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         size_t n, off, xfer = 0;
 525
 526         if (!sanity(i))
 527                 return 0;
 528
 529         n = push_pipe(i, bytes, &i_head, &off);
 530         while (n) {
 531                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 532                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 533                 unsigned long rem;
 534                 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
 535                 chunk -= rem;
 536                 kunmap_local(p);
 537                 i->head = i_head;
 538                 i->iov_offset = off + chunk;
 539                 xfer += chunk;
 540                 if (rem)
 541                         break;
 542                 n -= chunk;
 543                 off = 0;
 544                 i_head++;
 545         }
 546         i->count -= xfer;
 547         return xfer;
 548 }
 549
 550 /**
 551  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 552  * @addr: source kernel address
 553  * @bytes: total transfer length
 554  * @i: destination iterator
 555  *
 556  * The pmem driver deploys this for the dax operation
 557  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 558  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 559  * successfully copied.
 560  *
 561  * The main differences between this and typical _copy_to_iter().
 562  *
 563  * * Typical tail/residue handling after a fault retries the copy
 564  *   byte-by-byte until the fault happens again. Re-triggering machine
 565  *   checks is potentially fatal so the implementation uses source
 566  *   alignment and poison alignment assumptions to avoid re-triggering
 567  *   hardware exceptions.
 568  *
 569  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 570  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 571  *   a short copy.
 572  *
 573  * Return: number of bytes copied (may be %0)
 574  */
 575 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 576 {
 577         if (unlikely(iov_iter_is_pipe(i)))
 578                 return copy_mc_pipe_to_iter(addr, bytes, i);
 579         if (iter_is_iovec(i))
 580                 might_fault();
 581         __iterate_and_advance(i, bytes, base, len, off,
 582                 copyout_mc(base, addr + off, len),
 583                 copy_mc_to_kernel(base, addr + off, len)
 584         )
 585
 586         return bytes;
 587 }
 588 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 589 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 590
 591 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 592 {
 593         if (unlikely(iov_iter_is_pipe(i))) {
 594                 WARN_ON(1);
 595                 return 0;
 596         }
 597         if (iter_is_iovec(i))
 598                 might_fault();
 599         iterate_and_advance(i, bytes, base, len, off,
 600                 copyin(addr + off, base, len),
 601                 memcpy(addr + off, base, len)
 602         )
 603
 604         return bytes;
 605 }
 606 EXPORT_SYMBOL(_copy_from_iter);
 607
 608 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 609 {
 610         if (unlikely(iov_iter_is_pipe(i))) {
 611                 WARN_ON(1);
 612                 return 0;
 613         }
 614         iterate_and_advance(i, bytes, base, len, off,
 615                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 616                 memcpy(addr + off, base, len)
 617         )
 618
 619         return bytes;
 620 }
 621 EXPORT_SYMBOL(_copy_from_iter_nocache);
 622
 623 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 624 /**
 625  * _copy_from_iter_flushcache - write destination through cpu cache
 626  * @addr: destination kernel address
 627  * @bytes: total transfer length
 628  * @i: source iterator
 629  *
 630  * The pmem driver arranges for filesystem-dax to use this facility via
 631  * dax_copy_from_iter() for ensuring that writes to persistent memory
 632  * are flushed through the CPU cache. It is differentiated from
 633  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 634  * all iterator types. The _copy_from_iter_nocache() only attempts to
 635  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 636  * instructions that strand dirty-data in the cache.
 637  *
 638  * Return: number of bytes copied (may be %0)
 639  */
 640 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 641 {
 642         if (unlikely(iov_iter_is_pipe(i))) {
 643                 WARN_ON(1);
 644                 return 0;
 645         }
 646         iterate_and_advance(i, bytes, base, len, off,
 647                 __copy_from_user_flushcache(addr + off, base, len),
 648                 memcpy_flushcache(addr + off, base, len)
 649         )
 650
 651         return bytes;
 652 }
 653 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 654 #endif
 655
 656 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 657 {
 658         struct page *head;
 659         size_t v = n + offset;
 660
 661         /*
 662          * The general case needs to access the page order in order
 663          * to compute the page size.
 664          * However, we mostly deal with order-0 pages and thus can
 665          * avoid a possible cache line miss for requests that fit all
 666          * page orders.
