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 if (unlikely(i->type & ITER_DISCARD)) {  \
  87                 } else {                                        \
  88                         const struct iovec *iov;                \
  89                         struct iovec v;                         \
  90                         iterate_iovec(i, n, v, iov, skip, (I))  \
  91                 }                                               \
  92         }                                                       \
  93 }
  94
  95 #define iterate_and_advance(i, n, v, I, B, K) {                 \
  96         if (unlikely(i->count < n))                             \
  97                 n = i->count;                                   \
  98         if (i->count) {                                         \
  99                 size_t skip = i->iov_offset;                    \
 100                 if (unlikely(i->type & ITER_BVEC)) {            \
 101                         const struct bio_vec *bvec = i->bvec;   \
 102                         struct bio_vec v;                       \
 103                         struct bvec_iter __bi;                  \
 104                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 105                         i->bvec = __bvec_iter_bvec(i->bvec, __bi);      \
 106                         i->nr_segs -= i->bvec - bvec;           \
 107                         skip = __bi.bi_bvec_done;               \
 108                 } else if (unlikely(i->type & ITER_KVEC)) {     \
 109                         const struct kvec *kvec;                \
 110                         struct kvec v;                          \
 111                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 112                         if (skip == kvec->iov_len) {            \
 113                                 kvec++;                         \
 114                                 skip = 0;                       \
 115                         }                                       \
 116                         i->nr_segs -= kvec - i->kvec;           \
 117                         i->kvec = kvec;                         \
 118                 } else if (unlikely(i->type & ITER_DISCARD)) {  \
 119                         skip += n;                              \
 120                 } else {                                        \
 121                         const struct iovec *iov;                \
 122                         struct iovec v;                         \
 123                         iterate_iovec(i, n, v, iov, skip, (I))  \
 124                         if (skip == iov->iov_len) {             \
 125                                 iov++;                          \
 126                                 skip = 0;                       \
 127                         }                                       \
 128                         i->nr_segs -= iov - i->iov;             \
 129                         i->iov = iov;                           \
 130                 }                                               \
 131                 i->count -= n;                                  \
 132                 i->iov_offset = skip;                           \
 133         }                                                       \
 134 }
 135
 136 static int copyout(void __user *to, const void *from, size_t n)
 137 {
 138         if (access_ok(VERIFY_WRITE, to, n)) {
 139                 kasan_check_read(from, n);
 140                 n = raw_copy_to_user(to, from, n);
 141         }
 142         return n;
 143 }
 144
 145 static int copyin(void *to, const void __user *from, size_t n)
 146 {
 147         if (access_ok(VERIFY_READ, from, n)) {
 148                 kasan_check_write(to, n);
 149                 n = raw_copy_from_user(to, from, n);
 150         }
 151         return n;
 152 }
 153
 154 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 155                          struct iov_iter *i)
 156 {
 157         size_t skip, copy, left, wanted;
 158         const struct iovec *iov;
 159         char __user *buf;
 160         void *kaddr, *from;
 161
 162         if (unlikely(bytes > i->count))
 163                 bytes = i->count;
 164
 165         if (unlikely(!bytes))
 166                 return 0;
 167
 168         might_fault();
 169         wanted = bytes;
 170         iov = i->iov;
 171         skip = i->iov_offset;
 172         buf = iov->iov_base + skip;
 173         copy = min(bytes, iov->iov_len - skip);
 174
 175         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 176                 kaddr = kmap_atomic(page);
 177                 from = kaddr + offset;
 178
 179                 /* first chunk, usually the only one */
 180                 left = copyout(buf, from, copy);
 181                 copy -= left;
 182                 skip += copy;
 183                 from += copy;
 184                 bytes -= copy;
 185
 186                 while (unlikely(!left && bytes)) {
 187                         iov++;
 188                         buf = iov->iov_base;
 189                         copy = min(bytes, iov->iov_len);
 190                         left = copyout(buf, from, copy);
 191                         copy -= left;
 192                         skip = copy;
 193                         from += copy;
 194                         bytes -= copy;
 195                 }
 196                 if (likely(!bytes)) {
 197                         kunmap_atomic(kaddr);
 198                         goto done;
 199                 }
 200                 offset = from - kaddr;
 201                 buf += copy;
 202                 kunmap_atomic(kaddr);
 203                 copy = min(bytes, iov->iov_len - skip);
 204         }
 205         /* Too bad - revert to non-atomic kmap */
 206
 207         kaddr = kmap(page);
 208         from = kaddr + offset;
 209         left = copyout(buf, from, copy);
 210         copy -= left;
 211         skip += copy;
 212         from += copy;
 213         bytes -= copy;
 214         while (unlikely(!left && bytes)) {
 215                 iov++;
 216                 buf = iov->iov_base;
 217                 copy = min(bytes, iov->iov_len);
 218                 left = copyout(buf, from, copy);
 219                 copy -= left;
 220                 skip = copy;
 221                 from += copy;
 222                 bytes -= copy;
 223         }
 224         kunmap(page);
 225
 226 done:
 227         if (skip == iov->iov_len) {
 228                 iov++;
