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  * @i: 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  * Return: number of bytes copied (may be %0)
 695  */
 696 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 697 {
 698         if (unlikely(iov_iter_is_pipe(i)))
 699                 return copy_mc_pipe_to_iter(addr, bytes, i);
 700         if (iter_is_iovec(i))
 701                 might_fault();
 702         __iterate_and_advance(i, bytes, base, len, off,
 703                 copyout_mc(base, addr + off, len),
 704                 copy_mc_to_kernel(base, addr + off, len)
 705         )
 706
 707         return bytes;
 708 }
 709 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 710 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 711
 712 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 713 {
 714         if (unlikely(iov_iter_is_pipe(i))) {
 715                 WARN_ON(1);
 716                 return 0;
 717         }
 718         if (iter_is_iovec(i))
 719                 might_fault();
 720         iterate_and_advance(i, bytes, base, len, off,
 721                 copyin(addr + off, base, len),
 722                 memcpy(addr + off, base, len)
 723         )
 724
 725         return bytes;
 726 }
 727 EXPORT_SYMBOL(_copy_from_iter);
 728
 729 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 730 {
 731         if (unlikely(iov_iter_is_pipe(i))) {
 732                 WARN_ON(1);
 733                 return 0;
 734         }
 735         iterate_and_advance(i, bytes, base, len, off,
 736                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 737                 memcpy(addr + off, base, len)
 738         )
 739
 740         return bytes;
 741 }
 742 EXPORT_SYMBOL(_copy_from_iter_nocache);
 743
 744 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 745 /**
 746  * _copy_from_iter_flushcache - write destination through cpu cache
 747  * @addr: destination kernel address
 748  * @bytes: total transfer length
 749  * @i: source iterator
 750  *
 751  * The pmem driver arranges for filesystem-dax to use this facility via
 752  * dax_copy_from_iter() for ensuring that writes to persistent memory
 753  * are flushed through the CPU cache. It is differentiated from
 754  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 755  * all iterator types. The _copy_from_iter_nocache() only attempts to
 756  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 757  * instructions that strand dirty-data in the cache.
 758  *
 759  * Return: number of bytes copied (may be %0)
 760  */
 761 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 762 {
 763         if (unlikely(iov_iter_is_pipe(i))) {
 764                 WARN_ON(1);
 765                 return 0;
 766         }
 767         iterate_and_advance(i, bytes, base, len, off,
 768                 __copy_from_user_flushcache(addr + off, base, len),
 769                 memcpy_flushcache(addr + off, base, len)
 770         )
 771
 772         return bytes;
 773 }
 774 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 775 #endif
 776
 777 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 778 {
 779         struct page *head;
 780         size_t v = n + offset;
 781
 782         /*
 783          * The general case needs to access the page order in order
 784          * to compute the page size.
 785          * However, we mostly deal with order-0 pages and thus can
 786          * avoid a possible cache line miss for requests that fit all
 787          * page orders.
 788          */
 789         if (n <= v && v <= PAGE_SIZE)
 790                 return true;
 791
 792         head = compound_head(page);
 793         v += (page - head) << PAGE_SHIFT;
 794
 795         if (likely(n <= v && v <= (page_size(head))))
 796                 return true;
 797         WARN_ON(1);
 798         return false;
 799 }
 800
 801 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 802                          struct iov_iter *i)
 803 {
 804         if (likely(iter_is_iovec(i)))
 805                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 806         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 807                 void *kaddr = kmap_local_page(page);
 808                 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
 809                 kunmap_local(kaddr);
 810                 return wanted;
 811         }
 812         if (iov_iter_is_pipe(i))
 813                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 814         if (unlikely(iov_iter_is_discard(i))) {
 815                 if (unlikely(i->count < bytes))
 816                         bytes = i->count;
 817                 i->count -= bytes;
 818                 return bytes;
 819         }
 820         WARN_ON(1);
 821         return 0;
 822 }
 823
 824 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 825                          struct iov_iter *i)
 826 {
 827         size_t res = 0;
 828         if (unlikely(!page_copy_sane(page, offset, bytes)))
 829                 return 0;
 830         page += offset / PAGE_SIZE; // first subpage
 831         offset %= PAGE_SIZE;
 832         while (1) {
 833                 size_t n = __copy_page_to_iter(page, offset,
 834                                 min(bytes, (size_t)PAGE_SIZE - offset), i);
 835                 res += n;
 836                 bytes -= n;
 837                 if (!bytes || !n)
 838                         break;
 839                 offset += n;
 840                 if (offset == PAGE_SIZE) {
 841                         page++;
 842                         offset = 0;
 843                 }
 844         }
 845         return res;
 846 }
 847 EXPORT_SYMBOL(copy_page_to_iter);
 848
 849 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 850                          struct iov_iter *i)
 851 {
 852         if (unlikely(!page_copy_sane(page, offset, bytes)))
 853                 return 0;
 854         if (likely(iter_is_iovec(i)))
