lib/iov_iter.c

   1 #include <linux/export.h>
   2 #include <linux/bvec.h>
   3 #include <linux/uio.h>
   4 #include <linux/pagemap.h>
   5 #include <linux/slab.h>
   6 #include <linux/vmalloc.h>
   7 #include <linux/splice.h>
   8 #include <net/checksum.h>
   9
  10 #define PIPE_PARANOIA /* for now */
  11
  12 #define iterate_iovec(i, n, __v, __p, skip, STEP) {     \
  13         size_t left;                                    \
  14         size_t wanted = n;                              \
  15         __p = i->iov;                                   \
  16         __v.iov_len = min(n, __p->iov_len - skip);      \
  17         if (likely(__v.iov_len)) {                      \
  18                 __v.iov_base = __p->iov_base + skip;    \
  19                 left = (STEP);                          \
  20                 __v.iov_len -= left;                    \
  21                 skip += __v.iov_len;                    \
  22                 n -= __v.iov_len;                       \
  23         } else {                                        \
  24                 left = 0;                               \
  25         }                                               \
  26         while (unlikely(!left && n)) {                  \
  27                 __p++;                                  \
  28                 __v.iov_len = min(n, __p->iov_len);     \
  29                 if (unlikely(!__v.iov_len))             \
  30                         continue;                       \
  31                 __v.iov_base = __p->iov_base;           \
  32                 left = (STEP);                          \
  33                 __v.iov_len -= left;                    \
  34                 skip = __v.iov_len;                     \
  35                 n -= __v.iov_len;                       \
  36         }                                               \
  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         __v.iov_len = min(n, __p->iov_len - skip);      \
  44         if (likely(__v.iov_len)) {                      \
  45                 __v.iov_base = __p->iov_base + skip;    \
  46                 (void)(STEP);                           \
  47                 skip += __v.iov_len;                    \
  48                 n -= __v.iov_len;                       \
  49         }                                               \
  50         while (unlikely(n)) {                           \
  51                 __p++;                                  \
  52                 __v.iov_len = min(n, __p->iov_len);     \
  53                 if (unlikely(!__v.iov_len))             \
  54                         continue;                       \
  55                 __v.iov_base = __p->iov_base;           \
  56                 (void)(STEP);                           \
  57                 skip = __v.iov_len;                     \
  58                 n -= __v.iov_len;                       \
  59         }                                               \
  60         n = wanted;                                     \
  61 }
  62
  63 #define iterate_bvec(i, n, __v, __bi, skip, STEP) {     \
  64         struct bvec_iter __start;                       \
  65         __start.bi_size = n;                            \
  66         __start.bi_bvec_done = skip;                    \
  67         __start.bi_idx = 0;                             \
  68         for_each_bvec(__v, i->bvec, __bi, __start) {    \
  69                 if (!__v.bv_len)                        \
  70                         continue;                       \
  71                 (void)(STEP);                           \
  72         }                                               \
  73 }
  74
  75 #define iterate_all_kinds(i, n, v, I, B, K) {                   \
  76         if (likely(n)) {                                        \
  77                 size_t skip = i->iov_offset;                    \
  78                 if (unlikely(i->type & ITER_BVEC)) {            \
  79                         struct bio_vec v;                       \
  80                         struct bvec_iter __bi;                  \
  81                         iterate_bvec(i, n, v, __bi, skip, (B))  \
  82                 } else if (unlikely(i->type & ITER_KVEC)) {     \
  83                         const struct kvec *kvec;                \
  84                         struct kvec v;                          \
  85                         iterate_kvec(i, n, v, kvec, skip, (K))  \
  86                 } else {                                        \
  87                         const struct iovec *iov;                \
  88                         struct iovec v;                         \
  89                         iterate_iovec(i, n, v, iov, skip, (I))  \
  90                 }                                               \
  91         }                                                       \
  92 }
  93
  94 #define iterate_and_advance(i, n, v, I, B, K) {                 \
  95         if (unlikely(i->count < n))                             \
  96                 n = i->count;                                   \
  97         if (i->count) {                                         \
  98                 size_t skip = i->iov_offset;                    \
  99                 if (unlikely(i->type & ITER_BVEC)) {            \
 100                         const struct bio_vec *bvec = i->bvec;   \
 101                         struct bio_vec v;                       \
 102                         struct bvec_iter __bi;                  \
 103                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 104                         i->bvec = __bvec_iter_bvec(i->bvec, __bi);      \
 105                         i->nr_segs -= i->bvec - bvec;           \
 106                         skip = __bi.bi_bvec_done;               \
 107                 } else if (unlikely(i->type & ITER_KVEC)) {     \
 108                         const struct kvec *kvec;                \
 109                         struct kvec v;                          \
 110                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 111                         if (skip == kvec->iov_len) {            \
 112                                 kvec++;                         \
 113                                 skip = 0;                       \
 114                         }                                       \
 115                         i->nr_segs -= kvec - i->kvec;           \
 116                         i->kvec = kvec;                         \
 117                 } else {                                        \
 118                         const struct iovec *iov;                \
 119                         struct iovec v;                         \
 120                         iterate_iovec(i, n, v, iov, skip, (I))  \
 121                         if (skip == iov->iov_len) {             \
 122                                 iov++;                          \
 123                                 skip = 0;                       \
 124                         }                                       \
 125                         i->nr_segs -= iov - i->iov;             \
 126                         i->iov = iov;                           \
 127                 }                                               \
 128                 i->count -= n;                                  \
 129                 i->iov_offset = skip;                           \
 130         }                                                       \
 131 }
 132
 133 static int copyout(void __user *to, const void *from, size_t n)
 134 {
 135         if (access_ok(VERIFY_WRITE, to, n)) {
 136                 kasan_check_read(from, n);
 137                 n = raw_copy_to_user(to, from, n);
 138         }
 139         return n;
 140 }
 141
 142 static int copyin(void *to, const void __user *from, size_t n)
 143 {
 144         if (access_ok(VERIFY_READ, from, n)) {
 145                 kasan_check_write(to, n);
 146                 n = raw_copy_from_user(to, from, n);
 147         }
 148         return n;
 149 }
 150
 151 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 152                          struct iov_iter *i)
 153 {
 154         size_t skip, copy, left, wanted;
 155         const struct iovec *iov;
 156         char __user *buf;
 157         void *kaddr, *from;
 158
 159         if (unlikely(bytes > i->count))
 160                 bytes = i->count;
 161
 162         if (unlikely(!bytes))
 163                 return 0;
 164
 165         might_fault();
 166         wanted = bytes;
 167         iov = i->iov;
 168         skip = i->iov_offset;
