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 #ifdef PIPE_PARANOIA
 190 static bool sanity(const struct iov_iter *i)
 191 {
 192         struct pipe_inode_info *pipe = i->pipe;
 193         unsigned int p_head = pipe->head;
 194         unsigned int p_tail = pipe->tail;
 195         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 196         unsigned int i_head = i->head;
 197         unsigned int idx;
 198
 199         if (i->last_offset) {
 200                 struct pipe_buffer *p;
 201                 if (unlikely(p_occupancy == 0))
 202                         goto Bad;       // pipe must be non-empty
 203                 if (unlikely(i_head != p_head - 1))
 204                         goto Bad;       // must be at the last buffer...
 205
 206                 p = pipe_buf(pipe, i_head);
 207                 if (unlikely(p->offset + p->len != abs(i->last_offset)))
 208                         goto Bad;       // ... at the end of segment
 209         } else {
 210                 if (i_head != p_head)
 211                         goto Bad;       // must be right after the last buffer
 212         }
 213         return true;
 214 Bad:
 215         printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset);
 216         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 217                         p_head, p_tail, pipe->ring_size);
 218         for (idx = 0; idx < pipe->ring_size; idx++)
 219                 printk(KERN_ERR "[%p %p %d %d]\n",
 220                         pipe->bufs[idx].ops,
 221                         pipe->bufs[idx].page,
 222                         pipe->bufs[idx].offset,
 223                         pipe->bufs[idx].len);
 224         WARN_ON(1);
 225         return false;
 226 }
 227 #else
 228 #define sanity(i) true
 229 #endif
 230
 231 static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size)
 232 {
 233         struct page *page = alloc_page(GFP_USER);
 234         if (page) {
 235                 struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 236                 *buf = (struct pipe_buffer) {
 237                         .ops = &default_pipe_buf_ops,
 238                         .page = page,
 239                         .offset = 0,
 240                         .len = size
 241                 };
 242         }
 243         return page;
 244 }
 245
 246 static void push_page(struct pipe_inode_info *pipe, struct page *page,
 247                         unsigned int offset, unsigned int size)
 248 {
 249         struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 250         *buf = (struct pipe_buffer) {
 251                 .ops = &page_cache_pipe_buf_ops,
 252                 .page = page,
 253                 .offset = offset,
 254                 .len = size
 255         };
 256         get_page(page);
 257 }
 258
 259 static inline int last_offset(const struct pipe_buffer *buf)
 260 {
 261         if (buf->ops == &default_pipe_buf_ops)
 262                 return buf->len;        // buf->offset is 0 for those
 263         else
 264                 return -(buf->offset + buf->len);
 265 }
 266
 267 static struct page *append_pipe(struct iov_iter *i, size_t size,
 268                                 unsigned int *off)
 269 {
 270         struct pipe_inode_info *pipe = i->pipe;
 271         int offset = i->last_offset;
 272         struct pipe_buffer *buf;
 273         struct page *page;
 274
 275         if (offset > 0 && offset < PAGE_SIZE) {
 276                 // some space in the last buffer; add to it
 277                 buf = pipe_buf(pipe, pipe->head - 1);
 278                 size = min_t(size_t, size, PAGE_SIZE - offset);
 279                 buf->len += size;
 280                 i->last_offset += size;
 281                 i->count -= size;
 282                 *off = offset;
 283                 return buf->page;
 284         }
 285         // OK, we need a new buffer
 286         *off = 0;
 287         size = min_t(size_t, size, PAGE_SIZE);
 288         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 289                 return NULL;
 290         page = push_anon(pipe, size);
 291         if (!page)
 292                 return NULL;
 293         i->head = pipe->head - 1;
 294         i->last_offset = size;
 295         i->count -= size;
 296         return page;
 297 }
 298
 299 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 300                          struct iov_iter *i)
 301 {
 302         struct pipe_inode_info *pipe = i->pipe;
 303         unsigned int head = pipe->head;
 304
 305         if (unlikely(bytes > i->count))
 306                 bytes = i->count;
 307
 308         if (unlikely(!bytes))
 309                 return 0;
 310
 311         if (!sanity(i))
 312                 return 0;
 313
 314         if (offset && i->last_offset == -offset) { // could we merge it?
 315                 struct pipe_buffer *buf = pipe_buf(pipe, head - 1);
 316                 if (buf->page == page) {
 317                         buf->len += bytes;
 318                         i->last_offset -= bytes;
 319                         i->count -= bytes;
 320                         return bytes;
 321                 }
 322         }
 323         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 324                 return 0;
 325
 326         push_page(pipe, page, offset, bytes);
 327         i->last_offset = -(offset + bytes);
 328         i->head = head;
 329         i->count -= bytes;
 330         return bytes;
 331 }
 332
 333 /*
 334  * fault_in_iov_iter_readable - fault in iov iterator for reading
 335  * @i: iterator
 336  * @size: maximum length
 337  *
 338  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 339  * @size.  For each iovec, fault in each page that constitutes the iovec.
 340  *
 341  * Returns the number of bytes not faulted in (like copy_to_user() and
 342  * copy_from_user()).
 343  *
 344  * Always returns 0 for non-userspace iterators.
 345  */
 346 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
 347 {
 348         if (iter_is_ubuf(i)) {
 349                 size_t n = min(size, iov_iter_count(i));
 350                 n -= fault_in_readable(i->ubuf + i->iov_offset, n);
 351                 return size - n;
 352         } else if (iter_is_iovec(i)) {
 353                 size_t count = min(size, iov_iter_count(i));
 354                 const struct iovec *p;
 355                 size_t skip;
 356
 357                 size -= count;
 358                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 359                         size_t len = min(count, p->iov_len - skip);
 360                         size_t ret;
 361
 362                         if (unlikely(!len))
 363                                 continue;
 364                         ret = fault_in_readable(p->iov_base + skip, len);
 365                         count -= len - ret;
 366                         if (ret)
 367                                 break;
 368                 }
 369                 return count + size;
 370         }
 371         return 0;
 372 }
 373 EXPORT_SYMBOL(fault_in_iov_iter_readable);
 374
 375 /*
 376  * fault_in_iov_iter_writeable - fault in iov iterator for writing
 377  * @i: iterator
 378  * @size: maximum length
 379  *
 380  * Faults in the iterator using get_user_pages(), i.e., without triggering
 381  * hardware page faults.  This is primarily useful when we already know that
 382  * some or all of the pages in @i aren't in memory.
 383  *
 384  * Returns the number of bytes not faulted in, like copy_to_user() and
 385  * copy_from_user().
 386  *
 387  * Always returns 0 for non-user-space iterators.