 855                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 856         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 857                 void *kaddr = kmap_local_page(page);
 858                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 859                 kunmap_local(kaddr);
 860                 return wanted;
 861         }
 862         WARN_ON(1);
 863         return 0;
 864 }
 865 EXPORT_SYMBOL(copy_page_from_iter);
 866
 867 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 868 {
 869         struct pipe_inode_info *pipe = i->pipe;
 870         unsigned int p_mask = pipe->ring_size - 1;
 871         unsigned int i_head;
 872         size_t n, off;
 873
 874         if (!sanity(i))
 875                 return 0;
 876
 877         bytes = n = push_pipe(i, bytes, &i_head, &off);
 878         if (unlikely(!n))
 879                 return 0;
 880
 881         do {
 882                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 883                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 884                 memset(p + off, 0, chunk);
 885                 kunmap_local(p);
 886                 i->head = i_head;
 887                 i->iov_offset = off + chunk;
 888                 n -= chunk;
 889                 off = 0;
 890                 i_head++;
 891         } while (n);
 892         i->count -= bytes;
 893         return bytes;
 894 }
 895
 896 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 897 {
 898         if (unlikely(iov_iter_is_pipe(i)))
 899                 return pipe_zero(bytes, i);
 900         iterate_and_advance(i, bytes, base, len, count,
 901                 clear_user(base, len),
 902                 memset(base, 0, len)
 903         )
 904
 905         return bytes;
 906 }
 907 EXPORT_SYMBOL(iov_iter_zero);
 908
 909 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 910                                   struct iov_iter *i)
 911 {
 912         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 913         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 914                 kunmap_atomic(kaddr);
 915                 return 0;
 916         }
 917         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 918                 kunmap_atomic(kaddr);
 919                 WARN_ON(1);
 920                 return 0;
 921         }
 922         iterate_and_advance(i, bytes, base, len, off,
 923                 copyin(p + off, base, len),
 924                 memcpy(p + off, base, len)
 925         )
 926         kunmap_atomic(kaddr);
 927         return bytes;
 928 }
 929 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 930
 931 static inline void pipe_truncate(struct iov_iter *i)
 932 {
 933         struct pipe_inode_info *pipe = i->pipe;
 934         unsigned int p_tail = pipe->tail;
 935         unsigned int p_head = pipe->head;
 936         unsigned int p_mask = pipe->ring_size - 1;
 937
 938         if (!pipe_empty(p_head, p_tail)) {
 939                 struct pipe_buffer *buf;
 940                 unsigned int i_head = i->head;
 941                 size_t off = i->iov_offset;
 942
 943                 if (off) {
 944                         buf = &pipe->bufs[i_head & p_mask];
 945                         buf->len = off - buf->offset;
 946                         i_head++;
 947                 }
 948                 while (p_head != i_head) {
 949                         p_head--;
 950                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
 951                 }
 952
 953                 pipe->head = p_head;
 954         }
 955 }
 956
 957 static void pipe_advance(struct iov_iter *i, size_t size)
 958 {
 959         struct pipe_inode_info *pipe = i->pipe;
 960         if (size) {
 961                 struct pipe_buffer *buf;
 962                 unsigned int p_mask = pipe->ring_size - 1;
 963                 unsigned int i_head = i->head;
 964                 size_t off = i->iov_offset, left = size;
 965
 966                 if (off) /* make it relative to the beginning of buffer */
 967                         left += off - pipe->bufs[i_head & p_mask].offset;
 968                 while (1) {
 969                         buf = &pipe->bufs[i_head & p_mask];
 970                         if (left <= buf->len)
 971                                 break;
 972                         left -= buf->len;
 973                         i_head++;
 974                 }
 975                 i->head = i_head;
 976                 i->iov_offset = buf->offset + left;
 977         }
 978         i->count -= size;
 979         /* ... and discard everything past that point */
 980         pipe_truncate(i);
 981 }
 982
 983 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 984 {
 985         struct bvec_iter bi;
 986
 987         bi.bi_size = i->count;
 988         bi.bi_bvec_done = i->iov_offset;
 989         bi.bi_idx = 0;
 990         bvec_iter_advance(i->bvec, &bi, size);
 991
 992         i->bvec += bi.bi_idx;
 993         i->nr_segs -= bi.bi_idx;
 994         i->count = bi.bi_size;
 995         i->iov_offset = bi.bi_bvec_done;
 996 }
 997
 998 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 999 {
1000         const struct iovec *iov, *end;
1001
1002         if (!i->count)
1003                 return;
1004         i->count -= size;
1005
1006         size += i->iov_offset; // from beginning of current segment
1007         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1008                 if (likely(size < iov->iov_len))
1009                         break;
1010                 size -= iov->iov_len;
1011         }
1012         i->iov_offset = size;
1013         i->nr_segs -= iov - i->iov;
1014         i->iov = iov;
1015 }
1016
1017 void iov_iter_advance(struct iov_iter *i, size_t size)
1018 {
1019         if (unlikely(i->count < size))
1020                 size = i->count;
1021         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1022                 /* iovec and kvec have identical layouts */
1023                 iov_iter_iovec_advance(i, size);
1024         } else if (iov_iter_is_bvec(i)) {
