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, base, len, off, __p, STEP) {        \
  21         size_t off = 0;                                         \
  22         size_t skip = i->iov_offset;                            \
  23         do {                                                    \
  24                 len = min(n, __p->iov_len - skip);              \
  25                 if (likely(len)) {                              \
  26                         base = __p->iov_base + skip;            \
  27                         len -= (STEP);                          \
  28                         off += len;                             \
  29                         skip += len;                            \
  30                         n -= len;                               \
  31                         if (skip < __p->iov_len)                \
  32                                 break;                          \
  33                 }                                               \
  34                 __p++;                                          \
  35                 skip = 0;                                       \
  36         } while (n);                                            \
  37         i->iov_offset = skip;                                   \
  38         n = off;                                                \
  39 }
  40
  41 #define iterate_bvec(i, n, base, len, off, p, STEP) {           \
  42         size_t off = 0;                                         \
  43         unsigned skip = i->iov_offset;                          \
  44         while (n) {                                             \
  45                 unsigned offset = p->bv_offset + skip;          \
  46                 unsigned left;                                  \
  47                 void *kaddr = kmap_local_page(p->bv_page +      \
  48                                         offset / PAGE_SIZE);    \
  49                 base = kaddr + offset % PAGE_SIZE;              \
  50                 len = min(min(n, (size_t)(p->bv_len - skip)),   \
  51                      (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
  52                 left = (STEP);                                  \
  53                 kunmap_local(kaddr);                            \
  54                 len -= left;                                    \
  55                 off += len;                                     \
  56                 skip += len;                                    \
  57                 if (skip == p->bv_len) {                        \
  58                         skip = 0;                               \
  59                         p++;                                    \
  60                 }                                               \
  61                 n -= len;                                       \
  62                 if (left)                                       \
  63                         break;                                  \
  64         }                                                       \
  65         i->iov_offset = skip;                                   \
  66         n = off;                                                \
  67 }
  68
  69 #define iterate_xarray(i, n, base, len, __off, STEP) {          \
  70         __label__ __out;                                        \
  71         size_t __off = 0;                                       \
  72         struct folio *folio;                                    \
  73         loff_t start = i->xarray_start + i->iov_offset;         \
  74         pgoff_t index = start / PAGE_SIZE;                      \
  75         XA_STATE(xas, i->xarray, index);                        \
  76                                                                 \
  77         len = PAGE_SIZE - offset_in_page(start);                \
  78         rcu_read_lock();                                        \
  79         xas_for_each(&xas, folio, ULONG_MAX) {                  \
  80                 unsigned left;                                  \
  81                 size_t offset;                                  \
  82                 if (xas_retry(&xas, folio))                     \
  83                         continue;                               \
  84                 if (WARN_ON(xa_is_value(folio)))                \
  85                         break;                                  \
  86                 if (WARN_ON(folio_test_hugetlb(folio)))         \
  87                         break;                                  \
  88                 offset = offset_in_folio(folio, start + __off); \
  89                 while (offset < folio_size(folio)) {            \
  90                         base = kmap_local_folio(folio, offset); \
  91                         len = min(n, len);                      \
  92                         left = (STEP);                          \
  93                         kunmap_local(base);                     \
  94                         len -= left;                            \
  95                         __off += len;                           \
  96                         n -= len;                               \
  97                         if (left || n == 0)                     \
  98                                 goto __out;                     \
  99                         offset += len;                          \
 100                         len = PAGE_SIZE;                        \
 101                 }                                               \
 102         }                                                       \
 103 __out:                                                          \
 104         rcu_read_unlock();                                      \
 105         i->iov_offset += __off;                                 \
 106         n = __off;                                              \
 107 }
 108
 109 #define __iterate_and_advance(i, n, base, len, off, I, K) {     \
 110         if (unlikely(i->count < n))                             \
 111                 n = i->count;                                   \
 112         if (likely(n)) {                                        \
 113                 if (likely(iter_is_iovec(i))) {                 \
 114                         const struct iovec *iov = i->iov;       \
 115                         void __user *base;                      \
 116                         size_t len;                             \
 117                         iterate_iovec(i, n, base, len, off,     \
 118                                                 iov, (I))       \
 119                         i->nr_segs -= iov - i->iov;             \
 120                         i->iov = iov;                           \
 121                 } else if (iov_iter_is_bvec(i)) {               \
 122                         const struct bio_vec *bvec = i->bvec;   \
 123                         void *base;                             \
 124                         size_t len;                             \
 125                         iterate_bvec(i, n, base, len, off,      \
 126                                                 bvec, (K))      \
 127                         i->nr_segs -= bvec - i->bvec;           \
 128                         i->bvec = bvec;                         \
 129                 } else if (iov_iter_is_kvec(i)) {               \
 130                         const struct kvec *kvec = i->kvec;      \
 131                         void *base;                             \
 132                         size_t len;                             \
 133                         iterate_iovec(i, n, base, len, off,     \
 134                                                 kvec, (K))      \
 135                         i->nr_segs -= kvec - i->kvec;           \
 136                         i->kvec = kvec;                         \
 137                 } else if (iov_iter_is_xarray(i)) {             \
 138                         void *base;                             \
 139                         size_t len;                             \
 140                         iterate_xarray(i, n, base, len, off,    \
 141                                                         (K))    \
 142                 }                                               \
 143                 i->count -= n;                                  \
 144         }                                                       \
 145 }
 146 #define iterate_and_advance(i, n, base, len, off, I, K) \
 147         __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
 148
 149 static int copyout(void __user *to, const void *from, size_t n)
 150 {
 151         if (should_fail_usercopy())
 152                 return n;
 153         if (access_ok(to, n)) {
 154                 instrument_copy_to_user(to, from, n);
 155                 n = raw_copy_to_user(to, from, n);
 156         }
 157         return n;
 158 }
 159
 160 static int copyin(void *to, const void __user *from, size_t n)
 161 {
 162         if (should_fail_usercopy())
 163                 return n;
 164         if (access_ok(from, n)) {
 165                 instrument_copy_from_user(to, from, n);
 166                 n = raw_copy_from_user(to, from, n);
 167         }
 168         return n;
 169 }
 170
 171 #ifdef PIPE_PARANOIA
 172 static bool sanity(const struct iov_iter *i)
 173 {
 174         struct pipe_inode_info *pipe = i->pipe;
 175         unsigned int p_head = pipe->head;
 176         unsigned int p_tail = pipe->tail;
 177         unsigned int p_mask = pipe->ring_size - 1;
 178         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 179         unsigned int i_head = i->head;
 180         unsigned int idx;
 181
 182         if (i->iov_offset) {
 183                 struct pipe_buffer *p;
 184                 if (unlikely(p_occupancy == 0))
 185                         goto Bad;       // pipe must be non-empty
 186                 if (unlikely(i_head != p_head - 1))
 187                         goto Bad;       // must be at the last buffer...