 229                 skip = 0;
 230         }
 231         i->count -= wanted - bytes;
 232         i->nr_segs -= iov - i->iov;
 233         i->iov = iov;
 234         i->iov_offset = skip;
 235         return wanted - bytes;
 236 }
 237
 238 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 239                          struct iov_iter *i)
 240 {
 241         size_t skip, copy, left, wanted;
 242         const struct iovec *iov;
 243         char __user *buf;
 244         void *kaddr, *to;
 245
 246         if (unlikely(bytes > i->count))
 247                 bytes = i->count;
 248
 249         if (unlikely(!bytes))
 250                 return 0;
 251
 252         might_fault();
 253         wanted = bytes;
 254         iov = i->iov;
 255         skip = i->iov_offset;
 256         buf = iov->iov_base + skip;
 257         copy = min(bytes, iov->iov_len - skip);
 258
 259         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 260                 kaddr = kmap_atomic(page);
 261                 to = kaddr + offset;
 262
 263                 /* first chunk, usually the only one */
 264                 left = copyin(to, buf, copy);
 265                 copy -= left;
 266                 skip += copy;
 267                 to += copy;
 268                 bytes -= copy;
 269
 270                 while (unlikely(!left && bytes)) {
 271                         iov++;
 272                         buf = iov->iov_base;
 273                         copy = min(bytes, iov->iov_len);
 274                         left = copyin(to, buf, copy);
 275                         copy -= left;
 276                         skip = copy;
 277                         to += copy;
 278                         bytes -= copy;
 279                 }
 280                 if (likely(!bytes)) {
 281                         kunmap_atomic(kaddr);
 282                         goto done;
 283                 }
 284                 offset = to - kaddr;
 285                 buf += copy;
 286                 kunmap_atomic(kaddr);
 287                 copy = min(bytes, iov->iov_len - skip);
 288         }
 289         /* Too bad - revert to non-atomic kmap */
 290
 291         kaddr = kmap(page);
 292         to = kaddr + offset;
 293         left = copyin(to, buf, copy);
 294         copy -= left;
 295         skip += copy;
 296         to += copy;
 297         bytes -= copy;
 298         while (unlikely(!left && bytes)) {
 299                 iov++;
 300                 buf = iov->iov_base;
 301                 copy = min(bytes, iov->iov_len);
 302                 left = copyin(to, buf, copy);
 303                 copy -= left;
 304                 skip = copy;
 305                 to += copy;
 306                 bytes -= copy;
 307         }
 308         kunmap(page);
 309
 310 done:
 311         if (skip == iov->iov_len) {
 312                 iov++;
 313                 skip = 0;
 314         }
 315         i->count -= wanted - bytes;
 316         i->nr_segs -= iov - i->iov;
 317         i->iov = iov;
 318         i->iov_offset = skip;
 319         return wanted - bytes;
 320 }
 321
 322 #ifdef PIPE_PARANOIA
 323 static bool sanity(const struct iov_iter *i)
 324 {
 325         struct pipe_inode_info *pipe = i->pipe;
 326         int idx = i->idx;
 327         int next = pipe->curbuf + pipe->nrbufs;
 328         if (i->iov_offset) {
 329                 struct pipe_buffer *p;
 330                 if (unlikely(!pipe->nrbufs))
 331                         goto Bad;       // pipe must be non-empty
 332                 if (unlikely(idx != ((next - 1) & (pipe->buffers - 1))))
 333                         goto Bad;       // must be at the last buffer...
 334
 335                 p = &pipe->bufs[idx];
 336                 if (unlikely(p->offset + p->len != i->iov_offset))
 337                         goto Bad;       // ... at the end of segment
 338         } else {
 339                 if (idx != (next & (pipe->buffers - 1)))
 340                         goto Bad;       // must be right after the last buffer
 341         }
 342         return true;
 343 Bad:
 344         printk(KERN_ERR "idx = %d, offset = %zd\n", i->idx, i->iov_offset);
 345         printk(KERN_ERR "curbuf = %d, nrbufs = %d, buffers = %d\n",
 346                         pipe->curbuf, pipe->nrbufs, pipe->buffers);
 347         for (idx = 0; idx < pipe->buffers; idx++)
 348                 printk(KERN_ERR "[%p %p %d %d]\n",
 349                         pipe->bufs[idx].ops,
 350                         pipe->bufs[idx].page,
 351                         pipe->bufs[idx].offset,
 352                         pipe->bufs[idx].len);
 353         WARN_ON(1);
 354         return false;
 355 }
 356 #else
 357 #define sanity(i) true
 358 #endif
 359
 360 static inline int next_idx(int idx, struct pipe_inode_info *pipe)
 361 {
 362         return (idx + 1) & (pipe->buffers - 1);
 363 }
 364
 365 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 366                          struct iov_iter *i)
 367 {
 368         struct pipe_inode_info *pipe = i->pipe;
 369         struct pipe_buffer *buf;
 370         size_t off;
 371         int idx;
 372
 373         if (unlikely(bytes > i->count))
 374                 bytes = i->count;
 375
 376         if (unlikely(!bytes))
 377                 return 0;
 378
 379         if (!sanity(i))
 380                 return 0;
 381
 382         off = i->iov_offset;
 383         idx = i->idx;
 384         buf = &pipe->bufs[idx];
 385         if (off) {
 386                 if (offset == off && buf->page == page) {
 387                         /* merge with the last one */
 388                         buf->len += bytes;
 389                         i->iov_offset += bytes;
 390                         goto out;
 391                 }
 392                 idx = next_idx(idx, pipe);
 393                 buf = &pipe->bufs[idx];
 394         }
 395         if (idx == pipe->curbuf && pipe->nrbufs)
 396                 return 0;
 397         pipe->nrbufs++;
 398         buf->ops = &page_cache_pipe_buf_ops;
 399         get_page(buf->page = page);
 400         buf->offset = offset;
 401         buf->len = bytes;
 402         i->iov_offset = offset + bytes;
 403         i->idx = idx;
 404 out:
 405         i->count -= bytes;
 406         return bytes;
 407 }
 408
 409 /*
 410  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 411  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 412  *
 413  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 414  * because it is an invalid address).
