lib/iov_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <crypto/hash.h>
   3 #include <linux/export.h>
   4 #include <linux/bvec.h>
   5 #include <linux/fault-inject-usercopy.h>
   6 #include <linux/uio.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/highmem.h>
   9 #include <linux/slab.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/splice.h>
  12 #include <linux/compat.h>
  13 #include <net/checksum.h>
  14 #include <linux/scatterlist.h>
  15 #include <linux/instrumented.h>
  16
  17 #define PIPE_PARANOIA /* for now */
  18
  19 /* covers iovec and kvec alike */
  20 #define iterate_iovec(i, n, __v, __off, __p, skip, STEP) {      \
  21         size_t __off = 0;                                       \
  22         do {                                                    \
  23                 __v.iov_len = min(n, __p->iov_len - skip);      \
  24                 if (likely(__v.iov_len)) {                      \
  25                         __v.iov_base = __p->iov_base + skip;    \
  26                         __v.iov_len -= (STEP);                  \
  27                         __off += __v.iov_len;                   \
  28                         skip += __v.iov_len;                    \
  29                         n -= __v.iov_len;                       \
  30                         if (skip < __p->iov_len)                \
  31                                 break;                          \
  32                 }                                               \
  33                 __p++;                                          \
  34                 skip = 0;                                       \
  35         } while (n);                                            \
  36         n = __off;                                              \
  37 }
  38
  39 #define iterate_bvec(i, n, __v, __off, p, skip, STEP) {         \
  40         size_t __off = 0;                                       \
  41         while (n) {                                             \
  42                 unsigned offset = p->bv_offset + skip;          \
  43                 unsigned left;                                  \
  44                 void *kaddr = kmap_local_page(p->bv_page +      \
  45                                         offset / PAGE_SIZE);    \
  46                 __v.iov_base = kaddr + offset % PAGE_SIZE;      \
  47                 __v.iov_len = min(min(n, p->bv_len - skip),     \
  48                      (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
  49                 left = (STEP);                                  \
  50                 kunmap_local(kaddr);                            \
  51                 __v.iov_len -= left;                            \
  52                 __off += __v.iov_len;                           \
  53                 skip += __v.iov_len;                            \
  54                 if (skip == p->bv_len) {                        \
  55                         skip = 0;                               \
  56                         p++;                                    \
  57                 }                                               \
  58                 n -= __v.iov_len;                               \
  59                 if (left)                                       \
  60                         break;                                  \
  61         }                                                       \
  62         n = __off;                                              \
  63 }
  64
  65 #define iterate_xarray(i, n, __v, __off, skip, STEP) {          \
  66         __label__ __out;                                        \
  67         size_t __off = 0;                                       \
  68         struct page *head = NULL;                               \
  69         size_t seg, offset;                                     \
  70         loff_t start = i->xarray_start + skip;                  \
  71         pgoff_t index = start >> PAGE_SHIFT;                    \
  72         int j;                                                  \
  73                                                                 \
  74         XA_STATE(xas, i->xarray, index);                        \
  75                                                                 \
  76         rcu_read_lock();                                                \
  77         xas_for_each(&xas, head, ULONG_MAX) {                           \
  78                 unsigned left;                                          \
  79                 if (xas_retry(&xas, head))                              \
  80                         continue;                                       \
  81                 if (WARN_ON(xa_is_value(head)))                         \
  82                         break;                                          \
  83                 if (WARN_ON(PageHuge(head)))                            \
  84                         break;                                          \
  85                 for (j = (head->index < index) ? index - head->index : 0; \
  86                      j < thp_nr_pages(head); j++) {                     \
  87                         void *kaddr = kmap_local_page(head + j);        \
  88                         offset = (start + __off) % PAGE_SIZE;           \
  89                         __v.iov_base = kaddr + offset;                  \
  90                         seg = PAGE_SIZE - offset;                       \
  91                         __v.iov_len = min(n, seg);                      \
  92                         left = (STEP);                                  \
  93                         kunmap_local(kaddr);                            \
  94                         __v.iov_len -= left;                            \
  95                         __off += __v.iov_len;                           \
  96                         n -= __v.iov_len;                               \
  97                         if (left || n == 0)                             \
  98                                 goto __out;                             \
  99                 }                                                       \
 100         }                                                       \
 101 __out:                                                          \
 102         rcu_read_unlock();                                      \
 103         skip += __off;                                          \
 104         n = __off;                                              \
 105 }
 106
 107 #define __iterate_and_advance(i, n, v, off, I, K) {             \
 108         if (unlikely(i->count < n))                             \
 109                 n = i->count;                                   \
 110         if (likely(n)) {                                        \
 111                 size_t skip = i->iov_offset;                    \
 112                 if (likely(iter_is_iovec(i))) {                 \
 113                         const struct iovec *iov = i->iov;       \
 114                         struct iovec v;                         \
 115                         iterate_iovec(i, n, v, off, iov, skip, (I))     \
 116                         i->nr_segs -= iov - i->iov;             \
 117                         i->iov = iov;                           \
 118                 } else if (iov_iter_is_bvec(i)) {               \
 119                         const struct bio_vec *bvec = i->bvec;   \
 120                         struct kvec v;                          \
 121                         iterate_bvec(i, n, v, off, bvec, skip, (K))     \
 122                         i->nr_segs -= bvec - i->bvec;           \
 123                         i->bvec = bvec;                         \
 124                 } else if (iov_iter_is_kvec(i)) {               \
 125                         const struct kvec *kvec = i->kvec;      \
 126                         struct kvec v;                          \
 127                         iterate_iovec(i, n, v, off, kvec, skip, (K))    \
 128                         i->nr_segs -= kvec - i->kvec;           \
 129                         i->kvec = kvec;                         \
 130                 } else if (iov_iter_is_xarray(i)) {             \
 131                         struct kvec v;                          \
 132                         iterate_xarray(i, n, v, off, skip, (K)) \
 133                 }                                               \
 134                 i->count -= n;                                  \
 135                 i->iov_offset = skip;                           \
 136         }                                                       \
 137 }
 138 #define iterate_and_advance(i, n, v, off, I, K) \
 139         __iterate_and_advance(i, n, v, off, I, ((void)(K),0))
 140
 141 static int copyout(void __user *to, const void *from, size_t n)
 142 {
 143         if (should_fail_usercopy())
 144                 return n;
 145         if (access_ok(to, n)) {
 146                 instrument_copy_to_user(to, from, n);
 147                 n = raw_copy_to_user(to, from, n);
 148         }
 149         return n;
 150 }
 151
 152 static int copyin(void *to, const void __user *from, size_t n)
 153 {
 154         if (should_fail_usercopy())
 155                 return n;
 156         if (access_ok(from, n)) {
 157                 instrument_copy_from_user(to, from, n);