1025                 iov_iter_bvec_advance(i, size);
1026         } else if (iov_iter_is_pipe(i)) {
1027                 pipe_advance(i, size);
1028         } else if (unlikely(iov_iter_is_xarray(i))) {
1029                 i->iov_offset += size;
1030                 i->count -= size;
1031         } else if (iov_iter_is_discard(i)) {
1032                 i->count -= size;
1033         }
1034 }
1035 EXPORT_SYMBOL(iov_iter_advance);
1036
1037 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1038 {
1039         if (!unroll)
1040                 return;
1041         if (WARN_ON(unroll > MAX_RW_COUNT))
1042                 return;
1043         i->count += unroll;
1044         if (unlikely(iov_iter_is_pipe(i))) {
1045                 struct pipe_inode_info *pipe = i->pipe;
1046                 unsigned int p_mask = pipe->ring_size - 1;
1047                 unsigned int i_head = i->head;
1048                 size_t off = i->iov_offset;
1049                 while (1) {
1050                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1051                         size_t n = off - b->offset;
1052                         if (unroll < n) {
1053                                 off -= unroll;
1054                                 break;
1055                         }
1056                         unroll -= n;
1057                         if (!unroll && i_head == i->start_head) {
1058                                 off = 0;
1059                                 break;
1060                         }
1061                         i_head--;
1062                         b = &pipe->bufs[i_head & p_mask];
1063                         off = b->offset + b->len;
1064                 }
1065                 i->iov_offset = off;
1066                 i->head = i_head;
1067                 pipe_truncate(i);
1068                 return;
1069         }
1070         if (unlikely(iov_iter_is_discard(i)))
1071                 return;
1072         if (unroll <= i->iov_offset) {
1073                 i->iov_offset -= unroll;
1074                 return;
1075         }
1076         unroll -= i->iov_offset;
1077         if (iov_iter_is_xarray(i)) {
1078                 BUG(); /* We should never go beyond the start of the specified
1079                         * range since we might then be straying into pages that
1080                         * aren't pinned.
1081                         */
1082         } else if (iov_iter_is_bvec(i)) {
1083                 const struct bio_vec *bvec = i->bvec;
1084                 while (1) {
1085                         size_t n = (--bvec)->bv_len;
1086                         i->nr_segs++;
1087                         if (unroll <= n) {
1088                                 i->bvec = bvec;
1089                                 i->iov_offset = n - unroll;
1090                                 return;
1091                         }
1092                         unroll -= n;
1093                 }
1094         } else { /* same logics for iovec and kvec */
1095                 const struct iovec *iov = i->iov;
1096                 while (1) {
1097                         size_t n = (--iov)->iov_len;
1098                         i->nr_segs++;
1099                         if (unroll <= n) {
1100                                 i->iov = iov;
1101                                 i->iov_offset = n - unroll;
1102                                 return;
1103                         }
1104                         unroll -= n;
1105                 }
1106         }
1107 }
1108 EXPORT_SYMBOL(iov_iter_revert);
1109
1110 /*
1111  * Return the count of just the current iov_iter segment.
1112  */
1113 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1114 {
1115         if (i->nr_segs > 1) {
1116                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1117                         return min(i->count, i->iov->iov_len - i->iov_offset);
1118                 if (iov_iter_is_bvec(i))
1119                         return min(i->count, i->bvec->bv_len - i->iov_offset);
1120         }
1121         return i->count;
1122 }
1123 EXPORT_SYMBOL(iov_iter_single_seg_count);
1124
1125 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1126                         const struct kvec *kvec, unsigned long nr_segs,
1127                         size_t count)
1128 {
1129         WARN_ON(direction & ~(READ | WRITE));
1130         *i = (struct iov_iter){
1131                 .iter_type = ITER_KVEC,
1132                 .data_source = direction,
1133                 .kvec = kvec,
1134                 .nr_segs = nr_segs,
1135                 .iov_offset = 0,
1136                 .count = count
1137         };
1138 }
1139 EXPORT_SYMBOL(iov_iter_kvec);
1140
1141 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1142                         const struct bio_vec *bvec, unsigned long nr_segs,
1143                         size_t count)
1144 {
1145         WARN_ON(direction & ~(READ | WRITE));
1146         *i = (struct iov_iter){
1147                 .iter_type = ITER_BVEC,
1148                 .data_source = direction,
1149                 .bvec = bvec,
1150                 .nr_segs = nr_segs,
1151                 .iov_offset = 0,
1152                 .count = count
1153         };
1154 }
1155 EXPORT_SYMBOL(iov_iter_bvec);
1156
1157 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1158                         struct pipe_inode_info *pipe,
1159                         size_t count)
1160 {
1161         BUG_ON(direction != READ);
1162         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1163         *i = (struct iov_iter){
1164                 .iter_type = ITER_PIPE,
1165                 .data_source = false,
1166                 .pipe = pipe,
1167                 .head = pipe->head,
1168                 .start_head = pipe->head,
1169                 .iov_offset = 0,
1170                 .count = count
1171         };
1172 }
1173 EXPORT_SYMBOL(iov_iter_pipe);
1174
1175 /**
1176  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1177  * @i: The iterator to initialise.
