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 /* covers ubuf and kbuf alike */
  18 #define iterate_buf(i, n, base, len, off, __p, STEP) {          \
  19         size_t __maybe_unused off = 0;                          \
  20         len = n;                                                \
  21         base = __p + i->iov_offset;                             \
  22         len -= (STEP);                                          \
  23         i->iov_offset += len;                                   \
  24         n = len;                                                \
  25 }
  26
  27 /* covers iovec and kvec alike */
  28 #define iterate_iovec(i, n, base, len, off, __p, STEP) {        \
  29         size_t off = 0;                                         \
  30         size_t skip = i->iov_offset;                            \
  31         do {                                                    \
  32                 len = min(n, __p->iov_len - skip);              \
  33                 if (likely(len)) {                              \
  34                         base = __p->iov_base + skip;            \
  35                         len -= (STEP);                          \
  36                         off += len;                             \
  37                         skip += len;                            \
  38                         n -= len;                               \
  39                         if (skip < __p->iov_len)                \
  40                                 break;                          \
  41                 }                                               \
  42                 __p++;                                          \
  43                 skip = 0;                                       \
  44         } while (n);                                            \
  45         i->iov_offset = skip;                                   \
  46         n = off;                                                \
  47 }
  48
  49 #define iterate_bvec(i, n, base, len, off, p, STEP) {           \
  50         size_t off = 0;                                         \
  51         unsigned skip = i->iov_offset;                          \
  52         while (n) {                                             \
  53                 unsigned offset = p->bv_offset + skip;          \
  54                 unsigned left;                                  \
  55                 void *kaddr = kmap_local_page(p->bv_page +      \
  56                                         offset / PAGE_SIZE);    \
  57                 base = kaddr + offset % PAGE_SIZE;              \
  58                 len = min(min(n, (size_t)(p->bv_len - skip)),   \
  59                      (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
  60                 left = (STEP);                                  \
  61                 kunmap_local(kaddr);                            \
  62                 len -= left;                                    \
  63                 off += len;                                     \
  64                 skip += len;                                    \
  65                 if (skip == p->bv_len) {                        \
  66                         skip = 0;                               \
  67                         p++;                                    \
  68                 }                                               \
  69                 n -= len;                                       \
  70                 if (left)                                       \
  71                         break;                                  \
  72         }                                                       \
  73         i->iov_offset = skip;                                   \
  74         n = off;                                                \
  75 }
  76
  77 #define iterate_xarray(i, n, base, len, __off, STEP) {          \
  78         __label__ __out;                                        \
  79         size_t __off = 0;                                       \
  80         struct folio *folio;                                    \
  81         loff_t start = i->xarray_start + i->iov_offset;         \
  82         pgoff_t index = start / PAGE_SIZE;                      \
  83         XA_STATE(xas, i->xarray, index);                        \
  84                                                                 \
  85         len = PAGE_SIZE - offset_in_page(start);                \
  86         rcu_read_lock();                                        \
  87         xas_for_each(&xas, folio, ULONG_MAX) {                  \
  88                 unsigned left;                                  \
  89                 size_t offset;                                  \
  90                 if (xas_retry(&xas, folio))                     \
  91                         continue;                               \
  92                 if (WARN_ON(xa_is_value(folio)))                \
  93                         break;                                  \
  94                 if (WARN_ON(folio_test_hugetlb(folio)))         \
  95                         break;                                  \
  96                 offset = offset_in_folio(folio, start + __off); \
  97                 while (offset < folio_size(folio)) {            \
  98                         base = kmap_local_folio(folio, offset); \
  99                         len = min(n, len);                      \
 100                         left = (STEP);                          \
 101                         kunmap_local(base);                     \
 102                         len -= left;                            \
 103                         __off += len;                           \
 104                         n -= len;                               \
 105                         if (left || n == 0)                     \
 106                                 goto __out;                     \
 107                         offset += len;                          \
 108                         len = PAGE_SIZE;                        \
 109                 }                                               \
 110         }                                                       \
 111 __out:                                                          \
 112         rcu_read_unlock();                                      \
 113         i->iov_offset += __off;                                 \
 114         n = __off;                                              \
 115 }
 116
 117 #define __iterate_and_advance(i, n, base, len, off, I, K) {     \
 118         if (unlikely(i->count < n))                             \
 119                 n = i->count;                                   \
 120         if (likely(n)) {                                        \
 121                 if (likely(iter_is_ubuf(i))) {                  \
 122                         void __user *base;                      \
 123                         size_t len;                             \
 124                         iterate_buf(i, n, base, len, off,       \
 125                                                 i->ubuf, (I))   \
 126                 } else if (likely(iter_is_iovec(i))) {          \
 127                         const struct iovec *iov = iter_iov(i);  \
 128                         void __user *base;                      \
 129                         size_t len;                             \
 130                         iterate_iovec(i, n, base, len, off,     \
 131                                                 iov, (I))       \
 132                         i->nr_segs -= iov - iter_iov(i);        \
 133                         i->__iov = iov;                         \
 134                 } else if (iov_iter_is_bvec(i)) {               \
 135                         const struct bio_vec *bvec = i->bvec;   \
 136                         void *base;                             \
 137                         size_t len;                             \
 138                         iterate_bvec(i, n, base, len, off,      \
 139                                                 bvec, (K))      \
 140                         i->nr_segs -= bvec - i->bvec;           \
 141                         i->bvec = bvec;                         \
 142                 } else if (iov_iter_is_kvec(i)) {               \
 143                         const struct kvec *kvec = i->kvec;      \
 144                         void *base;                             \
 145                         size_t len;                             \
 146                         iterate_iovec(i, n, base, len, off,     \
 147                                                 kvec, (K))      \
 148                         i->nr_segs -= kvec - i->kvec;           \
 149                         i->kvec = kvec;                         \
 150                 } else if (iov_iter_is_xarray(i)) {             \
 151                         void *base;                             \
 152                         size_t len;                             \
 153                         iterate_xarray(i, n, base, len, off,    \
 154                                                         (K))    \
 155                 }                                               \
 156                 i->count -= n;                                  \
 157         }                                                       \
 158 }
 159 #define iterate_and_advance(i, n, base, len, off, I, K) \
 160         __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
 161
 162 static int copyout(void __user *to, const void *from, size_t n)
 163 {
 164         if (should_fail_usercopy())
 165                 return n;
 166         if (access_ok(to, n)) {
 167                 instrument_copy_to_user(to, from, n);
 168                 n = raw_copy_to_user(to, from, n);
 169         }
 170         return n;
 171 }
 172
 173 static int copyout_nofault(void __user *to, const void *from, size_t n)
 174 {
 175         long res;
 176
 177         if (should_fail_usercopy())
 178                 return n;
 179
 180         res = copy_to_user_nofault(to, from, n);
 181
 182         return res < 0 ? n : res;
 183 }
 184
 185 static int copyin(void *to, const void __user *from, size_t n)
 186 {
 187         size_t res = n;
 188
 189         if (should_fail_usercopy())
 190                 return n;
 191         if (access_ok(from, n)) {
 192                 instrument_copy_from_user_before(to, from, n);
 193                 res = raw_copy_from_user(to, from, n);
 194                 instrument_copy_from_user_after(to, from, n, res);
 195         }
 196         return res;
 197 }
 198
 199 /*
 200  * fault_in_iov_iter_readable - fault in iov iterator for reading
 201  * @i: iterator
 202  * @size: maximum length
 203  *
 204  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 205  * @size.  For each iovec, fault in each page that constitutes the iovec.
 206  *
 207  * Returns the number of bytes not faulted in (like copy_to_user() and
 208  * copy_from_user()).
 209  *
 210  * Always returns 0 for non-userspace iterators.