 169         buf = iov->iov_base + skip;
 170         copy = min(bytes, iov->iov_len - skip);
 171
 172         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 173                 kaddr = kmap_atomic(page);
 174                 from = kaddr + offset;
 175
 176                 /* first chunk, usually the only one */
 177                 left = copyout(buf, from, copy);
 178                 copy -= left;
 179                 skip += copy;
 180                 from += copy;
 181                 bytes -= copy;
 182
 183                 while (unlikely(!left && bytes)) {
 184                         iov++;
 185                         buf = iov->iov_base;
 186                         copy = min(bytes, iov->iov_len);
 187                         left = copyout(buf, from, copy);
 188                         copy -= left;
 189                         skip = copy;
 190                         from += copy;
 191                         bytes -= copy;
 192                 }
 193                 if (likely(!bytes)) {
 194                         kunmap_atomic(kaddr);
 195                         goto done;
 196                 }
 197                 offset = from - kaddr;
 198                 buf += copy;
 199                 kunmap_atomic(kaddr);
 200                 copy = min(bytes, iov->iov_len - skip);
 201         }
 202         /* Too bad - revert to non-atomic kmap */
 203
 204         kaddr = kmap(page);
 205         from = kaddr + offset;
 206         left = copyout(buf, from, copy);
 207         copy -= left;
 208         skip += copy;
 209         from += copy;
 210         bytes -= copy;
 211         while (unlikely(!left && bytes)) {
 212                 iov++;
 213                 buf = iov->iov_base;
 214                 copy = min(bytes, iov->iov_len);
 215                 left = copyout(buf, from, copy);
 216                 copy -= left;
 217                 skip = copy;
 218                 from += copy;
 219                 bytes -= copy;
 220         }
 221         kunmap(page);
 222
 223 done:
 224         if (skip == iov->iov_len) {
 225                 iov++;
 226                 skip = 0;
 227         }
 228         i->count -= wanted - bytes;
 229         i->nr_segs -= iov - i->iov;
 230         i->iov = iov;
 231         i->iov_offset = skip;
 232         return wanted - bytes;
 233 }
 234
 235 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 236                          struct iov_iter *i)
 237 {
 238         size_t skip, copy, left, wanted;
 239         const struct iovec *iov;
 240         char __user *buf;
 241         void *kaddr, *to;
 242
 243         if (unlikely(bytes > i->count))
 244                 bytes = i->count;
 245
 246         if (unlikely(!bytes))
 247                 return 0;
 248
 249         might_fault();
 250         wanted = bytes;
 251         iov = i->iov;
 252         skip = i->iov_offset;
 253         buf = iov->iov_base + skip;
 254         copy = min(bytes, iov->iov_len - skip);
 255
 256         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 257                 kaddr = kmap_atomic(page);
 258                 to = kaddr + offset;
 259
 260                 /* first chunk, usually the only one */
 261                 left = copyin(to, buf, copy);
 262                 copy -= left;
 263                 skip += copy;
 264                 to += copy;
 265                 bytes -= copy;
 266
 267                 while (unlikely(!left && bytes)) {
 268                         iov++;
 269                         buf = iov->iov_base;
 270                         copy = min(bytes, iov->iov_len);
 271                         left = copyin(to, buf, copy);
 272                         copy -= left;
 273                         skip = copy;
 274                         to += copy;
 275                         bytes -= copy;
 276                 }
 277                 if (likely(!bytes)) {
 278                         kunmap_atomic(kaddr);
 279                         goto done;
 280                 }
 281                 offset = to - kaddr;
 282                 buf += copy;
 283                 kunmap_atomic(kaddr);
 284                 copy = min(bytes, iov->iov_len - skip);
 285         }
 286         /* Too bad - revert to non-atomic kmap */
 287
 288         kaddr = kmap(page);
 289         to = kaddr + offset;
 290         left = copyin(to, buf, copy);
 291         copy -= left;
 292         skip += copy;
 293         to += copy;
 294         bytes -= copy;
 295         while (unlikely(!left && bytes)) {
 296                 iov++;
 297                 buf = iov->iov_base;
 298                 copy = min(bytes, iov->iov_len);
 299                 left = copyin(to, buf, copy);
 300                 copy -= left;
 301                 skip = copy;
 302                 to += copy;
 303                 bytes -= copy;
 304         }
 305         kunmap(page);
 306
 307 done:
 308         if (skip == iov->iov_len) {
 309                 iov++;
 310                 skip = 0;
 311         }
 312         i->count -= wanted - bytes;
 313         i->nr_segs -= iov - i->iov;
 314         i->iov = iov;
 315         i->iov_offset = skip;
 316         return wanted - bytes;
 317 }
 318
 319 #ifdef PIPE_PARANOIA
 320 static bool sanity(const struct iov_iter *i)
 321 {
 322         struct pipe_inode_info *pipe = i->pipe;
 323         int idx = i->idx;
 324         int next = pipe->curbuf + pipe->nrbufs;
 325         if (i->iov_offset) {
 326                 struct pipe_buffer *p;
 327                 if (unlikely(!pipe->nrbufs))
 328                         goto Bad;       // pipe must be non-empty
 329                 if (unlikely(idx != ((next - 1) & (pipe->buffers - 1))))
 330                         goto Bad;       // must be at the last buffer...
 331
 332                 p = &pipe->bufs[idx];
 333                 if (unlikely(p->offset + p->len != i->iov_offset))
 334                         goto Bad;       // ... at the end of segment
 335         } else {
 336                 if (idx != (next & (pipe->buffers - 1)))
 337                         goto Bad;       // must be right after the last buffer
 338         }
 339         return true;
 340 Bad:
 341         printk(KERN_ERR "idx = %d, offset = %zd\n", i->idx, i->iov_offset);
 342         printk(KERN_ERR "curbuf = %d, nrbufs = %d, buffers = %d\n",
 343                         pipe->curbuf, pipe->nrbufs, pipe->buffers);
 344         for (idx = 0; idx < pipe->buffers; idx++)
 345                 printk(KERN_ERR "[%p %p %d %d]\n",
 346                         pipe->bufs[idx].ops,
 347                         pipe->bufs[idx].page,
 348                         pipe->bufs[idx].offset,
 349                         pipe->bufs[idx].len);
 350         WARN_ON(1);
 351         return false;
 352 }
 353 #else
 354 #define sanity(i) true
 355 #endif
 356
 357 static inline int next_idx(int idx, struct pipe_inode_info *pipe)
 358 {
 359         return (idx + 1) & (pipe->buffers - 1);
 360 }
 361
 362 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 363                          struct iov_iter *i)
 364 {
 365         struct pipe_inode_info *pipe = i->pipe;
 366         struct pipe_buffer *buf;
 367         size_t off;
 368         int idx;
 369
 370         if (unlikely(bytes > i->count))
 371                 bytes = i->count;
 372
 373         if (unlikely(!bytes))
 374                 return 0;
 375
 376         if (!sanity(i))
 377                 return 0;
 378
 379         off = i->iov_offset;
 380         idx = i->idx;
 381         buf = &pipe->bufs[idx];
 382         if (off) {
 383                 if (offset == off && buf->page == page) {
 384                         /* merge with the last one */
 385                         buf->len += bytes;
 386                         i->iov_offset += bytes;
 387                         goto out;
 388                 }
 389                 idx = next_idx(idx, pipe);
 390                 buf = &pipe->bufs[idx];
 391         }
 392         if (idx == pipe->curbuf && pipe->nrbufs)
 393                 return 0;
 394         pipe->nrbufs++;
 395         buf->ops = &page_cache_pipe_buf_ops;
 396         get_page(buf->page = page);
 397         buf->offset = offset;
 398         buf->len = bytes;
 399         i->iov_offset = offset + bytes;
 400         i->idx = idx;
 401 out:
 402         i->count -= bytes;
 403         return bytes;
 404 }
 405
 406 /*
 407  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 408  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 409  *
 410  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 411  * because it is an invalid address).
