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         struct bvec_iter bi;
 850
 851         bi.bi_size = i->count;
 852         bi.bi_bvec_done = i->iov_offset;
 853         bi.bi_idx = 0;
 854         bvec_iter_advance(i->bvec, &bi, size);
 855
 856         i->bvec += bi.bi_idx;
 857         i->nr_segs -= bi.bi_idx;
 858         i->count = bi.bi_size;
 859         i->iov_offset = bi.bi_bvec_done;
 860 }
 861
 862 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 863 {
 864         const struct iovec *iov, *end;
 865
 866         if (!i->count)
 867                 return;
 868         i->count -= size;
 869
 870         size += i->iov_offset; // from beginning of current segment
 871         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
 872                 if (likely(size < iov->iov_len))
 873                         break;
 874                 size -= iov->iov_len;
 875         }
 876         i->iov_offset = size;
 877         i->nr_segs -= iov - i->iov;
 878         i->iov = iov;
 879 }
 880
 881 void iov_iter_advance(struct iov_iter *i, size_t size)
 882 {
 883         if (unlikely(i->count < size))
 884                 size = i->count;
 885         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 886                 /* iovec and kvec have identical layouts */
 887                 iov_iter_iovec_advance(i, size);
 888         } else if (iov_iter_is_bvec(i)) {
 889                 iov_iter_bvec_advance(i, size);
 890         } else if (iov_iter_is_pipe(i)) {
 891                 pipe_advance(i, size);
 892         } else if (unlikely(iov_iter_is_xarray(i))) {
 893                 i->iov_offset += size;
 894                 i->count -= size;
 895         } else if (iov_iter_is_discard(i)) {
 896                 i->count -= size;
 897         }
 898 }
 899 EXPORT_SYMBOL(iov_iter_advance);
 900
 901 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 902 {
 903         if (!unroll)
 904                 return;
 905         if (WARN_ON(unroll > MAX_RW_COUNT))
 906                 return;
 907         i->count += unroll;
 908         if (unlikely(iov_iter_is_pipe(i))) {
 909                 struct pipe_inode_info *pipe = i->pipe;
 910                 unsigned int p_mask = pipe->ring_size - 1;
 911                 unsigned int i_head = i->head;
 912                 size_t off = i->iov_offset;
 913                 while (1) {
 914                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
 915                         size_t n = off - b->offset;
 916                         if (unroll < n) {
 917                                 off -= unroll;
 918                                 break;
 919                         }
 920                         unroll -= n;
 921                         if (!unroll && i_head == i->start_head) {
 922                                 off = 0;
 923                                 break;
 924                         }
 925                         i_head--;
 926                         b = &pipe->bufs[i_head & p_mask];
 927                         off = b->offset + b->len;
 928                 }
 929                 i->iov_offset = off;
 930                 i->head = i_head;
 931                 pipe_truncate(i);
 932                 return;
 933         }
 934         if (unlikely(iov_iter_is_discard(i)))
 935                 return;
 936         if (unroll <= i->iov_offset) {
 937                 i->iov_offset -= unroll;
 938                 return;
 939         }
 940         unroll -= i->iov_offset;
 941         if (iov_iter_is_xarray(i)) {
 942                 BUG(); /* We should never go beyond the start of the specified
 943                         * range since we might then be straying into pages that
 944                         * aren't pinned.
 945                         */
 946         } else if (iov_iter_is_bvec(i)) {
 947                 const struct bio_vec *bvec = i->bvec;
 948                 while (1) {
 949                         size_t n = (--bvec)->bv_len;
 950                         i->nr_segs++;
 951                         if (unroll <= n) {
 952                                 i->bvec = bvec;
 953                                 i->iov_offset = n - unroll;
 954                                 return;
 955                         }
 956                         unroll -= n;
 957                 }
 958         } else { /* same logics for iovec and kvec */
 959                 const struct iovec *iov = i->iov;
 960                 while (1) {
 961                         size_t n = (--iov)->iov_len;
 962                         i->nr_segs++;
 963                         if (unroll <= n) {
 964                                 i->iov = iov;
 965                                 i->iov_offset = n - unroll;
 966                                 return;
 967                         }
 968                         unroll -= n;
 969                 }
 970         }
 971 }
 972 EXPORT_SYMBOL(iov_iter_revert);
 973
 974 /*
 975  * Return the count of just the current iov_iter segment.
