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 #include <linux/scatterlist.h>
  10
  11 #define PIPE_PARANOIA /* for now */
  12
  13 #define iterate_iovec(i, n, __v, __p, skip, STEP) {     \
  14         size_t left;                                    \
  15         size_t wanted = n;                              \
  16         __p = i->iov;                                   \
  17         __v.iov_len = min(n, __p->iov_len - skip);      \
  18         if (likely(__v.iov_len)) {                      \
  19                 __v.iov_base = __p->iov_base + skip;    \
  20                 left = (STEP);                          \
  21                 __v.iov_len -= left;                    \
  22                 skip += __v.iov_len;                    \
  23                 n -= __v.iov_len;                       \
  24         } else {                                        \
  25                 left = 0;                               \
  26         }                                               \
  27         while (unlikely(!left && n)) {                  \
  28                 __p++;                                  \
  29                 __v.iov_len = min(n, __p->iov_len);     \
  30                 if (unlikely(!__v.iov_len))             \
  31                         continue;                       \
  32                 __v.iov_base = __p->iov_base;           \
  33                 left = (STEP);                          \
  34                 __v.iov_len -= left;                    \
  35                 skip = __v.iov_len;                     \
  36                 n -= __v.iov_len;                       \
  37         }                                               \
  38         n = wanted - n;                                 \
  39 }
  40
  41 #define iterate_kvec(i, n, __v, __p, skip, STEP) {      \
  42         size_t wanted = n;                              \
  43         __p = i->kvec;                                  \
  44         __v.iov_len = min(n, __p->iov_len - skip);      \
  45         if (likely(__v.iov_len)) {                      \
  46                 __v.iov_base = __p->iov_base + skip;    \
  47                 (void)(STEP);                           \
  48                 skip += __v.iov_len;                    \
  49                 n -= __v.iov_len;                       \
  50         }                                               \
  51         while (unlikely(n)) {                           \
  52                 __p++;                                  \
  53                 __v.iov_len = min(n, __p->iov_len);     \
  54                 if (unlikely(!__v.iov_len))             \
  55                         continue;                       \
  56                 __v.iov_base = __p->iov_base;           \
  57                 (void)(STEP);                           \
  58                 skip = __v.iov_len;                     \
  59                 n -= __v.iov_len;                       \
  60         }                                               \
  61         n = wanted;                                     \
  62 }
  63
  64 #define iterate_bvec(i, n, __v, __bi, skip, STEP) {     \
  65         struct bvec_iter __start;                       \
  66         __start.bi_size = n;                            \
  67         __start.bi_bvec_done = skip;                    \
  68         __start.bi_idx = 0;                             \
  69         for_each_bvec(__v, i->bvec, __bi, __start) {    \
  70                 if (!__v.bv_len)                        \
  71                         continue;                       \
  72                 (void)(STEP);                           \
  73         }                                               \
  74 }
  75
  76 #define iterate_all_kinds(i, n, v, I, B, K) {                   \
  77         if (likely(n)) {                                        \
  78                 size_t skip = i->iov_offset;                    \
  79                 if (unlikely(i->type & ITER_BVEC)) {            \
  80                         struct bio_vec v;                       \
  81                         struct bvec_iter __bi;                  \
  82                         iterate_bvec(i, n, v, __bi, skip, (B))  \
  83                 } else if (unlikely(i->type & ITER_KVEC)) {     \
  84                         const struct kvec *kvec;                \
  85                         struct kvec v;                          \
  86                         iterate_kvec(i, n, v, kvec, skip, (K))  \
  87                 } else if (unlikely(i->type & ITER_DISCARD)) {  \
  88                 } else {                                        \
  89                         const struct iovec *iov;                \
  90                         struct iovec v;                         \
  91                         iterate_iovec(i, n, v, iov, skip, (I))  \
  92                 }                                               \
  93         }                                                       \
  94 }
  95
  96 #define iterate_and_advance(i, n, v, I, B, K) {                 \
  97         if (unlikely(i->count < n))                             \
  98                 n = i->count;                                   \
  99         if (i->count) {                                         \
 100                 size_t skip = i->iov_offset;                    \
 101                 if (unlikely(i->type & ITER_BVEC)) {            \
 102                         const struct bio_vec *bvec = i->bvec;   \
 103                         struct bio_vec v;                       \
 104                         struct bvec_iter __bi;                  \
 105                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 106                         i->bvec = __bvec_iter_bvec(i->bvec, __bi);      \
 107                         i->nr_segs -= i->bvec - bvec;           \
 108                         skip = __bi.bi_bvec_done;               \
 109                 } else if (unlikely(i->type & ITER_KVEC)) {     \
 110                         const struct kvec *kvec;                \
 111                         struct kvec v;                          \
 112                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 113                         if (skip == kvec->iov_len) {            \
 114                                 kvec++;                         \
 115                                 skip = 0;                       \
 116                         }                                       \
 117                         i->nr_segs -= kvec - i->kvec;           \
 118                         i->kvec = kvec;                         \
 119                 } else if (unlikely(i->type & ITER_DISCARD)) {  \
 120                         skip += n;                              \
 121                 } else {                                        \
 122                         const struct iovec *iov;                \
 123                         struct iovec v;                         \
 124                         iterate_iovec(i, n, v, iov, skip, (I))  \
 125                         if (skip == iov->iov_len) {             \
 126                                 iov++;                          \
 127                                 skip = 0;                       \
 128                         }                                       \
 129                         i->nr_segs -= iov - i->iov;             \
 130                         i->iov = iov;                           \
 131                 }                                               \
 132                 i->count -= n;                                  \
 133                 i->iov_offset = skip;                           \
 134         }                                                       \
 135 }
 136
 137 static int copyout(void __user *to, const void *from, size_t n)
 138 {
 139         if (access_ok(to, n)) {
 140                 kasan_check_read(from, n);
 141                 n = raw_copy_to_user(to, from, n);
 142         }
 143         return n;
 144 }
 145
 146 static int copyin(void *to, const void __user *from, size_t n)
 147 {
 148         if (access_ok(from, n)) {
 149                 kasan_check_write(to, n);
 150                 n = raw_copy_from_user(to, from, n);
 151         }
 152         return n;
 153 }
 154
 155 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 156                          struct iov_iter *i)
 157 {
 158         size_t skip, copy, left, wanted;
 159         const struct iovec *iov;
 160         char __user *buf;
 161         void *kaddr, *from;
 162
 163         if (unlikely(bytes > i->count))
 164                 bytes = i->count;
 165
 166         if (unlikely(!bytes))
 167                 return 0;
