lib/iov_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <crypto/hash.h>
   3 #include <linux/export.h>
   4 #include <linux/bvec.h>
   5 #include <linux/fault-inject-usercopy.h>
   6 #include <linux/uio.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/highmem.h>
   9 #include <linux/slab.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/splice.h>
  12 #include <linux/compat.h>
  13 #include <net/checksum.h>
  14 #include <linux/scatterlist.h>
  15 #include <linux/instrumented.h>
  16
  17 #define PIPE_PARANOIA /* for now */
  18
  19 #define iterate_iovec(i, n, __v, __p, skip, STEP) {     \
  20         size_t left;                                    \
  21         size_t wanted = n;                              \
  22         __p = i->iov;                                   \
  23         __v.iov_len = min(n, __p->iov_len - skip);      \
  24         if (likely(__v.iov_len)) {                      \
  25                 __v.iov_base = __p->iov_base + skip;    \
  26                 left = (STEP);                          \
  27                 __v.iov_len -= left;                    \
  28                 skip += __v.iov_len;                    \
  29                 n -= __v.iov_len;                       \
  30         } else {                                        \
  31                 left = 0;                               \
  32         }                                               \
  33         while (unlikely(!left && n)) {                  \
  34                 __p++;                                  \
  35                 __v.iov_len = min(n, __p->iov_len);     \
  36                 if (unlikely(!__v.iov_len))             \
  37                         continue;                       \
  38                 __v.iov_base = __p->iov_base;           \
  39                 left = (STEP);                          \
  40                 __v.iov_len -= left;                    \
  41                 skip = __v.iov_len;                     \
  42                 n -= __v.iov_len;                       \
  43         }                                               \
  44         n = wanted - n;                                 \
  45 }
  46
  47 #define iterate_kvec(i, n, __v, __p, skip, STEP) {      \
  48         size_t wanted = n;                              \
  49         __p = i->kvec;                                  \
  50         __v.iov_len = min(n, __p->iov_len - skip);      \
  51         if (likely(__v.iov_len)) {                      \
  52                 __v.iov_base = __p->iov_base + skip;    \
  53                 (void)(STEP);                           \
  54                 skip += __v.iov_len;                    \
  55                 n -= __v.iov_len;                       \
  56         }                                               \
  57         while (unlikely(n)) {                           \
  58                 __p++;                                  \
  59                 __v.iov_len = min(n, __p->iov_len);     \
  60                 if (unlikely(!__v.iov_len))             \
  61                         continue;                       \
  62                 __v.iov_base = __p->iov_base;           \
  63                 (void)(STEP);                           \
  64                 skip = __v.iov_len;                     \
  65                 n -= __v.iov_len;                       \
  66         }                                               \
  67         n = wanted;                                     \
  68 }
  69
  70 #define iterate_bvec(i, n, __v, __bi, skip, STEP) {     \
  71         struct bvec_iter __start;                       \
  72         __start.bi_size = n;                            \
  73         __start.bi_bvec_done = skip;                    \
  74         __start.bi_idx = 0;                             \
  75         for_each_bvec(__v, i->bvec, __bi, __start) {    \
  76                 (void)(STEP);                           \
  77         }                                               \
  78 }
  79
  80 #define iterate_xarray(i, n, __v, skip, STEP) {         \
  81         struct page *head = NULL;                               \
  82         size_t wanted = n, seg, offset;                         \
  83         loff_t start = i->xarray_start + skip;                  \
  84         pgoff_t index = start >> PAGE_SHIFT;                    \
  85         int j;                                                  \
  86                                                                 \
  87         XA_STATE(xas, i->xarray, index);                        \
  88                                                                 \
  89         rcu_read_lock();                                                \
  90         xas_for_each(&xas, head, ULONG_MAX) {                           \
  91                 if (xas_retry(&xas, head))                              \
  92                         continue;                                       \
  93                 if (WARN_ON(xa_is_value(head)))                         \
  94                         break;                                          \
  95                 if (WARN_ON(PageHuge(head)))                            \
  96                         break;                                          \
  97                 for (j = (head->index < index) ? index - head->index : 0; \
  98                      j < thp_nr_pages(head); j++) {                     \
  99                         __v.bv_page = head + j;                         \
 100                         offset = (i->xarray_start + skip) & ~PAGE_MASK; \
 101                         seg = PAGE_SIZE - offset;                       \
 102                         __v.bv_offset = offset;                         \
 103                         __v.bv_len = min(n, seg);                       \
 104                         (void)(STEP);                                   \
 105                         n -= __v.bv_len;                                \
 106                         skip += __v.bv_len;                             \
 107                         if (n == 0)                                     \
 108                                 break;                                  \
 109                 }                                                       \
 110                 if (n == 0)                                             \
 111                         break;                                          \
 112         }                                                       \
 113         rcu_read_unlock();                                      \
 114         n = wanted - n;                                         \
 115 }
 116
 117 #define iterate_and_advance(i, n, v, I, B, K, X) {              \
 118         if (unlikely(i->count < n))                             \
 119                 n = i->count;                                   \
 120         if (i->count) {                                         \
 121                 size_t skip = i->iov_offset;                    \
 122                 if (likely(iter_is_iovec(i))) {                 \
 123                         const struct iovec *iov;                \
 124                         struct iovec v;                         \
 125                         iterate_iovec(i, n, v, iov, skip, (I))  \
 126                         if (skip == iov->iov_len) {             \
 127                                 iov++;                          \
 128                                 skip = 0;                       \
 129                         }                                       \
 130                         i->nr_segs -= iov - i->iov;             \
 131                         i->iov = iov;                           \
 132                 } else if (iov_iter_is_bvec(i)) {               \
 133                         const struct bio_vec *bvec = i->bvec;   \
 134                         struct bio_vec v;                       \
 135                         struct bvec_iter __bi;                  \
 136                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 137                         i->bvec = __bvec_iter_bvec(i->bvec, __bi);      \
 138                         i->nr_segs -= i->bvec - bvec;           \
 139                         skip = __bi.bi_bvec_done;               \
 140                 } else if (iov_iter_is_kvec(i)) {               \
 141                         const struct kvec *kvec;                \
 142                         struct kvec v;                          \
 143                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 144                         if (skip == kvec->iov_len) {            \
 145                                 kvec++;                         \
 146                                 skip = 0;                       \
 147                         }                                       \
 148                         i->nr_segs -= kvec - i->kvec;           \
 149                         i->kvec = kvec;                         \
 150                 } else if (iov_iter_is_xarray(i)) {             \
 151                         struct bio_vec v;                       \
 152                         iterate_xarray(i, n, v, skip, (X))      \
 153                 }                                               \
 154                 i->count -= n;                                  \
 155                 i->iov_offset = skip;                           \
 156         }                                                       \
 157 }
 158
 159 static int copyout(void __user *to, const void *from, size_t n)
 160 {
 161         if (should_fail_usercopy())
 162                 return n;
 163         if (access_ok(to, n)) {
 164                 instrument_copy_to_user(to, from, n);
