lib/iov_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <linux/export.h>
   3 #include <linux/bvec.h>
   4 #include <linux/uio.h>
   5 #include <linux/pagemap.h>
   6 #include <linux/slab.h>
   7 #include <linux/vmalloc.h>
   8 #include <linux/splice.h>
   9 #include <net/checksum.h>
  10 #include <linux/scatterlist.h>
  11
  12 #define PIPE_PARANOIA /* for now */
  13
  14 #define iterate_iovec(i, n, __v, __p, skip, STEP) {     \
  15         size_t left;                                    \
  16         size_t wanted = n;                              \
  17         __p = i->iov;                                   \
  18         __v.iov_len = min(n, __p->iov_len - skip);      \
  19         if (likely(__v.iov_len)) {                      \
  20                 __v.iov_base = __p->iov_base + skip;    \
  21                 left = (STEP);                          \
  22                 __v.iov_len -= left;                    \
  23                 skip += __v.iov_len;                    \
  24                 n -= __v.iov_len;                       \
  25         } else {                                        \
  26                 left = 0;                               \
  27         }                                               \
  28         while (unlikely(!left && n)) {                  \
  29                 __p++;                                  \
  30                 __v.iov_len = min(n, __p->iov_len);     \
  31                 if (unlikely(!__v.iov_len))             \
  32                         continue;                       \
  33                 __v.iov_base = __p->iov_base;           \
  34                 left = (STEP);                          \
  35                 __v.iov_len -= left;                    \
  36                 skip = __v.iov_len;                     \
  37                 n -= __v.iov_len;                       \
  38         }                                               \
  39         n = wanted - n;                                 \
  40 }
  41
  42 #define iterate_kvec(i, n, __v, __p, skip, STEP) {      \
  43         size_t wanted = n;                              \
  44         __p = i->kvec;                                  \
  45         __v.iov_len = min(n, __p->iov_len - skip);      \
  46         if (likely(__v.iov_len)) {                      \
  47                 __v.iov_base = __p->iov_base + skip;    \
  48                 (void)(STEP);                           \
  49                 skip += __v.iov_len;                    \
  50                 n -= __v.iov_len;                       \
  51         }                                               \
  52         while (unlikely(n)) {                           \
  53                 __p++;                                  \
  54                 __v.iov_len = min(n, __p->iov_len);     \
  55                 if (unlikely(!__v.iov_len))             \
  56                         continue;                       \
  57                 __v.iov_base = __p->iov_base;           \
  58                 (void)(STEP);                           \
  59                 skip = __v.iov_len;                     \
  60                 n -= __v.iov_len;                       \
  61         }                                               \
  62         n = wanted;                                     \
  63 }
  64
  65 #define iterate_bvec(i, n, __v, __bi, skip, STEP) {     \
  66         struct bvec_iter __start;                       \
  67         __start.bi_size = n;                            \
  68         __start.bi_bvec_done = skip;                    \
  69         __start.bi_idx = 0;                             \
  70         for_each_bvec(__v, i->bvec, __bi, __start) {    \
  71                 if (!__v.bv_len)                        \
  72                         continue;                       \
  73                 (void)(STEP);                           \
  74         }                                               \
  75 }
  76
  77 #define iterate_all_kinds(i, n, v, I, B, K) {                   \
  78         if (likely(n)) {                                        \
  79                 size_t skip = i->iov_offset;                    \
  80                 if (unlikely(i->type & ITER_BVEC)) {            \
  81                         struct bio_vec v;                       \
  82                         struct bvec_iter __bi;                  \
  83                         iterate_bvec(i, n, v, __bi, skip, (B))  \
  84                 } else if (unlikely(i->type & ITER_KVEC)) {     \
  85                         const struct kvec *kvec;                \
  86                         struct kvec v;                          \
  87                         iterate_kvec(i, n, v, kvec, skip, (K))  \
  88                 } else if (unlikely(i->type & ITER_DISCARD)) {  \
  89                 } else {                                        \
  90                         const struct iovec *iov;                \
  91                         struct iovec v;                         \
  92                         iterate_iovec(i, n, v, iov, skip, (I))  \
  93                 }                                               \
  94         }                                                       \
  95 }
  96
  97 #define iterate_and_advance(i, n, v, I, B, K) {                 \
  98         if (unlikely(i->count < n))                             \
  99                 n = i->count;                                   \
 100         if (i->count) {                                         \
 101                 size_t skip = i->iov_offset;                    \
 102                 if (unlikely(i->type & ITER_BVEC)) {            \
 103                         const struct bio_vec *bvec = i->bvec;   \
 104                         struct bio_vec v;                       \
 105                         struct bvec_iter __bi;                  \
 106                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 107                         i->bvec = __bvec_iter_bvec(i->bvec, __bi);      \
 108                         i->nr_segs -= i->bvec - bvec;           \
 109                         skip = __bi.bi_bvec_done;               \
 110                 } else if (unlikely(i->type & ITER_KVEC)) {     \
 111                         const struct kvec *kvec;                \
 112                         struct kvec v;                          \
 113                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 114                         if (skip == kvec->iov_len) {            \
 115                                 kvec++;                         \
 116                                 skip = 0;                       \
 117                         }                                       \
 118                         i->nr_segs -= kvec - i->kvec;           \
 119                         i->kvec = kvec;                         \
 120                 } else if (unlikely(i->type & ITER_DISCARD)) {  \
 121                         skip += n;                              \
 122                 } else {                                        \
 123                         const struct iovec *iov;                \
 124                         struct iovec v;                         \
 125                         iterate_iovec(i, n, v, iov, skip, (I))  \
 126                         if (skip == iov->iov_len) {             \
 127                                 iov++;                          \
 128                                 skip = 0;                       \
 129                         }                                       \
 130                         i->nr_segs -= iov - i->iov;             \
 131                         i->iov = iov;                           \
 132                 }                                               \
 133                 i->count -= n;                                  \
 134                 i->iov_offset = skip;                           \
 135         }                                                       \
 136 }
 137
 138 static int copyout(void __user *to, const void *from, size_t n)
 139 {
 140         if (access_ok(to, n)) {
 141                 kasan_check_read(from, n);
 142                 n = raw_copy_to_user(to, from, n);
 143         }
 144         return n;
 145 }
 146
 147 static int copyin(void *to, const void __user *from, size_t n)
 148 {
 149         if (access_ok(from, n)) {
 150                 kasan_check_write(to, n);
 151                 n = raw_copy_from_user(to, from, n);
 152         }
 153         return n;
 154 }
 155
 156 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 157                          struct iov_iter *i)
 158 {
 159         size_t skip, copy, left, wanted;
 160         const struct iovec *iov;
 161         char __user *buf;
 162         void *kaddr, *from;
 163
 164         if (unlikely(bytes > i->count))
 165                 bytes = i->count;
 166
 167         if (unlikely(!bytes))
