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         if (should_fail_usercopy())
 178                 return n;
 179         if (access_ok(from, n)) {
 180                 instrument_copy_from_user(to, from, n);
 181                 n = raw_copy_from_user(to, from, n);
 182         }
 183         return n;
 184 }
 185
 186 static inline struct pipe_buffer *pipe_buf(const struct pipe_inode_info *pipe,
 187                                            unsigned int slot)
 188 {
 189         return &pipe->bufs[slot & (pipe->ring_size - 1)];
 190 }
 191
 192 #ifdef PIPE_PARANOIA
 193 static bool sanity(const struct iov_iter *i)
 194 {
 195         struct pipe_inode_info *pipe = i->pipe;
 196         unsigned int p_head = pipe->head;
 197         unsigned int p_tail = pipe->tail;
 198         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 199         unsigned int i_head = i->head;
 200         unsigned int idx;
 201
 202         if (i->last_offset) {
 203                 struct pipe_buffer *p;
 204                 if (unlikely(p_occupancy == 0))
 205                         goto Bad;       // pipe must be non-empty
 206                 if (unlikely(i_head != p_head - 1))
 207                         goto Bad;       // must be at the last buffer...
 208
 209                 p = pipe_buf(pipe, i_head);
 210                 if (unlikely(p->offset + p->len != abs(i->last_offset)))
 211                         goto Bad;       // ... at the end of segment
 212         } else {
 213                 if (i_head != p_head)
 214                         goto Bad;       // must be right after the last buffer
 215         }
 216         return true;
 217 Bad:
 218         printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset);
 219         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 220                         p_head, p_tail, pipe->ring_size);
 221         for (idx = 0; idx < pipe->ring_size; idx++)
 222                 printk(KERN_ERR "[%p %p %d %d]\n",
 223                         pipe->bufs[idx].ops,
 224                         pipe->bufs[idx].page,
 225                         pipe->bufs[idx].offset,
 226                         pipe->bufs[idx].len);
 227         WARN_ON(1);
 228         return false;
 229 }
 230 #else
 231 #define sanity(i) true
 232 #endif
 233
 234 static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size)
 235 {
 236         struct page *page = alloc_page(GFP_USER);
 237         if (page) {
 238                 struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 239                 *buf = (struct pipe_buffer) {
 240                         .ops = &default_pipe_buf_ops,
 241                         .page = page,
 242                         .offset = 0,
 243                         .len = size
 244                 };
 245         }
 246         return page;
 247 }
 248
 249 static void push_page(struct pipe_inode_info *pipe, struct page *page,
 250                         unsigned int offset, unsigned int size)
 251 {
 252         struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 253         *buf = (struct pipe_buffer) {
 254                 .ops = &page_cache_pipe_buf_ops,
 255                 .page = page,
 256                 .offset = offset,
 257                 .len = size
 258         };
 259         get_page(page);
 260 }
 261
 262 static inline int last_offset(const struct pipe_buffer *buf)
 263 {
 264         if (buf->ops == &default_pipe_buf_ops)
 265                 return buf->len;        // buf->offset is 0 for those
 266         else
 267                 return -(buf->offset + buf->len);
 268 }
 269
 270 static struct page *append_pipe(struct iov_iter *i, size_t size,
 271                                 unsigned int *off)
 272 {
 273         struct pipe_inode_info *pipe = i->pipe;
 274         int offset = i->last_offset;
 275         struct pipe_buffer *buf;
 276         struct page *page;
 277
 278         if (offset > 0 && offset < PAGE_SIZE) {
 279                 // some space in the last buffer; add to it
 280                 buf = pipe_buf(pipe, pipe->head - 1);
 281                 size = min_t(size_t, size, PAGE_SIZE - offset);
 282                 buf->len += size;
 283                 i->last_offset += size;
 284                 i->count -= size;
 285                 *off = offset;
 286                 return buf->page;
 287         }
 288         // OK, we need a new buffer
 289         *off = 0;
 290         size = min_t(size_t, size, PAGE_SIZE);
 291         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 292                 return NULL;
 293         page = push_anon(pipe, size);
 294         if (!page)
 295                 return NULL;
 296         i->head = pipe->head - 1;
 297         i->last_offset = size;
 298         i->count -= size;
 299         return page;
 300 }
 301
 302 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 303                          struct iov_iter *i)
 304 {
 305         struct pipe_inode_info *pipe = i->pipe;
 306         unsigned int head = pipe->head;
 307
 308         if (unlikely(bytes > i->count))
 309                 bytes = i->count;
 310
 311         if (unlikely(!bytes))
 312                 return 0;
 313
 314         if (!sanity(i))
 315                 return 0;
 316
 317         if (offset && i->last_offset == -offset) { // could we merge it?
 318                 struct pipe_buffer *buf = pipe_buf(pipe, head - 1);
 319                 if (buf->page == page) {
 320                         buf->len += bytes;
 321                         i->last_offset -= bytes;
 322                         i->count -= bytes;
 323                         return bytes;
 324                 }
 325         }
 326         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 327                 return 0;
 328
 329         push_page(pipe, page, offset, bytes);
 330         i->last_offset = -(offset + bytes);
 331         i->head = head;
 332         i->count -= bytes;
 333         return bytes;
 334 }
 335
 336 /*
 337  * fault_in_iov_iter_readable - fault in iov iterator for reading
 338  * @i: iterator
 339  * @size: maximum length
 340  *
 341  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 342  * @size.  For each iovec, fault in each page that constitutes the iovec.
 343  *
 344  * Returns the number of bytes not faulted in (like copy_to_user() and
 345  * copy_from_user()).
 346  *
 347  * Always returns 0 for non-userspace iterators.
 348  */
 349 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
 350 {
 351         if (iter_is_ubuf(i)) {
 352                 size_t n = min(size, iov_iter_count(i));
 353                 n -= fault_in_readable(i->ubuf + i->iov_offset, n);
 354                 return size - n;
 355         } else if (iter_is_iovec(i)) {
 356                 size_t count = min(size, iov_iter_count(i));
 357                 const struct iovec *p;
 358                 size_t skip;
 359
 360                 size -= count;
 361                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 362                         size_t len = min(count, p->iov_len - skip);
 363                         size_t ret;
 364
 365                         if (unlikely(!len))
 366                                 continue;
 367                         ret = fault_in_readable(p->iov_base + skip, len);
 368                         count -= len - ret;
 369                         if (ret)
 370                                 break;
 371                 }
 372                 return count + size;
 373         }
 374         return 0;
 375 }
 376 EXPORT_SYMBOL(fault_in_iov_iter_readable);
 377
 378 /*
 379  * fault_in_iov_iter_writeable - fault in iov iterator for writing
 380  * @i: iterator
 381  * @size: maximum length
 382  *
 383  * Faults in the iterator using get_user_pages(), i.e., without triggering
 384  * hardware page faults.  This is primarily useful when we already know that
 385  * some or all of the pages in @i aren't in memory.