 158                 n = raw_copy_from_user(to, from, n);
 159         }
 160         return n;
 161 }
 162
 163 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 164                          struct iov_iter *i)
 165 {
 166         size_t skip, copy, left, wanted;
 167         const struct iovec *iov;
 168         char __user *buf;
 169         void *kaddr, *from;
 170
 171         if (unlikely(bytes > i->count))
 172                 bytes = i->count;
 173
 174         if (unlikely(!bytes))
 175                 return 0;
 176
 177         might_fault();
 178         wanted = bytes;
 179         iov = i->iov;
 180         skip = i->iov_offset;
 181         buf = iov->iov_base + skip;
 182         copy = min(bytes, iov->iov_len - skip);
 183
 184         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 185                 kaddr = kmap_atomic(page);
 186                 from = kaddr + offset;
 187
 188                 /* first chunk, usually the only one */
 189                 left = copyout(buf, from, copy);
 190                 copy -= left;
 191                 skip += copy;
 192                 from += copy;
 193                 bytes -= copy;
 194
 195                 while (unlikely(!left && bytes)) {
 196                         iov++;
 197                         buf = iov->iov_base;
 198                         copy = min(bytes, iov->iov_len);
 199                         left = copyout(buf, from, copy);
 200                         copy -= left;
 201                         skip = copy;
 202                         from += copy;
 203                         bytes -= copy;
 204                 }
 205                 if (likely(!bytes)) {
 206                         kunmap_atomic(kaddr);
 207                         goto done;
 208                 }
 209                 offset = from - kaddr;
 210                 buf += copy;
 211                 kunmap_atomic(kaddr);
 212                 copy = min(bytes, iov->iov_len - skip);
 213         }
 214         /* Too bad - revert to non-atomic kmap */
 215
 216         kaddr = kmap(page);
 217         from = kaddr + offset;
 218         left = copyout(buf, from, copy);
 219         copy -= left;
 220         skip += copy;
 221         from += copy;
 222         bytes -= copy;
 223         while (unlikely(!left && bytes)) {
 224                 iov++;
 225                 buf = iov->iov_base;
 226                 copy = min(bytes, iov->iov_len);
 227                 left = copyout(buf, from, copy);
 228                 copy -= left;
 229                 skip = copy;
 230                 from += copy;
 231                 bytes -= copy;
 232         }
 233         kunmap(page);
 234
 235 done:
 236         if (skip == iov->iov_len) {
 237                 iov++;
 238                 skip = 0;
 239         }
 240         i->count -= wanted - bytes;
 241         i->nr_segs -= iov - i->iov;
 242         i->iov = iov;
 243         i->iov_offset = skip;
 244         return wanted - bytes;
 245 }
 246
 247 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 248                          struct iov_iter *i)
 249 {
 250         size_t skip, copy, left, wanted;
 251         const struct iovec *iov;
 252         char __user *buf;
 253         void *kaddr, *to;
 254
 255         if (unlikely(bytes > i->count))
 256                 bytes = i->count;
 257
 258         if (unlikely(!bytes))
 259                 return 0;
 260
 261         might_fault();
 262         wanted = bytes;
 263         iov = i->iov;
 264         skip = i->iov_offset;
 265         buf = iov->iov_base + skip;
 266         copy = min(bytes, iov->iov_len - skip);
 267
 268         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 269                 kaddr = kmap_atomic(page);
 270                 to = kaddr + offset;
 271
 272                 /* first chunk, usually the only one */
 273                 left = copyin(to, buf, copy);
 274                 copy -= left;
 275                 skip += copy;
 276                 to += copy;
 277                 bytes -= copy;
 278
 279                 while (unlikely(!left && bytes)) {
 280                         iov++;
 281                         buf = iov->iov_base;
 282                         copy = min(bytes, iov->iov_len);
 283                         left = copyin(to, buf, copy);
 284                         copy -= left;
 285                         skip = copy;
 286                         to += copy;
 287                         bytes -= copy;
 288                 }
 289                 if (likely(!bytes)) {
 290                         kunmap_atomic(kaddr);
 291                         goto done;
 292                 }
 293                 offset = to - kaddr;
 294                 buf += copy;
 295                 kunmap_atomic(kaddr);
 296                 copy = min(bytes, iov->iov_len - skip);
 297         }
 298         /* Too bad - revert to non-atomic kmap */
 299
 300         kaddr = kmap(page);
 301         to = kaddr + offset;
 302         left = copyin(to, buf, copy);
 303         copy -= left;
 304         skip += copy;
 305         to += copy;
 306         bytes -= copy;
 307         while (unlikely(!left && bytes)) {
 308                 iov++;
 309                 buf = iov->iov_base;
 310                 copy = min(bytes, iov->iov_len);
 311                 left = copyin(to, buf, copy);
 312                 copy -= left;
 313                 skip = copy;
 314                 to += copy;
 315                 bytes -= copy;
 316         }
 317         kunmap(page);
 318
 319 done:
 320         if (skip == iov->iov_len) {
 321                 iov++;
 322                 skip = 0;
 323         }
 324         i->count -= wanted - bytes;
 325         i->nr_segs -= iov - i->iov;
 326         i->iov = iov;
 327         i->iov_offset = skip;
 328         return wanted - bytes;
 329 }
 330
 331 #ifdef PIPE_PARANOIA
 332 static bool sanity(const struct iov_iter *i)
 333 {
 334         struct pipe_inode_info *pipe = i->pipe;
 335         unsigned int p_head = pipe->head;
 336         unsigned int p_tail = pipe->tail;
 337         unsigned int p_mask = pipe->ring_size - 1;
 338         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 339         unsigned int i_head = i->head;
 340         unsigned int idx;
 341
 342         if (i->iov_offset) {
 343                 struct pipe_buffer *p;
 344                 if (unlikely(p_occupancy == 0))
 345                         goto Bad;       // pipe must be non-empty
 346                 if (unlikely(i_head != p_head - 1))
 347                         goto Bad;       // must be at the last buffer...
 348
 349                 p = &pipe->bufs[i_head & p_mask];
 350                 if (unlikely(p->offset + p->len != i->iov_offset))
 351                         goto Bad;       // ... at the end of segment
 352         } else {
 353                 if (i_head != p_head)
 354                         goto Bad;       // must be right after the last buffer
 355         }
 356         return true;
 357 Bad:
 358         printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
 359         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 360                         p_head, p_tail, pipe->ring_size);
 361         for (idx = 0; idx < pipe->ring_size; idx++)
 362                 printk(KERN_ERR "[%p %p %d %d]\n",
 363                         pipe->bufs[idx].ops,
 364                         pipe->bufs[idx].page,
 365                         pipe->bufs[idx].offset,
 366                         pipe->bufs[idx].len);
 367         WARN_ON(1);
 368         return false;
 369 }
 370 #else
 371 #define sanity(i) true
 372 #endif
 373
 374 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 375                          struct iov_iter *i)
 376 {
 377         struct pipe_inode_info *pipe = i->pipe;
 378         struct pipe_buffer *buf;
 379         unsigned int p_tail = pipe->tail;
 380         unsigned int p_mask = pipe->ring_size - 1;
 381         unsigned int i_head = i->head;
 382         size_t off;
 383
 384         if (unlikely(bytes > i->count))
 385                 bytes = i->count;
 386
 387         if (unlikely(!bytes))
 388                 return 0;
 389
 390         if (!sanity(i))
 391                 return 0;
 392
 393         off = i->iov_offset;
 394         buf = &pipe->bufs[i_head & p_mask];
 395         if (off) {
 396                 if (offset == off && buf->page == page) {
 397                         /* merge with the last one */
 398                         buf->len += bytes;
 399                         i->iov_offset += bytes;
 400                         goto out;
 401                 }
 402                 i_head++;
 403                 buf = &pipe->bufs[i_head & p_mask];
 404         }
 405         if (pipe_full(i_head, p_tail, pipe->max_usage))
 406                 return 0;
 407
 408         buf->ops = &page_cache_pipe_buf_ops;
 409         get_page(page);
 410         buf->page = page;
 411         buf->offset = offset;
 412         buf->len = bytes;
 413
 414         pipe->head = i_head + 1;
 415         i->iov_offset = offset + bytes;
 416         i->head = i_head;
 417 out:
 418         i->count -= bytes;
 419         return bytes;
 420 }
 421
 422 /*
 423  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 424  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 425  *
 426  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 427  * because it is an invalid address).