 388  */
 389 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
 390 {
 391         if (iter_is_ubuf(i)) {
 392                 size_t n = min(size, iov_iter_count(i));
 393                 n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n);
 394                 return size - n;
 395         } else if (iter_is_iovec(i)) {
 396                 size_t count = min(size, iov_iter_count(i));
 397                 const struct iovec *p;
 398                 size_t skip;
 399
 400                 size -= count;
 401                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 402                         size_t len = min(count, p->iov_len - skip);
 403                         size_t ret;
 404
 405                         if (unlikely(!len))
 406                                 continue;
 407                         ret = fault_in_safe_writeable(p->iov_base + skip, len);
 408                         count -= len - ret;
 409                         if (ret)
 410                                 break;
 411                 }
 412                 return count + size;
 413         }
 414         return 0;
 415 }
 416 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
 417
 418 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 419                         const struct iovec *iov, unsigned long nr_segs,
 420                         size_t count)
 421 {
 422         WARN_ON(direction & ~(READ | WRITE));
 423         *i = (struct iov_iter) {
 424                 .iter_type = ITER_IOVEC,
 425                 .nofault = false,
 426                 .user_backed = true,
 427                 .data_source = direction,
 428                 .iov = iov,
 429                 .nr_segs = nr_segs,
 430                 .iov_offset = 0,
 431                 .count = count
 432         };
 433 }
 434 EXPORT_SYMBOL(iov_iter_init);
 435
 436 // returns the offset in partial buffer (if any)
 437 static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages)
 438 {
 439         struct pipe_inode_info *pipe = i->pipe;
 440         int used = pipe->head - pipe->tail;
 441         int off = i->last_offset;
 442
 443         *npages = max((int)pipe->max_usage - used, 0);
 444
 445         if (off > 0 && off < PAGE_SIZE) { // anon and not full
 446                 (*npages)++;
 447                 return off;
 448         }
 449         return 0;
 450 }
 451
 452 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 453                                 struct iov_iter *i)
 454 {
 455         unsigned int off, chunk;
 456
 457         if (unlikely(bytes > i->count))
 458                 bytes = i->count;
 459         if (unlikely(!bytes))
 460                 return 0;
 461
 462         if (!sanity(i))
 463                 return 0;
 464
 465         for (size_t n = bytes; n; n -= chunk) {
 466                 struct page *page = append_pipe(i, n, &off);
 467                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 468                 if (!page)
 469                         return bytes - n;
 470                 memcpy_to_page(page, off, addr, chunk);
 471                 addr += chunk;
 472         }
 473         return bytes;
 474 }
 475
 476 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 477                               __wsum sum, size_t off)
 478 {
 479         __wsum next = csum_partial_copy_nocheck(from, to, len);
 480         return csum_block_add(sum, next, off);
 481 }
 482
 483 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 484                                          struct iov_iter *i, __wsum *sump)
 485 {
 486         __wsum sum = *sump;
 487         size_t off = 0;
 488         unsigned int chunk, r;
 489
 490         if (unlikely(bytes > i->count))
 491                 bytes = i->count;
 492         if (unlikely(!bytes))
 493                 return 0;
 494
 495         if (!sanity(i))
 496                 return 0;
 497
 498         while (bytes) {
 499                 struct page *page = append_pipe(i, bytes, &r);
 500                 char *p;
 501
 502                 if (!page)
 503                         break;
 504                 chunk = min_t(size_t, bytes, PAGE_SIZE - r);
 505                 p = kmap_local_page(page);
 506                 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
 507                 kunmap_local(p);
 508                 off += chunk;
 509                 bytes -= chunk;
 510         }
 511         *sump = sum;
 512         return off;
 513 }
 514
 515 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 516 {
 517         if (WARN_ON_ONCE(i->data_source))
 518                 return 0;
 519         if (unlikely(iov_iter_is_pipe(i)))
 520                 return copy_pipe_to_iter(addr, bytes, i);
 521         if (user_backed_iter(i))
 522                 might_fault();
 523         iterate_and_advance(i, bytes, base, len, off,
 524                 copyout(base, addr + off, len),
 525                 memcpy(base, addr + off, len)
 526         )
 527
 528         return bytes;
 529 }
 530 EXPORT_SYMBOL(_copy_to_iter);
 531
 532 #ifdef CONFIG_ARCH_HAS_COPY_MC
 533 static int copyout_mc(void __user *to, const void *from, size_t n)
 534 {
 535         if (access_ok(to, n)) {
 536                 instrument_copy_to_user(to, from, n);
 537                 n = copy_mc_to_user((__force void *) to, from, n);
 538         }
 539         return n;
 540 }
 541
 542 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 543                                 struct iov_iter *i)
 544 {
 545         size_t xfer = 0;
 546         unsigned int off, chunk;
 547
 548         if (unlikely(bytes > i->count))
 549                 bytes = i->count;
 550         if (unlikely(!bytes))
 551                 return 0;
 552
 553         if (!sanity(i))
 554                 return 0;
 555
 556         while (bytes) {
 557                 struct page *page = append_pipe(i, bytes, &off);
 558                 unsigned long rem;
 559                 char *p;
 560
 561                 if (!page)
 562                         break;
 563                 chunk = min_t(size_t, bytes, PAGE_SIZE - off);
 564                 p = kmap_local_page(page);
 565                 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
 566                 chunk -= rem;
 567                 kunmap_local(p);
 568                 xfer += chunk;
 569                 bytes -= chunk;
 570                 if (rem) {
 571                         iov_iter_revert(i, rem);
 572                         break;
 573                 }
 574         }
 575         return xfer;
 576 }
 577
 578 /**
 579  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 580  * @addr: source kernel address
 581  * @bytes: total transfer length
 582  * @i: destination iterator
 583  *
 584  * The pmem driver deploys this for the dax operation
 585  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 586  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 587  * successfully copied.
 588  *
 589  * The main differences between this and typical _copy_to_iter().
 590  *
 591  * * Typical tail/residue handling after a fault retries the copy
 592  *   byte-by-byte until the fault happens again. Re-triggering machine
 593  *   checks is potentially fatal so the implementation uses source
 594  *   alignment and poison alignment assumptions to avoid re-triggering
 595  *   hardware exceptions.
 596  *
 597  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 598  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 599  *   a short copy.
 600  *
 601  * Return: number of bytes copied (may be %0)
 602  */
 603 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 604 {
 605         if (WARN_ON_ONCE(i->data_source))
 606                 return 0;
 607         if (unlikely(iov_iter_is_pipe(i)))
 608                 return copy_mc_pipe_to_iter(addr, bytes, i);
 609         if (user_backed_iter(i))
 610                 might_fault();
 611         __iterate_and_advance(i, bytes, base, len, off,
 612                 copyout_mc(base, addr + off, len),
 613                 copy_mc_to_kernel(base, addr + off, len)
 614         )
 615
 616         return bytes;
 617 }
 618 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 619 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 620
 621 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 622 {
 623         if (WARN_ON_ONCE(!i->data_source))
 624                 return 0;
 625
 626         if (user_backed_iter(i))
 627                 might_fault();
 628         iterate_and_advance(i, bytes, base, len, off,
 629                 copyin(addr + off, base, len),
 630                 memcpy(addr + off, base, len)
 631         )
 632
 633         return bytes;
 634 }
 635 EXPORT_SYMBOL(_copy_from_iter);
 636
 637 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 638 {
 639         if (WARN_ON_ONCE(!i->data_source))
 640                 return 0;
 641
 642         iterate_and_advance(i, bytes, base, len, off,
 643                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 644                 memcpy(addr + off, base, len)
 645         )
 646
 647         return bytes;
 648 }
 649 EXPORT_SYMBOL(_copy_from_iter_nocache);
 650
 651 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 652 /**
 653  * _copy_from_iter_flushcache - write destination through cpu cache
 654  * @addr: destination kernel address
 655  * @bytes: total transfer length
 656  * @i: source iterator
 657  *
 658  * The pmem driver arranges for filesystem-dax to use this facility via
 659  * dax_copy_from_iter() for ensuring that writes to persistent memory
 660  * are flushed through the CPU cache. It is differentiated from
 661  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 662  * all iterator types. The _copy_from_iter_nocache() only attempts to
 663  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 664  * instructions that strand dirty-data in the cache.
 665  *
 666  * Return: number of bytes copied (may be %0)
 667  */
 668 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 669 {
 670         if (WARN_ON_ONCE(!i->data_source))
 671                 return 0;
 672
 673         iterate_and_advance(i, bytes, base, len, off,
 674                 __copy_from_user_flushcache(addr + off, base, len),
 675                 memcpy_flushcache(addr + off, base, len)
 676         )
 677
 678         return bytes;
 679 }
 680 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 681 #endif
 682
 683 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 684 {
 685         struct page *head;
 686         size_t v = n + offset;
 687
 688         /*
 689          * The general case needs to access the page order in order
 690          * to compute the page size.
 691          * However, we mostly deal with order-0 pages and thus can
 692          * avoid a possible cache line miss for requests that fit all
 693          * page orders.