1178  * @direction: The direction of the transfer.
1179  * @xarray: The xarray to access.
1180  * @start: The start file position.
1181  * @count: The size of the I/O buffer in bytes.
1182  *
1183  * Set up an I/O iterator to either draw data out of the pages attached to an
1184  * inode or to inject data into those pages.  The pages *must* be prevented
1185  * from evaporation, either by taking a ref on them or locking them by the
1186  * caller.
1187  */
1188 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1189                      struct xarray *xarray, loff_t start, size_t count)
1190 {
1191         BUG_ON(direction & ~1);
1192         *i = (struct iov_iter) {
1193                 .iter_type = ITER_XARRAY,
1194                 .data_source = direction,
1195                 .xarray = xarray,
1196                 .xarray_start = start,
1197                 .count = count,
1198                 .iov_offset = 0
1199         };
1200 }
1201 EXPORT_SYMBOL(iov_iter_xarray);
1202
1203 /**
1204  * iov_iter_discard - Initialise an I/O iterator that discards data
1205  * @i: The iterator to initialise.
1206  * @direction: The direction of the transfer.
1207  * @count: The size of the I/O buffer in bytes.
1208  *
1209  * Set up an I/O iterator that just discards everything that's written to it.
1210  * It's only available as a READ iterator.
1211  */
1212 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1213 {
1214         BUG_ON(direction != READ);
1215         *i = (struct iov_iter){
1216                 .iter_type = ITER_DISCARD,
1217                 .data_source = false,
1218                 .count = count,
1219                 .iov_offset = 0
1220         };
1221 }
1222 EXPORT_SYMBOL(iov_iter_discard);
1223
1224 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1225 {
1226         unsigned long res = 0;
1227         size_t size = i->count;
1228         size_t skip = i->iov_offset;
1229         unsigned k;
1230
1231         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1232                 size_t len = i->iov[k].iov_len - skip;
1233                 if (len) {
1234                         res |= (unsigned long)i->iov[k].iov_base + skip;
1235                         if (len > size)
1236                                 len = size;
1237                         res |= len;
1238                         size -= len;
1239                         if (!size)
1240                                 break;
1241                 }
1242         }
1243         return res;
1244 }
1245
1246 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1247 {
1248         unsigned res = 0;
1249         size_t size = i->count;
1250         unsigned skip = i->iov_offset;
1251         unsigned k;
1252
1253         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1254                 size_t len = i->bvec[k].bv_len - skip;
1255                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1256                 if (len > size)
1257                         len = size;
1258                 res |= len;
1259                 size -= len;
1260                 if (!size)
1261                         break;
1262         }
1263         return res;
1264 }
1265
1266 unsigned long iov_iter_alignment(const struct iov_iter *i)
1267 {
1268         /* iovec and kvec have identical layouts */
1269         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1270                 return iov_iter_alignment_iovec(i);
1271
1272         if (iov_iter_is_bvec(i))
1273                 return iov_iter_alignment_bvec(i);
1274
1275         if (iov_iter_is_pipe(i)) {
1276                 unsigned int p_mask = i->pipe->ring_size - 1;
1277                 size_t size = i->count;
1278
1279                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1280                         return size | i->iov_offset;
1281                 return size;
1282         }
1283
1284         if (iov_iter_is_xarray(i))
1285                 return (i->xarray_start + i->iov_offset) | i->count;
1286
1287         return 0;
1288 }
1289 EXPORT_SYMBOL(iov_iter_alignment);
1290
1291 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1292 {
1293         unsigned long res = 0;
1294         unsigned long v = 0;
1295         size_t size = i->count;
1296         unsigned k;
1297
1298         if (WARN_ON(!iter_is_iovec(i)))
1299                 return ~0U;
1300
1301         for (k = 0; k < i->nr_segs; k++) {
1302                 if (i->iov[k].iov_len) {
1303                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1304                         if (v) // if not the first one
1305                                 res |= base | v; // this start | previous end
1306                         v = base + i->iov[k].iov_len;
1307                         if (size <= i->iov[k].iov_len)
1308                                 break;
1309                         size -= i->iov[k].iov_len;
1310                 }
1311         }
1312         return res;
1313 }
1314 EXPORT_SYMBOL(iov_iter_gap_alignment);
1315
1316 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1317                                 size_t maxsize,
1318                                 struct page **pages,
1319                                 int iter_head,
1320                                 size_t *start)
1321 {
1322         struct pipe_inode_info *pipe = i->pipe;
1323         unsigned int p_mask = pipe->ring_size - 1;
1324         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1325         if (!n)
1326                 return -EFAULT;
1327
1328         maxsize = n;
1329         n += *start;
1330         while (n > 0) {
1331                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1332                 iter_head++;
1333                 n -= PAGE_SIZE;
1334         }
1335
1336         return maxsize;
1337 }
1338
1339 static ssize_t pipe_get_pages(struct iov_iter *i,
1340                    struct page **pages, size_t maxsize, unsigned maxpages,
1341                    size_t *start)
1342 {
1343         unsigned int iter_head, npages;
1344         size_t capacity;
1345
1346         if (!sanity(i))
1347                 return -EFAULT;
1348
1349         data_start(i, &iter_head, start);
1350         /* Amount of free space: some of this one + all after this one */