 415  */
 416 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
 417 {
 418         size_t skip = i->iov_offset;
 419         const struct iovec *iov;
 420         int err;
 421         struct iovec v;
 422
 423         if (!(i->type & (ITER_BVEC|ITER_KVEC))) {
 424                 iterate_iovec(i, bytes, v, iov, skip, ({
 425                         err = fault_in_pages_readable(v.iov_base, v.iov_len);
 426                         if (unlikely(err))
 427                         return err;
 428                 0;}))
 429         }
 430         return 0;
 431 }
 432 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 433
 434 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 435                         const struct iovec *iov, unsigned long nr_segs,
 436                         size_t count)
 437 {
 438         WARN_ON(direction & ~(READ | WRITE));
 439         direction &= READ | WRITE;
 440
 441         /* It will get better.  Eventually... */
 442         if (uaccess_kernel()) {
 443                 i->type = ITER_KVEC | direction;
 444                 i->kvec = (struct kvec *)iov;
 445         } else {
 446                 i->type = ITER_IOVEC | direction;
 447                 i->iov = iov;
 448         }
 449         i->nr_segs = nr_segs;
 450         i->iov_offset = 0;
 451         i->count = count;
 452 }
 453 EXPORT_SYMBOL(iov_iter_init);
 454
 455 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
 456 {
 457         char *from = kmap_atomic(page);
 458         memcpy(to, from + offset, len);
 459         kunmap_atomic(from);
 460 }
 461
 462 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
 463 {
 464         char *to = kmap_atomic(page);
 465         memcpy(to + offset, from, len);
 466         kunmap_atomic(to);
 467 }
 468
 469 static void memzero_page(struct page *page, size_t offset, size_t len)
 470 {
 471         char *addr = kmap_atomic(page);
 472         memset(addr + offset, 0, len);
 473         kunmap_atomic(addr);
 474 }
 475
 476 static inline bool allocated(struct pipe_buffer *buf)
 477 {
 478         return buf->ops == &default_pipe_buf_ops;
 479 }
 480
 481 static inline void data_start(const struct iov_iter *i, int *idxp, size_t *offp)
 482 {
 483         size_t off = i->iov_offset;
 484         int idx = i->idx;
 485         if (off && (!allocated(&i->pipe->bufs[idx]) || off == PAGE_SIZE)) {
 486                 idx = next_idx(idx, i->pipe);
 487                 off = 0;
 488         }
 489         *idxp = idx;
 490         *offp = off;
 491 }
 492
 493 static size_t push_pipe(struct iov_iter *i, size_t size,
 494                         int *idxp, size_t *offp)
 495 {
 496         struct pipe_inode_info *pipe = i->pipe;
 497         size_t off;
 498         int idx;
 499         ssize_t left;
 500
 501         if (unlikely(size > i->count))
 502                 size = i->count;
 503         if (unlikely(!size))
 504                 return 0;
 505
 506         left = size;
 507         data_start(i, &idx, &off);
 508         *idxp = idx;
 509         *offp = off;
 510         if (off) {
 511                 left -= PAGE_SIZE - off;
 512                 if (left <= 0) {
 513                         pipe->bufs[idx].len += size;
 514                         return size;
 515                 }
 516                 pipe->bufs[idx].len = PAGE_SIZE;
 517                 idx = next_idx(idx, pipe);
 518         }
 519         while (idx != pipe->curbuf || !pipe->nrbufs) {
 520                 struct page *page = alloc_page(GFP_USER);
 521                 if (!page)
 522                         break;
 523                 pipe->nrbufs++;
 524                 pipe->bufs[idx].ops = &default_pipe_buf_ops;
 525                 pipe->bufs[idx].page = page;
 526                 pipe->bufs[idx].offset = 0;
 527                 if (left <= PAGE_SIZE) {
 528                         pipe->bufs[idx].len = left;
 529                         return size;
 530                 }
 531                 pipe->bufs[idx].len = PAGE_SIZE;
 532                 left -= PAGE_SIZE;
 533                 idx = next_idx(idx, pipe);
 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         size_t n, off;
 543         int idx;
 544
 545         if (!sanity(i))
 546                 return 0;
 547
 548         bytes = n = push_pipe(i, bytes, &idx, &off);
 549         if (unlikely(!n))
 550                 return 0;
 551         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 552                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 553                 memcpy_to_page(pipe->bufs[idx].page, off, addr, chunk);
 554                 i->idx = idx;
 555                 i->iov_offset = off + chunk;
 556                 n -= chunk;
 557                 addr += chunk;
 558         }
 559         i->count -= bytes;
 560         return bytes;
 561 }
 562
 563 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 564 {
 565         const char *from = addr;
 566         if (unlikely(iov_iter_is_pipe(i)))
 567                 return copy_pipe_to_iter(addr, bytes, i);
 568         if (iter_is_iovec(i))
 569                 might_fault();
 570         iterate_and_advance(i, bytes, v,
 571                 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 572                 memcpy_to_page(v.bv_page, v.bv_offset,
 573                                (from += v.bv_len) - v.bv_len, v.bv_len),
 574                 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
 575         )
 576
 577         return bytes;
 578 }
 579 EXPORT_SYMBOL(_copy_to_iter);
 580
 581 #ifdef CONFIG_ARCH_HAS_UACCESS_MCSAFE
 582 static int copyout_mcsafe(void __user *to, const void *from, size_t n)
 583 {
 584         if (access_ok(VERIFY_WRITE, to, n)) {
 585                 kasan_check_read(from, n);
 586                 n = copy_to_user_mcsafe((__force void *) to, from, n);
 587         }
 588         return n;
 589 }
 590
 591 static unsigned long memcpy_mcsafe_to_page(struct page *page, size_t offset,
 592                 const char *from, size_t len)
 593 {
 594         unsigned long ret;
 595         char *to;
 596
 597         to = kmap_atomic(page);
 598         ret = memcpy_mcsafe(to + offset, from, len);
 599         kunmap_atomic(to);
 600
 601         return ret;
 602 }
 603
 604 static size_t copy_pipe_to_iter_mcsafe(const void *addr, size_t bytes,
 605                                 struct iov_iter *i)
 606 {
 607         struct pipe_inode_info *pipe = i->pipe;
 608         size_t n, off, xfer = 0;
 609         int idx;
 610
 611         if (!sanity(i))
 612                 return 0;
 613
 614         bytes = n = push_pipe(i, bytes, &idx, &off);
 615         if (unlikely(!n))
 616                 return 0;
 617         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 618                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 619                 unsigned long rem;
 620
 621                 rem = memcpy_mcsafe_to_page(pipe->bufs[idx].page, off, addr,
 622                                 chunk);
 623                 i->idx = idx;
 624                 i->iov_offset = off + chunk - rem;
 625                 xfer += chunk - rem;
 626                 if (rem)
 627                         break;
 628                 n -= chunk;
 629                 addr += chunk;
 630         }
 631         i->count -= xfer;
 632         return xfer;
 633 }
 634
 635 /**
 636  * _copy_to_iter_mcsafe - copy to user with source-read error exception handling
 637  * @addr: source kernel address
 638  * @bytes: total transfer length
 639  * @iter: destination iterator
 640  *
 641  * The pmem driver arranges for filesystem-dax to use this facility via
 642  * dax_copy_to_iter() for protecting read/write to persistent memory.
 643  * Unless / until an architecture can guarantee identical performance
 644  * between _copy_to_iter_mcsafe() and _copy_to_iter() it would be a
 645  * performance regression to switch more users to the mcsafe version.
 646  *
 647  * Otherwise, the main differences between this and typical _copy_to_iter().