 168                 return 0;
 169
 170         might_fault();
 171         wanted = bytes;
 172         iov = i->iov;
 173         skip = i->iov_offset;
 174         buf = iov->iov_base + skip;
 175         copy = min(bytes, iov->iov_len - skip);
 176
 177         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 178                 kaddr = kmap_atomic(page);
 179                 from = kaddr + offset;
 180
 181                 /* first chunk, usually the only one */
 182                 left = copyout(buf, from, copy);
 183                 copy -= left;
 184                 skip += copy;
 185                 from += copy;
 186                 bytes -= copy;
 187
 188                 while (unlikely(!left && bytes)) {
 189                         iov++;
 190                         buf = iov->iov_base;
 191                         copy = min(bytes, iov->iov_len);
 192                         left = copyout(buf, from, copy);
 193                         copy -= left;
 194                         skip = copy;
 195                         from += copy;
 196                         bytes -= copy;
 197                 }
 198                 if (likely(!bytes)) {
 199                         kunmap_atomic(kaddr);
 200                         goto done;
 201                 }
 202                 offset = from - kaddr;
 203                 buf += copy;
 204                 kunmap_atomic(kaddr);
 205                 copy = min(bytes, iov->iov_len - skip);
 206         }
 207         /* Too bad - revert to non-atomic kmap */
 208
 209         kaddr = kmap(page);
 210         from = kaddr + offset;
 211         left = copyout(buf, from, copy);
 212         copy -= left;
 213         skip += copy;
 214         from += copy;
 215         bytes -= copy;
 216         while (unlikely(!left && bytes)) {
 217                 iov++;
 218                 buf = iov->iov_base;
 219                 copy = min(bytes, iov->iov_len);
 220                 left = copyout(buf, from, copy);
 221                 copy -= left;
 222                 skip = copy;
 223                 from += copy;
 224                 bytes -= copy;
 225         }
 226         kunmap(page);
 227
 228 done:
 229         if (skip == iov->iov_len) {
 230                 iov++;
 231                 skip = 0;
 232         }
 233         i->count -= wanted - bytes;
 234         i->nr_segs -= iov - i->iov;
 235         i->iov = iov;
 236         i->iov_offset = skip;
 237         return wanted - bytes;
 238 }
 239
 240 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 241                          struct iov_iter *i)
 242 {
 243         size_t skip, copy, left, wanted;
 244         const struct iovec *iov;
 245         char __user *buf;
 246         void *kaddr, *to;
 247
 248         if (unlikely(bytes > i->count))
 249                 bytes = i->count;
 250
 251         if (unlikely(!bytes))
 252                 return 0;
 253
 254         might_fault();
 255         wanted = bytes;
 256         iov = i->iov;
 257         skip = i->iov_offset;
 258         buf = iov->iov_base + skip;
 259         copy = min(bytes, iov->iov_len - skip);
 260
 261         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 262                 kaddr = kmap_atomic(page);
 263                 to = kaddr + offset;
 264
 265                 /* first chunk, usually the only one */
 266                 left = copyin(to, buf, copy);
 267                 copy -= left;
 268                 skip += copy;
 269                 to += copy;
 270                 bytes -= copy;
 271
 272                 while (unlikely(!left && bytes)) {
 273                         iov++;
 274                         buf = iov->iov_base;
 275                         copy = min(bytes, iov->iov_len);
 276                         left = copyin(to, buf, copy);
 277                         copy -= left;
 278                         skip = copy;
 279                         to += copy;
 280                         bytes -= copy;
 281                 }
 282                 if (likely(!bytes)) {
 283                         kunmap_atomic(kaddr);
 284                         goto done;
 285                 }
 286                 offset = to - kaddr;
 287                 buf += copy;
 288                 kunmap_atomic(kaddr);
 289                 copy = min(bytes, iov->iov_len - skip);
 290         }
 291         /* Too bad - revert to non-atomic kmap */
 292
 293         kaddr = kmap(page);
 294         to = kaddr + offset;
 295         left = copyin(to, buf, copy);
 296         copy -= left;
 297         skip += copy;
 298         to += copy;
 299         bytes -= copy;
 300         while (unlikely(!left && bytes)) {
 301                 iov++;
 302                 buf = iov->iov_base;
 303                 copy = min(bytes, iov->iov_len);
 304                 left = copyin(to, buf, copy);
 305                 copy -= left;
 306                 skip = copy;
 307                 to += copy;
 308                 bytes -= copy;
 309         }
 310         kunmap(page);
 311
 312 done:
 313         if (skip == iov->iov_len) {
 314                 iov++;
 315                 skip = 0;
 316         }
 317         i->count -= wanted - bytes;
 318         i->nr_segs -= iov - i->iov;
 319         i->iov = iov;
 320         i->iov_offset = skip;
 321         return wanted - bytes;
 322 }
 323
 324 #ifdef PIPE_PARANOIA
 325 static bool sanity(const struct iov_iter *i)
 326 {
 327         struct pipe_inode_info *pipe = i->pipe;
 328         unsigned int p_head = pipe->head;
 329         unsigned int p_tail = pipe->tail;
 330         unsigned int p_mask = pipe->ring_size - 1;
 331         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 332         unsigned int i_head = i->head;
 333         unsigned int idx;
 334
 335         if (i->iov_offset) {
 336                 struct pipe_buffer *p;
 337                 if (unlikely(p_occupancy == 0))
 338                         goto Bad;       // pipe must be non-empty
 339                 if (unlikely(i_head != p_head - 1))
 340                         goto Bad;       // must be at the last buffer...
 341
 342                 p = &pipe->bufs[i_head & p_mask];
 343                 if (unlikely(p->offset + p->len != i->iov_offset))
 344                         goto Bad;       // ... at the end of segment
 345         } else {
 346                 if (i_head != p_head)
 347                         goto Bad;       // must be right after the last buffer
 348         }
 349         return true;
 350 Bad:
 351         printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
 352         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 353                         p_head, p_tail, pipe->ring_size);
 354         for (idx = 0; idx < pipe->ring_size; idx++)
 355                 printk(KERN_ERR "[%p %p %d %d]\n",
 356                         pipe->bufs[idx].ops,
 357                         pipe->bufs[idx].page,
 358                         pipe->bufs[idx].offset,
 359                         pipe->bufs[idx].len);
 360         WARN_ON(1);
 361         return false;
 362 }
 363 #else
 364 #define sanity(i) true
 365 #endif
 366
 367 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 368                          struct iov_iter *i)
 369 {
 370         struct pipe_inode_info *pipe = i->pipe;
 371         struct pipe_buffer *buf;
 372         unsigned int p_tail = pipe->tail;
 373         unsigned int p_mask = pipe->ring_size - 1;
 374         unsigned int i_head = i->head;
 375         size_t off;
 376
 377         if (unlikely(bytes > i->count))
 378                 bytes = i->count;
 379
 380         if (unlikely(!bytes))
 381                 return 0;
 382
 383         if (!sanity(i))
 384                 return 0;
 385
 386         off = i->iov_offset;
 387         buf = &pipe->bufs[i_head & p_mask];
 388         if (off) {
 389                 if (offset == off && buf->page == page) {
 390                         /* merge with the last one */
 391                         buf->len += bytes;
 392                         i->iov_offset += bytes;
 393                         goto out;
 394                 }
 395                 i_head++;
 396                 buf = &pipe->bufs[i_head & p_mask];
 397         }
 398         if (pipe_full(i_head, p_tail, pipe->max_usage))
 399                 return 0;
 400
 401         buf->ops = &page_cache_pipe_buf_ops;
 402         get_page(page);
 403         buf->page = page;
 404         buf->offset = offset;
 405         buf->len = bytes;
 406
 407         pipe->head = i_head + 1;
 408         i->iov_offset = offset + bytes;
 409         i->head = i_head;
 410 out:
 411         i->count -= bytes;
 412         return bytes;
 413 }
 414
 415 /*
 416  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 417  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 418  *
 419  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 420  * because it is an invalid address).
