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 ubuf and kbuf alike */
  20 #define iterate_buf(i, n, base, len, off, __p, STEP) {          \
  21         size_t __maybe_unused off = 0;                          \
  22         len = n;                                                \
  23         base = __p + i->iov_offset;                             \
  24         len -= (STEP);                                          \
  25         i->iov_offset += len;                                   \
  26         n = len;                                                \
  27 }
  28
  29 /* covers iovec and kvec alike */
  30 #define iterate_iovec(i, n, base, len, off, __p, STEP) {        \
  31         size_t off = 0;                                         \
  32         size_t skip = i->iov_offset;                            \
  33         do {                                                    \
  34                 len = min(n, __p->iov_len - skip);              \
  35                 if (likely(len)) {                              \
  36                         base = __p->iov_base + skip;            \
  37                         len -= (STEP);                          \
  38                         off += len;                             \
  39                         skip += len;                            \
  40                         n -= len;                               \
  41                         if (skip < __p->iov_len)                \
  42                                 break;                          \
  43                 }                                               \
  44                 __p++;                                          \
  45                 skip = 0;                                       \
  46         } while (n);                                            \
  47         i->iov_offset = skip;                                   \
  48         n = off;                                                \
  49 }
  50
  51 #define iterate_bvec(i, n, base, len, off, p, STEP) {           \
  52         size_t off = 0;                                         \
  53         unsigned skip = i->iov_offset;                          \
  54         while (n) {                                             \
  55                 unsigned offset = p->bv_offset + skip;          \
  56                 unsigned left;                                  \
  57                 void *kaddr = kmap_local_page(p->bv_page +      \
  58                                         offset / PAGE_SIZE);    \
  59                 base = kaddr + offset % PAGE_SIZE;              \
  60                 len = min(min(n, (size_t)(p->bv_len - skip)),   \
  61                      (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
  62                 left = (STEP);                                  \
  63                 kunmap_local(kaddr);                            \
  64                 len -= left;                                    \
  65                 off += len;                                     \
  66                 skip += len;                                    \
  67                 if (skip == p->bv_len) {                        \
  68                         skip = 0;                               \
  69                         p++;                                    \
  70                 }                                               \
  71                 n -= len;                                       \
  72                 if (left)                                       \
  73                         break;                                  \
  74         }                                                       \
  75         i->iov_offset = skip;                                   \
  76         n = off;                                                \
  77 }
  78
  79 #define iterate_xarray(i, n, base, len, __off, STEP) {          \
  80         __label__ __out;                                        \
  81         size_t __off = 0;                                       \
  82         struct folio *folio;                                    \
  83         loff_t start = i->xarray_start + i->iov_offset;         \
  84         pgoff_t index = start / PAGE_SIZE;                      \
  85         XA_STATE(xas, i->xarray, index);                        \
  86                                                                 \
  87         len = PAGE_SIZE - offset_in_page(start);                \
  88         rcu_read_lock();                                        \
  89         xas_for_each(&xas, folio, ULONG_MAX) {                  \
  90                 unsigned left;                                  \
  91                 size_t offset;                                  \
  92                 if (xas_retry(&xas, folio))                     \
  93                         continue;                               \
  94                 if (WARN_ON(xa_is_value(folio)))                \
  95                         break;                                  \
  96                 if (WARN_ON(folio_test_hugetlb(folio)))         \
  97                         break;                                  \
  98                 offset = offset_in_folio(folio, start + __off); \
  99                 while (offset < folio_size(folio)) {            \
 100                         base = kmap_local_folio(folio, offset); \
 101                         len = min(n, len);                      \
 102                         left = (STEP);                          \
 103                         kunmap_local(base);                     \
 104                         len -= left;                            \
 105                         __off += len;                           \
 106                         n -= len;                               \
 107                         if (left || n == 0)                     \
 108                                 goto __out;                     \
 109                         offset += len;                          \
 110                         len = PAGE_SIZE;                        \
 111                 }                                               \
 112         }                                                       \
 113 __out:                                                          \
 114         rcu_read_unlock();                                      \
 115         i->iov_offset += __off;                                 \
 116         n = __off;                                              \
 117 }
 118
 119 #define __iterate_and_advance(i, n, base, len, off, I, K) {     \
 120         if (unlikely(i->count < n))                             \
 121                 n = i->count;                                   \
 122         if (likely(n)) {                                        \
 123                 if (likely(iter_is_ubuf(i))) {                  \
 124                         void __user *base;                      \
 125                         size_t len;                             \
 126                         iterate_buf(i, n, base, len, off,       \
 127                                                 i->ubuf, (I))   \
 128                 } else if (likely(iter_is_iovec(i))) {          \
 129                         const struct iovec *iov = i->iov;       \
 130                         void __user *base;                      \
 131                         size_t len;                             \
 132                         iterate_iovec(i, n, base, len, off,     \
 133                                                 iov, (I))       \
 134                         i->nr_segs -= iov - i->iov;             \
 135                         i->iov = iov;                           \
 136                 } else if (iov_iter_is_bvec(i)) {               \
 137                         const struct bio_vec *bvec = i->bvec;   \
 138                         void *base;                             \
 139                         size_t len;                             \
 140                         iterate_bvec(i, n, base, len, off,      \
 141                                                 bvec, (K))      \
 142                         i->nr_segs -= bvec - i->bvec;           \
 143                         i->bvec = bvec;                         \
 144                 } else if (iov_iter_is_kvec(i)) {               \
 145                         const struct kvec *kvec = i->kvec;      \
 146                         void *base;                             \
 147                         size_t len;                             \
 148                         iterate_iovec(i, n, base, len, off,     \
 149                                                 kvec, (K))      \
 150                         i->nr_segs -= kvec - i->kvec;           \
 151                         i->kvec = kvec;                         \
 152                 } else if (iov_iter_is_xarray(i)) {             \
 153                         void *base;                             \
 154                         size_t len;                             \
 155                         iterate_xarray(i, n, base, len, off,    \
 156                                                         (K))    \
 157                 }                                               \
 158                 i->count -= n;                                  \
 159         }                                                       \
 160 }
 161 #define iterate_and_advance(i, n, base, len, off, I, K) \
 162         __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
 163
 164 static int copyout(void __user *to, const void *from, size_t n)
 165 {
 166         if (should_fail_usercopy())
 167                 return n;
 168         if (access_ok(to, n)) {
 169                 instrument_copy_to_user(to, from, n);
 170                 n = raw_copy_to_user(to, from, n);
 171         }
 172         return n;
 173 }
 174
 175 static int copyin(void *to, const void __user *from, size_t n)
 176 {
 177         size_t res = n;
 178
 179         if (should_fail_usercopy())
 180                 return n;
 181         if (access_ok(from, n)) {
 182                 instrument_copy_from_user_before(to, from, n);
 183                 res = raw_copy_from_user(to, from, n);
 184                 instrument_copy_from_user_after(to, from, n, res);
 185         }
 186         return res;
 187 }
 188
 189 static inline struct pipe_buffer *pipe_buf(const struct pipe_inode_info *pipe,
 190                                            unsigned int slot)
 191 {
 192         return &pipe->bufs[slot & (pipe->ring_size - 1)];
 193 }
 194
 195 #ifdef PIPE_PARANOIA
 196 static bool sanity(const struct iov_iter *i)
 197 {
 198         struct pipe_inode_info *pipe = i->pipe;
 199         unsigned int p_head = pipe->head;
 200         unsigned int p_tail = pipe->tail;
 201         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 202         unsigned int i_head = i->head;
 203         unsigned int idx;
 204
 205         if (i->last_offset) {
 206                 struct pipe_buffer *p;
 207                 if (unlikely(p_occupancy == 0))
 208                         goto Bad;       // pipe must be non-empty
 209                 if (unlikely(i_head != p_head - 1))
 210                         goto Bad;       // must be at the last buffer...