 694          */
 695         if (n <= v && v <= PAGE_SIZE)
 696                 return true;
 697
 698         head = compound_head(page);
 699         v += (page - head) << PAGE_SHIFT;
 700
 701         if (WARN_ON(n > v || v > page_size(head)))
 702                 return false;
 703         return true;
 704 }
 705
 706 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 707                          struct iov_iter *i)
 708 {
 709         size_t res = 0;
 710         if (!page_copy_sane(page, offset, bytes))
 711                 return 0;
 712         if (WARN_ON_ONCE(i->data_source))
 713                 return 0;
 714         if (unlikely(iov_iter_is_pipe(i)))
 715                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 716         page += offset / PAGE_SIZE; // first subpage
 717         offset %= PAGE_SIZE;
 718         while (1) {
 719                 void *kaddr = kmap_local_page(page);
 720                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 721                 n = _copy_to_iter(kaddr + offset, n, i);
 722                 kunmap_local(kaddr);
 723                 res += n;
 724                 bytes -= n;
 725                 if (!bytes || !n)
 726                         break;
 727                 offset += n;
 728                 if (offset == PAGE_SIZE) {
 729                         page++;
 730                         offset = 0;
 731                 }
 732         }
 733         return res;
 734 }
 735 EXPORT_SYMBOL(copy_page_to_iter);
 736
 737 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 738                          struct iov_iter *i)
 739 {
 740         size_t res = 0;
 741         if (!page_copy_sane(page, offset, bytes))
 742                 return 0;
 743         page += offset / PAGE_SIZE; // first subpage
 744         offset %= PAGE_SIZE;
 745         while (1) {
 746                 void *kaddr = kmap_local_page(page);
 747                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 748                 n = _copy_from_iter(kaddr + offset, n, i);
 749                 kunmap_local(kaddr);
 750                 res += n;
 751                 bytes -= n;
 752                 if (!bytes || !n)
 753                         break;
 754                 offset += n;
 755                 if (offset == PAGE_SIZE) {
 756                         page++;
 757                         offset = 0;
 758                 }
 759         }
 760         return res;
 761 }
 762 EXPORT_SYMBOL(copy_page_from_iter);
 763
 764 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 765 {
 766         unsigned int chunk, off;
 767
 768         if (unlikely(bytes > i->count))
 769                 bytes = i->count;
 770         if (unlikely(!bytes))
 771                 return 0;
 772
 773         if (!sanity(i))
 774                 return 0;
 775
 776         for (size_t n = bytes; n; n -= chunk) {
 777                 struct page *page = append_pipe(i, n, &off);
 778                 char *p;
 779
 780                 if (!page)
 781                         return bytes - n;
 782                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 783                 p = kmap_local_page(page);
 784                 memset(p + off, 0, chunk);
 785                 kunmap_local(p);
 786         }
 787         return bytes;
 788 }
 789
 790 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 791 {
 792         if (unlikely(iov_iter_is_pipe(i)))
 793                 return pipe_zero(bytes, i);
 794         iterate_and_advance(i, bytes, base, len, count,
 795                 clear_user(base, len),
 796                 memset(base, 0, len)
 797         )
 798
 799         return bytes;
 800 }
 801 EXPORT_SYMBOL(iov_iter_zero);
 802
 803 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 804                                   struct iov_iter *i)
 805 {
 806         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 807         if (!page_copy_sane(page, offset, bytes)) {
 808                 kunmap_atomic(kaddr);
 809                 return 0;
 810         }
 811         if (WARN_ON_ONCE(!i->data_source)) {
 812                 kunmap_atomic(kaddr);
 813                 return 0;
 814         }
 815         iterate_and_advance(i, bytes, base, len, off,
 816                 copyin(p + off, base, len),
 817                 memcpy(p + off, base, len)
 818         )
 819         kunmap_atomic(kaddr);
 820         return bytes;
 821 }
 822 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 823
 824 static void pipe_advance(struct iov_iter *i, size_t size)
 825 {
 826         struct pipe_inode_info *pipe = i->pipe;
 827         int off = i->last_offset;
 828
 829         if (!off && !size) {
 830                 pipe_discard_from(pipe, i->start_head); // discard everything
 831                 return;
 832         }
 833         i->count -= size;
 834         while (1) {
 835                 struct pipe_buffer *buf = pipe_buf(pipe, i->head);
 836                 if (off) /* make it relative to the beginning of buffer */
 837                         size += abs(off) - buf->offset;
 838                 if (size <= buf->len) {
 839                         buf->len = size;
 840                         i->last_offset = last_offset(buf);
 841                         break;
 842                 }
 843                 size -= buf->len;
 844                 i->head++;
 845                 off = 0;
 846         }
 847         pipe_discard_from(pipe, i->head + 1); // discard everything past this one
 848 }
 849
 850 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 851 {
 852         const struct bio_vec *bvec, *end;
 853
 854         if (!i->count)
 855                 return;
 856         i->count -= size;
 857
 858         size += i->iov_offset;
 859
 860         for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
 861                 if (likely(size < bvec->bv_len))
 862                         break;
 863                 size -= bvec->bv_len;
 864         }
 865         i->iov_offset = size;
 866         i->nr_segs -= bvec - i->bvec;
 867         i->bvec = bvec;
 868 }
 869
 870 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 871 {
 872         const struct iovec *iov, *end;
 873
 874         if (!i->count)
 875                 return;
 876         i->count -= size;
 877
 878         size += i->iov_offset; // from beginning of current segment
 879         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
 880                 if (likely(size < iov->iov_len))
 881                         break;
 882                 size -= iov->iov_len;
 883         }
 884         i->iov_offset = size;
 885         i->nr_segs -= iov - i->iov;
 886         i->iov = iov;
 887 }
 888
 889 void iov_iter_advance(struct iov_iter *i, size_t size)
 890 {
 891         if (unlikely(i->count < size))
 892                 size = i->count;
 893         if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
 894                 i->iov_offset += size;
 895                 i->count -= size;
 896         } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 897                 /* iovec and kvec have identical layouts */
 898                 iov_iter_iovec_advance(i, size);
 899         } else if (iov_iter_is_bvec(i)) {
 900                 iov_iter_bvec_advance(i, size);
 901         } else if (iov_iter_is_pipe(i)) {
 902                 pipe_advance(i, size);
 903         } else if (iov_iter_is_discard(i)) {
 904                 i->count -= size;
 905         }
 906 }
 907 EXPORT_SYMBOL(iov_iter_advance);
 908
 909 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 910 {
 911         if (!unroll)
 912                 return;
 913         if (WARN_ON(unroll > MAX_RW_COUNT))
 914                 return;
 915         i->count += unroll;
 916         if (unlikely(iov_iter_is_pipe(i))) {
 917                 struct pipe_inode_info *pipe = i->pipe;
 918                 unsigned int head = pipe->head;
 919
 920                 while (head > i->start_head) {
 921                         struct pipe_buffer *b = pipe_buf(pipe, --head);
 922                         if (unroll < b->len) {
 923                                 b->len -= unroll;
 924                                 i->last_offset = last_offset(b);
 925                                 i->head = head;
 926                                 return;
 927                         }
 928                         unroll -= b->len;
 929                         pipe_buf_release(pipe, b);
 930                         pipe->head--;
 931                 }
 932                 i->last_offset = 0;
 933                 i->head = head;
 934                 return;
 935         }
 936         if (unlikely(iov_iter_is_discard(i)))
 937                 return;
 938         if (unroll <= i->iov_offset) {
 939                 i->iov_offset -= unroll;
 940                 return;
 941         }
 942         unroll -= i->iov_offset;
 943         if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
 944                 BUG(); /* We should never go beyond the start of the specified
 945                         * range since we might then be straying into pages that
 946                         * aren't pinned.