 168
 169         might_fault();
 170         wanted = bytes;
 171         iov = i->iov;
 172         skip = i->iov_offset;
 173         buf = iov->iov_base + skip;
 174         copy = min(bytes, iov->iov_len - skip);
 175
 176         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 177                 kaddr = kmap_atomic(page);
 178                 from = kaddr + offset;
 179
 180                 /* first chunk, usually the only one */
 181                 left = copyout(buf, from, copy);
 182                 copy -= left;
 183                 skip += copy;
 184                 from += copy;
 185                 bytes -= copy;
 186
 187                 while (unlikely(!left && bytes)) {
 188                         iov++;
 189                         buf = iov->iov_base;
 190                         copy = min(bytes, iov->iov_len);
 191                         left = copyout(buf, from, copy);
 192                         copy -= left;
 193                         skip = copy;
 194                         from += copy;
 195                         bytes -= copy;
 196                 }
 197                 if (likely(!bytes)) {
 198                         kunmap_atomic(kaddr);
 199                         goto done;
 200                 }
 201                 offset = from - kaddr;
 202                 buf += copy;
 203                 kunmap_atomic(kaddr);
 204                 copy = min(bytes, iov->iov_len - skip);
 205         }
 206         /* Too bad - revert to non-atomic kmap */
 207
 208         kaddr = kmap(page);
 209         from = kaddr + offset;
 210         left = copyout(buf, from, copy);
 211         copy -= left;
 212         skip += copy;
 213         from += copy;
 214         bytes -= copy;
 215         while (unlikely(!left && bytes)) {
 216                 iov++;
 217                 buf = iov->iov_base;
 218                 copy = min(bytes, iov->iov_len);
 219                 left = copyout(buf, from, copy);
 220                 copy -= left;
 221                 skip = copy;
 222                 from += copy;
 223                 bytes -= copy;
 224         }
 225         kunmap(page);
 226
 227 done:
 228         if (skip == iov->iov_len) {
 229                 iov++;
 230                 skip = 0;
 231         }
 232         i->count -= wanted - bytes;
 233         i->nr_segs -= iov - i->iov;
 234         i->iov = iov;
 235         i->iov_offset = skip;
 236         return wanted - bytes;
 237 }
 238
 239 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 240                          struct iov_iter *i)
 241 {
 242         size_t skip, copy, left, wanted;
 243         const struct iovec *iov;
 244         char __user *buf;
 245         void *kaddr, *to;
 246
 247         if (unlikely(bytes > i->count))
 248                 bytes = i->count;
 249
 250         if (unlikely(!bytes))
 251                 return 0;
 252
 253         might_fault();
 254         wanted = bytes;
 255         iov = i->iov;
 256         skip = i->iov_offset;
 257         buf = iov->iov_base + skip;
 258         copy = min(bytes, iov->iov_len - skip);
 259
 260         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 261                 kaddr = kmap_atomic(page);
 262                 to = kaddr + offset;
 263
 264                 /* first chunk, usually the only one */
 265                 left = copyin(to, buf, copy);
 266                 copy -= left;
 267                 skip += copy;
 268                 to += copy;
 269                 bytes -= copy;
 270
 271                 while (unlikely(!left && bytes)) {
 272                         iov++;
 273                         buf = iov->iov_base;
 274                         copy = min(bytes, iov->iov_len);
 275                         left = copyin(to, buf, copy);
 276                         copy -= left;
 277                         skip = copy;
 278                         to += copy;
 279                         bytes -= copy;
 280                 }
 281                 if (likely(!bytes)) {
 282                         kunmap_atomic(kaddr);
 283                         goto done;
 284                 }
 285                 offset = to - kaddr;
 286                 buf += copy;
 287                 kunmap_atomic(kaddr);
 288                 copy = min(bytes, iov->iov_len - skip);
 289         }
 290         /* Too bad - revert to non-atomic kmap */
 291
 292         kaddr = kmap(page);
 293         to = kaddr + offset;
 294         left = copyin(to, buf, copy);
 295         copy -= left;
 296         skip += copy;
 297         to += copy;
 298         bytes -= copy;
 299         while (unlikely(!left && bytes)) {
 300                 iov++;
 301                 buf = iov->iov_base;
 302                 copy = min(bytes, iov->iov_len);
 303                 left = copyin(to, buf, copy);
 304                 copy -= left;
 305                 skip = copy;
 306                 to += copy;
 307                 bytes -= copy;
 308         }
 309         kunmap(page);
 310
 311 done:
 312         if (skip == iov->iov_len) {
 313                 iov++;
 314                 skip = 0;
 315         }
 316         i->count -= wanted - bytes;
 317         i->nr_segs -= iov - i->iov;
 318         i->iov = iov;
 319         i->iov_offset = skip;
 320         return wanted - bytes;
 321 }
 322
 323 #ifdef PIPE_PARANOIA
 324 static bool sanity(const struct iov_iter *i)
 325 {
 326         struct pipe_inode_info *pipe = i->pipe;
 327         int idx = i->idx;
 328         int next = pipe->curbuf + pipe->nrbufs;
 329         if (i->iov_offset) {
 330                 struct pipe_buffer *p;
 331                 if (unlikely(!pipe->nrbufs))
 332                         goto Bad;       // pipe must be non-empty
 333                 if (unlikely(idx != ((next - 1) & (pipe->buffers - 1))))
 334                         goto Bad;       // must be at the last buffer...
 335
 336                 p = &pipe->bufs[idx];
 337                 if (unlikely(p->offset + p->len != i->iov_offset))
 338                         goto Bad;       // ... at the end of segment
 339         } else {
 340                 if (idx != (next & (pipe->buffers - 1)))
 341                         goto Bad;       // must be right after the last buffer
 342         }
 343         return true;
 344 Bad:
 345         printk(KERN_ERR "idx = %d, offset = %zd\n", i->idx, i->iov_offset);
 346         printk(KERN_ERR "curbuf = %d, nrbufs = %d, buffers = %d\n",
 347                         pipe->curbuf, pipe->nrbufs, pipe->buffers);
 348         for (idx = 0; idx < pipe->buffers; idx++)
 349                 printk(KERN_ERR "[%p %p %d %d]\n",
 350                         pipe->bufs[idx].ops,
 351                         pipe->bufs[idx].page,
 352                         pipe->bufs[idx].offset,
 353                         pipe->bufs[idx].len);
 354         WARN_ON(1);
 355         return false;
 356 }
 357 #else
 358 #define sanity(i) true
 359 #endif
 360
 361 static inline int next_idx(int idx, struct pipe_inode_info *pipe)
 362 {
 363         return (idx + 1) & (pipe->buffers - 1);
 364 }
 365
 366 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 367                          struct iov_iter *i)
 368 {
 369         struct pipe_inode_info *pipe = i->pipe;
 370         struct pipe_buffer *buf;
 371         size_t off;
 372         int idx;
 373
 374         if (unlikely(bytes > i->count))
 375                 bytes = i->count;
 376
 377         if (unlikely(!bytes))
 378                 return 0;
 379
 380         if (!sanity(i))
 381                 return 0;
 382
 383         off = i->iov_offset;
 384         idx = i->idx;
 385         buf = &pipe->bufs[idx];
 386         if (off) {
 387                 if (offset == off && buf->page == page) {
 388                         /* merge with the last one */
 389                         buf->len += bytes;
 390                         i->iov_offset += bytes;
 391                         goto out;
 392                 }
 393                 idx = next_idx(idx, pipe);
 394                 buf = &pipe->bufs[idx];
 395         }
 396         if (idx == pipe->curbuf && pipe->nrbufs)
 397                 return 0;
 398         pipe->nrbufs++;
 399         buf->ops = &page_cache_pipe_buf_ops;
 400         get_page(buf->page = page);
 401         buf->offset = offset;
 402         buf->len = bytes;
 403         i->iov_offset = offset + bytes;
 404         i->idx = idx;
 405 out:
 406         i->count -= bytes;
 407         return bytes;
 408 }
 409
 410 /*
 411  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 412  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 413  *
 414  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 415  * because it is an invalid address).