 421  */
 422 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
 423 {
 424         size_t skip = i->iov_offset;
 425         const struct iovec *iov;
 426         int err;
 427         struct iovec v;
 428
 429         if (!(i->type & (ITER_BVEC|ITER_KVEC))) {
 430                 iterate_iovec(i, bytes, v, iov, skip, ({
 431                         err = fault_in_pages_readable(v.iov_base, v.iov_len);
 432                         if (unlikely(err))
 433                         return err;
 434                 0;}))
 435         }
 436         return 0;
 437 }
 438 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 439
 440 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 441                         const struct iovec *iov, unsigned long nr_segs,
 442                         size_t count)
 443 {
 444         WARN_ON(direction & ~(READ | WRITE));
 445         direction &= READ | WRITE;
 446
 447         /* It will get better.  Eventually... */
 448         if (uaccess_kernel()) {
 449                 i->type = ITER_KVEC | direction;
 450                 i->kvec = (struct kvec *)iov;
 451         } else {
 452                 i->type = ITER_IOVEC | direction;
 453                 i->iov = iov;
 454         }
 455         i->nr_segs = nr_segs;
 456         i->iov_offset = 0;
 457         i->count = count;
 458 }
 459 EXPORT_SYMBOL(iov_iter_init);
 460
 461 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
 462 {
 463         char *from = kmap_atomic(page);
 464         memcpy(to, from + offset, len);
 465         kunmap_atomic(from);
 466 }
 467
 468 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
 469 {
 470         char *to = kmap_atomic(page);
 471         memcpy(to + offset, from, len);
 472         kunmap_atomic(to);
 473 }
 474
 475 static void memzero_page(struct page *page, size_t offset, size_t len)
 476 {
 477         char *addr = kmap_atomic(page);
 478         memset(addr + offset, 0, len);
 479         kunmap_atomic(addr);
 480 }
 481
 482 static inline bool allocated(struct pipe_buffer *buf)
 483 {
 484         return buf->ops == &default_pipe_buf_ops;
 485 }
 486
 487 static inline void data_start(const struct iov_iter *i,
 488                               unsigned int *iter_headp, size_t *offp)
 489 {
 490         unsigned int p_mask = i->pipe->ring_size - 1;
 491         unsigned int iter_head = i->head;
 492         size_t off = i->iov_offset;
 493
 494         if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
 495                     off == PAGE_SIZE)) {
 496                 iter_head++;
 497                 off = 0;
 498         }
 499         *iter_headp = iter_head;
 500         *offp = off;
 501 }
 502
 503 static size_t push_pipe(struct iov_iter *i, size_t size,
 504                         int *iter_headp, size_t *offp)
 505 {
 506         struct pipe_inode_info *pipe = i->pipe;
 507         unsigned int p_tail = pipe->tail;
 508         unsigned int p_mask = pipe->ring_size - 1;
 509         unsigned int iter_head;
 510         size_t off;
 511         ssize_t left;
 512
 513         if (unlikely(size > i->count))
 514                 size = i->count;
 515         if (unlikely(!size))
 516                 return 0;
 517
 518         left = size;
 519         data_start(i, &iter_head, &off);
 520         *iter_headp = iter_head;
 521         *offp = off;
 522         if (off) {
 523                 left -= PAGE_SIZE - off;
 524                 if (left <= 0) {
 525                         pipe->bufs[iter_head & p_mask].len += size;
 526                         return size;
 527                 }
 528                 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
 529                 iter_head++;
 530         }
 531         while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
 532                 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
 533                 struct page *page = alloc_page(GFP_USER);
 534                 if (!page)
 535                         break;
 536
 537                 buf->ops = &default_pipe_buf_ops;
 538                 buf->page = page;
 539                 buf->offset = 0;
 540                 buf->len = min_t(ssize_t, left, PAGE_SIZE);
 541                 left -= buf->len;
 542                 iter_head++;
 543                 pipe->head = iter_head;
 544
 545                 if (left == 0)
 546                         return size;
 547         }
 548         return size - left;
 549 }
 550
 551 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 552                                 struct iov_iter *i)
 553 {
 554         struct pipe_inode_info *pipe = i->pipe;
 555         unsigned int p_mask = pipe->ring_size - 1;
 556         unsigned int i_head;
 557         size_t n, off;
 558
 559         if (!sanity(i))
 560                 return 0;
 561
 562         bytes = n = push_pipe(i, bytes, &i_head, &off);
 563         if (unlikely(!n))
 564                 return 0;
 565         do {
 566                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 567                 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
 568                 i->head = i_head;
 569                 i->iov_offset = off + chunk;
 570                 n -= chunk;
 571                 addr += chunk;
 572                 off = 0;
 573                 i_head++;
 574         } while (n);
 575         i->count -= bytes;
 576         return bytes;
 577 }
 578
 579 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 580                               __wsum sum, size_t off)
 581 {
 582         __wsum next = csum_partial_copy_nocheck(from, to, len, 0);
 583         return csum_block_add(sum, next, off);
 584 }
 585
 586 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 587                                 __wsum *csum, struct iov_iter *i)
 588 {
 589         struct pipe_inode_info *pipe = i->pipe;
 590         unsigned int p_mask = pipe->ring_size - 1;
 591         unsigned int i_head;
 592         size_t n, r;
 593         size_t off = 0;
 594         __wsum sum = *csum;
 595
 596         if (!sanity(i))
 597                 return 0;
 598
 599         bytes = n = push_pipe(i, bytes, &i_head, &r);
 600         if (unlikely(!n))
 601                 return 0;
 602         do {
 603                 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
 604                 char *p = kmap_atomic(pipe->bufs[i_head & p_mask].page);
 605                 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
 606                 kunmap_atomic(p);
 607                 i->head = i_head;
 608                 i->iov_offset = r + chunk;
 609                 n -= chunk;
 610                 off += chunk;
 611                 addr += chunk;
 612                 r = 0;
 613                 i_head++;
 614         } while (n);
 615         i->count -= bytes;
 616         *csum = sum;
 617         return bytes;
 618 }
 619
 620 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 621 {
 622         const char *from = addr;
 623         if (unlikely(iov_iter_is_pipe(i)))
 624                 return copy_pipe_to_iter(addr, bytes, i);
 625         if (iter_is_iovec(i))
 626                 might_fault();
 627         iterate_and_advance(i, bytes, v,
 628                 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 629                 memcpy_to_page(v.bv_page, v.bv_offset,
 630                                (from += v.bv_len) - v.bv_len, v.bv_len),
 631                 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
 632         )
 633
 634         return bytes;
 635 }
 636 EXPORT_SYMBOL(_copy_to_iter);
 637
 638 #ifdef CONFIG_ARCH_HAS_UACCESS_MCSAFE
 639 static int copyout_mcsafe(void __user *to, const void *from, size_t n)