 428  */
 429 int iov_iter_fault_in_readable(const struct iov_iter *i, size_t bytes)
 430 {
 431         if (iter_is_iovec(i)) {
 432                 const struct iovec *p;
 433                 size_t skip;
 434
 435                 if (bytes > i->count)
 436                         bytes = i->count;
 437                 for (p = i->iov, skip = i->iov_offset; bytes; p++, skip = 0) {
 438                         size_t len = min(bytes, p->iov_len - skip);
 439                         int err;
 440
 441                         if (unlikely(!len))
 442                                 continue;
 443                         err = fault_in_pages_readable(p->iov_base + skip, len);
 444                         if (unlikely(err))
 445                                 return err;
 446                         bytes -= len;
 447                 }
 448         }
 449         return 0;
 450 }
 451 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 452
 453 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 454                         const struct iovec *iov, unsigned long nr_segs,
 455                         size_t count)
 456 {
 457         WARN_ON(direction & ~(READ | WRITE));
 458         WARN_ON_ONCE(uaccess_kernel());
 459         *i = (struct iov_iter) {
 460                 .iter_type = ITER_IOVEC,
 461                 .data_source = direction,
 462                 .iov = iov,
 463                 .nr_segs = nr_segs,
 464                 .iov_offset = 0,
 465                 .count = count
 466         };
 467 }
 468 EXPORT_SYMBOL(iov_iter_init);
 469
 470 static inline bool allocated(struct pipe_buffer *buf)
 471 {
 472         return buf->ops == &default_pipe_buf_ops;
 473 }
 474
 475 static inline void data_start(const struct iov_iter *i,
 476                               unsigned int *iter_headp, size_t *offp)
 477 {
 478         unsigned int p_mask = i->pipe->ring_size - 1;
 479         unsigned int iter_head = i->head;
 480         size_t off = i->iov_offset;
 481
 482         if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
 483                     off == PAGE_SIZE)) {
 484                 iter_head++;
 485                 off = 0;
 486         }
 487         *iter_headp = iter_head;
 488         *offp = off;
 489 }
 490
 491 static size_t push_pipe(struct iov_iter *i, size_t size,
 492                         int *iter_headp, size_t *offp)
 493 {
 494         struct pipe_inode_info *pipe = i->pipe;
 495         unsigned int p_tail = pipe->tail;
 496         unsigned int p_mask = pipe->ring_size - 1;
 497         unsigned int iter_head;
 498         size_t off;
 499         ssize_t left;
 500
 501         if (unlikely(size > i->count))
 502                 size = i->count;
 503         if (unlikely(!size))
 504                 return 0;
 505
 506         left = size;
 507         data_start(i, &iter_head, &off);
 508         *iter_headp = iter_head;
 509         *offp = off;
 510         if (off) {
 511                 left -= PAGE_SIZE - off;
 512                 if (left <= 0) {
 513                         pipe->bufs[iter_head & p_mask].len += size;
 514                         return size;
 515                 }
 516                 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
 517                 iter_head++;
 518         }
 519         while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
 520                 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
 521                 struct page *page = alloc_page(GFP_USER);
 522                 if (!page)
 523                         break;
 524
 525                 buf->ops = &default_pipe_buf_ops;
 526                 buf->page = page;
 527                 buf->offset = 0;
 528                 buf->len = min_t(ssize_t, left, PAGE_SIZE);
 529                 left -= buf->len;
 530                 iter_head++;
 531                 pipe->head = iter_head;
 532
 533                 if (left == 0)
 534                         return size;
 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         unsigned int p_mask = pipe->ring_size - 1;
 544         unsigned int i_head;
 545         size_t n, off;
 546
 547         if (!sanity(i))
 548                 return 0;
 549
 550         bytes = n = push_pipe(i, bytes, &i_head, &off);
 551         if (unlikely(!n))
 552                 return 0;
 553         do {
 554                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 555                 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
 556                 i->head = i_head;
 557                 i->iov_offset = off + chunk;
 558                 n -= chunk;
 559                 addr += chunk;
 560                 off = 0;
 561                 i_head++;
 562         } while (n);
 563         i->count -= bytes;
 564         return bytes;
 565 }
 566
 567 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 568                               __wsum sum, size_t off)
 569 {
 570         __wsum next = csum_partial_copy_nocheck(from, to, len);
 571         return csum_block_add(sum, next, off);
 572 }
 573
 574 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 575                                          struct csum_state *csstate,
 576                                          struct iov_iter *i)
 577 {
 578         struct pipe_inode_info *pipe = i->pipe;
 579         unsigned int p_mask = pipe->ring_size - 1;
 580         __wsum sum = csstate->csum;
 581         size_t off = csstate->off;
 582         unsigned int i_head;
 583         size_t n, r;
 584
 585         if (!sanity(i))
 586                 return 0;
 587
 588         bytes = n = push_pipe(i, bytes, &i_head, &r);
 589         if (unlikely(!n))
 590                 return 0;
 591         do {
 592                 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
 593                 char *p = kmap_atomic(pipe->bufs[i_head & p_mask].page);
 594                 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
 595                 kunmap_atomic(p);
 596                 i->head = i_head;
 597                 i->iov_offset = r + chunk;
 598                 n -= chunk;
 599                 off += chunk;
 600                 addr += chunk;
 601                 r = 0;
 602                 i_head++;
 603         } while (n);
 604         i->count -= bytes;
 605         csstate->csum = sum;
 606         csstate->off = off;
 607         return bytes;
 608 }
 609
 610 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 611 {
 612         if (unlikely(iov_iter_is_pipe(i)))
 613                 return copy_pipe_to_iter(addr, bytes, i);
 614         if (iter_is_iovec(i))
 615                 might_fault();
 616         iterate_and_advance(i, bytes, v, off,
 617                 copyout(v.iov_base, addr + off, v.iov_len),
 618                 memcpy(v.iov_base, addr + off, v.iov_len)
 619         )
 620
 621         return bytes;
 622 }
 623 EXPORT_SYMBOL(_copy_to_iter);
 624
 625 #ifdef CONFIG_ARCH_HAS_COPY_MC
 626 static int copyout_mc(void __user *to, const void *from, size_t n)
 627 {
 628         if (access_ok(to, n)) {
 629                 instrument_copy_to_user(to, from, n);
 630                 n = copy_mc_to_user((__force void *) to, from, n);
 631         }
 632         return n;
 633 }
 634
 635 static unsigned long copy_mc_to_page(struct page *page, size_t offset,
 636                 const char *from, size_t len)
 637 {
 638         unsigned long ret;
 639         char *to;
 640
 641         to = kmap_atomic(page);
 642         ret = copy_mc_to_kernel(to + offset, from, len);
 643         kunmap_atomic(to);
 644
 645         return ret;
 646 }
 647
 648 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 649                                 struct iov_iter *i)
 650 {
 651         struct pipe_inode_info *pipe = i->pipe;
 652         unsigned int p_mask = pipe->ring_size - 1;
 653         unsigned int i_head;
 654         size_t n, off, xfer = 0;
 655
 656         if (!sanity(i))
 657                 return 0;
 658
 659         bytes = n = push_pipe(i, bytes, &i_head, &off);
 660         if (unlikely(!n))
 661                 return 0;
 662         do {
 663                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 664                 unsigned long rem;
 665
 666                 rem = copy_mc_to_page(pipe->bufs[i_head & p_mask].page,
 667                                             off, addr, chunk);
 668                 i->head = i_head;
 669                 i->iov_offset = off + chunk - rem;
 670                 xfer += chunk - rem;
 671                 if (rem)
 672                         break;
 673                 n -= chunk;
 674                 addr += chunk;
 675                 off = 0;
 676                 i_head++;
 677         } while (n);
 678         i->count -= xfer;
 679         return xfer;
 680 }
 681
 682 /**
 683  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 684  * @addr: source kernel address
 685  * @bytes: total transfer length
 686  * @iter: destination iterator
 687  *
 688  * The pmem driver deploys this for the dax operation
 689  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 690  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 691  * successfully copied.
 692  *
 693  * The main differences between this and typical _copy_to_iter().
 694  *
 695  * * Typical tail/residue handling after a fault retries the copy
 696  *   byte-by-byte until the fault happens again. Re-triggering machine
 697  *   checks is potentially fatal so the implementation uses source
 698  *   alignment and poison alignment assumptions to avoid re-triggering
 699  *   hardware exceptions.
 700  *
 701  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 702  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 703  *   a short copy.