 648  *
 649  * * Typical tail/residue handling after a fault retries the copy
 650  *   byte-by-byte until the fault happens again. Re-triggering machine
 651  *   checks is potentially fatal so the implementation uses source
 652  *   alignment and poison alignment assumptions to avoid re-triggering
 653  *   hardware exceptions.
 654  *
 655  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 656  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 657  *   a short copy.
 658  *
 659  * See MCSAFE_TEST for self-test.
 660  */
 661 size_t _copy_to_iter_mcsafe(const void *addr, size_t bytes, struct iov_iter *i)
 662 {
 663         const char *from = addr;
 664         unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
 665
 666         if (unlikely(iov_iter_is_pipe(i)))
 667                 return copy_pipe_to_iter_mcsafe(addr, bytes, i);
 668         if (iter_is_iovec(i))
 669                 might_fault();
 670         iterate_and_advance(i, bytes, v,
 671                 copyout_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 672                 ({
 673                 rem = memcpy_mcsafe_to_page(v.bv_page, v.bv_offset,
 674                                (from += v.bv_len) - v.bv_len, v.bv_len);
 675                 if (rem) {
 676                         curr_addr = (unsigned long) from;
 677                         bytes = curr_addr - s_addr - rem;
 678                         return bytes;
 679                 }
 680                 }),
 681                 ({
 682                 rem = memcpy_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len,
 683                                 v.iov_len);
 684                 if (rem) {
 685                         curr_addr = (unsigned long) from;
 686                         bytes = curr_addr - s_addr - rem;
 687                         return bytes;
 688                 }
 689                 })
 690         )
 691
 692         return bytes;
 693 }
 694 EXPORT_SYMBOL_GPL(_copy_to_iter_mcsafe);
 695 #endif /* CONFIG_ARCH_HAS_UACCESS_MCSAFE */
 696
 697 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 698 {
 699         char *to = addr;
 700         if (unlikely(iov_iter_is_pipe(i))) {
 701                 WARN_ON(1);
 702                 return 0;
 703         }
 704         if (iter_is_iovec(i))
 705                 might_fault();
 706         iterate_and_advance(i, bytes, v,
 707                 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 708                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 709                                  v.bv_offset, v.bv_len),
 710                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 711         )
 712
 713         return bytes;
 714 }
 715 EXPORT_SYMBOL(_copy_from_iter);
 716
 717 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
 718 {
 719         char *to = addr;
 720         if (unlikely(iov_iter_is_pipe(i))) {
 721                 WARN_ON(1);
 722                 return false;
 723         }
 724         if (unlikely(i->count < bytes))
 725                 return false;
 726
 727         if (iter_is_iovec(i))
 728                 might_fault();
 729         iterate_all_kinds(i, bytes, v, ({
 730                 if (copyin((to += v.iov_len) - v.iov_len,
 731                                       v.iov_base, v.iov_len))
 732                         return false;
 733                 0;}),
 734                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 735                                  v.bv_offset, v.bv_len),
 736                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 737         )
 738
 739         iov_iter_advance(i, bytes);
 740         return true;
 741 }
 742 EXPORT_SYMBOL(_copy_from_iter_full);
 743
 744 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 745 {
 746         char *to = addr;
 747         if (unlikely(iov_iter_is_pipe(i))) {
 748                 WARN_ON(1);
 749                 return 0;
 750         }
 751         iterate_and_advance(i, bytes, v,
 752                 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 753                                          v.iov_base, v.iov_len),
 754                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 755                                  v.bv_offset, v.bv_len),
 756                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 757         )
 758
 759         return bytes;
 760 }
 761 EXPORT_SYMBOL(_copy_from_iter_nocache);
 762
 763 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 764 /**
 765  * _copy_from_iter_flushcache - write destination through cpu cache
 766  * @addr: destination kernel address
 767  * @bytes: total transfer length
 768  * @iter: source iterator
 769  *
 770  * The pmem driver arranges for filesystem-dax to use this facility via
 771  * dax_copy_from_iter() for ensuring that writes to persistent memory
 772  * are flushed through the CPU cache. It is differentiated from
 773  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 774  * all iterator types. The _copy_from_iter_nocache() only attempts to
 775  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 776  * instructions that strand dirty-data in the cache.
 777  */
 778 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 779 {
 780         char *to = addr;
 781         if (unlikely(iov_iter_is_pipe(i))) {
 782                 WARN_ON(1);
 783                 return 0;
 784         }
 785         iterate_and_advance(i, bytes, v,
 786                 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
 787                                          v.iov_base, v.iov_len),
 788                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 789                                  v.bv_offset, v.bv_len),
 790                 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
 791                         v.iov_len)
 792         )
 793
 794         return bytes;
 795 }
 796 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 797 #endif
 798
 799 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
 800 {
 801         char *to = addr;
 802         if (unlikely(iov_iter_is_pipe(i))) {
 803                 WARN_ON(1);
 804                 return false;
 805         }
 806         if (unlikely(i->count < bytes))
 807                 return false;
 808         iterate_all_kinds(i, bytes, v, ({
 809                 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 810                                              v.iov_base, v.iov_len))
 811                         return false;
 812                 0;}),
 813                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 814                                  v.bv_offset, v.bv_len),
 815                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 816         )
 817
 818         iov_iter_advance(i, bytes);
 819         return true;
 820 }
 821 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
 822
 823 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 824 {
 825         struct page *head = compound_head(page);
 826         size_t v = n + offset + page_address(page) - page_address(head);
 827
 828         if (likely(n <= v && v <= (PAGE_SIZE << compound_order(head))))
 829                 return true;
 830         WARN_ON(1);
 831         return false;
 832 }
 833
 834 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 835                          struct iov_iter *i)
 836 {
 837         if (unlikely(!page_copy_sane(page, offset, bytes)))
 838                 return 0;
 839         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 840                 void *kaddr = kmap_atomic(page);
 841                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 842                 kunmap_atomic(kaddr);
 843                 return wanted;
 844         } else if (unlikely(iov_iter_is_discard(i)))
 845                 return bytes;
 846         else if (likely(!iov_iter_is_pipe(i)))
 847                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 848         else
 849                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 850 }
 851 EXPORT_SYMBOL(copy_page_to_iter);
 852
 853 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 854                          struct iov_iter *i)
 855 {
 856         if (unlikely(!page_copy_sane(page, offset, bytes)))
 857                 return 0;
 858         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 859                 WARN_ON(1);
 860                 return 0;
 861         }
 862         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 863                 void *kaddr = kmap_atomic(page);