 667          */
 668         if (n <= v && v <= PAGE_SIZE)
 669                 return true;
 670
 671         head = compound_head(page);
 672         v += (page - head) << PAGE_SHIFT;
 673
 674         if (likely(n <= v && v <= (page_size(head))))
 675                 return true;
 676         WARN_ON(1);
 677         return false;
 678 }
 679
 680 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 681                          struct iov_iter *i)
 682 {
 683         if (unlikely(iov_iter_is_pipe(i))) {
 684                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 685         } else {
 686                 void *kaddr = kmap_local_page(page);
 687                 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
 688                 kunmap_local(kaddr);
 689                 return wanted;
 690         }
 691 }
 692
 693 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 694                          struct iov_iter *i)
 695 {
 696         size_t res = 0;
 697         if (unlikely(!page_copy_sane(page, offset, bytes)))
 698                 return 0;
 699         page += offset / PAGE_SIZE; // first subpage
 700         offset %= PAGE_SIZE;
 701         while (1) {
 702                 size_t n = __copy_page_to_iter(page, offset,
 703                                 min(bytes, (size_t)PAGE_SIZE - offset), i);
 704                 res += n;
 705                 bytes -= n;
 706                 if (!bytes || !n)
 707                         break;
 708                 offset += n;
 709                 if (offset == PAGE_SIZE) {
 710                         page++;
 711                         offset = 0;
 712                 }
 713         }
 714         return res;
 715 }
 716 EXPORT_SYMBOL(copy_page_to_iter);
 717
 718 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 719                          struct iov_iter *i)
 720 {
 721         if (page_copy_sane(page, offset, bytes)) {
 722                 void *kaddr = kmap_local_page(page);
 723                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 724                 kunmap_local(kaddr);
 725                 return wanted;
 726         }
 727         return 0;
 728 }
 729 EXPORT_SYMBOL(copy_page_from_iter);
 730
 731 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 732 {
 733         struct pipe_inode_info *pipe = i->pipe;
 734         unsigned int p_mask = pipe->ring_size - 1;
 735         unsigned int i_head;
 736         size_t n, off;
 737
 738         if (!sanity(i))
 739                 return 0;
 740
 741         bytes = n = push_pipe(i, bytes, &i_head, &off);
 742         if (unlikely(!n))
 743                 return 0;
 744
 745         do {
 746                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 747                 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
 748                 memset(p + off, 0, chunk);
 749                 kunmap_local(p);
 750                 i->head = i_head;
 751                 i->iov_offset = off + chunk;
 752                 n -= chunk;
 753                 off = 0;
 754                 i_head++;
 755         } while (n);
 756         i->count -= bytes;
 757         return bytes;
 758 }
 759
 760 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 761 {
 762         if (unlikely(iov_iter_is_pipe(i)))
 763                 return pipe_zero(bytes, i);
 764         iterate_and_advance(i, bytes, base, len, count,
 765                 clear_user(base, len),
 766                 memset(base, 0, len)
 767         )
 768
 769         return bytes;
 770 }
 771 EXPORT_SYMBOL(iov_iter_zero);
 772
 773 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 774                                   struct iov_iter *i)
 775 {
 776         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 777         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 778                 kunmap_atomic(kaddr);
 779                 return 0;
 780         }
 781         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 782                 kunmap_atomic(kaddr);
 783                 WARN_ON(1);
 784                 return 0;
 785         }
 786         iterate_and_advance(i, bytes, base, len, off,
 787                 copyin(p + off, base, len),
 788                 memcpy(p + off, base, len)
 789         )
 790         kunmap_atomic(kaddr);
 791         return bytes;
 792 }
 793 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 794
 795 static inline void pipe_truncate(struct iov_iter *i)
 796 {
 797         struct pipe_inode_info *pipe = i->pipe;
 798         unsigned int p_tail = pipe->tail;
 799         unsigned int p_head = pipe->head;
 800         unsigned int p_mask = pipe->ring_size - 1;
 801
 802         if (!pipe_empty(p_head, p_tail)) {
 803                 struct pipe_buffer *buf;
 804                 unsigned int i_head = i->head;
 805                 size_t off = i->iov_offset;
 806
 807                 if (off) {
 808                         buf = &pipe->bufs[i_head & p_mask];
 809                         buf->len = off - buf->offset;
 810                         i_head++;
 811                 }
 812                 while (p_head != i_head) {
 813                         p_head--;
 814                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
 815                 }
 816
 817                 pipe->head = p_head;
 818         }
 819 }
 820
 821 static void pipe_advance(struct iov_iter *i, size_t size)
 822 {
 823         struct pipe_inode_info *pipe = i->pipe;
 824         if (size) {
 825                 struct pipe_buffer *buf;
 826                 unsigned int p_mask = pipe->ring_size - 1;
 827                 unsigned int i_head = i->head;
 828                 size_t off = i->iov_offset, left = size;
 829
 830                 if (off) /* make it relative to the beginning of buffer */
 831                         left += off - pipe->bufs[i_head & p_mask].offset;
 832                 while (1) {
 833                         buf = &pipe->bufs[i_head & p_mask];
 834                         if (left <= buf->len)
 835                                 break;
 836                         left -= buf->len;
 837                         i_head++;
 838                 }
 839                 i->head = i_head;
 840                 i->iov_offset = buf->offset + left;
 841         }
 842         i->count -= size;
 843         /* ... and discard everything past that point */
 844         pipe_truncate(i);
 845 }
 846
 847 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 848 {
 849         const struct bio_vec *bvec, *end;
 850
 851         if (!i->count)
 852                 return;
 853         i->count -= size;
 854
 855         size += i->iov_offset;
 856
 857         for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
 858                 if (likely(size < bvec->bv_len))
 859                         break;
 860                 size -= bvec->bv_len;
 861         }
 862         i->iov_offset = size;
 863         i->nr_segs -= bvec - i->bvec;
 864         i->bvec = bvec;
 865 }
 866
 867 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 868 {
 869         const struct iovec *iov, *end;
 870
 871         if (!i->count)
 872                 return;
 873         i->count -= size;
 874
