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