1351         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1352         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1353
1354         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1355 }
1356
1357 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1358                                           pgoff_t index, unsigned int nr_pages)
1359 {
1360         XA_STATE(xas, xa, index);
1361         struct page *page;
1362         unsigned int ret = 0;
1363
1364         rcu_read_lock();
1365         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1366                 if (xas_retry(&xas, page))
1367                         continue;
1368
1369                 /* Has the page moved or been split? */
1370                 if (unlikely(page != xas_reload(&xas))) {
1371                         xas_reset(&xas);
1372                         continue;
1373                 }
1374
1375                 pages[ret] = find_subpage(page, xas.xa_index);
1376                 get_page(pages[ret]);
1377                 if (++ret == nr_pages)
1378                         break;
1379         }
1380         rcu_read_unlock();
1381         return ret;
1382 }
1383
1384 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1385                                      struct page **pages, size_t maxsize,
1386                                      unsigned maxpages, size_t *_start_offset)
1387 {
1388         unsigned nr, offset;
1389         pgoff_t index, count;
1390         size_t size = maxsize, actual;
1391         loff_t pos;
1392
1393         if (!size || !maxpages)
1394                 return 0;
1395
1396         pos = i->xarray_start + i->iov_offset;
1397         index = pos >> PAGE_SHIFT;
1398         offset = pos & ~PAGE_MASK;
1399         *_start_offset = offset;
1400
1401         count = 1;
1402         if (size > PAGE_SIZE - offset) {
1403                 size -= PAGE_SIZE - offset;
1404                 count += size >> PAGE_SHIFT;
1405                 size &= ~PAGE_MASK;
1406                 if (size)
1407                         count++;
1408         }
1409
1410         if (count > maxpages)
1411                 count = maxpages;
1412
1413         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1414         if (nr == 0)
1415                 return 0;
1416
1417         actual = PAGE_SIZE * nr;
1418         actual -= offset;
1419         if (nr == count && size > 0) {
1420                 unsigned last_offset = (nr > 1) ? 0 : offset;
1421                 actual -= PAGE_SIZE - (last_offset + size);
1422         }
1423         return actual;
1424 }
1425
1426 /* must be done on non-empty ITER_IOVEC one */
1427 static unsigned long first_iovec_segment(const struct iov_iter *i,
1428                                          size_t *size, size_t *start,
1429                                          size_t maxsize, unsigned maxpages)
1430 {
1431         size_t skip;
1432         long k;
1433
1434         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1435                 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1436                 size_t len = i->iov[k].iov_len - skip;
1437
1438                 if (unlikely(!len))
1439                         continue;
1440                 if (len > maxsize)
1441                         len = maxsize;
1442                 len += (*start = addr % PAGE_SIZE);
1443                 if (len > maxpages * PAGE_SIZE)
1444                         len = maxpages * PAGE_SIZE;
1445                 *size = len;
1446                 return addr & PAGE_MASK;
1447         }
1448         BUG(); // if it had been empty, we wouldn't get called
1449 }
1450
1451 /* must be done on non-empty ITER_BVEC one */
1452 static struct page *first_bvec_segment(const struct iov_iter *i,
1453                                        size_t *size, size_t *start,
1454                                        size_t maxsize, unsigned maxpages)
1455 {
1456         struct page *page;
1457         size_t skip = i->iov_offset, len;
1458
1459         len = i->bvec->bv_len - skip;
1460         if (len > maxsize)
1461                 len = maxsize;
1462         skip += i->bvec->bv_offset;
1463         page = i->bvec->bv_page + skip / PAGE_SIZE;
1464         len += (*start = skip % PAGE_SIZE);
1465         if (len > maxpages * PAGE_SIZE)
1466                 len = maxpages * PAGE_SIZE;
1467         *size = len;
1468         return page;
1469 }
1470
1471 ssize_t iov_iter_get_pages(struct iov_iter *i,
1472                    struct page **pages, size_t maxsize, unsigned maxpages,
1473                    size_t *start)
1474 {
1475         size_t len;
1476         int n, res;
1477
1478         if (maxsize > i->count)
1479                 maxsize = i->count;
1480         if (!maxsize)
1481                 return 0;
1482
1483         if (likely(iter_is_iovec(i))) {
1484                 unsigned long addr;
1485
1486                 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1487                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1488                 res = get_user_pages_fast(addr, n,
1489                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0,
1490                                 pages);
1491                 if (unlikely(res < 0))
1492                         return res;
1493                 return (res == n ? len : res * PAGE_SIZE) - *start;
1494         }
1495         if (iov_iter_is_bvec(i)) {
1496                 struct page *page;
1497
1498                 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1499                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1500                 while (n--)
1501                         get_page(*pages++ = page++);
1502                 return len - *start;
1503         }
1504         if (iov_iter_is_pipe(i))
1505                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1506         if (iov_iter_is_xarray(i))
1507                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1508         return -EFAULT;
1509 }
1510 EXPORT_SYMBOL(iov_iter_get_pages);
1511
1512 static struct page **get_pages_array(size_t n)
1513 {
1514         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1515 }
1516