 211
 212                 p = pipe_buf(pipe, i_head);
 213                 if (unlikely(p->offset + p->len != abs(i->last_offset)))
 214                         goto Bad;       // ... at the end of segment
 215         } else {
 216                 if (i_head != p_head)
 217                         goto Bad;       // must be right after the last buffer
 218         }
 219         return true;
 220 Bad:
 221         printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset);
 222         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 223                         p_head, p_tail, pipe->ring_size);
 224         for (idx = 0; idx < pipe->ring_size; idx++)
 225                 printk(KERN_ERR "[%p %p %d %d]\n",
 226                         pipe->bufs[idx].ops,
 227                         pipe->bufs[idx].page,
 228                         pipe->bufs[idx].offset,
 229                         pipe->bufs[idx].len);
 230         WARN_ON(1);
 231         return false;
 232 }
 233 #else
 234 #define sanity(i) true
 235 #endif
 236
 237 static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size)
 238 {
 239         struct page *page = alloc_page(GFP_USER);
 240         if (page) {
 241                 struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 242                 *buf = (struct pipe_buffer) {
 243                         .ops = &default_pipe_buf_ops,
 244                         .page = page,
 245                         .offset = 0,
 246                         .len = size
 247                 };
 248         }
 249         return page;
 250 }
 251
 252 static void push_page(struct pipe_inode_info *pipe, struct page *page,
 253                         unsigned int offset, unsigned int size)
 254 {
 255         struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 256         *buf = (struct pipe_buffer) {
 257                 .ops = &page_cache_pipe_buf_ops,
 258                 .page = page,
 259                 .offset = offset,
 260                 .len = size
 261         };
 262         get_page(page);
 263 }
 264
 265 static inline int last_offset(const struct pipe_buffer *buf)
 266 {
 267         if (buf->ops == &default_pipe_buf_ops)
 268                 return buf->len;        // buf->offset is 0 for those
 269         else
 270                 return -(buf->offset + buf->len);
 271 }
 272
 273 static struct page *append_pipe(struct iov_iter *i, size_t size,
 274                                 unsigned int *off)
 275 {
 276         struct pipe_inode_info *pipe = i->pipe;
 277         int offset = i->last_offset;
 278         struct pipe_buffer *buf;
 279         struct page *page;
 280
 281         if (offset > 0 && offset < PAGE_SIZE) {
 282                 // some space in the last buffer; add to it
 283                 buf = pipe_buf(pipe, pipe->head - 1);
 284                 size = min_t(size_t, size, PAGE_SIZE - offset);
 285                 buf->len += size;
 286                 i->last_offset += size;
 287                 i->count -= size;
 288                 *off = offset;
 289                 return buf->page;
 290         }
 291         // OK, we need a new buffer
 292         *off = 0;
 293         size = min_t(size_t, size, PAGE_SIZE);
 294         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 295                 return NULL;
 296         page = push_anon(pipe, size);
 297         if (!page)
 298                 return NULL;
 299         i->head = pipe->head - 1;
 300         i->last_offset = size;
 301         i->count -= size;
 302         return page;
 303 }
 304
 305 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 306                          struct iov_iter *i)
 307 {
 308         struct pipe_inode_info *pipe = i->pipe;
 309         unsigned int head = pipe->head;
 310
 311         if (unlikely(bytes > i->count))
 312                 bytes = i->count;
 313
 314         if (unlikely(!bytes))
 315                 return 0;
 316
 317         if (!sanity(i))
 318                 return 0;
 319
 320         if (offset && i->last_offset == -offset) { // could we merge it?
 321                 struct pipe_buffer *buf = pipe_buf(pipe, head - 1);
 322                 if (buf->page == page) {
 323                         buf->len += bytes;
 324                         i->last_offset -= bytes;
 325                         i->count -= bytes;
 326                         return bytes;
 327                 }
 328         }
 329         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 330                 return 0;
 331
 332         push_page(pipe, page, offset, bytes);
 333         i->last_offset = -(offset + bytes);
 334         i->head = head;
 335         i->count -= bytes;
 336         return bytes;
 337 }
 338
 339 /*
 340  * fault_in_iov_iter_readable - fault in iov iterator for reading
 341  * @i: iterator
 342  * @size: maximum length
 343  *
 344  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 345  * @size.  For each iovec, fault in each page that constitutes the iovec.
 346  *
 347  * Returns the number of bytes not faulted in (like copy_to_user() and
 348  * copy_from_user()).
 349  *
 350  * Always returns 0 for non-userspace iterators.
 351  */
 352 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
 353 {
 354         if (iter_is_ubuf(i)) {
 355                 size_t n = min(size, iov_iter_count(i));
 356                 n -= fault_in_readable(i->ubuf + i->iov_offset, n);
 357                 return size - n;
 358         } else if (iter_is_iovec(i)) {
 359                 size_t count = min(size, iov_iter_count(i));
 360                 const struct iovec *p;
 361                 size_t skip;
 362
 363                 size -= count;
 364                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 365                         size_t len = min(count, p->iov_len - skip);
 366                         size_t ret;
 367
 368                         if (unlikely(!len))
 369                                 continue;
 370                         ret = fault_in_readable(p->iov_base + skip, len);
 371                         count -= len - ret;
 372                         if (ret)
 373                                 break;
 374                 }
 375                 return count + size;
 376         }
 377         return 0;
 378 }
 379 EXPORT_SYMBOL(fault_in_iov_iter_readable);
 380
 381 /*
 382  * fault_in_iov_iter_writeable - fault in iov iterator for writing
 383  * @i: iterator
 384  * @size: maximum length
 385  *
 386  * Faults in the iterator using get_user_pages(), i.e., without triggering
 387  * hardware page faults.  This is primarily useful when we already know that
 388  * some or all of the pages in @i aren't in memory.
 389  *
 390  * Returns the number of bytes not faulted in, like copy_to_user() and
 391  * copy_from_user().
 392  *
 393  * Always returns 0 for non-user-space iterators.