 640 {
 641         if (access_ok(to, n)) {
 642                 kasan_check_read(from, n);
 643                 n = copy_to_user_mcsafe((__force void *) to, from, n);
 644         }
 645         return n;
 646 }
 647
 648 static unsigned long memcpy_mcsafe_to_page(struct page *page, size_t offset,
 649                 const char *from, size_t len)
 650 {
 651         unsigned long ret;
 652         char *to;
 653
 654         to = kmap_atomic(page);
 655         ret = memcpy_mcsafe(to + offset, from, len);
 656         kunmap_atomic(to);
 657
 658         return ret;
 659 }
 660
 661 static size_t copy_pipe_to_iter_mcsafe(const void *addr, size_t bytes,
 662                                 struct iov_iter *i)
 663 {
 664         struct pipe_inode_info *pipe = i->pipe;
 665         unsigned int p_mask = pipe->ring_size - 1;
 666         unsigned int i_head;
 667         size_t n, off, xfer = 0;
 668
 669         if (!sanity(i))
 670                 return 0;
 671
 672         bytes = n = push_pipe(i, bytes, &i_head, &off);
 673         if (unlikely(!n))
 674                 return 0;
 675         do {
 676                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 677                 unsigned long rem;
 678
 679                 rem = memcpy_mcsafe_to_page(pipe->bufs[i_head & p_mask].page,
 680                                             off, addr, chunk);
 681                 i->head = i_head;
 682                 i->iov_offset = off + chunk - rem;
 683                 xfer += chunk - rem;
 684                 if (rem)
 685                         break;
 686                 n -= chunk;
 687                 addr += chunk;
 688                 off = 0;
 689                 i_head++;
 690         } while (n);
 691         i->count -= xfer;
 692         return xfer;
 693 }
 694
 695 /**
 696  * _copy_to_iter_mcsafe - copy to user with source-read error exception handling
 697  * @addr: source kernel address
 698  * @bytes: total transfer length
 699  * @iter: destination iterator
 700  *
 701  * The pmem driver arranges for filesystem-dax to use this facility via
 702  * dax_copy_to_iter() for protecting read/write to persistent memory.
 703  * Unless / until an architecture can guarantee identical performance
 704  * between _copy_to_iter_mcsafe() and _copy_to_iter() it would be a
 705  * performance regression to switch more users to the mcsafe version.
 706  *
 707  * Otherwise, the main differences between this and typical _copy_to_iter().
 708  *
 709  * * Typical tail/residue handling after a fault retries the copy
 710  *   byte-by-byte until the fault happens again. Re-triggering machine
 711  *   checks is potentially fatal so the implementation uses source
 712  *   alignment and poison alignment assumptions to avoid re-triggering
 713  *   hardware exceptions.
 714  *
 715  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 716  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 717  *   a short copy.
 718  *
 719  * See MCSAFE_TEST for self-test.
 720  */
 721 size_t _copy_to_iter_mcsafe(const void *addr, size_t bytes, struct iov_iter *i)
 722 {
 723         const char *from = addr;
 724         unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
 725
 726         if (unlikely(iov_iter_is_pipe(i)))
 727                 return copy_pipe_to_iter_mcsafe(addr, bytes, i);
 728         if (iter_is_iovec(i))
 729                 might_fault();
 730         iterate_and_advance(i, bytes, v,
 731                 copyout_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 732                 ({
 733                 rem = memcpy_mcsafe_to_page(v.bv_page, v.bv_offset,
 734                                (from += v.bv_len) - v.bv_len, v.bv_len);
 735                 if (rem) {
 736                         curr_addr = (unsigned long) from;
 737                         bytes = curr_addr - s_addr - rem;
 738                         return bytes;
 739                 }
 740                 }),
 741                 ({
 742                 rem = memcpy_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len,
 743                                 v.iov_len);
 744                 if (rem) {
 745                         curr_addr = (unsigned long) from;
 746                         bytes = curr_addr - s_addr - rem;
 747                         return bytes;
 748                 }
 749                 })
 750         )
 751
 752         return bytes;
 753 }
 754 EXPORT_SYMBOL_GPL(_copy_to_iter_mcsafe);
 755 #endif /* CONFIG_ARCH_HAS_UACCESS_MCSAFE */
 756
 757 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 758 {
 759         char *to = addr;
 760         if (unlikely(iov_iter_is_pipe(i))) {
 761                 WARN_ON(1);
 762                 return 0;
 763         }
 764         if (iter_is_iovec(i))
 765                 might_fault();
 766         iterate_and_advance(i, bytes, v,
 767                 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 768                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 769                                  v.bv_offset, v.bv_len),
 770                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 771         )
 772
 773         return bytes;
 774 }
 775 EXPORT_SYMBOL(_copy_from_iter);
 776
 777 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
 778 {
 779         char *to = addr;
 780         if (unlikely(iov_iter_is_pipe(i))) {
 781                 WARN_ON(1);
 782                 return false;
 783         }
 784         if (unlikely(i->count < bytes))
 785                 return false;
 786
 787         if (iter_is_iovec(i))
 788                 might_fault();
 789         iterate_all_kinds(i, bytes, v, ({
 790                 if (copyin((to += v.iov_len) - v.iov_len,
 791                                       v.iov_base, v.iov_len))
 792                         return false;
 793                 0;}),
 794                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 795                                  v.bv_offset, v.bv_len),
 796                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 797         )
 798
 799         iov_iter_advance(i, bytes);
 800         return true;
 801 }
 802 EXPORT_SYMBOL(_copy_from_iter_full);
 803
 804 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 805 {
 806         char *to = addr;
 807         if (unlikely(iov_iter_is_pipe(i))) {
 808                 WARN_ON(1);
 809                 return 0;
 810         }
 811         iterate_and_advance(i, bytes, v,
 812                 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 813                                          v.iov_base, v.iov_len),
 814                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 815                                  v.bv_offset, v.bv_len),
 816                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 817         )
 818
 819         return bytes;
 820 }
 821 EXPORT_SYMBOL(_copy_from_iter_nocache);
 822
 823 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 824 /**
 825  * _copy_from_iter_flushcache - write destination through cpu cache
 826  * @addr: destination kernel address
 827  * @bytes: total transfer length
 828  * @iter: source iterator
 829  *
 830  * The pmem driver arranges for filesystem-dax to use this facility via
 831  * dax_copy_from_iter() for ensuring that writes to persistent memory
 832  * are flushed through the CPU cache. It is differentiated from
 833  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 834  * all iterator types. The _copy_from_iter_nocache() only attempts to
 835  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 836  * instructions that strand dirty-data in the cache.