1517 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1518                    struct page ***pages, size_t maxsize,
1519                    size_t *start)
1520 {
1521         struct page **p;
1522         unsigned int iter_head, npages;
1523         ssize_t n;
1524
1525         if (!sanity(i))
1526                 return -EFAULT;
1527
1528         data_start(i, &iter_head, start);
1529         /* Amount of free space: some of this one + all after this one */
1530         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1531         n = npages * PAGE_SIZE - *start;
1532         if (maxsize > n)
1533                 maxsize = n;
1534         else
1535                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1536         p = get_pages_array(npages);
1537         if (!p)
1538                 return -ENOMEM;
1539         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1540         if (n > 0)
1541                 *pages = p;
1542         else
1543                 kvfree(p);
1544         return n;
1545 }
1546
1547 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1548                                            struct page ***pages, size_t maxsize,
1549                                            size_t *_start_offset)
1550 {
1551         struct page **p;
1552         unsigned nr, offset;
1553         pgoff_t index, count;
1554         size_t size = maxsize, actual;
1555         loff_t pos;
1556
1557         if (!size)
1558                 return 0;
1559
1560         pos = i->xarray_start + i->iov_offset;
1561         index = pos >> PAGE_SHIFT;
1562         offset = pos & ~PAGE_MASK;
1563         *_start_offset = offset;
1564
1565         count = 1;
1566         if (size > PAGE_SIZE - offset) {
1567                 size -= PAGE_SIZE - offset;
1568                 count += size >> PAGE_SHIFT;
1569                 size &= ~PAGE_MASK;
1570                 if (size)
1571                         count++;
1572         }
1573
1574         p = get_pages_array(count);
1575         if (!p)
1576                 return -ENOMEM;
1577         *pages = p;
1578
1579         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1580         if (nr == 0)
1581                 return 0;
1582
1583         actual = PAGE_SIZE * nr;
1584         actual -= offset;
1585         if (nr == count && size > 0) {
1586                 unsigned last_offset = (nr > 1) ? 0 : offset;
1587                 actual -= PAGE_SIZE - (last_offset + size);
1588         }
1589         return actual;
1590 }
1591
1592 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1593                    struct page ***pages, size_t maxsize,
1594                    size_t *start)
1595 {
1596         struct page **p;
1597         size_t len;
1598         int n, res;
1599
1600         if (maxsize > i->count)
1601                 maxsize = i->count;
1602         if (!maxsize)
1603                 return 0;
1604
1605         if (likely(iter_is_iovec(i))) {
1606                 unsigned long addr;
1607
1608                 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1609                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1610                 p = get_pages_array(n);
1611                 if (!p)
1612                         return -ENOMEM;
1613                 res = get_user_pages_fast(addr, n,
1614                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0, p);
1615                 if (unlikely(res < 0)) {
1616                         kvfree(p);
1617                         return res;
1618                 }
1619                 *pages = p;
1620                 return (res == n ? len : res * PAGE_SIZE) - *start;
1621         }
1622         if (iov_iter_is_bvec(i)) {
1623                 struct page *page;
1624
1625                 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1626                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1627                 *pages = p = get_pages_array(n);
1628                 if (!p)
1629                         return -ENOMEM;
1630                 while (n--)
1631                         get_page(*p++ = page++);
1632                 return len - *start;
1633         }
1634         if (iov_iter_is_pipe(i))
1635                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1636         if (iov_iter_is_xarray(i))
1637                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1638         return -EFAULT;
1639 }
1640 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1641
1642 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1643                                struct iov_iter *i)
1644 {
1645         __wsum sum, next;
1646         sum = *csum;
1647         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1648                 WARN_ON(1);
1649                 return 0;
1650         }
1651         iterate_and_advance(i, bytes, base, len, off, ({
1652                 next = csum_and_copy_from_user(base, addr + off, len);
1653                 sum = csum_block_add(sum, next, off);
1654                 next ? 0 : len;
1655         }), ({
1656                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1657         })
1658         )
1659         *csum = sum;
1660         return bytes;
1661 }
1662 EXPORT_SYMBOL(csum_and_copy_from_iter);
1663
1664 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1665                              struct iov_iter *i)
1666 {
1667         struct csum_state *csstate = _csstate;
1668         __wsum sum, next;
1669
1670         if (unlikely(iov_iter_is_discard(i))) {
1671                 WARN_ON(1);     /* for now */
1672                 return 0;
1673         }
1674
1675         sum = csum_shift(csstate->csum, csstate->off);
1676         if (unlikely(iov_iter_is_pipe(i)))
1677                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1678         else iterate_and_advance(i, bytes, base, len, off, ({
1679                 next = csum_and_copy_to_user(addr + off, base, len);
1680                 sum = csum_block_add(sum, next, off);
1681                 next ? 0 : len;
1682         }), ({
1683                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1684         })
1685         )
1686         csstate->csum = csum_shift(sum, csstate->off);
1687         csstate->off += bytes;