 412  */
 413 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
 414 {
 415         size_t skip = i->iov_offset;
 416         const struct iovec *iov;
 417         int err;
 418         struct iovec v;
 419
 420         if (!(i->type & (ITER_BVEC|ITER_KVEC))) {
 421                 iterate_iovec(i, bytes, v, iov, skip, ({
 422                         err = fault_in_pages_readable(v.iov_base, v.iov_len);
 423                         if (unlikely(err))
 424                         return err;
 425                 0;}))
 426         }
 427         return 0;
 428 }
 429 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 430
 431 void iov_iter_init(struct iov_iter *i, int direction,
 432                         const struct iovec *iov, unsigned long nr_segs,
 433                         size_t count)
 434 {
 435         /* It will get better.  Eventually... */
 436         if (uaccess_kernel()) {
 437                 direction |= ITER_KVEC;
 438                 i->type = direction;
 439                 i->kvec = (struct kvec *)iov;
 440         } else {
 441                 i->type = direction;
 442                 i->iov = iov;
 443         }
 444         i->nr_segs = nr_segs;
 445         i->iov_offset = 0;
 446         i->count = count;
 447 }
 448 EXPORT_SYMBOL(iov_iter_init);
 449
 450 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
 451 {
 452         char *from = kmap_atomic(page);
 453         memcpy(to, from + offset, len);
 454         kunmap_atomic(from);
 455 }
 456
 457 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
 458 {
 459         char *to = kmap_atomic(page);
 460         memcpy(to + offset, from, len);
 461         kunmap_atomic(to);
 462 }
 463
 464 static void memzero_page(struct page *page, size_t offset, size_t len)
 465 {
 466         char *addr = kmap_atomic(page);
 467         memset(addr + offset, 0, len);
 468         kunmap_atomic(addr);
 469 }
 470
 471 static inline bool allocated(struct pipe_buffer *buf)
 472 {
 473         return buf->ops == &default_pipe_buf_ops;
 474 }
 475
 476 static inline void data_start(const struct iov_iter *i, int *idxp, size_t *offp)
 477 {
 478         size_t off = i->iov_offset;
 479         int idx = i->idx;
 480         if (off && (!allocated(&i->pipe->bufs[idx]) || off == PAGE_SIZE)) {
 481                 idx = next_idx(idx, i->pipe);
 482                 off = 0;
 483         }
 484         *idxp = idx;
 485         *offp = off;
 486 }
 487
 488 static size_t push_pipe(struct iov_iter *i, size_t size,
 489                         int *idxp, size_t *offp)
 490 {
 491         struct pipe_inode_info *pipe = i->pipe;
 492         size_t off;
 493         int idx;
 494         ssize_t left;
 495
 496         if (unlikely(size > i->count))
 497                 size = i->count;
 498         if (unlikely(!size))
 499                 return 0;
 500
 501         left = size;
 502         data_start(i, &idx, &off);
 503         *idxp = idx;
 504         *offp = off;
 505         if (off) {
 506                 left -= PAGE_SIZE - off;
 507                 if (left <= 0) {
 508                         pipe->bufs[idx].len += size;
 509                         return size;
 510                 }
 511                 pipe->bufs[idx].len = PAGE_SIZE;
 512                 idx = next_idx(idx, pipe);
 513         }
 514         while (idx != pipe->curbuf || !pipe->nrbufs) {
 515                 struct page *page = alloc_page(GFP_USER);
 516                 if (!page)
 517                         break;
 518                 pipe->nrbufs++;
 519                 pipe->bufs[idx].ops = &default_pipe_buf_ops;
 520                 pipe->bufs[idx].page = page;
 521                 pipe->bufs[idx].offset = 0;
 522                 if (left <= PAGE_SIZE) {
 523                         pipe->bufs[idx].len = left;
 524                         return size;
 525                 }
 526                 pipe->bufs[idx].len = PAGE_SIZE;
 527                 left -= PAGE_SIZE;
 528                 idx = next_idx(idx, pipe);
 529         }
 530         return size - left;
 531 }
 532
 533 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 534                                 struct iov_iter *i)
 535 {
 536         struct pipe_inode_info *pipe = i->pipe;
 537         size_t n, off;
 538         int idx;
 539
 540         if (!sanity(i))
 541                 return 0;
 542
 543         bytes = n = push_pipe(i, bytes, &idx, &off);
 544         if (unlikely(!n))
 545                 return 0;
 546         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 547                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 548                 memcpy_to_page(pipe->bufs[idx].page, off, addr, chunk);
 549                 i->idx = idx;
 550                 i->iov_offset = off + chunk;
 551                 n -= chunk;
 552                 addr += chunk;
 553         }
 554         i->count -= bytes;
 555         return bytes;
 556 }
 557
 558 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 559 {
 560         const char *from = addr;
 561         if (unlikely(i->type & ITER_PIPE))
 562                 return copy_pipe_to_iter(addr, bytes, i);
 563         if (iter_is_iovec(i))
 564                 might_fault();
 565         iterate_and_advance(i, bytes, v,
 566                 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 567                 memcpy_to_page(v.bv_page, v.bv_offset,
 568                                (from += v.bv_len) - v.bv_len, v.bv_len),
 569                 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
 570         )
 571
 572         return bytes;
 573 }
 574 EXPORT_SYMBOL(_copy_to_iter);
 575
 576 #ifdef CONFIG_ARCH_HAS_UACCESS_MCSAFE
 577 static int copyout_mcsafe(void __user *to, const void *from, size_t n)
 578 {
 579         if (access_ok(VERIFY_WRITE, to, n)) {
 580                 kasan_check_read(from, n);
 581                 n = copy_to_user_mcsafe((__force void *) to, from, n);
 582         }
 583         return n;
 584 }
 585
 586 static unsigned long memcpy_mcsafe_to_page(struct page *page, size_t offset,
 587                 const char *from, size_t len)
 588 {
 589         unsigned long ret;
 590         char *to;
 591
 592         to = kmap_atomic(page);
 593         ret = memcpy_mcsafe(to + offset, from, len);
 594         kunmap_atomic(to);
 595
 596         return ret;
 597 }
 598
 599 static size_t copy_pipe_to_iter_mcsafe(const void *addr, size_t bytes,
 600                                 struct iov_iter *i)
 601 {
 602         struct pipe_inode_info *pipe = i->pipe;
 603         size_t n, off, xfer = 0;
 604         int idx;
 605
 606         if (!sanity(i))
 607                 return 0;
 608
 609         bytes = n = push_pipe(i, bytes, &idx, &off);
 610         if (unlikely(!n))
 611                 return 0;
 612         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 613                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 614                 unsigned long rem;
 615
 616                 rem = memcpy_mcsafe_to_page(pipe->bufs[idx].page, off, addr,
 617                                 chunk);
 618                 i->idx = idx;
 619                 i->iov_offset = off + chunk - rem;
 620                 xfer += chunk - rem;
 621                 if (rem)
 622                         break;
 623                 n -= chunk;
 624                 addr += chunk;
 625         }
 626         i->count -= xfer;
 627         return xfer;
 628 }
 629
 630 /**
 631  * _copy_to_iter_mcsafe - copy to user with source-read error exception handling
 632  * @addr: source kernel address
 633  * @bytes: total transfer length
 634  * @iter: destination iterator
 635  *
 636  * The pmem driver arranges for filesystem-dax to use this facility via
 637  * dax_copy_to_iter() for protecting read/write to persistent memory.