 165                 n = raw_copy_to_user(to, from, n);
 166         }
 167         return n;
 168 }
 169
 170 static int copyin(void *to, const void __user *from, size_t n)
 171 {
 172         if (should_fail_usercopy())
 173                 return n;
 174         if (access_ok(from, n)) {
 175                 instrument_copy_from_user(to, from, n);
 176                 n = raw_copy_from_user(to, from, n);
 177         }
 178         return n;
 179 }
 180
 181 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 182                          struct iov_iter *i)
 183 {
 184         size_t skip, copy, left, wanted;
 185         const struct iovec *iov;
 186         char __user *buf;
 187         void *kaddr, *from;
 188
 189         if (unlikely(bytes > i->count))
 190                 bytes = i->count;
 191
 192         if (unlikely(!bytes))
 193                 return 0;
 194
 195         might_fault();
 196         wanted = bytes;
 197         iov = i->iov;
 198         skip = i->iov_offset;
 199         buf = iov->iov_base + skip;
 200         copy = min(bytes, iov->iov_len - skip);
 201
 202         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 203                 kaddr = kmap_atomic(page);
 204                 from = kaddr + offset;
 205
 206                 /* first chunk, usually the only one */
 207                 left = copyout(buf, from, copy);
 208                 copy -= left;
 209                 skip += copy;
 210                 from += copy;
 211                 bytes -= copy;
 212
 213                 while (unlikely(!left && bytes)) {
 214                         iov++;
 215                         buf = iov->iov_base;
 216                         copy = min(bytes, iov->iov_len);
 217                         left = copyout(buf, from, copy);
 218                         copy -= left;
 219                         skip = copy;
 220                         from += copy;
 221                         bytes -= copy;
 222                 }
 223                 if (likely(!bytes)) {
 224                         kunmap_atomic(kaddr);
 225                         goto done;
 226                 }
 227                 offset = from - kaddr;
 228                 buf += copy;
 229                 kunmap_atomic(kaddr);
 230                 copy = min(bytes, iov->iov_len - skip);
 231         }
 232         /* Too bad - revert to non-atomic kmap */
 233
 234         kaddr = kmap(page);
 235         from = kaddr + offset;
 236         left = copyout(buf, from, copy);
 237         copy -= left;
 238         skip += copy;
 239         from += copy;
 240         bytes -= copy;
 241         while (unlikely(!left && bytes)) {
 242                 iov++;
 243                 buf = iov->iov_base;
 244                 copy = min(bytes, iov->iov_len);
 245                 left = copyout(buf, from, copy);
 246                 copy -= left;
 247                 skip = copy;
 248                 from += copy;
 249                 bytes -= copy;
 250         }
 251         kunmap(page);
 252
 253 done:
 254         if (skip == iov->iov_len) {
 255                 iov++;
 256                 skip = 0;
 257         }
 258         i->count -= wanted - bytes;
 259         i->nr_segs -= iov - i->iov;
 260         i->iov = iov;
 261         i->iov_offset = skip;
 262         return wanted - bytes;
 263 }
 264
 265 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 266                          struct iov_iter *i)
 267 {
 268         size_t skip, copy, left, wanted;
 269         const struct iovec *iov;
 270         char __user *buf;
 271         void *kaddr, *to;
 272
 273         if (unlikely(bytes > i->count))
 274                 bytes = i->count;
 275
 276         if (unlikely(!bytes))
 277                 return 0;
 278
 279         might_fault();
 280         wanted = bytes;
 281         iov = i->iov;
 282         skip = i->iov_offset;
 283         buf = iov->iov_base + skip;
 284         copy = min(bytes, iov->iov_len - skip);
 285
 286         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 287                 kaddr = kmap_atomic(page);
 288                 to = kaddr + offset;
 289
 290                 /* first chunk, usually the only one */
 291                 left = copyin(to, buf, copy);
 292                 copy -= left;
 293                 skip += copy;
 294                 to += copy;
 295                 bytes -= copy;
 296
 297                 while (unlikely(!left && bytes)) {
 298                         iov++;
 299                         buf = iov->iov_base;
 300                         copy = min(bytes, iov->iov_len);
 301                         left = copyin(to, buf, copy);
 302                         copy -= left;
 303                         skip = copy;
 304                         to += copy;
 305                         bytes -= copy;
 306                 }
 307                 if (likely(!bytes)) {
 308                         kunmap_atomic(kaddr);
 309                         goto done;
 310                 }
 311                 offset = to - kaddr;
 312                 buf += copy;
 313                 kunmap_atomic(kaddr);
 314                 copy = min(bytes, iov->iov_len - skip);
 315         }
 316         /* Too bad - revert to non-atomic kmap */
 317
 318         kaddr = kmap(page);
 319         to = kaddr + offset;
 320         left = copyin(to, buf, copy);
 321         copy -= left;
 322         skip += copy;
 323         to += copy;
 324         bytes -= copy;
 325         while (unlikely(!left && bytes)) {
 326                 iov++;
 327                 buf = iov->iov_base;
 328                 copy = min(bytes, iov->iov_len);
 329                 left = copyin(to, buf, copy);
 330                 copy -= left;
 331                 skip = copy;
 332                 to += copy;
 333                 bytes -= copy;
 334         }
 335         kunmap(page);
 336
 337 done:
 338         if (skip == iov->iov_len) {
 339                 iov++;
 340                 skip = 0;
 341         }
 342         i->count -= wanted - bytes;
 343         i->nr_segs -= iov - i->iov;
 344         i->iov = iov;
 345         i->iov_offset = skip;
 346         return wanted - bytes;
 347 }
 348
 349 #ifdef PIPE_PARANOIA
 350 static bool sanity(const struct iov_iter *i)
 351 {
 352         struct pipe_inode_info *pipe = i->pipe;
 353         unsigned int p_head = pipe->head;
 354         unsigned int p_tail = pipe->tail;
 355         unsigned int p_mask = pipe->ring_size - 1;
 356         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 357         unsigned int i_head = i->head;
 358         unsigned int idx;
 359
 360         if (i->iov_offset) {
 361                 struct pipe_buffer *p;
 362                 if (unlikely(p_occupancy == 0))
 363                         goto Bad;       // pipe must be non-empty
 364                 if (unlikely(i_head != p_head - 1))
 365                         goto Bad;       // must be at the last buffer...
 366
 367                 p = &pipe->bufs[i_head & p_mask];
 368                 if (unlikely(p->offset + p->len != i->iov_offset))
 369                         goto Bad;       // ... at the end of segment
 370         } else {
 371                 if (i_head != p_head)
 372                         goto Bad;       // must be right after the last buffer
 373         }
 374         return true;
 375 Bad:
 376         printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
 377         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 378                         p_head, p_tail, pipe->ring_size);
 379         for (idx = 0; idx < pipe->ring_size; idx++)
 380                 printk(KERN_ERR "[%p %p %d %d]\n",
 381                         pipe->bufs[idx].ops,
 382                         pipe->bufs[idx].page,
 383                         pipe->bufs[idx].offset,
 384                         pipe->bufs[idx].len);
 385         WARN_ON(1);
 386         return false;
 387 }
 388 #else
 389 #define sanity(i) true
 390 #endif
 391
 392 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 393                          struct iov_iter *i)
 394 {
 395         struct pipe_inode_info *pipe = i->pipe;
 396         struct pipe_buffer *buf;
 397         unsigned int p_tail = pipe->tail;
 398         unsigned int p_mask = pipe->ring_size - 1;
 399         unsigned int i_head = i->head;
 400         size_t off;
 401
 402         if (unlikely(bytes > i->count))
 403                 bytes = i->count;
 404
 405         if (unlikely(!bytes))
 406                 return 0;
 407
 408         if (!sanity(i))
 409                 return 0;
 410
 411         off = i->iov_offset;
 412         buf = &pipe->bufs[i_head & p_mask];
 413         if (off) {
 414                 if (offset == off && buf->page == page) {
 415                         /* merge with the last one */
 416                         buf->len += bytes;
 417                         i->iov_offset += bytes;
 418                         goto out;
 419                 }
 420                 i_head++;
 421                 buf = &pipe->bufs[i_head & p_mask];
 422         }
 423         if (pipe_full(i_head, p_tail, pipe->max_usage))
 424                 return 0;
 425
 426         buf->ops = &page_cache_pipe_buf_ops;
 427         get_page(page);
 428         buf->page = page;
 429         buf->offset = offset;
 430         buf->len = bytes;
 431
 432         pipe->head = i_head + 1;
 433         i->iov_offset = offset + bytes;
 434         i->head = i_head;
 435 out:
 436         i->count -= bytes;
 437         return bytes;
 438 }
 439
 440 /*
 441  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 442  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 443  *
 444  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 445  * because it is an invalid address).