1688         return bytes;
1689 }
1690 EXPORT_SYMBOL(csum_and_copy_to_iter);
1691
1692 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1693                 struct iov_iter *i)
1694 {
1695 #ifdef CONFIG_CRYPTO_HASH
1696         struct ahash_request *hash = hashp;
1697         struct scatterlist sg;
1698         size_t copied;
1699
1700         copied = copy_to_iter(addr, bytes, i);
1701         sg_init_one(&sg, addr, copied);
1702         ahash_request_set_crypt(hash, &sg, NULL, copied);
1703         crypto_ahash_update(hash);
1704         return copied;
1705 #else
1706         return 0;
1707 #endif
1708 }
1709 EXPORT_SYMBOL(hash_and_copy_to_iter);
1710
1711 static int iov_npages(const struct iov_iter *i, int maxpages)
1712 {
1713         size_t skip = i->iov_offset, size = i->count;
1714         const struct iovec *p;
1715         int npages = 0;
1716
1717         for (p = i->iov; size; skip = 0, p++) {
1718                 unsigned offs = offset_in_page(p->iov_base + skip);
1719                 size_t len = min(p->iov_len - skip, size);
1720
1721                 if (len) {
1722                         size -= len;
1723                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1724                         if (unlikely(npages > maxpages))
1725                                 return maxpages;
1726                 }
1727         }
1728         return npages;
1729 }
1730
1731 static int bvec_npages(const struct iov_iter *i, int maxpages)
1732 {
1733         size_t skip = i->iov_offset, size = i->count;
1734         const struct bio_vec *p;
1735         int npages = 0;
1736
1737         for (p = i->bvec; size; skip = 0, p++) {
1738                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1739                 size_t len = min(p->bv_len - skip, size);
1740
1741                 size -= len;
1742                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1743                 if (unlikely(npages > maxpages))
1744                         return maxpages;
1745         }
1746         return npages;
1747 }
1748
1749 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1750 {
1751         if (unlikely(!i->count))
1752                 return 0;
1753         /* iovec and kvec have identical layouts */
1754         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1755                 return iov_npages(i, maxpages);
1756         if (iov_iter_is_bvec(i))
1757                 return bvec_npages(i, maxpages);
1758         if (iov_iter_is_pipe(i)) {
1759                 unsigned int iter_head;
1760                 int npages;
1761                 size_t off;
1762
1763                 if (!sanity(i))
1764                         return 0;
1765
1766                 data_start(i, &iter_head, &off);
1767                 /* some of this one + all after this one */
1768                 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1769                 return min(npages, maxpages);
1770         }
1771         if (iov_iter_is_xarray(i)) {
1772                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1773                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1774                 return min(npages, maxpages);
1775         }
1776         return 0;
1777 }
1778 EXPORT_SYMBOL(iov_iter_npages);
1779
1780 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1781 {
1782         *new = *old;
1783         if (unlikely(iov_iter_is_pipe(new))) {
1784                 WARN_ON(1);
1785                 return NULL;
1786         }
1787         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1788                 return NULL;
1789         if (iov_iter_is_bvec(new))
1790                 return new->bvec = kmemdup(new->bvec,
1791                                     new->nr_segs * sizeof(struct bio_vec),
1792                                     flags);
1793         else
1794                 /* iovec and kvec have identical layout */
1795                 return new->iov = kmemdup(new->iov,
1796                                    new->nr_segs * sizeof(struct iovec),
1797                                    flags);
1798 }
1799 EXPORT_SYMBOL(dup_iter);
1800
1801 static int copy_compat_iovec_from_user(struct iovec *iov,
1802                 const struct iovec __user *uvec, unsigned long nr_segs)
1803 {
1804         const struct compat_iovec __user *uiov =
1805                 (const struct compat_iovec __user *)uvec;
1806         int ret = -EFAULT, i;
1807
1808         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1809                 return -EFAULT;
1810
1811         for (i = 0; i < nr_segs; i++) {
1812                 compat_uptr_t buf;
1813                 compat_ssize_t len;
1814
1815                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1816                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1817
1818                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1819                 if (len < 0) {
1820                         ret = -EINVAL;
1821                         goto uaccess_end;
1822                 }
1823                 iov[i].iov_base = compat_ptr(buf);
1824                 iov[i].iov_len = len;
1825         }
1826
1827         ret = 0;
1828 uaccess_end:
1829         user_access_end();
1830         return ret;
1831 }
1832
1833 static int copy_iovec_from_user(struct iovec *iov,
1834                 const struct iovec __user *uvec, unsigned long nr_segs)
1835 {
1836         unsigned long seg;
1837
1838         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1839                 return -EFAULT;
1840         for (seg = 0; seg < nr_segs; seg++) {
1841                 if ((ssize_t)iov[seg].iov_len < 0)
1842                         return -EINVAL;
1843         }
1844
1845         return 0;
1846 }
1847
1848 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1849                 unsigned long nr_segs, unsigned long fast_segs,
1850                 struct iovec *fast_iov, bool compat)
1851 {
1852         struct iovec *iov = fast_iov;
1853         int ret;
1854
1855         /*
1856          * SuS says "The readv() function *may* fail if the iovcnt argument was
1857          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1858          * traditionally returned zero for zero segments, so...