 704  */
 705 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 706 {
 707         if (unlikely(iov_iter_is_pipe(i)))
 708                 return copy_mc_pipe_to_iter(addr, bytes, i);
 709         if (iter_is_iovec(i))
 710                 might_fault();
 711         __iterate_and_advance(i, bytes, v, off,
 712                 copyout_mc(v.iov_base, addr + off, v.iov_len),
 713                 copy_mc_to_kernel(v.iov_base, addr + off, v.iov_len)
 714         )
 715
 716         return bytes;
 717 }
 718 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 719 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 720
 721 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 722 {
 723         if (unlikely(iov_iter_is_pipe(i))) {
 724                 WARN_ON(1);
 725                 return 0;
 726         }
 727         if (iter_is_iovec(i))
 728                 might_fault();
 729         iterate_and_advance(i, bytes, v, off,
 730                 copyin(addr + off, v.iov_base, v.iov_len),
 731                 memcpy(addr + off, v.iov_base, v.iov_len)
 732         )
 733
 734         return bytes;
 735 }
 736 EXPORT_SYMBOL(_copy_from_iter);
 737
 738 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 739 {
 740         if (unlikely(iov_iter_is_pipe(i))) {
 741                 WARN_ON(1);
 742                 return 0;
 743         }
 744         iterate_and_advance(i, bytes, v, off,
 745                 __copy_from_user_inatomic_nocache(addr + off,
 746                                          v.iov_base, v.iov_len),
 747                 memcpy(addr + off, v.iov_base, v.iov_len)
 748         )
 749
 750         return bytes;
 751 }
 752 EXPORT_SYMBOL(_copy_from_iter_nocache);
 753
 754 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 755 /**
 756  * _copy_from_iter_flushcache - write destination through cpu cache
 757  * @addr: destination kernel address
 758  * @bytes: total transfer length
 759  * @iter: source iterator
 760  *
 761  * The pmem driver arranges for filesystem-dax to use this facility via
 762  * dax_copy_from_iter() for ensuring that writes to persistent memory
 763  * are flushed through the CPU cache. It is differentiated from
 764  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 765  * all iterator types. The _copy_from_iter_nocache() only attempts to
 766  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 767  * instructions that strand dirty-data in the cache.
 768  */
 769 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 770 {
 771         if (unlikely(iov_iter_is_pipe(i))) {
 772                 WARN_ON(1);
 773                 return 0;
 774         }
 775         iterate_and_advance(i, bytes, v, off,
 776                 __copy_from_user_flushcache(addr + off, v.iov_base, v.iov_len),
 777                 memcpy_flushcache(addr + off, v.iov_base, v.iov_len)
 778         )
 779
 780         return bytes;
 781 }
 782 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 783 #endif
 784
 785 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 786 {
 787         struct page *head;
 788         size_t v = n + offset;
 789
 790         /*
 791          * The general case needs to access the page order in order
 792          * to compute the page size.
 793          * However, we mostly deal with order-0 pages and thus can
 794          * avoid a possible cache line miss for requests that fit all
 795          * page orders.
 796          */
 797         if (n <= v && v <= PAGE_SIZE)
 798                 return true;
 799
 800         head = compound_head(page);
 801         v += (page - head) << PAGE_SHIFT;
 802
 803         if (likely(n <= v && v <= (page_size(head))))
 804                 return true;
 805         WARN_ON(1);
 806         return false;
 807 }
 808
 809 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 810                          struct iov_iter *i)
 811 {
 812         if (likely(iter_is_iovec(i)))
 813                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 814         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 815                 void *kaddr = kmap_atomic(page);
 816                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 817                 kunmap_atomic(kaddr);
 818                 return wanted;
 819         }
 820         if (iov_iter_is_pipe(i))
 821                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 822         if (unlikely(iov_iter_is_discard(i))) {
 823                 if (unlikely(i->count < bytes))
 824                         bytes = i->count;
 825                 i->count -= bytes;
 826                 return bytes;
 827         }
 828         WARN_ON(1);
 829         return 0;
 830 }
 831
 832 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 833                          struct iov_iter *i)
 834 {
 835         size_t res = 0;
 836         if (unlikely(!page_copy_sane(page, offset, bytes)))
 837                 return 0;
 838         page += offset / PAGE_SIZE; // first subpage
 839         offset %= PAGE_SIZE;
 840         while (1) {
 841                 size_t n = __copy_page_to_iter(page, offset,
 842                                 min(bytes, (size_t)PAGE_SIZE - offset), i);
 843                 res += n;
 844                 bytes -= n;
 845                 if (!bytes || !n)
 846                         break;
 847                 offset += n;
 848                 if (offset == PAGE_SIZE) {
 849                         page++;
 850                         offset = 0;
 851                 }
 852         }
 853         return res;
 854 }
 855 EXPORT_SYMBOL(copy_page_to_iter);
 856
 857 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 858                          struct iov_iter *i)
 859 {
 860         if (unlikely(!page_copy_sane(page, offset, bytes)))
 861                 return 0;
 862         if (likely(iter_is_iovec(i)))
 863                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 864         if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
 865                 void *kaddr = kmap_atomic(page);
 866                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 867                 kunmap_atomic(kaddr);
 868                 return wanted;
 869         }
 870         WARN_ON(1);
 871         return 0;
 872 }
 873 EXPORT_SYMBOL(copy_page_from_iter);
 874
 875 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 876 {
 877         struct pipe_inode_info *pipe = i->pipe;
 878         unsigned int p_mask = pipe->ring_size - 1;
 879         unsigned int i_head;
 880         size_t n, off;
 881
 882         if (!sanity(i))
 883                 return 0;
 884
 885         bytes = n = push_pipe(i, bytes, &i_head, &off);
 886         if (unlikely(!n))
 887                 return 0;
 888
 889         do {
 890                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 891                 memzero_page(pipe->bufs[i_head & p_mask].page, off, chunk);
 892                 i->head = i_head;
 893                 i->iov_offset = off + chunk;
 894                 n -= chunk;
 895                 off = 0;
 896                 i_head++;
 897         } while (n);
 898         i->count -= bytes;
 899         return bytes;
 900 }
 901
 902 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 903 {
 904         if (unlikely(iov_iter_is_pipe(i)))
 905                 return pipe_zero(bytes, i);
 906         iterate_and_advance(i, bytes, v, count,
 907                 clear_user(v.iov_base, v.iov_len),
 908                 memset(v.iov_base, 0, v.iov_len)
 909         )
 910
 911         return bytes;
 912 }
 913 EXPORT_SYMBOL(iov_iter_zero);
 914
 915 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 916                                   struct iov_iter *i)
 917 {
 918         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 919         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 920                 kunmap_atomic(kaddr);
 921                 return 0;
 922         }
 923         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 924                 kunmap_atomic(kaddr);
 925                 WARN_ON(1);
 926                 return 0;
 927         }
 928         iterate_and_advance(i, bytes, v, off,
 929                 copyin(p + off, v.iov_base, v.iov_len),
 930                 memcpy(p + off, v.iov_base, v.iov_len)
 931         )
 932         kunmap_atomic(kaddr);
 933         return bytes;
 934 }
 935 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 936
 937 static inline void pipe_truncate(struct iov_iter *i)
 938 {
 939         struct pipe_inode_info *pipe = i->pipe;
 940         unsigned int p_tail = pipe->tail;
 941         unsigned int p_head = pipe->head;
 942         unsigned int p_mask = pipe->ring_size - 1;
 943
 944         if (!pipe_empty(p_head, p_tail)) {
 945                 struct pipe_buffer *buf;
 946                 unsigned int i_head = i->head;
 947                 size_t off = i->iov_offset;
 948
 949                 if (off) {
 950                         buf = &pipe->bufs[i_head & p_mask];
 951                         buf->len = off - buf->offset;
 952                         i_head++;
 953                 }
 954                 while (p_head != i_head) {
 955                         p_head--;
 956                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
 957                 }
 958
 959                 pipe->head = p_head;
 960         }
 961 }
 962
 963 static void pipe_advance(struct iov_iter *i, size_t size)
 964 {
 965         struct pipe_inode_info *pipe = i->pipe;