 446  */
 447 int iov_iter_fault_in_readable(const struct iov_iter *i, size_t bytes)
 448 {
 449         if (iter_is_iovec(i)) {
 450                 const struct iovec *p;
 451                 size_t skip;
 452
 453                 if (bytes > i->count)
 454                         bytes = i->count;
 455                 for (p = i->iov, skip = i->iov_offset; bytes; p++, skip = 0) {
 456                         size_t len = min(bytes, p->iov_len - skip);
 457                         int err;
 458
 459                         if (unlikely(!len))
 460                                 continue;
 461                         err = fault_in_pages_readable(p->iov_base + skip, len);
 462                         if (unlikely(err))
 463                                 return err;
 464                         bytes -= len;
 465                 }
 466         }
 467         return 0;
 468 }
 469 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 470
 471 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 472                         const struct iovec *iov, unsigned long nr_segs,
 473                         size_t count)
 474 {
 475         WARN_ON(direction & ~(READ | WRITE));
 476         WARN_ON_ONCE(uaccess_kernel());
 477         *i = (struct iov_iter) {
 478                 .iter_type = ITER_IOVEC,
 479                 .data_source = direction,
 480                 .iov = iov,
 481                 .nr_segs = nr_segs,
 482                 .iov_offset = 0,
 483                 .count = count
 484         };
 485 }
 486 EXPORT_SYMBOL(iov_iter_init);
 487
 488 static inline bool allocated(struct pipe_buffer *buf)
 489 {
 490         return buf->ops == &default_pipe_buf_ops;
 491 }
 492
 493 static inline void data_start(const struct iov_iter *i,
 494                               unsigned int *iter_headp, size_t *offp)
 495 {
 496         unsigned int p_mask = i->pipe->ring_size - 1;
 497         unsigned int iter_head = i->head;
 498         size_t off = i->iov_offset;
 499
 500         if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
 501                     off == PAGE_SIZE)) {
 502                 iter_head++;
 503                 off = 0;
 504         }
 505         *iter_headp = iter_head;
 506         *offp = off;
 507 }
 508
 509 static size_t push_pipe(struct iov_iter *i, size_t size,
 510                         int *iter_headp, size_t *offp)
 511 {
 512         struct pipe_inode_info *pipe = i->pipe;
 513         unsigned int p_tail = pipe->tail;
 514         unsigned int p_mask = pipe->ring_size - 1;
 515         unsigned int iter_head;
 516         size_t off;
 517         ssize_t left;
 518
 519         if (unlikely(size > i->count))
 520                 size = i->count;
 521         if (unlikely(!size))
 522                 return 0;
 523
 524         left = size;
 525         data_start(i, &iter_head, &off);
 526         *iter_headp = iter_head;
 527         *offp = off;
 528         if (off) {
 529                 left -= PAGE_SIZE - off;
 530                 if (left <= 0) {
 531                         pipe->bufs[iter_head & p_mask].len += size;
 532                         return size;
 533                 }
 534                 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
 535                 iter_head++;
 536         }
 537         while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
 538                 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
 539                 struct page *page = alloc_page(GFP_USER);
 540                 if (!page)
 541                         break;
 542
 543                 buf->ops = &default_pipe_buf_ops;
 544                 buf->page = page;
 545                 buf->offset = 0;
 546                 buf->len = min_t(ssize_t, left, PAGE_SIZE);
 547                 left -= buf->len;
 548                 iter_head++;
 549                 pipe->head = iter_head;
 550
 551                 if (left == 0)
 552                         return size;
 553         }
 554         return size - left;
 555 }
 556
 557 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 558                                 struct iov_iter *i)
 559 {
 560         struct pipe_inode_info *pipe = i->pipe;
 561         unsigned int p_mask = pipe->ring_size - 1;
 562         unsigned int i_head;
 563         size_t n, off;
 564
 565         if (!sanity(i))
 566                 return 0;
 567
 568         bytes = n = push_pipe(i, bytes, &i_head, &off);
 569         if (unlikely(!n))
 570                 return 0;
 571         do {
 572                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 573                 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
 574                 i->head = i_head;
 575                 i->iov_offset = off + chunk;
 576                 n -= chunk;
 577                 addr += chunk;
 578                 off = 0;
 579                 i_head++;
 580         } while (n);
 581         i->count -= bytes;
 582         return bytes;
 583 }
 584
 585 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 586                               __wsum sum, size_t off)
 587 {
 588         __wsum next = csum_partial_copy_nocheck(from, to, len);
 589         return csum_block_add(sum, next, off);
 590 }
 591
 592 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 593                                          struct csum_state *csstate,
 594                                          struct iov_iter *i)
 595 {
 596         struct pipe_inode_info *pipe = i->pipe;
 597         unsigned int p_mask = pipe->ring_size - 1;
 598         __wsum sum = csstate->csum;
 599         size_t off = csstate->off;
 600         unsigned int i_head;
 601         size_t n, r;
 602
 603         if (!sanity(i))
 604                 return 0;
 605
 606         bytes = n = push_pipe(i, bytes, &i_head, &r);
 607         if (unlikely(!n))
 608                 return 0;
 609         do {
 610                 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
 611                 char *p = kmap_atomic(pipe->bufs[i_head & p_mask].page);
 612                 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
 613                 kunmap_atomic(p);
 614                 i->head = i_head;
 615                 i->iov_offset = r + chunk;
 616                 n -= chunk;
 617                 off += chunk;
 618                 addr += chunk;
 619                 r = 0;
 620                 i_head++;
 621         } while (n);
 622         i->count -= bytes;
 623         csstate->csum = sum;
 624         csstate->off = off;
 625         return bytes;
 626 }
 627
 628 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 629 {
 630         const char *from = addr;
 631         if (unlikely(iov_iter_is_pipe(i)))
 632                 return copy_pipe_to_iter(addr, bytes, i);
 633         if (iter_is_iovec(i))
 634                 might_fault();
 635         iterate_and_advance(i, bytes, v,
 636                 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 637                 memcpy_to_page(v.bv_page, v.bv_offset,
 638                                (from += v.bv_len) - v.bv_len, v.bv_len),
 639                 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 640                 memcpy_to_page(v.bv_page, v.bv_offset,
 641                                (from += v.bv_len) - v.bv_len, v.bv_len)
 642         )
 643
 644         return bytes;
 645 }
 646 EXPORT_SYMBOL(_copy_to_iter);
 647
 648 #ifdef CONFIG_ARCH_HAS_COPY_MC
 649 static int copyout_mc(void __user *to, const void *from, size_t n)
 650 {
 651         if (access_ok(to, n)) {
 652                 instrument_copy_to_user(to, from, n);
 653                 n = copy_mc_to_user((__force void *) to, from, n);
 654         }
 655         return n;
 656 }
 657
 658 static unsigned long copy_mc_to_page(struct page *page, size_t offset,
 659                 const char *from, size_t len)
 660 {
 661         unsigned long ret;
 662         char *to;
 663
 664         to = kmap_atomic(page);
 665         ret = copy_mc_to_kernel(to + offset, from, len);
 666         kunmap_atomic(to);
 667
 668         return ret;
 669 }
 670
 671 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 672                                 struct iov_iter *i)
 673 {
 674         struct pipe_inode_info *pipe = i->pipe;
 675         unsigned int p_mask = pipe->ring_size - 1;
 676         unsigned int i_head;
 677         size_t n, off, xfer = 0;
 678
 679         if (!sanity(i))
 680                 return 0;
 681
 682         bytes = n = push_pipe(i, bytes, &i_head, &off);
 683         if (unlikely(!n))
 684                 return 0;
 685         do {
 686                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 687                 unsigned long rem;
 688
 689                 rem = copy_mc_to_page(pipe->bufs[i_head & p_mask].page,
 690                                             off, addr, chunk);
 691                 i->head = i_head;
 692                 i->iov_offset = off + chunk - rem;
 693                 xfer += chunk - rem;
 694                 if (rem)
 695                         break;
 696                 n -= chunk;
 697                 addr += chunk;
 698                 off = 0;
 699                 i_head++;
 700         } while (n);
 701         i->count -= xfer;
 702         return xfer;
 703 }
 704
 705 /**
 706  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 707  * @addr: source kernel address
 708  * @bytes: total transfer length
 709  * @iter: destination iterator
 710  *
 711  * The pmem driver deploys this for the dax operation
 712  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 713  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 714  * successfully copied.
 715  *
 716  * The main differences between this and typical _copy_to_iter().
 717  *
 718  * * Typical tail/residue handling after a fault retries the copy
 719  *   byte-by-byte until the fault happens again. Re-triggering machine
 720  *   checks is potentially fatal so the implementation uses source
 721  *   alignment and poison alignment assumptions to avoid re-triggering
 722  *   hardware exceptions.
 723  *
 724  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 725  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 726  *   a short copy.