 976  */
 977 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 978 {
 979         if (i->nr_segs > 1) {
 980                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 981                         return min(i->count, i->iov->iov_len - i->iov_offset);
 982                 if (iov_iter_is_bvec(i))
 983                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 984         }
 985         return i->count;
 986 }
 987 EXPORT_SYMBOL(iov_iter_single_seg_count);
 988
 989 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
 990                         const struct kvec *kvec, unsigned long nr_segs,
 991                         size_t count)
 992 {
 993         WARN_ON(direction & ~(READ | WRITE));
 994         *i = (struct iov_iter){
 995                 .iter_type = ITER_KVEC,
 996                 .data_source = direction,
 997                 .kvec = kvec,
 998                 .nr_segs = nr_segs,
 999                 .iov_offset = 0,
1000                 .count = count
1001         };
1002 }
1003 EXPORT_SYMBOL(iov_iter_kvec);
1004
1005 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1006                         const struct bio_vec *bvec, unsigned long nr_segs,
1007                         size_t count)
1008 {
1009         WARN_ON(direction & ~(READ | WRITE));
1010         *i = (struct iov_iter){
1011                 .iter_type = ITER_BVEC,
1012                 .data_source = direction,
1013                 .bvec = bvec,
1014                 .nr_segs = nr_segs,
1015                 .iov_offset = 0,
1016                 .count = count
1017         };
1018 }
1019 EXPORT_SYMBOL(iov_iter_bvec);
1020
1021 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1022                         struct pipe_inode_info *pipe,
1023                         size_t count)
1024 {
1025         BUG_ON(direction != READ);
1026         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1027         *i = (struct iov_iter){
1028                 .iter_type = ITER_PIPE,
1029                 .data_source = false,
1030                 .pipe = pipe,
1031                 .head = pipe->head,
1032                 .start_head = pipe->head,
1033                 .iov_offset = 0,
1034                 .count = count
1035         };
1036 }
1037 EXPORT_SYMBOL(iov_iter_pipe);
1038
1039 /**
1040  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1041  * @i: The iterator to initialise.
1042  * @direction: The direction of the transfer.
1043  * @xarray: The xarray to access.
1044  * @start: The start file position.
1045  * @count: The size of the I/O buffer in bytes.
1046  *
1047  * Set up an I/O iterator to either draw data out of the pages attached to an
1048  * inode or to inject data into those pages.  The pages *must* be prevented
1049  * from evaporation, either by taking a ref on them or locking them by the
1050  * caller.
1051  */
1052 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1053                      struct xarray *xarray, loff_t start, size_t count)
1054 {
1055         BUG_ON(direction & ~1);
1056         *i = (struct iov_iter) {
1057                 .iter_type = ITER_XARRAY,
1058                 .data_source = direction,
1059                 .xarray = xarray,
1060                 .xarray_start = start,
1061                 .count = count,
1062                 .iov_offset = 0
1063         };
1064 }
1065 EXPORT_SYMBOL(iov_iter_xarray);
1066
1067 /**
1068  * iov_iter_discard - Initialise an I/O iterator that discards data
1069  * @i: The iterator to initialise.
1070  * @direction: The direction of the transfer.
1071  * @count: The size of the I/O buffer in bytes.
1072  *
1073  * Set up an I/O iterator that just discards everything that's written to it.
1074  * It's only available as a READ iterator.