 188
 189                 p = &pipe->bufs[i_head & p_mask];
 190                 if (unlikely(p->offset + p->len != i->iov_offset))
 191                         goto Bad;       // ... at the end of segment
 192         } else {
 193                 if (i_head != p_head)
 194                         goto Bad;       // must be right after the last buffer
 195         }
 196         return true;
 197 Bad:
 198         printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
 199         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 200                         p_head, p_tail, pipe->ring_size);
 201         for (idx = 0; idx < pipe->ring_size; idx++)
 202                 printk(KERN_ERR "[%p %p %d %d]\n",
 203                         pipe->bufs[idx].ops,
 204                         pipe->bufs[idx].page,
 205                         pipe->bufs[idx].offset,
 206                         pipe->bufs[idx].len);
 207         WARN_ON(1);
 208         return false;
 209 }
 210 #else
 211 #define sanity(i) true
 212 #endif
 213
 214 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 215                          struct iov_iter *i)
 216 {
 217         struct pipe_inode_info *pipe = i->pipe;
 218         struct pipe_buffer *buf;
 219         unsigned int p_tail = pipe->tail;
 220         unsigned int p_mask = pipe->ring_size - 1;
 221         unsigned int i_head = i->head;
 222         size_t off;
 223
 224         if (unlikely(bytes > i->count))
 225                 bytes = i->count;
 226
 227         if (unlikely(!bytes))
 228                 return 0;
 229
 230         if (!sanity(i))
 231                 return 0;
 232
 233         off = i->iov_offset;
 234         buf = &pipe->bufs[i_head & p_mask];
 235         if (off) {
 236                 if (offset == off && buf->page == page) {
 237                         /* merge with the last one */
 238                         buf->len += bytes;
 239                         i->iov_offset += bytes;
 240                         goto out;
 241                 }
 242                 i_head++;
 243                 buf = &pipe->bufs[i_head & p_mask];
 244         }
 245         if (pipe_full(i_head, p_tail, pipe->max_usage))
 246                 return 0;
 247
 248         buf->ops = &page_cache_pipe_buf_ops;
 249         buf->flags = 0;
 250         get_page(page);
 251         buf->page = page;
 252         buf->offset = offset;
 253         buf->len = bytes;
 254
 255         pipe->head = i_head + 1;
 256         i->iov_offset = offset + bytes;
 257         i->head = i_head;
 258 out:
 259         i->count -= bytes;
 260         return bytes;
 261 }
 262
 263 /*
 264  * fault_in_iov_iter_readable - fault in iov iterator for reading
 265  * @i: iterator
 266  * @size: maximum length
 267  *
 268  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 269  * @size.  For each iovec, fault in each page that constitutes the iovec.
 270  *
 271  * Returns the number of bytes not faulted in (like copy_to_user() and
 272  * copy_from_user()).
 273  *
 274  * Always returns 0 for non-userspace iterators.
 275  */
 276 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
 277 {
 278         if (iter_is_iovec(i)) {
 279                 size_t count = min(size, iov_iter_count(i));
 280                 const struct iovec *p;
 281                 size_t skip;
 282
 283                 size -= count;
 284                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 285                         size_t len = min(count, p->iov_len - skip);
 286                         size_t ret;
 287
 288                         if (unlikely(!len))
 289                                 continue;
 290                         ret = fault_in_readable(p->iov_base + skip, len);
 291                         count -= len - ret;
 292                         if (ret)
 293                                 break;
 294                 }
 295                 return count + size;
 296         }
 297         return 0;
 298 }
 299 EXPORT_SYMBOL(fault_in_iov_iter_readable);
 300
 301 /*
 302  * fault_in_iov_iter_writeable - fault in iov iterator for writing
 303  * @i: iterator
 304  * @size: maximum length
 305  *
 306  * Faults in the iterator using get_user_pages(), i.e., without triggering
 307  * hardware page faults.  This is primarily useful when we already know that
 308  * some or all of the pages in @i aren't in memory.
 309  *
 310  * Returns the number of bytes not faulted in, like copy_to_user() and
 311  * copy_from_user().
 312  *
 313  * Always returns 0 for non-user-space iterators.
 314  */
 315 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
 316 {
 317         if (iter_is_iovec(i)) {
 318                 size_t count = min(size, iov_iter_count(i));
 319                 const struct iovec *p;
 320                 size_t skip;
 321
 322                 size -= count;
 323                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 324                         size_t len = min(count, p->iov_len - skip);
 325                         size_t ret;
 326
 327                         if (unlikely(!len))
 328                                 continue;
 329                         ret = fault_in_safe_writeable(p->iov_base + skip, len);
 330                         count -= len - ret;
 331                         if (ret)
 332                                 break;
 333                 }
 334                 return count + size;
 335         }
 336         return 0;
 337 }
 338 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
 339
 340 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 341                         const struct iovec *iov, unsigned long nr_segs,
 342                         size_t count)
 343 {
 344         WARN_ON(direction & ~(READ | WRITE));
 345         *i = (struct iov_iter) {
 346                 .iter_type = ITER_IOVEC,
 347                 .nofault = false,
 348                 .data_source = direction,
 349                 .iov = iov,
 350                 .nr_segs = nr_segs,
 351                 .iov_offset = 0,
 352                 .count = count
 353         };
 354 }
 355 EXPORT_SYMBOL(iov_iter_init);
 356
 357 static inline bool allocated(struct pipe_buffer *buf)
 358 {
 359         return buf->ops == &default_pipe_buf_ops;
 360 }
 361
 362 static inline void data_start(const struct iov_iter *i,
 363                               unsigned int *iter_headp, size_t *offp)
 364 {
 365         unsigned int p_mask = i->pipe->ring_size - 1;
 366         unsigned int iter_head = i->head;
 367         size_t off = i->iov_offset;
 368
 369         if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
 370                     off == PAGE_SIZE)) {
 371                 iter_head++;
 372                 off = 0;
 373         }
 374         *iter_headp = iter_head;
 375         *offp = off;
 376 }
 377
 378 static size_t push_pipe(struct iov_iter *i, size_t size,
 379                         int *iter_headp, size_t *offp)
 380 {
 381         struct pipe_inode_info *pipe = i->pipe;
 382         unsigned int p_tail = pipe->tail;
 383         unsigned int p_mask = pipe->ring_size - 1;
 384         unsigned int iter_head;
 385         size_t off;
 386         ssize_t left;
 387
 388         if (unlikely(size > i->count))
 389                 size = i->count;
 390         if (unlikely(!size))
 391                 return 0;
 392
 393         left = size;
 394         data_start(i, &iter_head, &off);
 395         *iter_headp = iter_head;
 396         *offp = off;
 397         if (off) {
 398                 left -= PAGE_SIZE - off;
 399                 if (left <= 0) {
 400                         pipe->bufs[iter_head & p_mask].len += size;
 401                         return size;
 402                 }
 403                 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
 404                 iter_head++;
 405         }
 406         while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
 407                 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
 408                 struct page *page = alloc_page(GFP_USER);
 409                 if (!page)
 410                         break;
 411
 412                 buf->ops = &default_pipe_buf_ops;
 413                 buf->flags = 0;
 414                 buf->page = page;
 415                 buf->offset = 0;
 416                 buf->len = min_t(ssize_t, left, PAGE_SIZE);
 417                 left -= buf->len;
 418                 iter_head++;
 419                 pipe->head = iter_head;
 420
 421                 if (left == 0)
 422                         return size;
 423         }
 424         return size - left;
 425 }
 426
 427 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 428                                 struct iov_iter *i)
 429 {
 430         struct pipe_inode_info *pipe = i->pipe;