 638  * Unless / until an architecture can guarantee identical performance
 639  * between _copy_to_iter_mcsafe() and _copy_to_iter() it would be a
 640  * performance regression to switch more users to the mcsafe version.
 641  *
 642  * Otherwise, the main differences between this and typical _copy_to_iter().
 643  *
 644  * * Typical tail/residue handling after a fault retries the copy
 645  *   byte-by-byte until the fault happens again. Re-triggering machine
 646  *   checks is potentially fatal so the implementation uses source
 647  *   alignment and poison alignment assumptions to avoid re-triggering
 648  *   hardware exceptions.
 649  *
 650  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 651  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 652  *   a short copy.
 653  *
 654  * See MCSAFE_TEST for self-test.
 655  */
 656 size_t _copy_to_iter_mcsafe(const void *addr, size_t bytes, struct iov_iter *i)
 657 {
 658         const char *from = addr;
 659         unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
 660
 661         if (unlikely(i->type & ITER_PIPE))
 662                 return copy_pipe_to_iter_mcsafe(addr, bytes, i);
 663         if (iter_is_iovec(i))
 664                 might_fault();
 665         iterate_and_advance(i, bytes, v,
 666                 copyout_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 667                 ({
 668                 rem = memcpy_mcsafe_to_page(v.bv_page, v.bv_offset,
 669                                (from += v.bv_len) - v.bv_len, v.bv_len);
 670                 if (rem) {
 671                         curr_addr = (unsigned long) from;
 672                         bytes = curr_addr - s_addr - rem;
 673                         return bytes;
 674                 }
 675                 }),
 676                 ({
 677                 rem = memcpy_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len,
 678                                 v.iov_len);
 679                 if (rem) {
 680                         curr_addr = (unsigned long) from;
 681                         bytes = curr_addr - s_addr - rem;
 682                         return bytes;
 683                 }
 684                 })
 685         )
 686
 687         return bytes;
 688 }
 689 EXPORT_SYMBOL_GPL(_copy_to_iter_mcsafe);
 690 #endif /* CONFIG_ARCH_HAS_UACCESS_MCSAFE */
 691
 692 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 693 {
 694         char *to = addr;
 695         if (unlikely(i->type & ITER_PIPE)) {
 696                 WARN_ON(1);
 697                 return 0;
 698         }
 699         if (iter_is_iovec(i))
 700                 might_fault();
 701         iterate_and_advance(i, bytes, v,
 702                 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 703                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 704                                  v.bv_offset, v.bv_len),
 705                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 706         )
 707
 708         return bytes;
 709 }
 710 EXPORT_SYMBOL(_copy_from_iter);
 711
 712 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
 713 {
 714         char *to = addr;
 715         if (unlikely(i->type & ITER_PIPE)) {
 716                 WARN_ON(1);
 717                 return false;
 718         }
 719         if (unlikely(i->count < bytes))
 720                 return false;
 721
 722         if (iter_is_iovec(i))
 723                 might_fault();
 724         iterate_all_kinds(i, bytes, v, ({
 725                 if (copyin((to += v.iov_len) - v.iov_len,
 726                                       v.iov_base, v.iov_len))
 727                         return false;
 728                 0;}),
 729                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 730                                  v.bv_offset, v.bv_len),
 731                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 732         )
 733
 734         iov_iter_advance(i, bytes);
 735         return true;
 736 }
 737 EXPORT_SYMBOL(_copy_from_iter_full);
 738
 739 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 740 {
 741         char *to = addr;
 742         if (unlikely(i->type & ITER_PIPE)) {
 743                 WARN_ON(1);
 744                 return 0;
 745         }
 746         iterate_and_advance(i, bytes, v,
 747                 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 748                                          v.iov_base, v.iov_len),
 749                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 750                                  v.bv_offset, v.bv_len),
 751                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 752         )
 753
 754         return bytes;
 755 }
 756 EXPORT_SYMBOL(_copy_from_iter_nocache);
 757
 758 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 759 /**
 760  * _copy_from_iter_flushcache - write destination through cpu cache
 761  * @addr: destination kernel address
 762  * @bytes: total transfer length
 763  * @iter: source iterator
 764  *
 765  * The pmem driver arranges for filesystem-dax to use this facility via
 766  * dax_copy_from_iter() for ensuring that writes to persistent memory
 767  * are flushed through the CPU cache. It is differentiated from
 768  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 769  * all iterator types. The _copy_from_iter_nocache() only attempts to
 770  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 771  * instructions that strand dirty-data in the cache.