1075  */
1076 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1077 {
1078         BUG_ON(direction != READ);
1079         *i = (struct iov_iter){
1080                 .iter_type = ITER_DISCARD,
1081                 .data_source = false,
1082                 .count = count,
1083                 .iov_offset = 0
1084         };
1085 }
1086 EXPORT_SYMBOL(iov_iter_discard);
1087
1088 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1089 {
1090         unsigned long res = 0;
1091         size_t size = i->count;
1092         size_t skip = i->iov_offset;
1093         unsigned k;
1094
1095         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1096                 size_t len = i->iov[k].iov_len - skip;
1097                 if (len) {
1098                         res |= (unsigned long)i->iov[k].iov_base + skip;
1099                         if (len > size)
1100                                 len = size;
1101                         res |= len;
1102                         size -= len;
1103                         if (!size)
1104                                 break;
1105                 }
1106         }
1107         return res;
1108 }
1109
1110 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1111 {
1112         unsigned res = 0;
1113         size_t size = i->count;
1114         unsigned skip = i->iov_offset;
1115         unsigned k;
1116
1117         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1118                 size_t len = i->bvec[k].bv_len - skip;
1119                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1120                 if (len > size)
1121                         len = size;
1122                 res |= len;
1123                 size -= len;
1124                 if (!size)
1125                         break;
1126         }
1127         return res;
1128 }
1129
1130 unsigned long iov_iter_alignment(const struct iov_iter *i)
1131 {
1132         /* iovec and kvec have identical layouts */
1133         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1134                 return iov_iter_alignment_iovec(i);
1135
1136         if (iov_iter_is_bvec(i))
1137                 return iov_iter_alignment_bvec(i);
1138
1139         if (iov_iter_is_pipe(i)) {
1140                 unsigned int p_mask = i->pipe->ring_size - 1;
1141                 size_t size = i->count;
1142
1143                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1144                         return size | i->iov_offset;
1145                 return size;
1146         }
1147
1148         if (iov_iter_is_xarray(i))
1149                 return (i->xarray_start + i->iov_offset) | i->count;
1150
1151         return 0;
1152 }
1153 EXPORT_SYMBOL(iov_iter_alignment);
1154
1155 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1156 {
1157         unsigned long res = 0;
1158         unsigned long v = 0;
1159         size_t size = i->count;
1160         unsigned k;
1161
1162         if (WARN_ON(!iter_is_iovec(i)))
1163                 return ~0U;
1164
1165         for (k = 0; k < i->nr_segs; k++) {
1166                 if (i->iov[k].iov_len) {
1167                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1168                         if (v) // if not the first one
1169                                 res |= base | v; // this start | previous end
1170                         v = base + i->iov[k].iov_len;
1171                         if (size <= i->iov[k].iov_len)
1172                                 break;
1173                         size -= i->iov[k].iov_len;
1174                 }
1175         }
1176         return res;
1177 }
1178 EXPORT_SYMBOL(iov_iter_gap_alignment);
1179
1180 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1181                                 size_t maxsize,
1182                                 struct page **pages,
1183                                 int iter_head,
1184                                 size_t *start)
1185 {
1186         struct pipe_inode_info *pipe = i->pipe;
1187         unsigned int p_mask = pipe->ring_size - 1;
1188         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1189         if (!n)
1190                 return -EFAULT;
1191
1192         maxsize = n;
1193         n += *start;
1194         while (n > 0) {
1195                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1196                 iter_head++;
1197                 n -= PAGE_SIZE;
1198         }
1199
1200         return maxsize;
1201 }
1202
1203 static ssize_t pipe_get_pages(struct iov_iter *i,
1204                    struct page **pages, size_t maxsize, unsigned maxpages,
1205                    size_t *start)
1206 {
1207         unsigned int iter_head, npages;
1208         size_t capacity;
1209
1210         if (!sanity(i))
1211                 return -EFAULT;
1212
1213         data_start(i, &iter_head, start);
1214         /* Amount of free space: some of this one + all after this one */
1215         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1216         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1217
1218         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1219 }