 431         unsigned int p_mask = pipe->ring_size - 1;
 432         unsigned int i_head;
 433         size_t n, off;
 434
 435         if (!sanity(i))
 436                 return 0;
 437
 438         bytes = n = push_pipe(i, bytes, &i_head, &off);
 439         if (unlikely(!n))
 440                 return 0;
 441         do {
 442                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 443                 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
 444                 i->head = i_head;
 445                 i->iov_offset = off + chunk;
 446                 n -= chunk;
 447                 addr += chunk;
 448                 off = 0;
 449                 i_head++;
 450         } while (n);
 451         i->count -= bytes;
 452         return bytes;
 453 }
 454
 455 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 456                               __wsum sum, size_t off)
 457 {
 458         __wsum next = csum_partial_copy_nocheck(from, to, len);
 459         return csum_block_add(sum, next, off);
 460 }
 461
 462 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 463                                          struct iov_iter *i, __wsum *sump)
 464 {
 465         struct pipe_inode_info *pipe = i->pipe;
 466         unsigned int p_mask = pipe->ring_size - 1;
 467         __wsum sum = *sump;
 468         size_t off = 0;
 469         unsigned int i_head;
 470         size_t r;
 471
 472         if (!sanity(i))
 473                 return 0;
 474
 475         bytes = push_pipe(i, bytes, &i_head, &r);
 476         while (bytes) {
 477                 size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r);
 478                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 479                 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
 480                 kunmap_local(p);
 481                 i->head = i_head;
 482                 i->iov_offset = r + chunk;
 483                 bytes -= chunk;
 484                 off += chunk;
 485                 r = 0;
 486                 i_head++;
 487         }
 488         *sump = sum;
 489         i->count -= off;
 490         return off;
 491 }
 492
 493 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 494 {
 495         if (unlikely(iov_iter_is_pipe(i)))
 496                 return copy_pipe_to_iter(addr, bytes, i);
 497         if (iter_is_iovec(i))
 498                 might_fault();
 499         iterate_and_advance(i, bytes, base, len, off,
 500                 copyout(base, addr + off, len),
 501                 memcpy(base, addr + off, len)
 502         )
 503
 504         return bytes;
 505 }
 506 EXPORT_SYMBOL(_copy_to_iter);
 507
 508 #ifdef CONFIG_ARCH_HAS_COPY_MC
 509 static int copyout_mc(void __user *to, const void *from, size_t n)
 510 {
 511         if (access_ok(to, n)) {
 512                 instrument_copy_to_user(to, from, n);
 513                 n = copy_mc_to_user((__force void *) to, from, n);
 514         }
 515         return n;
 516 }
 517
 518 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 519                                 struct iov_iter *i)
 520 {
 521         struct pipe_inode_info *pipe = i->pipe;
 522         unsigned int p_mask = pipe->ring_size - 1;
 523         unsigned int i_head;
 524         unsigned int valid = pipe->head;
 525         size_t n, off, xfer = 0;
 526
 527         if (!sanity(i))
 528                 return 0;
 529
 530         n = push_pipe(i, bytes, &i_head, &off);
 531         while (n) {
 532                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 533                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 534                 unsigned long rem;
 535                 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
 536                 chunk -= rem;
 537                 kunmap_local(p);
 538                 if (chunk) {
 539                         i->head = i_head;
 540                         i->iov_offset = off + chunk;
 541                         xfer += chunk;
 542                         valid = i_head + 1;
 543                 }
 544                 if (rem) {
 545                         pipe->bufs[i_head & p_mask].len -= rem;
 546                         pipe_discard_from(pipe, valid);
 547                         break;
 548                 }
 549                 n -= chunk;
 550                 off = 0;
 551                 i_head++;
 552         }
 553         i->count -= xfer;
 554         return xfer;
 555 }
 556
 557 /**
 558  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 559  * @addr: source kernel address
 560  * @bytes: total transfer length
 561  * @i: destination iterator
 562  *
 563  * The pmem driver deploys this for the dax operation
 564  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 565  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 566  * successfully copied.
 567  *
 568  * The main differences between this and typical _copy_to_iter().
 569  *
 570  * * Typical tail/residue handling after a fault retries the copy
 571  *   byte-by-byte until the fault happens again. Re-triggering machine
 572  *   checks is potentially fatal so the implementation uses source
 573  *   alignment and poison alignment assumptions to avoid re-triggering
 574  *   hardware exceptions.
 575  *
 576  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 577  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 578  *   a short copy.
 579  *
 580  * Return: number of bytes copied (may be %0)
 581  */
 582 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 583 {
 584         if (unlikely(iov_iter_is_pipe(i)))
 585                 return copy_mc_pipe_to_iter(addr, bytes, i);
 586         if (iter_is_iovec(i))
 587                 might_fault();
 588         __iterate_and_advance(i, bytes, base, len, off,
 589                 copyout_mc(base, addr + off, len),
 590                 copy_mc_to_kernel(base, addr + off, len)
 591         )
 592
 593         return bytes;
 594 }
 595 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 596 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 597
 598 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 599 {
 600         if (unlikely(iov_iter_is_pipe(i))) {
 601                 WARN_ON(1);
 602                 return 0;
 603         }
 604         if (iter_is_iovec(i))
 605                 might_fault();
 606         iterate_and_advance(i, bytes, base, len, off,
 607                 copyin(addr + off, base, len),
 608                 memcpy(addr + off, base, len)
 609         )
 610
 611         return bytes;
 612 }
 613 EXPORT_SYMBOL(_copy_from_iter);
 614
 615 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 616 {
 617         if (unlikely(iov_iter_is_pipe(i))) {
 618                 WARN_ON(1);
 619                 return 0;
 620         }
 621         iterate_and_advance(i, bytes, base, len, off,
 622                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 623                 memcpy(addr + off, base, len)
 624         )
 625
 626         return bytes;
 627 }
 628 EXPORT_SYMBOL(_copy_from_iter_nocache);
 629
 630 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 631 /**
 632  * _copy_from_iter_flushcache - write destination through cpu cache
 633  * @addr: destination kernel address
 634  * @bytes: total transfer length
 635  * @i: source iterator
 636  *
 637  * The pmem driver arranges for filesystem-dax to use this facility via
 638  * dax_copy_from_iter() for ensuring that writes to persistent memory
 639  * are flushed through the CPU cache. It is differentiated from
 640  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 641  * all iterator types. The _copy_from_iter_nocache() only attempts to
 642  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 643  * instructions that strand dirty-data in the cache.
 644  *
 645  * Return: number of bytes copied (may be %0)
 646  */
 647 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 648 {
 649         if (unlikely(iov_iter_is_pipe(i))) {
 650                 WARN_ON(1);
 651                 return 0;
 652         }
 653         iterate_and_advance(i, bytes, base, len, off,
 654                 __copy_from_user_flushcache(addr + off, base, len),
 655                 memcpy_flushcache(addr + off, base, len)
 656         )
 657
 658         return bytes;
 659 }
 660 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 661 #endif
 662
 663 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 664 {
 665         struct page *head;
 666         size_t v = n + offset;
 667
 668         /*
 669          * The general case needs to access the page order in order
 670          * to compute the page size.
 671          * However, we mostly deal with order-0 pages and thus can
 672          * avoid a possible cache line miss for requests that fit all
 673          * page orders.