 416  */
 417 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
 418 {
 419         size_t skip = i->iov_offset;
 420         const struct iovec *iov;
 421         int err;
 422         struct iovec v;
 423
 424         if (!(i->type & (ITER_BVEC|ITER_KVEC))) {
 425                 iterate_iovec(i, bytes, v, iov, skip, ({
 426                         err = fault_in_pages_readable(v.iov_base, v.iov_len);
 427                         if (unlikely(err))
 428                         return err;
 429                 0;}))
 430         }
 431         return 0;
 432 }
 433 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 434
 435 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 436                         const struct iovec *iov, unsigned long nr_segs,
 437                         size_t count)
 438 {
 439         WARN_ON(direction & ~(READ | WRITE));
 440         direction &= READ | WRITE;
 441
 442         /* It will get better.  Eventually... */
 443         if (uaccess_kernel()) {
 444                 i->type = ITER_KVEC | direction;
 445                 i->kvec = (struct kvec *)iov;
 446         } else {
 447                 i->type = ITER_IOVEC | direction;
 448                 i->iov = iov;
 449         }
 450         i->nr_segs = nr_segs;
 451         i->iov_offset = 0;
 452         i->count = count;
 453 }
 454 EXPORT_SYMBOL(iov_iter_init);
 455
 456 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
 457 {
 458         char *from = kmap_atomic(page);
 459         memcpy(to, from + offset, len);
 460         kunmap_atomic(from);
 461 }
 462
 463 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
 464 {
 465         char *to = kmap_atomic(page);
 466         memcpy(to + offset, from, len);
 467         kunmap_atomic(to);
 468 }
 469
 470 static void memzero_page(struct page *page, size_t offset, size_t len)
 471 {
 472         char *addr = kmap_atomic(page);
 473         memset(addr + offset, 0, len);
 474         kunmap_atomic(addr);
 475 }
 476
 477 static inline bool allocated(struct pipe_buffer *buf)
 478 {
 479         return buf->ops == &default_pipe_buf_ops;
 480 }
 481
 482 static inline void data_start(const struct iov_iter *i, int *idxp, size_t *offp)
 483 {
 484         size_t off = i->iov_offset;
 485         int idx = i->idx;
 486         if (off && (!allocated(&i->pipe->bufs[idx]) || off == PAGE_SIZE)) {
 487                 idx = next_idx(idx, i->pipe);
 488                 off = 0;
 489         }
 490         *idxp = idx;
 491         *offp = off;
 492 }
 493
 494 static size_t push_pipe(struct iov_iter *i, size_t size,
 495                         int *idxp, size_t *offp)
 496 {
 497         struct pipe_inode_info *pipe = i->pipe;
 498         size_t off;
 499         int idx;
 500         ssize_t left;
 501
 502         if (unlikely(size > i->count))
 503                 size = i->count;
 504         if (unlikely(!size))
 505                 return 0;
 506
 507         left = size;
 508         data_start(i, &idx, &off);
 509         *idxp = idx;
 510         *offp = off;
 511         if (off) {
 512                 left -= PAGE_SIZE - off;
 513                 if (left <= 0) {
 514                         pipe->bufs[idx].len += size;
 515                         return size;
 516                 }
 517                 pipe->bufs[idx].len = PAGE_SIZE;
 518                 idx = next_idx(idx, pipe);
 519         }
 520         while (idx != pipe->curbuf || !pipe->nrbufs) {
 521                 struct page *page = alloc_page(GFP_USER);
 522                 if (!page)
 523                         break;
 524                 pipe->nrbufs++;
 525                 pipe->bufs[idx].ops = &default_pipe_buf_ops;
 526                 pipe->bufs[idx].page = page;
 527                 pipe->bufs[idx].offset = 0;
 528                 if (left <= PAGE_SIZE) {
 529                         pipe->bufs[idx].len = left;
 530                         return size;
 531                 }
 532                 pipe->bufs[idx].len = PAGE_SIZE;
 533                 left -= PAGE_SIZE;
 534                 idx = next_idx(idx, pipe);
 535         }
 536         return size - left;
 537 }
 538
 539 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 540                                 struct iov_iter *i)
 541 {
 542         struct pipe_inode_info *pipe = i->pipe;
 543         size_t n, off;
 544         int idx;
 545
 546         if (!sanity(i))
 547                 return 0;
 548
 549         bytes = n = push_pipe(i, bytes, &idx, &off);
 550         if (unlikely(!n))
 551                 return 0;
 552         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 553                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 554                 memcpy_to_page(pipe->bufs[idx].page, off, addr, chunk);
 555                 i->idx = idx;
 556                 i->iov_offset = off + chunk;
 557                 n -= chunk;
 558                 addr += chunk;
 559         }
 560         i->count -= bytes;
 561         return bytes;
 562 }
 563
 564 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 565                               __wsum sum, size_t off)
 566 {
 567         __wsum next = csum_partial_copy_nocheck(from, to, len, 0);
 568         return csum_block_add(sum, next, off);
 569 }
 570
 571 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 572                                 __wsum *csum, struct iov_iter *i)
 573 {
 574         struct pipe_inode_info *pipe = i->pipe;
 575         size_t n, r;
 576         size_t off = 0;
 577         __wsum sum = *csum;
 578         int idx;
 579
 580         if (!sanity(i))
 581                 return 0;
 582
 583         bytes = n = push_pipe(i, bytes, &idx, &r);
 584         if (unlikely(!n))
 585                 return 0;
 586         for ( ; n; idx = next_idx(idx, pipe), r = 0) {
 587                 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
 588                 char *p = kmap_atomic(pipe->bufs[idx].page);
 589                 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
 590                 kunmap_atomic(p);
 591                 i->idx = idx;
 592                 i->iov_offset = r + chunk;
 593                 n -= chunk;
 594                 off += chunk;
 595                 addr += chunk;
 596         }
 597         i->count -= bytes;
 598         *csum = sum;
 599         return bytes;
 600 }
 601
 602 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 603 {
 604         const char *from = addr;
 605         if (unlikely(iov_iter_is_pipe(i)))
 606                 return copy_pipe_to_iter(addr, bytes, i);
 607         if (iter_is_iovec(i))
 608                 might_fault();
 609         iterate_and_advance(i, bytes, v,
 610                 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 611                 memcpy_to_page(v.bv_page, v.bv_offset,
 612                                (from += v.bv_len) - v.bv_len, v.bv_len),
 613                 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
 614         )
 615
 616         return bytes;
 617 }
 618 EXPORT_SYMBOL(_copy_to_iter);
 619
 620 #ifdef CONFIG_ARCH_HAS_UACCESS_MCSAFE
 621 static int copyout_mcsafe(void __user *to, const void *from, size_t n)
 622 {
 623         if (access_ok(to, n)) {
 624                 kasan_check_read(from, n);
 625                 n = copy_to_user_mcsafe((__force void *) to, from, n);