 837  */
 838 size_t _copy_from_iter_flushcache(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 0;
 844         }
 845         iterate_and_advance(i, bytes, v,
 846                 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
 847                                          v.iov_base, v.iov_len),
 848                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 849                                  v.bv_offset, v.bv_len),
 850                 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
 851                         v.iov_len)
 852         )
 853
 854         return bytes;
 855 }
 856 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 857 #endif
 858
 859 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
 860 {
 861         char *to = addr;
 862         if (unlikely(iov_iter_is_pipe(i))) {
 863                 WARN_ON(1);
 864                 return false;
 865         }
 866         if (unlikely(i->count < bytes))
 867                 return false;
 868         iterate_all_kinds(i, bytes, v, ({
 869                 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 870                                              v.iov_base, v.iov_len))
 871                         return false;
 872                 0;}),
 873                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 874                                  v.bv_offset, v.bv_len),
 875                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 876         )
 877
 878         iov_iter_advance(i, bytes);
 879         return true;
 880 }
 881 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
 882
 883 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 884 {
 885         struct page *head;
 886         size_t v = n + offset;
 887
 888         /*
 889          * The general case needs to access the page order in order
 890          * to compute the page size.
 891          * However, we mostly deal with order-0 pages and thus can
 892          * avoid a possible cache line miss for requests that fit all
 893          * page orders.
 894          */
 895         if (n <= v && v <= PAGE_SIZE)
 896                 return true;
 897
 898         head = compound_head(page);
 899         v += (page - head) << PAGE_SHIFT;
 900
 901         if (likely(n <= v && v <= (page_size(head))))
 902                 return true;
 903         WARN_ON(1);
 904         return false;
 905 }
 906
 907 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 908                          struct iov_iter *i)
 909 {
 910         if (unlikely(!page_copy_sane(page, offset, bytes)))
 911                 return 0;
 912         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 913                 void *kaddr = kmap_atomic(page);
 914                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 915                 kunmap_atomic(kaddr);
 916                 return wanted;
 917         } else if (unlikely(iov_iter_is_discard(i)))
 918                 return bytes;
 919         else if (likely(!iov_iter_is_pipe(i)))
 920                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 921         else
 922                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 923 }
 924 EXPORT_SYMBOL(copy_page_to_iter);
 925
 926 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 927                          struct iov_iter *i)
 928 {
 929         if (unlikely(!page_copy_sane(page, offset, bytes)))
 930                 return 0;
 931         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 932                 WARN_ON(1);
 933                 return 0;
 934         }
 935         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 936                 void *kaddr = kmap_atomic(page);
 937                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 938                 kunmap_atomic(kaddr);
 939                 return wanted;
 940         } else
 941                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 942 }
 943 EXPORT_SYMBOL(copy_page_from_iter);
 944
 945 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 946 {
 947         struct pipe_inode_info *pipe = i->pipe;
 948         unsigned int p_mask = pipe->ring_size - 1;
 949         unsigned int i_head;
 950         size_t n, off;
 951
 952         if (!sanity(i))
 953                 return 0;
 954
 955         bytes = n = push_pipe(i, bytes, &i_head, &off);
 956         if (unlikely(!n))
 957                 return 0;
 958
 959         do {
 960                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 961                 memzero_page(pipe->bufs[i_head & p_mask].page, off, chunk);
 962                 i->head = i_head;
 963                 i->iov_offset = off + chunk;
 964                 n -= chunk;
 965                 off = 0;
 966                 i_head++;
 967         } while (n);
 968         i->count -= bytes;
 969         return bytes;
 970 }
 971
 972 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 973 {
 974         if (unlikely(iov_iter_is_pipe(i)))
 975                 return pipe_zero(bytes, i);
 976         iterate_and_advance(i, bytes, v,
 977                 clear_user(v.iov_base, v.iov_len),
 978                 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
 979                 memset(v.iov_base, 0, v.iov_len)
 980         )
 981
 982         return bytes;
 983 }
 984 EXPORT_SYMBOL(iov_iter_zero);
 985
 986 size_t iov_iter_copy_from_user_atomic(struct page *page,
 987                 struct iov_iter *i, unsigned long offset, size_t bytes)
 988 {
 989         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 990         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 991                 kunmap_atomic(kaddr);
 992                 return 0;
 993         }
 994         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 995                 kunmap_atomic(kaddr);
 996                 WARN_ON(1);
 997                 return 0;
 998         }
 999         iterate_all_kinds(i, bytes, v,
1000                 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1001                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1002                                  v.bv_offset, v.bv_len),
1003                 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
1004         )
1005         kunmap_atomic(kaddr);
1006         return bytes;
1007 }
1008 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1009
1010 static inline void pipe_truncate(struct iov_iter *i)
1011 {
1012         struct pipe_inode_info *pipe = i->pipe;
1013         unsigned int p_tail = pipe->tail;
1014         unsigned int p_head = pipe->head;
1015         unsigned int p_mask = pipe->ring_size - 1;
1016
1017         if (!pipe_empty(p_head, p_tail)) {
1018                 struct pipe_buffer *buf;
1019                 unsigned int i_head = i->head;
1020                 size_t off = i->iov_offset;
1021
1022                 if (off) {
1023                         buf = &pipe->bufs[i_head & p_mask];
1024                         buf->len = off - buf->offset;
1025                         i_head++;
1026                 }
1027                 while (p_head != i_head) {
1028                         p_head--;
1029                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1030                 }
1031
1032                 pipe->head = p_head;
1033         }
1034 }
1035
1036 static void pipe_advance(struct iov_iter *i, size_t size)
1037 {
1038         struct pipe_inode_info *pipe = i->pipe;
1039         if (unlikely(i->count < size))
1040                 size = i->count;
1041         if (size) {
1042                 struct pipe_buffer *buf;
1043                 unsigned int p_mask = pipe->ring_size - 1;
1044                 unsigned int i_head = i->head;
1045                 size_t off = i->iov_offset, left = size;
1046
1047                 if (off) /* make it relative to the beginning of buffer */
1048                         left += off - pipe->bufs[i_head & p_mask].offset;
1049                 while (1) {
1050                         buf = &pipe->bufs[i_head & p_mask];
1051                         if (left <= buf->len)
1052                                 break;
1053                         left -= buf->len;
1054                         i_head++;
1055                 }
1056                 i->head = i_head;
1057                 i->iov_offset = buf->offset + left;
1058         }
1059         i->count -= size;
1060         /* ... and discard everything past that point */
1061         pipe_truncate(i);
1062 }
1063
1064 void iov_iter_advance(struct iov_iter *i, size_t size)
1065 {
1066         if (unlikely(iov_iter_is_pipe(i))) {
1067                 pipe_advance(i, size);
1068                 return;
1069         }
1070         if (unlikely(iov_iter_is_discard(i))) {
1071                 i->count -= size;
1072                 return;
1073         }
1074         iterate_and_advance(i, size, v, 0, 0, 0)
1075 }
1076 EXPORT_SYMBOL(iov_iter_advance);
1077
1078 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1079 {
1080         if (!unroll)
1081                 return;
1082         if (WARN_ON(unroll > MAX_RW_COUNT))
1083                 return;
1084         i->count += unroll;
1085         if (unlikely(iov_iter_is_pipe(i))) {
1086                 struct pipe_inode_info *pipe = i->pipe;
1087                 unsigned int p_mask = pipe->ring_size - 1;
1088                 unsigned int i_head = i->head;
1089                 size_t off = i->iov_offset;
1090                 while (1) {
1091                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1092                         size_t n = off - b->offset;
1093                         if (unroll < n) {
1094                                 off -= unroll;
1095                                 break;
1096                         }
1097                         unroll -= n;
1098                         if (!unroll && i_head == i->start_head) {
1099                                 off = 0;
1100                                 break;
1101                         }
1102                         i_head--;
1103                         b = &pipe->bufs[i_head & p_mask];
1104                         off = b->offset + b->len;
1105                 }
1106                 i->iov_offset = off;
1107                 i->head = i_head;
1108                 pipe_truncate(i);
1109                 return;
1110         }
1111         if (unlikely(iov_iter_is_discard(i)))
1112                 return;
1113         if (unroll <= i->iov_offset) {
1114                 i->iov_offset -= unroll;
1115                 return;
1116         }
1117         unroll -= i->iov_offset;
1118         if (iov_iter_is_bvec(i)) {
1119                 const struct bio_vec *bvec = i->bvec;
1120                 while (1) {
1121                         size_t n = (--bvec)->bv_len;
1122                         i->nr_segs++;
1123                         if (unroll <= n) {
1124                                 i->bvec = bvec;
1125                                 i->iov_offset = n - unroll;
1126                                 return;
1127                         }
1128                         unroll -= n;
1129                 }
1130         } else { /* same logics for iovec and kvec */
1131                 const struct iovec *iov = i->iov;
1132                 while (1) {
1133                         size_t n = (--iov)->iov_len;
1134                         i->nr_segs++;
1135                         if (unroll <= n) {
1136                                 i->iov = iov;
1137                                 i->iov_offset = n - unroll;
1138                                 return;
1139                         }
1140                         unroll -= n;
1141                 }
1142         }
1143 }
1144 EXPORT_SYMBOL(iov_iter_revert);
1145
1146 /*
1147  * Return the count of just the current iov_iter segment.