 727  */
 728 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 729 {
 730         const char *from = addr;
 731         unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
 732
 733         if (unlikely(iov_iter_is_pipe(i)))
 734                 return copy_mc_pipe_to_iter(addr, bytes, i);
 735         if (iter_is_iovec(i))
 736                 might_fault();
 737         iterate_and_advance(i, bytes, v,
 738                 copyout_mc(v.iov_base, (from += v.iov_len) - v.iov_len,
 739                            v.iov_len),
 740                 ({
 741                 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
 742                                       (from += v.bv_len) - v.bv_len, v.bv_len);
 743                 if (rem) {
 744                         curr_addr = (unsigned long) from;
 745                         bytes = curr_addr - s_addr - rem;
 746                         return bytes;
 747                 }
 748                 }),
 749                 ({
 750                 rem = copy_mc_to_kernel(v.iov_base, (from += v.iov_len)
 751                                         - v.iov_len, v.iov_len);
 752                 if (rem) {
 753                         curr_addr = (unsigned long) from;
 754                         bytes = curr_addr - s_addr - rem;
 755                         return bytes;
 756                 }
 757                 }),
 758                 ({
 759                 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
 760                                       (from += v.bv_len) - v.bv_len, v.bv_len);
 761                 if (rem) {
 762                         curr_addr = (unsigned long) from;
 763                         bytes = curr_addr - s_addr - rem;
 764                         rcu_read_unlock();
 765                         i->iov_offset += bytes;
 766                         i->count -= bytes;
 767                         return bytes;
 768                 }
 769                 })
 770         )
 771
 772         return bytes;
 773 }
 774 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 775 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 776
 777 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 778 {
 779         char *to = addr;
 780         if (unlikely(iov_iter_is_pipe(i))) {
 781                 WARN_ON(1);
 782                 return 0;
 783         }
 784         if (iter_is_iovec(i))
 785                 might_fault();
 786         iterate_and_advance(i, bytes, v,
 787                 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 788                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 789                                  v.bv_offset, v.bv_len),
 790                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 791                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 792                                  v.bv_offset, v.bv_len)
 793         )
 794
 795         return bytes;
 796 }
 797 EXPORT_SYMBOL(_copy_from_iter);
 798
 799 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 800 {
 801         char *to = addr;
 802         if (unlikely(iov_iter_is_pipe(i))) {
 803                 WARN_ON(1);
 804                 return 0;
 805         }
 806         iterate_and_advance(i, bytes, v,
 807                 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 808                                          v.iov_base, v.iov_len),
 809                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 810                                  v.bv_offset, v.bv_len),
 811                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 812                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 813                                  v.bv_offset, v.bv_len)
 814         )
 815
 816         return bytes;
 817 }
 818 EXPORT_SYMBOL(_copy_from_iter_nocache);
 819
 820 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 821 /**
 822  * _copy_from_iter_flushcache - write destination through cpu cache
 823  * @addr: destination kernel address
 824  * @bytes: total transfer length
 825  * @iter: source iterator
 826  *
 827  * The pmem driver arranges for filesystem-dax to use this facility via
 828  * dax_copy_from_iter() for ensuring that writes to persistent memory
 829  * are flushed through the CPU cache. It is differentiated from
 830  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 831  * all iterator types. The _copy_from_iter_nocache() only attempts to
 832  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 833  * instructions that strand dirty-data in the cache.
 834  */
 835 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 836 {
 837         char *to = addr;
 838         if (unlikely(iov_iter_is_pipe(i))) {
 839                 WARN_ON(1);
 840                 return 0;
 841         }
 842         iterate_and_advance(i, bytes, v,
 843                 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
 844                                          v.iov_base, v.iov_len),
 845                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 846                                  v.bv_offset, v.bv_len),
 847                 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
 848                         v.iov_len),
 849                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 850                                  v.bv_offset, v.bv_len)
 851         )
 852
 853         return bytes;
 854 }
 855 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 856 #endif
 857
 858 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 859 {
 860         struct page *head;
 861         size_t v = n + offset;
 862
 863         /*
 864          * The general case needs to access the page order in order
 865          * to compute the page size.
 866          * However, we mostly deal with order-0 pages and thus can
 867          * avoid a possible cache line miss for requests that fit all
 868          * page orders.
 869          */
 870         if (n <= v && v <= PAGE_SIZE)
 871                 return true;
 872
 873         head = compound_head(page);
 874         v += (page - head) << PAGE_SHIFT;
 875
 876         if (likely(n <= v && v <= (page_size(head))))
 877                 return true;
 878         WARN_ON(1);
 879         return false;
 880 }
 881
 882 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 883                          struct iov_iter *i)
 884 {
 885         if (likely(iter_is_iovec(i)))
 886                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 887         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 888                 void *kaddr = kmap_atomic(page);
 889                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 890                 kunmap_atomic(kaddr);
 891                 return wanted;
 892         }
 893         if (iov_iter_is_pipe(i))
 894                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 895         if (unlikely(iov_iter_is_discard(i))) {
 896                 if (unlikely(i->count < bytes))
 897                         bytes = i->count;
 898                 i->count -= bytes;
 899                 return bytes;
 900         }
 901         WARN_ON(1);
 902         return 0;
 903 }
 904
 905 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 906                          struct iov_iter *i)
 907 {
 908         size_t res = 0;
 909         if (unlikely(!page_copy_sane(page, offset, bytes)))
 910                 return 0;
 911         page += offset / PAGE_SIZE; // first subpage
 912         offset %= PAGE_SIZE;
 913         while (1) {
 914                 size_t n = __copy_page_to_iter(page, offset,
 915                                 min(bytes, (size_t)PAGE_SIZE - offset), i);
 916                 res += n;
 917                 bytes -= n;
 918                 if (!bytes || !n)
 919                         break;
 920                 offset += n;
 921                 if (offset == PAGE_SIZE) {
 922                         page++;
 923                         offset = 0;
 924                 }
 925         }
 926         return res;
 927 }
 928 EXPORT_SYMBOL(copy_page_to_iter);
 929
 930 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 931                          struct iov_iter *i)
 932 {
 933         if (unlikely(!page_copy_sane(page, offset, bytes)))
 934                 return 0;
 935         if (likely(iter_is_iovec(i)))
 936                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 937         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 938                 void *kaddr = kmap_atomic(page);
 939                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 940                 kunmap_atomic(kaddr);
 941                 return wanted;
 942         }
 943         WARN_ON(1);
 944         return 0;
 945 }
 946 EXPORT_SYMBOL(copy_page_from_iter);
 947
 948 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 949 {
 950         struct pipe_inode_info *pipe = i->pipe;
 951         unsigned int p_mask = pipe->ring_size - 1;
 952         unsigned int i_head;
 953         size_t n, off;
 954
 955         if (!sanity(i))
 956                 return 0;
 957
 958         bytes = n = push_pipe(i, bytes, &i_head, &off);
 959         if (unlikely(!n))
 960                 return 0;
 961
 962         do {
 963                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 964                 memzero_page(pipe->bufs[i_head & p_mask].page, off, chunk);
 965                 i->head = i_head;
 966                 i->iov_offset = off + chunk;
 967                 n -= chunk;
 968                 off = 0;
 969                 i_head++;
 970         } while (n);
 971         i->count -= bytes;
 972         return bytes;
 973 }
 974
 975 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 976 {
 977         if (unlikely(iov_iter_is_pipe(i)))
 978                 return pipe_zero(bytes, i);
 979         iterate_and_advance(i, bytes, v,
 980                 clear_user(v.iov_base, v.iov_len),
 981                 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
 982                 memset(v.iov_base, 0, v.iov_len),
 983                 memzero_page(v.bv_page, v.bv_offset, v.bv_len)
 984         )
 985
 986         return bytes;
 987 }
 988 EXPORT_SYMBOL(iov_iter_zero);
 989
 990 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 991                                   struct iov_iter *i)
 992 {
 993         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 994         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 995                 kunmap_atomic(kaddr);
 996                 return 0;
 997         }
 998         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 999                 kunmap_atomic(kaddr);
1000                 WARN_ON(1);
1001                 return 0;
1002         }
1003         iterate_and_advance(i, bytes, v,
1004                 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1005                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1006                                  v.bv_offset, v.bv_len),
