lib/iov_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <crypto/hash.h>
   3 #include <linux/export.h>
   4 #include <linux/bvec.h>
   5 #include <linux/fault-inject-usercopy.h>
   6 #include <linux/uio.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/highmem.h>
   9 #include <linux/slab.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/splice.h>
  12 #include <linux/compat.h>
  13 #include <net/checksum.h>
  14 #include <linux/scatterlist.h>
  15 #include <linux/instrumented.h>
  16
  17 #define PIPE_PARANOIA /* for now */
  18
  19 /* covers ubuf and kbuf alike */
  20 #define iterate_buf(i, n, base, len, off, __p, STEP) {          \
  21         size_t __maybe_unused off = 0;                          \
  22         len = n;                                                \
  23         base = __p + i->iov_offset;                             \
  24         len -= (STEP);                                          \
  25         i->iov_offset += len;                                   \
  26         n = len;                                                \
  27 }
  28
  29 /* covers iovec and kvec alike */
  30 #define iterate_iovec(i, n, base, len, off, __p, STEP) {        \
  31         size_t off = 0;                                         \
  32         size_t skip = i->iov_offset;                            \
  33         do {                                                    \
  34                 len = min(n, __p->iov_len - skip);              \
  35                 if (likely(len)) {                              \
  36                         base = __p->iov_base + skip;            \
  37                         len -= (STEP);                          \
  38                         off += len;                             \
  39                         skip += len;                            \
  40                         n -= len;                               \
  41                         if (skip < __p->iov_len)                \
  42                                 break;                          \
  43                 }                                               \
  44                 __p++;                                          \
  45                 skip = 0;                                       \
  46         } while (n);                                            \
  47         i->iov_offset = skip;                                   \
  48         n = off;                                                \
  49 }
  50
  51 #define iterate_bvec(i, n, base, len, off, p, STEP) {           \
  52         size_t off = 0;                                         \
  53         unsigned skip = i->iov_offset;                          \
  54         while (n) {                                             \
  55                 unsigned offset = p->bv_offset + skip;          \
  56                 unsigned left;                                  \
  57                 void *kaddr = kmap_local_page(p->bv_page +      \
  58                                         offset / PAGE_SIZE);    \
  59                 base = kaddr + offset % PAGE_SIZE;              \
  60                 len = min(min(n, (size_t)(p->bv_len - skip)),   \
  61                      (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
  62                 left = (STEP);                                  \
  63                 kunmap_local(kaddr);                            \
  64                 len -= left;                                    \
  65                 off += len;                                     \
  66                 skip += len;                                    \
  67                 if (skip == p->bv_len) {                        \
  68                         skip = 0;                               \
  69                         p++;                                    \
  70                 }                                               \
  71                 n -= len;                                       \
  72                 if (left)                                       \
  73                         break;                                  \
  74         }                                                       \
  75         i->iov_offset = skip;                                   \
  76         n = off;                                                \
  77 }
  78
  79 #define iterate_xarray(i, n, base, len, __off, STEP) {          \
  80         __label__ __out;                                        \
  81         size_t __off = 0;                                       \
  82         struct folio *folio;                                    \
  83         loff_t start = i->xarray_start + i->iov_offset;         \
  84         pgoff_t index = start / PAGE_SIZE;                      \
  85         XA_STATE(xas, i->xarray, index);                        \
  86                                                                 \
  87         len = PAGE_SIZE - offset_in_page(start);                \
  88         rcu_read_lock();                                        \
  89         xas_for_each(&xas, folio, ULONG_MAX) {                  \
  90                 unsigned left;                                  \
  91                 size_t offset;                                  \
  92                 if (xas_retry(&xas, folio))                     \
  93                         continue;                               \
  94                 if (WARN_ON(xa_is_value(folio)))                \
  95                         break;                                  \
  96                 if (WARN_ON(folio_test_hugetlb(folio)))         \
  97                         break;                                  \
  98                 offset = offset_in_folio(folio, start + __off); \
  99                 while (offset < folio_size(folio)) {            \
 100                         base = kmap_local_folio(folio, offset); \
 101                         len = min(n, len);                      \
 102                         left = (STEP);                          \
 103                         kunmap_local(base);                     \
 104                         len -= left;                            \
 105                         __off += len;                           \
 106                         n -= len;                               \
 107                         if (left || n == 0)                     \
 108                                 goto __out;                     \
 109                         offset += len;                          \
 110                         len = PAGE_SIZE;                        \
 111                 }                                               \
 112         }                                                       \
 113 __out:                                                          \
 114         rcu_read_unlock();                                      \
 115         i->iov_offset += __off;                                 \
 116         n = __off;                                              \
 117 }
 118
 119 #define __iterate_and_advance(i, n, base, len, off, I, K) {     \
 120         if (unlikely(i->count < n))                             \
 121                 n = i->count;                                   \
 122         if (likely(n)) {                                        \
 123                 if (likely(iter_is_ubuf(i))) {                  \
 124                         void __user *base;                      \
 125                         size_t len;                             \
 126                         iterate_buf(i, n, base, len, off,       \
 127                                                 i->ubuf, (I))   \
 128                 } else if (likely(iter_is_iovec(i))) {          \
 129                         const struct iovec *iov = i->iov;       \
 130                         void __user *base;                      \
 131                         size_t len;                             \
