = get_pages_array(count);
1620         if (!p)
1621                 return -ENOMEM;
1622         *pages = p;
1623
1624         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1625         if (nr == 0)
1626                 return 0;
1627
1628         return min(nr * PAGE_SIZE - offset, maxsize);
1629 }
1630
1631 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1632                    struct page ***pages, size_t maxsize,
1633                    size_t *start)
1634 {
1635         struct page **p;
1636         size_t len;
1637         int n, res;
1638
1639         if (maxsize > i->count)
1640                 maxsize = i->count;
1641         if (!maxsize)
1642                 return 0;
1643
1644         if (likely(iter_is_iovec(i))) {
1645                 unsigned int gup_flags = 0;
1646                 unsigned long addr;
1647
1648                 if (iov_iter_rw(i) != WRITE)
1649                         gup_flags |= FOLL_WRITE;
1650                 if (i->nofault)
1651                         gup_flags |= FOLL_NOFAULT;
1652
1653                 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1654                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1655                 p = get_pages_array(n);
1656                 if (!p)
1657                         return -ENOMEM;
1658                 res = get_user_pages_fast(addr, n, gup_flags, p);
1659                 if (unlikely(res <= 0)) {
1660                         kvfree(p);
1661                         *pages = NULL;
1662                         return res;
1663                 }
1664                 *pages = p;
1665                 return (res == n ? len : res * PAGE_SIZE) - *start;
1666         }
1667         if (iov_iter_is_bvec(i)) {
1668                 struct page *page;
1669
1670                 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1671                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1672                 *pages = p = get_pages_array(n);
1673                 if (!p)
1674                         return -ENOMEM;
1675                 while (n--)
1676                         get_page(*p++ = page++);
1677                 return len - *start;
1678         }
1679         if (iov_iter_is_pipe(i))
1680                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1681         if (iov_iter_is_xarray(i))
1682                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1683         return -EFAULT;
1684 }
1685 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1686
1687 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1688                                struct iov_iter *i)
1689 {
1690         __wsum sum, next;
1691         sum = *csum;
1692         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1693                 WARN_ON(1);
1694                 return 0;
1695         }
1696         iterate_and_advance(i, bytes, base, len, off, ({
1697                 next = csum_and_copy_from_user(base, addr + off, len);
1698                 sum = csum_block_add(sum, next, off);
1699                 next ? 0 : len;
1700         }), ({
1701                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1702         })
1703         )
1704         *csum = sum;
1705         return bytes;
1706 }
1707 EXPORT_SYMBOL(csum_and_copy_from_iter);
1708
1709 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1710                              struct iov_iter *i)
1711 {
1712         struct csum_state *csstate = _csstate;
1713         __wsum sum, next;
1714
1715         if (unlikely(iov_iter_is_discard(i))) {
1716                 WARN_ON(1);     /* for now */
1717                 return 0;
1718         }
1719
1720         sum = csum_shift(csstate->csum, csstate->off);
1721         if (unlikely(iov_iter_is_pipe(i)))
1722                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1723         else iterate_and_advance(i, bytes, base, len, off, ({
1724                 next = csum_and_copy_to_user(addr + off, base, len);
1725                 sum = csum_block_add(sum, next, off);
1726                 next ? 0 : len;
1727         }), ({
1728                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1729         })
1730         )
1731         csstate->csum = csum_shift(sum, csstate->off);
1732         csstate->off += bytes;
1733         return bytes;
1734 }
1735 EXPORT_SYMBOL(csum_and_copy_to_iter);
1736
1737 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1738                 struct iov_iter *i)
1739 {
1740 #ifdef CONFIG_CRYPTO_HASH
1741         struct ahash_request *hash = hashp;
1742         struct scatterlist sg;
1743         size_t copied;
1744
1745         copied = copy_to_iter(addr, bytes, i);
1746         sg_init_one(&sg, addr, copied);
1747         ahash_request_set_crypt(hash, &sg, NULL, copied);
1748         crypto_ahash_update(hash);
1749         return copied;
1750 #else
1751         return 0;
1752 #endif
1753 }
1754 EXPORT_SYMBOL(hash_and_copy_to_iter);
1755
1756 static int iov_npages(const struct iov_iter *i, int maxpages)
1757 {
1758         size_t skip = i->iov_offset, size = i->count;
1759         const struct iovec *p;
1760         int npages = 0;
1761
1762         for (p = i->iov; size; skip = 0, p++) {
1763                 unsigned offs = offset_in_page(p->iov_base + skip);