1007                 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1008                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1009                                  v.bv_offset, v.bv_len)
1010         )
1011         kunmap_atomic(kaddr);
1012         return bytes;
1013 }
1014 EXPORT_SYMBOL(copy_page_from_iter_atomic);
1015
1016 static inline void pipe_truncate(struct iov_iter *i)
1017 {
1018         struct pipe_inode_info *pipe = i->pipe;
1019         unsigned int p_tail = pipe->tail;
1020         unsigned int p_head = pipe->head;
1021         unsigned int p_mask = pipe->ring_size - 1;
1022
1023         if (!pipe_empty(p_head, p_tail)) {
1024                 struct pipe_buffer *buf;
1025                 unsigned int i_head = i->head;
1026                 size_t off = i->iov_offset;
1027
1028                 if (off) {
1029                         buf = &pipe->bufs[i_head & p_mask];
1030                         buf->len = off - buf->offset;
1031                         i_head++;
1032                 }
1033                 while (p_head != i_head) {
1034                         p_head--;
1035                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1036                 }
1037
1038                 pipe->head = p_head;
1039         }
1040 }
1041
1042 static void pipe_advance(struct iov_iter *i, size_t size)
1043 {
1044         struct pipe_inode_info *pipe = i->pipe;
1045         if (size) {
1046                 struct pipe_buffer *buf;
1047                 unsigned int p_mask = pipe->ring_size - 1;
1048                 unsigned int i_head = i->head;
1049                 size_t off = i->iov_offset, left = size;
1050
1051                 if (off) /* make it relative to the beginning of buffer */
1052                         left += off - pipe->bufs[i_head & p_mask].offset;
1053                 while (1) {
1054                         buf = &pipe->bufs[i_head & p_mask];
1055                         if (left <= buf->len)
1056                                 break;
1057                         left -= buf->len;
1058                         i_head++;
1059                 }
1060                 i->head = i_head;
1061                 i->iov_offset = buf->offset + left;
1062         }
1063         i->count -= size;
1064         /* ... and discard everything past that point */
1065         pipe_truncate(i);
1066 }
1067
1068 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1069 {
1070         struct bvec_iter bi;
1071
1072         bi.bi_size = i->count;
1073         bi.bi_bvec_done = i->iov_offset;
1074         bi.bi_idx = 0;
1075         bvec_iter_advance(i->bvec, &bi, size);
1076
1077         i->bvec += bi.bi_idx;
1078         i->nr_segs -= bi.bi_idx;
1079         i->count = bi.bi_size;
1080         i->iov_offset = bi.bi_bvec_done;
1081 }
1082
1083 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
1084 {
1085         const struct iovec *iov, *end;
1086
1087         if (!i->count)
1088                 return;
1089         i->count -= size;
1090
1091         size += i->iov_offset; // from beginning of current segment
1092         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1093                 if (likely(size < iov->iov_len))
1094                         break;
1095                 size -= iov->iov_len;
1096         }
1097         i->iov_offset = size;
1098         i->nr_segs -= iov - i->iov;
1099         i->iov = iov;
1100 }
1101
1102 void iov_iter_advance(struct iov_iter *i, size_t size)
1103 {
1104         if (unlikely(i->count < size))
1105                 size = i->count;
1106         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1107                 /* iovec and kvec have identical layouts */
1108                 iov_iter_iovec_advance(i, size);
1109         } else if (iov_iter_is_bvec(i)) {
1110                 iov_iter_bvec_advance(i, size);
1111         } else if (iov_iter_is_pipe(i)) {
1112                 pipe_advance(i, size);
1113         } else if (unlikely(iov_iter_is_xarray(i))) {
1114                 i->iov_offset += size;
1115                 i->count -= size;
1116         } else if (iov_iter_is_discard(i)) {
1117                 i->count -= size;
1118         }
1119 }
1120 EXPORT_SYMBOL(iov_iter_advance);
1121
1122 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1123 {
1124         if (!unroll)
1125                 return;
1126         if (WARN_ON(unroll > MAX_RW_COUNT))
1127                 return;
1128         i->count += unroll;
1129         if (unlikely(iov_iter_is_pipe(i))) {
1130                 struct pipe_inode_info *pipe = i->pipe;
1131                 unsigned int p_mask = pipe->ring_size - 1;
1132                 unsigned int i_head = i->head;
1133                 size_t off = i->iov_offset;
1134                 while (1) {
1135                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1136                         size_t n = off - b->offset;
1137                         if (unroll < n) {
1138                                 off -= unroll;
1139                                 break;
1140                         }
1141                         unroll -= n;
1142                         if (!unroll && i_head == i->start_head) {
1143                                 off = 0;
1144                                 break;
1145                         }
1146                         i_head--;
1147                         b = &pipe->bufs[i_head & p_mask];
1148                         off = b->offset + b->len;
1149                 }
1150                 i->iov_offset = off;
1151                 i->head = i_head;
1152                 pipe_truncate(i);
1153                 return;
1154         }
1155         if (unlikely(iov_iter_is_discard(i)))
1156                 return;
1157         if (unroll <= i->iov_offset) {
1158                 i->iov_offset -= unroll;
1159                 return;
1160         }
1161         unroll -= i->iov_offset;
1162         if (iov_iter_is_xarray(i)) {
1163                 BUG(); /* We should never go beyond the start of the specified
1164                         * range since we might then be straying into pages that
1165                         * aren't pinned.
1166                         */
1167         } else if (iov_iter_is_bvec(i)) {
1168                 const struct bio_vec *bvec = i->bvec;
1169                 while (1) {
1170                         size_t n = (--bvec)->bv_len;
1171                         i->nr_segs++;
1172                         if (unroll <= n) {
1173                                 i->bvec = bvec;
1174                                 i->iov_offset = n - unroll;
1175                                 return;
1176                         }
1177                         unroll -= n;
1178                 }
1179         } else { /* same logics for iovec and kvec */
1180                 const struct iovec *iov = i->iov;
1181                 while (1) {
1182                         size_t n = (--iov)->iov_len;
1183                         i->nr_segs++;
1184                         if (unroll <= n) {
1185                                 i->iov = iov;
1186                                 i->iov_offset = n - unroll;
1187                                 return;
1188                         }
1189                         unroll -= n;
1190                 }
1191         }
1192 }
1193 EXPORT_SYMBOL(iov_iter_revert);
1194
1195 /*
1196  * Return the count of just the current iov_iter segment.
1197  */
1198 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1199 {
1200         if (i->nr_segs > 1) {
1201                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1202                         return min(i->count, i->iov->iov_len - i->iov_offset);
1203                 if (iov_iter_is_bvec(i))
1204                         return min(i->count, i->bvec->bv_len - i->iov_offset);
1205         }
1206         return i->count;
1207 }
1208 EXPORT_SYMBOL(iov_iter_single_seg_count);
1209
1210 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1211                         const struct kvec *kvec, unsigned long nr_segs,
1212                         size_t count)
1213 {
1214         WARN_ON(direction & ~(READ | WRITE));
1215         *i = (struct iov_iter){
1216                 .iter_type = ITER_KVEC,
1217                 .data_source = direction,
1218                 .kvec = kvec,
1219                 .nr_segs = nr_segs,
1220                 .iov_offset = 0,
1221                 .count = count
1222         };
1223 }
1224 EXPORT_SYMBOL(iov_iter_kvec);
1225
1226 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1227                         const struct bio_vec *bvec, unsigned long nr_segs,
1228                         size_t count)
1229 {
1230         WARN_ON(direction & ~(READ | WRITE));
1231         *i = (struct iov_iter){
1232                 .iter_type = ITER_BVEC,
1233                 .data_source = direction,
1234                 .bvec = bvec,
1235                 .nr_segs = nr_segs,
1236                 .iov_offset = 0,
1237                 .count = count
1238         };
1239 }
1240 EXPORT_SYMBOL(iov_iter_bvec);
1241
1242 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1243                         struct pipe_inode_info *pipe,
1244                         size_t count)
1245 {
1246         BUG_ON(direction != READ);
1247         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1248         *i = (struct iov_iter){
1249                 .iter_type = ITER_PIPE,
1250                 .data_source = false,
1251                 .pipe = pipe,
1252                 .head = pipe->head,
1253                 .start_head = pipe->head,
1254                 .iov_offset = 0,
1255                 .count = count
1256         };
1257 }
1258 EXPORT_SYMBOL(iov_iter_pipe);
1259
1260 /**
1261  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1262  * @i: The iterator to initialise.
1263  * @direction: The direction of the transfer.
1264  * @xarray: The xarray to access.
1265  * @start: The start file position.
1266  * @count: The size of the I/O buffer in bytes.
1267  *
1268  * Set up an I/O iterator to either draw data out of the pages attached to an
1269  * inode or to inject data into those pages.  The pages *must* be prevented
1270  * from evaporation, either by taking a ref on them or locking them by the
1271  * caller.
1272  */
1273 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1274                      struct xarray *xarray, loff_t start, size_t count)
1275 {
1276         BUG_ON(direction & ~1);
1277         *i = (struct iov_iter) {
1278                 .iter_type = ITER_XARRAY,
1279                 .data_source = direction,
1280                 .xarray = xarray,
1281                 .xarray_start = start,
1282                 .count = count,
1283                 .iov_offset = 0
1284         };
1285 }
1286 EXPORT_SYMBOL(iov_iter_xarray);
1287
1288 /**
1289  * iov_iter_discard - Initialise an I/O iterator that discards data
1290  * @i: The iterator to initialise.