 864                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 865                 kunmap_atomic(kaddr);
 866                 return wanted;
 867         } else
 868                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 869 }
 870 EXPORT_SYMBOL(copy_page_from_iter);
 871
 872 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 873 {
 874         struct pipe_inode_info *pipe = i->pipe;
 875         size_t n, off;
 876         int idx;
 877
 878         if (!sanity(i))
 879                 return 0;
 880
 881         bytes = n = push_pipe(i, bytes, &idx, &off);
 882         if (unlikely(!n))
 883                 return 0;
 884
 885         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 886                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 887                 memzero_page(pipe->bufs[idx].page, off, chunk);
 888                 i->idx = idx;
 889                 i->iov_offset = off + chunk;
 890                 n -= chunk;
 891         }
 892         i->count -= bytes;
 893         return bytes;
 894 }
 895
 896 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 897 {
 898         if (unlikely(iov_iter_is_pipe(i)))
 899                 return pipe_zero(bytes, i);
 900         iterate_and_advance(i, bytes, v,
 901                 clear_user(v.iov_base, v.iov_len),
 902                 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
 903                 memset(v.iov_base, 0, v.iov_len)
 904         )
 905
 906         return bytes;
 907 }
 908 EXPORT_SYMBOL(iov_iter_zero);
 909
 910 size_t iov_iter_copy_from_user_atomic(struct page *page,
 911                 struct iov_iter *i, unsigned long offset, size_t bytes)
 912 {
 913         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 914         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 915                 kunmap_atomic(kaddr);
 916                 return 0;
 917         }
 918         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 919                 kunmap_atomic(kaddr);
 920                 WARN_ON(1);
 921                 return 0;
 922         }
 923         iterate_all_kinds(i, bytes, v,
 924                 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 925                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
 926                                  v.bv_offset, v.bv_len),
 927                 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 928         )
 929         kunmap_atomic(kaddr);
 930         return bytes;
 931 }
 932 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
 933
 934 static inline void pipe_truncate(struct iov_iter *i)
 935 {
 936         struct pipe_inode_info *pipe = i->pipe;
 937         if (pipe->nrbufs) {
 938                 size_t off = i->iov_offset;
 939                 int idx = i->idx;
 940                 int nrbufs = (idx - pipe->curbuf) & (pipe->buffers - 1);
 941                 if (off) {
 942                         pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
 943                         idx = next_idx(idx, pipe);
 944                         nrbufs++;
 945                 }
 946                 while (pipe->nrbufs > nrbufs) {
 947                         pipe_buf_release(pipe, &pipe->bufs[idx]);
 948                         idx = next_idx(idx, pipe);
 949                         pipe->nrbufs--;
 950                 }
 951         }
 952 }
 953
 954 static void pipe_advance(struct iov_iter *i, size_t size)
 955 {
 956         struct pipe_inode_info *pipe = i->pipe;
 957         if (unlikely(i->count < size))
 958                 size = i->count;
 959         if (size) {
 960                 struct pipe_buffer *buf;
 961                 size_t off = i->iov_offset, left = size;
 962                 int idx = i->idx;
 963                 if (off) /* make it relative to the beginning of buffer */
 964                         left += off - pipe->bufs[idx].offset;
 965                 while (1) {
 966                         buf = &pipe->bufs[idx];
 967                         if (left <= buf->len)
 968                                 break;
 969                         left -= buf->len;
 970                         idx = next_idx(idx, pipe);
 971                 }
 972                 i->idx = idx;
 973                 i->iov_offset = buf->offset + left;
 974         }
 975         i->count -= size;
 976         /* ... and discard everything past that point */
 977         pipe_truncate(i);
 978 }
 979
 980 void iov_iter_advance(struct iov_iter *i, size_t size)
 981 {
 982         if (unlikely(iov_iter_is_pipe(i))) {
 983                 pipe_advance(i, size);
 984                 return;
 985         }
 986         if (unlikely(iov_iter_is_discard(i))) {
 987                 i->count -= size;
 988                 return;
 989         }
 990         iterate_and_advance(i, size, v, 0, 0, 0)
 991 }
 992 EXPORT_SYMBOL(iov_iter_advance);
 993
 994 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 995 {
 996         if (!unroll)
 997                 return;
 998         if (WARN_ON(unroll > MAX_RW_COUNT))
 999                 return;
1000         i->count += unroll;
1001         if (unlikely(iov_iter_is_pipe(i))) {
1002                 struct pipe_inode_info *pipe = i->pipe;
1003                 int idx = i->idx;
1004                 size_t off = i->iov_offset;
1005                 while (1) {
1006                         size_t n = off - pipe->bufs[idx].offset;
1007                         if (unroll < n) {
1008                                 off -= unroll;
1009                                 break;
1010                         }
1011                         unroll -= n;
1012                         if (!unroll && idx == i->start_idx) {
1013                                 off = 0;
1014                                 break;
1015                         }
1016                         if (!idx--)
1017                                 idx = pipe->buffers - 1;
1018                         off = pipe->bufs[idx].offset + pipe->bufs[idx].len;
1019                 }
1020                 i->iov_offset = off;
1021                 i->idx = idx;
1022                 pipe_truncate(i);
1023                 return;
1024         }
1025         if (unlikely(iov_iter_is_discard(i)))
1026                 return;
1027         if (unroll <= i->iov_offset) {
1028                 i->iov_offset -= unroll;
1029                 return;
1030         }
1031         unroll -= i->iov_offset;
1032         if (iov_iter_is_bvec(i)) {
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(iov_iter_is_pipe(i)))
1066                 return i->count;        // it is a silly place, anyway
1067         if (i->nr_segs == 1)
1068                 return i->count;
1069         if (unlikely(iov_iter_is_discard(i)))
1070                 return i->count;
1071         else if (iov_iter_is_bvec(i))
1072                 return min(i->count, i->bvec->bv_len - i->iov_offset);
1073         else
1074                 return min(i->count, i->iov->iov_len - i->iov_offset);
1075 }
1076 EXPORT_SYMBOL(iov_iter_single_seg_count);
1077
1078 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1079                         const struct kvec *kvec, unsigned long nr_segs,
1080                         size_t count)
1081 {
1082         WARN_ON(direction & ~(READ | WRITE));
1083         i->type = ITER_KVEC | (direction & (READ | WRITE));
1084         i->kvec = kvec;
1085         i->nr_segs = nr_segs;
1086         i->iov_offset = 0;
1087         i->count = count;
1088 }
1089 EXPORT_SYMBOL(iov_iter_kvec);
1090
1091 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1092                         const struct bio_vec *bvec, unsigned long nr_segs,
1093                         size_t count)
1094 {
1095         WARN_ON(direction & ~(READ | WRITE));
1096         i->type = ITER_BVEC | (direction & (READ | WRITE));
1097         i->bvec = bvec;
1098         i->nr_segs = nr_segs;
1099         i->iov_offset = 0;
1100         i->count = count;
1101 }
1102 EXPORT_SYMBOL(iov_iter_bvec);
1103
1104 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1105                         struct pipe_inode_info *pipe,
1106                         size_t count)
1107 {
1108         BUG_ON(direction != READ);
1109         WARN_ON(pipe->nrbufs == pipe->buffers);
1110         i->type = ITER_PIPE | READ;
1111         i->pipe = pipe;
1112         i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1113         i->iov_offset = 0;
1114         i->count = count;
1115         i->start_idx = i->idx;
1116 }
1117 EXPORT_SYMBOL(iov_iter_pipe);
1118
1119 /**
1120  * iov_iter_discard - Initialise an I/O iterator that discards data
1121  * @i: The iterator to initialise.