 674          */
 675         if (n <= v && v <= PAGE_SIZE)
 676                 return true;
 677
 678         head = compound_head(page);
 679         v += (page - head) << PAGE_SHIFT;
 680
 681         if (likely(n <= v && v <= (page_size(head))))
 682                 return true;
 683         WARN_ON(1);
 684         return false;
 685 }
 686
 687 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 688                          struct iov_iter *i)
 689 {
 690         if (unlikely(iov_iter_is_pipe(i))) {
 691                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 692         } else {
 693                 void *kaddr = kmap_local_page(page);
 694                 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
 695                 kunmap_local(kaddr);
 696                 return wanted;
 697         }
 698 }
 699
 700 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 701                          struct iov_iter *i)
 702 {
 703         size_t res = 0;
 704         if (unlikely(!page_copy_sane(page, offset, bytes)))
 705                 return 0;
 706         page += offset / PAGE_SIZE; // first subpage
 707         offset %= PAGE_SIZE;
 708         while (1) {
 709                 size_t n = __copy_page_to_iter(page, offset,
 710                                 min(bytes, (size_t)PAGE_SIZE - offset), i);
 711                 res += n;
 712                 bytes -= n;
 713                 if (!bytes || !n)
 714                         break;
 715                 offset += n;
 716                 if (offset == PAGE_SIZE) {
 717                         page++;
 718                         offset = 0;
 719                 }
 720         }
 721         return res;
 722 }
 723 EXPORT_SYMBOL(copy_page_to_iter);
 724
 725 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 726                          struct iov_iter *i)
 727 {
 728         if (page_copy_sane(page, offset, bytes)) {
 729                 void *kaddr = kmap_local_page(page);
 730                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 731                 kunmap_local(kaddr);
 732                 return wanted;
 733         }
 734         return 0;
 735 }
 736 EXPORT_SYMBOL(copy_page_from_iter);
 737
 738 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 739 {
 740         struct pipe_inode_info *pipe = i->pipe;
 741         unsigned int p_mask = pipe->ring_size - 1;
 742         unsigned int i_head;
 743         size_t n, off;
 744
 745         if (!sanity(i))
 746                 return 0;
 747
 748         bytes = n = push_pipe(i, bytes, &i_head, &off);
 749         if (unlikely(!n))
 750                 return 0;
 751
 752         do {
 753                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 754                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 755                 memset(p + off, 0, chunk);
 756                 kunmap_local(p);
 757                 i->head = i_head;
 758                 i->iov_offset = off + chunk;
 759                 n -= chunk;
 760                 off = 0;
 761                 i_head++;
 762         } while (n);
 763         i->count -= bytes;
 764         return bytes;
 765 }
 766
 767 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 768 {
 769         if (unlikely(iov_iter_is_pipe(i)))
 770                 return pipe_zero(bytes, i);
 771         iterate_and_advance(i, bytes, base, len, count,
 772                 clear_user(base, len),
 773                 memset(base, 0, len)
 774         )
 775
 776         return bytes;
 777 }
 778 EXPORT_SYMBOL(iov_iter_zero);
 779
 780 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 781                                   struct iov_iter *i)
 782 {
 783         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 784         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 785                 kunmap_atomic(kaddr);
 786                 return 0;
 787         }
 788         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 789                 kunmap_atomic(kaddr);
 790                 WARN_ON(1);
 791                 return 0;
 792         }
 793         iterate_and_advance(i, bytes, base, len, off,
 794                 copyin(p + off, base, len),
 795                 memcpy(p + off, base, len)
 796         )
 797         kunmap_atomic(kaddr);
 798         return bytes;
 799 }
 800 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 801
 802 static inline void pipe_truncate(struct iov_iter *i)
 803 {
 804         struct pipe_inode_info *pipe = i->pipe;
 805         unsigned int p_tail = pipe->tail;
 806         unsigned int p_head = pipe->head;
 807         unsigned int p_mask = pipe->ring_size - 1;
 808
 809         if (!pipe_empty(p_head, p_tail)) {
 810                 struct pipe_buffer *buf;
 811                 unsigned int i_head = i->head;
 812                 size_t off = i->iov_offset;
 813
 814                 if (off) {
 815                         buf = &pipe->bufs[i_head & p_mask];
 816                         buf->len = off - buf->offset;
 817                         i_head++;
 818                 }
 819                 while (p_head != i_head) {
 820                         p_head--;
 821                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
 822                 }
 823
 824                 pipe->head = p_head;
 825         }
 826 }
 827
 828 static void pipe_advance(struct iov_iter *i, size_t size)
 829 {
 830         struct pipe_inode_info *pipe = i->pipe;
 831         if (size) {
 832                 struct pipe_buffer *buf;
 833                 unsigned int p_mask = pipe->ring_size - 1;
 834                 unsigned int i_head = i->head;
 835                 size_t off = i->iov_offset, left = size;
 836
 837                 if (off) /* make it relative to the beginning of buffer */
 838                         left += off - pipe->bufs[i_head & p_mask].offset;
 839                 while (1) {
 840                         buf = &pipe->bufs[i_head & p_mask];
 841                         if (left <= buf->len)
 842                                 break;
 843                         left -= buf->len;
 844                         i_head++;
 845                 }
 846                 i->head = i_head;
 847                 i->iov_offset = buf->offset + left;
 848         }
 849         i->count -= size;
 850         /* ... and discard everything past that point */
 851         pipe_truncate(i);
 852 }
 853
 854 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 855 {
 856         const struct bio_vec *bvec, *end;
 857
 858         if (!i->count)
 859                 return;
 860         i->count -= size;
 861
 862         size += i->iov_offset;
 863
 864         for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
 865                 if (likely(size < bvec->bv_len))
 866                         break;
 867                 size -= bvec->bv_len;
 868         }
 869         i->iov_offset = size;
 870         i->nr_segs -= bvec - i->bvec;
 871         i->bvec = bvec;
 872 }
 873
 874 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 875 {
 876         const struct iovec *iov, *end;
 877
 878         if (!i->count)
 879                 return;
 880         i->count -= size;
 881
 882         size += i->iov_offset; // from beginning of current segment
 883         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
 884                 if (likely(size < iov->iov_len))
 885                         break;
 886                 size -= iov->iov_len;
 887         }
 888         i->iov_offset = size;
 889         i->nr_segs -= iov - i->iov;
 890         i->iov = iov;
 891 }
 892
 893 void iov_iter_advance(struct iov_iter *i, size_t size)
 894 {
 895         if (unlikely(i->count < size))
 896                 size = i->count;
 897         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 898                 /* iovec and kvec have identical layouts */
 899                 iov_iter_iovec_advance(i, size);
 900         } else if (iov_iter_is_bvec(i)) {
 901                 iov_iter_bvec_advance(i, size);
 902         } else if (iov_iter_is_pipe(i)) {
 903                 pipe_advance(i, size);
 904         } else if (unlikely(iov_iter_is_xarray(i))) {
 905                 i->iov_offset += size;
 906                 i->count -= size;
 907         } else if (iov_iter_is_discard(i)) {
 908                 i->count -= size;
 909         }
 910 }
 911 EXPORT_SYMBOL(iov_iter_advance);
 912
 913 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 914 {
 915         if (!unroll)
 916                 return;
 917         if (WARN_ON(unroll > MAX_RW_COUNT))
 918                 return;
 919         i->count += unroll;
 920         if (unlikely(iov_iter_is_pipe(i))) {
 921                 struct pipe_inode_info *pipe = i->pipe;
 922                 unsigned int p_mask = pipe->ring_size - 1;
 923                 unsigned int i_head = i->head;
 924                 size_t off = i->iov_offset;
 925                 while (1) {
 926                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
 927                         size_t n = off - b->offset;
 928                         if (unroll < n) {
 929                                 off -= unroll;
 930                                 break;
 931                         }
 932                         unroll -= n;
 933                         if (!unroll && i_head == i->start_head) {
 934                                 off = 0;
 935                                 break;
 936                         }
 937                         i_head--;
 938                         b = &pipe->bufs[i_head & p_mask];
 939                         off = b->offset + b->len;
 940                 }
 941                 i->iov_offset = off;
 942                 i->head = i_head;
 943                 pipe_truncate(i);
 944                 return;
 945         }
 946         if (unlikely(iov_iter_is_discard(i)))
 947                 return;
 948         if (unroll <= i->iov_offset) {
 949                 i->iov_offset -= unroll;
 950                 return;
 951         }
 952         unroll -= i->iov_offset;
 953         if (iov_iter_is_xarray(i)) {
 954                 BUG(); /* We should never go beyond the start of the specified
 955                         * range since we might then be straying into pages that
 956                         * aren't pinned.