 394  */
 395 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
 396 {
 397         if (iter_is_ubuf(i)) {
 398                 size_t n = min(size, iov_iter_count(i));
 399                 n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n);
 400                 return size - n;
 401         } else if (iter_is_iovec(i)) {
 402                 size_t count = min(size, iov_iter_count(i));
 403                 const struct iovec *p;
 404                 size_t skip;
 405
 406                 size -= count;
 407                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 408                         size_t len = min(count, p->iov_len - skip);
 409                         size_t ret;
 410
 411                         if (unlikely(!len))
 412                                 continue;
 413                         ret = fault_in_safe_writeable(p->iov_base + skip, len);
 414                         count -= len - ret;
 415                         if (ret)
 416                                 break;
 417                 }
 418                 return count + size;
 419         }
 420         return 0;
 421 }
 422 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
 423
 424 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 425                         const struct iovec *iov, unsigned long nr_segs,
 426                         size_t count)
 427 {
 428         WARN_ON(direction & ~(READ | WRITE));
 429         *i = (struct iov_iter) {
 430                 .iter_type = ITER_IOVEC,
 431                 .nofault = false,
 432                 .user_backed = true,
 433                 .data_source = direction,
 434                 .iov = iov,
 435                 .nr_segs = nr_segs,
 436                 .iov_offset = 0,
 437                 .count = count
 438         };
 439 }
 440 EXPORT_SYMBOL(iov_iter_init);
 441
 442 // returns the offset in partial buffer (if any)
 443 static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages)
 444 {
 445         struct pipe_inode_info *pipe = i->pipe;
 446         int used = pipe->head - pipe->tail;
 447         int off = i->last_offset;
 448
 449         *npages = max((int)pipe->max_usage - used, 0);
 450
 451         if (off > 0 && off < PAGE_SIZE) { // anon and not full
 452                 (*npages)++;
 453                 return off;
 454         }
 455         return 0;
 456 }
 457
 458 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 459                                 struct iov_iter *i)
 460 {
 461         unsigned int off, chunk;
 462
 463         if (unlikely(bytes > i->count))
 464                 bytes = i->count;
 465         if (unlikely(!bytes))
 466                 return 0;
 467
 468         if (!sanity(i))
 469                 return 0;
 470
 471         for (size_t n = bytes; n; n -= chunk) {
 472                 struct page *page = append_pipe(i, n, &off);
 473                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 474                 if (!page)
 475                         return bytes - n;
 476                 memcpy_to_page(page, off, addr, chunk);
 477                 addr += chunk;
 478         }
 479         return bytes;
 480 }
 481
 482 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 483                               __wsum sum, size_t off)
 484 {
 485         __wsum next = csum_partial_copy_nocheck(from, to, len);
 486         return csum_block_add(sum, next, off);
 487 }
 488
 489 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 490                                          struct iov_iter *i, __wsum *sump)
 491 {
 492         __wsum sum = *sump;
 493         size_t off = 0;
 494         unsigned int chunk, r;
 495
 496         if (unlikely(bytes > i->count))
 497                 bytes = i->count;
 498         if (unlikely(!bytes))
 499                 return 0;
 500
 501         if (!sanity(i))
 502                 return 0;
 503
 504         while (bytes) {
 505                 struct page *page = append_pipe(i, bytes, &r);
 506                 char *p;
 507
 508                 if (!page)
 509                         break;
 510                 chunk = min_t(size_t, bytes, PAGE_SIZE - r);
 511                 p = kmap_local_page(page);
 512                 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
 513                 kunmap_local(p);
 514                 off += chunk;
 515                 bytes -= chunk;
 516         }
 517         *sump = sum;
 518         return off;
 519 }
 520
 521 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 522 {
 523         if (unlikely(iov_iter_is_pipe(i)))
 524                 return copy_pipe_to_iter(addr, bytes, i);
 525         if (user_backed_iter(i))
 526                 might_fault();
 527         iterate_and_advance(i, bytes, base, len, off,
 528                 copyout(base, addr + off, len),
 529                 memcpy(base, addr + off, len)
 530         )
 531
 532         return bytes;
 533 }
 534 EXPORT_SYMBOL(_copy_to_iter);
 535
 536 #ifdef CONFIG_ARCH_HAS_COPY_MC
 537 static int copyout_mc(void __user *to, const void *from, size_t n)
 538 {
 539         if (access_ok(to, n)) {
 540                 instrument_copy_to_user(to, from, n);
 541                 n = copy_mc_to_user((__force void *) to, from, n);
 542         }
 543         return n;
 544 }
 545
 546 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 547                                 struct iov_iter *i)
 548 {
 549         size_t xfer = 0;
 550         unsigned int off, chunk;
 551
 552         if (unlikely(bytes > i->count))
 553                 bytes = i->count;
 554         if (unlikely(!bytes))
 555                 return 0;
 556
 557         if (!sanity(i))
 558                 return 0;
 559
 560         while (bytes) {
 561                 struct page *page = append_pipe(i, bytes, &off);
 562                 unsigned long rem;
 563                 char *p;
 564
 565                 if (!page)
 566                         break;
 567                 chunk = min_t(size_t, bytes, PAGE_SIZE - off);
 568                 p = kmap_local_page(page);
 569                 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
 570                 chunk -= rem;
 571                 kunmap_local(p);
 572                 xfer += chunk;
 573                 bytes -= chunk;
 574                 if (rem) {
 575                         iov_iter_revert(i, rem);
 576                         break;
 577                 }
 578         }
 579         return xfer;
 580 }
 581
 582 /**
 583  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 584  * @addr: source kernel address
 585  * @bytes: total transfer length
 586  * @i: destination iterator
 587  *
 588  * The pmem driver deploys this for the dax operation
 589  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 590  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 591  * successfully copied.
 592  *
 593  * The main differences between this and typical _copy_to_iter().
 594  *
 595  * * Typical tail/residue handling after a fault retries the copy
 596  *   byte-by-byte until the fault happens again. Re-triggering machine
 597  *   checks is potentially fatal so the implementation uses source
 598  *   alignment and poison alignment assumptions to avoid re-triggering
 599  *   hardware exceptions.
 600  *
 601  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 602  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 603  *   a short copy.
 604  *
 605  * Return: number of bytes copied (may be %0)
 606  */
 607 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 608 {
 609         if (unlikely(iov_iter_is_pipe(i)))
 610                 return copy_mc_pipe_to_iter(addr, bytes, i);
 611         if (user_backed_iter(i))
 612                 might_fault();
 613         __iterate_and_advance(i, bytes, base, len, off,
 614                 copyout_mc(base, addr + off, len),
 615                 copy_mc_to_kernel(base, addr + off, len)
 616         )
 617
 618         return bytes;
 619 }
 620 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 621 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 622
 623 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 624 {
 625         if (unlikely(iov_iter_is_pipe(i))) {
 626                 WARN_ON(1);
 627                 return 0;
 628         }
 629         if (user_backed_iter(i))
 630                 might_fault();
 631         iterate_and_advance(i, bytes, base, len, off,
 632                 copyin(addr + off, base, len),
 633                 memcpy(addr + off, base, len)
 634         )
 635
 636         return bytes;
 637 }
 638 EXPORT_SYMBOL(_copy_from_iter);
 639
 640 size_t _copy_from_iter_nocache(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_inatomic_nocache(addr + off, base, len),
 648                 memcpy(addr + off, base, len)
 649         )
 650
 651         return bytes;
 652 }
 653 EXPORT_SYMBOL(_copy_from_iter_nocache);
 654
 655 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 656 /**
 657  * _copy_from_iter_flushcache - write destination through cpu cache
 658  * @addr: destination kernel address
 659  * @bytes: total transfer length
 660  * @i: source iterator
 661  *
 662  * The pmem driver arranges for filesystem-dax to use this facility via
 663  * dax_copy_from_iter() for ensuring that writes to persistent memory
 664  * are flushed through the CPU cache. It is differentiated from
 665  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 666  * all iterator types. The _copy_from_iter_nocache() only attempts to
 667  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 668  * instructions that strand dirty-data in the cache.