 772  */
 773 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 774 {
 775         char *to = addr;
 776         if (unlikely(i->type & ITER_PIPE)) {
 777                 WARN_ON(1);
 778                 return 0;
 779         }
 780         iterate_and_advance(i, bytes, v,
 781                 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
 782                                          v.iov_base, v.iov_len),
 783                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 784                                  v.bv_offset, v.bv_len),
 785                 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
 786                         v.iov_len)
 787         )
 788
 789         return bytes;
 790 }
 791 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 792 #endif
 793
 794 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
 795 {
 796         char *to = addr;
 797         if (unlikely(i->type & ITER_PIPE)) {
 798                 WARN_ON(1);
 799                 return false;
 800         }
 801         if (unlikely(i->count < bytes))
 802                 return false;
 803         iterate_all_kinds(i, bytes, v, ({
 804                 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 805                                              v.iov_base, v.iov_len))
 806                         return false;
 807                 0;}),
 808                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 809                                  v.bv_offset, v.bv_len),
 810                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 811         )
 812
 813         iov_iter_advance(i, bytes);
 814         return true;
 815 }
 816 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
 817
 818 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 819 {
 820         struct page *head;
 821         size_t v = n + offset;
 822
 823         /*
 824          * The general case needs to access the page order in order
 825          * to compute the page size.
 826          * However, we mostly deal with order-0 pages and thus can
 827          * avoid a possible cache line miss for requests that fit all
 828          * page orders.
 829          */
 830         if (n <= v && v <= PAGE_SIZE)
 831                 return true;
 832
 833         head = compound_head(page);
 834         v += (page - head) << PAGE_SHIFT;
 835
 836         if (likely(n <= v && v <= (PAGE_SIZE << compound_order(head))))
 837                 return true;
 838         WARN_ON(1);
 839         return false;
 840 }
 841
 842 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 843                          struct iov_iter *i)
 844 {
 845         if (unlikely(!page_copy_sane(page, offset, bytes)))
 846                 return 0;
 847         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 848                 void *kaddr = kmap_atomic(page);
 849                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 850                 kunmap_atomic(kaddr);
 851                 return wanted;
 852         } else if (likely(!(i->type & ITER_PIPE)))
 853                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 854         else
 855                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 856 }
 857 EXPORT_SYMBOL(copy_page_to_iter);
 858
 859 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 860                          struct iov_iter *i)
 861 {
 862         if (unlikely(!page_copy_sane(page, offset, bytes)))
 863                 return 0;
 864         if (unlikely(i->type & ITER_PIPE)) {
 865                 WARN_ON(1);
 866                 return 0;
 867         }
 868         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 869                 void *kaddr = kmap_atomic(page);
 870                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 871                 kunmap_atomic(kaddr);
 872                 return wanted;
 873         } else
 874                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 875 }
 876 EXPORT_SYMBOL(copy_page_from_iter);
 877
 878 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 879 {
 880         struct pipe_inode_info *pipe = i->pipe;
 881         size_t n, off;
 882         int idx;
 883
 884         if (!sanity(i))
 885                 return 0;
 886
 887         bytes = n = push_pipe(i, bytes, &idx, &off);
 888         if (unlikely(!n))
 889                 return 0;
 890
 891         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 892                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 893                 memzero_page(pipe->bufs[idx].page, off, chunk);
 894                 i->idx = idx;
 895                 i->iov_offset = off + chunk;
 896                 n -= chunk;
 897         }
 898         i->count -= bytes;
 899         return bytes;
 900 }
 901
 902 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 903 {
 904         if (unlikely(i->type & ITER_PIPE))
 905                 return pipe_zero(bytes, i);
 906         iterate_and_advance(i, bytes, v,
 907                 clear_user(v.iov_base, v.iov_len),
 908                 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
 909                 memset(v.iov_base, 0, v.iov_len)
 910         )
 911
 912         return bytes;
 913 }
 914 EXPORT_SYMBOL(iov_iter_zero);
 915
 916 size_t iov_iter_copy_from_user_atomic(struct page *page,
 917                 struct iov_iter *i, unsigned long offset, size_t bytes)
 918 {
 919         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 920         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 921                 kunmap_atomic(kaddr);
 922                 return 0;
 923         }
 924         if (unlikely(i->type & ITER_PIPE)) {
 925                 kunmap_atomic(kaddr);
 926                 WARN_ON(1);
 927                 return 0;
 928         }
 929         iterate_all_kinds(i, bytes, v,
 930                 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 931                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
 932                                  v.bv_offset, v.bv_len),
 933                 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 934         )
 935         kunmap_atomic(kaddr);
 936         return bytes;
 937 }
 938 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
 939
 940 static inline void pipe_truncate(struct iov_iter *i)
 941 {
 942         struct pipe_inode_info *pipe = i->pipe;
 943         if (pipe->nrbufs) {
 944                 size_t off = i->iov_offset;
 945                 int idx = i->idx;
 946                 int nrbufs = (idx - pipe->curbuf) & (pipe->buffers - 1);
 947                 if (off) {
 948                         pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
 949                         idx = next_idx(idx, pipe);
 950                         nrbufs++;
 951                 }
 952                 while (pipe->nrbufs > nrbufs) {
 953                         pipe_buf_release(pipe, &pipe->bufs[idx]);
 954                         idx = next_idx(idx, pipe);
 955                         pipe->nrbufs--;
 956                 }
 957         }
 958 }
 959
 960 static void pipe_advance(struct iov_iter *i, size_t size)
 961 {
 962         struct pipe_inode_info *pipe = i->pipe;
 963         if (unlikely(i->count < size))
 964                 size = i->count;
 965         if (size) {
 966                 struct pipe_buffer *buf;
 967                 size_t off = i->iov_offset, left = size;
 968                 int idx = i->idx;
 969                 if (off) /* make it relative to the beginning of buffer */
 970                         left += off - pipe->bufs[idx].offset;
 971                 while (1) {
 972                         buf = &pipe->bufs[idx];
 973                         if (left <= buf->len)
 974                                 break;
 975                         left -= buf->len;
 976                         idx = next_idx(idx, pipe);
 977                 }
 978                 i->idx = idx;
 979                 i->iov_offset = buf->offset + left;
 980         }
 981         i->count -= size;
 982         /* ... and discard everything past that point */
 983         pipe_truncate(i);
 984 }
 985
 986 void iov_iter_advance(struct iov_iter *i, size_t size)
 987 {
 988         if (unlikely(i->type & ITER_PIPE)) {
 989                 pipe_advance(i, size);
 990                 return;
 991         }
 992         iterate_and_advance(i, size, v, 0, 0, 0)
 993 }
 994 EXPORT_SYMBOL(iov_iter_advance);
 995
 996 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 997 {
 998         if (!unroll)
 999                 return;
1000         if (WARN_ON(unroll > MAX_RW_COUNT))
1001                 return;
1002         i->count += unroll;
1003         if (unlikely(i->type & ITER_PIPE)) {
1004                 struct pipe_inode_info *pipe = i->pipe;
1005                 int idx = i->idx;
1006                 size_t off = i->iov_offset;
1007                 while (1) {
1008                         size_t n = off - pipe->bufs[idx].offset;
1009                         if (unroll < n) {
1010                                 off -= unroll;
1011                                 break;
1012                         }
1013                         unroll -= n;
1014                         if (!unroll && idx == i->start_idx) {
1015                                 off = 0;
1016                                 break;
1017                         }
1018                         if (!idx--)
1019                                 idx = pipe->buffers - 1;
1020                         off = pipe->bufs[idx].offset + pipe->bufs[idx].len;
1021                 }
1022                 i->iov_offset = off;
1023                 i->idx = idx;
1024                 pipe_truncate(i);
1025                 return;
1026         }
1027         if (unroll <= i->iov_offset) {
1028                 i->iov_offset -= unroll;
1029                 return;
1030         }
1031         unroll -= i->iov_offset;
1032         if (i->type & ITER_BVEC) {
1033                 const struct bio_vec *bvec = i->bvec;
1034                 while (1) {
1035                         size_t n = (--bvec)->bv_len;
1036                         i->nr_segs++;
1037                         if (unroll <= n) {
1038                                 i->bvec = bvec;
1039                                 i->iov_offset = n - unroll;
1040                                 return;
1041                         }
1042                         unroll -= n;
1043                 }
1044         } else { /* same logics for iovec and kvec */
1045                 const struct iovec *iov = i->iov;
1046                 while (1) {
1047                         size_t n = (--iov)->iov_len;
1048                         i->nr_segs++;
1049                         if (unroll <= n) {
1050                                 i->iov = iov;
1051                                 i->iov_offset = n - unroll;
1052                                 return;
1053                         }
1054                         unroll -= n;
1055                 }
1056         }
1057 }
1058 EXPORT_SYMBOL(iov_iter_revert);
1059
1060 /*
1061  * Return the count of just the current iov_iter segment.