1148  */
1149 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1150 {
1151         if (unlikely(iov_iter_is_pipe(i)))
1152                 return i->count;        // it is a silly place, anyway
1153         if (i->nr_segs == 1)
1154                 return i->count;
1155         if (unlikely(iov_iter_is_discard(i)))
1156                 return i->count;
1157         else if (iov_iter_is_bvec(i))
1158                 return min(i->count, i->bvec->bv_len - i->iov_offset);
1159         else
1160                 return min(i->count, i->iov->iov_len - i->iov_offset);
1161 }
1162 EXPORT_SYMBOL(iov_iter_single_seg_count);
1163
1164 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1165                         const struct kvec *kvec, unsigned long nr_segs,
1166                         size_t count)
1167 {
1168         WARN_ON(direction & ~(READ | WRITE));
1169         i->type = ITER_KVEC | (direction & (READ | WRITE));
1170         i->kvec = kvec;
1171         i->nr_segs = nr_segs;
1172         i->iov_offset = 0;
1173         i->count = count;
1174 }
1175 EXPORT_SYMBOL(iov_iter_kvec);
1176
1177 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1178                         const struct bio_vec *bvec, unsigned long nr_segs,
1179                         size_t count)
1180 {
1181         WARN_ON(direction & ~(READ | WRITE));
1182         i->type = ITER_BVEC | (direction & (READ | WRITE));
1183         i->bvec = bvec;
1184         i->nr_segs = nr_segs;
1185         i->iov_offset = 0;
1186         i->count = count;
1187 }
1188 EXPORT_SYMBOL(iov_iter_bvec);
1189
1190 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1191                         struct pipe_inode_info *pipe,
1192                         size_t count)
1193 {
1194         BUG_ON(direction != READ);
1195         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1196         i->type = ITER_PIPE | READ;
1197         i->pipe = pipe;
1198         i->head = pipe->head;
1199         i->iov_offset = 0;
1200         i->count = count;
1201         i->start_head = i->head;
1202 }
1203 EXPORT_SYMBOL(iov_iter_pipe);
1204
1205 /**
1206  * iov_iter_discard - Initialise an I/O iterator that discards data
1207  * @i: The iterator to initialise.
1208  * @direction: The direction of the transfer.
1209  * @count: The size of the I/O buffer in bytes.
1210  *
1211  * Set up an I/O iterator that just discards everything that's written to it.
1212  * It's only available as a READ iterator.
1213  */
1214 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1215 {
1216         BUG_ON(direction != READ);
1217         i->type = ITER_DISCARD | READ;
1218         i->count = count;
1219         i->iov_offset = 0;
1220 }
1221 EXPORT_SYMBOL(iov_iter_discard);
1222
1223 unsigned long iov_iter_alignment(const struct iov_iter *i)
1224 {
1225         unsigned long res = 0;
1226         size_t size = i->count;
1227
1228         if (unlikely(iov_iter_is_pipe(i))) {
1229                 unsigned int p_mask = i->pipe->ring_size - 1;
1230
1231                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1232                         return size | i->iov_offset;
1233                 return size;
1234         }
1235         iterate_all_kinds(i, size, v,
1236                 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1237                 res |= v.bv_offset | v.bv_len,
1238                 res |= (unsigned long)v.iov_base | v.iov_len
1239         )
1240         return res;
1241 }
1242 EXPORT_SYMBOL(iov_iter_alignment);
1243
1244 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1245 {
1246         unsigned long res = 0;
1247         size_t size = i->count;
1248
1249         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1250                 WARN_ON(1);
1251                 return ~0U;
1252         }
1253
1254         iterate_all_kinds(i, size, v,
1255                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1256                         (size != v.iov_len ? size : 0), 0),
1257                 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1258                         (size != v.bv_len ? size : 0)),
1259                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1260                         (size != v.iov_len ? size : 0))
1261                 );
1262         return res;
1263 }
1264 EXPORT_SYMBOL(iov_iter_gap_alignment);
1265
1266 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1267                                 size_t maxsize,
1268                                 struct page **pages,
1269                                 int iter_head,
1270                                 size_t *start)
1271 {
1272         struct pipe_inode_info *pipe = i->pipe;
1273         unsigned int p_mask = pipe->ring_size - 1;
1274         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1275         if (!n)
1276                 return -EFAULT;
1277
1278         maxsize = n;
1279         n += *start;
1280         while (n > 0) {
1281                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1282                 iter_head++;
1283                 n -= PAGE_SIZE;
1284         }
1285
1286         return maxsize;
1287 }
1288
1289 static ssize_t pipe_get_pages(struct iov_iter *i,
1290                    struct page **pages, size_t maxsize, unsigned maxpages,
1291                    size_t *start)
1292 {
1293         unsigned int iter_head, npages;
1294         size_t capacity;
1295
1296         if (!maxsize)
1297                 return 0;
1298
1299         if (!sanity(i))
1300                 return -EFAULT;
1301
1302         data_start(i, &iter_head, start);
1303         /* Amount of free space: some of this one + all after this one */
1304         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1305         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1306
1307         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1308 }
1309
1310 ssize_t iov_iter_get_pages(struct iov_iter *i,
1311                    struct page **pages, size_t maxsize, unsigned maxpages,
1312                    size_t *start)
1313 {
1314         if (maxsize > i->count)
1315                 maxsize = i->count;
1316
1317         if (unlikely(iov_iter_is_pipe(i)))
1318                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1319         if (unlikely(iov_iter_is_discard(i)))
1320                 return -EFAULT;
1321
1322         iterate_all_kinds(i, maxsize, v, ({
1323                 unsigned long addr = (unsigned long)v.iov_base;
1324                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1325                 int n;
1326                 int res;
1327
1328                 if (len > maxpages * PAGE_SIZE)
1329                         len = maxpages * PAGE_SIZE;
1330                 addr &= ~(PAGE_SIZE - 1);
1331                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1332                 res = get_user_pages_fast(addr, n,
1333                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0,
1334                                 pages);
1335                 if (unlikely(res < 0))
1336                         return res;
1337                 return (res == n ? len : res * PAGE_SIZE) - *start;
1338         0;}),({
1339                 /* can't be more than PAGE_SIZE */
1340                 *start = v.bv_offset;
1341                 get_page(*pages = v.bv_page);
1342                 return v.bv_len;
1343         }),({
1344                 return -EFAULT;
1345         })
1346         )
1347         return 0;
1348 }
1349 EXPORT_SYMBOL(iov_iter_get_pages);
1350
1351 static struct page **get_pages_array(size_t n)
1352 {