 626         }
 627         return n;
 628 }
 629
 630 static unsigned long memcpy_mcsafe_to_page(struct page *page, size_t offset,
 631                 const char *from, size_t len)
 632 {
 633         unsigned long ret;
 634         char *to;
 635
 636         to = kmap_atomic(page);
 637         ret = memcpy_mcsafe(to + offset, from, len);
 638         kunmap_atomic(to);
 639
 640         return ret;
 641 }
 642
 643 static size_t copy_pipe_to_iter_mcsafe(const void *addr, size_t bytes,
 644                                 struct iov_iter *i)
 645 {
 646         struct pipe_inode_info *pipe = i->pipe;
 647         size_t n, off, xfer = 0;
 648         int idx;
 649
 650         if (!sanity(i))
 651                 return 0;
 652
 653         bytes = n = push_pipe(i, bytes, &idx, &off);
 654         if (unlikely(!n))
 655                 return 0;
 656         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 657                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 658                 unsigned long rem;
 659
 660                 rem = memcpy_mcsafe_to_page(pipe->bufs[idx].page, off, addr,
 661                                 chunk);
 662                 i->idx = idx;
 663                 i->iov_offset = off + chunk - rem;
 664                 xfer += chunk - rem;
 665                 if (rem)
 666                         break;
 667                 n -= chunk;
 668                 addr += chunk;
 669         }
 670         i->count -= xfer;
 671         return xfer;
 672 }
 673
 674 /**
 675  * _copy_to_iter_mcsafe - copy to user with source-read error exception handling
 676  * @addr: source kernel address
 677  * @bytes: total transfer length
 678  * @iter: destination iterator
 679  *
 680  * The pmem driver arranges for filesystem-dax to use this facility via
 681  * dax_copy_to_iter() for protecting read/write to persistent memory.
 682  * Unless / until an architecture can guarantee identical performance
 683  * between _copy_to_iter_mcsafe() and _copy_to_iter() it would be a
 684  * performance regression to switch more users to the mcsafe version.
 685  *
 686  * Otherwise, the main differences between this and typical _copy_to_iter().
 687  *
 688  * * Typical tail/residue handling after a fault retries the copy
 689  *   byte-by-byte until the fault happens again. Re-triggering machine
 690  *   checks is potentially fatal so the implementation uses source
 691  *   alignment and poison alignment assumptions to avoid re-triggering
 692  *   hardware exceptions.
 693  *
 694  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 695  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 696  *   a short copy.
 697  *
 698  * See MCSAFE_TEST for self-test.
 699  */
 700 size_t _copy_to_iter_mcsafe(const void *addr, size_t bytes, struct iov_iter *i)
 701 {
 702         const char *from = addr;
 703         unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
 704
 705         if (unlikely(iov_iter_is_pipe(i)))
 706                 return copy_pipe_to_iter_mcsafe(addr, bytes, i);
 707         if (iter_is_iovec(i))
 708                 might_fault();
 709         iterate_and_advance(i, bytes, v,
 710                 copyout_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 711                 ({
 712                 rem = memcpy_mcsafe_to_page(v.bv_page, v.bv_offset,
 713                                (from += v.bv_len) - v.bv_len, v.bv_len);
 714                 if (rem) {
 715                         curr_addr = (unsigned long) from;
 716                         bytes = curr_addr - s_addr - rem;
 717                         return bytes;
 718                 }
 719                 }),
 720                 ({
 721                 rem = memcpy_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len,
 722                                 v.iov_len);
 723                 if (rem) {
 724                         curr_addr = (unsigned long) from;
 725                         bytes = curr_addr - s_addr - rem;
 726                         return bytes;
 727                 }
 728                 })
 729         )
 730
 731         return bytes;
 732 }
 733 EXPORT_SYMBOL_GPL(_copy_to_iter_mcsafe);
 734 #endif /* CONFIG_ARCH_HAS_UACCESS_MCSAFE */
 735
 736 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 737 {
 738         char *to = addr;
 739         if (unlikely(iov_iter_is_pipe(i))) {
 740                 WARN_ON(1);
 741                 return 0;
 742         }
 743         if (iter_is_iovec(i))
 744                 might_fault();
 745         iterate_and_advance(i, bytes, v,
 746                 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 747                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 748                                  v.bv_offset, v.bv_len),
 749                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 750         )
 751
 752         return bytes;
 753 }
 754 EXPORT_SYMBOL(_copy_from_iter);
 755
 756 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
 757 {
 758         char *to = addr;
 759         if (unlikely(iov_iter_is_pipe(i))) {
 760                 WARN_ON(1);
 761                 return false;
 762         }
 763         if (unlikely(i->count < bytes))
 764                 return false;
 765
 766         if (iter_is_iovec(i))
 767                 might_fault();
 768         iterate_all_kinds(i, bytes, v, ({
 769                 if (copyin((to += v.iov_len) - v.iov_len,
 770                                       v.iov_base, v.iov_len))
 771                         return false;
 772                 0;}),
 773                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 774                                  v.bv_offset, v.bv_len),
 775                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 776         )
 777
 778         iov_iter_advance(i, bytes);
 779         return true;
 780 }
 781 EXPORT_SYMBOL(_copy_from_iter_full);
 782
 783 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 784 {
 785         char *to = addr;
 786         if (unlikely(iov_iter_is_pipe(i))) {
 787                 WARN_ON(1);
 788                 return 0;
 789         }
 790         iterate_and_advance(i, bytes, v,
 791                 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 792                                          v.iov_base, v.iov_len),
 793                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 794                                  v.bv_offset, v.bv_len),
 795                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 796         )
 797
 798         return bytes;
 799 }
 800 EXPORT_SYMBOL(_copy_from_iter_nocache);
 801
 802 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 803 /**
 804  * _copy_from_iter_flushcache - write destination through cpu cache
 805  * @addr: destination kernel address
 806  * @bytes: total transfer length
 807  * @iter: source iterator
 808  *
 809  * The pmem driver arranges for filesystem-dax to use this facility via
 810  * dax_copy_from_iter() for ensuring that writes to persistent memory
 811  * are flushed through the CPU cache. It is differentiated from
 812  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 813  * all iterator types. The _copy_from_iter_nocache() only attempts to
 814  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 815  * instructions that strand dirty-data in the cache.