1062  */
1063 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1064 {
1065         if (unlikely(i->type & ITER_PIPE))
1066                 return i->count;        // it is a silly place, anyway
1067         if (i->nr_segs == 1)
1068                 return i->count;
1069         else if (i->type & ITER_BVEC)
1070                 return min(i->count, i->bvec->bv_len - i->iov_offset);
1071         else
1072                 return min(i->count, i->iov->iov_len - i->iov_offset);
1073 }
1074 EXPORT_SYMBOL(iov_iter_single_seg_count);
1075
1076 void iov_iter_kvec(struct iov_iter *i, int direction,
1077                         const struct kvec *kvec, unsigned long nr_segs,
1078                         size_t count)
1079 {
1080         BUG_ON(!(direction & ITER_KVEC));
1081         i->type = direction;
1082         i->kvec = kvec;
1083         i->nr_segs = nr_segs;
1084         i->iov_offset = 0;
1085         i->count = count;
1086 }
1087 EXPORT_SYMBOL(iov_iter_kvec);
1088
1089 void iov_iter_bvec(struct iov_iter *i, int direction,
1090                         const struct bio_vec *bvec, unsigned long nr_segs,
1091                         size_t count)
1092 {
1093         BUG_ON(!(direction & ITER_BVEC));
1094         i->type = direction;
1095         i->bvec = bvec;
1096         i->nr_segs = nr_segs;
1097         i->iov_offset = 0;
1098         i->count = count;
1099 }
1100 EXPORT_SYMBOL(iov_iter_bvec);
1101
1102 void iov_iter_pipe(struct iov_iter *i, int direction,
1103                         struct pipe_inode_info *pipe,
1104                         size_t count)
1105 {
1106         BUG_ON(direction != ITER_PIPE);
1107         WARN_ON(pipe->nrbufs == pipe->buffers);
1108         i->type = direction;
1109         i->pipe = pipe;
1110         i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1111         i->iov_offset = 0;
1112         i->count = count;
1113         i->start_idx = i->idx;
1114 }
1115 EXPORT_SYMBOL(iov_iter_pipe);
1116
1117 unsigned long iov_iter_alignment(const struct iov_iter *i)
1118 {
1119         unsigned long res = 0;
1120         size_t size = i->count;
1121
1122         if (unlikely(i->type & ITER_PIPE)) {
1123                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
1124                         return size | i->iov_offset;
1125                 return size;
1126         }
1127         iterate_all_kinds(i, size, v,
1128                 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1129                 res |= v.bv_offset | v.bv_len,
1130                 res |= (unsigned long)v.iov_base | v.iov_len
1131         )
1132         return res;
1133 }
1134 EXPORT_SYMBOL(iov_iter_alignment);
1135
1136 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1137 {
1138         unsigned long res = 0;
1139         size_t size = i->count;
1140
1141         if (unlikely(i->type & ITER_PIPE)) {
1142                 WARN_ON(1);
1143                 return ~0U;
1144         }
1145
1146         iterate_all_kinds(i, size, v,
1147                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1148                         (size != v.iov_len ? size : 0), 0),
1149                 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1150                         (size != v.bv_len ? size : 0)),
1151                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1152                         (size != v.iov_len ? size : 0))
1153                 );
1154         return res;
1155 }
1156 EXPORT_SYMBOL(iov_iter_gap_alignment);
1157
1158 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1159                                 size_t maxsize,
1160                                 struct page **pages,
1161                                 int idx,
1162                                 size_t *start)
1163 {
1164         struct pipe_inode_info *pipe = i->pipe;
1165         ssize_t n = push_pipe(i, maxsize, &idx, start);
1166         if (!n)
1167                 return -EFAULT;
1168
1169         maxsize = n;
1170         n += *start;
1171         while (n > 0) {
1172                 get_page(*pages++ = pipe->bufs[idx].page);
1173                 idx = next_idx(idx, pipe);
1174                 n -= PAGE_SIZE;
1175         }
1176
1177         return maxsize;
1178 }
1179
1180 static ssize_t pipe_get_pages(struct iov_iter *i,
1181                    struct page **pages, size_t maxsize, unsigned maxpages,
1182                    size_t *start)
1183 {
1184         unsigned npages;
1185         size_t capacity;
1186         int idx;
1187
1188         if (!maxsize)
1189                 return 0;
1190
1191         if (!sanity(i))
1192                 return -EFAULT;
1193
1194         data_start(i, &idx, start);
1195         /* some of this one + all after this one */
1196         npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1197         capacity = min(npages,maxpages) * PAGE_SIZE - *start;
1198
1199         return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
1200 }
1201
1202 ssize_t iov_iter_get_pages(struct iov_iter *i,
1203                    struct page **pages, size_t maxsize, unsigned maxpages,
1204                    size_t *start)
1205 {
1206         if (maxsize > i->count)
1207                 maxsize = i->count;
1208
1209         if (unlikely(i->type & ITER_PIPE))
1210                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1211         iterate_all_kinds(i, maxsize, v, ({
1212                 unsigned long addr = (unsigned long)v.iov_base;
1213                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1214                 int n;
1215                 int res;
1216
1217                 if (len > maxpages * PAGE_SIZE)
1218                         len = maxpages * PAGE_SIZE;
1219                 addr &= ~(PAGE_SIZE - 1);
1220                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1221                 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages);
1222                 if (unlikely(res < 0))
1223                         return res;
1224                 return (res == n ? len : res * PAGE_SIZE) - *start;
1225         0;}),({
1226                 /* can't be more than PAGE_SIZE */
1227                 *start = v.bv_offset;
1228                 get_page(*pages = v.bv_page);
1229                 return v.bv_len;
1230         }),({
1231                 return -EFAULT;
1232         })
1233         )
1234         return 0;
1235 }
1236 EXPORT_SYMBOL(iov_iter_get_pages);
1237
1238 static struct page **get_pages_array(size_t n)
1239 {
1240         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1241 }
1242
1243 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1244                    struct page ***pages, size_t maxsize,
1245                    size_t *start)
1246 {
1247         struct page **p;
1248         ssize_t n;
1249         int idx;
1250         int npages;
1251
1252         if (!maxsize)
1253                 return 0;
1254
1255         if (!sanity(i))
1256                 return -EFAULT;
1257
1258         data_start(i, &idx, start);