 875         size += i->iov_offset; // from beginning of current segment
 876         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
 877                 if (likely(size < iov->iov_len))
 878                         break;
 879                 size -= iov->iov_len;
 880         }
 881         i->iov_offset = size;
 882         i->nr_segs -= iov - i->iov;
 883         i->iov = iov;
 884 }
 885
 886 void iov_iter_advance(struct iov_iter *i, size_t size)
 887 {
 888         if (unlikely(i->count < size))
 889                 size = i->count;
 890         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 891                 /* iovec and kvec have identical layouts */
 892                 iov_iter_iovec_advance(i, size);
 893         } else if (iov_iter_is_bvec(i)) {
 894                 iov_iter_bvec_advance(i, size);
 895         } else if (iov_iter_is_pipe(i)) {
 896                 pipe_advance(i, size);
 897         } else if (unlikely(iov_iter_is_xarray(i))) {
 898                 i->iov_offset += size;
 899                 i->count -= size;
 900         } else if (iov_iter_is_discard(i)) {
 901                 i->count -= size;
 902         }
 903 }
 904 EXPORT_SYMBOL(iov_iter_advance);
 905
 906 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 907 {
 908         if (!unroll)
 909                 return;
 910         if (WARN_ON(unroll > MAX_RW_COUNT))
 911                 return;
 912         i->count += unroll;
 913         if (unlikely(iov_iter_is_pipe(i))) {
 914                 struct pipe_inode_info *pipe = i->pipe;
 915                 unsigned int p_mask = pipe->ring_size - 1;
 916                 unsigned int i_head = i->head;
 917                 size_t off = i->iov_offset;
 918                 while (1) {
 919                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
 920                         size_t n = off - b->offset;
 921                         if (unroll < n) {
 922                                 off -= unroll;
 923                                 break;
 924                         }
 925                         unroll -= n;
 926                         if (!unroll && i_head == i->start_head) {
 927                                 off = 0;
 928                                 break;
 929                         }
 930                         i_head--;
 931                         b = &pipe->bufs[i_head & p_mask];
 932                         off = b->offset + b->len;
 933                 }
 934                 i->iov_offset = off;
 935                 i->head = i_head;
 936                 pipe_truncate(i);
 937                 return;
 938         }
 939         if (unlikely(iov_iter_is_discard(i)))
 940                 return;
 941         if (unroll <= i->iov_offset) {
 942                 i->iov_offset -= unroll;
 943                 return;
 944         }
 945         unroll -= i->iov_offset;
 946         if (iov_iter_is_xarray(i)) {
 947                 BUG(); /* We should never go beyond the start of the specified
 948                         * range since we might then be straying into pages that
 949                         * aren't pinned.
 950                         */
 951         } else if (iov_iter_is_bvec(i)) {
 952                 const struct bio_vec *bvec = i->bvec;
 953                 while (1) {
 954                         size_t n = (--bvec)->bv_len;
 955                         i->nr_segs++;
 956                         if (unroll <= n) {
 957                                 i->bvec = bvec;
 958                                 i->iov_offset = n - unroll;
 959                                 return;
 960                         }
 961                         unroll -= n;
 962                 }
 963         } else { /* same logics for iovec and kvec */
 964                 const struct iovec *iov = i->iov;
 965                 while (1) {
 966                         size_t n = (--iov)->iov_len;
 967                         i->nr_segs++;
 968                         if (unroll <= n) {
 969                                 i->iov = iov;
 970                                 i->iov_offset = n - unroll;
 971                                 return;
 972                         }
 973                         unroll -= n;
 974                 }
 975         }
 976 }
 977 EXPORT_SYMBOL(iov_iter_revert);
 978
 979 /*
 980  * Return the count of just the current iov_iter segment.
 981  */
 982 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 983 {
 984         if (i->nr_segs > 1) {
 985                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 986                         return min(i->count, i->iov->iov_len - i->iov_offset);
 987                 if (iov_iter_is_bvec(i))
 988                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 989         }
 990         return i->count;
 991 }
 992 EXPORT_SYMBOL(iov_iter_single_seg_count);
 993
 994 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
 995                         const struct kvec *kvec, unsigned long nr_segs,
 996                         size_t count)
 997 {
 998         WARN_ON(direction & ~(READ | WRITE));
 999         *i = (struct iov_iter){
1000                 .iter_type = ITER_KVEC,
1001                 .data_source = direction,
1002                 .kvec = kvec,
1003                 .nr_segs = nr_segs,
1004                 .iov_offset = 0,
1005                 .count = count
1006         };
1007 }
1008 EXPORT_SYMBOL(iov_iter_kvec);
1009
1010 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1011                         const struct bio_vec *bvec, unsigned long nr_segs,
1012                         size_t count)
1013 {
1014         WARN_ON(direction & ~(READ | WRITE));
1015         *i = (struct iov_iter){
1016                 .iter_type = ITER_BVEC,
1017                 .data_source = direction,
1018                 .bvec = bvec,
1019                 .nr_segs = nr_segs,
1020                 .iov_offset = 0,
1021                 .count = count
1022         };
1023 }
1024 EXPORT_SYMBOL(iov_iter_bvec);
1025
1026 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1027                         struct pipe_inode_info *pipe,
1028                         size_t count)
1029 {
1030         BUG_ON(direction != READ);
1031         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1032         *i = (struct iov_iter){
1033                 .iter_type = ITER_PIPE,
1034                 .data_source = false,
1035                 .pipe = pipe,
1036                 .head = pipe->head,
1037                 .start_head = pipe->head,
1038                 .iov_offset = 0,
1039                 .count = count
1040         };
1041 }
1042 EXPORT_SYMBOL(iov_iter_pipe);
1043
1044 /**
1045  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1046  * @i: The iterator to initialise.
1047  * @direction: The direction of the transfer.
1048  * @xarray: The xarray to access.
1049  * @start: The start file position.
1050  * @count: The size of the I/O buffer in bytes.
1051  *
1052  * Set up an I/O iterator to either draw data out of the pages attached to an
1053  * inode or to inject data into those pages.  The pages *must* be prevented
1054  * from evaporation, either by taking a ref on them or locking them by the
1055  * caller.