 816  */
 817 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 818 {
 819         char *to = addr;
 820         if (unlikely(iov_iter_is_pipe(i))) {
 821                 WARN_ON(1);
 822                 return 0;
 823         }
 824         iterate_and_advance(i, bytes, v,
 825                 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
 826                                          v.iov_base, v.iov_len),
 827                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 828                                  v.bv_offset, v.bv_len),
 829                 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
 830                         v.iov_len)
 831         )
 832
 833         return bytes;
 834 }
 835 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 836 #endif
 837
 838 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
 839 {
 840         char *to = addr;
 841         if (unlikely(iov_iter_is_pipe(i))) {
 842                 WARN_ON(1);
 843                 return false;
 844         }
 845         if (unlikely(i->count < bytes))
 846                 return false;
 847         iterate_all_kinds(i, bytes, v, ({
 848                 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 849                                              v.iov_base, v.iov_len))
 850                         return false;
 851                 0;}),
 852                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 853                                  v.bv_offset, v.bv_len),
 854                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 855         )
 856
 857         iov_iter_advance(i, bytes);
 858         return true;
 859 }
 860 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
 861
 862 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 863 {
 864         struct page *head = compound_head(page);
 865         size_t v = n + offset + page_address(page) - page_address(head);
 866
 867         if (likely(n <= v && v <= (PAGE_SIZE << compound_order(head))))
 868                 return true;
 869         WARN_ON(1);
 870         return false;
 871 }
 872
 873 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 874                          struct iov_iter *i)
 875 {
 876         if (unlikely(!page_copy_sane(page, offset, bytes)))
 877                 return 0;
 878         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 879                 void *kaddr = kmap_atomic(page);
 880                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 881                 kunmap_atomic(kaddr);
 882                 return wanted;
 883         } else if (unlikely(iov_iter_is_discard(i)))
 884                 return bytes;
 885         else if (likely(!iov_iter_is_pipe(i)))
 886                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 887         else
 888                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 889 }
 890 EXPORT_SYMBOL(copy_page_to_iter);
 891
 892 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 893                          struct iov_iter *i)
 894 {
 895         if (unlikely(!page_copy_sane(page, offset, bytes)))
 896                 return 0;
 897         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 898                 WARN_ON(1);
 899                 return 0;
 900         }
 901         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 902                 void *kaddr = kmap_atomic(page);
 903                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 904                 kunmap_atomic(kaddr);
 905                 return wanted;
 906         } else
 907                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 908 }
 909 EXPORT_SYMBOL(copy_page_from_iter);
 910
 911 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 912 {
 913         struct pipe_inode_info *pipe = i->pipe;
 914         size_t n, off;
 915         int idx;
 916
 917         if (!sanity(i))
 918                 return 0;
 919
 920         bytes = n = push_pipe(i, bytes, &idx, &off);
 921         if (unlikely(!n))
 922                 return 0;
 923
 924         for ( ; n; idx = next_idx(idx, pipe), off = 0) {
 925                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 926                 memzero_page(pipe->bufs[idx].page, off, chunk);
 927                 i->idx = idx;
 928                 i->iov_offset = off + chunk;
 929                 n -= chunk;
 930         }
 931         i->count -= bytes;
 932         return bytes;
 933 }
 934
 935 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 936 {
 937         if (unlikely(iov_iter_is_pipe(i)))
 938                 return pipe_zero(bytes, i);
 939         iterate_and_advance(i, bytes, v,
 940                 clear_user(v.iov_base, v.iov_len),
 941                 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
 942                 memset(v.iov_base, 0, v.iov_len)
 943         )
 944
 945         return bytes;
 946 }
 947 EXPORT_SYMBOL(iov_iter_zero);
 948
 949 size_t iov_iter_copy_from_user_atomic(struct page *page,
 950                 struct iov_iter *i, unsigned long offset, size_t bytes)
 951 {
 952         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 953         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 954                 kunmap_atomic(kaddr);
 955                 return 0;
 956         }
 957         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 958                 kunmap_atomic(kaddr);
 959                 WARN_ON(1);
 960                 return 0;
 961         }
 962         iterate_all_kinds(i, bytes, v,
 963                 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 964                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
 965                                  v.bv_offset, v.bv_len),
 966                 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 967         )
 968         kunmap_atomic(kaddr);
 969         return bytes;
 970 }
 971 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
 972
 973 static inline void pipe_truncate(struct iov_iter *i)
 974 {
 975         struct pipe_inode_info *pipe = i->pipe;
 976         if (pipe->nrbufs) {
 977                 size_t off = i->iov_offset;
 978                 int idx = i->idx;
 979                 int nrbufs = (idx - pipe->curbuf) & (pipe->buffers - 1);
 980                 if (off) {
 981                         pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
 982                         idx = next_idx(idx, pipe);
 983                         nrbufs++;
 984                 }
 985                 while (pipe->nrbufs > nrbufs) {
 986                         pipe_buf_release(pipe, &pipe->bufs[idx]);
 987                         idx = next_idx(idx, pipe);
 988                         pipe->nrbufs--;
 989                 }
 990         }
 991 }
 992
 993 static void pipe_advance(struct iov_iter *i, size_t size)
 994 {
 995         struct pipe_inode_info *pipe = i->pipe;
 996         if (unlikely(i->count < size))
 997                 size = i->count;
 998         if (size) {
 999                 struct pipe_buffer *buf;
1000                 size_t off = i->iov_offset, left = size;
1001                 int idx = i->idx;
1002                 if (off) /* make it relative to the beginning of buffer */
1003                         left += off - pipe->bufs[idx].offset;
1004                 while (1) {
1005                         buf = &pipe->bufs[idx];
1006                         if (left <= buf->len)
1007                                 break;
1008                         left -= buf->len;
1009                         idx = next_idx(idx, pipe);
1010                 }
1011                 i->idx = idx;
1012                 i->iov_offset = buf->offset + left;
1013         }
1014         i->count -= size;
1015         /* ... and discard everything past that point */
1016         pipe_truncate(i);
1017 }
1018
1019 void iov_iter_advance(struct iov_iter *i, size_t size)
1020 {
1021         if (unlikely(iov_iter_is_pipe(i))) {
1022                 pipe_advance(i, size);
1023                 return;
1024         }
1025         if (unlikely(iov_iter_is_discard(i))) {
1026                 i->count -= size;
1027                 return;
1028         }
1029         iterate_and_advance(i, size, v, 0, 0, 0)
1030 }
1031 EXPORT_SYMBOL(iov_iter_advance);
1032
1033 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1034 {
1035         if (!unroll)
1036                 return;
1037         if (WARN_ON(unroll > MAX_RW_COUNT))
1038                 return;
1039         i->count += unroll;
1040         if (unlikely(iov_iter_is_pipe(i))) {
1041                 struct pipe_inode_info *pipe = i->pipe;
1042                 int idx = i->idx;
1043                 size_t off = i->iov_offset;
1044                 while (1) {
1045                         size_t n = off - pipe->bufs[idx].offset;
1046                         if (unroll < n) {
1047                                 off -= unroll;
1048                                 break;
1049                         }
1050                         unroll -= n;
1051                         if (!unroll && idx == i->start_idx) {
1052                                 off = 0;
1053                                 break;
1054                         }
1055                         if (!idx--)
1056                                 idx = pipe->buffers - 1;
1057                         off = pipe->bufs[idx].offset + pipe->bufs[idx].len;
1058                 }
1059                 i->iov_offset = off;
1060                 i->idx = idx;
1061                 pipe_truncate(i);
1062                 return;
1063         }
1064         if (unlikely(iov_iter_is_discard(i)))
1065                 return;
1066         if (unroll <= i->iov_offset) {
1067                 i->iov_offset -= unroll;
1068                 return;
1069         }
1070         unroll -= i->iov_offset;
1071         if (iov_iter_is_bvec(i)) {
1072                 const struct bio_vec *bvec = i->bvec;
1073                 while (1) {
1074                         size_t n = (--bvec)->bv_len;
1075                         i->nr_segs++;
1076                         if (unroll <= n) {
1077                                 i->bvec = bvec;
1078                                 i->iov_offset = n - unroll;
1079                                 return;
1080                         }
1081                         unroll -= n;
1082                 }
1083         } else { /* same logics for iovec and kvec */
1084                 const struct iovec *iov = i->iov;
1085                 while (1) {
1086                         size_t n = (--iov)->iov_len;
1087                         i->nr_segs++;
1088                         if (unroll <= n) {
1089                                 i->iov = iov;
1090                                 i->iov_offset = n - unroll;
1091                                 return;
1092                         }
1093                         unroll -= n;
1094                 }
1095         }
1096 }
1097 EXPORT_SYMBOL(iov_iter_revert);
1098
1099 /*
1100  * Return the count of just the current iov_iter segment.