 132                         iterate_iovec(i, n, base, len, off,     \
 133                                                 iov, (I))       \
 134                         i->nr_segs -= iov - i->iov;             \
 135                         i->iov = iov;                           \
 136                 } else if (iov_iter_is_bvec(i)) {               \
 137                         const struct bio_vec *bvec = i->bvec;   \
 138                         void *base;                             \
 139                         size_t len;                             \
 140                         iterate_bvec(i, n, base, len, off,      \
 141                                                 bvec, (K))      \
 142                         i->nr_segs -= bvec - i->bvec;           \
 143                         i->bvec = bvec;                         \
 144                 } else if (iov_iter_is_kvec(i)) {               \
 145                         const struct kvec *kvec = i->kvec;      \
 146                         void *base;                             \
 147                         size_t len;                             \
 148                         iterate_iovec(i, n, base, len, off,     \
 149                                                 kvec, (K))      \
 150                         i->nr_segs -= kvec - i->kvec;           \
 151                         i->kvec = kvec;                         \
 152                 } else if (iov_iter_is_xarray(i)) {             \
 153                         void *base;                             \
 154                         size_t len;                             \
 155                         iterate_xarray(i, n, base, len, off,    \
 156                                                         (K))    \
 157                 }                                               \
 158                 i->count -= n;                                  \
 159         }                                                       \
 160 }
 161 #define iterate_and_advance(i, n, base, len, off, I, K) \
 162         __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
 163
 164 static int copyout(void __user *to, const void *from, size_t n)
 165 {
 166         if (should_fail_usercopy())
 167                 return n;
 168         if (access_ok(to, n)) {
 169                 instrument_copy_to_user(to, from, n);
 170                 n = raw_copy_to_user(to, from, n);
 171         }
 172         return n;
 173 }
 174
 175 static int copyin(void *to, const void __user *from, size_t n)
 176 {
 177         size_t res = n;
 178
 179         if (should_fail_usercopy())
 180                 return n;
 181         if (access_ok(from, n)) {
 182                 instrument_copy_from_user_before(to, from, n);
 183                 res = raw_copy_from_user(to, from, n);
 184                 instrument_copy_from_user_after(to, from, n, res);
 185         }
 186         return res;
 187 }
 188
 189 static inline struct pipe_buffer *pipe_buf(const struct pipe_inode_info *pipe,
 190                                            unsigned int slot)
 191 {
 192         return &pipe->bufs[slot & (pipe->ring_size - 1)];
 193 }
 194
 195 #ifdef PIPE_PARANOIA
 196 static bool sanity(const struct iov_iter *i)
 197 {
 198         struct pipe_inode_info *pipe = i->pipe;
 199         unsigned int p_head = pipe->head;
 200         unsigned int p_tail = pipe->tail;
 201         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 202         unsigned int i_head = i->head;
 203         unsigned int idx;
 204
 205         if (i->last_offset) {
 206                 struct pipe_buffer *p;
 207                 if (unlikely(p_occupancy == 0))
 208                         goto Bad;       // pipe must be non-empty
 209                 if (unlikely(i_head != p_head - 1))
 210                         goto Bad;       // must be at the last buffer...
 211
 212                 p = pipe_buf(pipe, i_head);
 213                 if (unlikely(p->offset + p->len != abs(i->last_offset)))
 214                         goto Bad;       // ... at the end of segment
 215         } else {
 216                 if (i_head != p_head)
 217                         goto Bad;       // must be right after the last buffer
 218         }
 219         return true;
 220 Bad:
 221         printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset);
 222         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 223                         p_head, p_tail, pipe->ring_size);
 224         for (idx = 0; idx < pipe->ring_size; idx++)
 225                 printk(KERN_ERR "[%p %p %d %d]\n",
 226                         pipe->bufs[idx].ops,
 227                         pipe->bufs[idx].page,
 228                         pipe->bufs[idx].offset,
 229                         pipe->bufs[idx].len);
 230         WARN_ON(1);
 231         return false;
 232 }
 233 #else
 234 #define sanity(i) true
 235 #endif
 236
 237 static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size)
 238 {
 239         struct page *page = alloc_page(GFP_USER);
 240         if (page) {
 241                 struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 242                 *buf = (struct pipe_buffer) {
 243                         .ops = &default_pipe_buf_ops,
 244                         .page = page,
 245                         .offset = 0,
 246                         .len = size
 247                 };
 248         }
 249         return page;
 250 }
 251
 252 static void push_page(struct pipe_inode_info *pipe, struct page *page,
 253                         unsigned int offset, unsigned int size)
 254 {
 255         struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
 256         *buf = (struct pipe_buffer) {
 257                 .ops = &page_cache_pipe_buf_ops,
 258                 .page = page,
 259                 .offset = offset,
 260                 .len = size
 261         };
 262         get_page(page);
 263 }
 264
 265 static inline int last_offset(const struct pipe_buffer *buf)
 266 {
 267         if (buf->ops == &default_pipe_buf_ops)
 268                 return buf->len;        // buf->offset is 0 for those
 269         else
 270                 return -(buf->offset + buf->len);
 271 }
 272
 273 static struct page *append_pipe(struct iov_iter *i, size_t size,
 274                                 unsigned int *off)
 275 {
 276         struct pipe_inode_info *pipe = i->pipe;
 277         int offset = i->last_offset;
 278         struct pipe_buffer *buf;
 279         struct page *page;
 280
 281         if (offset > 0 && offset < PAGE_SIZE) {
 282                 // some space in the last buffer; add to it
 283                 buf = pipe_buf(pipe, pipe->head - 1);
 284                 size = min_t(size_t, size, PAGE_SIZE - offset);
 285                 buf->len += size;
 286                 i->last_offset += size;
 287                 i->count -= size;
 288                 *off = offset;
 289                 return buf->page;
 290         }
 291         // OK, we need a new buffer
 292         *off = 0;
 293         size = min_t(size_t, size, PAGE_SIZE);
 294         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 295                 return NULL;
 296         page = push_anon(pipe, size);
 297         if (!page)
 298                 return NULL;
 299         i->head = pipe->head - 1;
 300         i->last_offset = size;
 301         i->count -= size;
 302         return page;
 303 }
 304
 305 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 306                          struct iov_iter *i)
 307 {
 308         struct pipe_inode_info *pipe = i->pipe;
 309         unsigned int head = pipe->head;
 310
 311         if (unlikely(bytes > i->count))
 312                 bytes = i->count;
 313
 314         if (unlikely(!bytes))
 315                 return 0;
 316
 317         if (!sanity(i))
 318                 return 0;
 319
 320         if (offset && i->last_offset == -offset) { // could we merge it?