 947                         */
 948         } else if (iov_iter_is_bvec(i)) {
 949                 const struct bio_vec *bvec = i->bvec;
 950                 while (1) {
 951                         size_t n = (--bvec)->bv_len;
 952                         i->nr_segs++;
 953                         if (unroll <= n) {
 954                                 i->bvec = bvec;
 955                                 i->iov_offset = n - unroll;
 956                                 return;
 957                         }
 958                         unroll -= n;
 959                 }
 960         } else { /* same logics for iovec and kvec */
 961                 const struct iovec *iov = i->iov;
 962                 while (1) {
 963                         size_t n = (--iov)->iov_len;
 964                         i->nr_segs++;
 965                         if (unroll <= n) {
 966                                 i->iov = iov;
 967                                 i->iov_offset = n - unroll;
 968                                 return;
 969                         }
 970                         unroll -= n;
 971                 }
 972         }
 973 }
 974 EXPORT_SYMBOL(iov_iter_revert);
 975
 976 /*
 977  * Return the count of just the current iov_iter segment.
 978  */
 979 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 980 {
 981         if (i->nr_segs > 1) {
 982                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 983                         return min(i->count, i->iov->iov_len - i->iov_offset);
 984                 if (iov_iter_is_bvec(i))
 985                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 986         }
 987         return i->count;
 988 }
 989 EXPORT_SYMBOL(iov_iter_single_seg_count);
 990
 991 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
 992                         const struct kvec *kvec, unsigned long nr_segs,
 993                         size_t count)
 994 {
 995         WARN_ON(direction & ~(READ | WRITE));
 996         *i = (struct iov_iter){
 997                 .iter_type = ITER_KVEC,
 998                 .data_source = direction,
 999                 .kvec = kvec,
1000                 .nr_segs = nr_segs,
1001                 .iov_offset = 0,
1002                 .count = count
1003         };
1004 }
1005 EXPORT_SYMBOL(iov_iter_kvec);
1006
1007 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1008                         const struct bio_vec *bvec, unsigned long nr_segs,
1009                         size_t count)
1010 {
1011         WARN_ON(direction & ~(READ | WRITE));
1012         *i = (struct iov_iter){
1013                 .iter_type = ITER_BVEC,
1014                 .data_source = direction,
1015                 .bvec = bvec,
1016                 .nr_segs = nr_segs,
1017                 .iov_offset = 0,
1018                 .count = count
1019         };
1020 }
1021 EXPORT_SYMBOL(iov_iter_bvec);
1022
1023 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1024                         struct pipe_inode_info *pipe,
1025                         size_t count)
1026 {
1027         BUG_ON(direction != READ);
1028         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1029         *i = (struct iov_iter){
1030                 .iter_type = ITER_PIPE,
1031                 .data_source = false,
1032                 .pipe = pipe,
1033                 .head = pipe->head,
1034                 .start_head = pipe->head,
1035                 .last_offset = 0,
1036                 .count = count
1037         };
1038 }
1039 EXPORT_SYMBOL(iov_iter_pipe);
1040
1041 /**
1042  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1043  * @i: The iterator to initialise.
1044  * @direction: The direction of the transfer.
1045  * @xarray: The xarray to access.
1046  * @start: The start file position.
1047  * @count: The size of the I/O buffer in bytes.
1048  *
1049  * Set up an I/O iterator to either draw data out of the pages attached to an
1050  * inode or to inject data into those pages.  The pages *must* be prevented
1051  * from evaporation, either by taking a ref on them or locking them by the
1052  * caller.
1053  */
1054 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1055                      struct xarray *xarray, loff_t start, size_t count)
1056 {
1057         BUG_ON(direction & ~1);
1058         *i = (struct iov_iter) {
1059                 .iter_type = ITER_XARRAY,
1060                 .data_source = direction,
1061                 .xarray = xarray,
1062                 .xarray_start = start,
1063                 .count = count,
1064                 .iov_offset = 0
1065         };
1066 }
1067 EXPORT_SYMBOL(iov_iter_xarray);
1068
1069 /**
1070  * iov_iter_discard - Initialise an I/O iterator that discards data
1071  * @i: The iterator to initialise.
1072  * @direction: The direction of the transfer.
1073  * @count: The size of the I/O buffer in bytes.
1074  *
1075  * Set up an I/O iterator that just discards everything that's written to it.
1076  * It's only available as a READ iterator.
1077  */
1078 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1079 {
1080         BUG_ON(direction != READ);
1081         *i = (struct iov_iter){
1082                 .iter_type = ITER_DISCARD,
1083                 .data_source = false,
1084                 .count = count,
1085                 .iov_offset = 0
1086         };
1087 }
1088 EXPORT_SYMBOL(iov_iter_discard);
1089
1090 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1091                                    unsigned len_mask)
1092 {
1093         size_t size = i->count;
1094         size_t skip = i->iov_offset;
1095         unsigned k;
1096
1097         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1098                 size_t len = i->iov[k].iov_len - skip;
1099
1100                 if (len > size)
1101                         len = size;
1102                 if (len & len_mask)
1103                         return false;
1104                 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1105                         return false;
1106
1107                 size -= len;
1108                 if (!size)
1109                         break;
1110         }
1111         return true;
1112 }
1113
1114 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1115                                   unsigned len_mask)
1116 {
1117         size_t size = i->count;
1118         unsigned skip = i->iov_offset;
1119         unsigned k;
1120
1121         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1122                 size_t len = i->bvec[k].bv_len - skip;
1123
1124                 if (len > size)
1125                         len = size;
1126                 if (len & len_mask)
1127                         return false;
1128                 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1129                         return false;
1130
1131                 size -= len;
1132                 if (!size)
1133                         break;
1134         }
1135         return true;
1136 }
1137
1138 /**
1139  * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1140  *      are aligned to the parameters.
1141  *
1142  * @i: &struct iov_iter to restore
1143  * @addr_mask: bit mask to check against the iov element's addresses
1144  * @len_mask: bit mask to check against the iov element's lengths
1145  *
1146  * Return: false if any addresses or lengths intersect with the provided masks
1147  */
1148 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1149                          unsigned len_mask)
1150 {
1151         if (likely(iter_is_ubuf(i))) {
1152                 if (i->count & len_mask)
1153                         return false;
1154                 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
1155                         return false;
1156                 return true;
1157         }
1158
1159         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1160                 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1161
1162         if (iov_iter_is_bvec(i))
1163                 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1164
1165         if (iov_iter_is_pipe(i)) {
1166                 size_t size = i->count;
1167
1168                 if (size & len_mask)
1169                         return false;
1170                 if (size && i->last_offset > 0) {
1171                         if (i->last_offset & addr_mask)
1172                                 return false;
1173                 }
1174
1175                 return true;
1176         }
1177
1178         if (iov_iter_is_xarray(i)) {
1179                 if (i->count & len_mask)
1180                         return false;
1181                 if ((i->xarray_start + i->iov_offset) & addr_mask)
1182                         return false;
1183         }
1184
1185         return true;
1186 }
1187 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1188
1189 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1190 {
1191         unsigned long res = 0;
1192         size_t size = i->count;
1193         size_t skip = i->iov_offset;
1194         unsigned k;
1195
1196         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1197                 size_t len = i->iov[k].iov_len - skip;
1198                 if (len) {
1199                         res |= (unsigned long)i->iov[k].iov_base + skip;
1200                         if (len > size)
1201                                 len = size;
1202                         res |= len;
1203                         size -= len;
1204                         if (!size)
1205                                 break;
1206                 }
1207         }
1208         return res;
1209 }
1210
1211 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1212 {
1213         unsigned res = 0;
1214         size_t size = i->count;
1215         unsigned skip = i->iov_offset;
1216         unsigned k;
1217
1218         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1219                 size_t len = i->bvec[k].bv_len - skip;