1291  * @direction: The direction of the transfer.
1292  * @count: The size of the I/O buffer in bytes.
1293  *
1294  * Set up an I/O iterator that just discards everything that's written to it.
1295  * It's only available as a READ iterator.
1296  */
1297 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1298 {
1299         BUG_ON(direction != READ);
1300         *i = (struct iov_iter){
1301                 .iter_type = ITER_DISCARD,
1302                 .data_source = false,
1303                 .count = count,
1304                 .iov_offset = 0
1305         };
1306 }
1307 EXPORT_SYMBOL(iov_iter_discard);
1308
1309 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1310 {
1311         unsigned long res = 0;
1312         size_t size = i->count;
1313         size_t skip = i->iov_offset;
1314         unsigned k;
1315
1316         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1317                 size_t len = i->iov[k].iov_len - skip;
1318                 if (len) {
1319                         res |= (unsigned long)i->iov[k].iov_base + skip;
1320                         if (len > size)
1321                                 len = size;
1322                         res |= len;
1323                         size -= len;
1324                         if (!size)
1325                                 break;
1326                 }
1327         }
1328         return res;
1329 }
1330
1331 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1332 {
1333         unsigned res = 0;
1334         size_t size = i->count;
1335         unsigned skip = i->iov_offset;
1336         unsigned k;
1337
1338         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1339                 size_t len = i->bvec[k].bv_len - skip;
1340                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1341                 if (len > size)
1342                         len = size;
1343                 res |= len;
1344                 size -= len;
1345                 if (!size)
1346                         break;
1347         }
1348         return res;
1349 }
1350
1351 unsigned long iov_iter_alignment(const struct iov_iter *i)
1352 {
1353         /* iovec and kvec have identical layouts */
1354         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1355                 return iov_iter_alignment_iovec(i);
1356
1357         if (iov_iter_is_bvec(i))
1358                 return iov_iter_alignment_bvec(i);
1359
1360         if (iov_iter_is_pipe(i)) {
1361                 unsigned int p_mask = i->pipe->ring_size - 1;
1362                 size_t size = i->count;
1363
1364                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1365                         return size | i->iov_offset;
1366                 return size;
1367         }
1368
1369         if (iov_iter_is_xarray(i))
1370                 return (i->xarray_start + i->iov_offset) | i->count;
1371
1372         return 0;
1373 }
1374 EXPORT_SYMBOL(iov_iter_alignment);
1375
1376 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1377 {
1378         unsigned long res = 0;
1379         unsigned long v = 0;
1380         size_t size = i->count;
1381         unsigned k;
1382
1383         if (WARN_ON(!iter_is_iovec(i)))
1384                 return ~0U;
1385
1386         for (k = 0; k < i->nr_segs; k++) {
1387                 if (i->iov[k].iov_len) {
1388                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1389                         if (v) // if not the first one
1390                                 res |= base | v; // this start | previous end
1391                         v = base + i->iov[k].iov_len;
1392                         if (size <= i->iov[k].iov_len)
1393                                 break;
1394                         size -= i->iov[k].iov_len;
1395                 }
1396         }
1397         return res;
1398 }
1399 EXPORT_SYMBOL(iov_iter_gap_alignment);
1400
1401 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1402                                 size_t maxsize,
1403                                 struct page **pages,
1404                                 int iter_head,
1405                                 size_t *start)
1406 {
1407         struct pipe_inode_info *pipe = i->pipe;
1408         unsigned int p_mask = pipe->ring_size - 1;
1409         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1410         if (!n)
1411                 return -EFAULT;
1412
1413         maxsize = n;
1414         n += *start;
1415         while (n > 0) {
1416                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1417                 iter_head++;
1418                 n -= PAGE_SIZE;
1419         }
1420
1421         return maxsize;
1422 }
1423
1424 static ssize_t pipe_get_pages(struct iov_iter *i,
1425                    struct page **pages, size_t maxsize, unsigned maxpages,
1426                    size_t *start)
1427 {
1428         unsigned int iter_head, npages;
1429         size_t capacity;
1430
1431         if (!sanity(i))
1432                 return -EFAULT;
1433
1434         data_start(i, &iter_head, start);
1435         /* Amount of free space: some of this one + all after this one */
1436         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1437         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1438
1439         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1440 }
1441
1442 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1443                                           pgoff_t index, unsigned int nr_pages)
1444 {
1445         XA_STATE(xas, xa, index);
1446         struct page *page;
1447         unsigned int ret = 0;
1448
1449         rcu_read_lock();
1450         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1451                 if (xas_retry(&xas, page))
1452                         continue;
1453
1454                 /* Has the page moved or been split? */
1455                 if (unlikely(page != xas_reload(&xas))) {
1456                         xas_reset(&xas);
1457                         continue;
1458                 }
1459
1460                 pages[ret] = find_subpage(page, xas.xa_index);
1461                 get_page(pages[ret]);
1462                 if (++ret == nr_pages)
1463                         break;
1464         }
1465         rcu_read_unlock();
1466         return ret;
1467 }
1468
1469 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1470                                      struct page **pages, size_t maxsize,
1471                                      unsigned maxpages, size_t *_start_offset)
1472 {
1473         unsigned nr, offset;
1474         pgoff_t index, count;
1475         size_t size = maxsize, actual;
1476         loff_t pos;
1477
1478         if (!size || !maxpages)
1479                 return 0;
1480
1481         pos = i->xarray_start + i->iov_offset;
1482         index = pos >> PAGE_SHIFT;
1483         offset = pos & ~PAGE_MASK;
1484         *_start_offset = offset;
1485
1486         count = 1;
1487         if (size > PAGE_SIZE - offset) {
1488                 size -= PAGE_SIZE - offset;
1489                 count += size >> PAGE_SHIFT;
1490                 size &= ~PAGE_MASK;
1491                 if (size)
1492                         count++;
1493         }
1494
1495         if (count > maxpages)
1496                 count = maxpages;
1497
1498         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1499         if (nr == 0)
1500                 return 0;
1501
1502         actual = PAGE_SIZE * nr;
1503         actual -= offset;
1504         if (nr == count && size > 0) {
1505                 unsigned last_offset = (nr > 1) ? 0 : offset;
1506                 actual -= PAGE_SIZE - (last_offset + size);
1507         }
1508         return actual;
1509 }
1510
1511 /* must be done on non-empty ITER_IOVEC one */
1512 static unsigned long first_iovec_segment(const struct iov_iter *i,
1513                                          size_t *size, size_t *start,
1514                                          size_t maxsize, unsigned maxpages)
1515 {
1516         size_t skip;
1517         long k;
1518
1519         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1520                 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1521                 size_t len = i->iov[k].iov_len - skip;
1522
1523                 if (unlikely(!len))
1524                         continue;
1525                 if (len > maxsize)
1526                         len = maxsize;
1527                 len += (*start = addr % PAGE_SIZE);
1528                 if (len > maxpages * PAGE_SIZE)
1529                         len = maxpages * PAGE_SIZE;
1530                 *size = len;
1531                 return addr & PAGE_MASK;
1532         }
1533         BUG(); // if it had been empty, we wouldn't get called
1534 }
1535
1536 /* must be done on non-empty ITER_BVEC one */
1537 static struct page *first_bvec_segment(const struct iov_iter *i,
1538                                        size_t *size, size_t *start,
1539                                        size_t maxsize, unsigned maxpages)
1540 {
1541         struct page *page;
1542         size_t skip = i->iov_offset, len;
1543
1544         len = i->bvec->bv_len - skip;
1545         if (len > maxsize)
1546                 len = maxsize;
1547         skip += i->bvec->bv_offset;
1548         page = i->bvec->bv_page + skip / PAGE_SIZE;
1549         len += (*start = skip % PAGE_SIZE);
1550         if (len > maxpages * PAGE_SIZE)
1551                 len = maxpages * PAGE_SIZE;
1552         *size = len;
1553         return page;
1554 }
1555
1556 ssize_t iov_iter_get_pages(struct iov_iter *i,
1557                    struct page **pages, size_t maxsize, unsigned maxpages,
1558                    size_t *start)
1559 {
1560         size_t len;
1561         int n, res;
1562
1563         if (maxsize > i->count)
1564                 maxsize = i->count;
1565         if (!maxsize)
1566                 return 0;
1567
1568         if (likely(iter_is_iovec(i))) {
1569                 unsigned long addr;
1570
1571                 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1572                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1573                 res = get_user_pages_fast(addr, n,