1056  */
1057 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1058                      struct xarray *xarray, loff_t start, size_t count)
1059 {
1060         BUG_ON(direction & ~1);
1061         *i = (struct iov_iter) {
1062                 .iter_type = ITER_XARRAY,
1063                 .data_source = direction,
1064                 .xarray = xarray,
1065                 .xarray_start = start,
1066                 .count = count,
1067                 .iov_offset = 0
1068         };
1069 }
1070 EXPORT_SYMBOL(iov_iter_xarray);
1071
1072 /**
1073  * iov_iter_discard - Initialise an I/O iterator that discards data
1074  * @i: The iterator to initialise.
1075  * @direction: The direction of the transfer.
1076  * @count: The size of the I/O buffer in bytes.
1077  *
1078  * Set up an I/O iterator that just discards everything that's written to it.
1079  * It's only available as a READ iterator.
1080  */
1081 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1082 {
1083         BUG_ON(direction != READ);
1084         *i = (struct iov_iter){
1085                 .iter_type = ITER_DISCARD,
1086                 .data_source = false,
1087                 .count = count,
1088                 .iov_offset = 0
1089         };
1090 }
1091 EXPORT_SYMBOL(iov_iter_discard);
1092
1093 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1094                                    unsigned len_mask)
1095 {
1096         size_t size = i->count;
1097         size_t skip = i->iov_offset;
1098         unsigned k;
1099
1100         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1101                 size_t len = i->iov[k].iov_len - skip;
1102
1103                 if (len > size)
1104                         len = size;
1105                 if (len & len_mask)
1106                         return false;
1107                 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1108                         return false;
1109
1110                 size -= len;
1111                 if (!size)
1112                         break;
1113         }
1114         return true;
1115 }
1116
1117 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1118                                   unsigned len_mask)
1119 {
1120         size_t size = i->count;
1121         unsigned skip = i->iov_offset;
1122         unsigned k;
1123
1124         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1125                 size_t len = i->bvec[k].bv_len - skip;
1126
1127                 if (len > size)
1128                         len = size;
1129                 if (len & len_mask)
1130                         return false;
1131                 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1132                         return false;
1133
1134                 size -= len;
1135                 if (!size)
1136                         break;
1137         }
1138         return true;
1139 }
1140
1141 /**
1142  * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1143  *      are aligned to the parameters.
1144  *
1145  * @i: &struct iov_iter to restore
1146  * @addr_mask: bit mask to check against the iov element's addresses
1147  * @len_mask: bit mask to check against the iov element's lengths
1148  *
1149  * Return: false if any addresses or lengths intersect with the provided masks
1150  */
1151 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1152                          unsigned len_mask)
1153 {
1154         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1155                 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1156
1157         if (iov_iter_is_bvec(i))
1158                 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1159
1160         if (iov_iter_is_pipe(i)) {
1161                 unsigned int p_mask = i->pipe->ring_size - 1;
1162                 size_t size = i->count;
1163
1164                 if (size & len_mask)
1165                         return false;
1166                 if (size && allocated(&i->pipe->bufs[i->head & p_mask])) {
1167                         if (i->iov_offset & addr_mask)
1168                                 return false;
1169                 }
1170
1171                 return true;
1172         }
1173
1174         if (iov_iter_is_xarray(i)) {
1175                 if (i->count & len_mask)
1176                         return false;
1177                 if ((i->xarray_start + i->iov_offset) & addr_mask)
1178                         return false;
1179         }
1180
1181         return true;
1182 }
1183 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1184
1185 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1186 {
1187         unsigned long res = 0;
1188         size_t size = i->count;
1189         size_t skip = i->iov_offset;
1190         unsigned k;
1191
1192         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1193                 size_t len = i->iov[k].iov_len - skip;
1194                 if (len) {
1195                         res |= (unsigned long)i->iov[k].iov_base + skip;
1196                         if (len > size)
1197                                 len = size;
1198                         res |= len;
1199                         size -= len;
1200                         if (!size)
1201                                 break;
1202                 }
1203         }
1204         return res;
1205 }
1206
1207 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1208 {
1209         unsigned res = 0;
1210         size_t size = i->count;
1211         unsigned skip = i->iov_offset;
1212         unsigned k;
1213
1214         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1215                 size_t len = i->bvec[k].bv_len - skip;
1216                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1217                 if (len > size)
1218                         len = size;
1219                 res |= len;
1220                 size -= len;
1221                 if (!size)
1222                         break;
1223         }
1224         return res;
1225 }
1226
1227 unsigned long iov_iter_alignment(const struct iov_iter *i)
1228 {
1229         /* iovec and kvec have identical layouts */
1230         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1231                 return iov_iter_alignment_iovec(i);
1232
1233         if (iov_iter_is_bvec(i))
1234                 return iov_iter_alignment_bvec(i);
1235
1236         if (iov_iter_is_pipe(i)) {
1237                 unsigned int p_mask = i->pipe->ring_size - 1;
1238                 size_t size = i->count;
1239
1240                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1241                         return size | i->iov_offset;
1242                 return size;
1243         }
1244
1245         if (iov_iter_is_xarray(i))
1246                 return (i->xarray_start + i->iov_offset) | i->count;
1247
1248         return 0;
1249 }
1250 EXPORT_SYMBOL(iov_iter_alignment);
1251
1252 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1253 {