 957                         */
 958         } else if (iov_iter_is_bvec(i)) {
 959                 const struct bio_vec *bvec = i->bvec;
 960                 while (1) {
 961                         size_t n = (--bvec)->bv_len;
 962                         i->nr_segs++;
 963                         if (unroll <= n) {
 964                                 i->bvec = bvec;
 965                                 i->iov_offset = n - unroll;
 966                                 return;
 967                         }
 968                         unroll -= n;
 969                 }
 970         } else { /* same logics for iovec and kvec */
 971                 const struct iovec *iov = i->iov;
 972                 while (1) {
 973                         size_t n = (--iov)->iov_len;
 974                         i->nr_segs++;
 975                         if (unroll <= n) {
 976                                 i->iov = iov;
 977                                 i->iov_offset = n - unroll;
 978                                 return;
 979                         }
 980                         unroll -= n;
 981                 }
 982         }
 983 }
 984 EXPORT_SYMBOL(iov_iter_revert);
 985
 986 /*
 987  * Return the count of just the current iov_iter segment.
 988  */
 989 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 990 {
 991         if (i->nr_segs > 1) {
 992                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 993                         return min(i->count, i->iov->iov_len - i->iov_offset);
 994                 if (iov_iter_is_bvec(i))
 995                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 996         }
 997         return i->count;
 998 }
 999 EXPORT_SYMBOL(iov_iter_single_seg_count);
1000
1001 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1002                         const struct kvec *kvec, unsigned long nr_segs,
1003                         size_t count)
1004 {
1005         WARN_ON(direction & ~(READ | WRITE));
1006         *i = (struct iov_iter){
1007                 .iter_type = ITER_KVEC,
1008                 .data_source = direction,
1009                 .kvec = kvec,
1010                 .nr_segs = nr_segs,
1011                 .iov_offset = 0,
1012                 .count = count
1013         };
1014 }
1015 EXPORT_SYMBOL(iov_iter_kvec);
1016
1017 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1018                         const struct bio_vec *bvec, unsigned long nr_segs,
1019                         size_t count)
1020 {
1021         WARN_ON(direction & ~(READ | WRITE));
1022         *i = (struct iov_iter){
1023                 .iter_type = ITER_BVEC,
1024                 .data_source = direction,
1025                 .bvec = bvec,
1026                 .nr_segs = nr_segs,
1027                 .iov_offset = 0,
1028                 .count = count
1029         };
1030 }
1031 EXPORT_SYMBOL(iov_iter_bvec);
1032
1033 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1034                         struct pipe_inode_info *pipe,
1035                         size_t count)
1036 {
1037         BUG_ON(direction != READ);
1038         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1039         *i = (struct iov_iter){
1040                 .iter_type = ITER_PIPE,
1041                 .data_source = false,
1042                 .pipe = pipe,
1043                 .head = pipe->head,
1044                 .start_head = pipe->head,
1045                 .iov_offset = 0,
1046                 .count = count
1047         };
1048 }
1049 EXPORT_SYMBOL(iov_iter_pipe);
1050
1051 /**
1052  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1053  * @i: The iterator to initialise.
1054  * @direction: The direction of the transfer.
1055  * @xarray: The xarray to access.
1056  * @start: The start file position.
1057  * @count: The size of the I/O buffer in bytes.
1058  *
1059  * Set up an I/O iterator to either draw data out of the pages attached to an
1060  * inode or to inject data into those pages.  The pages *must* be prevented
1061  * from evaporation, either by taking a ref on them or locking them by the
1062  * caller.
1063  */
1064 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1065                      struct xarray *xarray, loff_t start, size_t count)
1066 {
1067         BUG_ON(direction & ~1);
1068         *i = (struct iov_iter) {
1069                 .iter_type = ITER_XARRAY,
1070                 .data_source = direction,
1071                 .xarray = xarray,
1072                 .xarray_start = start,
1073                 .count = count,
1074                 .iov_offset = 0
1075         };
1076 }
1077 EXPORT_SYMBOL(iov_iter_xarray);
1078
1079 /**
1080  * iov_iter_discard - Initialise an I/O iterator that discards data
1081  * @i: The iterator to initialise.
1082  * @direction: The direction of the transfer.
1083  * @count: The size of the I/O buffer in bytes.
1084  *
1085  * Set up an I/O iterator that just discards everything that's written to it.
1086  * It's only available as a READ iterator.
1087  */
1088 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1089 {
1090         BUG_ON(direction != READ);
1091         *i = (struct iov_iter){
1092                 .iter_type = ITER_DISCARD,
1093                 .data_source = false,
1094                 .count = count,
1095                 .iov_offset = 0
1096         };
1097 }
1098 EXPORT_SYMBOL(iov_iter_discard);
1099
1100 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1101                                    unsigned len_mask)
1102 {
1103         size_t size = i->count;
1104         size_t skip = i->iov_offset;
1105         unsigned k;
1106
1107         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1108                 size_t len = i->iov[k].iov_len - skip;
1109
1110                 if (len > size)
1111                         len = size;
1112                 if (len & len_mask)
1113                         return false;
1114                 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1115                         return false;
1116
1117                 size -= len;
1118                 if (!size)
1119                         break;
1120         }
1121         return true;
1122 }
1123
1124 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1125                                   unsigned len_mask)
1126 {
1127         size_t size = i->count;
1128         unsigned skip = i->iov_offset;
1129         unsigned k;
1130
1131         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1132                 size_t len = i->bvec[k].bv_len - skip;
1133
1134                 if (len > size)
1135                         len = size;
1136                 if (len & len_mask)
1137                         return false;
1138                 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1139                         return false;
1140
1141                 size -= len;
1142                 if (!size)
1143                         break;
1144         }
1145         return true;
1146 }
1147
1148 /**
1149  * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1150  *      are aligned to the parameters.