 211  */
 212 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
 213 {
 214         if (iter_is_ubuf(i)) {
 215                 size_t n = min(size, iov_iter_count(i));
 216                 n -= fault_in_readable(i->ubuf + i->iov_offset, n);
 217                 return size - n;
 218         } else if (iter_is_iovec(i)) {
 219                 size_t count = min(size, iov_iter_count(i));
 220                 const struct iovec *p;
 221                 size_t skip;
 222
 223                 size -= count;
 224                 for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) {
 225                         size_t len = min(count, p->iov_len - skip);
 226                         size_t ret;
 227
 228                         if (unlikely(!len))
 229                                 continue;
 230                         ret = fault_in_readable(p->iov_base + skip, len);
 231                         count -= len - ret;
 232                         if (ret)
 233                                 break;
 234                 }
 235                 return count + size;
 236         }
 237         return 0;
 238 }
 239 EXPORT_SYMBOL(fault_in_iov_iter_readable);
 240
 241 /*
 242  * fault_in_iov_iter_writeable - fault in iov iterator for writing
 243  * @i: iterator
 244  * @size: maximum length
 245  *
 246  * Faults in the iterator using get_user_pages(), i.e., without triggering
 247  * hardware page faults.  This is primarily useful when we already know that
 248  * some or all of the pages in @i aren't in memory.
 249  *
 250  * Returns the number of bytes not faulted in, like copy_to_user() and
 251  * copy_from_user().
 252  *
 253  * Always returns 0 for non-user-space iterators.
 254  */
 255 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
 256 {
 257         if (iter_is_ubuf(i)) {
 258                 size_t n = min(size, iov_iter_count(i));
 259                 n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n);
 260                 return size - n;
 261         } else if (iter_is_iovec(i)) {
 262                 size_t count = min(size, iov_iter_count(i));
 263                 const struct iovec *p;
 264                 size_t skip;
 265
 266                 size -= count;
 267                 for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) {
 268                         size_t len = min(count, p->iov_len - skip);
 269                         size_t ret;
 270
 271                         if (unlikely(!len))
 272                                 continue;
 273                         ret = fault_in_safe_writeable(p->iov_base + skip, len);
 274                         count -= len - ret;
 275                         if (ret)
 276                                 break;
 277                 }
 278                 return count + size;
 279         }
 280         return 0;
 281 }
 282 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
 283
 284 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 285                         const struct iovec *iov, unsigned long nr_segs,
 286                         size_t count)
 287 {
 288         WARN_ON(direction & ~(READ | WRITE));
 289         *i = (struct iov_iter) {
 290                 .iter_type = ITER_IOVEC,
 291                 .copy_mc = false,
 292                 .nofault = false,
 293                 .user_backed = true,
 294                 .data_source = direction,
 295                 .__iov = iov,
 296                 .nr_segs = nr_segs,
 297                 .iov_offset = 0,
 298                 .count = count
 299         };
 300 }
 301 EXPORT_SYMBOL(iov_iter_init);
 302
 303 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 304                               __wsum sum, size_t off)
 305 {
 306         __wsum next = csum_partial_copy_nocheck(from, to, len);
 307         return csum_block_add(sum, next, off);
 308 }
 309
 310 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 311 {
 312         if (WARN_ON_ONCE(i->data_source))
 313                 return 0;
 314         if (user_backed_iter(i))
 315                 might_fault();
 316         iterate_and_advance(i, bytes, base, len, off,
 317                 copyout(base, addr + off, len),
 318                 memcpy(base, addr + off, len)
 319         )
 320
 321         return bytes;
 322 }
 323 EXPORT_SYMBOL(_copy_to_iter);
 324
 325 #ifdef CONFIG_ARCH_HAS_COPY_MC
 326 static int copyout_mc(void __user *to, const void *from, size_t n)
 327 {
 328         if (access_ok(to, n)) {
 329                 instrument_copy_to_user(to, from, n);
 330                 n = copy_mc_to_user((__force void *) to, from, n);
 331         }
 332         return n;
 333 }
 334
 335 /**
 336  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 337  * @addr: source kernel address
 338  * @bytes: total transfer length
 339  * @i: destination iterator
 340  *
 341  * The pmem driver deploys this for the dax operation
 342  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 343  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 344  * successfully copied.
 345  *
 346  * The main differences between this and typical _copy_to_iter().
 347  *
 348  * * Typical tail/residue handling after a fault retries the copy
 349  *   byte-by-byte until the fault happens again. Re-triggering machine
 350  *   checks is potentially fatal so the implementation uses source
 351  *   alignment and poison alignment assumptions to avoid re-triggering
 352  *   hardware exceptions.
 353  *
 354  * * ITER_KVEC and ITER_BVEC can return short copies.  Compare to
 355  *   copy_to_iter() where only ITER_IOVEC attempts might return a short copy.
 356  *
 357  * Return: number of bytes copied (may be %0)
 358  */
 359 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 360 {
 361         if (WARN_ON_ONCE(i->data_source))
 362                 return 0;
 363         if (user_backed_iter(i))
 364                 might_fault();
 365         __iterate_and_advance(i, bytes, base, len, off,
 366                 copyout_mc(base, addr + off, len),
 367                 copy_mc_to_kernel(base, addr + off, len)
 368         )
 369
 370         return bytes;
 371 }
 372 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 373 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 374
 375 static void *memcpy_from_iter(struct iov_iter *i, void *to, const void *from,
 376                                  size_t size)
 377 {
 378         if (iov_iter_is_copy_mc(i))
 379                 return (void *)copy_mc_to_kernel(to, from, size);
 380         return memcpy(to, from, size);
 381 }
 382
 383 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 384 {
 385         if (WARN_ON_ONCE(!i->data_source))
 386                 return 0;
 387
 388         if (user_backed_iter(i))
 389                 might_fault();
 390         iterate_and_advance(i, bytes, base, len, off,
 391                 copyin(addr + off, base, len),
 392                 memcpy_from_iter(i, addr + off, base, len)
 393         )
 394
 395         return bytes;
 396 }
 397 EXPORT_SYMBOL(_copy_from_iter);
 398
 399 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 400 {
 401         if (WARN_ON_ONCE(!i->data_source))
 402                 return 0;
 403
 404         iterate_and_advance(i, bytes, base, len, off,
 405                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 406                 memcpy(addr + off, base, len)
 407         )
 408
 409         return bytes;
 410 }
 411 EXPORT_SYMBOL(_copy_from_iter_nocache);
 412
 413 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 414 /**
 415  * _copy_from_iter_flushcache - write destination through cpu cache
 416  * @addr: destination kernel address
 417  * @bytes: total transfer length
 418  * @i: source iterator
 419  *
 420  * The pmem driver arranges for filesystem-dax to use this facility via
 421  * dax_copy_from_iter() for ensuring that writes to persistent memory
 422  * are flushed through the CPU cache. It is differentiated from
 423  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 424  * all iterator types. The _copy_from_iter_nocache() only attempts to
 425  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 426  * instructions that strand dirty-data in the cache.
 427  *
 428  * Return: number of bytes copied (may be %0)
 429  */
 430 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 431 {
 432         if (WARN_ON_ONCE(!i->data_source))
 433                 return 0;
 434
 435         iterate_and_advance(i, bytes, base, len, off,
 436                 __copy_from_user_flushcache(addr + off, base, len),
 437                 memcpy_flushcache(addr + off, base, len)
 438         )
 439
 440         return bytes;
 441 }
 442 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 443 #endif
 444
 445 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 446 {
 447         struct page *head;
 448         size_t v = n + offset;
 449
 450         /*
 451          * The general case needs to access the page order in order
 452          * to compute the page size.
 453          * However, we mostly deal with order-0 pages and thus can
 454          * avoid a possible cache line miss for requests that fit all
 455          * page orders.