 321                 struct pipe_buffer *buf = pipe_buf(pipe, head - 1);
 322                 if (buf->page == page) {
 323                         buf->len += bytes;
 324                         i->last_offset -= bytes;
 325                         i->count -= bytes;
 326                         return bytes;
 327                 }
 328         }
 329         if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
 330                 return 0;
 331
 332         push_page(pipe, page, offset, bytes);
 333         i->last_offset = -(offset + bytes);
 334         i->head = head;
 335         i->count -= bytes;
 336         return bytes;
 337 }
 338
 339 /*
 340  * fault_in_iov_iter_readable - fault in iov iterator for reading
 341  * @i: iterator
 342  * @size: maximum length
 343  *
 344  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 345  * @size.  For each iovec, fault in each page that constitutes the iovec.
 346  *
 347  * Returns the number of bytes not faulted in (like copy_to_user() and
 348  * copy_from_user()).
 349  *
 350  * Always returns 0 for non-userspace iterators.
 351  */
 352 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
 353 {
 354         if (iter_is_ubuf(i)) {
 355                 size_t n = min(size, iov_iter_count(i));
 356                 n -= fault_in_readable(i->ubuf + i->iov_offset, n);
 357                 return size - n;
 358         } else if (iter_is_iovec(i)) {
 359                 size_t count = min(size, iov_iter_count(i));
 360                 const struct iovec *p;
 361                 size_t skip;
 362
 363                 size -= count;
 364                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 365                         size_t len = min(count, p->iov_len - skip);
 366                         size_t ret;
 367
 368                         if (unlikely(!len))
 369                                 continue;
 370                         ret = fault_in_readable(p->iov_base + skip, len);
 371                         count -= len - ret;
 372                         if (ret)
 373                                 break;
 374                 }
 375                 return count + size;
 376         }
 377         return 0;
 378 }
 379 EXPORT_SYMBOL(fault_in_iov_iter_readable);
 380
 381 /*
 382  * fault_in_iov_iter_writeable - fault in iov iterator for writing
 383  * @i: iterator
 384  * @size: maximum length
 385  *
 386  * Faults in the iterator using get_user_pages(), i.e., without triggering
 387  * hardware page faults.  This is primarily useful when we already know that
 388  * some or all of the pages in @i aren't in memory.
 389  *
 390  * Returns the number of bytes not faulted in, like copy_to_user() and
 391  * copy_from_user().
 392  *
 393  * Always returns 0 for non-user-space iterators.
 394  */
 395 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
 396 {
 397         if (iter_is_ubuf(i)) {
 398                 size_t n = min(size, iov_iter_count(i));
 399                 n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n);
 400                 return size - n;
 401         } else if (iter_is_iovec(i)) {
 402                 size_t count = min(size, iov_iter_count(i));
 403                 const struct iovec *p;
 404                 size_t skip;
 405
 406                 size -= count;
 407                 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
 408                         size_t len = min(count, p->iov_len - skip);
 409                         size_t ret;
 410
 411                         if (unlikely(!len))
 412                                 continue;
 413                         ret = fault_in_safe_writeable(p->iov_base + skip, len);
 414                         count -= len - ret;
 415                         if (ret)
 416                                 break;
 417                 }
 418                 return count + size;
 419         }
 420         return 0;
 421 }
 422 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
 423
 424 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 425                         const struct iovec *iov, unsigned long nr_segs,
 426                         size_t count)
 427 {
 428         WARN_ON(direction & ~(READ | WRITE));
 429         *i = (struct iov_iter) {
 430                 .iter_type = ITER_IOVEC,
 431                 .nofault = false,
 432                 .user_backed = true,
 433                 .data_source = direction,
 434                 .iov = iov,
 435                 .nr_segs = nr_segs,
 436                 .iov_offset = 0,
 437                 .count = count
 438         };
 439 }
 440 EXPORT_SYMBOL(iov_iter_init);
 441
 442 // returns the offset in partial buffer (if any)
 443 static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages)
 444 {
 445         struct pipe_inode_info *pipe = i->pipe;
 446         int used = pipe->head - pipe->tail;
 447         int off = i->last_offset;
 448
 449         *npages = max((int)pipe->max_usage - used, 0);
 450
 451         if (off > 0 && off < PAGE_SIZE) { // anon and not full
 452                 (*npages)++;
 453                 return off;
 454         }
 455         return 0;