1220                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1221                 if (len > size)
1222                         len = size;
1223                 res |= len;
1224                 size -= len;
1225                 if (!size)
1226                         break;
1227         }
1228         return res;
1229 }
1230
1231 unsigned long iov_iter_alignment(const struct iov_iter *i)
1232 {
1233         if (likely(iter_is_ubuf(i))) {
1234                 size_t size = i->count;
1235                 if (size)
1236                         return ((unsigned long)i->ubuf + i->iov_offset) | size;
1237                 return 0;
1238         }
1239
1240         /* iovec and kvec have identical layouts */
1241         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1242                 return iov_iter_alignment_iovec(i);
1243
1244         if (iov_iter_is_bvec(i))
1245                 return iov_iter_alignment_bvec(i);
1246
1247         if (iov_iter_is_pipe(i)) {
1248                 size_t size = i->count;
1249
1250                 if (size && i->last_offset > 0)
1251                         return size | i->last_offset;
1252                 return size;
1253         }
1254
1255         if (iov_iter_is_xarray(i))
1256                 return (i->xarray_start + i->iov_offset) | i->count;
1257
1258         return 0;
1259 }
1260 EXPORT_SYMBOL(iov_iter_alignment);
1261
1262 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1263 {
1264         unsigned long res = 0;
1265         unsigned long v = 0;
1266         size_t size = i->count;
1267         unsigned k;
1268
1269         if (iter_is_ubuf(i))
1270                 return 0;
1271
1272         if (WARN_ON(!iter_is_iovec(i)))
1273                 return ~0U;
1274
1275         for (k = 0; k < i->nr_segs; k++) {
1276                 if (i->iov[k].iov_len) {
1277                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1278                         if (v) // if not the first one
1279                                 res |= base | v; // this start | previous end
1280                         v = base + i->iov[k].iov_len;
1281                         if (size <= i->iov[k].iov_len)
1282                                 break;
1283                         size -= i->iov[k].iov_len;
1284                 }
1285         }
1286         return res;
1287 }
1288 EXPORT_SYMBOL(iov_iter_gap_alignment);
1289
1290 static int want_pages_array(struct page ***res, size_t size,
1291                             size_t start, unsigned int maxpages)
1292 {
1293         unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
1294
1295         if (count > maxpages)
1296                 count = maxpages;
1297         WARN_ON(!count);        // caller should've prevented that
1298         if (!*res) {
1299                 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
1300                 if (!*res)
1301                         return 0;
1302         }
1303         return count;
1304 }
1305
1306 static ssize_t pipe_get_pages(struct iov_iter *i,
1307                    struct page ***pages, size_t maxsize, unsigned maxpages,
1308                    size_t *start)
1309 {
1310         unsigned int npages, count, off, chunk;
1311         struct page **p;
1312         size_t left;
1313
1314         if (!sanity(i))
1315                 return -EFAULT;
1316
1317         *start = off = pipe_npages(i, &npages);
1318         if (!npages)
1319                 return -EFAULT;
1320         count = want_pages_array(pages, maxsize, off, min(npages, maxpages));
1321         if (!count)
1322                 return -ENOMEM;
1323         p = *pages;
1324         for (npages = 0, left = maxsize ; npages < count; npages++, left -= chunk) {
1325                 struct page *page = append_pipe(i, left, &off);
1326                 if (!page)
1327                         break;
1328                 chunk = min_t(size_t, left, PAGE_SIZE - off);
1329                 get_page(*p++ = page);
1330         }
1331         if (!npages)
1332                 return -EFAULT;
1333         return maxsize - left;
1334 }
1335
1336 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1337                                           pgoff_t index, unsigned int nr_pages)
1338 {
1339         XA_STATE(xas, xa, index);
1340         struct page *page;
1341         unsigned int ret = 0;
1342
1343         rcu_read_lock();
1344         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1345                 if (xas_retry(&xas, page))
1346                         continue;
1347
1348                 /* Has the page moved or been split? */
1349                 if (unlikely(page != xas_reload(&xas))) {
1350                         xas_reset(&xas);
1351                         continue;
1352                 }
1353
1354                 pages[ret] = find_subpage(page, xas.xa_index);
1355                 get_page(pages[ret]);
1356                 if (++ret == nr_pages)
1357                         break;
1358         }
1359         rcu_read_unlock();
1360         return ret;
1361 }
1362
1363 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1364                                      struct page ***pages, size_t maxsize,
1365                                      unsigned maxpages, size_t *_start_offset)
1366 {
1367         unsigned nr, offset, count;
1368         pgoff_t index;
1369         loff_t pos;
1370
1371         pos = i->xarray_start + i->iov_offset;
1372         index = pos >> PAGE_SHIFT;
1373         offset = pos & ~PAGE_MASK;
1374         *_start_offset = offset;
1375
1376         count = want_pages_array(pages, maxsize, offset, maxpages);
1377         if (!count)
1378                 return -ENOMEM;
1379         nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
1380         if (nr == 0)
1381                 return 0;
1382
1383         maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1384         i->iov_offset += maxsize;
1385         i->count -= maxsize;
1386         return maxsize;
1387 }
1388
1389 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1390 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1391 {
1392         size_t skip;
1393         long k;
1394
1395         if (iter_is_ubuf(i))
1396                 return (unsigned long)i->ubuf + i->iov_offset;
1397
1398         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1399                 size_t len = i->iov[k].iov_len - skip;
1400
1401                 if (unlikely(!len))
1402                         continue;
1403                 if (*size > len)
1404                         *size = len;
1405                 return (unsigned long)i->iov[k].iov_base + skip;
1406         }
1407         BUG(); // if it had been empty, we wouldn't get called
1408 }
1409
1410 /* must be done on non-empty ITER_BVEC one */
1411 static struct page *first_bvec_segment(const struct iov_iter *i,
1412                                        size_t *size, size_t *start)
1413 {
1414         struct page *page;
1415         size_t skip = i->iov_offset, len;
1416
1417         len = i->bvec->bv_len - skip;
1418         if (*size > len)
1419                 *size = len;
1420         skip += i->bvec->bv_offset;
1421         page = i->bvec->bv_page + skip / PAGE_SIZE;
1422         *start = skip % PAGE_SIZE;
1423         return page;
1424 }
1425
1426 static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
1427                    struct page ***pages, size_t maxsize,
1428                    unsigned int maxpages, size_t *start,
1429                    iov_iter_extraction_t extraction_flags)
1430 {
1431         unsigned int n, gup_flags = 0;
1432
1433         if (maxsize > i->count)
1434                 maxsize = i->count;
1435         if (!maxsize)
1436                 return 0;
1437         if (maxsize > MAX_RW_COUNT)
1438                 maxsize = MAX_RW_COUNT;
1439         if (extraction_flags & ITER_ALLOW_P2PDMA)
1440                 gup_flags |= FOLL_PCI_P2PDMA;
1441
1442         if (likely(user_backed_iter(i))) {
1443                 unsigned long addr;
1444                 int res;
1445
1446                 if (iov_iter_rw(i) != WRITE)
1447                         gup_flags |= FOLL_WRITE;
1448                 if (i->nofault)
1449                         gup_flags |= FOLL_NOFAULT;
1450
1451                 addr = first_iovec_segment(i, &maxsize);
1452                 *start = addr % PAGE_SIZE;
1453                 addr &= PAGE_MASK;
1454                 n = want_pages_array(pages, maxsize, *start, maxpages);
1455                 if (!n)
1456                         return -ENOMEM;
1457                 res = get_user_pages_fast(addr, n, gup_flags, *pages);
1458                 if (unlikely(res <= 0))
1459                         return res;
1460                 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1461                 iov_iter_advance(i, maxsize);
1462                 return maxsize;
1463         }
1464         if (iov_iter_is_bvec(i)) {
1465                 struct page **p;
1466                 struct page *page;
1467
1468                 page = first_bvec_segment(i, &maxsize, start);
1469                 n = want_pages_array(pages, maxsize, *start, maxpages);
1470                 if (!n)
1471                         return -ENOMEM;
1472                 p = *pages;
1473                 for (int k = 0; k < n; k++)
1474                         get_page(p[k] = page + k);
1475                 maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1476                 i->count -= maxsize;
1477                 i->iov_offset += maxsize;
1478                 if (i->iov_offset == i->bvec->bv_len) {
1479                         i->iov_offset = 0;
1480                         i->bvec++;
1481                         i->nr_segs--;
1482                 }
1483                 return maxsize;
1484         }
1485         if (iov_iter_is_pipe(i))
1486                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1487         if (iov_iter_is_xarray(i))
1488                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1489         return -EFAULT;
1490 }
1491