 456          */
 457         if (n <= v && v <= PAGE_SIZE)
 458                 return true;
 459
 460         head = compound_head(page);
 461         v += (page - head) << PAGE_SHIFT;
 462
 463         if (WARN_ON(n > v || v > page_size(head)))
 464                 return false;
 465         return true;
 466 }
 467
 468 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 469                          struct iov_iter *i)
 470 {
 471         size_t res = 0;
 472         if (!page_copy_sane(page, offset, bytes))
 473                 return 0;
 474         if (WARN_ON_ONCE(i->data_source))
 475                 return 0;
 476         page += offset / PAGE_SIZE; // first subpage
 477         offset %= PAGE_SIZE;
 478         while (1) {
 479                 void *kaddr = kmap_local_page(page);
 480                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 481                 n = _copy_to_iter(kaddr + offset, n, i);
 482                 kunmap_local(kaddr);
 483                 res += n;
 484                 bytes -= n;
 485                 if (!bytes || !n)
 486                         break;
 487                 offset += n;
 488                 if (offset == PAGE_SIZE) {
 489                         page++;
 490                         offset = 0;
 491                 }
 492         }
 493         return res;
 494 }
 495 EXPORT_SYMBOL(copy_page_to_iter);
 496
 497 size_t copy_page_to_iter_nofault(struct page *page, unsigned offset, size_t bytes,
 498                                  struct iov_iter *i)
 499 {
 500         size_t res = 0;
 501
 502         if (!page_copy_sane(page, offset, bytes))
 503                 return 0;
 504         if (WARN_ON_ONCE(i->data_source))
 505                 return 0;
 506         page += offset / PAGE_SIZE; // first subpage
 507         offset %= PAGE_SIZE;
 508         while (1) {
 509                 void *kaddr = kmap_local_page(page);
 510                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 511
 512                 iterate_and_advance(i, n, base, len, off,
 513                         copyout_nofault(base, kaddr + offset + off, len),
 514                         memcpy(base, kaddr + offset + off, len)
 515                 )
 516                 kunmap_local(kaddr);
 517                 res += n;
 518                 bytes -= n;
 519                 if (!bytes || !n)
 520                         break;
 521                 offset += n;
 522                 if (offset == PAGE_SIZE) {
 523                         page++;
 524                         offset = 0;
 525                 }
 526         }
 527         return res;
 528 }
 529 EXPORT_SYMBOL(copy_page_to_iter_nofault);
 530
 531 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 532                          struct iov_iter *i)
 533 {
 534         size_t res = 0;
 535         if (!page_copy_sane(page, offset, bytes))
 536                 return 0;
 537         page += offset / PAGE_SIZE; // first subpage
 538         offset %= PAGE_SIZE;
 539         while (1) {
 540                 void *kaddr = kmap_local_page(page);
 541                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 542                 n = _copy_from_iter(kaddr + offset, n, i);
 543                 kunmap_local(kaddr);
 544                 res += n;
 545                 bytes -= n;
 546                 if (!bytes || !n)
 547                         break;
 548                 offset += n;
 549                 if (offset == PAGE_SIZE) {
 550                         page++;
 551                         offset = 0;
 552                 }
 553         }
 554         return res;
 555 }
 556 EXPORT_SYMBOL(copy_page_from_iter);
 557
 558 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 559 {
 560         iterate_and_advance(i, bytes, base, len, count,
 561                 clear_user(base, len),
 562                 memset(base, 0, len)
 563         )
 564
 565         return bytes;
 566 }
 567 EXPORT_SYMBOL(iov_iter_zero);
 568
 569 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 570                                   struct iov_iter *i)
 571 {
 572         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 573         if (!page_copy_sane(page, offset, bytes)) {
 574                 kunmap_atomic(kaddr);
 575                 return 0;
 576         }
 577         if (WARN_ON_ONCE(!i->data_source)) {
 578                 kunmap_atomic(kaddr);
 579                 return 0;
 580         }
 581         iterate_and_advance(i, bytes, base, len, off,
 582                 copyin(p + off, base, len),
 583                 memcpy_from_iter(i, p + off, base, len)
 584         )
 585         kunmap_atomic(kaddr);
 586         return bytes;
 587 }
 588 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 589
 590 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 591 {
 592         const struct bio_vec *bvec, *end;
 593
 594         if (!i->count)
 595                 return;
 596         i->count -= size;
 597
 598         size += i->iov_offset;
 599
 600         for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
 601                 if (likely(size < bvec->bv_len))
 602                         break;
 603                 size -= bvec->bv_len;
 604         }
 605         i->iov_offset = size;
 606         i->nr_segs -= bvec - i->bvec;
 607         i->bvec = bvec;
 608 }
 609
 610 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 611 {
 612         const struct iovec *iov, *end;
 613
 614         if (!i->count)
 615                 return;
 616         i->count -= size;
 617
 618         size += i->iov_offset; // from beginning of current segment
 619         for (iov = iter_iov(i), end = iov + i->nr_segs; iov < end; iov++) {
 620                 if (likely(size < iov->iov_len))
 621                         break;
 622                 size -= iov->iov_len;
 623         }
 624         i->iov_offset = size;
 625         i->nr_segs -= iov - iter_iov(i);
 626         i->__iov = iov;
 627 }
 628
 629 void iov_iter_advance(struct iov_iter *i, size_t size)
 630 {
 631         if (unlikely(i->count < size))
 632                 size = i->count;
 633         if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
 634                 i->iov_offset += size;
 635                 i->count -= size;
 636         } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 637                 /* iovec and kvec have identical layouts */
 638                 iov_iter_iovec_advance(i, size);
 639         } else if (iov_iter_is_bvec(i)) {
 640                 iov_iter_bvec_advance(i, size);
 641         } else if (iov_iter_is_discard(i)) {
 642                 i->count -= size;
 643         }
 644 }
 645 EXPORT_SYMBOL(iov_iter_advance);
 646
 647 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 648 {
 649         if (!unroll)
 650                 return;
 651         if (WARN_ON(unroll > MAX_RW_COUNT))
 652                 return;
 653         i->count += unroll;
 654         if (unlikely(iov_iter_is_discard(i)))
 655                 return;
 656         if (unroll <= i->iov_offset) {
 657                 i->iov_offset -= unroll;
 658                 return;
 659         }
 660         unroll -= i->iov_offset;
 661         if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
 662                 BUG(); /* We should never go beyond the start of the specified
 663                         * range since we might then be straying into pages that
 664                         * aren't pinned.
 665                         */
 666         } else if (iov_iter_is_bvec(i)) {
 667                 const struct bio_vec *bvec = i->bvec;
 668                 while (1) {
 669                         size_t n = (--bvec)->bv_len;
 670                         i->nr_segs++;
 671                         if (unroll <= n) {
 672                                 i->bvec = bvec;
 673                                 i->iov_offset = n - unroll;
 674                                 return;
 675                         }
 676                         unroll -= n;
 677                 }
 678         } else { /* same logics for iovec and kvec */
 679                 const struct iovec *iov = iter_iov(i);
 680                 while (1) {
 681                         size_t n = (--iov)->iov_len;
 682                         i->nr_segs++;
 683                         if (unroll <= n) {
 684                                 i->__iov = iov;
 685                                 i->iov_offset = n - unroll;
 686                                 return;
 687                         }
 688                         unroll -= n;
 689                 }
 690         }
 691 }
 692 EXPORT_SYMBOL(iov_iter_revert);
 693
 694 /*
 695  * Return the count of just the current iov_iter segment.