 386  *
 387  * Returns the number of bytes not faulted in, like copy_to_user() and
 388  * copy_from_user().
 389  *
 390  * Always returns 0 for non-user-space iterators.
 391  */
 392 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
 393 {
 394         if (iter_is_ubuf(i)) {
 395                 size_t n = min(size, iov_iter_count(i));
 396                 n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n);
 397                 return size - n;
 398         } else if (iter_is_iovec(i)) {
 399                 size_t count = min(size, iov_iter_count(i));
 400                 const struct iovec *p;
 401                 size_t skip;
 402
 403                 size -= count;
 404                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 405                         size_t len = min(count, p->iov_len - skip);
 406                         size_t ret;
 407
 408                         if (unlikely(!len))
 409                                 continue;
 410                         ret = fault_in_safe_writeable(p->iov_base + skip, len);
 411                         count -= len - ret;
 412                         if (ret)
 413                                 break;
 414                 }
 415                 return count + size;
 416         }
 417         return 0;
 418 }
 419 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
 420
 421 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 422                         const struct iovec *iov, unsigned long nr_segs,
 423                         size_t count)
 424 {
 425         WARN_ON(direction & ~(READ | WRITE));
 426         *i = (struct iov_iter) {
 427                 .iter_type = ITER_IOVEC,
 428                 .nofault = false,
 429                 .user_backed = true,
 430                 .data_source = direction,
 431                 .iov = iov,
 432                 .nr_segs = nr_segs,
 433                 .iov_offset = 0,
 434                 .count = count
 435         };
 436 }
 437 EXPORT_SYMBOL(iov_iter_init);
 438
 439 // returns the offset in partial buffer (if any)
 440 static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages)
 441 {
 442         struct pipe_inode_info *pipe = i->pipe;
 443         int used = pipe->head - pipe->tail;
 444         int off = i->last_offset;
 445
 446         *npages = max((int)pipe->max_usage - used, 0);
 447
 448         if (off > 0 && off < PAGE_SIZE) { // anon and not full
 449                 (*npages)++;
 450                 return off;
 451         }
 452         return 0;
 453 }
 454
 455 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 456                                 struct iov_iter *i)
 457 {
 458         unsigned int off, chunk;
 459
 460         if (unlikely(bytes > i->count))
 461                 bytes = i->count;
 462         if (unlikely(!bytes))
 463                 return 0;
 464
 465         if (!sanity(i))
 466                 return 0;
 467
 468         for (size_t n = bytes; n; n -= chunk) {
 469                 struct page *page = append_pipe(i, n, &off);
 470                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 471                 if (!page)
 472                         return bytes - n;
 473                 memcpy_to_page(page, off, addr, chunk);
 474                 addr += chunk;
 475         }
 476         return bytes;
 477 }
 478
 479 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 480                               __wsum sum, size_t off)
 481 {
 482         __wsum next = csum_partial_copy_nocheck(from, to, len);
 483         return csum_block_add(sum, next, off);
 484 }
 485
 486 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 487                                          struct iov_iter *i, __wsum *sump)
 488 {
 489         __wsum sum = *sump;
 490         size_t off = 0;
 491         unsigned int chunk, r;
 492
 493         if (unlikely(bytes > i->count))
 494                 bytes = i->count;
 495         if (unlikely(!bytes))
 496                 return 0;
 497
 498         if (!sanity(i))
 499                 return 0;
 500
 501         while (bytes) {
 502                 struct page *page = append_pipe(i, bytes, &r);
 503                 char *p;
 504
 505                 if (!page)
 506                         break;
 507                 chunk = min_t(size_t, bytes, PAGE_SIZE - r);
 508                 p = kmap_local_page(page);
 509                 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
 510                 kunmap_local(p);
 511                 off += chunk;
 512                 bytes -= chunk;
 513         }
 514         *sump = sum;
 515         return off;
 516 }
 517
 518 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 519 {
 520         if (unlikely(iov_iter_is_pipe(i)))
 521                 return copy_pipe_to_iter(addr, bytes, i);
 522         if (user_backed_iter(i))
 523                 might_fault();
 524         iterate_and_advance(i, bytes, base, len, off,
 525                 copyout(base, addr + off, len),
 526                 memcpy(base, addr + off, len)
 527         )
 528
 529         return bytes;
 530 }
 531 EXPORT_SYMBOL(_copy_to_iter);
 532
 533 #ifdef CONFIG_ARCH_HAS_COPY_MC
 534 static int copyout_mc(void __user *to, const void *from, size_t n)
 535 {
 536         if (access_ok(to, n)) {
 537                 instrument_copy_to_user(to, from, n);
 538                 n = copy_mc_to_user((__force void *) to, from, n);
 539         }
 540         return n;
 541 }
 542
 543 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 544                                 struct iov_iter *i)
 545 {
 546         size_t xfer = 0;
 547         unsigned int off, chunk;
 548
 549         if (unlikely(bytes > i->count))
 550                 bytes = i->count;
 551         if (unlikely(!bytes))
 552                 return 0;
 553
 554         if (!sanity(i))
 555                 return 0;
 556
 557         while (bytes) {
 558                 struct page *page = append_pipe(i, bytes, &off);
 559                 unsigned long rem;
 560                 char *p;
 561
 562                 if (!page)
 563                         break;
 564                 chunk = min_t(size_t, bytes, PAGE_SIZE - off);
 565                 p = kmap_local_page(page);
 566                 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
 567                 chunk -= rem;
 568                 kunmap_local(p);
 569                 xfer += chunk;
 570                 bytes -= chunk;
 571                 if (rem) {
 572                         iov_iter_revert(i, rem);
 573                         break;
 574                 }
 575         }
 576         return xfer;
 577 }
 578
 579 /**
 580  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 581  * @addr: source kernel address
 582  * @bytes: total transfer length
 583  * @i: destination iterator
 584  *
 585  * The pmem driver deploys this for the dax operation
 586  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 587  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 588  * successfully copied.