1254         unsigned long res = 0;
1255         unsigned long v = 0;
1256         size_t size = i->count;
1257         unsigned k;
1258
1259         if (WARN_ON(!iter_is_iovec(i)))
1260                 return ~0U;
1261
1262         for (k = 0; k < i->nr_segs; k++) {
1263                 if (i->iov[k].iov_len) {
1264                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1265                         if (v) // if not the first one
1266                                 res |= base | v; // this start | previous end
1267                         v = base + i->iov[k].iov_len;
1268                         if (size <= i->iov[k].iov_len)
1269                                 break;
1270                         size -= i->iov[k].iov_len;
1271                 }
1272         }
1273         return res;
1274 }
1275 EXPORT_SYMBOL(iov_iter_gap_alignment);
1276
1277 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1278                                 size_t maxsize,
1279                                 struct page **pages,
1280                                 int iter_head,
1281                                 size_t *start)
1282 {
1283         struct pipe_inode_info *pipe = i->pipe;
1284         unsigned int p_mask = pipe->ring_size - 1;
1285         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1286         if (!n)
1287                 return -EFAULT;
1288
1289         maxsize = n;
1290         n += *start;
1291         while (n > 0) {
1292                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1293                 iter_head++;
1294                 n -= PAGE_SIZE;
1295         }
1296
1297         return maxsize;
1298 }
1299
1300 static ssize_t pipe_get_pages(struct iov_iter *i,
1301                    struct page **pages, size_t maxsize, unsigned maxpages,
1302                    size_t *start)
1303 {
1304         unsigned int iter_head, npages;
1305         size_t capacity;
1306
1307         if (!sanity(i))
1308                 return -EFAULT;
1309
1310         data_start(i, &iter_head, start);
1311         /* Amount of free space: some of this one + all after this one */
1312         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1313         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1314
1315         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1316 }
1317
1318 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1319                                           pgoff_t index, unsigned int nr_pages)
1320 {
1321         XA_STATE(xas, xa, index);
1322         struct page *page;
1323         unsigned int ret = 0;
1324
1325         rcu_read_lock();
1326         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1327                 if (xas_retry(&xas, page))
1328                         continue;
1329
1330                 /* Has the page moved or been split? */
1331                 if (unlikely(page != xas_reload(&xas))) {
1332                         xas_reset(&xas);
1333                         continue;
1334                 }
1335
1336                 pages[ret] = find_subpage(page, xas.xa_index);
1337                 get_page(pages[ret]);
1338                 if (++ret == nr_pages)
1339                         break;
1340         }
1341         rcu_read_unlock();
1342         return ret;
1343 }
1344
1345 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1346                                      struct page **pages, size_t maxsize,
1347                                      unsigned maxpages, size_t *_start_offset)
1348 {
1349         unsigned nr, offset;
1350         pgoff_t index, count;
1351         size_t size = maxsize;
1352         loff_t pos;
1353
1354         if (!size || !maxpages)
1355                 return 0;
1356
1357         pos = i->xarray_start + i->iov_offset;
1358         index = pos >> PAGE_SHIFT;
1359         offset = pos & ~PAGE_MASK;
1360         *_start_offset = offset;
1361
1362         count = 1;
1363         if (size > PAGE_SIZE - offset) {
1364                 size -= PAGE_SIZE - offset;
1365                 count += size >> PAGE_SHIFT;
1366                 size &= ~PAGE_MASK;
1367                 if (size)
1368                         count++;
1369         }
1370
1371         if (count > maxpages)
1372                 count = maxpages;
1373
1374         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1375         if (nr == 0)
1376                 return 0;
1377
1378         return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1379 }
1380
1381 /* must be done on non-empty ITER_IOVEC one */
1382 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1383 {
1384         size_t skip;
1385         long k;
1386
1387         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1388                 size_t len = i->iov[k].iov_len - skip;
1389
1390                 if (unlikely(!len))
1391                         continue;
1392                 if (*size > len)
1393                         *size = len;
1394                 return (unsigned long)i->iov[k].iov_base + skip;
1395         }
1396         BUG(); // if it had been empty, we wouldn't get called
1397 }
1398
1399 /* must be done on non-empty ITER_BVEC one */
1400 static struct page *first_bvec_segment(const struct iov_iter *i,
1401                                        size_t *size, size_t *start)
1402 {
1403         struct page *page;
1404         size_t skip = i->iov_offset, len;
1405
1406         len = i->bvec->bv_len - skip;
1407         if (*size > len)
1408                 *size = len;
1409         skip += i->bvec->bv_offset;
1410         page = i->bvec->bv_page + skip / PAGE_SIZE;
1411         *start = skip % PAGE_SIZE;
1412         return page;
1413 }
1414
1415 ssize_t iov_iter_get_pages(struct iov_iter *i,
1416                    struct page **pages, size_t maxsize, unsigned maxpages,
1417                    size_t *start)
1418 {
1419         int n, res;
1420
1421         if (maxsize > i->count)
1422                 maxsize = i->count;
1423         if (!maxsize)
1424                 return 0;
1425         if (maxsize > MAX_RW_COUNT)
1426                 maxsize = MAX_RW_COUNT;
1427
1428         if (likely(iter_is_iovec(i))) {
1429                 unsigned int gup_flags = 0;
1430                 unsigned long addr;
1431
1432                 if (iov_iter_rw(i) != WRITE)
1433                         gup_flags |= FOLL_WRITE;
1434                 if (i->nofault)
1435                         gup_flags |= FOLL_NOFAULT;
1436
1437                 addr = first_iovec_segment(i, &maxsize);
1438                 *start = addr % PAGE_SIZE;
1439                 addr &= PAGE_MASK;
1440                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1441                 if (n > maxpages)
1442                         n = maxpages;
1443                 res = get_user_pages_fast(addr, n, gup_flags, pages);
1444                 if (unlikely(res <= 0))