 696  */
 697 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 698 {
 699         if (i->nr_segs > 1) {
 700                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 701                         return min(i->count, iter_iov(i)->iov_len - i->iov_offset);
 702                 if (iov_iter_is_bvec(i))
 703                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 704         }
 705         return i->count;
 706 }
 707 EXPORT_SYMBOL(iov_iter_single_seg_count);
 708
 709 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
 710                         const struct kvec *kvec, unsigned long nr_segs,
 711                         size_t count)
 712 {
 713         WARN_ON(direction & ~(READ | WRITE));
 714         *i = (struct iov_iter){
 715                 .iter_type = ITER_KVEC,
 716                 .copy_mc = false,
 717                 .data_source = direction,
 718                 .kvec = kvec,
 719                 .nr_segs = nr_segs,
 720                 .iov_offset = 0,
 721                 .count = count
 722         };
 723 }
 724 EXPORT_SYMBOL(iov_iter_kvec);
 725
 726 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
 727                         const struct bio_vec *bvec, unsigned long nr_segs,
 728                         size_t count)
 729 {
 730         WARN_ON(direction & ~(READ | WRITE));
 731         *i = (struct iov_iter){
 732                 .iter_type = ITER_BVEC,
 733                 .copy_mc = false,
 734                 .data_source = direction,
 735                 .bvec = bvec,
 736                 .nr_segs = nr_segs,
 737                 .iov_offset = 0,
 738                 .count = count
 739         };
 740 }
 741 EXPORT_SYMBOL(iov_iter_bvec);
 742
 743 /**
 744  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
 745  * @i: The iterator to initialise.
 746  * @direction: The direction of the transfer.
 747  * @xarray: The xarray to access.
 748  * @start: The start file position.
 749  * @count: The size of the I/O buffer in bytes.
 750  *
 751  * Set up an I/O iterator to either draw data out of the pages attached to an
 752  * inode or to inject data into those pages.  The pages *must* be prevented
 753  * from evaporation, either by taking a ref on them or locking them by the
 754  * caller.
 755  */
 756 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
 757                      struct xarray *xarray, loff_t start, size_t count)
 758 {
 759         BUG_ON(direction & ~1);
 760         *i = (struct iov_iter) {
 761                 .iter_type = ITER_XARRAY,
 762                 .copy_mc = false,
 763                 .data_source = direction,
 764                 .xarray = xarray,
 765                 .xarray_start = start,
 766                 .count = count,
 767                 .iov_offset = 0
 768         };
 769 }
 770 EXPORT_SYMBOL(iov_iter_xarray);
 771
 772 /**
 773  * iov_iter_discard - Initialise an I/O iterator that discards data
 774  * @i: The iterator to initialise.
 775  * @direction: The direction of the transfer.
 776  * @count: The size of the I/O buffer in bytes.
 777  *
 778  * Set up an I/O iterator that just discards everything that's written to it.
 779  * It's only available as a READ iterator.
 780  */
 781 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
 782 {
 783         BUG_ON(direction != READ);
 784         *i = (struct iov_iter){
 785                 .iter_type = ITER_DISCARD,
 786                 .copy_mc = false,
 787                 .data_source = false,
 788                 .count = count,
 789                 .iov_offset = 0
 790         };
 791 }
 792 EXPORT_SYMBOL(iov_iter_discard);
 793
 794 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
 795                                    unsigned len_mask)
 796 {
 797         size_t size = i->count;
 798         size_t skip = i->iov_offset;
 799         unsigned k;
 800
 801         for (k = 0; k < i->nr_segs; k++, skip = 0) {
 802                 const struct iovec *iov = iter_iov(i) + k;
 803                 size_t len = iov->iov_len - skip;
 804
 805                 if (len > size)
 806                         len = size;
 807                 if (len & len_mask)
 808                         return false;
 809                 if ((unsigned long)(iov->iov_base + skip) & addr_mask)
 810                         return false;
 811
 812                 size -= len;
 813                 if (!size)
 814                         break;
 815         }
 816         return true;
 817 }
 818
 819 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
 820                                   unsigned len_mask)
 821 {
 822         size_t size = i->count;
 823         unsigned skip = i->iov_offset;
 824         unsigned k;
 825
 826         for (k = 0; k < i->nr_segs; k++, skip = 0) {
 827                 size_t len = i->bvec[k].bv_len - skip;
 828
 829                 if (len > size)
 830                         len = size;
 831                 if (len & len_mask)
 832                         return false;
 833                 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
 834                         return false;
 835
 836                 size -= len;
 837                 if (!size)
 838                         break;
 839         }
 840         return true;
 841 }
 842
 843 /**
 844  * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
 845  *      are aligned to the parameters.
 846  *
 847  * @i: &struct iov_iter to restore
 848  * @addr_mask: bit mask to check against the iov element's addresses
 849  * @len_mask: bit mask to check against the iov element's lengths
 850  *
 851  * Return: false if any addresses or lengths intersect with the provided masks
 852  */
 853 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
 854                          unsigned len_mask)
 855 {
 856         if (likely(iter_is_ubuf(i))) {
 857                 if (i->count & len_mask)
 858                         return false;
 859                 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
 860                         return false;
 861                 return true;
 862         }
 863
 864         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 865                 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
 866
 867         if (iov_iter_is_bvec(i))
 868                 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
 869
 870         if (iov_iter_is_xarray(i)) {
 871                 if (i->count & len_mask)
 872                         return false;
 873                 if ((i->xarray_start + i->iov_offset) & addr_mask)
 874                         return false;
 875         }
 876
 877         return true;
 878 }
 879 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
 880
 881 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
 882 {
 883         unsigned long res = 0;
 884         size_t size = i->count;
 885         size_t skip = i->iov_offset;
 886         unsigned k;
 887
 888         for (k = 0; k < i->nr_segs; k++, skip = 0) {
 889                 const struct iovec *iov = iter_iov(i) + k;
 890                 size_t len = iov->iov_len - skip;
 891                 if (len) {
 892                         res |= (unsigned long)iov->iov_base + skip;
 893                         if (len > size)
 894                                 len = size;
 895                         res |= len;
 896                         size -= len;
 897                         if (!size)
 898                                 break;
 899                 }
 900         }
 901         return res;
 902 }
 903
 904 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
 905 {
 906         unsigned res = 0;
 907         size_t size = i->count;
 908         unsigned skip = i->iov_offset;
 909         unsigned k;
 910
 911         for (k = 0; k < i->nr_segs; k++, skip = 0) {
 912                 size_t len = i->bvec[k].bv_len - skip;
 913                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
 914                 if (len > size)
 915                         len = size;
 916                 res |= len;
 917                 size -= len;
 918                 if (!size)
 919                         break;
 920         }
 921         return res;
 922 }
 923
 924 unsigned long iov_iter_alignment(const struct iov_iter *i)
 925 {
 926         if (likely(iter_is_ubuf(i))) {
 927                 size_t size = i->count;
 928                 if (size)
 929                         return ((unsigned long)i->ubuf + i->iov_offset) | size;
 930                 return 0;
 931         }
 932
 933         /* iovec and kvec have identical layouts */
 934         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 935                 return iov_iter_alignment_iovec(i);
 936
 937         if (iov_iter_is_bvec(i))
 938                 return iov_iter_alignment_bvec(i);
 939
 940         if (iov_iter_is_xarray(i))
 941                 return (i->xarray_start + i->iov_offset) | i->count;
 942
 943         return 0;
 944 }
 945 EXPORT_SYMBOL(iov_iter_alignment);
 946
 947 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
 948 {
 949         unsigned long res = 0;
 950         unsigned long v = 0;