 669  *
 670  * Return: number of bytes copied (may be %0)
 671  */
 672 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 673 {
 674         if (unlikely(iov_iter_is_pipe(i))) {
 675                 WARN_ON(1);
 676                 return 0;
 677         }
 678         iterate_and_advance(i, bytes, base, len, off,
 679                 __copy_from_user_flushcache(addr + off, base, len),
 680                 memcpy_flushcache(addr + off, base, len)
 681         )
 682
 683         return bytes;
 684 }
 685 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 686 #endif
 687
 688 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 689 {
 690         struct page *head;
 691         size_t v = n + offset;
 692
 693         /*
 694          * The general case needs to access the page order in order
 695          * to compute the page size.
 696          * However, we mostly deal with order-0 pages and thus can
 697          * avoid a possible cache line miss for requests that fit all
 698          * page orders.
 699          */
 700         if (n <= v && v <= PAGE_SIZE)
 701                 return true;
 702
 703         head = compound_head(page);
 704         v += (page - head) << PAGE_SHIFT;
 705
 706         if (likely(n <= v && v <= (page_size(head))))
 707                 return true;
 708         WARN_ON(1);
 709         return false;
 710 }
 711
 712 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 713                          struct iov_iter *i)
 714 {
 715         size_t res = 0;
 716         if (unlikely(!page_copy_sane(page, offset, bytes)))
 717                 return 0;
 718         if (unlikely(iov_iter_is_pipe(i)))
 719                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 720         page += offset / PAGE_SIZE; // first subpage
 721         offset %= PAGE_SIZE;
 722         while (1) {
 723                 void *kaddr = kmap_local_page(page);
 724                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 725                 n = _copy_to_iter(kaddr + offset, n, i);
 726                 kunmap_local(kaddr);
 727                 res += n;
 728                 bytes -= n;
 729                 if (!bytes || !n)
 730                         break;
 731                 offset += n;
 732                 if (offset == PAGE_SIZE) {
 733                         page++;
 734                         offset = 0;
 735                 }
 736         }
 737         return res;
 738 }
 739 EXPORT_SYMBOL(copy_page_to_iter);
 740
 741 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 742                          struct iov_iter *i)
 743 {
 744         size_t res = 0;
 745         if (!page_copy_sane(page, offset, bytes))
 746                 return 0;
 747         page += offset / PAGE_SIZE; // first subpage
 748         offset %= PAGE_SIZE;
 749         while (1) {
 750                 void *kaddr = kmap_local_page(page);
 751                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 752                 n = _copy_from_iter(kaddr + offset, n, i);
 753                 kunmap_local(kaddr);
 754                 res += n;
 755                 bytes -= n;
 756                 if (!bytes || !n)
 757                         break;
 758                 offset += n;
 759                 if (offset == PAGE_SIZE) {
 760                         page++;
 761                         offset = 0;
 762                 }
 763         }
 764         return res;
 765 }
 766 EXPORT_SYMBOL(copy_page_from_iter);
 767
 768 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 769 {
 770         unsigned int chunk, off;
 771
 772         if (unlikely(bytes > i->count))
 773                 bytes = i->count;
 774         if (unlikely(!bytes))
 775                 return 0;
 776
 777         if (!sanity(i))
 778                 return 0;
 779
 780         for (size_t n = bytes; n; n -= chunk) {
 781                 struct page *page = append_pipe(i, n, &off);
 782                 char *p;
 783
 784                 if (!page)
 785                         return bytes - n;
 786                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 787                 p = kmap_local_page(page);
 788                 memset(p + off, 0, chunk);
 789                 kunmap_local(p);
 790         }
 791         return bytes;
 792 }
 793
 794 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 795 {
 796         if (unlikely(iov_iter_is_pipe(i)))
 797                 return pipe_zero(bytes, i);
 798         iterate_and_advance(i, bytes, base, len, count,
 799                 clear_user(base, len),
 800                 memset(base, 0, len)
 801         )
 802
 803         return bytes;
 804 }
 805 EXPORT_SYMBOL(iov_iter_zero);
 806
 807 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 808                                   struct iov_iter *i)
 809 {
 810         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 811         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 812                 kunmap_atomic(kaddr);
 813                 return 0;
 814         }
 815         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 816                 kunmap_atomic(kaddr);
 817                 WARN_ON(1);
 818                 return 0;
 819         }
 820         iterate_and_advance(i, bytes, base, len, off,
 821                 copyin(p + off, base, len),
 822                 memcpy(p + off, base, len)
 823         )
 824         kunmap_atomic(kaddr);
 825         return bytes;
 826 }
 827 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 828
 829 static void pipe_advance(struct iov_iter *i, size_t size)
 830 {
 831         struct pipe_inode_info *pipe = i->pipe;
 832         int off = i->last_offset;
 833
 834         if (!off && !size) {
 835                 pipe_discard_from(pipe, i->start_head); // discard everything
 836                 return;
 837         }
 838         i->count -= size;
 839         while (1) {
 840                 struct pipe_buffer *buf = pipe_buf(pipe, i->head);
 841                 if (off) /* make it relative to the beginning of buffer */
 842                         size += abs(off) - buf->offset;
 843                 if (size <= buf->len) {
 844                         buf->len = size;
 845                         i->last_offset = last_offset(buf);
 846                         break;
 847                 }
 848                 size -= buf->len;
 849                 i->head++;
 850                 off = 0;
 851         }
 852         pipe_discard_from(pipe, i->head + 1); // discard everything past this one
 853 }
 854
 855 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 856 {
 857         const struct bio_vec *bvec, *end;
 858
 859         if (!i->count)
 860                 return;
 861         i->count -= size;
 862
 863         size += i->iov_offset;
 864
 865         for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
 866                 if (likely(size < bvec->bv_len))
 867                         break;
 868                 size -= bvec->bv_len;
 869         }
 870         i->iov_offset = size;
 871         i->nr_segs -= bvec - i->bvec;
 872         i->bvec = bvec;
 873 }
 874
 875 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 876 {
 877         const struct iovec *iov, *end;
 878
 879         if (!i->count)
 880                 return;
 881         i->count -= size;
 882
 883         size += i->iov_offset; // from beginning of current segment
 884         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
 885                 if (likely(size < iov->iov_len))
 886                         break;
 887                 size -= iov->iov_len;
 888         }
 889         i->iov_offset = size;
 890         i->nr_segs -= iov - i->iov;
 891         i->iov = iov;
 892 }
 893
 894 void iov_iter_advance(struct iov_iter *i, size_t size)
 895 {
 896         if (unlikely(i->count < size))
 897                 size = i->count;
 898         if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
 899                 i->iov_offset += size;
 900                 i->count -= size;
 901         } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 902                 /* iovec and kvec have identical layouts */
 903                 iov_iter_iovec_advance(i, size);
 904         } else if (iov_iter_is_bvec(i)) {
 905                 iov_iter_bvec_advance(i, size);
 906         } else if (iov_iter_is_pipe(i)) {
 907                 pipe_advance(i, size);
 908         } else if (iov_iter_is_discard(i)) {
 909                 i->count -= size;
 910         }
 911 }
 912 EXPORT_SYMBOL(iov_iter_advance);
 913
 914 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 915 {
 916         if (!unroll)
 917                 return;
 918         if (WARN_ON(unroll > MAX_RW_COUNT))
 919                 return;
 920         i->count += unroll;
 921         if (unlikely(iov_iter_is_pipe(i))) {
 922                 struct pipe_inode_info *pipe = i->pipe;
 923                 unsigned int head = pipe->head;
 924
 925                 while (head > i->start_head) {
 926                         struct pipe_buffer *b = pipe_buf(pipe, --head);
 927                         if (unroll < b->len) {
 928                                 b->len -= unroll;
 929                                 i->last_offset = last_offset(b);
 930                                 i->head = head;
 931                                 return;
 932                         }
 933                         unroll -= b->len;
 934                         pipe_buf_release(pipe, b);
 935                         pipe->head--;
 936                 }
 937                 i->last_offset = 0;
 938                 i->head = head;
 939                 return;
 940         }
 941         if (unlikely(iov_iter_is_discard(i)))
 942                 return;
 943         if (unroll <= i->iov_offset) {
 944                 i->iov_offset -= unroll;
 945                 return;
 946         }
 947         unroll -= i->iov_offset;
 948         if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
 949                 BUG(); /* We should never go beyond the start of the specified
 950                         * range since we might then be straying into pages that
 951                         * aren't pinned.