1259         /* some of this one + all after this one */
1260         npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1261         n = npages * PAGE_SIZE - *start;
1262         if (maxsize > n)
1263                 maxsize = n;
1264         else
1265                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1266         p = get_pages_array(npages);
1267         if (!p)
1268                 return -ENOMEM;
1269         n = __pipe_get_pages(i, maxsize, p, idx, start);
1270         if (n > 0)
1271                 *pages = p;
1272         else
1273                 kvfree(p);
1274         return n;
1275 }
1276
1277 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1278                    struct page ***pages, size_t maxsize,
1279                    size_t *start)
1280 {
1281         struct page **p;
1282
1283         if (maxsize > i->count)
1284                 maxsize = i->count;
1285
1286         if (unlikely(i->type & ITER_PIPE))
1287                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1288         iterate_all_kinds(i, maxsize, v, ({
1289                 unsigned long addr = (unsigned long)v.iov_base;
1290                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1291                 int n;
1292                 int res;
1293
1294                 addr &= ~(PAGE_SIZE - 1);
1295                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1296                 p = get_pages_array(n);
1297                 if (!p)
1298                         return -ENOMEM;
1299                 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p);
1300                 if (unlikely(res < 0)) {
1301                         kvfree(p);
1302                         return res;
1303                 }
1304                 *pages = p;
1305                 return (res == n ? len : res * PAGE_SIZE) - *start;
1306         0;}),({
1307                 /* can't be more than PAGE_SIZE */
1308                 *start = v.bv_offset;
1309                 *pages = p = get_pages_array(1);
1310                 if (!p)
1311                         return -ENOMEM;
1312                 get_page(*p = v.bv_page);
1313                 return v.bv_len;
1314         }),({
1315                 return -EFAULT;
1316         })
1317         )
1318         return 0;
1319 }
1320 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1321
1322 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1323                                struct iov_iter *i)
1324 {
1325         char *to = addr;
1326         __wsum sum, next;
1327         size_t off = 0;
1328         sum = *csum;
1329         if (unlikely(i->type & ITER_PIPE)) {
1330                 WARN_ON(1);
1331                 return 0;
1332         }
1333         iterate_and_advance(i, bytes, v, ({
1334                 int err = 0;
1335                 next = csum_and_copy_from_user(v.iov_base,
1336                                                (to += v.iov_len) - v.iov_len,
1337                                                v.iov_len, 0, &err);
1338                 if (!err) {
1339                         sum = csum_block_add(sum, next, off);
1340                         off += v.iov_len;
1341                 }
1342                 err ? v.iov_len : 0;
1343         }), ({
1344                 char *p = kmap_atomic(v.bv_page);
1345                 next = csum_partial_copy_nocheck(p + v.bv_offset,
1346                                                  (to += v.bv_len) - v.bv_len,
1347                                                  v.bv_len, 0);
1348                 kunmap_atomic(p);
1349                 sum = csum_block_add(sum, next, off);
1350                 off += v.bv_len;
1351         }),({
1352                 next = csum_partial_copy_nocheck(v.iov_base,
1353                                                  (to += v.iov_len) - v.iov_len,
1354                                                  v.iov_len, 0);
1355                 sum = csum_block_add(sum, next, off);
1356                 off += v.iov_len;
1357         })
1358         )
1359         *csum = sum;
1360         return bytes;
1361 }
1362 EXPORT_SYMBOL(csum_and_copy_from_iter);
1363
1364 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1365                                struct iov_iter *i)
1366 {
1367         char *to = addr;
1368         __wsum sum, next;
1369         size_t off = 0;
1370         sum = *csum;
1371         if (unlikely(i->type & ITER_PIPE)) {
1372                 WARN_ON(1);
1373                 return false;
1374         }
1375         if (unlikely(i->count < bytes))
1376                 return false;
1377         iterate_all_kinds(i, bytes, v, ({
1378                 int err = 0;
1379                 next = csum_and_copy_from_user(v.iov_base,
1380                                                (to += v.iov_len) - v.iov_len,
1381                                                v.iov_len, 0, &err);
1382                 if (err)
1383                         return false;
1384                 sum = csum_block_add(sum, next, off);
1385                 off += v.iov_len;
1386                 0;
1387         }), ({
1388                 char *p = kmap_atomic(v.bv_page);
1389                 next = csum_partial_copy_nocheck(p + v.bv_offset,
1390                                                  (to += v.bv_len) - v.bv_len,
1391                                                  v.bv_len, 0);
1392                 kunmap_atomic(p);
1393                 sum = csum_block_add(sum, next, off);
1394                 off += v.bv_len;
1395         }),({
1396                 next = csum_partial_copy_nocheck(v.iov_base,
1397                                                  (to += v.iov_len) - v.iov_len,
1398                                                  v.iov_len, 0);
1399                 sum = csum_block_add(sum, next, off);
1400                 off += v.iov_len;
1401         })
1402         )
1403         *csum = sum;
1404         iov_iter_advance(i, bytes);
1405         return true;
1406 }
1407 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1408
1409 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, __wsum *csum,
1410                              struct iov_iter *i)
1411 {
1412         const char *from = addr;
1413         __wsum sum, next;
1414         size_t off = 0;
1415         sum = *csum;
1416         if (unlikely(i->type & ITER_PIPE)) {
1417                 WARN_ON(1);     /* for now */
1418                 return 0;
1419         }
1420         iterate_and_advance(i, bytes, v, ({
1421                 int err = 0;
1422                 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1423                                              v.iov_base,
1424                                              v.iov_len, 0, &err);
1425                 if (!err) {
1426                         sum = csum_block_add(sum, next, off);
1427                         off += v.iov_len;
1428                 }
1429                 err ? v.iov_len : 0;
1430         }), ({
1431                 char *p = kmap_atomic(v.bv_page);