1574                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0,
1575                                 pages);
1576                 if (unlikely(res < 0))
1577                         return res;
1578                 return (res == n ? len : res * PAGE_SIZE) - *start;
1579         }
1580         if (iov_iter_is_bvec(i)) {
1581                 struct page *page;
1582
1583                 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1584                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1585                 while (n--)
1586                         get_page(*pages++ = page++);
1587                 return len - *start;
1588         }
1589         if (iov_iter_is_pipe(i))
1590                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1591         if (iov_iter_is_xarray(i))
1592                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1593         return -EFAULT;
1594 }
1595 EXPORT_SYMBOL(iov_iter_get_pages);
1596
1597 static struct page **get_pages_array(size_t n)
1598 {
1599         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1600 }
1601
1602 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1603                    struct page ***pages, size_t maxsize,
1604                    size_t *start)
1605 {
1606         struct page **p;
1607         unsigned int iter_head, npages;
1608         ssize_t n;
1609
1610         if (!sanity(i))
1611                 return -EFAULT;
1612
1613         data_start(i, &iter_head, start);
1614         /* Amount of free space: some of this one + all after this one */
1615         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1616         n = npages * PAGE_SIZE - *start;
1617         if (maxsize > n)
1618                 maxsize = n;
1619         else
1620                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1621         p = get_pages_array(npages);
1622         if (!p)
1623                 return -ENOMEM;
1624         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1625         if (n > 0)
1626                 *pages = p;
1627         else
1628                 kvfree(p);
1629         return n;
1630 }
1631
1632 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1633                                            struct page ***pages, size_t maxsize,
1634                                            size_t *_start_offset)
1635 {
1636         struct page **p;
1637         unsigned nr, offset;
1638         pgoff_t index, count;
1639         size_t size = maxsize, actual;
1640         loff_t pos;
1641
1642         if (!size)
1643                 return 0;
1644
1645         pos = i->xarray_start + i->iov_offset;
1646         index = pos >> PAGE_SHIFT;
1647         offset = pos & ~PAGE_MASK;
1648         *_start_offset = offset;
1649
1650         count = 1;
1651         if (size > PAGE_SIZE - offset) {
1652                 size -= PAGE_SIZE - offset;
1653                 count += size >> PAGE_SHIFT;
1654                 size &= ~PAGE_MASK;
1655                 if (size)
1656                         count++;
1657         }
1658
1659         p = get_pages_array(count);
1660         if (!p)
1661                 return -ENOMEM;
1662         *pages = p;
1663
1664         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1665         if (nr == 0)
1666                 return 0;
1667
1668         actual = PAGE_SIZE * nr;
1669         actual -= offset;
1670         if (nr == count && size > 0) {
1671                 unsigned last_offset = (nr > 1) ? 0 : offset;
1672                 actual -= PAGE_SIZE - (last_offset + size);
1673         }
1674         return actual;
1675 }
1676
1677 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1678                    struct page ***pages, size_t maxsize,
1679                    size_t *start)
1680 {
1681         struct page **p;
1682         size_t len;
1683         int n, res;
1684
1685         if (maxsize > i->count)
1686                 maxsize = i->count;
1687         if (!maxsize)
1688                 return 0;
1689
1690         if (likely(iter_is_iovec(i))) {
1691                 unsigned long addr;
1692
1693                 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1694                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1695                 p = get_pages_array(n);
1696                 if (!p)
1697                         return -ENOMEM;
1698                 res = get_user_pages_fast(addr, n,
1699                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0, p);
1700                 if (unlikely(res < 0)) {
1701                         kvfree(p);
1702                         return res;
1703                 }
1704                 *pages = p;
1705                 return (res == n ? len : res * PAGE_SIZE) - *start;
1706         }
1707         if (iov_iter_is_bvec(i)) {
1708                 struct page *page;
1709
1710                 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1711                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1712                 *pages = p = get_pages_array(n);
1713                 if (!p)
1714                         return -ENOMEM;
1715                 while (n--)
1716                         get_page(*p++ = page++);
1717                 return len - *start;
1718         }
1719         if (iov_iter_is_pipe(i))
1720                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1721         if (iov_iter_is_xarray(i))
1722                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1723         return -EFAULT;
1724 }
1725 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1726
1727 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1728                                struct iov_iter *i)
1729 {
1730         char *to = addr;
1731         __wsum sum, next;
1732         size_t off = 0;
1733         sum = *csum;
1734         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1735                 WARN_ON(1);
1736                 return 0;
1737         }
1738         iterate_and_advance(i, bytes, v, ({
1739                 next = csum_and_copy_from_user(v.iov_base,
1740                                                (to += v.iov_len) - v.iov_len,
1741                                                v.iov_len);
1742                 if (next) {
1743                         sum = csum_block_add(sum, next, off);
1744                         off += v.iov_len;
1745                 }
1746                 next ? 0 : v.iov_len;
1747         }), ({
1748                 char *p = kmap_atomic(v.bv_page);
1749                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1750                                       p + v.bv_offset, v.bv_len,
1751                                       sum, off);
1752                 kunmap_atomic(p);
1753                 off += v.bv_len;
1754         }),({
1755                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1756                                       v.iov_base, v.iov_len,
1757                                       sum, off);
1758                 off += v.iov_len;
1759         }), ({
1760                 char *p = kmap_atomic(v.bv_page);
1761                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1762                                       p + v.bv_offset, v.bv_len,
1763                                       sum, off);
1764                 kunmap_atomic(p);
1765                 off += v.bv_len;
1766         })
1767         )
1768         *csum = sum;
1769         return bytes;
1770 }
1771 EXPORT_SYMBOL(csum_and_copy_from_iter);
1772
1773 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1774                              struct iov_iter *i)
1775 {
1776         struct csum_state *csstate = _csstate;
1777         const char *from = addr;
1778         __wsum sum, next;
1779         size_t off;
1780
1781         if (unlikely(iov_iter_is_pipe(i)))
1782                 return csum_and_copy_to_pipe_iter(addr, bytes, _csstate, i);
1783
1784         sum = csum_shift(csstate->csum, csstate->off);
1785         off = 0;
1786         if (unlikely(iov_iter_is_discard(i))) {
1787                 WARN_ON(1);     /* for now */
1788                 return 0;
1789         }
1790         iterate_and_advance(i, bytes, v, ({
1791                 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1792                                              v.iov_base,
1793                                              v.iov_len);
1794                 if (next) {
1795                         sum = csum_block_add(sum, next, off);
1796                         off += v.iov_len;
1797                 }
1798                 next ? 0 : v.iov_len;
1799         }), ({
1800                 char *p = kmap_atomic(v.bv_page);
1801                 sum = csum_and_memcpy(p + v.bv_offset,
1802                                       (from += v.bv_len) - v.bv_len,
1803                                       v.bv_len, sum, off);
1804                 kunmap_atomic(p);
1805                 off += v.bv_len;
1806         }),({
1807                 sum = csum_and_memcpy(v.iov_base,
1808                                      (from += v.iov_len) - v.iov_len,
1809                                      v.iov_len, sum, off);
1810                 off += v.iov_len;
1811         }), ({
1812                 char *p = kmap_atomic(v.bv_page);
1813                 sum = csum_and_memcpy(p + v.bv_offset,
1814                                       (from += v.bv_len) - v.bv_len,
1815                                       v.bv_len, sum, off);
1816                 kunmap_atomic(p);
1817                 off += v.bv_len;
1818         })
1819         )
1820         csstate->csum = csum_shift(sum, csstate->off);
1821         csstate->off += bytes;
1822         return bytes;
1823 }
1824 EXPORT_SYMBOL(csum_and_copy_to_iter);
1825
1826 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1827                 struct iov_iter *i)
1828 {
1829 #ifdef CONFIG_CRYPTO_HASH
1830         struct ahash_request *hash = hashp;
1831         struct scatterlist sg;
1832         size_t copied;
1833
1834         copied = copy_to_iter(addr, bytes, i);
1835         sg_init_one(&sg, addr, copied);
1836         ahash_request_set_crypt(hash, &sg, NULL, copied);