 952                         */
 953         } else if (iov_iter_is_bvec(i)) {
 954                 const struct bio_vec *bvec = i->bvec;
 955                 while (1) {
 956                         size_t n = (--bvec)->bv_len;
 957                         i->nr_segs++;
 958                         if (unroll <= n) {
 959                                 i->bvec = bvec;
 960                                 i->iov_offset = n - unroll;
 961                                 return;
 962                         }
 963                         unroll -= n;
 964                 }
 965         } else { /* same logics for iovec and kvec */
 966                 const struct iovec *iov = i->iov;
 967                 while (1) {
 968                         size_t n = (--iov)->iov_len;
 969                         i->nr_segs++;
 970                         if (unroll <= n) {
 971                                 i->iov = iov;
 972                                 i->iov_offset = n - unroll;
 973                                 return;
 974                         }
 975                         unroll -= n;
 976                 }
 977         }
 978 }
 979 EXPORT_SYMBOL(iov_iter_revert);
 980
 981 /*
 982  * Return the count of just the current iov_iter segment.
 983  */
 984 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 985 {
 986         if (i->nr_segs > 1) {
 987                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 988                         return min(i->count, i->iov->iov_len - i->iov_offset);
 989                 if (iov_iter_is_bvec(i))
 990                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 991         }
 992         return i->count;
 993 }
 994 EXPORT_SYMBOL(iov_iter_single_seg_count);
 995
 996 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
 997                         const struct kvec *kvec, unsigned long nr_segs,
 998                         size_t count)
 999 {
1000         WARN_ON(direction & ~(READ | WRITE));
1001         *i = (struct iov_iter){
1002                 .iter_type = ITER_KVEC,
1003                 .data_source = direction,
1004                 .kvec = kvec,
1005                 .nr_segs = nr_segs,
1006                 .iov_offset = 0,
1007                 .count = count
1008         };
1009 }
1010 EXPORT_SYMBOL(iov_iter_kvec);
1011
1012 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1013                         const struct bio_vec *bvec, unsigned long nr_segs,
1014                         size_t count)
1015 {
1016         WARN_ON(direction & ~(READ | WRITE));
1017         *i = (struct iov_iter){
1018                 .iter_type = ITER_BVEC,
1019                 .data_source = direction,
1020                 .bvec = bvec,
1021                 .nr_segs = nr_segs,
1022                 .iov_offset = 0,
1023                 .count = count
1024         };
1025 }
1026 EXPORT_SYMBOL(iov_iter_bvec);
1027
1028 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1029                         struct pipe_inode_info *pipe,
1030                         size_t count)
1031 {
1032         BUG_ON(direction != READ);
1033         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1034         *i = (struct iov_iter){
1035                 .iter_type = ITER_PIPE,
1036                 .data_source = false,
1037                 .pipe = pipe,
1038                 .head = pipe->head,
1039                 .start_head = pipe->head,
1040                 .last_offset = 0,
1041                 .count = count
1042         };
1043 }
1044 EXPORT_SYMBOL(iov_iter_pipe);
1045
1046 /**
1047  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1048  * @i: The iterator to initialise.
1049  * @direction: The direction of the transfer.
1050  * @xarray: The xarray to access.
1051  * @start: The start file position.
1052  * @count: The size of the I/O buffer in bytes.
1053  *
1054  * Set up an I/O iterator to either draw data out of the pages attached to an
1055  * inode or to inject data into those pages.  The pages *must* be prevented
1056  * from evaporation, either by taking a ref on them or locking them by the
1057  * caller.
1058  */
1059 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1060                      struct xarray *xarray, loff_t start, size_t count)
1061 {
1062         BUG_ON(direction & ~1);
1063         *i = (struct iov_iter) {
1064                 .iter_type = ITER_XARRAY,
1065                 .data_source = direction,
1066                 .xarray = xarray,
1067                 .xarray_start = start,
1068                 .count = count,
1069                 .iov_offset = 0
1070         };
1071 }
1072 EXPORT_SYMBOL(iov_iter_xarray);
1073
1074 /**
1075  * iov_iter_discard - Initialise an I/O iterator that discards data
1076  * @i: The iterator to initialise.
1077  * @direction: The direction of the transfer.
1078  * @count: The size of the I/O buffer in bytes.
1079  *
1080  * Set up an I/O iterator that just discards everything that's written to it.
1081  * It's only available as a READ iterator.
1082  */
1083 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1084 {
1085         BUG_ON(direction != READ);
1086         *i = (struct iov_iter){
1087                 .iter_type = ITER_DISCARD,
1088                 .data_source = false,
1089                 .count = count,
1090                 .iov_offset = 0
1091         };
1092 }
1093 EXPORT_SYMBOL(iov_iter_discard);
1094
1095 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1096                                    unsigned len_mask)
1097 {
1098         size_t size = i->count;
1099         size_t skip = i->iov_offset;
1100         unsigned k;
1101
1102         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1103                 size_t len = i->iov[k].iov_len - skip;
1104
1105                 if (len > size)
1106                         len = size;
1107                 if (len & len_mask)
1108                         return false;
1109                 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1110                         return false;
1111
1112                 size -= len;
1113                 if (!size)
1114                         break;
1115         }
1116         return true;
1117 }
1118
1119 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1120                                   unsigned len_mask)
1121 {
1122         size_t size = i->count;
1123         unsigned skip = i->iov_offset;
1124         unsigned k;
1125
1126         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1127                 size_t len = i->bvec[k].bv_len - skip;
1128
1129                 if (len > size)
1130                         len = size;
1131                 if (len & len_mask)
1132                         return false;
1133                 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1134                         return false;
1135
1136                 size -= len;
1137                 if (!size)
1138                         break;
1139         }
1140         return true;
1141 }
1142
1143 /**
1144  * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1145  *      are aligned to the parameters.