1220
1221 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1222                                           pgoff_t index, unsigned int nr_pages)
1223 {
1224         XA_STATE(xas, xa, index);
1225         struct page *page;
1226         unsigned int ret = 0;
1227
1228         rcu_read_lock();
1229         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1230                 if (xas_retry(&xas, page))
1231                         continue;
1232
1233                 /* Has the page moved or been split? */
1234                 if (unlikely(page != xas_reload(&xas))) {
1235                         xas_reset(&xas);
1236                         continue;
1237                 }
1238
1239                 pages[ret] = find_subpage(page, xas.xa_index);
1240                 get_page(pages[ret]);
1241                 if (++ret == nr_pages)
1242                         break;
1243         }
1244         rcu_read_unlock();
1245         return ret;
1246 }
1247
1248 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1249                                      struct page **pages, size_t maxsize,
1250                                      unsigned maxpages, size_t *_start_offset)
1251 {
1252         unsigned nr, offset;
1253         pgoff_t index, count;
1254         size_t size = maxsize, actual;
1255         loff_t pos;
1256
1257         if (!size || !maxpages)
1258                 return 0;
1259
1260         pos = i->xarray_start + i->iov_offset;
1261         index = pos >> PAGE_SHIFT;
1262         offset = pos & ~PAGE_MASK;
1263         *_start_offset = offset;
1264
1265         count = 1;
1266         if (size > PAGE_SIZE - offset) {
1267                 size -= PAGE_SIZE - offset;
1268                 count += size >> PAGE_SHIFT;
1269                 size &= ~PAGE_MASK;
1270                 if (size)
1271                         count++;
1272         }
1273
1274         if (count > maxpages)
1275                 count = maxpages;
1276
1277         nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1278         if (nr == 0)
1279                 return 0;
1280
1281         actual = PAGE_SIZE * nr;
1282         actual -= offset;
1283         if (nr == count && size > 0) {
1284                 unsigned last_offset = (nr > 1) ? 0 : offset;
1285                 actual -= PAGE_SIZE - (last_offset + size);
1286         }
1287         return actual;
1288 }
1289
1290 /* must be done on non-empty ITER_IOVEC one */
1291 static unsigned long first_iovec_segment(const struct iov_iter *i,
1292                                          size_t *size, size_t *start,
1293                                          size_t maxsize, unsigned maxpages)
1294 {
1295         size_t skip;
1296         long k;
1297
1298         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1299                 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1300                 size_t len = i->iov[k].iov_len - skip;
1301
1302                 if (unlikely(!len))
1303                         continue;
1304                 if (len > maxsize)
1305                         len = maxsize;
1306                 len += (*start = addr % PAGE_SIZE);
1307                 if (len > maxpages * PAGE_SIZE)
1308                         len = maxpages * PAGE_SIZE;
1309                 *size = len;
1310                 return addr & PAGE_MASK;
1311         }
1312         BUG(); // if it had been empty, we wouldn't get called
1313 }
1314
1315 /* must be done on non-empty ITER_BVEC one */
1316 static struct page *first_bvec_segment(const struct iov_iter *i,
1317                                        size_t *size, size_t *start,
1318                                        size_t maxsize, unsigned maxpages)
1319 {
1320         struct page *page;
1321         size_t skip = i->iov_offset, len;
1322
1323         len = i->bvec->bv_len - skip;
1324         if (len > maxsize)
1325                 len = maxsize;
1326         skip += i->bvec->bv_offset;
1327         page = i->bvec->bv_page + skip / PAGE_SIZE;
1328         len += (*start = skip % PAGE_SIZE);
1329         if (len > maxpages * PAGE_SIZE)
1330                 len = maxpages * PAGE_SIZE;
1331         *size = len;
1332         return page;
1333 }
1334
1335 ssize_t iov_iter_get_pages(struct iov_iter *i,
1336                    struct page **pages, size_t maxsize, unsigned maxpages,
1337                    size_t *start)
1338 {
1339         size_t len;
1340         int n, res;
1341
1342         if (maxsize > i->count)
1343                 maxsize = i->count;
1344         if (!maxsize)
1345                 return 0;
1346
1347         if (likely(iter_is_iovec(i))) {
1348                 unsigned int gup_flags = 0;
1349                 unsigned long addr;
1350
1351                 if (iov_iter_rw(i) != WRITE)
1352                         gup_flags |= FOLL_WRITE;
1353                 if (i->nofault)
1354                         gup_flags |= FOLL_NOFAULT;
1355
1356                 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1357                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1358                 res = get_user_pages_fast(addr, n, gup_flags, pages);