 951         size_t size = i->count;
 952         unsigned k;
 953
 954         if (iter_is_ubuf(i))
 955                 return 0;
 956
 957         if (WARN_ON(!iter_is_iovec(i)))
 958                 return ~0U;
 959
 960         for (k = 0; k < i->nr_segs; k++) {
 961                 const struct iovec *iov = iter_iov(i) + k;
 962                 if (iov->iov_len) {
 963                         unsigned long base = (unsigned long)iov->iov_base;
 964                         if (v) // if not the first one
 965                                 res |= base | v; // this start | previous end
 966                         v = base + iov->iov_len;
 967                         if (size <= iov->iov_len)
 968                                 break;
 969                         size -= iov->iov_len;
 970                 }
 971         }
 972         return res;
 973 }
 974 EXPORT_SYMBOL(iov_iter_gap_alignment);
 975
 976 static int want_pages_array(struct page ***res, size_t size,
 977                             size_t start, unsigned int maxpages)
 978 {
 979         unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
 980
 981         if (count > maxpages)
 982                 count = maxpages;
 983         WARN_ON(!count);        // caller should've prevented that
 984         if (!*res) {
 985                 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
 986                 if (!*res)
 987                         return 0;
 988         }
 989         return count;
 990 }
 991
 992 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
 993                                           pgoff_t index, unsigned int nr_pages)
 994 {
 995         XA_STATE(xas, xa, index);
 996         struct page *page;
 997         unsigned int ret = 0;
 998
 999         rcu_read_lock();
1000         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1001                 if (xas_retry(&xas, page))
1002                         continue;
1003
1004                 /* Has the page moved or been split? */
1005                 if (unlikely(page != xas_reload(&xas))) {
1006                         xas_reset(&xas);
1007                         continue;
1008                 }
1009
1010                 pages[ret] = find_subpage(page, xas.xa_index);
1011                 get_page(pages[ret]);
1012                 if (++ret == nr_pages)
1013                         break;
1014         }
1015         rcu_read_unlock();
1016         return ret;
1017 }
1018
1019 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1020                                      struct page ***pages, size_t maxsize,
1021                                      unsigned maxpages, size_t *_start_offset)
1022 {
1023         unsigned nr, offset, count;
1024         pgoff_t index;
1025         loff_t pos;
1026
1027         pos = i->xarray_start + i->iov_offset;
1028         index = pos >> PAGE_SHIFT;
1029         offset = pos & ~PAGE_MASK;
1030         *_start_offset = offset;
1031
1032         count = want_pages_array(pages, maxsize, offset, maxpages);
1033         if (!count)
1034                 return -ENOMEM;
1035         nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
1036         if (nr == 0)
1037                 return 0;
1038
1039         maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1040         i->iov_offset += maxsize;
1041         i->count -= maxsize;
1042         return maxsize;
1043 }
1044
1045 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1046 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1047 {
1048         size_t skip;
1049         long k;
1050
1051         if (iter_is_ubuf(i))
1052                 return (unsigned long)i->ubuf + i->iov_offset;
1053
1054         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1055                 const struct iovec *iov = iter_iov(i) + k;
1056                 size_t len = iov->iov_len - skip;
1057
1058                 if (unlikely(!len))
1059                         continue;
1060                 if (*size > len)
1061                         *size = len;
1062                 return (unsigned long)iov->iov_base + skip;
1063         }
1064         BUG(); // if it had been empty, we wouldn't get called
1065 }
1066
1067 /* must be done on non-empty ITER_BVEC one */
1068 static struct page *first_bvec_segment(const struct iov_iter *i,
1069                                        size_t *size, size_t *start)
1070 {
1071         struct page *page;
1072         size_t skip = i->iov_offset, len;
1073
1074         len = i->bvec->bv_len - skip;
1075         if (*size > len)
1076                 *size = len;
1077         skip += i->bvec->bv_offset;
1078         page = i->bvec->bv_page + skip / PAGE_SIZE;
1079         *start = skip % PAGE_SIZE;
1080         return page;
1081 }
1082
1083 static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
1084                    struct page ***pages, size_t maxsize,
1085                    unsigned int maxpages, size_t *start,
1086                    iov_iter_extraction_t extraction_flags)
1087 {
1088         unsigned int n, gup_flags = 0;
1089
1090         if (maxsize > i->count)
1091                 maxsize = i->count;
1092         if (!maxsize)
1093                 return 0;
1094         if (maxsize > MAX_RW_COUNT)
1095                 maxsize = MAX_RW_COUNT;
1096         if (extraction_flags & ITER_ALLOW_P2PDMA)
1097                 gup_flags |= FOLL_PCI_P2PDMA;
1098
1099         if (likely(user_backed_iter(i))) {
1100                 unsigned long addr;
1101                 int res;
1102
1103                 if (iov_iter_rw(i) != WRITE)
1104                         gup_flags |= FOLL_WRITE;
1105                 if (i->nofault)
1106                         gup_flags |= FOLL_NOFAULT;
1107
1108                 addr = first_iovec_segment(i, &maxsize);
1109                 *start = addr % PAGE_SIZE;
1110                 addr &= PAGE_MASK;
1111                 n = want_pages_array(pages, maxsize, *start, maxpages);
1112                 if (!n)
1113                         return -ENOMEM;
1114                 res = get_user_pages_fast(addr, n, gup_flags, *pages);
1115                 if (unlikely(res <= 0))
1116                         return res;
1117                 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1118                 iov_iter_advance(i, maxsize);
1119                 return maxsize;
1120         }
1121         if (iov_iter_is_bvec(i)) {
1122                 struct page **p;
1123                 struct page *page;
1124
1125                 page = first_bvec_segment(i, &maxsize, start);
1126                 n = want_pages_array(pages, maxsize, *start, maxpages);
1127                 if (!n)
1128                         return -ENOMEM;
1129                 p = *pages;
1130                 for (int k = 0; k < n; k++)
1131                         get_page(p[k] = page + k);
1132                 maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1133                 i->count -= maxsize;
1134                 i->iov_offset += maxsize;
1135                 if (i->iov_offset == i->bvec->bv_len) {
1136                         i->iov_offset = 0;
1137                         i->bvec++;
1138                         i->nr_segs--;
1139                 }
1140                 return maxsize;
1141         }
1142         if (iov_iter_is_xarray(i))
1143                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1144         return -EFAULT;
1145 }
1146
1147 ssize_t iov_iter_get_pages(struct iov_iter *i,
1148                    struct page **pages, size_t maxsize, unsigned maxpages,
1149                    size_t *start, iov_iter_extraction_t extraction_flags)
1150 {
1151         if (!maxpages)
1152                 return 0;
1153         BUG_ON(!pages);
1154
1155         return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages,
1156                                           start, extraction_flags);
1157 }
1158 EXPORT_SYMBOL_GPL(iov_iter_get_pages);
1159
1160 ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages,
1161                 size_t maxsize, unsigned maxpages, size_t *start)
1162 {
1163         return iov_iter_get_pages(i, pages, maxsize, maxpages, start, 0);
1164 }
1165 EXPORT_SYMBOL(iov_iter_get_pages2);
1166
1167 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1168                    struct page ***pages, size_t maxsize,
1169                    size_t *start, iov_iter_extraction_t extraction_flags)
1170 {
1171         ssize_t len;
1172
1173         *pages = NULL;
1174
1175         len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start,
1176                                          extraction_flags);
1177         if (len <= 0) {
1178                 kvfree(*pages);
1179                 *pages = NULL;
1180         }
1181         return len;
1182 }
1183 EXPORT_SYMBOL_GPL(iov_iter_get_pages_alloc);
1184
1185 ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1186                 struct page ***pages, size_t maxsize, size_t *start)
1187 {
1188         return iov_iter_get_pages_alloc(i, pages, maxsize, start, 0);
1189 }
1190 EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1191
1192 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1193                                struct iov_iter *i)