 456 }
 457
 458 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 459                                 struct iov_iter *i)
 460 {
 461         unsigned int off, chunk;
 462
 463         if (unlikely(bytes > i->count))
 464                 bytes = i->count;
 465         if (unlikely(!bytes))
 466                 return 0;
 467
 468         if (!sanity(i))
 469                 return 0;
 470
 471         for (size_t n = bytes; n; n -= chunk) {
 472                 struct page *page = append_pipe(i, n, &off);
 473                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 474                 if (!page)
 475                         return bytes - n;
 476                 memcpy_to_page(page, off, addr, chunk);
 477                 addr += chunk;
 478         }
 479         return bytes;
 480 }
 481
 482 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 483                               __wsum sum, size_t off)
 484 {
 485         __wsum next = csum_partial_copy_nocheck(from, to, len);
 486         return csum_block_add(sum, next, off);
 487 }
 488
 489 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 490                                          struct iov_iter *i, __wsum *sump)
 491 {
 492         __wsum sum = *sump;
 493         size_t off = 0;
 494         unsigned int chunk, r;
 495
 496         if (unlikely(bytes > i->count))
 497                 bytes = i->count;
 498         if (unlikely(!bytes))
 499                 return 0;
 500
 501         if (!sanity(i))
 502                 return 0;
 503
 504         while (bytes) {
 505                 struct page *page = append_pipe(i, bytes, &r);
 506                 char *p;
 507
 508                 if (!page)
 509                         break;
 510                 chunk = min_t(size_t, bytes, PAGE_SIZE - r);
 511                 p = kmap_local_page(page);
 512                 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
 513                 kunmap_local(p);
 514                 off += chunk;
 515                 bytes -= chunk;
 516         }
 517         *sump = sum;
 518         return off;
 519 }
 520
 521 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 522 {
 523         if (WARN_ON_ONCE(i->data_source))
 524                 return 0;
 525         if (unlikely(iov_iter_is_pipe(i)))
 526                 return copy_pipe_to_iter(addr, bytes, i);
 527         if (user_backed_iter(i))
 528                 might_fault();
 529         iterate_and_advance(i, bytes, base, len, off,
 530                 copyout(base, addr + off, len),
 531                 memcpy(base, addr + off, len)
 532         )
 533
 534         return bytes;
 535 }
 536 EXPORT_SYMBOL(_copy_to_iter);
 537
 538 #ifdef CONFIG_ARCH_HAS_COPY_MC
 539 static int copyout_mc(void __user *to, const void *from, size_t n)
 540 {
 541         if (access_ok(to, n)) {
 542                 instrument_copy_to_user(to, from, n);
 543                 n = copy_mc_to_user((__force void *) to, from, n);
 544         }
 545         return n;
 546 }
 547
 548 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 549                                 struct iov_iter *i)
 550 {
 551         size_t xfer = 0;
 552         unsigned int off, chunk;
 553
 554         if (unlikely(bytes > i->count))
 555                 bytes = i->count;
 556         if (unlikely(!bytes))
 557                 return 0;
 558
 559         if (!sanity(i))
 560                 return 0;
 561
 562         while (bytes) {
 563                 struct page *page = append_pipe(i, bytes, &off);
 564                 unsigned long rem;
 565                 char *p;
 566
 567                 if (!page)
 568                         break;
 569                 chunk = min_t(size_t, bytes, PAGE_SIZE - off);
 570                 p = kmap_local_page(page);
 571                 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
 572                 chunk -= rem;
 573                 kunmap_local(p);
 574                 xfer += chunk;
 575                 bytes -= chunk;
 576                 if (rem) {
 577                         iov_iter_revert(i, rem);
 578                         break;
 579                 }
 580         }
 581         return xfer;
 582 }
 583
 584 /**
 585  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 586  * @addr: source kernel address
 587  * @bytes: total transfer length
 588  * @i: destination iterator
 589  *
 590  * The pmem driver deploys this for the dax operation
 591  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 592  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 593  * successfully copied.
 594  *
 595  * The main differences between this and typical _copy_to_iter().
 596  *
 597  * * Typical tail/residue handling after a fault retries the copy
 598  *   byte-by-byte until the fault happens again. Re-triggering machine
 599  *   checks is potentially fatal so the implementation uses source
 600  *   alignment and poison alignment assumptions to avoid re-triggering
 601  *   hardware exceptions.
 602  *
 603  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 604  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 605  *   a short copy.