1146  *
1147  * @i: &struct iov_iter to restore
1148  * @addr_mask: bit mask to check against the iov element's addresses
1149  * @len_mask: bit mask to check against the iov element's lengths
1150  *
1151  * Return: false if any addresses or lengths intersect with the provided masks
1152  */
1153 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1154                          unsigned len_mask)
1155 {
1156         if (likely(iter_is_ubuf(i))) {
1157                 if (i->count & len_mask)
1158                         return false;
1159                 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
1160                         return false;
1161                 return true;
1162         }
1163
1164         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1165                 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1166
1167         if (iov_iter_is_bvec(i))
1168                 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1169
1170         if (iov_iter_is_pipe(i)) {
1171                 size_t size = i->count;
1172
1173                 if (size & len_mask)
1174                         return false;
1175                 if (size && i->last_offset > 0) {
1176                         if (i->last_offset & addr_mask)
1177                                 return false;
1178                 }
1179
1180                 return true;
1181         }
1182
1183         if (iov_iter_is_xarray(i)) {
1184                 if (i->count & len_mask)
1185                         return false;
1186                 if ((i->xarray_start + i->iov_offset) & addr_mask)
1187                         return false;
1188         }
1189
1190         return true;
1191 }
1192 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1193
1194 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1195 {
1196         unsigned long res = 0;
1197         size_t size = i->count;
1198         size_t skip = i->iov_offset;
1199         unsigned k;
1200
1201         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1202                 size_t len = i->iov[k].iov_len - skip;
1203                 if (len) {
1204                         res |= (unsigned long)i->iov[k].iov_base + skip;
1205                         if (len > size)
1206                                 len = size;
1207                         res |= len;
1208                         size -= len;
1209                         if (!size)
1210                                 break;
1211                 }
1212         }
1213         return res;
1214 }
1215
1216 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1217 {
1218         unsigned res = 0;
1219         size_t size = i->count;
1220         unsigned skip = i->iov_offset;
1221         unsigned k;
1222
1223         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1224                 size_t len = i->bvec[k].bv_len - skip;
1225                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1226                 if (len > size)
1227                         len = size;
1228                 res |= len;
1229                 size -= len;
1230                 if (!size)
1231                         break;
1232         }
1233         return res;
1234 }
1235
1236 unsigned long iov_iter_alignment(const struct iov_iter *i)
1237 {
1238         if (likely(iter_is_ubuf(i))) {
1239                 size_t size = i->count;
1240                 if (size)
1241                         return ((unsigned long)i->ubuf + i->iov_offset) | size;
1242                 return 0;
1243         }
1244
1245         /* iovec and kvec have identical layouts */
1246         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1247                 return iov_iter_alignment_iovec(i);
1248
1249         if (iov_iter_is_bvec(i))
1250                 return iov_iter_alignment_bvec(i);
1251
1252         if (iov_iter_is_pipe(i)) {
1253                 size_t size = i->count;
1254
1255                 if (size && i->last_offset > 0)
1256                         return size | i->last_offset;
1257                 return size;
1258         }
1259
1260         if (iov_iter_is_xarray(i))
1261                 return (i->xarray_start + i->iov_offset) | i->count;
1262
1263         return 0;
1264 }
1265 EXPORT_SYMBOL(iov_iter_alignment);
1266
1267 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1268 {
1269         unsigned long res = 0;
1270         unsigned long v = 0;
1271         size_t size = i->count;
1272         unsigned k;
1273
1274         if (iter_is_ubuf(i))
1275                 return 0;
1276
1277         if (WARN_ON(!iter_is_iovec(i)))
1278                 return ~0U;
1279
1280         for (k = 0; k < i->nr_segs; k++) {
1281                 if (i->iov[k].iov_len) {
1282                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1283                         if (v) // if not the first one
1284                                 res |= base | v; // this start | previous end
1285                         v = base + i->iov[k].iov_len;
1286                         if (size <= i->iov[k].iov_len)
1287                                 break;
1288                         size -= i->iov[k].iov_len;
1289                 }
1290         }
1291         return res;
1292 }
1293 EXPORT_SYMBOL(iov_iter_gap_alignment);
1294
1295 static int want_pages_array(struct page ***res, size_t size,
1296                             size_t start, unsigned int maxpages)
1297 {
1298         unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
1299
1300         if (count > maxpages)
1301                 count = maxpages;
1302         WARN_ON(!count);        // caller should've prevented that
1303         if (!*res) {
1304                 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
1305                 if (!*res)
1306                         return 0;
1307         }
1308         return count;
1309 }
1310
1311 static ssize_t pipe_get_pages(struct iov_iter *i,
1312                    struct page ***pages, size_t maxsize, unsigned maxpages,
1313                    size_t *start)
1314 {
1315         unsigned int npages, count, off, chunk;
1316         struct page **p;
1317         size_t left;
1318
1319         if (!sanity(i))
1320                 return -EFAULT;
1321
1322         *start = off = pipe_npages(i, &npages);
1323         if (!npages)
1324                 return -EFAULT;
1325         count = want_pages_array(pages, maxsize, off, min(npages, maxpages));
1326         if (!count)
1327                 return -ENOMEM;
1328         p = *pages;
1329         for (npages = 0, left = maxsize ; npages < count; npages++, left -= chunk) {
1330                 struct page *page = append_pipe(i, left, &off);
1331                 if (!page)
1332                         break;
1333                 chunk = min_t(size_t, left, PAGE_SIZE - off);
1334                 get_page(*p++ = page);
1335         }
1336         if (!npages)
1337                 return -EFAULT;
1338         return maxsize - left;
1339 }
1340
1341 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1342                                           pgoff_t index, unsigned int nr_pages)
1343 {
1344         XA_STATE(xas, xa, index);
1345         struct page *page;
1346         unsigned int ret = 0;
1347
1348         rcu_read_lock();
1349         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1350                 if (xas_retry(&xas, page))
1351                         continue;
1352
1353                 /* Has the page moved or been split? */
1354                 if (unlikely(page != xas_reload(&xas))) {
1355                         xas_reset(&xas);
1356                         continue;
1357                 }
1358
1359                 pages[ret] = find_subpage(page, xas.xa_index);
1360                 get_page(pages[ret]);
1361                 if (++ret == nr_pages)
1362                         break;
1363         }
1364         rcu_read_unlock();
1365         return ret;
1366 }
1367
1368 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1369                                      struct page ***pages, size_t maxsize,
1370                                      unsigned maxpages, size_t *_start_offset)
1371 {
1372         unsigned nr, offset, count;
1373         pgoff_t index;
1374         loff_t pos;
1375
1376         pos = i->xarray_start + i->iov_offset;
1377         index = pos >> PAGE_SHIFT;
1378         offset = pos & ~PAGE_MASK;
1379         *_start_offset = offset;
1380
1381         count = want_pages_array(pages, maxsize, offset, maxpages);
1382         if (!count)
1383                 return -ENOMEM;
1384         nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
1385         if (nr == 0)
1386                 return 0;
1387
1388         maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1389         i->iov_offset += maxsize;
1390         i->count -= maxsize;
1391         return maxsize;
1392 }
1393
1394 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1395 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1396 {
1397         size_t skip;
1398         long k;
1399
1400         if (iter_is_ubuf(i))
1401                 return (unsigned long)i->ubuf + i->iov_offset;
1402
1403         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1404                 size_t len = i->iov[k].iov_len - skip;
1405