 589  *
 590  * The main differences between this and typical _copy_to_iter().
 591  *
 592  * * Typical tail/residue handling after a fault retries the copy
 593  *   byte-by-byte until the fault happens again. Re-triggering machine
 594  *   checks is potentially fatal so the implementation uses source
 595  *   alignment and poison alignment assumptions to avoid re-triggering
 596  *   hardware exceptions.
 597  *
 598  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 599  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 600  *   a short copy.
 601  *
 602  * Return: number of bytes copied (may be %0)
 603  */
 604 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 605 {
 606         if (unlikely(iov_iter_is_pipe(i)))
 607                 return copy_mc_pipe_to_iter(addr, bytes, i);
 608         if (user_backed_iter(i))
 609                 might_fault();
 610         __iterate_and_advance(i, bytes, base, len, off,
 611                 copyout_mc(base, addr + off, len),
 612                 copy_mc_to_kernel(base, addr + off, len)
 613         )
 614
 615         return bytes;
 616 }
 617 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 618 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 619
 620 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 621 {
 622         if (unlikely(iov_iter_is_pipe(i))) {
 623                 WARN_ON(1);
 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 (unlikely(iov_iter_is_pipe(i))) {
 640                 WARN_ON(1);
 641                 return 0;
 642         }
 643         iterate_and_advance(i, bytes, base, len, off,
 644                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 645                 memcpy(addr + off, base, len)
 646         )
 647
 648         return bytes;
 649 }
 650 EXPORT_SYMBOL(_copy_from_iter_nocache);
 651
 652 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 653 /**
 654  * _copy_from_iter_flushcache - write destination through cpu cache
 655  * @addr: destination kernel address
 656  * @bytes: total transfer length
 657  * @i: source iterator
 658  *
 659  * The pmem driver arranges for filesystem-dax to use this facility via
 660  * dax_copy_from_iter() for ensuring that writes to persistent memory
 661  * are flushed through the CPU cache. It is differentiated from
 662  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 663  * all iterator types. The _copy_from_iter_nocache() only attempts to
 664  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 665  * instructions that strand dirty-data in the cache.
 666  *
 667  * Return: number of bytes copied (may be %0)
 668  */
 669 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 670 {
 671         if (unlikely(iov_iter_is_pipe(i))) {
 672                 WARN_ON(1);
 673                 return 0;
 674         }
 675         iterate_and_advance(i, bytes, base, len, off,
 676                 __copy_from_user_flushcache(addr + off, base, len),
 677                 memcpy_flushcache(addr + off, base, len)
 678         )
 679
 680         return bytes;
 681 }
 682 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 683 #endif
 684
 685 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 686 {
 687         struct page *head;
 688         size_t v = n + offset;
 689
 690         /*
 691          * The general case needs to access the page order in order
 692          * to compute the page size.
 693          * However, we mostly deal with order-0 pages and thus can
 694          * avoid a possible cache line miss for requests that fit all
 695          * page orders.
 696          */
 697         if (n <= v && v <= PAGE_SIZE)
 698                 return true;
 699
 700         head = compound_head(page);
 701         v += (page - head) << PAGE_SHIFT;
 702
 703         if (likely(n <= v && v <= (page_size(head))))
 704                 return true;
 705         WARN_ON(1);
 706         return false;
 707 }
 708
 709 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 710                          struct iov_iter *i)
 711 {
 712         size_t res = 0;
 713         if (unlikely(!page_copy_sane(page, offset, bytes)))
 714                 return 0;
 715         if (unlikely(iov_iter_is_pipe(i)))
 716                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 717         page += offset / PAGE_SIZE; // first subpage
 718         offset %= PAGE_SIZE;
 719         while (1) {
 720                 void *kaddr = kmap_local_page(page);
 721                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 722                 n = _copy_to_iter(kaddr + offset, n, i);
 723                 kunmap_local(kaddr);
 724                 res += n;
 725                 bytes -= n;
 726                 if (!bytes || !n)
 727                         break;
 728                 offset += n;
 729                 if (offset == PAGE_SIZE) {
 730                         page++;
 731                         offset = 0;
 732                 }
 733         }
 734         return res;
 735 }
 736 EXPORT_SYMBOL(copy_page_to_iter);
 737
 738 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 739                          struct iov_iter *i)
 740 {
 741         size_t res = 0;
 742         if (!page_copy_sane(page, offset, bytes))
 743                 return 0;
 744         page += offset / PAGE_SIZE; // first subpage
 745         offset %= PAGE_SIZE;
 746         while (1) {
 747                 void *kaddr = kmap_local_page(page);
 748                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 749                 n = _copy_from_iter(kaddr + offset, n, i);
 750                 kunmap_local(kaddr);
 751                 res += n;
 752                 bytes -= n;
 753                 if (!bytes || !n)
 754                         break;
 755                 offset += n;
 756                 if (offset == PAGE_SIZE) {
 757                         page++;
 758                         offset = 0;
 759                 }
 760         }
 761         return res;
 762 }
 763 EXPORT_SYMBOL(copy_page_from_iter);
 764
 765 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 766 {
 767         unsigned int chunk, off;
 768
 769         if (unlikely(bytes > i->count))
 770                 bytes = i->count;
 771         if (unlikely(!bytes))
 772                 return 0;
 773
 774         if (!sanity(i))
 775                 return 0;
 776
 777         for (size_t n = bytes; n; n -= chunk) {
 778                 struct page *page = append_pipe(i, n, &off);
 779                 char *p;
 780
 781                 if (!page)
 782                         return bytes - n;
 783                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 784                 p = kmap_local_page(page);
 785                 memset(p + off, 0, chunk);
 786                 kunmap_local(p);
 787         }
 788         return bytes;
 789 }
 790
 791 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 792 {
 793         if (unlikely(iov_iter_is_pipe(i)))
 794                 return pipe_zero(bytes, i);