 966         if (size) {
 967                 struct pipe_buffer *buf;
 968                 unsigned int p_mask = pipe->ring_size - 1;
 969                 unsigned int i_head = i->head;
 970                 size_t off = i->iov_offset, left = size;
 971
 972                 if (off) /* make it relative to the beginning of buffer */
 973                         left += off - pipe->bufs[i_head & p_mask].offset;
 974                 while (1) {
 975                         buf = &pipe->bufs[i_head & p_mask];
 976                         if (left <= buf->len)
 977                                 break;
 978                         left -= buf->len;
 979                         i_head++;
 980                 }
 981                 i->head = i_head;
 982                 i->iov_offset = buf->offset + left;
 983         }
 984         i->count -= size;
 985         /* ... and discard everything past that point */
 986         pipe_truncate(i);
 987 }
 988
 989 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 990 {
 991         struct bvec_iter bi;
 992
 993         bi.bi_size = i->count;
 994         bi.bi_bvec_done = i->iov_offset;
 995         bi.bi_idx = 0;
 996         bvec_iter_advance(i->bvec, &bi, size);
 997
 998         i->bvec += bi.bi_idx;
 999         i->nr_segs -= bi.bi_idx;
1000         i->count = bi.bi_size;
1001         i->iov_offset = bi.bi_bvec_done;
1002 }
1003
1004 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
1005 {
1006         const struct iovec *iov, *end;
1007
1008         if (!i->count)
1009                 return;
1010         i->count -= size;
1011
1012         size += i->iov_offset; // from beginning of current segment
1013         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1014                 if (likely(size < iov->iov_len))
1015                         break;
1016                 size -= iov->iov_len;
1017         }
1018         i->iov_offset = size;
1019         i->nr_segs -= iov - i->iov;
1020         i->iov = iov;
1021 }
1022
1023 void iov_iter_advance(struct iov_iter *i, size_t size)
1024 {
1025         if (unlikely(i->count < size))
1026                 size = i->count;
1027         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1028                 /* iovec and kvec have identical layouts */
1029                 iov_iter_iovec_advance(i, size);
1030         } else if (iov_iter_is_bvec(i)) {
1031                 iov_iter_bvec_advance(i, size);
1032         } else if (iov_iter_is_pipe(i)) {
1033                 pipe_advance(i, size);
1034         } else if (unlikely(iov_iter_is_xarray(i))) {
1035                 i->iov_offset += size;
1036                 i->count -= size;
1037         } else if (iov_iter_is_discard(i)) {
1038                 i->count -= size;
1039         }
1040 }
1041 EXPORT_SYMBOL(iov_iter_advance);
1042
1043 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1044 {
1045         if (!unroll)
1046                 return;
1047         if (WARN_ON(unroll > MAX_RW_COUNT))
1048                 return;
1049         i->count += unroll;
1050         if (unlikely(iov_iter_is_pipe(i))) {
1051                 struct pipe_inode_info *pipe = i->pipe;
1052                 unsigned int p_mask = pipe->ring_size - 1;
1053                 unsigned int i_head = i->head;
1054                 size_t off = i->iov_offset;
1055                 while (1) {
1056                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1057                         size_t n = off - b->offset;
1058                         if (unroll < n) {
1059                                 off -= unroll;
1060                                 break;
1061                         }
1062                         unroll -= n;
1063                         if (!unroll && i_head == i->start_head) {
1064                                 off = 0;
1065                                 break;
1066                         }
1067                         i_head--;
1068                         b = &pipe->bufs[i_head & p_mask];
1069                         off = b->offset + b->len;
1070                 }
1071                 i->iov_offset = off;
1072                 i->head = i_head;
1073                 pipe_truncate(i);
1074                 return;
1075         }
1076         if (unlikely(iov_iter_is_discard(i)))
1077                 return;
1078         if (unroll <= i->iov_offset) {
1079                 i->iov_offset -= unroll;
1080                 return;
1081         }
1082         unroll -= i->iov_offset;
1083         if (iov_iter_is_xarray(i)) {
1084                 BUG(); /* We should never go beyond the start of the specified
1085                         * range since we might then be straying into pages that
1086                         * aren't pinned.
1087                         */
1088         } else if (iov_iter_is_bvec(i)) {
1089                 const struct bio_vec *bvec = i->bvec;
1090                 while (1) {
1091                         size_t n = (--bvec)->bv_len;
1092                         i->nr_segs++;
1093                         if (unroll <= n) {
1094                                 i->bvec = bvec;
1095                                 i->iov_offset = n - unroll;
1096                                 return;
1097                         }
1098                         unroll -= n;
1099                 }
1100         } else { /* same logics for iovec and kvec */
1101                 const struct iovec *iov = i->iov;
1102                 while (1) {
1103                         size_t n = (--iov)->iov_len;
1104                         i->nr_segs++;
1105                         if (unroll <= n) {
1106                                 i->iov = iov;
1107                                 i->iov_offset = n - unroll;
1108                                 return;
1109                         }
1110                         unroll -= n;
1111                 }
1112         }
1113 }
1114 EXPORT_SYMBOL(iov_iter_revert);
1115
1116 /*
1117  * Return the count of just the current iov_iter segment.
1118  */
1119 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1120 {
1121         if (i->nr_segs > 1) {
1122                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1123                         return min(i->count, i->iov->iov_len - i->iov_offset);
1124                 if (iov_iter_is_bvec(i))
1125                         return min(i->count, i->bvec->bv_len - i->iov_offset);
1126         }
1127         return i->count;
1128 }
1129 EXPORT_SYMBOL(iov_iter_single_seg_count);
1130
1131 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1132                         const struct kvec *kvec, unsigned long nr_segs,
1133                         size_t count)
1134 {
1135         WARN_ON(direction & ~(READ | WRITE));
1136         *i = (struct iov_iter){
1137                 .iter_type = ITER_KVEC,
1138                 .data_source = direction,
1139                 .kvec = kvec,
1140                 .nr_segs = nr_segs,
1141                 .iov_offset = 0,
1142                 .count = count
1143         };
1144 }
1145 EXPORT_SYMBOL(iov_iter_kvec);
1146
1147 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1148                         const struct bio_vec *bvec, unsigned long nr_segs,
1149                         size_t count)
1150 {
1151         WARN_ON(direction & ~(READ | WRITE));
1152         *i = (struct iov_iter){
1153                 .iter_type = ITER_BVEC,
1154                 .data_source = direction,
1155                 .bvec = bvec,
1156                 .nr_segs = nr_segs,
1157                 .iov_offset = 0,
1158                 .count = count
1159         };
1160 }
1161 EXPORT_SYMBOL(iov_iter_bvec);
1162
1163 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1164                         struct pipe_inode_info *pipe,
1165                         size_t count)
1166 {
1167         BUG_ON(direction != READ);
1168         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1169         *i = (struct iov_iter){
1170                 .iter_type = ITER_PIPE,
1171                 .data_source = false,
1172                 .pipe = pipe,
1173                 .head = pipe->head,
1174                 .start_head = pipe->head,
1175                 .iov_offset = 0,
1176                 .count = count
1177         };
1178 }
1179 EXPORT_SYMBOL(iov_iter_pipe);
1180
1181 /**
1182  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1183  * @i: The iterator to initialise.
1184  * @direction: The direction of the transfer.
1185  * @xarray: The xarray to access.
1186  * @start: The start file position.
1187  * @count: The size of the I/O buffer in bytes.
1188  *
1189  * Set up an I/O iterator to either draw data out of the pages attached to an
1190  * inode or to inject data into those pages.  The pages *must* be prevented
1191  * from evaporation, either by taking a ref on them or locking them by the
1192  * caller.
1193  */
1194 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1195                      struct xarray *xarray, loff_t start, size_t count)
1196 {
1197         BUG_ON(direction & ~1);
1198         *i = (struct iov_iter) {
1199                 .iter_type = ITER_XARRAY,
1200                 .data_source = direction,
1201                 .xarray = xarray,
1202                 .xarray_start = start,
1203                 .count = count,
1204                 .iov_offset = 0
1205         };
1206 }
1207 EXPORT_SYMBOL(iov_iter_xarray);
1208
1209 /**
1210  * iov_iter_discard - Initialise an I/O iterator that discards data
1211  * @i: The iterator to initialise.