 606  *
 607  * Return: number of bytes copied (may be %0)
 608  */
 609 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 610 {
 611         if (WARN_ON_ONCE(i->data_source))
 612                 return 0;
 613         if (unlikely(iov_iter_is_pipe(i)))
 614                 return copy_mc_pipe_to_iter(addr, bytes, i);
 615         if (user_backed_iter(i))
 616                 might_fault();
 617         __iterate_and_advance(i, bytes, base, len, off,
 618                 copyout_mc(base, addr + off, len),
 619                 copy_mc_to_kernel(base, addr + off, len)
 620         )
 621
 622         return bytes;
 623 }
 624 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 625 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 626
 627 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 628 {
 629         if (WARN_ON_ONCE(!i->data_source))
 630                 return 0;
 631
 632         if (user_backed_iter(i))
 633                 might_fault();
 634         iterate_and_advance(i, bytes, base, len, off,
 635                 copyin(addr + off, base, len),
 636                 memcpy(addr + off, base, len)
 637         )
 638
 639         return bytes;
 640 }
 641 EXPORT_SYMBOL(_copy_from_iter);
 642
 643 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 644 {
 645         if (WARN_ON_ONCE(!i->data_source))
 646                 return 0;
 647
 648         iterate_and_advance(i, bytes, base, len, off,
 649                 __copy_from_user_inatomic_nocache(addr + off, base, len),
 650                 memcpy(addr + off, base, len)
 651         )
 652
 653         return bytes;
 654 }
 655 EXPORT_SYMBOL(_copy_from_iter_nocache);
 656
 657 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 658 /**
 659  * _copy_from_iter_flushcache - write destination through cpu cache
 660  * @addr: destination kernel address
 661  * @bytes: total transfer length
 662  * @i: source iterator
 663  *
 664  * The pmem driver arranges for filesystem-dax to use this facility via
 665  * dax_copy_from_iter() for ensuring that writes to persistent memory
 666  * are flushed through the CPU cache. It is differentiated from
 667  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 668  * all iterator types. The _copy_from_iter_nocache() only attempts to
 669  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 670  * instructions that strand dirty-data in the cache.
 671  *
 672  * Return: number of bytes copied (may be %0)
 673  */
 674 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 675 {
 676         if (WARN_ON_ONCE(!i->data_source))
 677                 return 0;
 678
 679         iterate_and_advance(i, bytes, base, len, off,
 680                 __copy_from_user_flushcache(addr + off, base, len),
 681                 memcpy_flushcache(addr + off, base, len)
 682         )
 683
 684         return bytes;
 685 }
 686 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 687 #endif
 688
 689 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 690 {
 691         struct page *head;
 692         size_t v = n + offset;
 693
 694         /*
 695          * The general case needs to access the page order in order
 696          * to compute the page size.
 697          * However, we mostly deal with order-0 pages and thus can
 698          * avoid a possible cache line miss for requests that fit all
 699          * page orders.
 700          */
 701         if (n <= v && v <= PAGE_SIZE)
 702                 return true;
 703
 704         head = compound_head(page);
 705         v += (page - head) << PAGE_SHIFT;
 706
 707         if (WARN_ON(n > v || v > page_size(head)))
 708                 return false;
 709         return true;
 710 }
 711
 712 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 713                          struct iov_iter *i)
 714 {
 715         size_t res = 0;
 716         if (!page_copy_sane(page, offset, bytes))
 717                 return 0;
 718         if (WARN_ON_ONCE(i->data_source))
 719                 return 0;
 720         if (unlikely(iov_iter_is_pipe(i)))
 721                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 722         page += offset / PAGE_SIZE; // first subpage
 723         offset %= PAGE_SIZE;
 724         while (1) {
 725                 void *kaddr = kmap_local_page(page);
 726                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 727                 n = _copy_to_iter(kaddr + offset, n, i);
 728                 kunmap_local(kaddr);
 729                 res += n;
 730                 bytes -= n;
 731                 if (!bytes || !n)
 732                         break;
 733                 offset += n;
 734                 if (offset == PAGE_SIZE) {
 735                         page++;
 736                         offset = 0;
 737                 }
 738         }
 739         return res;
 740 }
 741 EXPORT_SYMBOL(copy_page_to_iter);
 742
 743 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 744                          struct iov_iter *i)
 745 {
 746         size_t res = 0;
 747         if (!page_copy_sane(page, offset, bytes))
 748                 return 0;
 749         page += offset / PAGE_SIZE; // first subpage
 750         offset %= PAGE_SIZE;
 751         while (1) {
 752                 void *kaddr = kmap_local_page(page);
 753                 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
 754                 n = _copy_from_iter(kaddr + offset, n, i);
 755                 kunmap_local(kaddr);
 756                 res += n;
 757                 bytes -= n;
 758                 if (!bytes || !n)
 759                         break;
 760                 offset += n;
 761                 if (offset == PAGE_SIZE) {
 762                         page++;
 763                         offset = 0;
 764                 }
 765         }
 766         return res;
 767 }
 768 EXPORT_SYMBOL(copy_page_from_iter);
 769
 770 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 771 {
 772         unsigned int chunk, off;
 773
 774         if (unlikely(bytes > i->count))
 775                 bytes = i->count;
 776         if (unlikely(!bytes))
 777                 return 0;
 778
 779         if (!sanity(i))
 780                 return 0;
 781
 782         for (size_t n = bytes; n; n -= chunk) {
 783                 struct page *page = append_pipe(i, n, &off);
 784                 char *p;
 785
 786                 if (!page)
 787                         return bytes - n;
 788                 chunk = min_t(size_t, n, PAGE_SIZE - off);
 789                 p = kmap_local_page(page);
 790                 memset(p + off, 0, chunk);
 791                 kunmap_local(p);
 792         }
 793         return bytes;
 794 }
 795
 796 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 797 {
 798         if (unlikely(iov_iter_is_pipe(i)))
 799                 return pipe_zero(bytes, i);
 800         iterate_and_advance(i, bytes, base, len, count,
