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