1859          */
1860         if (nr_segs == 0)
1861                 return iov;
1862         if (nr_segs > UIO_MAXIOV)
1863                 return ERR_PTR(-EINVAL);
1864         if (nr_segs > fast_segs) {
1865                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1866                 if (!iov)
1867                         return ERR_PTR(-ENOMEM);
1868         }
1869
1870         if (compat)
1871                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1872         else
1873                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1874         if (ret) {
1875                 if (iov != fast_iov)
1876                         kfree(iov);
1877                 return ERR_PTR(ret);
1878         }
1879
1880         return iov;
1881 }
1882
1883 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1884                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1885                  struct iov_iter *i, bool compat)
1886 {
1887         ssize_t total_len = 0;
1888         unsigned long seg;
1889         struct iovec *iov;
1890
1891         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1892         if (IS_ERR(iov)) {
1893                 *iovp = NULL;
1894                 return PTR_ERR(iov);
1895         }
1896
1897         /*
1898          * According to the Single Unix Specification we should return EINVAL if
1899          * an element length is < 0 when cast to ssize_t or if the total length
1900          * would overflow the ssize_t return value of the system call.
1901          *
1902          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1903          * overflow case.
1904          */
1905         for (seg = 0; seg < nr_segs; seg++) {
1906                 ssize_t len = (ssize_t)iov[seg].iov_len;
1907
1908                 if (!access_ok(iov[seg].iov_base, len)) {
1909                         if (iov != *iovp)
1910                                 kfree(iov);
1911                         *iovp = NULL;
1912                         return -EFAULT;
1913                 }
1914
1915                 if (len > MAX_RW_COUNT - total_len) {
1916                         len = MAX_RW_COUNT - total_len;
1917                         iov[seg].iov_len = len;
1918                 }
1919                 total_len += len;
1920         }
1921
1922         iov_iter_init(i, type, iov, nr_segs, total_len);
1923         if (iov == *iovp)
1924                 *iovp = NULL;
1925         else
1926                 *iovp = iov;
1927         return total_len;
1928 }
1929
1930 /**
1931  * import_iovec() - Copy an array of &struct iovec from userspace
1932  *     into the kernel, check that it is valid, and initialize a new
1933  *     &struct iov_iter iterator to access it.
1934  *
1935  * @type: One of %READ or %WRITE.
1936  * @uvec: Pointer to the userspace array.
1937  * @nr_segs: Number of elements in userspace array.
1938  * @fast_segs: Number of elements in @iov.
1939  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1940  *     on-stack) kernel array.
1941  * @i: Pointer to iterator that will be initialized on success.
1942  *
1943  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1944  * then this function places %NULL in *@iov on return. Otherwise, a new
1945  * array will be allocated and the result placed in *@iov. This means that
1946  * the caller may call kfree() on *@iov regardless of whether the small
1947  * on-stack array was used or not (and regardless of whether this function
1948  * returns an error or not).
1949  *
1950  * Return: Negative error code on error, bytes imported on success
1951  */
1952 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1953                  unsigned nr_segs, unsigned fast_segs,
1954                  struct iovec **iovp, struct iov_iter *i)
1955 {
1956         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1957                               in_compat_syscall());
1958 }
1959 EXPORT_SYMBOL(import_iovec);
1960
1961 int import_single_range(int rw, void __user *buf, size_t len,
1962                  struct iovec *iov, struct iov_iter *i)
1963 {
1964         if (len > MAX_RW_COUNT)
1965                 len = MAX_RW_COUNT;
1966         if (unlikely(!access_ok(buf, len)))
1967                 return -EFAULT;
1968
1969         iov->iov_base = buf;
1970         iov->iov_len = len;
1971         iov_iter_init(i, rw, iov, 1, len);
1972         return 0;
1973 }
1974 EXPORT_SYMBOL(import_single_range);