 801                 clear_user(base, len),
 802                 memset(base, 0, len)
 803         )
 804
 805         return bytes;
 806 }
 807 EXPORT_SYMBOL(iov_iter_zero);
 808
 809 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
 810                                   struct iov_iter *i)
 811 {
 812         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 813         if (!page_copy_sane(page, offset, bytes)) {
 814                 kunmap_atomic(kaddr);
 815                 return 0;
 816         }
 817         if (WARN_ON_ONCE(!i->data_source)) {
 818                 kunmap_atomic(kaddr);
 819                 return 0;
 820         }
 821         iterate_and_advance(i, bytes, base, len, off,
 822                 copyin(p + off, base, len),
 823                 memcpy(p + off, base, len)
 824         )
 825         kunmap_atomic(kaddr);
 826         return bytes;
 827 }
 828 EXPORT_SYMBOL(copy_page_from_iter_atomic);
 829
 830 static void pipe_advance(struct iov_iter *i, size_t size)
 831 {
 832         struct pipe_inode_info *pipe = i->pipe;
 833         int off = i->last_offset;
 834
 835         if (!off && !size) {
 836                 pipe_discard_from(pipe, i->start_head); // discard everything
 837                 return;
 838         }
 839         i->count -= size;
 840         while (1) {
 841                 struct pipe_buffer *buf = pipe_buf(pipe, i->head);
 842                 if (off) /* make it relative to the beginning of buffer */
 843                         size += abs(off) - buf->offset;
 844                 if (size <= buf->len) {
 845                         buf->len = size;
 846                         i->last_offset = last_offset(buf);
 847                         break;
 848                 }
 849                 size -= buf->len;
 850                 i->head++;
 851                 off = 0;
 852         }
 853         pipe_discard_from(pipe, i->head + 1); // discard everything past this one
 854 }
 855
 856 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
 857 {
 858         const struct bio_vec *bvec, *end;
 859
 860         if (!i->count)
 861                 return;
 862         i->count -= size;
 863
 864         size += i->iov_offset;
 865
 866         for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
 867                 if (likely(size < bvec->bv_len))
 868                         break;
 869                 size -= bvec->bv_len;
 870         }
 871         i->iov_offset = size;
 872         i->nr_segs -= bvec - i->bvec;
 873         i->bvec = bvec;
 874 }
 875
 876 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
 877 {
 878         const struct iovec *iov, *end;
 879
 880         if (!i->count)
 881                 return;
 882         i->count -= size;
 883
 884         size += i->iov_offset; // from beginning of current segment
 885         for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
 886                 if (likely(size < iov->iov_len))
 887                         break;
 888                 size -= iov->iov_len;
 889         }
 890         i->iov_offset = size;
 891         i->nr_segs -= iov - i->iov;
 892         i->iov = iov;
 893 }
 894
 895 void iov_iter_advance(struct iov_iter *i, size_t size)
 896 {
 897         if (unlikely(i->count < size))
 898                 size = i->count;
 899         if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
 900                 i->iov_offset += size;
 901                 i->count -= size;
 902         } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
 903                 /* iovec and kvec have identical layouts */
 904                 iov_iter_iovec_advance(i, size);
 905         } else if (iov_iter_is_bvec(i)) {
 906                 iov_iter_bvec_advance(i, size);
 907         } else if (iov_iter_is_pipe(i)) {
 908                 pipe_advance(i, size);
 909         } else if (iov_iter_is_discard(i)) {
 910                 i->count -= size;
 911         }
 912 }
 913 EXPORT_SYMBOL(iov_iter_advance);
 914
 915 void iov_iter_revert(struct iov_iter *i, size_t unroll)
 916 {
 917         if (!unroll)
 918                 return;
 919         if (WARN_ON(unroll > MAX_RW_COUNT))
 920                 return;
 921         i->count += unroll;
 922         if (unlikely(iov_iter_is_pipe(i))) {
 923                 struct pipe_inode_info *pipe = i->pipe;
 924                 unsigned int head = pipe->head;
 925
 926                 while (head > i->start_head) {
 927                         struct pipe_buffer *b = pipe_buf(pipe, --head);
 928                         if (unroll < b->len) {
 929                                 b->len -= unroll;
 930                                 i->last_offset = last_offset(b);
 931                                 i->head = head;
 932                                 return;
 933                         }
 934                         unroll -= b->len;
 935                         pipe_buf_release(pipe, b);
 936                         pipe->head--;
 937                 }
 938                 i->last_offset = 0;
 939                 i->head = head;
 940                 return;
 941         }
 942         if (unlikely(iov_iter_is_discard(i)))
 943                 return;
 944         if (unroll <= i->iov_offset) {
 945                 i->iov_offset -= unroll;
 946                 return;
 947         }
 948         unroll -= i->iov_offset;
 949         if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
 950                 BUG(); /* We should never go beyond the start of the specified
 951                         * range since we might then be straying into pages that
 952                         * aren't pinned.
 953                         */
 954         } else if (iov_iter_is_bvec(i)) {
 955                 const struct bio_vec *bvec = i->bvec;
 956                 while (1) {
 957                         size_t n = (--bvec)->bv_len;
 958                         i->nr_segs++;
 959                         if (unroll <= n) {
 960                                 i->bvec = bvec;
 961                                 i->iov_offset = n - unroll;
 962                                 return;
 963                         }
 964                         unroll -= n;
 965                 }
 966         } else { /* same logics for iovec and kvec */
 967                 const struct iovec *iov = i->iov;
 968                 while (1) {
 969                         size_t n = (--iov)->iov_len;
 970                         i->nr_segs++;
 971                         if (unroll <= n) {
 972                                 i->iov = iov;
 973                                 i->iov_offset = n - unroll;
 974                                 return;
 975                         }
 976                         unroll -= n;
 977                 }
 978         }
 979 }
 980 EXPORT_SYMBOL(iov_iter_revert);
 981
 982 /*
 983  * Return the count of just the current iov_iter segment.