1359                 if (unlikely(res <= 0))
1360                         return res;
1361                 return (res == n ? len : res * PAGE_SIZE) - *start;
1362         }
1363         if (iov_iter_is_bvec(i)) {
1364                 struct page *page;
1365
1366                 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1367                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1368                 while (n--)
1369                         get_page(*pages++ = page++);
1370                 return len - *start;
1371         }
1372         if (iov_iter_is_pipe(i))
1373                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1374         if (iov_iter_is_xarray(i))
1375                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1376         return -EFAULT;
1377 }
1378 EXPORT_SYMBOL(iov_iter_get_pages);
1379
1380 static struct page **get_pages_array(size_t n)
1381 {
1382         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1383 }
1384
1385 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1386                    struct page ***pages, size_t maxsize,
1387                    size_t *start)
1388 {
1389         struct page **p;
1390         unsigned int iter_head, npages;
1391         ssize_t n;
1392
1393         if (!sanity(i))
1394                 return -EFAULT;
1395
1396         data_start(i, &iter_head, start);
1397         /* Amount of free space: some of this one + all after this one */
1398         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1399         n = npages * PAGE_SIZE - *start;
1400         if (maxsize > n)
1401                 maxsize = n;
1402         else
1403                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1404         p = get_pages_array(npages);
1405         if (!p)
1406                 return -ENOMEM;
1407         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1408         if (n > 0)
1409                 *pages = p;
1410         else
1411                 kvfree(p);
1412         return n;
1413 }
1414
1415 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1416                                            struct page ***pages, size_t maxsize,
1417                                            size_t *_start_offset)
1418 {
1419         struct page **p;
1420         unsigned nr, offset;
1421         pgoff_t index, count;
1422         size_t size = maxsize, actual;
1423         loff_t pos;
1424
1425         if (!size)
1426                 return 0;
1427
1428         pos = i->xarray_start + i->iov_offset;
1429         index = pos >> PAGE_SHIFT;
1430         offset = pos & ~PAGE_MASK;
1431         *_start_offset = offset;
1432
1433         count = 1;
1434         if (size > PAGE_SIZE - offset) {
1435                 size -= PAGE_SIZE - offset;
1436                 count += size >> PAGE_SHIFT;
1437                 size &= ~PAGE_MASK;
1438                 if (size)
1439                         count++;
1440         }
1441
1442         p = get_pages_array(count);
1443         if (!p)
1444                 return -ENOMEM;
1445         *pages = p;
1446
1447         nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1448         if (nr == 0)
1449                 return 0;
1450
1451         actual = PAGE_SIZE * nr;
1452         actual -= offset;
1453         if (nr == count && size > 0) {
1454                 unsigned last_offset = (nr > 1) ? 0 : offset;
1455                 actual -= PAGE_SIZE - (last_offset + size);
1456         }
1457         return actual;
1458 }
1459
1460 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1461                    struct page ***pages, size_t maxsize,
1462                    size_t *start)
1463 {
1464         struct page **p;
1465         size_t len;
1466         int n, res;
1467
1468         if (maxsize > i->count)
1469                 maxsize = i->count;
1470         if (!maxsize)
1471                 return 0;
1472
1473         if (likely(iter_is_iovec(i))) {
1474                 unsigned int gup_flags = 0;
1475                 unsigned long addr;
1476
1477                 if (iov_iter_rw(i) != WRITE)
1478                         gup_flags |= FOLL_WRITE;
1479                 if (i->nofault)
1480                         gup_flags |= FOLL_NOFAULT;
1481
1482                 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1483                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1484                 p = get_pages_array(n);
1485                 if (!p)
1486                         return -ENOMEM;
1487                 res = get_user_pages_fast(addr, n, gup_flags, p);
1488                 if (unlikely(res <= 0)) {
1489                         kvfree(p);
1490                         *pages = NULL;
1491                         return res;
1492                 }
1493                 *pages = p;
1494                 return (res == n ? len : res * PAGE_SIZE) - *start;
1495         }
1496         if (iov_iter_is_bvec(i)) {
1497                 struct page *page;
1498