1353         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1354 }
1355
1356 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1357                    struct page ***pages, size_t maxsize,
1358                    size_t *start)
1359 {
1360         struct page **p;
1361         unsigned int iter_head, npages;
1362         ssize_t n;
1363
1364         if (!maxsize)
1365                 return 0;
1366
1367         if (!sanity(i))
1368                 return -EFAULT;
1369
1370         data_start(i, &iter_head, start);
1371         /* Amount of free space: some of this one + all after this one */
1372         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1373         n = npages * PAGE_SIZE - *start;
1374         if (maxsize > n)
1375                 maxsize = n;
1376         else
1377                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1378         p = get_pages_array(npages);
1379         if (!p)
1380                 return -ENOMEM;
1381         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1382         if (n > 0)
1383                 *pages = p;
1384         else
1385                 kvfree(p);
1386         return n;
1387 }
1388
1389 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1390                    struct page ***pages, size_t maxsize,
1391                    size_t *start)
1392 {
1393         struct page **p;
1394
1395         if (maxsize > i->count)
1396                 maxsize = i->count;
1397
1398         if (unlikely(iov_iter_is_pipe(i)))
1399                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1400         if (unlikely(iov_iter_is_discard(i)))
1401                 return -EFAULT;
1402
1403         iterate_all_kinds(i, maxsize, v, ({
1404                 unsigned long addr = (unsigned long)v.iov_base;
1405                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1406                 int n;
1407                 int res;
1408
1409                 addr &= ~(PAGE_SIZE - 1);
1410                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1411                 p = get_pages_array(n);
1412                 if (!p)
1413                         return -ENOMEM;
1414                 res = get_user_pages_fast(addr, n,
1415                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0, p);
1416                 if (unlikely(res < 0)) {
1417                         kvfree(p);
1418                         return res;
1419                 }
1420                 *pages = p;
1421                 return (res == n ? len : res * PAGE_SIZE) - *start;
1422         0;}),({
1423                 /* can't be more than PAGE_SIZE */
1424                 *start = v.bv_offset;
1425                 *pages = p = get_pages_array(1);
1426                 if (!p)
1427                         return -ENOMEM;
1428                 get_page(*p = v.bv_page);
1429                 return v.bv_len;
1430         }),({
1431                 return -EFAULT;
1432         })
1433         )
1434         return 0;
1435 }
1436 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1437
1438 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1439                                struct iov_iter *i)
1440 {
1441         char *to = addr;
1442         __wsum sum, next;
1443         size_t off = 0;
1444         sum = *csum;
1445         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1446                 WARN_ON(1);
1447                 return 0;
1448         }
1449         iterate_and_advance(i, bytes, v, ({
1450                 int err = 0;
1451                 next = csum_and_copy_from_user(v.iov_base,
1452                                                (to += v.iov_len) - v.iov_len,
1453                                                v.iov_len, 0, &err);
1454                 if (!err) {
1455                         sum = csum_block_add(sum, next, off);
1456                         off += v.iov_len;
1457                 }
1458                 err ? v.iov_len : 0;
1459         }), ({
1460                 char *p = kmap_atomic(v.bv_page);
1461                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1462                                       p + v.bv_offset, v.bv_len,
1463                                       sum, off);
1464                 kunmap_atomic(p);
1465                 off += v.bv_len;
1466         }),({
1467                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1468                                       v.iov_base, v.iov_len,
1469                                       sum, off);
1470                 off += v.iov_len;
1471         })
1472         )
1473         *csum = sum;
1474         return bytes;
1475 }
1476 EXPORT_SYMBOL(csum_and_copy_from_iter);
1477
1478 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1479                                struct iov_iter *i)
1480 {
1481         char *to = addr;
1482         __wsum sum, next;
1483         size_t off = 0;
1484         sum = *csum;
1485         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1486                 WARN_ON(1);
1487                 return false;
1488         }
1489         if (unlikely(i->count < bytes))
1490                 return false;
1491         iterate_all_kinds(i, bytes, v, ({
1492                 int err = 0;
1493                 next = csum_and_copy_from_user(v.iov_base,
1494                                                (to += v.iov_len) - v.iov_len,
1495                                                v.iov_len, 0, &err);
1496                 if (err)
1497                         return false;
1498                 sum = csum_block_add(sum, next, off);
1499                 off += v.iov_len;
1500                 0;
1501         }), ({
1502                 char *p = kmap_atomic(v.bv_page);
1503                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1504                                       p + v.bv_offset, v.bv_len,
1505                                       sum, off);
1506                 kunmap_atomic(p);
1507                 off += v.bv_len;
1508         }),({
1509                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1510                                       v.iov_base, v.iov_len,
1511                                       sum, off);
1512                 off += v.iov_len;
1513         })
1514         )
1515         *csum = sum;
1516         iov_iter_advance(i, bytes);
1517         return true;
1518 }
1519 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1520
1521 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *csump,
1522                              struct iov_iter *i)
1523 {
1524         const char *from = addr;
1525         __wsum *csum = csump;
1526         __wsum sum, next;
1527         size_t off = 0;
1528
1529         if (unlikely(iov_iter_is_pipe(i)))
1530                 return csum_and_copy_to_pipe_iter(addr, bytes, csum, i);
1531
1532         sum = *csum;
1533         if (unlikely(iov_iter_is_discard(i))) {
1534                 WARN_ON(1);     /* for now */
1535                 return 0;
1536         }
1537         iterate_and_advance(i, bytes, v, ({
1538                 int err = 0;
1539                 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1540                                              v.iov_base,
1541                                              v.iov_len, 0, &err);
1542                 if (!err) {
1543                         sum = csum_block_add(sum, next, off);
1544                         off += v.iov_len;
1545                 }
1546                 err ? v.iov_len : 0;
1547         }), ({
1548                 char *p = kmap_atomic(v.bv_page);
1549                 sum = csum_and_memcpy(p + v.bv_offset,