1764                 size_t len = min(p->iov_len - skip, size);
1765
1766                 if (len) {
1767                         size -= len;
1768                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1769                         if (unlikely(npages > maxpages))
1770                                 return maxpages;
1771                 }
1772         }
1773         return npages;
1774 }
1775
1776 static int bvec_npages(const struct iov_iter *i, int maxpages)
1777 {
1778         size_t skip = i->iov_offset, size = i->count;
1779         const struct bio_vec *p;
1780         int npages = 0;
1781
1782         for (p = i->bvec; size; skip = 0, p++) {
1783                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1784                 size_t len = min(p->bv_len - skip, size);
1785
1786                 size -= len;
1787                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1788                 if (unlikely(npages > maxpages))
1789                         return maxpages;
1790         }
1791         return npages;
1792 }
1793
1794 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1795 {
1796         if (unlikely(!i->count))
1797                 return 0;
1798         /* iovec and kvec have identical layouts */
1799         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1800                 return iov_npages(i, maxpages);
1801         if (iov_iter_is_bvec(i))
1802                 return bvec_npages(i, maxpages);
1803         if (iov_iter_is_pipe(i)) {
1804                 unsigned int iter_head;
1805                 int npages;
1806                 size_t off;
1807
1808                 if (!sanity(i))
1809                         return 0;
1810
1811                 data_start(i, &iter_head, &off);
1812                 /* some of this one + all after this one */
1813                 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1814                 return min(npages, maxpages);
1815         }
1816         if (iov_iter_is_xarray(i)) {
1817                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1818                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1819                 return min(npages, maxpages);
1820         }
1821         return 0;
1822 }
1823 EXPORT_SYMBOL(iov_iter_npages);
1824
1825 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1826 {
1827         *new = *old;
1828         if (unlikely(iov_iter_is_pipe(new))) {
1829                 WARN_ON(1);
1830                 return NULL;
1831         }
1832         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1833                 return NULL;
1834         if (iov_iter_is_bvec(new))
1835                 return new->bvec = kmemdup(new->bvec,
1836                                     new->nr_segs * sizeof(struct bio_vec),
1837                                     flags);
1838         else
1839                 /* iovec and kvec have identical layout */
1840                 return new->iov = kmemdup(new->iov,
1841                                    new->nr_segs * sizeof(struct iovec),
1842                                    flags);
1843 }
1844 EXPORT_SYMBOL(dup_iter);
1845
1846 static int copy_compat_iovec_from_user(struct iovec *iov,
1847                 const struct iovec __user *uvec, unsigned long nr_segs)
1848 {
1849         const struct compat_iovec __user *uiov =
1850                 (const struct compat_iovec __user *)uvec;
1851         int ret = -EFAULT, i;
1852
1853         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1854                 return -EFAULT;
1855
1856         for (i = 0; i < nr_segs; i++) {
1857                 compat_uptr_t buf;
1858                 compat_ssize_t len;
1859
1860                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1861                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1862
1863                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1864                 if (len < 0) {
1865                         ret = -EINVAL;
1866                         goto uaccess_end;
1867                 }
1868                 iov[i].iov_base = compat_ptr(buf);
1869                 iov[i].iov_len = len;
1870         }
1871
1872         ret = 0;
1873 uaccess_end:
1874         user_access_end();
1875         return ret;
1876 }
1877
1878 static int copy_iovec_from_user(struct iovec *iov,
1879                 const struct iovec __user *uvec, unsigned long nr_segs)
1880 {
1881         unsigned long seg;
1882
1883         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1884                 return -EFAULT;
1885         for (seg = 0; seg < nr_segs; seg++) {
1886                 if ((ssize_t)iov[seg].iov_len < 0)
1887                         return -EINVAL;
1888         }
1889
1890         return 0;
1891 }
1892
1893 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1894                 unsigned long nr_segs, unsigned long fast_segs,
1895                 struct iovec *fast_iov, bool compat)
1896 {
1897         struct iovec *iov = fast_iov;
1898         int ret;
1899
1900         /*
1901          * SuS says "The readv() function *may* fail if the iovcnt argument was
1902          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1903          * traditionally returned zero for zero segments, so...