1194 {
1195         __wsum sum, next;
1196         sum = *csum;
1197         if (WARN_ON_ONCE(!i->data_source))
1198                 return 0;
1199
1200         iterate_and_advance(i, bytes, base, len, off, ({
1201                 next = csum_and_copy_from_user(base, addr + off, len);
1202                 sum = csum_block_add(sum, next, off);
1203                 next ? 0 : len;
1204         }), ({
1205                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1206         })
1207         )
1208         *csum = sum;
1209         return bytes;
1210 }
1211 EXPORT_SYMBOL(csum_and_copy_from_iter);
1212
1213 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1214                              struct iov_iter *i)
1215 {
1216         struct csum_state *csstate = _csstate;
1217         __wsum sum, next;
1218
1219         if (WARN_ON_ONCE(i->data_source))
1220                 return 0;
1221         if (unlikely(iov_iter_is_discard(i))) {
1222                 // can't use csum_memcpy() for that one - data is not copied
1223                 csstate->csum = csum_block_add(csstate->csum,
1224                                                csum_partial(addr, bytes, 0),
1225                                                csstate->off);
1226                 csstate->off += bytes;
1227                 return bytes;
1228         }
1229
1230         sum = csum_shift(csstate->csum, csstate->off);
1231         iterate_and_advance(i, bytes, base, len, off, ({
1232                 next = csum_and_copy_to_user(addr + off, base, len);
1233                 sum = csum_block_add(sum, next, off);
1234                 next ? 0 : len;
1235         }), ({
1236                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1237         })
1238         )
1239         csstate->csum = csum_shift(sum, csstate->off);
1240         csstate->off += bytes;
1241         return bytes;
1242 }
1243 EXPORT_SYMBOL(csum_and_copy_to_iter);
1244
1245 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1246                 struct iov_iter *i)
1247 {
1248 #ifdef CONFIG_CRYPTO_HASH
1249         struct ahash_request *hash = hashp;
1250         struct scatterlist sg;
1251         size_t copied;
1252
1253         copied = copy_to_iter(addr, bytes, i);
1254         sg_init_one(&sg, addr, copied);
1255         ahash_request_set_crypt(hash, &sg, NULL, copied);
1256         crypto_ahash_update(hash);
1257         return copied;
1258 #else
1259         return 0;
1260 #endif
1261 }
1262 EXPORT_SYMBOL(hash_and_copy_to_iter);
1263
1264 static int iov_npages(const struct iov_iter *i, int maxpages)
1265 {
1266         size_t skip = i->iov_offset, size = i->count;
1267         const struct iovec *p;
1268         int npages = 0;
1269
1270         for (p = iter_iov(i); size; skip = 0, p++) {
1271                 unsigned offs = offset_in_page(p->iov_base + skip);
1272                 size_t len = min(p->iov_len - skip, size);
1273
1274                 if (len) {
1275                         size -= len;
1276                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1277                         if (unlikely(npages > maxpages))
1278                                 return maxpages;
1279                 }
1280         }
1281         return npages;
1282 }
1283
1284 static int bvec_npages(const struct iov_iter *i, int maxpages)
1285 {
1286         size_t skip = i->iov_offset, size = i->count;
1287         const struct bio_vec *p;
1288         int npages = 0;
1289
1290         for (p = i->bvec; size; skip = 0, p++) {
1291                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1292                 size_t len = min(p->bv_len - skip, size);
1293
1294                 size -= len;
1295                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1296                 if (unlikely(npages > maxpages))
1297                         return maxpages;
1298         }
1299         return npages;
1300 }
1301
1302 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1303 {
1304         if (unlikely(!i->count))
1305                 return 0;
1306         if (likely(iter_is_ubuf(i))) {
1307                 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1308                 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1309                 return min(npages, maxpages);
1310         }
1311         /* iovec and kvec have identical layouts */
1312         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1313                 return iov_npages(i, maxpages);
1314         if (iov_iter_is_bvec(i))
1315                 return bvec_npages(i, maxpages);
1316         if (iov_iter_is_xarray(i)) {
1317                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1318                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1319                 return min(npages, maxpages);
1320         }
1321         return 0;
1322 }
1323 EXPORT_SYMBOL(iov_iter_npages);
1324
1325 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1326 {
1327         *new = *old;
1328         if (iov_iter_is_bvec(new))
1329                 return new->bvec = kmemdup(new->bvec,
1330                                     new->nr_segs * sizeof(struct bio_vec),
1331                                     flags);
1332         else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1333                 /* iovec and kvec have identical layout */
1334                 return new->__iov = kmemdup(new->__iov,
1335                                    new->nr_segs * sizeof(struct iovec),
1336                                    flags);
1337         return NULL;
1338 }
1339 EXPORT_SYMBOL(dup_iter);
1340
1341 static __noclone int copy_compat_iovec_from_user(struct iovec *iov,
1342                 const struct iovec __user *uvec, unsigned long nr_segs)
1343 {
1344         const struct compat_iovec __user *uiov =
1345                 (const struct compat_iovec __user *)uvec;
1346         int ret = -EFAULT, i;
1347
1348         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1349                 return -EFAULT;
1350
1351         for (i = 0; i < nr_segs; i++) {
1352                 compat_uptr_t buf;
1353                 compat_ssize_t len;
1354
1355                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1356                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1357
1358                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1359                 if (len < 0) {
1360                         ret = -EINVAL;
1361                         goto uaccess_end;
1362                 }
1363                 iov[i].iov_base = compat_ptr(buf);
1364                 iov[i].iov_len = len;
1365         }
1366
1367         ret = 0;
1368 uaccess_end:
1369         user_access_end();
1370         return ret;
1371 }
1372
1373 static int copy_iovec_from_user(struct iovec *iov,
1374                 const struct iovec __user *uiov, unsigned long nr_segs)
1375 {
1376         int ret = -EFAULT;
1377
1378         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1379                 return -EFAULT;
1380
1381         do {
1382                 void __user *buf;
1383                 ssize_t len;
1384
1385                 unsafe_get_user(len, &uiov->iov_len, uaccess_end);
1386                 unsafe_get_user(buf, &uiov->iov_base, uaccess_end);
1387
1388                 /* check for size_t not fitting in ssize_t .. */
1389                 if (unlikely(len < 0)) {
1390                         ret = -EINVAL;
1391                         goto uaccess_end;
1392                 }
1393                 iov->iov_base = buf;
1394                 iov->iov_len = len;
1395
1396                 uiov++; iov++;
1397         } while (--nr_segs);
1398
1399         ret = 0;
1400 uaccess_end:
1401         user_access_end();
1402         return ret;
1403 }
1404
1405 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1406                 unsigned long nr_segs, unsigned long fast_segs,
1407                 struct iovec *fast_iov, bool compat)
1408 {
1409         struct iovec *iov = fast_iov;
1410         int ret;
1411
1412         /*
1413          * SuS says "The readv() function *may* fail if the iovcnt argument was
1414          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1415          * traditionally returned zero for zero segments, so...
1416          */
1417         if (nr_segs == 0)
1418                 return iov;
1419         if (nr_segs > UIO_MAXIOV)
1420                 return ERR_PTR(-EINVAL);
1421         if (nr_segs > fast_segs) {
1422                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1423                 if (!iov)
1424                         return ERR_PTR(-ENOMEM);
1425         }
1426
1427         if (unlikely(compat))
1428                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1429         else
1430                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1431         if (ret) {
1432                 if (iov != fast_iov)
1433                         kfree(iov);
1434                 return ERR_PTR(ret);
1435         }
1436
1437         return iov;
1438 }
1439
1440 /*
1441  * Single segment iovec supplied by the user, import it as ITER_UBUF.