1101  */
1102 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1103 {
1104         if (unlikely(iov_iter_is_pipe(i)))
1105                 return i->count;        // it is a silly place, anyway
1106         if (i->nr_segs == 1)
1107                 return i->count;
1108         if (unlikely(iov_iter_is_discard(i)))
1109                 return i->count;
1110         else if (iov_iter_is_bvec(i))
1111                 return min(i->count, i->bvec->bv_len - i->iov_offset);
1112         else
1113                 return min(i->count, i->iov->iov_len - i->iov_offset);
1114 }
1115 EXPORT_SYMBOL(iov_iter_single_seg_count);
1116
1117 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1118                         const struct kvec *kvec, unsigned long nr_segs,
1119                         size_t count)
1120 {
1121         WARN_ON(direction & ~(READ | WRITE));
1122         i->type = ITER_KVEC | (direction & (READ | WRITE));
1123         i->kvec = kvec;
1124         i->nr_segs = nr_segs;
1125         i->iov_offset = 0;
1126         i->count = count;
1127 }
1128 EXPORT_SYMBOL(iov_iter_kvec);
1129
1130 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1131                         const struct bio_vec *bvec, unsigned long nr_segs,
1132                         size_t count)
1133 {
1134         WARN_ON(direction & ~(READ | WRITE));
1135         i->type = ITER_BVEC | (direction & (READ | WRITE));
1136         i->bvec = bvec;
1137         i->nr_segs = nr_segs;
1138         i->iov_offset = 0;
1139         i->count = count;
1140 }
1141 EXPORT_SYMBOL(iov_iter_bvec);
1142
1143 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1144                         struct pipe_inode_info *pipe,
1145                         size_t count)
1146 {
1147         BUG_ON(direction != READ);
1148         WARN_ON(pipe->nrbufs == pipe->buffers);
1149         i->type = ITER_PIPE | READ;
1150         i->pipe = pipe;
1151         i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1152         i->iov_offset = 0;
1153         i->count = count;
1154         i->start_idx = i->idx;
1155 }
1156 EXPORT_SYMBOL(iov_iter_pipe);
1157
1158 /**
1159  * iov_iter_discard - Initialise an I/O iterator that discards data
1160  * @i: The iterator to initialise.
1161  * @direction: The direction of the transfer.
1162  * @count: The size of the I/O buffer in bytes.
1163  *
1164  * Set up an I/O iterator that just discards everything that's written to it.
1165  * It's only available as a READ iterator.
1166  */
1167 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1168 {
1169         BUG_ON(direction != READ);
1170         i->type = ITER_DISCARD | READ;
1171         i->count = count;
1172         i->iov_offset = 0;
1173 }
1174 EXPORT_SYMBOL(iov_iter_discard);
1175
1176 unsigned long iov_iter_alignment(const struct iov_iter *i)
1177 {
1178         unsigned long res = 0;
1179         size_t size = i->count;
1180
1181         if (unlikely(iov_iter_is_pipe(i))) {
1182                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
1183                         return size | i->iov_offset;
1184                 return size;
1185         }
1186         iterate_all_kinds(i, size, v,
1187                 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1188                 res |= v.bv_offset | v.bv_len,
1189                 res |= (unsigned long)v.iov_base | v.iov_len
1190         )
1191         return res;
1192 }
1193 EXPORT_SYMBOL(iov_iter_alignment);
1194
1195 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1196 {
1197         unsigned long res = 0;
1198         size_t size = i->count;
1199
1200         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1201                 WARN_ON(1);
1202                 return ~0U;
1203         }
1204
1205         iterate_all_kinds(i, size, v,
1206                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1207                         (size != v.iov_len ? size : 0), 0),
1208                 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1209                         (size != v.bv_len ? size : 0)),
1210                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1211                         (size != v.iov_len ? size : 0))
1212                 );
1213         return res;
1214 }
1215 EXPORT_SYMBOL(iov_iter_gap_alignment);
1216
1217 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1218                                 size_t maxsize,
1219                                 struct page **pages,
1220                                 int idx,
1221                                 size_t *start)
1222 {
1223         struct pipe_inode_info *pipe = i->pipe;
1224         ssize_t n = push_pipe(i, maxsize, &idx, start);
1225         if (!n)
1226                 return -EFAULT;
1227
1228         maxsize = n;
1229         n += *start;
1230         while (n > 0) {
1231                 get_page(*pages++ = pipe->bufs[idx].page);
1232                 idx = next_idx(idx, pipe);
1233                 n -= PAGE_SIZE;
1234         }
1235
1236         return maxsize;
1237 }
1238
1239 static ssize_t pipe_get_pages(struct iov_iter *i,
1240                    struct page **pages, size_t maxsize, unsigned maxpages,
1241                    size_t *start)
1242 {
1243         unsigned npages;
1244         size_t capacity;
1245         int idx;
1246
1247         if (!maxsize)
1248                 return 0;
1249
1250         if (!sanity(i))
1251                 return -EFAULT;
1252
1253         data_start(i, &idx, start);
1254         /* some of this one + all after this one */
1255         npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1256         capacity = min(npages,maxpages) * PAGE_SIZE - *start;
1257
1258         return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
1259 }
1260
1261 ssize_t iov_iter_get_pages(struct iov_iter *i,
1262                    struct page **pages, size_t maxsize, unsigned maxpages,
1263                    size_t *start)
1264 {
1265         if (maxsize > i->count)
1266                 maxsize = i->count;
1267
1268         if (unlikely(iov_iter_is_pipe(i)))
1269                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1270         if (unlikely(iov_iter_is_discard(i)))
1271                 return -EFAULT;
1272
1273         iterate_all_kinds(i, maxsize, v, ({
1274                 unsigned long addr = (unsigned long)v.iov_base;
1275                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1276                 int n;
1277                 int res;
1278
1279                 if (len > maxpages * PAGE_SIZE)
1280                         len = maxpages * PAGE_SIZE;
1281                 addr &= ~(PAGE_SIZE - 1);
1282                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1283                 res = get_user_pages_fast(addr, n, iov_iter_rw(i) != WRITE, pages);
1284                 if (unlikely(res < 0))
1285                         return res;
1286                 return (res == n ? len : res * PAGE_SIZE) - *start;
1287         0;}),({
1288                 /* can't be more than PAGE_SIZE */
1289                 *start = v.bv_offset;
1290                 get_page(*pages = v.bv_page);
1291                 return v.bv_len;
1292         }),({
1293                 return -EFAULT;
1294         })
1295         )
1296         return 0;
1297 }
1298 EXPORT_SYMBOL(iov_iter_get_pages);
1299
1300 static struct page **get_pages_array(size_t n)
1301 {
1302         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1303 }
1304