1432                 next = csum_partial_copy_nocheck((from += v.bv_len) - v.bv_len,
1433                                                  p + v.bv_offset,
1434                                                  v.bv_len, 0);
1435                 kunmap_atomic(p);
1436                 sum = csum_block_add(sum, next, off);
1437                 off += v.bv_len;
1438         }),({
1439                 next = csum_partial_copy_nocheck((from += v.iov_len) - v.iov_len,
1440                                                  v.iov_base,
1441                                                  v.iov_len, 0);
1442                 sum = csum_block_add(sum, next, off);
1443                 off += v.iov_len;
1444         })
1445         )
1446         *csum = sum;
1447         return bytes;
1448 }
1449 EXPORT_SYMBOL(csum_and_copy_to_iter);
1450
1451 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1452 {
1453         size_t size = i->count;
1454         int npages = 0;
1455
1456         if (!size)
1457                 return 0;
1458
1459         if (unlikely(i->type & ITER_PIPE)) {
1460                 struct pipe_inode_info *pipe = i->pipe;
1461                 size_t off;
1462                 int idx;
1463
1464                 if (!sanity(i))
1465                         return 0;
1466
1467                 data_start(i, &idx, &off);
1468                 /* some of this one + all after this one */
1469                 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1470                 if (npages >= maxpages)
1471                         return maxpages;
1472         } else iterate_all_kinds(i, size, v, ({
1473                 unsigned long p = (unsigned long)v.iov_base;
1474                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1475                         - p / PAGE_SIZE;
1476                 if (npages >= maxpages)
1477                         return maxpages;
1478         0;}),({
1479                 npages++;
1480                 if (npages >= maxpages)
1481                         return maxpages;
1482         }),({
1483                 unsigned long p = (unsigned long)v.iov_base;
1484                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1485                         - p / PAGE_SIZE;
1486                 if (npages >= maxpages)
1487                         return maxpages;
1488         })
1489         )
1490         return npages;
1491 }
1492 EXPORT_SYMBOL(iov_iter_npages);
1493
1494 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1495 {
1496         *new = *old;
1497         if (unlikely(new->type & ITER_PIPE)) {
1498                 WARN_ON(1);
1499                 return NULL;
1500         }
1501         if (new->type & ITER_BVEC)
1502                 return new->bvec = kmemdup(new->bvec,
1503                                     new->nr_segs * sizeof(struct bio_vec),
1504                                     flags);
1505         else
1506                 /* iovec and kvec have identical layout */
1507                 return new->iov = kmemdup(new->iov,
1508                                    new->nr_segs * sizeof(struct iovec),
1509                                    flags);
1510 }
1511 EXPORT_SYMBOL(dup_iter);
1512
1513 /**
1514  * import_iovec() - Copy an array of &struct iovec from userspace
1515  *     into the kernel, check that it is valid, and initialize a new
1516  *     &struct iov_iter iterator to access it.
1517  *
1518  * @type: One of %READ or %WRITE.
1519  * @uvector: Pointer to the userspace array.
1520  * @nr_segs: Number of elements in userspace array.
1521  * @fast_segs: Number of elements in @iov.
1522  * @iov: (input and output parameter) Pointer to pointer to (usually small
1523  *     on-stack) kernel array.
1524  * @i: Pointer to iterator that will be initialized on success.
1525  *
1526  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1527  * then this function places %NULL in *@iov on return. Otherwise, a new
1528  * array will be allocated and the result placed in *@iov. This means that
1529  * the caller may call kfree() on *@iov regardless of whether the small
1530  * on-stack array was used or not (and regardless of whether this function
1531  * returns an error or not).
1532  *
1533  * Return: 0 on success or negative error code on error.
1534  */
1535 int import_iovec(int type, const struct iovec __user * uvector,
1536                  unsigned nr_segs, unsigned fast_segs,
1537                  struct iovec **iov, struct iov_iter *i)
1538 {
1539         ssize_t n;
1540         struct iovec *p;
1541         n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1542                                   *iov, &p);
1543         if (n < 0) {
1544                 if (p != *iov)
1545                         kfree(p);
1546                 *iov = NULL;
1547                 return n;
1548         }
1549         iov_iter_init(i, type, p, nr_segs, n);
1550         *iov = p == *iov ? NULL : p;
1551         return 0;
1552 }
1553 EXPORT_SYMBOL(import_iovec);
1554
1555 #ifdef CONFIG_COMPAT
1556 #include <linux/compat.h>
1557
1558 int compat_import_iovec(int type, const struct compat_iovec __user * uvector,
1559                  unsigned nr_segs, unsigned fast_segs,
1560                  struct iovec **iov, struct iov_iter *i)
1561 {
1562         ssize_t n;
1563         struct iovec *p;
1564         n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1565                                   *iov, &p);
1566         if (n < 0) {
1567                 if (p != *iov)
1568                         kfree(p);
1569                 *iov = NULL;
1570                 return n;
1571         }
1572         iov_iter_init(i, type, p, nr_segs, n);
1573         *iov = p == *iov ? NULL : p;
1574         return 0;
1575 }
1576 #endif
1577
1578 int import_single_range(int rw, void __user *buf, size_t len,
1579                  struct iovec *iov, struct iov_iter *i)
1580 {
1581         if (len > MAX_RW_COUNT)
1582                 len = MAX_RW_COUNT;
1583         if (unlikely(!access_ok(!rw, buf, len)))
1584                 return -EFAULT;
1585
1586         iov->iov_base = buf;
1587         iov->iov_len = len;
1588         iov_iter_init(i, rw, iov, 1, len);
1589         return 0;
1590 }
1591 EXPORT_SYMBOL(import_single_range);
1592
1593 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1594                             int (*f)(struct kvec *vec, void *context),
1595                             void *context)
1596 {
1597         struct kvec w;
1598         int err = -EINVAL;
1599         if (!bytes)
1600                 return 0;
1601
1602         iterate_all_kinds(i, bytes, v, -EINVAL, ({
1603                 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1604                 w.iov_len = v.bv_len;
1605                 err = f(&w, context);
1606                 kunmap(v.bv_page);
1607                 err;}), ({
1608                 w = v;
1609                 err = f(&w, context);})
1610         )
1611         return err;
1612 }
1613 EXPORT_SYMBOL(iov_iter_for_each_range);