1442  */
1443 static ssize_t __import_iovec_ubuf(int type, const struct iovec __user *uvec,
1444                                    struct iovec **iovp, struct iov_iter *i,
1445                                    bool compat)
1446 {
1447         struct iovec *iov = *iovp;
1448         ssize_t ret;
1449
1450         if (compat)
1451                 ret = copy_compat_iovec_from_user(iov, uvec, 1);
1452         else
1453                 ret = copy_iovec_from_user(iov, uvec, 1);
1454         if (unlikely(ret))
1455                 return ret;
1456
1457         ret = import_ubuf(type, iov->iov_base, iov->iov_len, i);
1458         if (unlikely(ret))
1459                 return ret;
1460         *iovp = NULL;
1461         return i->count;
1462 }
1463
1464 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1465                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1466                  struct iov_iter *i, bool compat)
1467 {
1468         ssize_t total_len = 0;
1469         unsigned long seg;
1470         struct iovec *iov;
1471
1472         if (nr_segs == 1)
1473                 return __import_iovec_ubuf(type, uvec, iovp, i, compat);
1474
1475         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1476         if (IS_ERR(iov)) {
1477                 *iovp = NULL;
1478                 return PTR_ERR(iov);
1479         }
1480
1481         /*
1482          * According to the Single Unix Specification we should return EINVAL if
1483          * an element length is < 0 when cast to ssize_t or if the total length
1484          * would overflow the ssize_t return value of the system call.
1485          *
1486          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1487          * overflow case.
1488          */
1489         for (seg = 0; seg < nr_segs; seg++) {
1490                 ssize_t len = (ssize_t)iov[seg].iov_len;
1491
1492                 if (!access_ok(iov[seg].iov_base, len)) {
1493                         if (iov != *iovp)
1494                                 kfree(iov);
1495                         *iovp = NULL;
1496                         return -EFAULT;
1497                 }
1498
1499                 if (len > MAX_RW_COUNT - total_len) {
1500                         len = MAX_RW_COUNT - total_len;
1501                         iov[seg].iov_len = len;
1502                 }
1503                 total_len += len;
1504         }
1505
1506         iov_iter_init(i, type, iov, nr_segs, total_len);
1507         if (iov == *iovp)
1508                 *iovp = NULL;
1509         else
1510                 *iovp = iov;
1511         return total_len;
1512 }
1513
1514 /**
1515  * import_iovec() - Copy an array of &struct iovec from userspace
1516  *     into the kernel, check that it is valid, and initialize a new
1517  *     &struct iov_iter iterator to access it.
1518  *
1519  * @type: One of %READ or %WRITE.
1520  * @uvec: Pointer to the userspace array.
1521  * @nr_segs: Number of elements in userspace array.
1522  * @fast_segs: Number of elements in @iov.
1523  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1524  *     on-stack) kernel array.
1525  * @i: Pointer to iterator that will be initialized on success.
1526  *
1527  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1528  * then this function places %NULL in *@iov on return. Otherwise, a new
1529  * array will be allocated and the result placed in *@iov. This means that
1530  * the caller may call kfree() on *@iov regardless of whether the small
1531  * on-stack array was used or not (and regardless of whether this function
1532  * returns an error or not).
1533  *
1534  * Return: Negative error code on error, bytes imported on success
1535  */
1536 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1537                  unsigned nr_segs, unsigned fast_segs,
1538                  struct iovec **iovp, struct iov_iter *i)
1539 {
1540         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1541                               in_compat_syscall());
1542 }
1543 EXPORT_SYMBOL(import_iovec);
1544
1545 int import_single_range(int rw, void __user *buf, size_t len,
1546                  struct iovec *iov, struct iov_iter *i)
1547 {
1548         if (len > MAX_RW_COUNT)
1549                 len = MAX_RW_COUNT;
1550         if (unlikely(!access_ok(buf, len)))
1551                 return -EFAULT;
1552
1553         iov_iter_ubuf(i, rw, buf, len);
1554         return 0;
1555 }
1556 EXPORT_SYMBOL(import_single_range);
1557
1558 int import_ubuf(int rw, void __user *buf, size_t len, struct iov_iter *i)
1559 {
1560         if (len > MAX_RW_COUNT)
1561                 len = MAX_RW_COUNT;
1562         if (unlikely(!access_ok(buf, len)))
1563                 return -EFAULT;
1564
1565         iov_iter_ubuf(i, rw, buf, len);
1566         return 0;
1567 }
1568
1569 /**
1570  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1571  *     iov_iter_save_state() was called.
1572  *
1573  * @i: &struct iov_iter to restore
1574  * @state: state to restore from
1575  *
1576  * Used after iov_iter_save_state() to bring restore @i, if operations may
1577  * have advanced it.
1578  *
1579  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1580  */
1581 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1582 {
1583         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i) &&
1584                          !iter_is_ubuf(i)) && !iov_iter_is_kvec(i))
1585                 return;
1586         i->iov_offset = state->iov_offset;
1587         i->count = state->count;
1588         if (iter_is_ubuf(i))
1589                 return;
1590         /*
1591          * For the *vec iters, nr_segs + iov is constant - if we increment
1592          * the vec, then we also decrement the nr_segs count. Hence we don't
1593          * need to track both of these, just one is enough and we can deduct
1594          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1595          * size, so we can just increment the iov pointer as they are unionzed.
1596          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1597          * not. Be safe and handle it separately.
1598          */
1599         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1600         if (iov_iter_is_bvec(i))
1601                 i->bvec -= state->nr_segs - i->nr_segs;
1602         else
1603                 i->__iov -= state->nr_segs - i->nr_segs;
1604         i->nr_segs = state->nr_segs;
1605 }
1606
1607 /*
1608  * Extract a list of contiguous pages from an ITER_XARRAY iterator.  This does not
1609  * get references on the pages, nor does it get a pin on them.
1610  */
1611 static ssize_t iov_iter_extract_xarray_pages(struct iov_iter *i,
1612                                              struct page ***pages, size_t maxsize,
1613                                              unsigned int maxpages,
1614                                              iov_iter_extraction_t extraction_flags,
1615                                              size_t *offset0)
1616 {
1617         struct page *page, **p;
1618         unsigned int nr = 0, offset;
1619         loff_t pos = i->xarray_start + i->iov_offset;
1620         pgoff_t index = pos >> PAGE_SHIFT;
1621         XA_STATE(xas, i->xarray, index);
1622
1623         offset = pos & ~PAGE_MASK;
1624         *offset0 = offset;
1625
1626         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1627         if (!maxpages)
1628                 return -ENOMEM;
1629         p = *pages;
1630
1631         rcu_read_lock();
1632         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1633                 if (xas_retry(&xas, page))
1634                         continue;
1635
1636                 /* Has the page moved or been split? */
1637                 if (unlikely(page != xas_reload(&xas))) {
1638                         xas_reset(&xas);
1639                         continue;
1640                 }
1641
1642                 p[nr++] = find_subpage(page, xas.xa_index);
1643                 if (nr == maxpages)
1644                         break;
1645         }
1646         rcu_read_unlock();
1647
1648         maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1649         iov_iter_advance(i, maxsize);
1650         return maxsize;
1651 }
1652
1653 /*
1654  * Extract a list of contiguous pages from an ITER_BVEC iterator.  This does
1655  * not get references on the pages, nor does it get a pin on them.