1445                         return res;
1446                 return min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1447         }
1448         if (iov_iter_is_bvec(i)) {
1449                 struct page *page;
1450
1451                 page = first_bvec_segment(i, &maxsize, start);
1452                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1453                 if (n > maxpages)
1454                         n = maxpages;
1455                 for (int k = 0; k < n; k++)
1456                         get_page(*pages++ = page++);
1457                 return min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1458         }
1459         if (iov_iter_is_pipe(i))
1460                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1461         if (iov_iter_is_xarray(i))
1462                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1463         return -EFAULT;
1464 }
1465 EXPORT_SYMBOL(iov_iter_get_pages);
1466
1467 static struct page **get_pages_array(size_t n)
1468 {
1469         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1470 }
1471
1472 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1473                    struct page ***pages, size_t maxsize,
1474                    size_t *start)
1475 {
1476         struct page **p;
1477         unsigned int iter_head, npages;
1478         ssize_t n;
1479
1480         if (!sanity(i))
1481                 return -EFAULT;
1482
1483         data_start(i, &iter_head, start);
1484         /* Amount of free space: some of this one + all after this one */
1485         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1486         n = npages * PAGE_SIZE - *start;
1487         if (maxsize > n)
1488                 maxsize = n;
1489         else
1490                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1491         p = get_pages_array(npages);
1492         if (!p)
1493                 return -ENOMEM;
1494         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1495         if (n > 0)
1496                 *pages = p;
1497         else
1498                 kvfree(p);
1499         return n;
1500 }
1501
1502 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1503                                            struct page ***pages, size_t maxsize,
1504                                            size_t *_start_offset)
1505 {
1506         struct page **p;
1507         unsigned nr, offset;
1508         pgoff_t index, count;
1509         size_t size = maxsize;
1510         loff_t pos;
1511
1512         if (!size)
1513                 return 0;
1514
1515         pos = i->xarray_start + i->iov_offset;
1516         index = pos >> PAGE_SHIFT;
1517         offset = pos & ~PAGE_MASK;
1518         *_start_offset = offset;
1519
1520         count = 1;
1521         if (size > PAGE_SIZE - offset) {
1522                 size -= PAGE_SIZE - offset;
1523                 count += size >> PAGE_SHIFT;
1524                 size &= ~PAGE_MASK;
1525                 if (size)
1526                         count++;
1527         }
1528
1529         p = get_pages_array(count);
1530         if (!p)
1531                 return -ENOMEM;
1532         *pages = p;
1533
1534         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1535         if (nr == 0)
1536                 return 0;
1537
1538         return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1539 }
1540
1541 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1542                    struct page ***pages, size_t maxsize,
1543                    size_t *start)
1544 {
1545         struct page **p;
1546         int n, res;
1547
1548         if (maxsize > i->count)
1549                 maxsize = i->count;
1550         if (!maxsize)
1551                 return 0;
1552         if (maxsize > MAX_RW_COUNT)
1553                 maxsize = MAX_RW_COUNT;
1554
1555         if (likely(iter_is_iovec(i))) {
1556                 unsigned int gup_flags = 0;
1557                 unsigned long addr;
1558
1559                 if (iov_iter_rw(i) != WRITE)
1560                         gup_flags |= FOLL_WRITE;
1561                 if (i->nofault)
1562                         gup_flags |= FOLL_NOFAULT;
1563
1564                 addr = first_iovec_segment(i, &maxsize);
1565                 *start = addr % PAGE_SIZE;
1566                 addr &= PAGE_MASK;
1567                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1568                 p = get_pages_array(n);
1569                 if (!p)
1570                         return -ENOMEM;
1571                 res = get_user_pages_fast(addr, n, gup_flags, p);
1572                 if (unlikely(res <= 0)) {
1573                         kvfree(p);
1574                         *pages = NULL;
1575                         return res;
1576                 }
1577                 *pages = p;
1578                 return min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1579         }
1580         if (iov_iter_is_bvec(i)) {
1581                 struct page *page;
1582
1583                 page = first_bvec_segment(i, &maxsize, start);
1584                 n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1585                 *pages = p = get_pages_array(n);
1586                 if (!p)
1587                         return -ENOMEM;
1588                 for (int k = 0; k < n; k++)
1589                         get_page(*p++ = page++);
1590                 return min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1591         }
1592         if (iov_iter_is_pipe(i))
1593                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1594         if (iov_iter_is_xarray(i))
1595                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1596         return -EFAULT;
1597 }
1598 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1599
1600 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1601                                struct iov_iter *i)
1602 {
1603         __wsum sum, next;
1604         sum = *csum;
1605         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1606                 WARN_ON(1);
1607                 return 0;
1608         }
1609         iterate_and_advance(i, bytes, base, len, off, ({
1610                 next = csum_and_copy_from_user(base, addr + off, len);
1611                 sum = csum_block_add(sum, next, off);
1612                 next ? 0 : len;
1613         }), ({
1614                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1615         })
1616         )
1617         *csum = sum;
1618         return bytes;
1619 }
1620 EXPORT_SYMBOL(csum_and_copy_from_iter);
1621
1622 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1623                              struct iov_iter *i)
1624 {
1625         struct csum_state *csstate = _csstate;
1626         __wsum sum, next;
1627
1628         if (unlikely(iov_iter_is_discard(i))) {
1629                 WARN_ON(1);     /* for now */
1630                 return 0;