1904          */
1905         if (nr_segs == 0)
1906                 return iov;
1907         if (nr_segs > UIO_MAXIOV)
1908                 return ERR_PTR(-EINVAL);
1909         if (nr_segs > fast_segs) {
1910                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1911                 if (!iov)
1912                         return ERR_PTR(-ENOMEM);
1913         }
1914
1915         if (compat)
1916                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1917         else
1918                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1919         if (ret) {
1920                 if (iov != fast_iov)
1921                         kfree(iov);
1922                 return ERR_PTR(ret);
1923         }
1924
1925         return iov;
1926 }
1927
1928 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1929                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1930                  struct iov_iter *i, bool compat)
1931 {
1932         ssize_t total_len = 0;
1933         unsigned long seg;
1934         struct iovec *iov;
1935
1936         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1937         if (IS_ERR(iov)) {
1938                 *iovp = NULL;
1939                 return PTR_ERR(iov);
1940         }
1941
1942         /*
1943          * According to the Single Unix Specification we should return EINVAL if
1944          * an element length is < 0 when cast to ssize_t or if the total length
1945          * would overflow the ssize_t return value of the system call.
1946          *
1947          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1948          * overflow case.
1949          */
1950         for (seg = 0; seg < nr_segs; seg++) {
1951                 ssize_t len = (ssize_t)iov[seg].iov_len;
1952
1953                 if (!access_ok(iov[seg].iov_base, len)) {
1954                         if (iov != *iovp)
1955                                 kfree(iov);
1956                         *iovp = NULL;
1957                         return -EFAULT;
1958                 }
1959
1960                 if (len > MAX_RW_COUNT - total_len) {
1961                         len = MAX_RW_COUNT - total_len;
1962                         iov[seg].iov_len = len;
1963                 }
1964                 total_len += len;
1965         }
1966
1967         iov_iter_init(i, type, iov, nr_segs, total_len);
1968         if (iov == *iovp)
1969                 *iovp = NULL;
1970         else
1971                 *iovp = iov;
1972         return total_len;
1973 }
1974
1975 /**
1976  * import_iovec() - Copy an array of &struct iovec from userspace
1977  *     into the kernel, check that it is valid, and initialize a new
1978  *     &struct iov_iter iterator to access it.
1979  *
1980  * @type: One of %READ or %WRITE.
1981  * @uvec: Pointer to the userspace array.
1982  * @nr_segs: Number of elements in userspace array.
1983  * @fast_segs: Number of elements in @iov.
1984  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1985  *     on-stack) kernel array.
1986  * @i: Pointer to iterator that will be initialized on success.
1987  *
1988  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1989  * then this function places %NULL in *@iov on return. Otherwise, a new
1990  * array will be allocated and the result placed in *@iov. This means that
1991  * the caller may call kfree() on *@iov regardless of whether the small
1992  * on-stack array was used or not (and regardless of whether this function
1993  * returns an error or not).
1994  *
1995  * Return: Negative error code on error, bytes imported on success
1996  */
1997 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1998                  unsigned nr_segs, unsigned fast_segs,
1999                  struct iovec **iovp, struct iov_iter *i)
2000 {
2001         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
2002                               in_compat_syscall());
2003 }
2004 EXPORT_SYMBOL(import_iovec);
2005
2006 int import_single_range(int rw, void __user *buf, size_t len,
2007                  struct iovec *iov, struct iov_iter *i)
2008 {
2009         if (len > MAX_RW_COUNT)
2010                 len = MAX_RW_COUNT;
2011         if (unlikely(!access_ok(buf, len)))
2012                 return -EFAULT;
2013
2014         iov->iov_base = buf;
2015         iov->iov_len = len;
2016         iov_iter_init(i, rw, iov, 1, len);
2017         return 0;
2018 }
2019 EXPORT_SYMBOL(import_single_range);
2020
2021 /**
2022  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
2023  *     iov_iter_save_state() was called.
2024  *
2025  * @i: &struct iov_iter to restore
2026  * @state: state to restore from
2027  *
2028  * Used after iov_iter_save_state() to bring restore @i, if operations may
2029  * have advanced it.
2030  *
2031  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
2032  */
2033 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
2034 {
2035         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
2036                          !iov_iter_is_kvec(i))
2037                 return;
2038         i->iov_offset = state->iov_offset;
2039         i->count = state->count;
2040         /*
2041          * For the *vec iters, nr_segs + iov is constant - if we increment
2042          * the vec, then we also decrement the nr_segs count. Hence we don't
2043          * need to track both of these, just one is enough and we can deduct
2044          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
2045          * size, so we can just increment the iov pointer as they are unionzed.
2046          * ITER_BVEC _may_ be the same size on some archs, but on others it is
2047          * not. Be safe and handle it separately.
2048          */
2049         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
2050         if (iov_iter_is_bvec(i))
2051                 i->bvec -= state->nr_segs - i->nr_segs;
2052         else
2053                 i->iov -= state->nr_segs - i->nr_segs;
2054         i->nr_segs = state->nr_segs;
2055 }