1656  */
1657 static ssize_t iov_iter_extract_bvec_pages(struct iov_iter *i,
1658                                            struct page ***pages, size_t maxsize,
1659                                            unsigned int maxpages,
1660                                            iov_iter_extraction_t extraction_flags,
1661                                            size_t *offset0)
1662 {
1663         struct page **p, *page;
1664         size_t skip = i->iov_offset, offset;
1665         int k;
1666
1667         for (;;) {
1668                 if (i->nr_segs == 0)
1669                         return 0;
1670                 maxsize = min(maxsize, i->bvec->bv_len - skip);
1671                 if (maxsize)
1672                         break;
1673                 i->iov_offset = 0;
1674                 i->nr_segs--;
1675                 i->bvec++;
1676                 skip = 0;
1677         }
1678
1679         skip += i->bvec->bv_offset;
1680         page = i->bvec->bv_page + skip / PAGE_SIZE;
1681         offset = skip % PAGE_SIZE;
1682         *offset0 = offset;
1683
1684         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1685         if (!maxpages)
1686                 return -ENOMEM;
1687         p = *pages;
1688         for (k = 0; k < maxpages; k++)
1689                 p[k] = page + k;
1690
1691         maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset);
1692         iov_iter_advance(i, maxsize);
1693         return maxsize;
1694 }
1695
1696 /*
1697  * Extract a list of virtually contiguous pages from an ITER_KVEC iterator.
1698  * This does not get references on the pages, nor does it get a pin on them.
1699  */
1700 static ssize_t iov_iter_extract_kvec_pages(struct iov_iter *i,
1701                                            struct page ***pages, size_t maxsize,
1702                                            unsigned int maxpages,
1703                                            iov_iter_extraction_t extraction_flags,
1704                                            size_t *offset0)
1705 {
1706         struct page **p, *page;
1707         const void *kaddr;
1708         size_t skip = i->iov_offset, offset, len;
1709         int k;
1710
1711         for (;;) {
1712                 if (i->nr_segs == 0)
1713                         return 0;
1714                 maxsize = min(maxsize, i->kvec->iov_len - skip);
1715                 if (maxsize)
1716                         break;
1717                 i->iov_offset = 0;
1718                 i->nr_segs--;
1719                 i->kvec++;
1720                 skip = 0;
1721         }
1722
1723         kaddr = i->kvec->iov_base + skip;
1724         offset = (unsigned long)kaddr & ~PAGE_MASK;
1725         *offset0 = offset;
1726
1727         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1728         if (!maxpages)
1729                 return -ENOMEM;
1730         p = *pages;
1731
1732         kaddr -= offset;
1733         len = offset + maxsize;
1734         for (k = 0; k < maxpages; k++) {
1735                 size_t seg = min_t(size_t, len, PAGE_SIZE);
1736
1737                 if (is_vmalloc_or_module_addr(kaddr))
1738                         page = vmalloc_to_page(kaddr);
1739                 else
1740                         page = virt_to_page(kaddr);
1741
1742                 p[k] = page;
1743                 len -= seg;
1744                 kaddr += PAGE_SIZE;
1745         }
1746
1747         maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset);
1748         iov_iter_advance(i, maxsize);
1749         return maxsize;
1750 }
1751
1752 /*
1753  * Extract a list of contiguous pages from a user iterator and get a pin on
1754  * each of them.  This should only be used if the iterator is user-backed
1755  * (IOBUF/UBUF).
1756  *
1757  * It does not get refs on the pages, but the pages must be unpinned by the
1758  * caller once the transfer is complete.
1759  *
1760  * This is safe to be used where background IO/DMA *is* going to be modifying
1761  * the buffer; using a pin rather than a ref makes forces fork() to give the
1762  * child a copy of the page.
1763  */
1764 static ssize_t iov_iter_extract_user_pages(struct iov_iter *i,
1765                                            struct page ***pages,
1766                                            size_t maxsize,
1767                                            unsigned int maxpages,
1768                                            iov_iter_extraction_t extraction_flags,
1769                                            size_t *offset0)
1770 {
1771         unsigned long addr;
1772         unsigned int gup_flags = 0;
1773         size_t offset;
1774         int res;
1775
1776         if (i->data_source == ITER_DEST)
1777                 gup_flags |= FOLL_WRITE;
1778         if (extraction_flags & ITER_ALLOW_P2PDMA)
1779                 gup_flags |= FOLL_PCI_P2PDMA;
1780         if (i->nofault)
1781                 gup_flags |= FOLL_NOFAULT;
1782
1783         addr = first_iovec_segment(i, &maxsize);
1784         *offset0 = offset = addr % PAGE_SIZE;
1785         addr &= PAGE_MASK;
1786         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1787         if (!maxpages)
1788                 return -ENOMEM;
1789         res = pin_user_pages_fast(addr, maxpages, gup_flags, *pages);
1790         if (unlikely(res <= 0))
1791                 return res;
1792         maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - offset);
1793         iov_iter_advance(i, maxsize);
1794         return maxsize;
1795 }
1796
1797 /**
1798  * iov_iter_extract_pages - Extract a list of contiguous pages from an iterator
1799  * @i: The iterator to extract from
1800  * @pages: Where to return the list of pages
1801  * @maxsize: The maximum amount of iterator to extract
1802  * @maxpages: The maximum size of the list of pages
1803  * @extraction_flags: Flags to qualify request
1804  * @offset0: Where to return the starting offset into (*@pages)[0]
1805  *
1806  * Extract a list of contiguous pages from the current point of the iterator,
1807  * advancing the iterator.  The maximum number of pages and the maximum amount
1808  * of page contents can be set.
1809  *
1810  * If *@pages is NULL, a page list will be allocated to the required size and
1811  * *@pages will be set to its base.  If *@pages is not NULL, it will be assumed
1812  * that the caller allocated a page list at least @maxpages in size and this
1813  * will be filled in.
1814  *
1815  * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA
1816  * be allowed on the pages extracted.
1817  *
1818  * The iov_iter_extract_will_pin() function can be used to query how cleanup
1819  * should be performed.
1820  *
1821  * Extra refs or pins on the pages may be obtained as follows:
1822  *
1823  *  (*) If the iterator is user-backed (ITER_IOVEC/ITER_UBUF), pins will be
1824  *      added to the pages, but refs will not be taken.
1825  *      iov_iter_extract_will_pin() will return true.
1826  *
1827  *  (*) If the iterator is ITER_KVEC, ITER_BVEC or ITER_XARRAY, the pages are
1828  *      merely listed; no extra refs or pins are obtained.
1829  *      iov_iter_extract_will_pin() will return 0.
1830  *
1831  * Note also:
1832  *
1833  *  (*) Use with ITER_DISCARD is not supported as that has no content.
1834  *
1835  * On success, the function sets *@pages to the new pagelist, if allocated, and
1836  * sets *offset0 to the offset into the first page.
1837  *
1838  * It may also return -ENOMEM and -EFAULT.
1839  */
1840 ssize_t iov_iter_extract_pages(struct iov_iter *i,
1841                                struct page ***pages,
1842                                size_t maxsize,
1843                                unsigned int maxpages,
1844                                iov_iter_extraction_t extraction_flags,
1845                                size_t *offset0)
1846 {
1847         maxsize = min_t(size_t, min_t(size_t, maxsize, i->count), MAX_RW_COUNT);
1848         if (!maxsize)
1849                 return 0;
1850
1851         if (likely(user_backed_iter(i)))
1852                 return iov_iter_extract_user_pages(i, pages, maxsize,
1853                                                    maxpages, extraction_flags,
1854                                                    offset0);
1855         if (iov_iter_is_kvec(i))
1856                 return iov_iter_extract_kvec_pages(i, pages, maxsize,
1857                                                    maxpages, extraction_flags,
1858                                                    offset0);
1859         if (iov_iter_is_bvec(i))
1860                 return iov_iter_extract_bvec_pages(i, pages, maxsize,
1861                                                    maxpages, extraction_flags,
1862                                                    offset0);
1863         if (iov_iter_is_xarray(i))
1864                 return iov_iter_extract_xarray_pages(i, pages, maxsize,
1865                                                      maxpages, extraction_flags,
1866                                                      offset0);
1867         return -EFAULT;
1868 }
1869 EXPORT_SYMBOL_GPL(iov_iter_extract_pages);