1122  * @direction: The direction of the transfer.
1123  * @count: The size of the I/O buffer in bytes.
1124  *
1125  * Set up an I/O iterator that just discards everything that's written to it.
1126  * It's only available as a READ iterator.
1127  */
1128 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1129 {
1130         BUG_ON(direction != READ);
1131         i->type = ITER_DISCARD | READ;
1132         i->count = count;
1133         i->iov_offset = 0;
1134 }
1135 EXPORT_SYMBOL(iov_iter_discard);
1136
1137 unsigned long iov_iter_alignment(const struct iov_iter *i)
1138 {
1139         unsigned long res = 0;
1140         size_t size = i->count;
1141
1142         if (unlikely(iov_iter_is_pipe(i))) {
1143                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
1144                         return size | i->iov_offset;
1145                 return size;
1146         }
1147         iterate_all_kinds(i, size, v,
1148                 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1149                 res |= v.bv_offset | v.bv_len,
1150                 res |= (unsigned long)v.iov_base | v.iov_len
1151         )
1152         return res;
1153 }
1154 EXPORT_SYMBOL(iov_iter_alignment);
1155
1156 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1157 {
1158         unsigned long res = 0;
1159         size_t size = i->count;
1160
1161         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1162                 WARN_ON(1);
1163                 return ~0U;
1164         }
1165
1166         iterate_all_kinds(i, size, v,
1167                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1168                         (size != v.iov_len ? size : 0), 0),
1169                 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1170                         (size != v.bv_len ? size : 0)),
1171                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1172                         (size != v.iov_len ? size : 0))
1173                 );
1174         return res;
1175 }
1176 EXPORT_SYMBOL(iov_iter_gap_alignment);
1177
1178 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1179                                 size_t maxsize,
1180                                 struct page **pages,
1181                                 int idx,
1182                                 size_t *start)
1183 {
1184         struct pipe_inode_info *pipe = i->pipe;
1185         ssize_t n = push_pipe(i, maxsize, &idx, start);
1186         if (!n)
1187                 return -EFAULT;
1188
1189         maxsize = n;
1190         n += *start;
1191         while (n > 0) {
1192                 get_page(*pages++ = pipe->bufs[idx].page);
1193                 idx = next_idx(idx, pipe);
1194                 n -= PAGE_SIZE;
1195         }
1196
1197         return maxsize;
1198 }
1199
1200 static ssize_t pipe_get_pages(struct iov_iter *i,
1201                    struct page **pages, size_t maxsize, unsigned maxpages,
1202                    size_t *start)
1203 {
1204         unsigned npages;
1205         size_t capacity;
1206         int idx;
1207
1208         if (!maxsize)
1209                 return 0;
1210
1211         if (!sanity(i))
1212                 return -EFAULT;
1213
1214         data_start(i, &idx, start);
1215         /* some of this one + all after this one */
1216         npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1217         capacity = min(npages,maxpages) * PAGE_SIZE - *start;
1218
1219         return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
1220 }
1221
1222 ssize_t iov_iter_get_pages(struct iov_iter *i,
1223                    struct page **pages, size_t maxsize, unsigned maxpages,
1224                    size_t *start)
1225 {
1226         if (maxsize > i->count)
1227                 maxsize = i->count;
1228
1229         if (unlikely(iov_iter_is_pipe(i)))
1230                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1231         if (unlikely(iov_iter_is_discard(i)))
1232                 return -EFAULT;
1233
1234         iterate_all_kinds(i, maxsize, v, ({
1235                 unsigned long addr = (unsigned long)v.iov_base;
1236                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1237                 int n;
1238                 int res;
1239
1240                 if (len > maxpages * PAGE_SIZE)
1241                         len = maxpages * PAGE_SIZE;
1242                 addr &= ~(PAGE_SIZE - 1);
1243                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1244                 res = get_user_pages_fast(addr, n, iov_iter_rw(i) != WRITE, pages);
1245                 if (unlikely(res < 0))
1246                         return res;
1247                 return (res == n ? len : res * PAGE_SIZE) - *start;
1248         0;}),({
1249                 /* can't be more than PAGE_SIZE */
1250                 *start = v.bv_offset;
1251                 get_page(*pages = v.bv_page);
1252                 return v.bv_len;
1253         }),({
1254                 return -EFAULT;
1255         })
1256         )
1257         return 0;
1258 }
1259 EXPORT_SYMBOL(iov_iter_get_pages);
1260
1261 static struct page **get_pages_array(size_t n)
1262 {
1263         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1264 }
1265
1266 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1267                    struct page ***pages, size_t maxsize,
1268                    size_t *start)
1269 {
1270         struct page **p;
1271         ssize_t n;
1272         int idx;
1273         int npages;
1274
1275         if (!maxsize)
1276                 return 0;
1277
1278         if (!sanity(i))
1279                 return -EFAULT;
1280
1281         data_start(i, &idx, start);
1282         /* some of this one + all after this one */
1283         npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1284         n = npages * PAGE_SIZE - *start;
1285         if (maxsize > n)
1286                 maxsize = n;
1287         else
1288                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1289         p = get_pages_array(npages);
1290         if (!p)
1291                 return -ENOMEM;
1292         n = __pipe_get_pages(i, maxsize, p, idx, start);
1293         if (n > 0)
1294                 *pages = p;
1295         else
1296                 kvfree(p);
1297         return n;
1298 }
1299
1300 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1301                    struct page ***pages, size_t maxsize,
1302                    size_t *start)
1303 {
1304         struct page **p;
1305
1306         if (maxsize > i->count)
1307                 maxsize = i->count;
1308
1309         if (unlikely(iov_iter_is_pipe(i)))