1492 ssize_t iov_iter_get_pages(struct iov_iter *i,
1493                    struct page **pages, size_t maxsize, unsigned maxpages,
1494                    size_t *start, iov_iter_extraction_t extraction_flags)
1495 {
1496         if (!maxpages)
1497                 return 0;
1498         BUG_ON(!pages);
1499
1500         return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages,
1501                                           start, extraction_flags);
1502 }
1503 EXPORT_SYMBOL_GPL(iov_iter_get_pages);
1504
1505 ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages,
1506                 size_t maxsize, unsigned maxpages, size_t *start)
1507 {
1508         return iov_iter_get_pages(i, pages, maxsize, maxpages, start, 0);
1509 }
1510 EXPORT_SYMBOL(iov_iter_get_pages2);
1511
1512 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1513                    struct page ***pages, size_t maxsize,
1514                    size_t *start, iov_iter_extraction_t extraction_flags)
1515 {
1516         ssize_t len;
1517
1518         *pages = NULL;
1519
1520         len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start,
1521                                          extraction_flags);
1522         if (len <= 0) {
1523                 kvfree(*pages);
1524                 *pages = NULL;
1525         }
1526         return len;
1527 }
1528 EXPORT_SYMBOL_GPL(iov_iter_get_pages_alloc);
1529
1530 ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1531                 struct page ***pages, size_t maxsize, size_t *start)
1532 {
1533         return iov_iter_get_pages_alloc(i, pages, maxsize, start, 0);
1534 }
1535 EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1536
1537 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1538                                struct iov_iter *i)
1539 {
1540         __wsum sum, next;
1541         sum = *csum;
1542         if (WARN_ON_ONCE(!i->data_source))
1543                 return 0;
1544
1545         iterate_and_advance(i, bytes, base, len, off, ({
1546                 next = csum_and_copy_from_user(base, addr + off, len);
1547                 sum = csum_block_add(sum, next, off);
1548                 next ? 0 : len;
1549         }), ({
1550                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1551         })
1552         )
1553         *csum = sum;
1554         return bytes;
1555 }
1556 EXPORT_SYMBOL(csum_and_copy_from_iter);
1557
1558 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1559                              struct iov_iter *i)
1560 {
1561         struct csum_state *csstate = _csstate;
1562         __wsum sum, next;
1563
1564         if (WARN_ON_ONCE(i->data_source))
1565                 return 0;
1566         if (unlikely(iov_iter_is_discard(i))) {
1567                 // can't use csum_memcpy() for that one - data is not copied
1568                 csstate->csum = csum_block_add(csstate->csum,
1569                                                csum_partial(addr, bytes, 0),
1570                                                csstate->off);
1571                 csstate->off += bytes;
1572                 return bytes;
1573         }
1574
1575         sum = csum_shift(csstate->csum, csstate->off);
1576         if (unlikely(iov_iter_is_pipe(i)))
1577                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1578         else iterate_and_advance(i, bytes, base, len, off, ({
1579                 next = csum_and_copy_to_user(addr + off, base, len);
1580                 sum = csum_block_add(sum, next, off);
1581                 next ? 0 : len;
1582         }), ({
1583                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1584         })
1585         )
1586         csstate->csum = csum_shift(sum, csstate->off);
1587         csstate->off += bytes;
1588         return bytes;
1589 }
1590 EXPORT_SYMBOL(csum_and_copy_to_iter);
1591
1592 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1593                 struct iov_iter *i)
1594 {
1595 #ifdef CONFIG_CRYPTO_HASH
1596         struct ahash_request *hash = hashp;
1597         struct scatterlist sg;
1598         size_t copied;
1599
1600         copied = copy_to_iter(addr, bytes, i);
1601         sg_init_one(&sg, addr, copied);
1602         ahash_request_set_crypt(hash, &sg, NULL, copied);
1603         crypto_ahash_update(hash);
1604         return copied;
1605 #else
1606         return 0;
1607 #endif
1608 }
1609 EXPORT_SYMBOL(hash_and_copy_to_iter);
1610
1611 static int iov_npages(const struct iov_iter *i, int maxpages)
1612 {
1613         size_t skip = i->iov_offset, size = i->count;
1614         const struct iovec *p;
1615         int npages = 0;
1616
1617         for (p = i->iov; size; skip = 0, p++) {
1618                 unsigned offs = offset_in_page(p->iov_base + skip);
1619                 size_t len = min(p->iov_len - skip, size);
1620
1621                 if (len) {
1622                         size -= len;
1623                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1624                         if (unlikely(npages > maxpages))
1625                                 return maxpages;
1626                 }
1627         }
1628         return npages;
1629 }
1630
1631 static int bvec_npages(const struct iov_iter *i, int maxpages)
1632 {
1633         size_t skip = i->iov_offset, size = i->count;
1634         const struct bio_vec *p;
1635         int npages = 0;
1636
1637         for (p = i->bvec; size; skip = 0, p++) {
1638                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1639                 size_t len = min(p->bv_len - skip, size);
1640
1641                 size -= len;
1642                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1643                 if (unlikely(npages > maxpages))
1644                         return maxpages;
1645         }
1646         return npages;
1647 }
1648
1649 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1650 {
1651         if (unlikely(!i->count))
1652                 return 0;
1653         if (likely(iter_is_ubuf(i))) {
1654                 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1655                 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1656                 return min(npages, maxpages);
1657         }
1658         /* iovec and kvec have identical layouts */
1659         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1660                 return iov_npages(i, maxpages);
1661         if (iov_iter_is_bvec(i))
1662                 return bvec_npages(i, maxpages);
1663         if (iov_iter_is_pipe(i)) {
1664                 int npages;
1665
1666                 if (!sanity(i))
1667                         return 0;
1668
1669                 pipe_npages(i, &npages);
1670                 return min(npages, maxpages);
1671         }
1672         if (iov_iter_is_xarray(i)) {
1673                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1674                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1675                 return min(npages, maxpages);
1676         }
1677         return 0;
1678 }
1679 EXPORT_SYMBOL(iov_iter_npages);
1680
1681 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1682 {
1683         *new = *old;
1684         if (unlikely(iov_iter_is_pipe(new))) {
1685                 WARN_ON(1);
1686                 return NULL;
1687         }
1688         if (iov_iter_is_bvec(new))
1689                 return new->bvec = kmemdup(new->bvec,
1690                                     new->nr_segs * sizeof(struct bio_vec),
1691                                     flags);
1692         else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1693                 /* iovec and kvec have identical layout */
1694                 return new->iov = kmemdup(new->iov,
1695                                    new->nr_segs * sizeof(struct iovec),
1696                                    flags);
1697         return NULL;
1698 }
1699 EXPORT_SYMBOL(dup_iter);
1700
1701 static int copy_compat_iovec_from_user(struct iovec *iov,
1702                 const struct iovec __user *uvec, unsigned long nr_segs)
1703 {
1704         const struct compat_iovec __user *uiov =
1705                 (const struct compat_iovec __user *)uvec;
1706         int ret = -EFAULT, i;
1707
1708         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1709                 return -EFAULT;
1710
1711         for (i = 0; i < nr_segs; i++) {
1712                 compat_uptr_t buf;
1713                 compat_ssize_t len;
1714
1715                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1716                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1717
1718                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1719                 if (len < 0) {
1720                         ret = -EINVAL;
1721                         goto uaccess_end;
1722                 }
1723                 iov[i].iov_base = compat_ptr(buf);
1724                 iov[i].iov_len = len;
1725         }
1726
1727         ret = 0;
1728 uaccess_end:
1729         user_access_end();
1730         return ret;
1731 }
1732
1733 static int copy_iovec_from_user(struct iovec *iov,
1734                 const struct iovec __user *uvec, unsigned long nr_segs)
1735 {
1736         unsigned long seg;
1737
1738         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1739                 return -EFAULT;
1740         for (seg = 0; seg < nr_segs; seg++) {
1741                 if ((ssize_t)iov[seg].iov_len < 0)
1742                         return -EINVAL;
1743         }
1744
1745         return 0;
1746 }
1747
1748 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1749                 unsigned long nr_segs, unsigned long fast_segs,
1750                 struct iovec *fast_iov, bool compat)
1751 {
1752         struct iovec *iov = fast_iov;
1753         int ret;
1754
1755         /*
1756          * SuS says "The readv() function *may* fail if the iovcnt argument was
1757          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1758          * traditionally returned zero for zero segments, so...