 795         iterate_and_advance(i, bytes, base, len, count,
 796                 clear_user(base, len),
 797                 memset(base, 0, len)
 798         )
 799
 800         return bytes;
 801 }
 802 EXPORT_SYMBOL(iov_iter_zero);
 803
 804 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 805                                   struct iov_iter *i)
 806 {
 807         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 808         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 809                 kunmap_atomic(kaddr);
 810                 return 0;
 811         }
 812         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 813                 kunmap_atomic(kaddr);
 814                 WARN_ON(1);
 815                 return 0;
 816         }
 817         iterate_and_advance(i, bytes, base, len, off,
 818                 copyin(p + off, base, len),
 819                 memcpy(p + off, base, len)
 820         )
 821         kunmap_atomic(kaddr);
 822         return bytes;
 823 }
 824 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 825
 826 static void pipe_advance(struct iov_iter *i, size_t size)
 827 {
 828         struct pipe_inode_info *pipe = i->pipe;
 829         int off = i->last_offset;
 830
 831         if (!off && !size) {
 832                 pipe_discard_from(pipe, i->start_head); // discard everything
 833                 return;
 834         }
 835         i->count -= size;
 836         while (1) {
 837                 struct pipe_buffer *buf = pipe_buf(pipe, i->head);
 838                 if (off) /* make it relative to the beginning of buffer */
 839                         size += abs(off) - buf->offset;
 840                 if (size <= buf->len) {
 841                         buf->len = size;
 842                         i->last_offset = last_offset(buf);
 843                         break;
 844                 }
 845                 size -= buf->len;
 846                 i->head++;
 847                 off = 0;
 848         }
 849         pipe_discard_from(pipe, i->head + 1); // discard everything past this one
 850 }
 851
 852 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 853 {
 854         const struct bio_vec *bvec, *end;
 855
 856         if (!i->count)
 857                 return;
 858         i->count -= size;
 859
 860         size += i->iov_offset;
 861
 862         for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
 863                 if (likely(size < bvec->bv_len))
 864                         break;
 865                 size -= bvec->bv_len;
 866         }
 867         i->iov_offset = size;
 868         i->nr_segs -= bvec - i->bvec;
 869         i->bvec = bvec;
 870 }
 871
 872 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 873 {
 874         const struct iovec *iov, *end;
 875
 876         if (!i->count)
 877                 return;
 878         i->count -= size;
 879
 880         size += i->iov_offset; // from beginning of current segment
 881         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
 882                 if (likely(size < iov->iov_len))
 883                         break;
 884                 size -= iov->iov_len;
 885         }
 886         i->iov_offset = size;
 887         i->nr_segs -= iov - i->iov;
 888         i->iov = iov;
 889 }
 890
 891 void iov_iter_advance(struct iov_iter *i, size_t size)
 892 {
 893         if (unlikely(i->count < size))
 894                 size = i->count;
 895         if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
 896                 i->iov_offset += size;
 897                 i->count -= size;
 898         } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 899                 /* iovec and kvec have identical layouts */
 900                 iov_iter_iovec_advance(i, size);
 901         } else if (iov_iter_is_bvec(i)) {
 902                 iov_iter_bvec_advance(i, size);
 903         } else if (iov_iter_is_pipe(i)) {
 904                 pipe_advance(i, size);
 905         } else if (iov_iter_is_discard(i)) {
 906                 i->count -= size;
 907         }
 908 }
 909 EXPORT_SYMBOL(iov_iter_advance);
 910
 911 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 912 {
 913         if (!unroll)
 914                 return;
 915         if (WARN_ON(unroll > MAX_RW_COUNT))
 916                 return;
 917         i->count += unroll;
 918         if (unlikely(iov_iter_is_pipe(i))) {
 919                 struct pipe_inode_info *pipe = i->pipe;
 920                 unsigned int head = pipe->head;
 921
 922                 while (head > i->start_head) {
 923                         struct pipe_buffer *b = pipe_buf(pipe, --head);
 924                         if (unroll < b->len) {
 925                                 b->len -= unroll;
 926                                 i->last_offset = last_offset(b);
 927                                 i->head = head;
 928                                 return;
 929                         }
 930                         unroll -= b->len;
 931                         pipe_buf_release(pipe, b);
 932                         pipe->head--;
 933                 }
 934                 i->last_offset = 0;
 935                 i->head = head;
 936                 return;
 937         }
 938         if (unlikely(iov_iter_is_discard(i)))
 939                 return;
 940         if (unroll <= i->iov_offset) {
 941                 i->iov_offset -= unroll;
 942                 return;
 943         }
 944         unroll -= i->iov_offset;
 945         if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
 946                 BUG(); /* We should never go beyond the start of the specified
 947                         * range since we might then be straying into pages that
 948                         * aren't pinned.
 949                         */
 950         } else if (iov_iter_is_bvec(i)) {
 951                 const struct bio_vec *bvec = i->bvec;
 952                 while (1) {
 953                         size_t n = (--bvec)->bv_len;
 954                         i->nr_segs++;
 955                         if (unroll <= n) {
 956                                 i->bvec = bvec;
 957                                 i->iov_offset = n - unroll;
 958                                 return;
 959                         }
 960                         unroll -= n;
 961                 }
 962         } else { /* same logics for iovec and kvec */
 963                 const struct iovec *iov = i->iov;
 964                 while (1) {
 965                         size_t n = (--iov)->iov_len;
 966                         i->nr_segs++;
 967                         if (unroll <= n) {
 968                                 i->iov = iov;
 969                                 i->iov_offset = n - unroll;
 970                                 return;
 971                         }
 972                         unroll -= n;
 973                 }
 974         }
 975 }
 976 EXPORT_SYMBOL(iov_iter_revert);
 977
 978 /*
 979  * Return the count of just the current iov_iter segment.