1406                 if (unlikely(!len))
1407                         continue;
1408                 if (*size > len)
1409                         *size = len;
1410                 return (unsigned long)i->iov[k].iov_base + skip;
1411         }
1412         BUG(); // if it had been empty, we wouldn't get called
1413 }
1414
1415 /* must be done on non-empty ITER_BVEC one */
1416 static struct page *first_bvec_segment(const struct iov_iter *i,
1417                                        size_t *size, size_t *start)
1418 {
1419         struct page *page;
1420         size_t skip = i->iov_offset, len;
1421
1422         len = i->bvec->bv_len - skip;
1423         if (*size > len)
1424                 *size = len;
1425         skip += i->bvec->bv_offset;
1426         page = i->bvec->bv_page + skip / PAGE_SIZE;
1427         *start = skip % PAGE_SIZE;
1428         return page;
1429 }
1430
1431 static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
1432                    struct page ***pages, size_t maxsize,
1433                    unsigned int maxpages, size_t *start)
1434 {
1435         unsigned int n;
1436
1437         if (maxsize > i->count)
1438                 maxsize = i->count;
1439         if (!maxsize)
1440                 return 0;
1441         if (maxsize > MAX_RW_COUNT)
1442                 maxsize = MAX_RW_COUNT;
1443
1444         if (likely(user_backed_iter(i))) {
1445                 unsigned int gup_flags = 0;
1446                 unsigned long addr;
1447                 int res;
1448
1449                 if (iov_iter_rw(i) != WRITE)
1450                         gup_flags |= FOLL_WRITE;
1451                 if (i->nofault)
1452                         gup_flags |= FOLL_NOFAULT;
1453
1454                 addr = first_iovec_segment(i, &maxsize);
1455                 *start = addr % PAGE_SIZE;
1456                 addr &= PAGE_MASK;
1457                 n = want_pages_array(pages, maxsize, *start, maxpages);
1458                 if (!n)
1459                         return -ENOMEM;
1460                 res = get_user_pages_fast(addr, n, gup_flags, *pages);
1461                 if (unlikely(res <= 0))
1462                         return res;
1463                 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1464                 iov_iter_advance(i, maxsize);
1465                 return maxsize;
1466         }
1467         if (iov_iter_is_bvec(i)) {
1468                 struct page **p;
1469                 struct page *page;
1470
1471                 page = first_bvec_segment(i, &maxsize, start);
1472                 n = want_pages_array(pages, maxsize, *start, maxpages);
1473                 if (!n)
1474                         return -ENOMEM;
1475                 p = *pages;
1476                 for (int k = 0; k < n; k++)
1477                         get_page(p[k] = page + k);
1478                 maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1479                 i->count -= maxsize;
1480                 i->iov_offset += maxsize;
1481                 if (i->iov_offset == i->bvec->bv_len) {
1482                         i->iov_offset = 0;
1483                         i->bvec++;
1484                         i->nr_segs--;
1485                 }
1486                 return maxsize;
1487         }
1488         if (iov_iter_is_pipe(i))
1489                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1490         if (iov_iter_is_xarray(i))
1491                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1492         return -EFAULT;
1493 }
1494
1495 ssize_t iov_iter_get_pages2(struct iov_iter *i,
1496                    struct page **pages, size_t maxsize, unsigned maxpages,
1497                    size_t *start)
1498 {
1499         if (!maxpages)
1500                 return 0;
1501         BUG_ON(!pages);
1502
1503         return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, start);
1504 }
1505 EXPORT_SYMBOL(iov_iter_get_pages2);
1506
1507 ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1508                    struct page ***pages, size_t maxsize,
1509                    size_t *start)
1510 {
1511         ssize_t len;
1512
1513         *pages = NULL;
1514
1515         len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start);
1516         if (len <= 0) {
1517                 kvfree(*pages);
1518                 *pages = NULL;
1519         }
1520         return len;
1521 }
1522 EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1523
1524 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1525                                struct iov_iter *i)
1526 {
1527         __wsum sum, next;
1528         sum = *csum;
1529         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1530                 WARN_ON(1);
1531                 return 0;
1532         }
1533         iterate_and_advance(i, bytes, base, len, off, ({
1534                 next = csum_and_copy_from_user(base, addr + off, len);
1535                 sum = csum_block_add(sum, next, off);
1536                 next ? 0 : len;
1537         }), ({
1538                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1539         })
1540         )
1541         *csum = sum;
1542         return bytes;
1543 }
1544 EXPORT_SYMBOL(csum_and_copy_from_iter);
1545
1546 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1547                              struct iov_iter *i)
1548 {
1549         struct csum_state *csstate = _csstate;
1550         __wsum sum, next;
1551
1552         if (unlikely(iov_iter_is_discard(i))) {
1553                 WARN_ON(1);     /* for now */
1554                 return 0;
1555         }
1556
1557         sum = csum_shift(csstate->csum, csstate->off);
1558         if (unlikely(iov_iter_is_pipe(i)))
1559                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1560         else iterate_and_advance(i, bytes, base, len, off, ({
1561                 next = csum_and_copy_to_user(addr + off, base, len);
1562                 sum = csum_block_add(sum, next, off);
1563                 next ? 0 : len;
1564         }), ({
1565                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1566         })
1567         )
1568         csstate->csum = csum_shift(sum, csstate->off);
1569         csstate->off += bytes;
1570         return bytes;
1571 }
1572 EXPORT_SYMBOL(csum_and_copy_to_iter);
1573
1574 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1575                 struct iov_iter *i)
1576 {
1577 #ifdef CONFIG_CRYPTO_HASH
1578         struct ahash_request *hash = hashp;
1579         struct scatterlist sg;
1580         size_t copied;
1581
1582         copied = copy_to_iter(addr, bytes, i);
1583         sg_init_one(&sg, addr, copied);
1584         ahash_request_set_crypt(hash, &sg, NULL, copied);
1585         crypto_ahash_update(hash);
1586         return copied;
1587 #else
1588         return 0;
1589 #endif
1590 }
1591 EXPORT_SYMBOL(hash_and_copy_to_iter);
1592
1593 static int iov_npages(const struct iov_iter *i, int maxpages)
1594 {
1595         size_t skip = i->iov_offset, size = i->count;
1596         const struct iovec *p;
1597         int npages = 0;
1598
1599         for (p = i->iov; size; skip = 0, p++) {
1600                 unsigned offs = offset_in_page(p->iov_base + skip);
1601                 size_t len = min(p->iov_len - skip, size);
1602
1603                 if (len) {
1604                         size -= len;
1605                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1606                         if (unlikely(npages > maxpages))
1607                                 return maxpages;
1608                 }
1609         }
1610         return npages;
1611 }
1612
1613 static int bvec_npages(const struct iov_iter *i, int maxpages)
1614 {
1615         size_t skip = i->iov_offset, size = i->count;
1616         const struct bio_vec *p;
1617         int npages = 0;
1618
1619         for (p = i->bvec; size; skip = 0, p++) {
1620                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1621                 size_t len = min(p->bv_len - skip, size);
1622
1623                 size -= len;
1624                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1625                 if (unlikely(npages > maxpages))
1626                         return maxpages;
1627         }
1628         return npages;
1629 }
1630
1631 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1632 {
1633         if (unlikely(!i->count))
1634                 return 0;
1635         if (likely(iter_is_ubuf(i))) {
1636                 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1637                 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1638                 return min(npages, maxpages);
1639         }
1640         /* iovec and kvec have identical layouts */
1641         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1642                 return iov_npages(i, maxpages);
1643         if (iov_iter_is_bvec(i))
1644                 return bvec_npages(i, maxpages);
1645         if (iov_iter_is_pipe(i)) {
1646                 int npages;
1647
1648                 if (!sanity(i))
1649                         return 0;
1650
1651                 pipe_npages(i, &npages);
1652                 return min(npages, maxpages);
1653         }
1654         if (iov_iter_is_xarray(i)) {