1499                 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1500                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1501                 *pages = p = get_pages_array(n);
1502                 if (!p)
1503                         return -ENOMEM;
1504                 while (n--)
1505                         get_page(*p++ = page++);
1506                 return len - *start;
1507         }
1508         if (iov_iter_is_pipe(i))
1509                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1510         if (iov_iter_is_xarray(i))
1511                 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1512         return -EFAULT;
1513 }
1514 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1515
1516 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1517                                struct iov_iter *i)
1518 {
1519         __wsum sum, next;
1520         sum = *csum;
1521         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1522                 WARN_ON(1);
1523                 return 0;
1524         }
1525         iterate_and_advance(i, bytes, base, len, off, ({
1526                 next = csum_and_copy_from_user(base, addr + off, len);
1527                 sum = csum_block_add(sum, next, off);
1528                 next ? 0 : len;
1529         }), ({
1530                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1531         })
1532         )
1533         *csum = sum;
1534         return bytes;
1535 }
1536 EXPORT_SYMBOL(csum_and_copy_from_iter);
1537
1538 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1539                              struct iov_iter *i)
1540 {
1541         struct csum_state *csstate = _csstate;
1542         __wsum sum, next;
1543
1544         if (unlikely(iov_iter_is_discard(i))) {
1545                 WARN_ON(1);     /* for now */
1546                 return 0;
1547         }
1548
1549         sum = csum_shift(csstate->csum, csstate->off);
1550         if (unlikely(iov_iter_is_pipe(i)))
1551                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1552         else iterate_and_advance(i, bytes, base, len, off, ({
1553                 next = csum_and_copy_to_user(addr + off, base, len);
1554                 sum = csum_block_add(sum, next, off);
1555                 next ? 0 : len;
1556         }), ({
1557                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1558         })
1559         )
1560         csstate->csum = csum_shift(sum, csstate->off);
1561         csstate->off += bytes;
1562         return bytes;
1563 }
1564 EXPORT_SYMBOL(csum_and_copy_to_iter);
1565
1566 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1567                 struct iov_iter *i)
1568 {
1569 #ifdef CONFIG_CRYPTO_HASH
1570         struct ahash_request *hash = hashp;
1571         struct scatterlist sg;
1572         size_t copied;
1573
1574         copied = copy_to_iter(addr, bytes, i);
1575         sg_init_one(&sg, addr, copied);
1576         ahash_request_set_crypt(hash, &sg, NULL, copied);
1577         crypto_ahash_update(hash);
1578         return copied;
1579 #else
1580         return 0;
1581 #endif
1582 }
1583 EXPORT_SYMBOL(hash_and_copy_to_iter);
1584
1585 static int iov_npages(const struct iov_iter *i, int maxpages)
1586 {
1587         size_t skip = i->iov_offset, size = i->count;
1588         const struct iovec *p;
1589         int npages = 0;
1590
1591         for (p = i->iov; size; skip = 0, p++) {
1592                 unsigned offs = offset_in_page(p->iov_base + skip);
1593                 size_t len = min(p->iov_len - skip, size);
1594
1595                 if (len) {
1596                         size -= len;
1597                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1598                         if (unlikely(npages > maxpages))
1599                                 return maxpages;
1600                 }
1601         }
1602         return npages;
1603 }
1604
1605 static int bvec_npages(const struct iov_iter *i, int maxpages)
1606 {
1607         size_t skip = i->iov_offset, size = i->count;
1608         const struct bio_vec *p;
1609         int npages = 0;
1610
1611         for (p = i->bvec; size; skip = 0, p++) {
1612                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1613                 size_t len = min(p->bv_len - skip, size);
1614
1615                 size -= len;
1616                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1617                 if (unlikely(npages > maxpages))
1618                         return maxpages;
1619         }
1620         return npages;
1621 }
1622
1623 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1624 {
1625         if (unlikely(!i->count))
1626                 return 0;
1627         /* iovec and kvec have identical layouts */
1628         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1629                 return iov_npages(i, maxpages);
1630         if (iov_iter_is_bvec(i))
1631                 return bvec_npages(i, maxpages);
1632         if (iov_iter_is_pipe(i)) {
1633                 unsigned int iter_head;
1634                 int npages;
1635                 size_t off;
1636
1637                 if (!sanity(i))
1638                         return 0;