1212  * @direction: The direction of the transfer.
1213  * @count: The size of the I/O buffer in bytes.
1214  *
1215  * Set up an I/O iterator that just discards everything that's written to it.
1216  * It's only available as a READ iterator.
1217  */
1218 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1219 {
1220         BUG_ON(direction != READ);
1221         *i = (struct iov_iter){
1222                 .iter_type = ITER_DISCARD,
1223                 .data_source = false,
1224                 .count = count,
1225                 .iov_offset = 0
1226         };
1227 }
1228 EXPORT_SYMBOL(iov_iter_discard);
1229
1230 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1231 {
1232         unsigned long res = 0;
1233         size_t size = i->count;
1234         size_t skip = i->iov_offset;
1235         unsigned k;
1236
1237         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1238                 size_t len = i->iov[k].iov_len - skip;
1239                 if (len) {
1240                         res |= (unsigned long)i->iov[k].iov_base + skip;
1241                         if (len > size)
1242                                 len = size;
1243                         res |= len;
1244                         size -= len;
1245                         if (!size)
1246                                 break;
1247                 }
1248         }
1249         return res;
1250 }
1251
1252 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1253 {
1254         unsigned res = 0;
1255         size_t size = i->count;
1256         unsigned skip = i->iov_offset;
1257         unsigned k;
1258
1259         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1260                 size_t len = i->bvec[k].bv_len - skip;
1261                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1262                 if (len > size)
1263                         len = size;
1264                 res |= len;
1265                 size -= len;
1266                 if (!size)
1267                         break;
1268         }
1269         return res;
1270 }
1271
1272 unsigned long iov_iter_alignment(const struct iov_iter *i)
1273 {
1274         /* iovec and kvec have identical layouts */
1275         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1276                 return iov_iter_alignment_iovec(i);
1277
1278         if (iov_iter_is_bvec(i))
1279                 return iov_iter_alignment_bvec(i);
1280
1281         if (iov_iter_is_pipe(i)) {
1282                 unsigned int p_mask = i->pipe->ring_size - 1;
1283                 size_t size = i->count;
1284
1285                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1286                         return size | i->iov_offset;
1287                 return size;
1288         }
1289
1290         if (iov_iter_is_xarray(i))
1291                 return (i->xarray_start + i->iov_offset) | i->count;
1292
1293         return 0;
1294 }
1295 EXPORT_SYMBOL(iov_iter_alignment);
1296
1297 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1298 {
1299         unsigned long res = 0;
1300         unsigned long v = 0;
1301         size_t size = i->count;
1302         unsigned k;
1303
1304         if (WARN_ON(!iter_is_iovec(i)))
1305                 return ~0U;
1306
1307         for (k = 0; k < i->nr_segs; k++) {
1308                 if (i->iov[k].iov_len) {
1309                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1310                         if (v) // if not the first one
1311                                 res |= base | v; // this start | previous end
1312                         v = base + i->iov[k].iov_len;
1313                         if (size <= i->iov[k].iov_len)
1314                                 break;
1315                         size -= i->iov[k].iov_len;
1316                 }
1317         }
1318         return res;
1319 }
1320 EXPORT_SYMBOL(iov_iter_gap_alignment);
1321
1322 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1323                                 size_t maxsize,
1324                                 struct page **pages,
1325                                 int iter_head,
1326                                 size_t *start)
1327 {
1328         struct pipe_inode_info *pipe = i->pipe;
1329         unsigned int p_mask = pipe->ring_size - 1;
1330         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1331         if (!n)
1332                 return -EFAULT;
1333
1334         maxsize = n;
1335         n += *start;
1336         while (n > 0) {
1337                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1338                 iter_head++;
1339                 n -= PAGE_SIZE;
1340         }
1341
1342         return maxsize;
1343 }
1344
1345 static ssize_t pipe_get_pages(struct iov_iter *i,
1346                    struct page **pages, size_t maxsize, unsigned maxpages,
1347                    size_t *start)
1348 {
1349         unsigned int iter_head, npages;
1350         size_t capacity;
1351
1352         if (!sanity(i))
1353                 return -EFAULT;
1354
1355         data_start(i, &iter_head, start);
1356         /* Amount of free space: some of this one + all after this one */
1357         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1358         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1359
1360         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1361 }
1362
1363 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1364                                           pgoff_t index, unsigned int nr_pages)
1365 {
1366         XA_STATE(xas, xa, index);
1367         struct page *page;
1368         unsigned int ret = 0;
1369
1370         rcu_read_lock();
1371         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1372                 if (xas_retry(&xas, page))
1373                         continue;
1374
1375                 /* Has the page moved or been split? */
1376                 if (unlikely(page != xas_reload(&xas))) {
1377                         xas_reset(&xas);
1378                         continue;
1379                 }
1380
1381                 pages[ret] = find_subpage(page, xas.xa_index);
1382                 get_page(pages[ret]);
1383                 if (++ret == nr_pages)
1384                         break;
1385         }
1386         rcu_read_unlock();
1387         return ret;
1388 }
1389
1390 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1391                                      struct page **pages, size_t maxsize,
1392                                      unsigned maxpages, size_t *_start_offset)
1393 {
1394         unsigned nr, offset;
1395         pgoff_t index, count;
1396         size_t size = maxsize, actual;
1397         loff_t pos;
1398
1399         if (!size || !maxpages)
1400                 return 0;
1401
1402         pos = i->xarray_start + i->iov_offset;
1403         index = pos >> PAGE_SHIFT;
1404         offset = pos & ~PAGE_MASK;
1405         *_start_offset = offset;
1406
1407         count = 1;
1408         if (size > PAGE_SIZE - offset) {
1409                 size -= PAGE_SIZE - offset;
1410                 count += size >> PAGE_SHIFT;
1411                 size &= ~PAGE_MASK;
1412                 if (size)
1413                         count++;
1414         }
1415
1416         if (count > maxpages)
1417                 count = maxpages;
1418
1419         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1420         if (nr == 0)
1421                 return 0;
1422
1423         actual = PAGE_SIZE * nr;
1424         actual -= offset;
1425         if (nr == count && size > 0) {
1426                 unsigned last_offset = (nr > 1) ? 0 : offset;
1427                 actual -= PAGE_SIZE - (last_offset + size);
1428         }
1429         return actual;
1430 }
1431
1432 /* must be done on non-empty ITER_IOVEC one */
1433 static unsigned long first_iovec_segment(const struct iov_iter *i,
1434                                          size_t *size, size_t *start,
1435                                          size_t maxsize, unsigned maxpages)
1436 {
1437         size_t skip;
1438         long k;
1439
1440         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1441                 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1442                 size_t len = i->iov[k].iov_len - skip;
1443
1444                 if (unlikely(!len))
1445                         continue;
1446                 if (len > maxsize)
1447                         len = maxsize;
1448                 len += (*start = addr % PAGE_SIZE);
1449                 if (len > maxpages * PAGE_SIZE)
1450                         len = maxpages * PAGE_SIZE;
1451                 *size = len;
1452                 return addr & PAGE_MASK;
1453         }
1454         BUG(); // if it had been empty, we wouldn't get called
1455 }
1456
1457 /* must be done on non-empty ITER_BVEC one */
1458 static struct page *first_bvec_segment(const struct iov_iter *i,
1459                                        size_t *size, size_t *start,
1460                                        size_t maxsize, unsigned maxpages)
1461 {
1462         struct page *page;
1463         size_t skip = i->iov_offset, len;
1464
1465         len = i->bvec->bv_len - skip;
1466         if (len > maxsize)
1467                 len = maxsize;
1468         skip += i->bvec->bv_offset;
1469         page = i->bvec->bv_page + skip / PAGE_SIZE;
1470         len += (*start = skip % PAGE_SIZE);
1471         if (len > maxpages * PAGE_SIZE)
1472                 len = maxpages * PAGE_SIZE;
1473         *size = len;
1474         return page;
1475 }
1476
1477 ssize_t iov_iter_get_pages(struct iov_iter *i,