1305 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1306                    struct page ***pages, size_t maxsize,
1307                    size_t *start)
1308 {
1309         struct page **p;
1310         ssize_t n;
1311         int idx;
1312         int npages;
1313
1314         if (!maxsize)
1315                 return 0;
1316
1317         if (!sanity(i))
1318                 return -EFAULT;
1319
1320         data_start(i, &idx, start);
1321         /* some of this one + all after this one */
1322         npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1323         n = npages * PAGE_SIZE - *start;
1324         if (maxsize > n)
1325                 maxsize = n;
1326         else
1327                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1328         p = get_pages_array(npages);
1329         if (!p)
1330                 return -ENOMEM;
1331         n = __pipe_get_pages(i, maxsize, p, idx, start);
1332         if (n > 0)
1333                 *pages = p;
1334         else
1335                 kvfree(p);
1336         return n;
1337 }
1338
1339 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1340                    struct page ***pages, size_t maxsize,
1341                    size_t *start)
1342 {
1343         struct page **p;
1344
1345         if (maxsize > i->count)
1346                 maxsize = i->count;
1347
1348         if (unlikely(iov_iter_is_pipe(i)))
1349                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1350         if (unlikely(iov_iter_is_discard(i)))
1351                 return -EFAULT;
1352
1353         iterate_all_kinds(i, maxsize, v, ({
1354                 unsigned long addr = (unsigned long)v.iov_base;
1355                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1356                 int n;
1357                 int res;
1358
1359                 addr &= ~(PAGE_SIZE - 1);
1360                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1361                 p = get_pages_array(n);
1362                 if (!p)
1363                         return -ENOMEM;
1364                 res = get_user_pages_fast(addr, n, iov_iter_rw(i) != WRITE, p);
1365                 if (unlikely(res < 0)) {
1366                         kvfree(p);
1367                         return res;
1368                 }
1369                 *pages = p;
1370                 return (res == n ? len : res * PAGE_SIZE) - *start;
1371         0;}),({
1372                 /* can't be more than PAGE_SIZE */
1373                 *start = v.bv_offset;
1374                 *pages = p = get_pages_array(1);
1375                 if (!p)
1376                         return -ENOMEM;
1377                 get_page(*p = v.bv_page);
1378                 return v.bv_len;
1379         }),({
1380                 return -EFAULT;
1381         })
1382         )
1383         return 0;
1384 }
1385 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1386
1387 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1388                                struct iov_iter *i)
1389 {
1390         char *to = addr;
1391         __wsum sum, next;
1392         size_t off = 0;
1393         sum = *csum;
1394         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1395                 WARN_ON(1);
1396                 return 0;
1397         }
1398         iterate_and_advance(i, bytes, v, ({
1399                 int err = 0;
1400                 next = csum_and_copy_from_user(v.iov_base,
1401                                                (to += v.iov_len) - v.iov_len,
1402                                                v.iov_len, 0, &err);
1403                 if (!err) {
1404                         sum = csum_block_add(sum, next, off);
1405                         off += v.iov_len;
1406                 }
1407                 err ? v.iov_len : 0;
1408         }), ({
1409                 char *p = kmap_atomic(v.bv_page);
1410                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1411                                       p + v.bv_offset, v.bv_len,
1412                                       sum, off);
1413                 kunmap_atomic(p);
1414                 off += v.bv_len;
1415         }),({
1416                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1417                                       v.iov_base, v.iov_len,
1418                                       sum, off);
1419                 off += v.iov_len;
1420         })
1421         )
1422         *csum = sum;
1423         return bytes;
1424 }
1425 EXPORT_SYMBOL(csum_and_copy_from_iter);
1426
1427 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1428                                struct iov_iter *i)
1429 {
1430         char *to = addr;
1431         __wsum sum, next;
1432         size_t off = 0;
1433         sum = *csum;
1434         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1435                 WARN_ON(1);
1436                 return false;
1437         }
1438         if (unlikely(i->count < bytes))
1439                 return false;
1440         iterate_all_kinds(i, bytes, v, ({
1441                 int err = 0;
1442                 next = csum_and_copy_from_user(v.iov_base,
1443                                                (to += v.iov_len) - v.iov_len,
1444                                                v.iov_len, 0, &err);
1445                 if (err)
1446                         return false;
1447                 sum = csum_block_add(sum, next, off);
1448                 off += v.iov_len;
1449                 0;
1450         }), ({
1451                 char *p = kmap_atomic(v.bv_page);
1452                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1453                                       p + v.bv_offset, v.bv_len,
1454                                       sum, off);
1455                 kunmap_atomic(p);
1456                 off += v.bv_len;
1457         }),({
1458                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1459                                       v.iov_base, v.iov_len,
1460                                       sum, off);
1461                 off += v.iov_len;
1462         })
1463         )
1464         *csum = sum;
1465         iov_iter_advance(i, bytes);
1466         return true;
1467 }
1468 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1469
1470 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *csump,
1471                              struct iov_iter *i)
1472 {
1473         const char *from = addr;
1474         __wsum *csum = csump;
1475         __wsum sum, next;
1476         size_t off = 0;
1477
1478         if (unlikely(iov_iter_is_pipe(i)))
1479                 return csum_and_copy_to_pipe_iter(addr, bytes, csum, i);
1480
1481         sum = *csum;
1482         if (unlikely(iov_iter_is_discard(i))) {
1483                 WARN_ON(1);     /* for now */
1484                 return 0;
1485         }
1486         iterate_and_advance(i, bytes, v, ({
1487                 int err = 0;
1488                 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1489                                              v.iov_base,
1490                                              v.iov_len, 0, &err);
1491                 if (!err) {
1492                         sum = csum_block_add(sum, next, off);
1493                         off += v.iov_len;
1494                 }
1495                 err ? v.iov_len : 0;
1496         }), ({
1497                 char *p = kmap_atomic(v.bv_page);
1498                 sum = csum_and_memcpy(p + v.bv_offset,