1631         }
1632
1633         sum = csum_shift(csstate->csum, csstate->off);
1634         if (unlikely(iov_iter_is_pipe(i)))
1635                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1636         else iterate_and_advance(i, bytes, base, len, off, ({
1637                 next = csum_and_copy_to_user(addr + off, base, len);
1638                 sum = csum_block_add(sum, next, off);
1639                 next ? 0 : len;
1640         }), ({
1641                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1642         })
1643         )
1644         csstate->csum = csum_shift(sum, csstate->off);
1645         csstate->off += bytes;
1646         return bytes;
1647 }
1648 EXPORT_SYMBOL(csum_and_copy_to_iter);
1649
1650 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1651                 struct iov_iter *i)
1652 {
1653 #ifdef CONFIG_CRYPTO_HASH
1654         struct ahash_request *hash = hashp;
1655         struct scatterlist sg;
1656         size_t copied;
1657
1658         copied = copy_to_iter(addr, bytes, i);
1659         sg_init_one(&sg, addr, copied);
1660         ahash_request_set_crypt(hash, &sg, NULL, copied);
1661         crypto_ahash_update(hash);
1662         return copied;
1663 #else
1664         return 0;
1665 #endif
1666 }
1667 EXPORT_SYMBOL(hash_and_copy_to_iter);
1668
1669 static int iov_npages(const struct iov_iter *i, int maxpages)
1670 {
1671         size_t skip = i->iov_offset, size = i->count;
1672         const struct iovec *p;
1673         int npages = 0;
1674
1675         for (p = i->iov; size; skip = 0, p++) {
1676                 unsigned offs = offset_in_page(p->iov_base + skip);
1677                 size_t len = min(p->iov_len - skip, size);
1678
1679                 if (len) {
1680                         size -= len;
1681                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1682                         if (unlikely(npages > maxpages))
1683                                 return maxpages;
1684                 }
1685         }
1686         return npages;
1687 }
1688
1689 static int bvec_npages(const struct iov_iter *i, int maxpages)
1690 {
1691         size_t skip = i->iov_offset, size = i->count;
1692         const struct bio_vec *p;
1693         int npages = 0;
1694
1695         for (p = i->bvec; size; skip = 0, p++) {
1696                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1697                 size_t len = min(p->bv_len - skip, size);
1698
1699                 size -= len;
1700                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1701                 if (unlikely(npages > maxpages))
1702                         return maxpages;
1703         }
1704         return npages;
1705 }
1706
1707 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1708 {
1709         if (unlikely(!i->count))
1710                 return 0;
1711         /* iovec and kvec have identical layouts */
1712         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1713                 return iov_npages(i, maxpages);
1714         if (iov_iter_is_bvec(i))
1715                 return bvec_npages(i, maxpages);
1716         if (iov_iter_is_pipe(i)) {
1717                 unsigned int iter_head;
1718                 int npages;
1719                 size_t off;
1720
1721                 if (!sanity(i))
1722                         return 0;
1723
1724                 data_start(i, &iter_head, &off);
1725                 /* some of this one + all after this one */
1726                 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1727                 return min(npages, maxpages);
1728         }
1729         if (iov_iter_is_xarray(i)) {
1730                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1731                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1732                 return min(npages, maxpages);
1733         }
1734         return 0;
1735 }
1736 EXPORT_SYMBOL(iov_iter_npages);
1737
1738 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1739 {
1740         *new = *old;
1741         if (unlikely(iov_iter_is_pipe(new))) {
1742                 WARN_ON(1);
1743                 return NULL;
1744         }
1745         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1746                 return NULL;
1747         if (iov_iter_is_bvec(new))
1748                 return new->bvec = kmemdup(new->bvec,
1749                                     new->nr_segs * sizeof(struct bio_vec),
1750                                     flags);
1751         else
1752                 /* iovec and kvec have identical layout */
1753                 return new->iov = kmemdup(new->iov,
1754                                    new->nr_segs * sizeof(struct iovec),
1755                                    flags);
1756 }
1757 EXPORT_SYMBOL(dup_iter);
1758
1759 static int copy_compat_iovec_from_user(struct iovec *iov,
1760                 const struct iovec __user *uvec, unsigned long nr_segs)
1761 {
1762         const struct compat_iovec __user *uiov =
1763                 (const struct compat_iovec __user *)uvec;
1764         int ret = -EFAULT, i;
1765
1766         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1767                 return -EFAULT;
1768
1769         for (i = 0; i < nr_segs; i++) {
1770                 compat_uptr_t buf;
1771                 compat_ssize_t len;
1772
1773                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1774                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1775
1776                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1777                 if (len < 0) {
1778                         ret = -EINVAL;
1779                         goto uaccess_end;
1780                 }
1781                 iov[i].iov_base = compat_ptr(buf);
1782                 iov[i].iov_len = len;
1783         }
1784
1785         ret = 0;
1786 uaccess_end:
1787         user_access_end();
1788         return ret;
1789 }
1790
1791 static int copy_iovec_from_user(struct iovec *iov,
1792                 const struct iovec __user *uvec, unsigned long nr_segs)
1793 {
1794         unsigned long seg;
1795
1796         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1797                 return -EFAULT;
1798         for (seg = 0; seg < nr_segs; seg++) {
1799                 if ((ssize_t)iov[seg].iov_len < 0)
1800                         return -EINVAL;
1801         }
1802
1803         return 0;
1804 }
1805
1806 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1807                 unsigned long nr_segs, unsigned long fast_segs,
1808                 struct iovec *fast_iov, bool compat)
1809 {
1810         struct iovec *iov = fast_iov;
1811         int ret;
1812
1813         /*
1814          * SuS says "The readv() function *may* fail if the iovcnt argument was
1815          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1816          * traditionally returned zero for zero segments, so...