 980  */
 981 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 982 {
 983         if (i->nr_segs > 1) {
 984                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 985                         return min(i->count, i->iov->iov_len - i->iov_offset);
 986                 if (iov_iter_is_bvec(i))
 987                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 988         }
 989         return i->count;
 990 }
 991 EXPORT_SYMBOL(iov_iter_single_seg_count);
 992
 993 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
 994                         const struct kvec *kvec, unsigned long nr_segs,
 995                         size_t count)
 996 {
 997         WARN_ON(direction & ~(READ | WRITE));
 998         *i = (struct iov_iter){
 999                 .iter_type = ITER_KVEC,
1000                 .data_source = direction,
1001                 .kvec = kvec,
1002                 .nr_segs = nr_segs,
1003                 .iov_offset = 0,
1004                 .count = count
1005         };
1006 }
1007 EXPORT_SYMBOL(iov_iter_kvec);
1008
1009 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1010                         const struct bio_vec *bvec, unsigned long nr_segs,
1011                         size_t count)
1012 {
1013         WARN_ON(direction & ~(READ | WRITE));
1014         *i = (struct iov_iter){
1015                 .iter_type = ITER_BVEC,
1016                 .data_source = direction,
1017                 .bvec = bvec,
1018                 .nr_segs = nr_segs,
1019                 .iov_offset = 0,
1020                 .count = count
1021         };
1022 }
1023 EXPORT_SYMBOL(iov_iter_bvec);
1024
1025 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1026                         struct pipe_inode_info *pipe,
1027                         size_t count)
1028 {
1029         BUG_ON(direction != READ);
1030         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1031         *i = (struct iov_iter){
1032                 .iter_type = ITER_PIPE,
1033                 .data_source = false,
1034                 .pipe = pipe,
1035                 .head = pipe->head,
1036                 .start_head = pipe->head,
1037                 .last_offset = 0,
1038                 .count = count
1039         };
1040 }
1041 EXPORT_SYMBOL(iov_iter_pipe);
1042
1043 /**
1044  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1045  * @i: The iterator to initialise.
1046  * @direction: The direction of the transfer.
1047  * @xarray: The xarray to access.
1048  * @start: The start file position.
1049  * @count: The size of the I/O buffer in bytes.
1050  *
1051  * Set up an I/O iterator to either draw data out of the pages attached to an
1052  * inode or to inject data into those pages.  The pages *must* be prevented
1053  * from evaporation, either by taking a ref on them or locking them by the
1054  * caller.
1055  */
1056 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1057                      struct xarray *xarray, loff_t start, size_t count)
1058 {
1059         BUG_ON(direction & ~1);
1060         *i = (struct iov_iter) {
1061                 .iter_type = ITER_XARRAY,
1062                 .data_source = direction,
1063                 .xarray = xarray,
1064                 .xarray_start = start,
1065                 .count = count,
1066                 .iov_offset = 0
1067         };
1068 }
1069 EXPORT_SYMBOL(iov_iter_xarray);
1070
1071 /**
1072  * iov_iter_discard - Initialise an I/O iterator that discards data
1073  * @i: The iterator to initialise.
1074  * @direction: The direction of the transfer.
1075  * @count: The size of the I/O buffer in bytes.
1076  *
1077  * Set up an I/O iterator that just discards everything that's written to it.
1078  * It's only available as a READ iterator.
1079  */
1080 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1081 {
1082         BUG_ON(direction != READ);
1083         *i = (struct iov_iter){
1084                 .iter_type = ITER_DISCARD,
1085                 .data_source = false,
1086                 .count = count,
1087                 .iov_offset = 0
1088         };
1089 }
1090 EXPORT_SYMBOL(iov_iter_discard);
1091
1092 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1093                                    unsigned len_mask)
1094 {
1095         size_t size = i->count;
1096         size_t skip = i->iov_offset;
1097         unsigned k;
1098
1099         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1100                 size_t len = i->iov[k].iov_len - skip;
1101
1102                 if (len > size)
1103                         len = size;
1104                 if (len & len_mask)
1105                         return false;
1106                 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1107                         return false;
1108
1109                 size -= len;
1110                 if (!size)
1111                         break;
1112         }
1113         return true;
1114 }
1115
1116 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1117                                   unsigned len_mask)
1118 {
1119         size_t size = i->count;
1120         unsigned skip = i->iov_offset;
1121         unsigned k;
1122
1123         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1124                 size_t len = i->bvec[k].bv_len - skip;
1125
1126                 if (len > size)
1127                         len = size;
1128                 if (len & len_mask)
1129                         return false;
1130                 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1131                         return false;
1132
1133                 size -= len;
1134                 if (!size)
1135                         break;
1136         }
1137         return true;
1138 }
1139
1140 /**
1141  * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1142  *      are aligned to the parameters.
1143  *
1144  * @i: &struct iov_iter to restore
1145  * @addr_mask: bit mask to check against the iov element's addresses
1146  * @len_mask: bit mask to check against the iov element's lengths
1147  *
1148  * Return: false if any addresses or lengths intersect with the provided masks
1149  */
1150 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1151                          unsigned len_mask)
1152 {
1153         if (likely(iter_is_ubuf(i))) {
1154                 if (i->count & len_mask)
1155                         return false;
1156                 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
1157                         return false;
1158                 return true;
1159         }
1160
1161         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1162                 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1163
1164         if (iov_iter_is_bvec(i))
1165                 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1166
1167         if (iov_iter_is_pipe(i)) {
1168                 size_t size = i->count;
1169
1170                 if (size & len_mask)
1171                         return false;
1172                 if (size && i->last_offset > 0) {
1173                         if (i->last_offset & addr_mask)
1174                                 return false;
1175                 }
1176
1177                 return true;
1178         }
1179
1180         if (iov_iter_is_xarray(i)) {
1181                 if (i->count & len_mask)
1182                         return false;
1183                 if ((i->xarray_start + i->iov_offset) & addr_mask)
1184                         return false;
1185         }
1186
1187         return true;
1188 }
1189 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1190
1191 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1192 {
1193         unsigned long res = 0;
1194         size_t size = i->count;
1195         size_t skip = i->iov_offset;
1196         unsigned k;
1197
1198         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1199                 size_t len = i->iov[k].iov_len - skip;
1200                 if (len) {
1201                         res |= (unsigned long)i->iov[k].iov_base + skip;
1202                         if (len > size)
1203                                 len = size;
1204                         res |= len;
1205                         size -= len;
1206                         if (!size)
1207                                 break;
1208                 }
1209         }
1210         return res;
1211 }
1212
1213 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1214 {
1215         unsigned res = 0;
1216         size_t size = i->count;