1499                                       (from += v.bv_len) - v.bv_len,
1500                                       v.bv_len, sum, off);
1501                 kunmap_atomic(p);
1502                 off += v.bv_len;
1503         }),({
1504                 sum = csum_and_memcpy(v.iov_base,
1505                                      (from += v.iov_len) - v.iov_len,
1506                                      v.iov_len, sum, off);
1507                 off += v.iov_len;
1508         })
1509         )
1510         *csum = sum;
1511         return bytes;
1512 }
1513 EXPORT_SYMBOL(csum_and_copy_to_iter);
1514
1515 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1516                 struct iov_iter *i)
1517 {
1518         struct ahash_request *hash = hashp;
1519         struct scatterlist sg;
1520         size_t copied;
1521
1522         copied = copy_to_iter(addr, bytes, i);
1523         sg_init_one(&sg, addr, copied);
1524         ahash_request_set_crypt(hash, &sg, NULL, copied);
1525         crypto_ahash_update(hash);
1526         return copied;
1527 }
1528 EXPORT_SYMBOL(hash_and_copy_to_iter);
1529
1530 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1531 {
1532         size_t size = i->count;
1533         int npages = 0;
1534
1535         if (!size)
1536                 return 0;
1537         if (unlikely(iov_iter_is_discard(i)))
1538                 return 0;
1539
1540         if (unlikely(iov_iter_is_pipe(i))) {
1541                 struct pipe_inode_info *pipe = i->pipe;
1542                 size_t off;
1543                 int idx;
1544
1545                 if (!sanity(i))
1546                         return 0;
1547
1548                 data_start(i, &idx, &off);
1549                 /* some of this one + all after this one */
1550                 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1551                 if (npages >= maxpages)
1552                         return maxpages;
1553         } else iterate_all_kinds(i, size, v, ({
1554                 unsigned long p = (unsigned long)v.iov_base;
1555                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1556                         - p / PAGE_SIZE;
1557                 if (npages >= maxpages)
1558                         return maxpages;
1559         0;}),({
1560                 npages++;
1561                 if (npages >= maxpages)
1562                         return maxpages;
1563         }),({
1564                 unsigned long p = (unsigned long)v.iov_base;
1565                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1566                         - p / PAGE_SIZE;
1567                 if (npages >= maxpages)
1568                         return maxpages;
1569         })
1570         )
1571         return npages;
1572 }
1573 EXPORT_SYMBOL(iov_iter_npages);
1574
1575 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1576 {
1577         *new = *old;
1578         if (unlikely(iov_iter_is_pipe(new))) {
1579                 WARN_ON(1);
1580                 return NULL;
1581         }
1582         if (unlikely(iov_iter_is_discard(new)))
1583                 return NULL;
1584         if (iov_iter_is_bvec(new))
1585                 return new->bvec = kmemdup(new->bvec,
1586                                     new->nr_segs * sizeof(struct bio_vec),
1587                                     flags);
1588         else
1589                 /* iovec and kvec have identical layout */
1590                 return new->iov = kmemdup(new->iov,
1591                                    new->nr_segs * sizeof(struct iovec),
1592                                    flags);
1593 }
1594 EXPORT_SYMBOL(dup_iter);
1595
1596 /**
1597  * import_iovec() - Copy an array of &struct iovec from userspace
1598  *     into the kernel, check that it is valid, and initialize a new
1599  *     &struct iov_iter iterator to access it.
1600  *
1601  * @type: One of %READ or %WRITE.
1602  * @uvector: Pointer to the userspace array.
1603  * @nr_segs: Number of elements in userspace array.
1604  * @fast_segs: Number of elements in @iov.
1605  * @iov: (input and output parameter) Pointer to pointer to (usually small
1606  *     on-stack) kernel array.
1607  * @i: Pointer to iterator that will be initialized on success.
1608  *
1609  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1610  * then this function places %NULL in *@iov on return. Otherwise, a new
1611  * array will be allocated and the result placed in *@iov. This means that
1612  * the caller may call kfree() on *@iov regardless of whether the small
1613  * on-stack array was used or not (and regardless of whether this function
1614  * returns an error or not).
1615  *
1616  * Return: 0 on success or negative error code on error.
1617  */
1618 int import_iovec(int type, const struct iovec __user * uvector,
1619                  unsigned nr_segs, unsigned fast_segs,
1620                  struct iovec **iov, struct iov_iter *i)
1621 {
1622         ssize_t n;
1623         struct iovec *p;
1624         n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1625                                   *iov, &p);
1626         if (n < 0) {
1627                 if (p != *iov)
1628                         kfree(p);
1629                 *iov = NULL;
1630                 return n;
1631         }
1632         iov_iter_init(i, type, p, nr_segs, n);
1633         *iov = p == *iov ? NULL : p;
1634         return 0;
1635 }
1636 EXPORT_SYMBOL(import_iovec);
1637
1638 #ifdef CONFIG_COMPAT
1639 #include <linux/compat.h>
1640
1641 int compat_import_iovec(int type, const struct compat_iovec __user * uvector,
1642                  unsigned nr_segs, unsigned fast_segs,
1643                  struct iovec **iov, struct iov_iter *i)
1644 {
1645         ssize_t n;
1646         struct iovec *p;
1647         n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1648                                   *iov, &p);
1649         if (n < 0) {
1650                 if (p != *iov)
1651                         kfree(p);
1652                 *iov = NULL;
1653                 return n;
1654         }
1655         iov_iter_init(i, type, p, nr_segs, n);
1656         *iov = p == *iov ? NULL : p;
1657         return 0;
1658 }
1659 #endif
1660
1661 int import_single_range(int rw, void __user *buf, size_t len,
1662                  struct iovec *iov, struct iov_iter *i)
1663 {
1664         if (len > MAX_RW_COUNT)
1665                 len = MAX_RW_COUNT;
1666         if (unlikely(!access_ok(buf, len)))
1667                 return -EFAULT;
1668
1669         iov->iov_base = buf;
1670         iov->iov_len = len;
1671         iov_iter_init(i, rw, iov, 1, len);
1672         return 0;
1673 }
1674 EXPORT_SYMBOL(import_single_range);
1675
1676 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1677                             int (*f)(struct kvec *vec, void *context),
1678                             void *context)
1679 {
1680         struct kvec w;
1681         int err = -EINVAL;
1682         if (!bytes)
1683                 return 0;
1684
1685         iterate_all_kinds(i, bytes, v, -EINVAL, ({
1686                 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1687                 w.iov_len = v.bv_len;
1688                 err = f(&w, context);
1689                 kunmap(v.bv_page);
1690                 err;}), ({
1691                 w = v;
1692                 err = f(&w, context);})
1693         )
1694         return err;
1695 }
1696 EXPORT_SYMBOL(iov_iter_for_each_range);