1759          */
1760         if (nr_segs == 0)
1761                 return iov;
1762         if (nr_segs > UIO_MAXIOV)
1763                 return ERR_PTR(-EINVAL);
1764         if (nr_segs > fast_segs) {
1765                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1766                 if (!iov)
1767                         return ERR_PTR(-ENOMEM);
1768         }
1769
1770         if (compat)
1771                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1772         else
1773                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1774         if (ret) {
1775                 if (iov != fast_iov)
1776                         kfree(iov);
1777                 return ERR_PTR(ret);
1778         }
1779
1780         return iov;
1781 }
1782
1783 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1784                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1785                  struct iov_iter *i, bool compat)
1786 {
1787         ssize_t total_len = 0;
1788         unsigned long seg;
1789         struct iovec *iov;
1790
1791         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1792         if (IS_ERR(iov)) {
1793                 *iovp = NULL;
1794                 return PTR_ERR(iov);
1795         }
1796
1797         /*
1798          * According to the Single Unix Specification we should return EINVAL if
1799          * an element length is < 0 when cast to ssize_t or if the total length
1800          * would overflow the ssize_t return value of the system call.
1801          *
1802          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1803          * overflow case.
1804          */
1805         for (seg = 0; seg < nr_segs; seg++) {
1806                 ssize_t len = (ssize_t)iov[seg].iov_len;
1807
1808                 if (!access_ok(iov[seg].iov_base, len)) {
1809                         if (iov != *iovp)
1810                                 kfree(iov);
1811                         *iovp = NULL;
1812                         return -EFAULT;
1813                 }
1814
1815                 if (len > MAX_RW_COUNT - total_len) {
1816                         len = MAX_RW_COUNT - total_len;
1817                         iov[seg].iov_len = len;
1818                 }
1819                 total_len += len;
1820         }
1821
1822         iov_iter_init(i, type, iov, nr_segs, total_len);
1823         if (iov == *iovp)
1824                 *iovp = NULL;
1825         else
1826                 *iovp = iov;
1827         return total_len;
1828 }
1829
1830 /**
1831  * import_iovec() - Copy an array of &struct iovec from userspace
1832  *     into the kernel, check that it is valid, and initialize a new
1833  *     &struct iov_iter iterator to access it.
1834  *
1835  * @type: One of %READ or %WRITE.
1836  * @uvec: Pointer to the userspace array.
1837  * @nr_segs: Number of elements in userspace array.
1838  * @fast_segs: Number of elements in @iov.
1839  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1840  *     on-stack) kernel array.
1841  * @i: Pointer to iterator that will be initialized on success.
1842  *
1843  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1844  * then this function places %NULL in *@iov on return. Otherwise, a new
1845  * array will be allocated and the result placed in *@iov. This means that
1846  * the caller may call kfree() on *@iov regardless of whether the small
1847  * on-stack array was used or not (and regardless of whether this function
1848  * returns an error or not).
1849  *
1850  * Return: Negative error code on error, bytes imported on success
1851  */
1852 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1853                  unsigned nr_segs, unsigned fast_segs,
1854                  struct iovec **iovp, struct iov_iter *i)
1855 {
1856         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1857                               in_compat_syscall());
1858 }
1859 EXPORT_SYMBOL(import_iovec);
1860
1861 int import_single_range(int rw, void __user *buf, size_t len,
1862                  struct iovec *iov, struct iov_iter *i)
1863 {
1864         if (len > MAX_RW_COUNT)
1865                 len = MAX_RW_COUNT;
1866         if (unlikely(!access_ok(buf, len)))
1867                 return -EFAULT;
1868
1869         iov->iov_base = buf;
1870         iov->iov_len = len;
1871         iov_iter_init(i, rw, iov, 1, len);
1872         return 0;
1873 }
1874 EXPORT_SYMBOL(import_single_range);
1875
1876 int import_ubuf(int rw, void __user *buf, size_t len, struct iov_iter *i)
1877 {
1878         if (len > MAX_RW_COUNT)
1879                 len = MAX_RW_COUNT;
1880         if (unlikely(!access_ok(buf, len)))
1881                 return -EFAULT;
1882
1883         iov_iter_ubuf(i, rw, buf, len);
1884         return 0;
1885 }
1886
1887 /**
1888  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1889  *     iov_iter_save_state() was called.
1890  *
1891  * @i: &struct iov_iter to restore
1892  * @state: state to restore from
1893  *
1894  * Used after iov_iter_save_state() to bring restore @i, if operations may
1895  * have advanced it.
1896  *
1897  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1898  */
1899 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1900 {
1901         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i) &&
1902                          !iter_is_ubuf(i)) && !iov_iter_is_kvec(i))
1903                 return;
1904         i->iov_offset = state->iov_offset;
1905         i->count = state->count;
1906         if (iter_is_ubuf(i))
1907                 return;
1908         /*
1909          * For the *vec iters, nr_segs + iov is constant - if we increment
1910          * the vec, then we also decrement the nr_segs count. Hence we don't
1911          * need to track both of these, just one is enough and we can deduct
1912          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1913          * size, so we can just increment the iov pointer as they are unionzed.
1914          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1915          * not. Be safe and handle it separately.
1916          */
1917         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1918         if (iov_iter_is_bvec(i))
1919                 i->bvec -= state->nr_segs - i->nr_segs;
1920         else
1921                 i->iov -= state->nr_segs - i->nr_segs;
1922         i->nr_segs = state->nr_segs;
1923 }
1924
1925 /*
1926  * Extract a list of contiguous pages from an ITER_XARRAY iterator.  This does not
1927  * get references on the pages, nor does it get a pin on them.
1928  */
1929 static ssize_t iov_iter_extract_xarray_pages(struct iov_iter *i,
1930                                              struct page ***pages, size_t maxsize,
1931                                              unsigned int maxpages,
1932                                              iov_iter_extraction_t extraction_flags,
1933                                              size_t *offset0)
1934 {
1935         struct page *page, **p;
1936         unsigned int nr = 0, offset;
1937         loff_t pos = i->xarray_start + i->iov_offset;
1938         pgoff_t index = pos >> PAGE_SHIFT;
1939         XA_STATE(xas, i->xarray, index);
1940
1941         offset = pos & ~PAGE_MASK;
1942         *offset0 = offset;
1943
1944         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1945         if (!maxpages)
1946                 return -ENOMEM;
1947         p = *pages;
1948
1949         rcu_read_lock();
1950         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1951                 if (xas_retry(&xas, page))
1952                         continue;
1953
1954                 /* Has the page moved or been split? */
1955                 if (unlikely(page != xas_reload(&xas))) {
1956                         xas_reset(&xas);
1957                         continue;
1958                 }
1959
1960                 p[nr++] = find_subpage(page, xas.xa_index);
1961                 if (nr == maxpages)
1962                         break;
1963         }
1964         rcu_read_unlock();
1965
1966         maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1967         iov_iter_advance(i, maxsize);
1968         return maxsize;
1969 }
1970
1971 /*
1972  * Extract a list of contiguous pages from an ITER_BVEC iterator.  This does
1973  * not get references on the pages, nor does it get a pin on them.