 984  */
 985 size_t iov_iter_single_seg_count(const struct iov_iter *i)
 986 {
 987         if (i->nr_segs > 1) {
 988                 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
 989                         return min(i->count, i->iov->iov_len - i->iov_offset);
 990                 if (iov_iter_is_bvec(i))
 991                         return min(i->count, i->bvec->bv_len - i->iov_offset);
 992         }
 993         return i->count;
 994 }
 995 EXPORT_SYMBOL(iov_iter_single_seg_count);
 996
 997 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
 998                         const struct kvec *kvec, unsigned long nr_segs,
 999                         size_t count)
1000 {
1001         WARN_ON(direction & ~(READ | WRITE));
1002         *i = (struct iov_iter){
1003                 .iter_type = ITER_KVEC,
1004                 .data_source = direction,
1005                 .kvec = kvec,
1006                 .nr_segs = nr_segs,
1007                 .iov_offset = 0,
1008                 .count = count
1009         };
1010 }
1011 EXPORT_SYMBOL(iov_iter_kvec);
1012
1013 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1014                         const struct bio_vec *bvec, unsigned long nr_segs,
1015                         size_t count)
1016 {
1017         WARN_ON(direction & ~(READ | WRITE));
1018         *i = (struct iov_iter){
1019                 .iter_type = ITER_BVEC,
1020                 .data_source = direction,
1021                 .bvec = bvec,
1022                 .nr_segs = nr_segs,
1023                 .iov_offset = 0,
1024                 .count = count
1025         };
1026 }
1027 EXPORT_SYMBOL(iov_iter_bvec);
1028
1029 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1030                         struct pipe_inode_info *pipe,
1031                         size_t count)
1032 {
1033         BUG_ON(direction != READ);
1034         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1035         *i = (struct iov_iter){
1036                 .iter_type = ITER_PIPE,
1037                 .data_source = false,
1038                 .pipe = pipe,
1039                 .head = pipe->head,
1040                 .start_head = pipe->head,
1041                 .last_offset = 0,
1042                 .count = count
1043         };
1044 }
1045 EXPORT_SYMBOL(iov_iter_pipe);
1046
1047 /**
1048  * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1049  * @i: The iterator to initialise.
1050  * @direction: The direction of the transfer.
1051  * @xarray: The xarray to access.
1052  * @start: The start file position.
1053  * @count: The size of the I/O buffer in bytes.
1054  *
1055  * Set up an I/O iterator to either draw data out of the pages attached to an
1056  * inode or to inject data into those pages.  The pages *must* be prevented
1057  * from evaporation, either by taking a ref on them or locking them by the
1058  * caller.
1059  */
1060 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1061                      struct xarray *xarray, loff_t start, size_t count)
1062 {
1063         BUG_ON(direction & ~1);
1064         *i = (struct iov_iter) {
1065                 .iter_type = ITER_XARRAY,
1066                 .data_source = direction,
1067                 .xarray = xarray,
1068                 .xarray_start = start,
1069                 .count = count,
1070                 .iov_offset = 0
1071         };
1072 }
1073 EXPORT_SYMBOL(iov_iter_xarray);
1074
1075 /**
1076  * iov_iter_discard - Initialise an I/O iterator that discards data
1077  * @i: The iterator to initialise.
1078  * @direction: The direction of the transfer.
1079  * @count: The size of the I/O buffer in bytes.
1080  *
1081  * Set up an I/O iterator that just discards everything that's written to it.
1082  * It's only available as a READ iterator.
1083  */
1084 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1085 {
1086         BUG_ON(direction != READ);
1087         *i = (struct iov_iter){
1088                 .iter_type = ITER_DISCARD,
1089                 .data_source = false,
1090                 .count = count,
1091                 .iov_offset = 0
1092         };
1093 }
1094 EXPORT_SYMBOL(iov_iter_discard);
1095
1096 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1097                                    unsigned len_mask)
1098 {
1099         size_t size = i->count;
1100         size_t skip = i->iov_offset;
1101         unsigned k;
1102
1103         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1104                 size_t len = i->iov[k].iov_len - skip;
1105
1106                 if (len > size)
1107                         len = size;
1108                 if (len & len_mask)
1109                         return false;
1110                 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1111                         return false;
1112
1113                 size -= len;
1114                 if (!size)
1115                         break;
1116         }
1117         return true;
1118 }
1119
1120 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1121                                   unsigned len_mask)
1122 {
1123         size_t size = i->count;
1124         unsigned skip = i->iov_offset;
1125         unsigned k;
1126
1127         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1128                 size_t len = i->bvec[k].bv_len - skip;
1129
1130                 if (len > size)
1131                         len = size;
1132                 if (len & len_mask)
1133                         return false;
1134                 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1135                         return false;
1136
1137                 size -= len;
1138                 if (!size)
1139                         break;
1140         }
1141         return true;
1142 }
1143
1144 /**
1145  * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1146  *      are aligned to the parameters.