1217         unsigned skip = i->iov_offset;
1218         unsigned k;
1219
1220         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1221                 size_t len = i->bvec[k].bv_len - skip;
1222                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1223                 if (len > size)
1224                         len = size;
1225                 res |= len;
1226                 size -= len;
1227                 if (!size)
1228                         break;
1229         }
1230         return res;
1231 }
1232
1233 unsigned long iov_iter_alignment(const struct iov_iter *i)
1234 {
1235         if (likely(iter_is_ubuf(i))) {
1236                 size_t size = i->count;
1237                 if (size)
1238                         return ((unsigned long)i->ubuf + i->iov_offset) | size;
1239                 return 0;
1240         }
1241
1242         /* iovec and kvec have identical layouts */
1243         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1244                 return iov_iter_alignment_iovec(i);
1245
1246         if (iov_iter_is_bvec(i))
1247                 return iov_iter_alignment_bvec(i);
1248
1249         if (iov_iter_is_pipe(i)) {
1250                 size_t size = i->count;
1251
1252                 if (size && i->last_offset > 0)
1253                         return size | i->last_offset;
1254                 return size;
1255         }
1256
1257         if (iov_iter_is_xarray(i))
1258                 return (i->xarray_start + i->iov_offset) | i->count;
1259
1260         return 0;
1261 }
1262 EXPORT_SYMBOL(iov_iter_alignment);
1263
1264 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1265 {
1266         unsigned long res = 0;
1267         unsigned long v = 0;
1268         size_t size = i->count;
1269         unsigned k;
1270
1271         if (iter_is_ubuf(i))
1272                 return 0;
1273
1274         if (WARN_ON(!iter_is_iovec(i)))
1275                 return ~0U;
1276
1277         for (k = 0; k < i->nr_segs; k++) {
1278                 if (i->iov[k].iov_len) {
1279                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1280                         if (v) // if not the first one
1281                                 res |= base | v; // this start | previous end
1282                         v = base + i->iov[k].iov_len;
1283                         if (size <= i->iov[k].iov_len)
1284                                 break;
1285                         size -= i->iov[k].iov_len;
1286                 }
1287         }
1288         return res;
1289 }
1290 EXPORT_SYMBOL(iov_iter_gap_alignment);
1291
1292 static int want_pages_array(struct page ***res, size_t size,
1293                             size_t start, unsigned int maxpages)
1294 {
1295         unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
1296
1297         if (count > maxpages)
1298                 count = maxpages;
1299         WARN_ON(!count);        // caller should've prevented that
1300         if (!*res) {
1301                 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
1302                 if (!*res)
1303                         return 0;
1304         }
1305         return count;
1306 }
1307
1308 static ssize_t pipe_get_pages(struct iov_iter *i,
1309                    struct page ***pages, size_t maxsize, unsigned maxpages,
1310                    size_t *start)
1311 {
1312         unsigned int npages, count, off, chunk;
1313         struct page **p;
1314         size_t left;
1315
1316         if (!sanity(i))
1317                 return -EFAULT;
1318
1319         *start = off = pipe_npages(i, &npages);
1320         if (!npages)
1321                 return -EFAULT;
1322         count = want_pages_array(pages, maxsize, off, min(npages, maxpages));
1323         if (!count)
1324                 return -ENOMEM;
1325         p = *pages;
1326         for (npages = 0, left = maxsize ; npages < count; npages++, left -= chunk) {
1327                 struct page *page = append_pipe(i, left, &off);
1328                 if (!page)
1329                         break;
1330                 chunk = min_t(size_t, left, PAGE_SIZE - off);
1331                 get_page(*p++ = page);
1332         }
1333         if (!npages)
1334                 return -EFAULT;
1335         return maxsize - left;
1336 }
1337
1338 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1339                                           pgoff_t index, unsigned int nr_pages)
1340 {
1341         XA_STATE(xas, xa, index);
1342         struct page *page;
1343         unsigned int ret = 0;
1344
1345         rcu_read_lock();
1346         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1347                 if (xas_retry(&xas, page))
1348                         continue;
1349
1350                 /* Has the page moved or been split? */
1351                 if (unlikely(page != xas_reload(&xas))) {
1352                         xas_reset(&xas);
1353                         continue;
1354                 }
1355
1356                 pages[ret] = find_subpage(page, xas.xa_index);
1357                 get_page(pages[ret]);
1358                 if (++ret == nr_pages)
1359                         break;
1360         }
1361         rcu_read_unlock();
1362         return ret;
1363 }
1364
1365 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1366                                      struct page ***pages, size_t maxsize,
1367                                      unsigned maxpages, size_t *_start_offset)
1368 {
1369         unsigned nr, offset, count;
1370         pgoff_t index;
1371         loff_t pos;
1372
1373         pos = i->xarray_start + i->iov_offset;
1374         index = pos >> PAGE_SHIFT;
1375         offset = pos & ~PAGE_MASK;
1376         *_start_offset = offset;
1377
1378         count = want_pages_array(pages, maxsize, offset, maxpages);
1379         if (!count)
1380                 return -ENOMEM;
1381         nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
1382         if (nr == 0)
1383                 return 0;
1384
1385         maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1386         i->iov_offset += maxsize;
1387         i->count -= maxsize;
1388         return maxsize;
1389 }
1390
1391 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1392 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1393 {
1394         size_t skip;
1395         long k;
1396
1397         if (iter_is_ubuf(i))
1398                 return (unsigned long)i->ubuf + i->iov_offset;
1399
1400         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1401                 size_t len = i->iov[k].iov_len - skip;
1402
1403                 if (unlikely(!len))
1404                         continue;
1405                 if (*size > len)
1406                         *size = len;
1407                 return (unsigned long)i->iov[k].iov_base + skip;
1408         }
1409         BUG(); // if it had been empty, we wouldn't get called
1410 }
1411
1412 /* must be done on non-empty ITER_BVEC one */
1413 static struct page *first_bvec_segment(const struct iov_iter *i,
1414                                        size_t *size, size_t *start)
1415 {
1416         struct page *page;
1417         size_t skip = i->iov_offset, len;
1418
1419         len = i->bvec->bv_len - skip;
1420         if (*size > len)
1421                 *size = len;
1422         skip += i->bvec->bv_offset;
1423         page = i->bvec->bv_page + skip / PAGE_SIZE;
1424         *start = skip % PAGE_SIZE;
1425         return page;
1426 }
1427
1428 static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
1429                    struct page ***pages, size_t maxsize,
1430                    unsigned int maxpages, size_t *start)
1431 {
1432         unsigned int n;
1433
1434         if (maxsize > i->count)
1435                 maxsize = i->count;
1436         if (!maxsize)
1437                 return 0;
1438         if (maxsize > MAX_RW_COUNT)
1439                 maxsize = MAX_RW_COUNT;
1440
1441         if (likely(user_backed_iter(i))) {
1442                 unsigned int gup_flags = 0;
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_pages2(struct iov_iter *i,
1493                    struct page **pages, size_t maxsize, unsigned maxpages,
1494                    size_t *start)
1495 {
1496         if (!maxpages)
1497                 return 0;
1498         BUG_ON(!pages);
1499
1500         return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, start);
1501 }
1502 EXPORT_SYMBOL(iov_iter_get_pages2);
1503