1655                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1656                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1657                 return min(npages, maxpages);
1658         }
1659         return 0;
1660 }
1661 EXPORT_SYMBOL(iov_iter_npages);
1662
1663 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1664 {
1665         *new = *old;
1666         if (unlikely(iov_iter_is_pipe(new))) {
1667                 WARN_ON(1);
1668                 return NULL;
1669         }
1670         if (iov_iter_is_bvec(new))
1671                 return new->bvec = kmemdup(new->bvec,
1672                                     new->nr_segs * sizeof(struct bio_vec),
1673                                     flags);
1674         else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1675                 /* iovec and kvec have identical layout */
1676                 return new->iov = kmemdup(new->iov,
1677                                    new->nr_segs * sizeof(struct iovec),
1678                                    flags);
1679         return NULL;
1680 }
1681 EXPORT_SYMBOL(dup_iter);
1682
1683 static int copy_compat_iovec_from_user(struct iovec *iov,
1684                 const struct iovec __user *uvec, unsigned long nr_segs)
1685 {
1686         const struct compat_iovec __user *uiov =
1687                 (const struct compat_iovec __user *)uvec;
1688         int ret = -EFAULT, i;
1689
1690         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1691                 return -EFAULT;
1692
1693         for (i = 0; i < nr_segs; i++) {
1694                 compat_uptr_t buf;
1695                 compat_ssize_t len;
1696
1697                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1698                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1699
1700                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1701                 if (len < 0) {
1702                         ret = -EINVAL;
1703                         goto uaccess_end;
1704                 }
1705                 iov[i].iov_base = compat_ptr(buf);
1706                 iov[i].iov_len = len;
1707         }
1708
1709         ret = 0;
1710 uaccess_end:
1711         user_access_end();
1712         return ret;
1713 }
1714
1715 static int copy_iovec_from_user(struct iovec *iov,
1716                 const struct iovec __user *uvec, unsigned long nr_segs)
1717 {
1718         unsigned long seg;
1719
1720         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1721                 return -EFAULT;
1722         for (seg = 0; seg < nr_segs; seg++) {
1723                 if ((ssize_t)iov[seg].iov_len < 0)
1724                         return -EINVAL;
1725         }
1726
1727         return 0;
1728 }
1729
1730 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1731                 unsigned long nr_segs, unsigned long fast_segs,
1732                 struct iovec *fast_iov, bool compat)
1733 {
1734         struct iovec *iov = fast_iov;
1735         int ret;
1736
1737         /*
1738          * SuS says "The readv() function *may* fail if the iovcnt argument was
1739          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1740          * traditionally returned zero for zero segments, so...
1741          */
1742         if (nr_segs == 0)
1743                 return iov;
1744         if (nr_segs > UIO_MAXIOV)
1745                 return ERR_PTR(-EINVAL);
1746         if (nr_segs > fast_segs) {
1747                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1748                 if (!iov)
1749                         return ERR_PTR(-ENOMEM);
1750         }
1751
1752         if (compat)
1753                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1754         else
1755                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1756         if (ret) {
1757                 if (iov != fast_iov)
1758                         kfree(iov);
1759                 return ERR_PTR(ret);
1760         }
1761
1762         return iov;
1763 }
1764
1765 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1766                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1767                  struct iov_iter *i, bool compat)
1768 {
1769         ssize_t total_len = 0;
1770         unsigned long seg;
1771         struct iovec *iov;
1772
1773         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1774         if (IS_ERR(iov)) {
1775                 *iovp = NULL;
1776                 return PTR_ERR(iov);
1777         }
1778
1779         /*
1780          * According to the Single Unix Specification we should return EINVAL if
1781          * an element length is < 0 when cast to ssize_t or if the total length
1782          * would overflow the ssize_t return value of the system call.
1783          *
1784          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1785          * overflow case.
1786          */
1787         for (seg = 0; seg < nr_segs; seg++) {
1788                 ssize_t len = (ssize_t)iov[seg].iov_len;
1789
1790                 if (!access_ok(iov[seg].iov_base, len)) {
1791                         if (iov != *iovp)
1792                                 kfree(iov);
1793                         *iovp = NULL;
1794                         return -EFAULT;
1795                 }
1796
1797                 if (len > MAX_RW_COUNT - total_len) {
1798                         len = MAX_RW_COUNT - total_len;
1799                         iov[seg].iov_len = len;
1800                 }
1801                 total_len += len;
1802         }
1803
1804         iov_iter_init(i, type, iov, nr_segs, total_len);
1805         if (iov == *iovp)
1806                 *iovp = NULL;
1807         else
1808                 *iovp = iov;
1809         return total_len;
1810 }
1811
1812 /**
1813  * import_iovec() - Copy an array of &struct iovec from userspace
1814  *     into the kernel, check that it is valid, and initialize a new
1815  *     &struct iov_iter iterator to access it.
1816  *
1817  * @type: One of %READ or %WRITE.
1818  * @uvec: Pointer to the userspace array.
1819  * @nr_segs: Number of elements in userspace array.
1820  * @fast_segs: Number of elements in @iov.
1821  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1822  *     on-stack) kernel array.
1823  * @i: Pointer to iterator that will be initialized on success.
1824  *
1825  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1826  * then this function places %NULL in *@iov on return. Otherwise, a new
1827  * array will be allocated and the result placed in *@iov. This means that
1828  * the caller may call kfree() on *@iov regardless of whether the small
1829  * on-stack array was used or not (and regardless of whether this function
1830  * returns an error or not).
1831  *
1832  * Return: Negative error code on error, bytes imported on success
1833  */
1834 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1835                  unsigned nr_segs, unsigned fast_segs,
1836                  struct iovec **iovp, struct iov_iter *i)
1837 {
1838         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1839                               in_compat_syscall());
1840 }
1841 EXPORT_SYMBOL(import_iovec);
1842
1843 int import_single_range(int rw, void __user *buf, size_t len,
1844                  struct iovec *iov, struct iov_iter *i)
1845 {
1846         if (len > MAX_RW_COUNT)
1847                 len = MAX_RW_COUNT;
1848         if (unlikely(!access_ok(buf, len)))
1849                 return -EFAULT;
1850
1851         iov->iov_base = buf;
1852         iov->iov_len = len;
1853         iov_iter_init(i, rw, iov, 1, len);
1854         return 0;
1855 }
1856 EXPORT_SYMBOL(import_single_range);
1857
1858 /**
1859  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1860  *     iov_iter_save_state() was called.
1861  *
1862  * @i: &struct iov_iter to restore
1863  * @state: state to restore from
1864  *
1865  * Used after iov_iter_save_state() to bring restore @i, if operations may
1866  * have advanced it.
1867  *
1868  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1869  */
1870 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1871 {
1872         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
1873                          !iov_iter_is_kvec(i) && !iter_is_ubuf(i))
1874                 return;
1875         i->iov_offset = state->iov_offset;
1876         i->count = state->count;
1877         if (iter_is_ubuf(i))
1878                 return;
1879         /*
1880          * For the *vec iters, nr_segs + iov is constant - if we increment
1881          * the vec, then we also decrement the nr_segs count. Hence we don't
1882          * need to track both of these, just one is enough and we can deduct
1883          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1884          * size, so we can just increment the iov pointer as they are unionzed.
1885          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1886          * not. Be safe and handle it separately.
1887          */
1888         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1889         if (iov_iter_is_bvec(i))
1890                 i->bvec -= state->nr_segs - i->nr_segs;
1891         else
1892                 i->iov -= state->nr_segs - i->nr_segs;
1893         i->nr_segs = state->nr_segs;
1894 }