1478                    struct page **pages, size_t maxsize, unsigned maxpages,
1479                    size_t *start)
1480 {
1481         size_t len;
1482         int n, res;
1483
1484         if (maxsize > i->count)
1485                 maxsize = i->count;
1486         if (!maxsize)
1487                 return 0;
1488
1489         if (likely(iter_is_iovec(i))) {
1490                 unsigned long addr;
1491
1492                 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1493                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1494                 res = get_user_pages_fast(addr, n,
1495                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0,
1496                                 pages);
1497                 if (unlikely(res < 0))
1498                         return res;
1499                 return (res == n ? len : res * PAGE_SIZE) - *start;
1500         }
1501         if (iov_iter_is_bvec(i)) {
1502                 struct page *page;
1503
1504                 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1505                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1506                 while (n--)
1507                         get_page(*pages++ = page++);
1508                 return len - *start;
1509         }
1510         if (iov_iter_is_pipe(i))
1511                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1512         if (iov_iter_is_xarray(i))
1513                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1514         return -EFAULT;
1515 }
1516 EXPORT_SYMBOL(iov_iter_get_pages);
1517
1518 static struct page **get_pages_array(size_t n)
1519 {
1520         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1521 }
1522
1523 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1524                    struct page ***pages, size_t maxsize,
1525                    size_t *start)
1526 {
1527         struct page **p;
1528         unsigned int iter_head, npages;
1529         ssize_t n;
1530
1531         if (!sanity(i))
1532                 return -EFAULT;
1533
1534         data_start(i, &iter_head, start);
1535         /* Amount of free space: some of this one + all after this one */
1536         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1537         n = npages * PAGE_SIZE - *start;
1538         if (maxsize > n)
1539                 maxsize = n;
1540         else
1541                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1542         p = get_pages_array(npages);
1543         if (!p)
1544                 return -ENOMEM;
1545         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1546         if (n > 0)
1547                 *pages = p;
1548         else
1549                 kvfree(p);
1550         return n;
1551 }
1552
1553 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1554                                            struct page ***pages, size_t maxsize,
1555                                            size_t *_start_offset)
1556 {
1557         struct page **p;
1558         unsigned nr, offset;
1559         pgoff_t index, count;
1560         size_t size = maxsize, actual;
1561         loff_t pos;
1562
1563         if (!size)
1564                 return 0;
1565
1566         pos = i->xarray_start + i->iov_offset;
1567         index = pos >> PAGE_SHIFT;
1568         offset = pos & ~PAGE_MASK;
1569         *_start_offset = offset;
1570
1571         count = 1;
1572         if (size > PAGE_SIZE - offset) {
1573                 size -= PAGE_SIZE - offset;
1574                 count += size >> PAGE_SHIFT;
1575                 size &= ~PAGE_MASK;
1576                 if (size)
1577                         count++;
1578         }
1579
1580         p = get_pages_array(count);
1581         if (!p)
1582                 return -ENOMEM;
1583         *pages = p;
1584
1585         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1586         if (nr == 0)
1587                 return 0;
1588
1589         actual = PAGE_SIZE * nr;
1590         actual -= offset;
1591         if (nr == count && size > 0) {
1592                 unsigned last_offset = (nr > 1) ? 0 : offset;
1593                 actual -= PAGE_SIZE - (last_offset + size);
1594         }
1595         return actual;
1596 }
1597
1598 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1599                    struct page ***pages, size_t maxsize,
1600                    size_t *start)
1601 {
1602         struct page **p;
1603         size_t len;
1604         int n, res;
1605
1606         if (maxsize > i->count)
1607                 maxsize = i->count;
1608         if (!maxsize)
1609                 return 0;
1610
1611         if (likely(iter_is_iovec(i))) {
1612                 unsigned long addr;
1613
1614                 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1615                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1616                 p = get_pages_array(n);
1617                 if (!p)
1618                         return -ENOMEM;
1619                 res = get_user_pages_fast(addr, n,
1620                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0, p);
1621                 if (unlikely(res < 0)) {
1622                         kvfree(p);
1623                         return res;
1624                 }
1625                 *pages = p;
1626                 return (res == n ? len : res * PAGE_SIZE) - *start;
1627         }
1628         if (iov_iter_is_bvec(i)) {
1629                 struct page *page;
1630
1631                 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1632                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1633                 *pages = p = get_pages_array(n);
1634                 if (!p)
1635                         return -ENOMEM;
1636                 while (n--)
1637                         get_page(*p++ = page++);
1638                 return len - *start;
1639         }
1640         if (iov_iter_is_pipe(i))
1641                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1642         if (iov_iter_is_xarray(i))
1643                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1644         return -EFAULT;
1645 }
1646 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1647
1648 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1649                                struct iov_iter *i)
1650 {
1651         __wsum sum, next;
1652         sum = *csum;
1653         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1654                 WARN_ON(1);
1655                 return 0;
1656         }
1657         iterate_and_advance(i, bytes, v, off, ({
1658                 next = csum_and_copy_from_user(v.iov_base,
1659                                                addr + off,
1660                                                v.iov_len);
1661                 if (next)
1662                         sum = csum_block_add(sum, next, off);
1663                 next ? 0 : v.iov_len;
1664         }), ({
1665                 sum = csum_and_memcpy(addr + off, v.iov_base, v.iov_len,
1666                                       sum, off);
1667         })
1668         )
1669         *csum = sum;
1670         return bytes;
1671 }
1672 EXPORT_SYMBOL(csum_and_copy_from_iter);
1673
1674 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1675                              struct iov_iter *i)
1676 {
1677         struct csum_state *csstate = _csstate;
1678         __wsum sum, next;
1679
1680         if (unlikely(iov_iter_is_pipe(i)))
1681                 return csum_and_copy_to_pipe_iter(addr, bytes, _csstate, i);
1682
1683         sum = csum_shift(csstate->csum, csstate->off);
1684         if (unlikely(iov_iter_is_discard(i))) {
1685                 WARN_ON(1);     /* for now */
1686                 return 0;
1687         }
1688         iterate_and_advance(i, bytes, v, off, ({
1689                 next = csum_and_copy_to_user(addr + off,
1690                                              v.iov_base,
1691                                              v.iov_len);
1692                 if (next)
1693                         sum = csum_block_add(sum, next, off);
1694                 next ? 0 : v.iov_len;
1695         }), ({
1696                 sum = csum_and_memcpy(v.iov_base,
1697                                      addr + off,
1698                                      v.iov_len, sum, off);
1699         })
1700         )
1701         csstate->csum = csum_shift(sum, csstate->off);
1702         csstate->off += bytes;
1703         return bytes;
1704 }
1705 EXPORT_SYMBOL(csum_and_copy_to_iter);
1706
1707 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1708                 struct iov_iter *i)
1709 {
1710 #ifdef CONFIG_CRYPTO_HASH
1711         struct ahash_request *hash = hashp;
1712         struct scatterlist sg;
1713         size_t copied;
1714
1715         copied = copy_to_iter(addr, bytes, i);
1716         sg_init_one(&sg, addr, copied);
1717         ahash_request_set_crypt(hash, &sg, NULL, copied);
1718         crypto_ahash_update(hash);
1719         return copied;
1720 #else
1721         return 0;
1722 #endif
1723 }
1724 EXPORT_SYMBOL(hash_and_copy_to_iter);
1725
1726 static int iov_npages(const struct iov_iter *i, int maxpages)
1727 {
1728         size_t skip = i->iov_offset, size = i->count;
1729         const struct iovec *p;
1730         int npages = 0;
1731
1732         for (p = i->iov; size; skip = 0, p++) {
1733                 unsigned offs = offset_in_page(p->iov_base + skip);
1734                 size_t len = min(p->iov_len - skip, size);
1735
1736                 if (len) {