1310                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1311         if (unlikely(iov_iter_is_discard(i)))
1312                 return -EFAULT;
1313
1314         iterate_all_kinds(i, maxsize, v, ({
1315                 unsigned long addr = (unsigned long)v.iov_base;
1316                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1317                 int n;
1318                 int res;
1319
1320                 addr &= ~(PAGE_SIZE - 1);
1321                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1322                 p = get_pages_array(n);
1323                 if (!p)
1324                         return -ENOMEM;
1325                 res = get_user_pages_fast(addr, n, iov_iter_rw(i) != WRITE, p);
1326                 if (unlikely(res < 0)) {
1327                         kvfree(p);
1328                         return res;
1329                 }
1330                 *pages = p;
1331                 return (res == n ? len : res * PAGE_SIZE) - *start;
1332         0;}),({
1333                 /* can't be more than PAGE_SIZE */
1334                 *start = v.bv_offset;
1335                 *pages = p = get_pages_array(1);
1336                 if (!p)
1337                         return -ENOMEM;
1338                 get_page(*p = v.bv_page);
1339                 return v.bv_len;
1340         }),({
1341                 return -EFAULT;
1342         })
1343         )
1344         return 0;
1345 }
1346 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1347
1348 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1349                                struct iov_iter *i)
1350 {
1351         char *to = addr;
1352         __wsum sum, next;
1353         size_t off = 0;
1354         sum = *csum;
1355         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1356                 WARN_ON(1);
1357                 return 0;
1358         }
1359         iterate_and_advance(i, bytes, v, ({
1360                 int err = 0;
1361                 next = csum_and_copy_from_user(v.iov_base,
1362                                                (to += v.iov_len) - v.iov_len,
1363                                                v.iov_len, 0, &err);
1364                 if (!err) {
1365                         sum = csum_block_add(sum, next, off);
1366                         off += v.iov_len;
1367                 }
1368                 err ? v.iov_len : 0;
1369         }), ({
1370                 char *p = kmap_atomic(v.bv_page);
1371                 next = csum_partial_copy_nocheck(p + v.bv_offset,
1372                                                  (to += v.bv_len) - v.bv_len,
1373                                                  v.bv_len, 0);
1374                 kunmap_atomic(p);
1375                 sum = csum_block_add(sum, next, off);
1376                 off += v.bv_len;
1377         }),({
1378                 next = csum_partial_copy_nocheck(v.iov_base,
1379                                                  (to += v.iov_len) - v.iov_len,
1380                                                  v.iov_len, 0);
1381                 sum = csum_block_add(sum, next, off);
1382                 off += v.iov_len;
1383         })
1384         )
1385         *csum = sum;
1386         return bytes;
1387 }
1388 EXPORT_SYMBOL(csum_and_copy_from_iter);
1389
1390 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1391                                struct iov_iter *i)
1392 {
1393         char *to = addr;
1394         __wsum sum, next;
1395         size_t off = 0;
1396         sum = *csum;
1397         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1398                 WARN_ON(1);
1399                 return false;
1400         }
1401         if (unlikely(i->count < bytes))
1402                 return false;
1403         iterate_all_kinds(i, bytes, v, ({
1404                 int err = 0;
1405                 next = csum_and_copy_from_user(v.iov_base,
1406                                                (to += v.iov_len) - v.iov_len,
1407                                                v.iov_len, 0, &err);
1408                 if (err)
1409                         return false;
1410                 sum = csum_block_add(sum, next, off);
1411                 off += v.iov_len;
1412                 0;
1413         }), ({
1414                 char *p = kmap_atomic(v.bv_page);
1415                 next = csum_partial_copy_nocheck(p + v.bv_offset,
1416                                                  (to += v.bv_len) - v.bv_len,
1417                                                  v.bv_len, 0);
1418                 kunmap_atomic(p);
1419                 sum = csum_block_add(sum, next, off);
1420                 off += v.bv_len;
1421         }),({
1422                 next = csum_partial_copy_nocheck(v.iov_base,
1423                                                  (to += v.iov_len) - v.iov_len,
1424                                                  v.iov_len, 0);
1425                 sum = csum_block_add(sum, next, off);
1426                 off += v.iov_len;
1427         })
1428         )
1429         *csum = sum;
1430         iov_iter_advance(i, bytes);
1431         return true;
1432 }
1433 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1434
1435 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, __wsum *csum,
1436                              struct iov_iter *i)
1437 {
1438         const char *from = addr;
1439         __wsum sum, next;
1440         size_t off = 0;
1441         sum = *csum;
1442         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1443                 WARN_ON(1);     /* for now */
1444                 return 0;
1445         }
1446         iterate_and_advance(i, bytes, v, ({
1447                 int err = 0;
1448                 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1449                                              v.iov_base,
1450                                              v.iov_len, 0, &err);
1451                 if (!err) {
1452                         sum = csum_block_add(sum, next, off);
1453                         off += v.iov_len;
1454                 }
1455                 err ? v.iov_len : 0;
1456         }), ({
1457                 char *p = kmap_atomic(v.bv_page);
1458                 next = csum_partial_copy_nocheck((from += v.bv_len) - v.bv_len,
1459                                                  p + v.bv_offset,
1460                                                  v.bv_len, 0);
1461                 kunmap_atomic(p);
1462                 sum = csum_block_add(sum, next, off);
1463                 off += v.bv_len;
1464         }),({
1465                 next = csum_partial_copy_nocheck((from += v.iov_len) - v.iov_len,
1466                                                  v.iov_base,