1550                                       (from += v.bv_len) - v.bv_len,
1551                                       v.bv_len, sum, off);
1552                 kunmap_atomic(p);
1553                 off += v.bv_len;
1554         }),({
1555                 sum = csum_and_memcpy(v.iov_base,
1556                                      (from += v.iov_len) - v.iov_len,
1557                                      v.iov_len, sum, off);
1558                 off += v.iov_len;
1559         })
1560         )
1561         *csum = sum;
1562         return bytes;
1563 }
1564 EXPORT_SYMBOL(csum_and_copy_to_iter);
1565
1566 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1567                 struct iov_iter *i)
1568 {
1569 #ifdef CONFIG_CRYPTO
1570         struct ahash_request *hash = hashp;
1571         struct scatterlist sg;
1572         size_t copied;
1573
1574         copied = copy_to_iter(addr, bytes, i);
1575         sg_init_one(&sg, addr, copied);
1576         ahash_request_set_crypt(hash, &sg, NULL, copied);
1577         crypto_ahash_update(hash);
1578         return copied;
1579 #else
1580         return 0;
1581 #endif
1582 }
1583 EXPORT_SYMBOL(hash_and_copy_to_iter);
1584
1585 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1586 {
1587         size_t size = i->count;
1588         int npages = 0;
1589
1590         if (!size)
1591                 return 0;
1592         if (unlikely(iov_iter_is_discard(i)))
1593                 return 0;
1594
1595         if (unlikely(iov_iter_is_pipe(i))) {
1596                 struct pipe_inode_info *pipe = i->pipe;
1597                 unsigned int iter_head;
1598                 size_t off;
1599
1600                 if (!sanity(i))
1601                         return 0;
1602
1603                 data_start(i, &iter_head, &off);
1604                 /* some of this one + all after this one */
1605                 npages = pipe_space_for_user(iter_head, pipe->tail, pipe);
1606                 if (npages >= maxpages)
1607                         return maxpages;
1608         } else iterate_all_kinds(i, size, v, ({
1609                 unsigned long p = (unsigned long)v.iov_base;
1610                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1611                         - p / PAGE_SIZE;
1612                 if (npages >= maxpages)
1613                         return maxpages;
1614         0;}),({
1615                 npages++;
1616                 if (npages >= maxpages)
1617                         return maxpages;
1618         }),({
1619                 unsigned long p = (unsigned long)v.iov_base;
1620                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1621                         - p / PAGE_SIZE;
1622                 if (npages >= maxpages)
1623                         return maxpages;
1624         })
1625         )
1626         return npages;
1627 }
1628 EXPORT_SYMBOL(iov_iter_npages);
1629
1630 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1631 {
1632         *new = *old;
1633         if (unlikely(iov_iter_is_pipe(new))) {
1634                 WARN_ON(1);
1635                 return NULL;
1636         }
1637         if (unlikely(iov_iter_is_discard(new)))
1638                 return NULL;
1639         if (iov_iter_is_bvec(new))
1640                 return new->bvec = kmemdup(new->bvec,
1641                                     new->nr_segs * sizeof(struct bio_vec),
1642                                     flags);
1643         else
1644                 /* iovec and kvec have identical layout */
1645                 return new->iov = kmemdup(new->iov,
1646                                    new->nr_segs * sizeof(struct iovec),
1647                                    flags);
1648 }
1649 EXPORT_SYMBOL(dup_iter);
1650
1651 /**
1652  * import_iovec() - Copy an array of &struct iovec from userspace
1653  *     into the kernel, check that it is valid, and initialize a new
1654  *     &struct iov_iter iterator to access it.
1655  *
1656  * @type: One of %READ or %WRITE.
1657  * @uvector: Pointer to the userspace array.
1658  * @nr_segs: Number of elements in userspace array.
1659  * @fast_segs: Number of elements in @iov.
1660  * @iov: (input and output parameter) Pointer to pointer to (usually small
1661  *     on-stack) kernel array.
1662  * @i: Pointer to iterator that will be initialized on success.
1663  *
1664  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1665  * then this function places %NULL in *@iov on return. Otherwise, a new
1666  * array will be allocated and the result placed in *@iov. This means that
1667  * the caller may call kfree() on *@iov regardless of whether the small
1668  * on-stack array was used or not (and regardless of whether this function
1669  * returns an error or not).
1670  *
1671  * Return: Negative error code on error, bytes imported on success
1672  */
1673 ssize_t import_iovec(int type, const struct iovec __user * uvector,
1674                  unsigned nr_segs, unsigned fast_segs,
1675                  struct iovec **iov, struct iov_iter *i)
1676 {
1677         ssize_t n;
1678         struct iovec *p;
1679         n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1680                                   *iov, &p);
1681         if (n < 0) {
1682                 if (p != *iov)
1683                         kfree(p);
1684                 *iov = NULL;
1685                 return n;
1686         }
1687         iov_iter_init(i, type, p, nr_segs, n);
1688         *iov = p == *iov ? NULL : p;
1689         return n;
1690 }
1691 EXPORT_SYMBOL(import_iovec);
1692
1693 #ifdef CONFIG_COMPAT
1694 #include <linux/compat.h>
1695
1696 ssize_t compat_import_iovec(int type,
1697                 const struct compat_iovec __user * uvector,
1698                 unsigned nr_segs, unsigned fast_segs,
1699                 struct iovec **iov, struct iov_iter *i)
1700 {
1701         ssize_t n;
1702         struct iovec *p;
1703         n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1704                                   *iov, &p);
1705         if (n < 0) {
1706                 if (p != *iov)
1707                         kfree(p);
1708                 *iov = NULL;
1709                 return n;
1710         }
1711         iov_iter_init(i, type, p, nr_segs, n);
1712         *iov = p == *iov ? NULL : p;
1713         return n;
1714 }
1715 EXPORT_SYMBOL(compat_import_iovec);
1716 #endif
1717
1718 int import_single_range(int rw, void __user *buf, size_t len,
1719                  struct iovec *iov, struct iov_iter *i)
1720 {
1721         if (len > MAX_RW_COUNT)
1722                 len = MAX_RW_COUNT;
1723         if (unlikely(!access_ok(buf, len)))
1724                 return -EFAULT;
1725
1726         iov->iov_base = buf;
1727         iov->iov_len = len;
1728         iov_iter_init(i, rw, iov, 1, len);
1729         return 0;
1730 }
1731 EXPORT_SYMBOL(import_single_range);
1732
1733 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1734                             int (*f)(struct kvec *vec, void *context),
1735                             void *context)
1736 {
1737         struct kvec w;
1738         int err = -EINVAL;
1739         if (!bytes)
1740                 return 0;
1741
1742         iterate_all_kinds(i, bytes, v, -EINVAL, ({
1743                 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1744                 w.iov_len = v.bv_len;
1745                 err = f(&w, context);
1746                 kunmap(v.bv_page);
1747                 err;}), ({
1748                 w = v;
1749                 err = f(&w, context);})
1750         )
1751         return err;
1752 }
1753 EXPORT_SYMBOL(iov_iter_for_each_range);