1639
1640                 data_start(i, &iter_head, &off);
1641                 /* some of this one + all after this one */
1642                 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1643                 return min(npages, maxpages);
1644         }
1645         if (iov_iter_is_xarray(i)) {
1646                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1647                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1648                 return min(npages, maxpages);
1649         }
1650         return 0;
1651 }
1652 EXPORT_SYMBOL(iov_iter_npages);
1653
1654 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1655 {
1656         *new = *old;
1657         if (unlikely(iov_iter_is_pipe(new))) {
1658                 WARN_ON(1);
1659                 return NULL;
1660         }
1661         if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1662                 return NULL;
1663         if (iov_iter_is_bvec(new))
1664                 return new->bvec = kmemdup(new->bvec,
1665                                     new->nr_segs * sizeof(struct bio_vec),
1666                                     flags);
1667         else
1668                 /* iovec and kvec have identical layout */
1669                 return new->iov = kmemdup(new->iov,
1670                                    new->nr_segs * sizeof(struct iovec),
1671                                    flags);
1672 }
1673 EXPORT_SYMBOL(dup_iter);
1674
1675 static int copy_compat_iovec_from_user(struct iovec *iov,
1676                 const struct iovec __user *uvec, unsigned long nr_segs)
1677 {
1678         const struct compat_iovec __user *uiov =
1679                 (const struct compat_iovec __user *)uvec;
1680         int ret = -EFAULT, i;
1681
1682         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1683                 return -EFAULT;
1684
1685         for (i = 0; i < nr_segs; i++) {
1686                 compat_uptr_t buf;
1687                 compat_ssize_t len;
1688
1689                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1690                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1691
1692                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1693                 if (len < 0) {
1694                         ret = -EINVAL;
1695                         goto uaccess_end;
1696                 }
1697                 iov[i].iov_base = compat_ptr(buf);
1698                 iov[i].iov_len = len;
1699         }
1700
1701         ret = 0;
1702 uaccess_end:
1703         user_access_end();
1704         return ret;
1705 }
1706
1707 static int copy_iovec_from_user(struct iovec *iov,
1708                 const struct iovec __user *uvec, unsigned long nr_segs)
1709 {
1710         unsigned long seg;
1711
1712         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1713                 return -EFAULT;
1714         for (seg = 0; seg < nr_segs; seg++) {
1715                 if ((ssize_t)iov[seg].iov_len < 0)
1716                         return -EINVAL;
1717         }
1718
1719         return 0;
1720 }
1721
1722 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1723                 unsigned long nr_segs, unsigned long fast_segs,
1724                 struct iovec *fast_iov, bool compat)
1725 {
1726         struct iovec *iov = fast_iov;
1727         int ret;
1728
1729         /*
1730          * SuS says "The readv() function *may* fail if the iovcnt argument was
1731          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1732          * traditionally returned zero for zero segments, so...
1733          */
1734         if (nr_segs == 0)
1735                 return iov;
1736         if (nr_segs > UIO_MAXIOV)
1737                 return ERR_PTR(-EINVAL);
1738         if (nr_segs > fast_segs) {
1739                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1740                 if (!iov)
1741                         return ERR_PTR(-ENOMEM);
1742         }
1743
1744         if (compat)
1745                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1746         else
1747                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1748         if (ret) {
1749                 if (iov != fast_iov)
1750                         kfree(iov);
1751                 return ERR_PTR(ret);
1752         }
1753
1754         return iov;
1755 }
1756
1757 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1758                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1759                  struct iov_iter *i, bool compat)
1760 {
1761         ssize_t total_len = 0;
1762         unsigned long seg;
1763         struct iovec *iov;
1764
1765         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1766         if (IS_ERR(iov)) {
1767                 *iovp = NULL;
1768                 return PTR_ERR(iov);
1769         }
1770
1771         /*
1772          * According to the Single Unix Specification we should return EINVAL if
1773          * an element length is < 0 when cast to ssize_t or if the total length
1774          * would overflow the ssize_t return value of the system call.