1467                                                  v.iov_len, 0);
1468                 sum = csum_block_add(sum, next, off);
1469                 off += v.iov_len;
1470         })
1471         )
1472         *csum = sum;
1473         return bytes;
1474 }
1475 EXPORT_SYMBOL(csum_and_copy_to_iter);
1476
1477 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1478 {
1479         size_t size = i->count;
1480         int npages = 0;
1481
1482         if (!size)
1483                 return 0;
1484         if (unlikely(iov_iter_is_discard(i)))
1485                 return 0;
1486
1487         if (unlikely(iov_iter_is_pipe(i))) {
1488                 struct pipe_inode_info *pipe = i->pipe;
1489                 size_t off;
1490                 int idx;
1491
1492                 if (!sanity(i))
1493                         return 0;
1494
1495                 data_start(i, &idx, &off);
1496                 /* some of this one + all after this one */
1497                 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1498                 if (npages >= maxpages)
1499                         return maxpages;
1500         } else iterate_all_kinds(i, size, v, ({
1501                 unsigned long p = (unsigned long)v.iov_base;
1502                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1503                         - p / PAGE_SIZE;
1504                 if (npages >= maxpages)
1505                         return maxpages;
1506         0;}),({
1507                 npages++;
1508                 if (npages >= maxpages)
1509                         return maxpages;
1510         }),({
1511                 unsigned long p = (unsigned long)v.iov_base;
1512                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1513                         - p / PAGE_SIZE;
1514                 if (npages >= maxpages)
1515                         return maxpages;
1516         })
1517         )
1518         return npages;
1519 }
1520 EXPORT_SYMBOL(iov_iter_npages);
1521
1522 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1523 {
1524         *new = *old;
1525         if (unlikely(iov_iter_is_pipe(new))) {
1526                 WARN_ON(1);
1527                 return NULL;
1528         }
1529         if (unlikely(iov_iter_is_discard(new)))
1530                 return NULL;
1531         if (iov_iter_is_bvec(new))
1532                 return new->bvec = kmemdup(new->bvec,
1533                                     new->nr_segs * sizeof(struct bio_vec),
1534                                     flags);
1535         else
1536                 /* iovec and kvec have identical layout */
1537                 return new->iov = kmemdup(new->iov,
1538                                    new->nr_segs * sizeof(struct iovec),
1539                                    flags);
1540 }
1541 EXPORT_SYMBOL(dup_iter);
1542
1543 /**
1544  * import_iovec() - Copy an array of &struct iovec from userspace
1545  *     into the kernel, check that it is valid, and initialize a new
1546  *     &struct iov_iter iterator to access it.
1547  *
1548  * @type: One of %READ or %WRITE.
1549  * @uvector: Pointer to the userspace array.
1550  * @nr_segs: Number of elements in userspace array.
1551  * @fast_segs: Number of elements in @iov.
1552  * @iov: (input and output parameter) Pointer to pointer to (usually small
1553  *     on-stack) kernel array.
1554  * @i: Pointer to iterator that will be initialized on success.
1555  *
1556  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1557  * then this function places %NULL in *@iov on return. Otherwise, a new
1558  * array will be allocated and the result placed in *@iov. This means that
1559  * the caller may call kfree() on *@iov regardless of whether the small
1560  * on-stack array was used or not (and regardless of whether this function
1561  * returns an error or not).
1562  *
1563  * Return: 0 on success or negative error code on error.
1564  */
1565 int import_iovec(int type, const struct iovec __user * uvector,
1566                  unsigned nr_segs, unsigned fast_segs,
1567                  struct iovec **iov, struct iov_iter *i)
1568 {
1569         ssize_t n;
1570         struct iovec *p;
1571         n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1572                                   *iov, &p);
1573         if (n < 0) {
1574                 if (p != *iov)
1575                         kfree(p);
1576                 *iov = NULL;
1577                 return n;
1578         }
1579         iov_iter_init(i, type, p, nr_segs, n);
1580         *iov = p == *iov ? NULL : p;
1581         return 0;
1582 }
1583 EXPORT_SYMBOL(import_iovec);
1584
1585 #ifdef CONFIG_COMPAT
1586 #include <linux/compat.h>
1587
1588 int compat_import_iovec(int type, const struct compat_iovec __user * uvector,
1589                  unsigned nr_segs, unsigned fast_segs,
1590                  struct iovec **iov, struct iov_iter *i)
1591 {
1592         ssize_t n;
1593         struct iovec *p;
1594         n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1595                                   *iov, &p);
1596         if (n < 0) {
1597                 if (p != *iov)
1598                         kfree(p);
1599                 *iov = NULL;
1600                 return n;
1601         }
1602         iov_iter_init(i, type, p, nr_segs, n);
1603         *iov = p == *iov ? NULL : p;
1604         return 0;
1605 }
1606 #endif
1607
1608 int import_single_range(int rw, void __user *buf, size_t len,
1609                  struct iovec *iov, struct iov_iter *i)
1610 {
1611         if (len > MAX_RW_COUNT)
1612                 len = MAX_RW_COUNT;
1613         if (unlikely(!access_ok(!rw, buf, len)))
1614                 return -EFAULT;
1615
1616         iov->iov_base = buf;
1617         iov->iov_len = len;
1618         iov_iter_init(i, rw, iov, 1, len);
1619         return 0;
1620 }
1621 EXPORT_SYMBOL(import_single_range);
1622
1623 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1624                             int (*f)(struct kvec *vec, void *context),
1625                             void *context)
1626 {
1627         struct kvec w;
1628         int err = -EINVAL;
1629         if (!bytes)
1630                 return 0;
1631
1632         iterate_all_kinds(i, bytes, v, -EINVAL, ({
1633                 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1634                 w.iov_len = v.bv_len;
1635                 err = f(&w, context);
1636                 kunmap(v.bv_page);
1637                 err;}), ({
1638                 w = v;
1639                 err = f(&w, context);})
1640         )
1641         return err;
1642 }
1643 EXPORT_SYMBOL(iov_iter_for_each_range);