1504 ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1505                    struct page ***pages, size_t maxsize,
1506                    size_t *start)
1507 {
1508         ssize_t len;
1509
1510         *pages = NULL;
1511
1512         len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start);
1513         if (len <= 0) {
1514                 kvfree(*pages);
1515                 *pages = NULL;
1516         }
1517         return len;
1518 }
1519 EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1520
1521 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1522                                struct iov_iter *i)
1523 {
1524         __wsum sum, next;
1525         sum = *csum;
1526         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1527                 WARN_ON(1);
1528                 return 0;
1529         }
1530         iterate_and_advance(i, bytes, base, len, off, ({
1531                 next = csum_and_copy_from_user(base, addr + off, len);
1532                 sum = csum_block_add(sum, next, off);
1533                 next ? 0 : len;
1534         }), ({
1535                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1536         })
1537         )
1538         *csum = sum;
1539         return bytes;
1540 }
1541 EXPORT_SYMBOL(csum_and_copy_from_iter);
1542
1543 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1544                              struct iov_iter *i)
1545 {
1546         struct csum_state *csstate = _csstate;
1547         __wsum sum, next;
1548
1549         if (unlikely(iov_iter_is_discard(i))) {
1550                 WARN_ON(1);     /* for now */
1551                 return 0;
1552         }
1553
1554         sum = csum_shift(csstate->csum, csstate->off);
1555         if (unlikely(iov_iter_is_pipe(i)))
1556                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1557         else iterate_and_advance(i, bytes, base, len, off, ({
1558                 next = csum_and_copy_to_user(addr + off, base, len);
1559                 sum = csum_block_add(sum, next, off);
1560                 next ? 0 : len;
1561         }), ({
1562                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1563         })
1564         )
1565         csstate->csum = csum_shift(sum, csstate->off);
1566         csstate->off += bytes;
1567         return bytes;
1568 }
1569 EXPORT_SYMBOL(csum_and_copy_to_iter);
1570
1571 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1572                 struct iov_iter *i)
1573 {
1574 #ifdef CONFIG_CRYPTO_HASH
1575         struct ahash_request *hash = hashp;
1576         struct scatterlist sg;
1577         size_t copied;
1578
1579         copied = copy_to_iter(addr, bytes, i);
1580         sg_init_one(&sg, addr, copied);
1581         ahash_request_set_crypt(hash, &sg, NULL, copied);
1582         crypto_ahash_update(hash);
1583         return copied;
1584 #else
1585         return 0;
1586 #endif
1587 }
1588 EXPORT_SYMBOL(hash_and_copy_to_iter);
1589
1590 static int iov_npages(const struct iov_iter *i, int maxpages)
1591 {
1592         size_t skip = i->iov_offset, size = i->count;
1593         const struct iovec *p;
1594         int npages = 0;
1595
1596         for (p = i->iov; size; skip = 0, p++) {
1597                 unsigned offs = offset_in_page(p->iov_base + skip);
1598                 size_t len = min(p->iov_len - skip, size);
1599
1600                 if (len) {
1601                         size -= len;
1602                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1603                         if (unlikely(npages > maxpages))
1604                                 return maxpages;
1605                 }
1606         }
1607         return npages;
1608 }
1609
1610 static int bvec_npages(const struct iov_iter *i, int maxpages)
1611 {
1612         size_t skip = i->iov_offset, size = i->count;
1613         const struct bio_vec *p;
1614         int npages = 0;
1615
1616         for (p = i->bvec; size; skip = 0, p++) {
1617                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1618                 size_t len = min(p->bv_len - skip, size);
1619
1620                 size -= len;
1621                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1622                 if (unlikely(npages > maxpages))
1623                         return maxpages;
1624         }
1625         return npages;
1626 }
1627
1628 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1629 {
1630         if (unlikely(!i->count))
1631                 return 0;
1632         if (likely(iter_is_ubuf(i))) {
1633                 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1634                 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1635                 return min(npages, maxpages);
1636         }
1637         /* iovec and kvec have identical layouts */
1638         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1639                 return iov_npages(i, maxpages);
1640         if (iov_iter_is_bvec(i))
1641                 return bvec_npages(i, maxpages);
1642         if (iov_iter_is_pipe(i)) {
1643                 int npages;
1644
1645                 if (!sanity(i))
1646                         return 0;
1647
1648                 pipe_npages(i, &npages);
1649                 return min(npages, maxpages);
1650         }
1651         if (iov_iter_is_xarray(i)) {
1652                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1653                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1654                 return min(npages, maxpages);
1655         }
1656         return 0;
1657 }
1658 EXPORT_SYMBOL(iov_iter_npages);
1659
1660 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1661 {
1662         *new = *old;
1663         if (unlikely(iov_iter_is_pipe(new))) {
1664                 WARN_ON(1);
1665                 return NULL;
1666         }
1667         if (iov_iter_is_bvec(new))
1668                 return new->bvec = kmemdup(new->bvec,
1669                                     new->nr_segs * sizeof(struct bio_vec),
1670                                     flags);
1671         else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1672                 /* iovec and kvec have identical layout */
1673                 return new->iov = kmemdup(new->iov,
1674                                    new->nr_segs * sizeof(struct iovec),
1675                                    flags);
1676         return NULL;
1677 }
1678 EXPORT_SYMBOL(dup_iter);
1679
1680 static int copy_compat_iovec_from_user(struct iovec *iov,
1681                 const struct iovec __user *uvec, unsigned long nr_segs)
1682 {
1683         const struct compat_iovec __user *uiov =
1684                 (const struct compat_iovec __user *)uvec;
1685         int ret = -EFAULT, i;
1686
1687         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1688                 return -EFAULT;
1689
1690         for (i = 0; i < nr_segs; i++) {
1691                 compat_uptr_t buf;
1692                 compat_ssize_t len;
1693
1694                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1695                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1696
1697                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1698                 if (len < 0) {
1699                         ret = -EINVAL;
1700                         goto uaccess_end;
1701                 }
1702                 iov[i].iov_base = compat_ptr(buf);
1703                 iov[i].iov_len = len;
1704         }
1705
1706         ret = 0;
1707 uaccess_end:
1708         user_access_end();
1709         return ret;
1710 }
1711
1712 static int copy_iovec_from_user(struct iovec *iov,
1713                 const struct iovec __user *uvec, unsigned long nr_segs)
1714 {
1715         unsigned long seg;
1716
1717         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1718                 return -EFAULT;
1719         for (seg = 0; seg < nr_segs; seg++) {
1720                 if ((ssize_t)iov[seg].iov_len < 0)
1721                         return -EINVAL;
1722         }
1723
1724         return 0;
1725 }
1726
1727 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1728                 unsigned long nr_segs, unsigned long fast_segs,
1729                 struct iovec *fast_iov, bool compat)
1730 {
1731         struct iovec *iov = fast_iov;
1732         int ret;
1733
1734         /*
1735          * SuS says "The readv() function *may* fail if the iovcnt argument was
1736          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1737          * traditionally returned zero for zero segments, so...