1151  *
1152  * @i: &struct iov_iter to restore
1153  * @addr_mask: bit mask to check against the iov element's addresses
1154  * @len_mask: bit mask to check against the iov element's lengths
1155  *
1156  * Return: false if any addresses or lengths intersect with the provided masks
1157  */
1158 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1159                          unsigned len_mask)
1160 {
1161         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1162                 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1163
1164         if (iov_iter_is_bvec(i))
1165                 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1166
1167         if (iov_iter_is_pipe(i)) {
1168                 unsigned int p_mask = i->pipe->ring_size - 1;
1169                 size_t size = i->count;
1170
1171                 if (size & len_mask)
1172                         return false;
1173                 if (size && allocated(&i->pipe->bufs[i->head & p_mask])) {
1174                         if (i->iov_offset & addr_mask)
1175                                 return false;
1176                 }
1177
1178                 return true;
1179         }
1180
1181         if (iov_iter_is_xarray(i)) {
1182                 if (i->count & len_mask)
1183                         return false;
1184                 if ((i->xarray_start + i->iov_offset) & addr_mask)
1185                         return false;
1186         }
1187
1188         return true;
1189 }
1190 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1191
1192 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1193 {
1194         unsigned long res = 0;
1195         size_t size = i->count;
1196         size_t skip = i->iov_offset;
1197         unsigned k;
1198
1199         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1200                 size_t len = i->iov[k].iov_len - skip;
1201                 if (len) {
1202                         res |= (unsigned long)i->iov[k].iov_base + skip;
1203                         if (len > size)
1204                                 len = size;
1205                         res |= len;
1206                         size -= len;
1207                         if (!size)
1208                                 break;
1209                 }
1210         }
1211         return res;
1212 }
1213
1214 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1215 {
1216         unsigned res = 0;
1217         size_t size = i->count;
1218         unsigned skip = i->iov_offset;
1219         unsigned k;
1220
1221         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1222                 size_t len = i->bvec[k].bv_len - skip;
1223                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1224                 if (len > size)
1225                         len = size;
1226                 res |= len;
1227                 size -= len;
1228                 if (!size)
1229                         break;
1230         }
1231         return res;
1232 }
1233
1234 unsigned long iov_iter_alignment(const struct iov_iter *i)
1235 {
1236         /* iovec and kvec have identical layouts */
1237         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1238                 return iov_iter_alignment_iovec(i);
1239
1240         if (iov_iter_is_bvec(i))
1241                 return iov_iter_alignment_bvec(i);
1242
1243         if (iov_iter_is_pipe(i)) {
1244                 unsigned int p_mask = i->pipe->ring_size - 1;
1245                 size_t size = i->count;
1246
1247                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1248                         return size | i->iov_offset;
1249                 return size;
1250         }
1251
1252         if (iov_iter_is_xarray(i))
1253                 return (i->xarray_start + i->iov_offset) | i->count;
1254
1255         return 0;
1256 }
1257 EXPORT_SYMBOL(iov_iter_alignment);
1258
1259 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1260 {
1261         unsigned long res = 0;
1262         unsigned long v = 0;
1263         size_t size = i->count;
1264         unsigned k;
1265
1266         if (WARN_ON(!iter_is_iovec(i)))
1267                 return ~0U;
1268
1269         for (k = 0; k < i->nr_segs; k++) {
1270                 if (i->iov[k].iov_len) {
1271                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1272                         if (v) // if not the first one
1273                                 res |= base | v; // this start | previous end
1274                         v = base + i->iov[k].iov_len;
1275                         if (size <= i->iov[k].iov_len)
1276                                 break;
1277                         size -= i->iov[k].iov_len;
1278                 }
1279         }
1280         return res;
1281 }
1282 EXPORT_SYMBOL(iov_iter_gap_alignment);
1283
1284 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1285                                 size_t maxsize,
1286                                 struct page **pages,
1287                                 int iter_head,
1288                                 size_t *start)
1289 {
1290         struct pipe_inode_info *pipe = i->pipe;
1291         unsigned int p_mask = pipe->ring_size - 1;
1292         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1293         if (!n)
1294                 return -EFAULT;
1295
1296         maxsize = n;
1297         n += *start;
1298         while (n > 0) {
1299                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1300                 iter_head++;
1301                 n -= PAGE_SIZE;
1302         }
1303
1304         return maxsize;
1305 }
1306
1307 static ssize_t pipe_get_pages(struct iov_iter *i,
1308                    struct page **pages, size_t maxsize, unsigned maxpages,
1309                    size_t *start)
1310 {
1311         unsigned int iter_head, npages;
1312         size_t capacity;
1313
1314         if (!sanity(i))
1315                 return -EFAULT;
1316
1317         data_start(i, &iter_head, start);
1318         /* Amount of free space: some of this one + all after this one */
1319         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1320         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1321
1322         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1323 }
1324
1325 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1326                                           pgoff_t index, unsigned int nr_pages)
1327 {
1328         XA_STATE(xas, xa, index);
1329         struct page *page;
1330         unsigned int ret = 0;
1331
1332         rcu_read_lock();
1333         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1334                 if (xas_retry(&xas, page))
1335                         continue;
1336
1337                 /* Has the page moved or been split? */
1338                 if (unlikely(page != xas_reload(&xas))) {
1339                         xas_reset(&xas);
1340                         continue;
1341                 }
1342
1343                 pages[ret] = find_subpage(page, xas.xa_index);
1344                 get_page(pages[ret]);
1345                 if (++ret == nr_pages)
1346                         break;
1347         }
1348         rcu_read_unlock();
1349         return ret;
1350 }
1351
1352 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1353                                      struct page **pages, size_t maxsize,
1354                                      unsigned maxpages, size_t *_start_offset)
1355 {
1356         unsigned nr, offset;
1357         pgoff_t index, count;
1358         size_t size = maxsize;
1359         loff_t pos;
1360
1361         if (!size || !maxpages)
1362                 return 0;
1363
1364         pos = i->xarray_start + i->iov_offset;
1365         index = pos >> PAGE_SHIFT;
1366         offset = pos & ~PAGE_MASK;
1367         *_start_offset = offset;
1368
1369         count = 1;
1370         if (size > PAGE_SIZE - offset) {
1371                 size -= PAGE_SIZE - offset;
1372                 count += size >> PAGE_SHIFT;
1373                 size &= ~PAGE_MASK;
1374                 if (size)
1375                         count++;
1376         }
1377
1378         if (count > maxpages)
1379                 count = maxpages;
1380
1381         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1382         if (nr == 0)
1383                 return 0;
1384
1385         return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1386 }
1387
1388 /* must be done on non-empty ITER_IOVEC one */
1389 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1390 {
1391         size_t skip;
1392         long k;
1393
1394         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1395                 size_t len = i->iov[k].iov_len - skip;
1396
1397                 if (unlikely(!len))
1398                         continue;
1399                 if (*size > len)
1400                         *size = len;
1401                 return (unsigned long)i->iov[k].iov_base + skip;
1402         }
1403         BUG(); // if it had been empty, we wouldn't get called
1404 }
1405
1406 /* must be done on non-empty ITER_BVEC one */
1407 static struct page *first_bvec_segment(const struct iov_iter *i,
1408                                        size_t *size, size_t *start)
1409 {
1410         struct page *page;
1411         size_t skip = i->iov_offset, len;
1412
1413         len = i->bvec->bv_len - skip;
1414         if (*size > len)
1415                 *size = len;
1416         skip += i->bvec->bv_offset;
1417         page = i->bvec->bv_page + skip / PAGE_SIZE;
1418         *start = skip % PAGE_SIZE;
1419         return page;
1420 }
1421
1422 ssize_t iov_iter_get_pages(struct iov_iter *i,
1423                    struct page **pages, size_t maxsize, unsigned maxpages,
1424                    size_t *start)
1425 {
1426         int n, res;
1427
1428         if (maxsize > i->count)
1429                 maxsize = i->count;
1430         if (!maxsize)
1431                 return 0;
1432         if (maxsize > MAX_RW_COUNT)
1433                 maxsize = MAX_RW_COUNT;
1434
1435         if (likely(iter_is_iovec(i))) {
1436                 unsigned int gup_flags = 0;
1437                 unsigned long addr;
1438