1737                         size -= len;
1738                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1739                         if (unlikely(npages > maxpages))
1740                                 return maxpages;
1741                 }
1742         }
1743         return npages;
1744 }
1745
1746 static int bvec_npages(const struct iov_iter *i, int maxpages)
1747 {
1748         size_t skip = i->iov_offset, size = i->count;
1749         const struct bio_vec *p;
1750         int npages = 0;
1751
1752         for (p = i->bvec; size; skip = 0, p++) {
1753                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1754                 size_t len = min(p->bv_len - skip, size);
1755
1756                 size -= len;
1757                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1758                 if (unlikely(npages > maxpages))
1759                         return maxpages;
1760         }
1761         return npages;
1762 }
1763
1764 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1765 {
1766         if (unlikely(!i->count))
1767                 return 0;
1768         /* iovec and kvec have identical layouts */
1769         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1770                 return iov_npages(i, maxpages);
1771         if (iov_iter_is_bvec(i))
1772                 return bvec_npages(i, maxpages);
1773         if (iov_iter_is_pipe(i)) {
1774                 unsigned int iter_head;
1775                 int npages;
1776                 size_t off;
1777
1778                 if (!sanity(i))
1779                         return 0;
1780
1781                 data_start(i, &iter_head, &off);
1782                 /* some of this one + all after this one */
1783                 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1784                 return min(npages, maxpages);
1785         }
1786         if (iov_iter_is_xarray(i)) {
1787                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1788                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1789                 return min(npages, maxpages);
1790         }
1791         return 0;
1792 }
1793 EXPORT_SYMBOL(iov_iter_npages);
1794
1795 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1796 {
1797         *new = *old;
1798         if (unlikely(iov_iter_is_pipe(new))) {
1799                 WARN_ON(1);
1800                 return NULL;
1801         }
1802         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1803                 return NULL;
1804         if (iov_iter_is_bvec(new))
1805                 return new->bvec = kmemdup(new->bvec,
1806                                     new->nr_segs * sizeof(struct bio_vec),
1807                                     flags);
1808         else
1809                 /* iovec and kvec have identical layout */
1810                 return new->iov = kmemdup(new->iov,
1811                                    new->nr_segs * sizeof(struct iovec),
1812                                    flags);
1813 }
1814 EXPORT_SYMBOL(dup_iter);
1815
1816 static int copy_compat_iovec_from_user(struct iovec *iov,
1817                 const struct iovec __user *uvec, unsigned long nr_segs)
1818 {
1819         const struct compat_iovec __user *uiov =
1820                 (const struct compat_iovec __user *)uvec;
1821         int ret = -EFAULT, i;
1822
1823         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1824                 return -EFAULT;
1825
1826         for (i = 0; i < nr_segs; i++) {
1827                 compat_uptr_t buf;
1828                 compat_ssize_t len;
1829
1830                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1831                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1832
1833                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1834                 if (len < 0) {
1835                         ret = -EINVAL;
1836                         goto uaccess_end;
1837                 }
1838                 iov[i].iov_base = compat_ptr(buf);
1839                 iov[i].iov_len = len;
1840         }
1841
1842         ret = 0;
1843 uaccess_end:
1844         user_access_end();
1845         return ret;
1846 }
1847
1848 static int copy_iovec_from_user(struct iovec *iov,
1849                 const struct iovec __user *uvec, unsigned long nr_segs)
1850 {
1851         unsigned long seg;
1852
1853         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1854                 return -EFAULT;
1855         for (seg = 0; seg < nr_segs; seg++) {
1856                 if ((ssize_t)iov[seg].iov_len < 0)
1857                         return -EINVAL;
1858         }
1859
1860         return 0;
1861 }
1862
1863 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1864                 unsigned long nr_segs, unsigned long fast_segs,
1865                 struct iovec *fast_iov, bool compat)
1866 {
1867         struct iovec *iov = fast_iov;
1868         int ret;
1869
1870         /*
1871          * SuS says "The readv() function *may* fail if the iovcnt argument was
1872          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1873          * traditionally returned zero for zero segments, so...
1874          */
1875         if (nr_segs == 0)
1876                 return iov;
1877         if (nr_segs > UIO_MAXIOV)
1878                 return ERR_PTR(-EINVAL);
1879         if (nr_segs > fast_segs) {
1880                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1881                 if (!iov)
1882                         return ERR_PTR(-ENOMEM);
1883         }
1884
1885         if (compat)
1886                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1887         else
1888                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1889         if (ret) {
1890                 if (iov != fast_iov)
1891                         kfree(iov);
1892                 return ERR_PTR(ret);
1893         }
1894
1895         return iov;
1896 }
1897
1898 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1899                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1900                  struct iov_iter *i, bool compat)
1901 {
1902         ssize_t total_len = 0;
1903         unsigned long seg;
1904         struct iovec *iov;
1905
1906         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1907         if (IS_ERR(iov)) {
1908                 *iovp = NULL;
1909                 return PTR_ERR(iov);
1910         }
1911
1912         /*
1913          * According to the Single Unix Specification we should return EINVAL if
1914          * an element length is < 0 when cast to ssize_t or if the total length
1915          * would overflow the ssize_t return value of the system call.
1916          *
1917          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1918          * overflow case.
1919          */
1920         for (seg = 0; seg < nr_segs; seg++) {
1921                 ssize_t len = (ssize_t)iov[seg].iov_len;
1922
1923                 if (!access_ok(iov[seg].iov_base, len)) {
1924                         if (iov != *iovp)
1925                                 kfree(iov);
1926                         *iovp = NULL;
1927                         return -EFAULT;
1928                 }
1929
1930                 if (len > MAX_RW_COUNT - total_len) {
1931                         len = MAX_RW_COUNT - total_len;
1932                         iov[seg].iov_len = len;
1933                 }
1934                 total_len += len;
1935         }
1936
1937         iov_iter_init(i, type, iov, nr_segs, total_len);
1938         if (iov == *iovp)
1939                 *iovp = NULL;
1940         else
1941                 *iovp = iov;
1942         return total_len;
1943 }
1944
1945 /**
1946  * import_iovec() - Copy an array of &struct iovec from userspace
1947  *     into the kernel, check that it is valid, and initialize a new
1948  *     &struct iov_iter iterator to access it.
1949  *
1950  * @type: One of %READ or %WRITE.
1951  * @uvec: Pointer to the userspace array.
1952  * @nr_segs: Number of elements in userspace array.
1953  * @fast_segs: Number of elements in @iov.
1954  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1955  *     on-stack) kernel array.
1956  * @i: Pointer to iterator that will be initialized on success.
1957  *
1958  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1959  * then this function places %NULL in *@iov on return. Otherwise, a new
1960  * array will be allocated and the result placed in *@iov. This means that
1961  * the caller may call kfree() on *@iov regardless of whether the small
1962  * on-stack array was used or not (and regardless of whether this function
1963  * returns an error or not).
1964  *
1965  * Return: Negative error code on error, bytes imported on success
1966  */
1967 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1968                  unsigned nr_segs, unsigned fast_segs,
1969                  struct iovec **iovp, struct iov_iter *i)
1970 {
1971         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1972                               in_compat_syscall());
1973 }
1974 EXPORT_SYMBOL(import_iovec);
1975
1976 int import_single_range(int rw, void __user *buf, size_t len,
1977                  struct iovec *iov, struct iov_iter *i)
1978 {
1979         if (len > MAX_RW_COUNT)
1980                 len = MAX_RW_COUNT;
1981         if (unlikely(!access_ok(buf, len)))
1982                 return -EFAULT;
1983
1984         iov->iov_base = buf;
1985         iov->iov_len = len;
1986         iov_iter_init(i, rw, iov, 1, len);
1987         return 0;
1988 }
1989 EXPORT_SYMBOL(import_single_range);