1738          */
1739         if (nr_segs == 0)
1740                 return iov;
1741         if (nr_segs > UIO_MAXIOV)
1742                 return ERR_PTR(-EINVAL);
1743         if (nr_segs > fast_segs) {
1744                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1745                 if (!iov)
1746                         return ERR_PTR(-ENOMEM);
1747         }
1748
1749         if (compat)
1750                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1751         else
1752                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1753         if (ret) {
1754                 if (iov != fast_iov)
1755                         kfree(iov);
1756                 return ERR_PTR(ret);
1757         }
1758
1759         return iov;
1760 }
1761
1762 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1763                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1764                  struct iov_iter *i, bool compat)
1765 {
1766         ssize_t total_len = 0;
1767         unsigned long seg;
1768         struct iovec *iov;
1769
1770         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1771         if (IS_ERR(iov)) {
1772                 *iovp = NULL;
1773                 return PTR_ERR(iov);
1774         }
1775
1776         /*
1777          * According to the Single Unix Specification we should return EINVAL if
1778          * an element length is < 0 when cast to ssize_t or if the total length
1779          * would overflow the ssize_t return value of the system call.
1780          *
1781          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1782          * overflow case.
1783          */
1784         for (seg = 0; seg < nr_segs; seg++) {
1785                 ssize_t len = (ssize_t)iov[seg].iov_len;
1786
1787                 if (!access_ok(iov[seg].iov_base, len)) {
1788                         if (iov != *iovp)
1789                                 kfree(iov);
1790                         *iovp = NULL;
1791                         return -EFAULT;
1792                 }
1793
1794                 if (len > MAX_RW_COUNT - total_len) {
1795                         len = MAX_RW_COUNT - total_len;
1796                         iov[seg].iov_len = len;
1797                 }
1798                 total_len += len;
1799         }
1800
1801         iov_iter_init(i, type, iov, nr_segs, total_len);
1802         if (iov == *iovp)
1803                 *iovp = NULL;
1804         else
1805                 *iovp = iov;
1806         return total_len;
1807 }
1808
1809 /**
1810  * import_iovec() - Copy an array of &struct iovec from userspace
1811  *     into the kernel, check that it is valid, and initialize a new
1812  *     &struct iov_iter iterator to access it.
1813  *
1814  * @type: One of %READ or %WRITE.
1815  * @uvec: Pointer to the userspace array.
1816  * @nr_segs: Number of elements in userspace array.
1817  * @fast_segs: Number of elements in @iov.
1818  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1819  *     on-stack) kernel array.
1820  * @i: Pointer to iterator that will be initialized on success.
1821  *
1822  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1823  * then this function places %NULL in *@iov on return. Otherwise, a new
1824  * array will be allocated and the result placed in *@iov. This means that
1825  * the caller may call kfree() on *@iov regardless of whether the small
1826  * on-stack array was used or not (and regardless of whether this function
1827  * returns an error or not).
1828  *
1829  * Return: Negative error code on error, bytes imported on success
1830  */
1831 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1832                  unsigned nr_segs, unsigned fast_segs,
1833                  struct iovec **iovp, struct iov_iter *i)
1834 {
1835         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1836                               in_compat_syscall());
1837 }
1838 EXPORT_SYMBOL(import_iovec);
1839
1840 int import_single_range(int rw, void __user *buf, size_t len,
1841                  struct iovec *iov, struct iov_iter *i)
1842 {
1843         if (len > MAX_RW_COUNT)
1844                 len = MAX_RW_COUNT;
1845         if (unlikely(!access_ok(buf, len)))
1846                 return -EFAULT;
1847
1848         iov->iov_base = buf;
1849         iov->iov_len = len;
1850         iov_iter_init(i, rw, iov, 1, len);
1851         return 0;
1852 }
1853 EXPORT_SYMBOL(import_single_range);
1854
1855 /**
1856  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1857  *     iov_iter_save_state() was called.
1858  *
1859  * @i: &struct iov_iter to restore
1860  * @state: state to restore from
1861  *
1862  * Used after iov_iter_save_state() to bring restore @i, if operations may
1863  * have advanced it.
1864  *
1865  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1866  */
1867 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1868 {
1869         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
1870                          !iov_iter_is_kvec(i) && !iter_is_ubuf(i))
1871                 return;
1872         i->iov_offset = state->iov_offset;
1873         i->count = state->count;
1874         if (iter_is_ubuf(i))
1875                 return;
1876         /*
1877          * For the *vec iters, nr_segs + iov is constant - if we increment
1878          * the vec, then we also decrement the nr_segs count. Hence we don't
1879          * need to track both of these, just one is enough and we can deduct
1880          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1881          * size, so we can just increment the iov pointer as they are unionzed.
1882          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1883          * not. Be safe and handle it separately.
1884          */
1885         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1886         if (iov_iter_is_bvec(i))
1887                 i->bvec -= state->nr_segs - i->nr_segs;
1888         else
1889                 i->iov -= state->nr_segs - i->nr_segs;
1890         i->nr_segs = state->nr_segs;
1891 }