1439                 if (iov_iter_rw(i) != WRITE)
1440                         gup_flags |= FOLL_WRITE;
1441                 if (i->nofault)
1442                         gup_flags |= FOLL_NOFAULT;
1443
1444                 addr = first_iovec_segment(i, &maxsize);
1445                 *start = addr % PAGE_SIZE;
1446                 addr &= PAGE_MASK;
1447                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1448                 if (n > maxpages)
1449                         n = maxpages;
1450                 res = get_user_pages_fast(addr, n, gup_flags, pages);
1451                 if (unlikely(res <= 0))
1452                         return res;
1453                 return min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1454         }
1455         if (iov_iter_is_bvec(i)) {
1456                 struct page *page;
1457
1458                 page = first_bvec_segment(i, &maxsize, start);
1459                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1460                 if (n > maxpages)
1461                         n = maxpages;
1462                 for (int k = 0; k < n; k++)
1463                         get_page(*pages++ = page++);
1464                 return min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1465         }
1466         if (iov_iter_is_pipe(i))
1467                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1468         if (iov_iter_is_xarray(i))
1469                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1470         return -EFAULT;
1471 }
1472 EXPORT_SYMBOL(iov_iter_get_pages);
1473
1474 static struct page **get_pages_array(size_t n)
1475 {
1476         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1477 }
1478
1479 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1480                    struct page ***pages, size_t maxsize,
1481                    size_t *start)
1482 {
1483         struct page **p;
1484         unsigned int iter_head, npages;
1485         ssize_t n;
1486
1487         if (!sanity(i))
1488                 return -EFAULT;
1489
1490         data_start(i, &iter_head, start);
1491         /* Amount of free space: some of this one + all after this one */
1492         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1493         n = npages * PAGE_SIZE - *start;
1494         if (maxsize > n)
1495                 maxsize = n;
1496         else
1497                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1498         p = get_pages_array(npages);
1499         if (!p)
1500                 return -ENOMEM;
1501         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1502         if (n > 0)
1503                 *pages = p;
1504         else
1505                 kvfree(p);
1506         return n;
1507 }
1508
1509 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1510                                            struct page ***pages, size_t maxsize,
1511                                            size_t *_start_offset)
1512 {
1513         struct page **p;
1514         unsigned nr, offset;
1515         pgoff_t index, count;
1516         size_t size = maxsize;
1517         loff_t pos;
1518
1519         if (!size)
1520                 return 0;
1521
1522         pos = i->xarray_start + i->iov_offset;
1523         index = pos >> PAGE_SHIFT;
1524         offset = pos & ~PAGE_MASK;
1525         *_start_offset = offset;
1526
1527         count = 1;
1528         if (size > PAGE_SIZE - offset) {
1529                 size -= PAGE_SIZE - offset;
1530                 count += size >> PAGE_SHIFT;
1531                 size &= ~PAGE_MASK;
1532                 if (size)
1533                         count++;
1534         }
1535
1536         p = get_pages_array(count);
1537         if (!p)
1538                 return -ENOMEM;
1539         *pages = p;
1540
1541         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1542         if (nr == 0)
1543                 return 0;
1544
1545         return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1546 }
1547
1548 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1549                    struct page ***pages, size_t maxsize,
1550                    size_t *start)
1551 {
1552         struct page **p;
1553         int n, res;
1554
1555         if (maxsize > i->count)
1556                 maxsize = i->count;
1557         if (!maxsize)
1558                 return 0;
1559         if (maxsize > MAX_RW_COUNT)
1560                 maxsize = MAX_RW_COUNT;
1561
1562         if (likely(iter_is_iovec(i))) {
1563                 unsigned int gup_flags = 0;
1564                 unsigned long addr;
1565
1566                 if (iov_iter_rw(i) != WRITE)
1567                         gup_flags |= FOLL_WRITE;
1568                 if (i->nofault)
1569                         gup_flags |= FOLL_NOFAULT;
1570
1571                 addr = first_iovec_segment(i, &maxsize);
1572                 *start = addr % PAGE_SIZE;
1573                 addr &= PAGE_MASK;
1574                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1575                 p = get_pages_array(n);
1576                 if (!p)
1577                         return -ENOMEM;
1578                 res = get_user_pages_fast(addr, n, gup_flags, p);
1579                 if (unlikely(res <= 0)) {
1580                         kvfree(p);
1581                         *pages = NULL;
1582                         return res;
1583                 }
1584                 *pages = p;
1585                 return min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1586         }
1587         if (iov_iter_is_bvec(i)) {
1588                 struct page *page;
1589
1590                 page = first_bvec_segment(i, &maxsize, start);
1591                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1592                 *pages = p = get_pages_array(n);
1593                 if (!p)
1594                         return -ENOMEM;
1595                 for (int k = 0; k < n; k++)
1596                         get_page(*p++ = page++);
1597                 return min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1598         }
1599         if (iov_iter_is_pipe(i))
1600                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1601         if (iov_iter_is_xarray(i))
1602                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1603         return -EFAULT;
1604 }
1605 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1606
1607 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1608                                struct iov_iter *i)
1609 {
1610         __wsum sum, next;
1611         sum = *csum;
1612         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1613                 WARN_ON(1);
1614                 return 0;
1615         }
1616         iterate_and_advance(i, bytes, base, len, off, ({
1617                 next = csum_and_copy_from_user(base, addr + off, len);
1618                 sum = csum_block_add(sum, next, off);
1619                 next ? 0 : len;
1620         }), ({
1621                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1622         })
1623         )
1624         *csum = sum;
1625         return bytes;
1626 }
1627 EXPORT_SYMBOL(csum_and_copy_from_iter);
1628
1629 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1630                              struct iov_iter *i)
1631 {
1632         struct csum_state *csstate = _csstate;
1633         __wsum sum, next;
1634
1635         if (unlikely(iov_iter_is_discard(i))) {
1636                 WARN_ON(1);     /* for now */
1637                 return 0;
1638         }
1639
1640         sum = csum_shift(csstate->csum, csstate->off);
1641         if (unlikely(iov_iter_is_pipe(i)))
1642                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1643         else iterate_and_advance(i, bytes, base, len, off, ({
1644                 next = csum_and_copy_to_user(addr + off, base, len);
1645                 sum = csum_block_add(sum, next, off);
1646                 next ? 0 : len;
1647         }), ({
1648                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1649         })
1650         )
1651         csstate->csum = csum_shift(sum, csstate->off);
1652         csstate->off += bytes;
1653         return bytes;
1654 }
1655 EXPORT_SYMBOL(csum_and_copy_to_iter);
1656
1657 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1658                 struct iov_iter *i)
1659 {
1660 #ifdef CONFIG_CRYPTO_HASH
1661         struct ahash_request *hash = hashp;
1662         struct scatterlist sg;
1663         size_t copied;
1664
1665         copied = copy_to_iter(addr, bytes, i);
1666         sg_init_one(&sg, addr, copied);
1667         ahash_request_set_crypt(hash, &sg, NULL, copied);
1668         crypto_ahash_update(hash);
1669         return copied;
1670 #else
1671         return 0;
1672 #endif
1673 }
1674 EXPORT_SYMBOL(hash_and_copy_to_iter);
1675
1676 static int iov_npages(const struct iov_iter *i, int maxpages)
1677 {
1678         size_t skip = i->iov_offset, size = i->count;
1679         const struct iovec *p;
1680         int npages = 0;
1681
1682         for (p = i->iov; size; skip = 0, p++) {
1683                 unsigned offs = offset_in_page(p->iov_base + skip);
1684                 size_t len = min(p->iov_len - skip, size);
1685
1686                 if (len) {
1687                         size -= len;
1688                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1689                         if (unlikely(npages > maxpages))
1690                                 return maxpages;
1691                 }
1692         }
1693         return npages;
1694 }
1695
1696 static int bvec_npages(const struct iov_iter *i, int maxpages)
1697 {
1698         size_t skip = i->iov_offset, size = i->count;
1699         const struct bio_vec *p;
1700         int npages = 0;
1701
1702         for (p = i->bvec; size; skip = 0, p++) {
1703                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1704                 size_t len = min(p->bv_len - skip, size);
1705
1706                 size -= len;
1707                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1708                 if (unlikely(npages > maxpages))
1709                         return maxpages;
1710         }