1974  */
1975 static ssize_t iov_iter_extract_bvec_pages(struct iov_iter *i,
1976                                            struct page ***pages, size_t maxsize,
1977                                            unsigned int maxpages,
1978                                            iov_iter_extraction_t extraction_flags,
1979                                            size_t *offset0)
1980 {
1981         struct page **p, *page;
1982         size_t skip = i->iov_offset, offset;
1983         int k;
1984
1985         for (;;) {
1986                 if (i->nr_segs == 0)
1987                         return 0;
1988                 maxsize = min(maxsize, i->bvec->bv_len - skip);
1989                 if (maxsize)
1990                         break;
1991                 i->iov_offset = 0;
1992                 i->nr_segs--;
1993                 i->bvec++;
1994                 skip = 0;
1995         }
1996
1997         skip += i->bvec->bv_offset;
1998         page = i->bvec->bv_page + skip / PAGE_SIZE;
1999         offset = skip % PAGE_SIZE;
2000         *offset0 = offset;
2001
2002         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
2003         if (!maxpages)
2004                 return -ENOMEM;
2005         p = *pages;
2006         for (k = 0; k < maxpages; k++)
2007                 p[k] = page + k;
2008
2009         maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset);
2010         iov_iter_advance(i, maxsize);
2011         return maxsize;
2012 }
2013
2014 /*
2015  * Extract a list of virtually contiguous pages from an ITER_KVEC iterator.
2016  * This does not get references on the pages, nor does it get a pin on them.
2017  */
2018 static ssize_t iov_iter_extract_kvec_pages(struct iov_iter *i,
2019                                            struct page ***pages, size_t maxsize,
2020                                            unsigned int maxpages,
2021                                            iov_iter_extraction_t extraction_flags,
2022                                            size_t *offset0)
2023 {
2024         struct page **p, *page;
2025         const void *kaddr;
2026         size_t skip = i->iov_offset, offset, len;
2027         int k;
2028
2029         for (;;) {
2030                 if (i->nr_segs == 0)
2031                         return 0;
2032                 maxsize = min(maxsize, i->kvec->iov_len - skip);
2033                 if (maxsize)
2034                         break;
2035                 i->iov_offset = 0;
2036                 i->nr_segs--;
2037                 i->kvec++;
2038                 skip = 0;
2039         }
2040
2041         kaddr = i->kvec->iov_base + skip;
2042         offset = (unsigned long)kaddr & ~PAGE_MASK;
2043         *offset0 = offset;
2044
2045         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
2046         if (!maxpages)
2047                 return -ENOMEM;
2048         p = *pages;
2049
2050         kaddr -= offset;
2051         len = offset + maxsize;
2052         for (k = 0; k < maxpages; k++) {
2053                 size_t seg = min_t(size_t, len, PAGE_SIZE);
2054
2055                 if (is_vmalloc_or_module_addr(kaddr))
2056                         page = vmalloc_to_page(kaddr);
2057                 else
2058                         page = virt_to_page(kaddr);
2059
2060                 p[k] = page;
2061                 len -= seg;
2062                 kaddr += PAGE_SIZE;
2063         }
2064
2065         maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset);
2066         iov_iter_advance(i, maxsize);
2067         return maxsize;
2068 }
2069
2070 /*
2071  * Extract a list of contiguous pages from a user iterator and get a pin on
2072  * each of them.  This should only be used if the iterator is user-backed
2073  * (IOBUF/UBUF).
2074  *
2075  * It does not get refs on the pages, but the pages must be unpinned by the
2076  * caller once the transfer is complete.
2077  *
2078  * This is safe to be used where background IO/DMA *is* going to be modifying
2079  * the buffer; using a pin rather than a ref makes forces fork() to give the
2080  * child a copy of the page.
2081  */
2082 static ssize_t iov_iter_extract_user_pages(struct iov_iter *i,
2083                                            struct page ***pages,
2084                                            size_t maxsize,
2085                                            unsigned int maxpages,
2086                                            iov_iter_extraction_t extraction_flags,
2087                                            size_t *offset0)
2088 {
2089         unsigned long addr;
2090         unsigned int gup_flags = 0;
2091         size_t offset;
2092         int res;
2093
2094         if (i->data_source == ITER_DEST)
2095                 gup_flags |= FOLL_WRITE;
2096         if (extraction_flags & ITER_ALLOW_P2PDMA)
2097                 gup_flags |= FOLL_PCI_P2PDMA;
2098         if (i->nofault)
2099                 gup_flags |= FOLL_NOFAULT;
2100
2101         addr = first_iovec_segment(i, &maxsize);
2102         *offset0 = offset = addr % PAGE_SIZE;
2103         addr &= PAGE_MASK;
2104         maxpages = want_pages_array(pages, maxsize, offset, maxpages);
2105         if (!maxpages)
2106                 return -ENOMEM;
2107         res = pin_user_pages_fast(addr, maxpages, gup_flags, *pages);
2108         if (unlikely(res <= 0))
2109                 return res;
2110         maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - offset);
2111         iov_iter_advance(i, maxsize);
2112         return maxsize;
2113 }
2114
2115 /**
2116  * iov_iter_extract_pages - Extract a list of contiguous pages from an iterator
2117  * @i: The iterator to extract from
2118  * @pages: Where to return the list of pages
2119  * @maxsize: The maximum amount of iterator to extract
2120  * @maxpages: The maximum size of the list of pages
2121  * @extraction_flags: Flags to qualify request
2122  * @offset0: Where to return the starting offset into (*@pages)[0]
2123  *
2124  * Extract a list of contiguous pages from the current point of the iterator,
2125  * advancing the iterator.  The maximum number of pages and the maximum amount
2126  * of page contents can be set.
2127  *
2128  * If *@pages is NULL, a page list will be allocated to the required size and
2129  * *@pages will be set to its base.  If *@pages is not NULL, it will be assumed
2130  * that the caller allocated a page list at least @maxpages in size and this
2131  * will be filled in.
2132  *
2133  * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA
2134  * be allowed on the pages extracted.
2135  *
2136  * The iov_iter_extract_will_pin() function can be used to query how cleanup
2137  * should be performed.
2138  *
2139  * Extra refs or pins on the pages may be obtained as follows:
2140  *
2141  *  (*) If the iterator is user-backed (ITER_IOVEC/ITER_UBUF), pins will be
2142  *      added to the pages, but refs will not be taken.
2143  *      iov_iter_extract_will_pin() will return true.
2144  *
2145  *  (*) If the iterator is ITER_KVEC, ITER_BVEC or ITER_XARRAY, the pages are
2146  *      merely listed; no extra refs or pins are obtained.
2147  *      iov_iter_extract_will_pin() will return 0.
2148  *
2149  * Note also:
2150  *
2151  *  (*) Use with ITER_DISCARD is not supported as that has no content.
2152  *
2153  * On success, the function sets *@pages to the new pagelist, if allocated, and
2154  * sets *offset0 to the offset into the first page.
2155  *
2156  * It may also return -ENOMEM and -EFAULT.
2157  */
2158 ssize_t iov_iter_extract_pages(struct iov_iter *i,
2159                                struct page ***pages,
2160                                size_t maxsize,
2161                                unsigned int maxpages,
2162                                iov_iter_extraction_t extraction_flags,
2163                                size_t *offset0)
2164 {
2165         maxsize = min_t(size_t, min_t(size_t, maxsize, i->count), MAX_RW_COUNT);
2166         if (!maxsize)
2167                 return 0;
2168
2169         if (likely(user_backed_iter(i)))
2170                 return iov_iter_extract_user_pages(i, pages, maxsize,
2171                                                    maxpages, extraction_flags,
2172                                                    offset0);
2173         if (iov_iter_is_kvec(i))
2174                 return iov_iter_extract_kvec_pages(i, pages, maxsize,
2175                                                    maxpages, extraction_flags,
2176                                                    offset0);
2177         if (iov_iter_is_bvec(i))
2178                 return iov_iter_extract_bvec_pages(i, pages, maxsize,
2179                                                    maxpages, extraction_flags,
2180                                                    offset0);
2181         if (iov_iter_is_xarray(i))
2182                 return iov_iter_extract_xarray_pages(i, pages, maxsize,
2183                                                      maxpages, extraction_flags,
2184                                                      offset0);
2185         return -EFAULT;
2186 }
2187 EXPORT_SYMBOL_GPL(iov_iter_extract_pages);