1775          *
1776          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1777          * overflow case.
1778          */
1779         for (seg = 0; seg < nr_segs; seg++) {
1780                 ssize_t len = (ssize_t)iov[seg].iov_len;
1781
1782                 if (!access_ok(iov[seg].iov_base, len)) {
1783                         if (iov != *iovp)
1784                                 kfree(iov);
1785                         *iovp = NULL;
1786                         return -EFAULT;
1787                 }
1788
1789                 if (len > MAX_RW_COUNT - total_len) {
1790                         len = MAX_RW_COUNT - total_len;
1791                         iov[seg].iov_len = len;
1792                 }
1793                 total_len += len;
1794         }
1795
1796         iov_iter_init(i, type, iov, nr_segs, total_len);
1797         if (iov == *iovp)
1798                 *iovp = NULL;
1799         else
1800                 *iovp = iov;
1801         return total_len;
1802 }
1803
1804 /**
1805  * import_iovec() - Copy an array of &struct iovec from userspace
1806  *     into the kernel, check that it is valid, and initialize a new
1807  *     &struct iov_iter iterator to access it.
1808  *
1809  * @type: One of %READ or %WRITE.
1810  * @uvec: Pointer to the userspace array.
1811  * @nr_segs: Number of elements in userspace array.
1812  * @fast_segs: Number of elements in @iov.
1813  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1814  *     on-stack) kernel array.
1815  * @i: Pointer to iterator that will be initialized on success.
1816  *
1817  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1818  * then this function places %NULL in *@iov on return. Otherwise, a new
1819  * array will be allocated and the result placed in *@iov. This means that
1820  * the caller may call kfree() on *@iov regardless of whether the small
1821  * on-stack array was used or not (and regardless of whether this function
1822  * returns an error or not).
1823  *
1824  * Return: Negative error code on error, bytes imported on success
1825  */
1826 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1827                  unsigned nr_segs, unsigned fast_segs,
1828                  struct iovec **iovp, struct iov_iter *i)
1829 {
1830         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1831                               in_compat_syscall());
1832 }
1833 EXPORT_SYMBOL(import_iovec);
1834
1835 int import_single_range(int rw, void __user *buf, size_t len,
1836                  struct iovec *iov, struct iov_iter *i)
1837 {
1838         if (len > MAX_RW_COUNT)
1839                 len = MAX_RW_COUNT;
1840         if (unlikely(!access_ok(buf, len)))
1841                 return -EFAULT;
1842
1843         iov->iov_base = buf;
1844         iov->iov_len = len;
1845         iov_iter_init(i, rw, iov, 1, len);
1846         return 0;
1847 }
1848 EXPORT_SYMBOL(import_single_range);
1849
1850 /**
1851  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1852  *     iov_iter_save_state() was called.
1853  *
1854  * @i: &struct iov_iter to restore
1855  * @state: state to restore from
1856  *
1857  * Used after iov_iter_save_state() to bring restore @i, if operations may
1858  * have advanced it.
1859  *
1860  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1861  */
1862 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1863 {
1864         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
1865                          !iov_iter_is_kvec(i))
1866                 return;
1867         i->iov_offset = state->iov_offset;
1868         i->count = state->count;
1869         /*
1870          * For the *vec iters, nr_segs + iov is constant - if we increment
1871          * the vec, then we also decrement the nr_segs count. Hence we don't
1872          * need to track both of these, just one is enough and we can deduct
1873          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1874          * size, so we can just increment the iov pointer as they are unionzed.
1875          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1876          * not. Be safe and handle it separately.
1877          */
1878         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1879         if (iov_iter_is_bvec(i))
1880                 i->bvec -= state->nr_segs - i->nr_segs;
1881         else
1882                 i->iov -= state->nr_segs - i->nr_segs;
1883         i->nr_segs = state->nr_segs;
1884 }