1711         return npages;
1712 }
1713
1714 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1715 {
1716         if (unlikely(!i->count))
1717                 return 0;
1718         /* iovec and kvec have identical layouts */
1719         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1720                 return iov_npages(i, maxpages);
1721         if (iov_iter_is_bvec(i))
1722                 return bvec_npages(i, maxpages);
1723         if (iov_iter_is_pipe(i)) {
1724                 unsigned int iter_head;
1725                 int npages;
1726                 size_t off;
1727
1728                 if (!sanity(i))
1729                         return 0;
1730
1731                 data_start(i, &iter_head, &off);
1732                 /* some of this one + all after this one */
1733                 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1734                 return min(npages, maxpages);
1735         }
1736         if (iov_iter_is_xarray(i)) {
1737                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1738                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1739                 return min(npages, maxpages);
1740         }
1741         return 0;
1742 }
1743 EXPORT_SYMBOL(iov_iter_npages);
1744
1745 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1746 {
1747         *new = *old;
1748         if (unlikely(iov_iter_is_pipe(new))) {
1749                 WARN_ON(1);
1750                 return NULL;
1751         }
1752         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1753                 return NULL;
1754         if (iov_iter_is_bvec(new))
1755                 return new->bvec = kmemdup(new->bvec,
1756                                     new->nr_segs * sizeof(struct bio_vec),
1757                                     flags);
1758         else
1759                 /* iovec and kvec have identical layout */
1760                 return new->iov = kmemdup(new->iov,
1761                                    new->nr_segs * sizeof(struct iovec),
1762                                    flags);
1763 }
1764 EXPORT_SYMBOL(dup_iter);
1765
1766 static int copy_compat_iovec_from_user(struct iovec *iov,
1767                 const struct iovec __user *uvec, unsigned long nr_segs)
1768 {
1769         const struct compat_iovec __user *uiov =
1770                 (const struct compat_iovec __user *)uvec;
1771         int ret = -EFAULT, i;
1772
1773         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1774                 return -EFAULT;
1775
1776         for (i = 0; i < nr_segs; i++) {
1777                 compat_uptr_t buf;
1778                 compat_ssize_t len;
1779
1780                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1781                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1782
1783                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1784                 if (len < 0) {
1785                         ret = -EINVAL;
1786                         goto uaccess_end;
1787                 }
1788                 iov[i].iov_base = compat_ptr(buf);
1789                 iov[i].iov_len = len;
1790         }
1791
1792         ret = 0;
1793 uaccess_end:
1794         user_access_end();
1795         return ret;
1796 }
1797
1798 static int copy_iovec_from_user(struct iovec *iov,
1799                 const struct iovec __user *uvec, unsigned long nr_segs)
1800 {
1801         unsigned long seg;
1802
1803         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1804                 return -EFAULT;
1805         for (seg = 0; seg < nr_segs; seg++) {
1806                 if ((ssize_t)iov[seg].iov_len < 0)
1807                         return -EINVAL;
1808         }
1809
1810         return 0;
1811 }
1812
1813 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1814                 unsigned long nr_segs, unsigned long fast_segs,
1815                 struct iovec *fast_iov, bool compat)
1816 {
1817         struct iovec *iov = fast_iov;
1818         int ret;
1819
1820         /*
1821          * SuS says "The readv() function *may* fail if the iovcnt argument was
1822          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1823          * traditionally returned zero for zero segments, so...
1824          */
1825         if (nr_segs == 0)
1826                 return iov;
1827         if (nr_segs > UIO_MAXIOV)
1828                 return ERR_PTR(-EINVAL);
1829         if (nr_segs > fast_segs) {
1830                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1831                 if (!iov)
1832                         return ERR_PTR(-ENOMEM);
1833         }
1834
1835         if (compat)
1836                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1837         else
1838                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1839         if (ret) {
1840                 if (iov != fast_iov)
1841                         kfree(iov);
1842                 return ERR_PTR(ret);
1843         }
1844
1845         return iov;
1846 }
1847
1848 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1849                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1850                  struct iov_iter *i, bool compat)
1851 {
1852         ssize_t total_len = 0;
1853         unsigned long seg;
1854         struct iovec *iov;
1855
1856         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1857         if (IS_ERR(iov)) {
1858                 *iovp = NULL;
1859                 return PTR_ERR(iov);
1860         }
1861
1862         /*
1863          * According to the Single Unix Specification we should return EINVAL if
1864          * an element length is < 0 when cast to ssize_t or if the total length
1865          * would overflow the ssize_t return value of the system call.
1866          *
1867          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1868          * overflow case.
1869          */
1870         for (seg = 0; seg < nr_segs; seg++) {
1871                 ssize_t len = (ssize_t)iov[seg].iov_len;
1872
1873                 if (!access_ok(iov[seg].iov_base, len)) {
1874                         if (iov != *iovp)
1875                                 kfree(iov);
1876                         *iovp = NULL;
1877                         return -EFAULT;
1878                 }
1879
1880                 if (len > MAX_RW_COUNT - total_len) {
1881                         len = MAX_RW_COUNT - total_len;
1882                         iov[seg].iov_len = len;
1883                 }
1884                 total_len += len;
1885         }
1886
1887         iov_iter_init(i, type, iov, nr_segs, total_len);
1888         if (iov == *iovp)
1889                 *iovp = NULL;
1890         else
1891                 *iovp = iov;
1892         return total_len;
1893 }
1894
1895 /**
1896  * import_iovec() - Copy an array of &struct iovec from userspace
1897  *     into the kernel, check that it is valid, and initialize a new
1898  *     &struct iov_iter iterator to access it.
1899  *
1900  * @type: One of %READ or %WRITE.
1901  * @uvec: Pointer to the userspace array.
1902  * @nr_segs: Number of elements in userspace array.
1903  * @fast_segs: Number of elements in @iov.
1904  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1905  *     on-stack) kernel array.
1906  * @i: Pointer to iterator that will be initialized on success.
1907  *
1908  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1909  * then this function places %NULL in *@iov on return. Otherwise, a new
1910  * array will be allocated and the result placed in *@iov. This means that
1911  * the caller may call kfree() on *@iov regardless of whether the small
1912  * on-stack array was used or not (and regardless of whether this function
1913  * returns an error or not).
1914  *
1915  * Return: Negative error code on error, bytes imported on success
1916  */
1917 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1918                  unsigned nr_segs, unsigned fast_segs,
1919                  struct iovec **iovp, struct iov_iter *i)
1920 {
1921         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1922                               in_compat_syscall());
1923 }
1924 EXPORT_SYMBOL(import_iovec);
1925
1926 int import_single_range(int rw, void __user *buf, size_t len,
1927                  struct iovec *iov, struct iov_iter *i)
1928 {
1929         if (len > MAX_RW_COUNT)
1930                 len = MAX_RW_COUNT;
1931         if (unlikely(!access_ok(buf, len)))
1932                 return -EFAULT;
1933
1934         iov->iov_base = buf;
1935         iov->iov_len = len;
1936         iov_iter_init(i, rw, iov, 1, len);
1937         return 0;
1938 }
1939 EXPORT_SYMBOL(import_single_range);
1940
1941 /**
1942  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1943  *     iov_iter_save_state() was called.
1944  *
1945  * @i: &struct iov_iter to restore
1946  * @state: state to restore from
1947  *
1948  * Used after iov_iter_save_state() to bring restore @i, if operations may
1949  * have advanced it.
1950  *
1951  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1952  */
1953 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1954 {
1955         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
1956                          !iov_iter_is_kvec(i))
1957                 return;
1958         i->iov_offset = state->iov_offset;
1959         i->count = state->count;
1960         /*
1961          * For the *vec iters, nr_segs + iov is constant - if we increment
1962          * the vec, then we also decrement the nr_segs count. Hence we don't
1963          * need to track both of these, just one is enough and we can deduct
1964          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1965          * size, so we can just increment the iov pointer as they are unionzed.
1966          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1967          * not. Be safe and handle it separately.
1968          */
1969         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1970         if (iov_iter_is_bvec(i))
1971                 i->bvec -= state->nr_segs - i->nr_segs;
1972         else
1973                 i->iov -= state->nr_segs - i->nr_segs;
1974         i->nr_segs = state->nr_segs;
1975 }