1837         crypto_ahash_update(hash);
1838         return copied;
1839 #else
1840         return 0;
1841 #endif
1842 }
1843 EXPORT_SYMBOL(hash_and_copy_to_iter);
1844
1845 static int iov_npages(const struct iov_iter *i, int maxpages)
1846 {
1847         size_t skip = i->iov_offset, size = i->count;
1848         const struct iovec *p;
1849         int npages = 0;
1850
1851         for (p = i->iov; size; skip = 0, p++) {
1852                 unsigned offs = offset_in_page(p->iov_base + skip);
1853                 size_t len = min(p->iov_len - skip, size);
1854
1855                 if (len) {
1856                         size -= len;
1857                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1858                         if (unlikely(npages > maxpages))
1859                                 return maxpages;
1860                 }
1861         }
1862         return npages;
1863 }
1864
1865 static int bvec_npages(const struct iov_iter *i, int maxpages)
1866 {
1867         size_t skip = i->iov_offset, size = i->count;
1868         const struct bio_vec *p;
1869         int npages = 0;
1870
1871         for (p = i->bvec; size; skip = 0, p++) {
1872                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1873                 size_t len = min(p->bv_len - skip, size);
1874
1875                 size -= len;
1876                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1877                 if (unlikely(npages > maxpages))
1878                         return maxpages;
1879         }
1880         return npages;
1881 }
1882
1883 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1884 {
1885         if (unlikely(!i->count))
1886                 return 0;
1887         /* iovec and kvec have identical layouts */
1888         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1889                 return iov_npages(i, maxpages);
1890         if (iov_iter_is_bvec(i))
1891                 return bvec_npages(i, maxpages);
1892         if (iov_iter_is_pipe(i)) {
1893                 unsigned int iter_head;
1894                 int npages;
1895                 size_t off;
1896
1897                 if (!sanity(i))
1898                         return 0;
1899
1900                 data_start(i, &iter_head, &off);
1901                 /* some of this one + all after this one */
1902                 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1903                 return min(npages, maxpages);
1904         }
1905         if (iov_iter_is_xarray(i)) {
1906                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1907                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1908                 return min(npages, maxpages);
1909         }
1910         return 0;
1911 }
1912 EXPORT_SYMBOL(iov_iter_npages);
1913
1914 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1915 {
1916         *new = *old;
1917         if (unlikely(iov_iter_is_pipe(new))) {
1918                 WARN_ON(1);
1919                 return NULL;
1920         }
1921         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1922                 return NULL;
1923         if (iov_iter_is_bvec(new))
1924                 return new->bvec = kmemdup(new->bvec,
1925                                     new->nr_segs * sizeof(struct bio_vec),
1926                                     flags);
1927         else
1928                 /* iovec and kvec have identical layout */
1929                 return new->iov = kmemdup(new->iov,
1930                                    new->nr_segs * sizeof(struct iovec),
1931                                    flags);
1932 }
1933 EXPORT_SYMBOL(dup_iter);
1934
1935 static int copy_compat_iovec_from_user(struct iovec *iov,
1936                 const struct iovec __user *uvec, unsigned long nr_segs)
1937 {
1938         const struct compat_iovec __user *uiov =
1939                 (const struct compat_iovec __user *)uvec;
1940         int ret = -EFAULT, i;
1941
1942         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1943                 return -EFAULT;
1944
1945         for (i = 0; i < nr_segs; i++) {
1946                 compat_uptr_t buf;
1947                 compat_ssize_t len;
1948
1949                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1950                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1951
1952                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1953                 if (len < 0) {
1954                         ret = -EINVAL;
1955                         goto uaccess_end;
1956                 }
1957                 iov[i].iov_base = compat_ptr(buf);
1958                 iov[i].iov_len = len;
1959         }
1960
1961         ret = 0;
1962 uaccess_end:
1963         user_access_end();
1964         return ret;
1965 }
1966
1967 static int copy_iovec_from_user(struct iovec *iov,
1968                 const struct iovec __user *uvec, unsigned long nr_segs)
1969 {
1970         unsigned long seg;
1971
1972         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1973                 return -EFAULT;
1974         for (seg = 0; seg < nr_segs; seg++) {
1975                 if ((ssize_t)iov[seg].iov_len < 0)
1976                         return -EINVAL;
1977         }
1978
1979         return 0;
1980 }
1981
1982 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1983                 unsigned long nr_segs, unsigned long fast_segs,
1984                 struct iovec *fast_iov, bool compat)
1985 {
1986         struct iovec *iov = fast_iov;
1987         int ret;
1988
1989         /*
1990          * SuS says "The readv() function *may* fail if the iovcnt argument was
1991          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1992          * traditionally returned zero for zero segments, so...
1993          */
1994         if (nr_segs == 0)
1995                 return iov;
1996         if (nr_segs > UIO_MAXIOV)
1997                 return ERR_PTR(-EINVAL);
1998         if (nr_segs > fast_segs) {
1999                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
2000                 if (!iov)
2001                         return ERR_PTR(-ENOMEM);
2002         }
2003
2004         if (compat)
2005                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
2006         else
2007                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
2008         if (ret) {
2009                 if (iov != fast_iov)
2010                         kfree(iov);
2011                 return ERR_PTR(ret);
2012         }
2013
2014         return iov;
2015 }
2016
2017 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
2018                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
2019                  struct iov_iter *i, bool compat)
2020 {
2021         ssize_t total_len = 0;
2022         unsigned long seg;
2023         struct iovec *iov;
2024
2025         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
2026         if (IS_ERR(iov)) {
2027                 *iovp = NULL;
2028                 return PTR_ERR(iov);
2029         }
2030
2031         /*
2032          * According to the Single Unix Specification we should return EINVAL if
2033          * an element length is < 0 when cast to ssize_t or if the total length
2034          * would overflow the ssize_t return value of the system call.
2035          *
2036          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
2037          * overflow case.
2038          */
2039         for (seg = 0; seg < nr_segs; seg++) {
2040                 ssize_t len = (ssize_t)iov[seg].iov_len;
2041
2042                 if (!access_ok(iov[seg].iov_base, len)) {
2043                         if (iov != *iovp)
2044                                 kfree(iov);
2045                         *iovp = NULL;
2046                         return -EFAULT;
2047                 }
2048
2049                 if (len > MAX_RW_COUNT - total_len) {
2050                         len = MAX_RW_COUNT - total_len;
2051                         iov[seg].iov_len = len;
2052                 }
2053                 total_len += len;
2054         }
2055
2056         iov_iter_init(i, type, iov, nr_segs, total_len);
2057         if (iov == *iovp)
2058                 *iovp = NULL;
2059         else
2060                 *iovp = iov;
2061         return total_len;
2062 }
2063
2064 /**
2065  * import_iovec() - Copy an array of &struct iovec from userspace
2066  *     into the kernel, check that it is valid, and initialize a new
2067  *     &struct iov_iter iterator to access it.
2068  *
2069  * @type: One of %READ or %WRITE.
2070  * @uvec: Pointer to the userspace array.
2071  * @nr_segs: Number of elements in userspace array.
2072  * @fast_segs: Number of elements in @iov.
2073  * @iovp: (input and output parameter) Pointer to pointer to (usually small
2074  *     on-stack) kernel array.
2075  * @i: Pointer to iterator that will be initialized on success.
2076  *
2077  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
2078  * then this function places %NULL in *@iov on return. Otherwise, a new
2079  * array will be allocated and the result placed in *@iov. This means that
2080  * the caller may call kfree() on *@iov regardless of whether the small
2081  * on-stack array was used or not (and regardless of whether this function
2082  * returns an error or not).
2083  *
2084  * Return: Negative error code on error, bytes imported on success
2085  */
2086 ssize_t import_iovec(int type, const struct iovec __user *uvec,
2087                  unsigned nr_segs, unsigned fast_segs,
2088                  struct iovec **iovp, struct iov_iter *i)
2089 {
2090         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
2091                               in_compat_syscall());
2092 }
2093 EXPORT_SYMBOL(import_iovec);
2094
2095 int import_single_range(int rw, void __user *buf, size_t len,
2096                  struct iovec *iov, struct iov_iter *i)
2097 {
2098         if (len > MAX_RW_COUNT)
2099                 len = MAX_RW_COUNT;
2100         if (unlikely(!access_ok(buf, len)))
2101                 return -EFAULT;
2102
2103         iov->iov_base = buf;
2104         iov->iov_len = len;
2105         iov_iter_init(i, rw, iov, 1, len);
2106         return 0;
2107 }
2108 EXPORT_SYMBOL(import_single_range);