1817          */
1818         if (nr_segs == 0)
1819                 return iov;
1820         if (nr_segs > UIO_MAXIOV)
1821                 return ERR_PTR(-EINVAL);
1822         if (nr_segs > fast_segs) {
1823                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1824                 if (!iov)
1825                         return ERR_PTR(-ENOMEM);
1826         }
1827
1828         if (compat)
1829                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1830         else
1831                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1832         if (ret) {
1833                 if (iov != fast_iov)
1834                         kfree(iov);
1835                 return ERR_PTR(ret);
1836         }
1837
1838         return iov;
1839 }
1840
1841 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1842                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1843                  struct iov_iter *i, bool compat)
1844 {
1845         ssize_t total_len = 0;
1846         unsigned long seg;
1847         struct iovec *iov;
1848
1849         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1850         if (IS_ERR(iov)) {
1851                 *iovp = NULL;
1852                 return PTR_ERR(iov);
1853         }
1854
1855         /*
1856          * According to the Single Unix Specification we should return EINVAL if
1857          * an element length is < 0 when cast to ssize_t or if the total length
1858          * would overflow the ssize_t return value of the system call.
1859          *
1860          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1861          * overflow case.
1862          */
1863         for (seg = 0; seg < nr_segs; seg++) {
1864                 ssize_t len = (ssize_t)iov[seg].iov_len;
1865
1866                 if (!access_ok(iov[seg].iov_base, len)) {
1867                         if (iov != *iovp)
1868                                 kfree(iov);
1869                         *iovp = NULL;
1870                         return -EFAULT;
1871                 }
1872
1873                 if (len > MAX_RW_COUNT - total_len) {
1874                         len = MAX_RW_COUNT - total_len;
1875                         iov[seg].iov_len = len;
1876                 }
1877                 total_len += len;
1878         }
1879
1880         iov_iter_init(i, type, iov, nr_segs, total_len);
1881         if (iov == *iovp)
1882                 *iovp = NULL;
1883         else
1884                 *iovp = iov;
1885         return total_len;
1886 }
1887
1888 /**
1889  * import_iovec() - Copy an array of &struct iovec from userspace
1890  *     into the kernel, check that it is valid, and initialize a new
1891  *     &struct iov_iter iterator to access it.
1892  *
1893  * @type: One of %READ or %WRITE.
1894  * @uvec: Pointer to the userspace array.
1895  * @nr_segs: Number of elements in userspace array.
1896  * @fast_segs: Number of elements in @iov.
1897  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1898  *     on-stack) kernel array.
1899  * @i: Pointer to iterator that will be initialized on success.
1900  *
1901  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1902  * then this function places %NULL in *@iov on return. Otherwise, a new
1903  * array will be allocated and the result placed in *@iov. This means that
1904  * the caller may call kfree() on *@iov regardless of whether the small
1905  * on-stack array was used or not (and regardless of whether this function
1906  * returns an error or not).
1907  *
1908  * Return: Negative error code on error, bytes imported on success
1909  */
1910 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1911                  unsigned nr_segs, unsigned fast_segs,
1912                  struct iovec **iovp, struct iov_iter *i)
1913 {
1914         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1915                               in_compat_syscall());
1916 }
1917 EXPORT_SYMBOL(import_iovec);
1918
1919 int import_single_range(int rw, void __user *buf, size_t len,
1920                  struct iovec *iov, struct iov_iter *i)
1921 {
1922         if (len > MAX_RW_COUNT)
1923                 len = MAX_RW_COUNT;
1924         if (unlikely(!access_ok(buf, len)))
1925                 return -EFAULT;
1926
1927         iov->iov_base = buf;
1928         iov->iov_len = len;
1929         iov_iter_init(i, rw, iov, 1, len);
1930         return 0;
1931 }
1932 EXPORT_SYMBOL(import_single_range);
1933
1934 /**
1935  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1936  *     iov_iter_save_state() was called.
1937  *
1938  * @i: &struct iov_iter to restore
1939  * @state: state to restore from
1940  *
1941  * Used after iov_iter_save_state() to bring restore @i, if operations may
1942  * have advanced it.
1943  *
1944  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1945  */
1946 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1947 {
1948         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
1949                          !iov_iter_is_kvec(i))
1950                 return;
1951         i->iov_offset = state->iov_offset;
1952         i->count = state->count;
1953         /*
1954          * For the *vec iters, nr_segs + iov is constant - if we increment
1955          * the vec, then we also decrement the nr_segs count. Hence we don't
1956          * need to track both of these, just one is enough and we can deduct
1957          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1958          * size, so we can just increment the iov pointer as they are unionzed.
1959          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1960          * not. Be safe and handle it separately.
1961          */
1962         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1963         if (iov_iter_is_bvec(i))
1964                 i->bvec -= state->nr_segs - i->nr_segs;
1965         else
1966                 i->iov -= state->nr_segs - i->nr_segs;
1967         i->nr_segs = state->nr_segs;
1968 }