1147  *
1148  * @i: &struct iov_iter to restore
1149  * @addr_mask: bit mask to check against the iov element's addresses
1150  * @len_mask: bit mask to check against the iov element's lengths
1151  *
1152  * Return: false if any addresses or lengths intersect with the provided masks
1153  */
1154 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1155                          unsigned len_mask)
1156 {
1157         if (likely(iter_is_ubuf(i))) {
1158                 if (i->count & len_mask)
1159                         return false;
1160                 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
1161                         return false;
1162                 return true;
1163         }
1164
1165         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1166                 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1167
1168         if (iov_iter_is_bvec(i))
1169                 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1170
1171         if (iov_iter_is_pipe(i)) {
1172                 size_t size = i->count;
1173
1174                 if (size & len_mask)
1175                         return false;
1176                 if (size && i->last_offset > 0) {
1177                         if (i->last_offset & addr_mask)
1178                                 return false;
1179                 }
1180
1181                 return true;
1182         }
1183
1184         if (iov_iter_is_xarray(i)) {
1185                 if (i->count & len_mask)
1186                         return false;
1187                 if ((i->xarray_start + i->iov_offset) & addr_mask)
1188                         return false;
1189         }
1190
1191         return true;
1192 }
1193 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1194
1195 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1196 {
1197         unsigned long res = 0;
1198         size_t size = i->count;
1199         size_t skip = i->iov_offset;
1200         unsigned k;
1201
1202         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1203                 size_t len = i->iov[k].iov_len - skip;
1204                 if (len) {
1205                         res |= (unsigned long)i->iov[k].iov_base + skip;
1206                         if (len > size)
1207                                 len = size;
1208                         res |= len;
1209                         size -= len;
1210                         if (!size)
1211                                 break;
1212                 }
1213         }
1214         return res;
1215 }
1216
1217 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1218 {
1219         unsigned res = 0;
1220         size_t size = i->count;
1221         unsigned skip = i->iov_offset;
1222         unsigned k;
1223
1224         for (k = 0; k < i->nr_segs; k++, skip = 0) {
1225                 size_t len = i->bvec[k].bv_len - skip;
1226                 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1227                 if (len > size)
1228                         len = size;
1229                 res |= len;
1230                 size -= len;
1231                 if (!size)
1232                         break;
1233         }
1234         return res;
1235 }
1236
1237 unsigned long iov_iter_alignment(const struct iov_iter *i)
1238 {
1239         if (likely(iter_is_ubuf(i))) {
1240                 size_t size = i->count;
1241                 if (size)
1242                         return ((unsigned long)i->ubuf + i->iov_offset) | size;
1243                 return 0;
1244         }
1245
1246         /* iovec and kvec have identical layouts */
1247         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1248                 return iov_iter_alignment_iovec(i);
1249
1250         if (iov_iter_is_bvec(i))
1251                 return iov_iter_alignment_bvec(i);
1252
1253         if (iov_iter_is_pipe(i)) {
1254                 size_t size = i->count;
1255
1256                 if (size && i->last_offset > 0)
1257                         return size | i->last_offset;
1258                 return size;
1259         }
1260
1261         if (iov_iter_is_xarray(i))
1262                 return (i->xarray_start + i->iov_offset) | i->count;
1263
1264         return 0;
1265 }
1266 EXPORT_SYMBOL(iov_iter_alignment);
1267
1268 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1269 {
1270         unsigned long res = 0;
1271         unsigned long v = 0;
1272         size_t size = i->count;
1273         unsigned k;
1274
1275         if (iter_is_ubuf(i))
1276                 return 0;
1277
1278         if (WARN_ON(!iter_is_iovec(i)))
1279                 return ~0U;
1280
1281         for (k = 0; k < i->nr_segs; k++) {
1282                 if (i->iov[k].iov_len) {
1283                         unsigned long base = (unsigned long)i->iov[k].iov_base;
1284                         if (v) // if not the first one
1285                                 res |= base | v; // this start | previous end
1286                         v = base + i->iov[k].iov_len;
1287                         if (size <= i->iov[k].iov_len)
1288                                 break;
1289                         size -= i->iov[k].iov_len;
1290                 }
1291         }
1292         return res;
1293 }
1294 EXPORT_SYMBOL(iov_iter_gap_alignment);
1295
1296 static int want_pages_array(struct page ***res, size_t size,
1297                             size_t start, unsigned int maxpages)
1298 {
1299         unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
1300
1301         if (count > maxpages)
1302                 count = maxpages;
1303         WARN_ON(!count);        // caller should've prevented that
1304         if (!*res) {
1305                 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
1306                 if (!*res)
1307                         return 0;
1308         }
1309         return count;
1310 }
1311
1312 static ssize_t pipe_get_pages(struct iov_iter *i,
1313                    struct page ***pages, size_t maxsize, unsigned maxpages,
1314                    size_t *start)
1315 {
1316         unsigned int npages, count, off, chunk;
1317         struct page **p;
1318         size_t left;
1319
1320         if (!sanity(i))
1321                 return -EFAULT;
1322
1323         *start = off = pipe_npages(i, &npages);
1324         if (!npages)
1325                 return -EFAULT;
1326         count = want_pages_array(pages, maxsize, off, min(npages, maxpages));
1327         if (!count)
1328                 return -ENOMEM;
1329         p = *pages;
1330         for (npages = 0, left = maxsize ; npages < count; npages++, left -= chunk) {
1331                 struct page *page = append_pipe(i, left, &off);
1332                 if (!page)
1333                         break;
1334                 chunk = min_t(size_t, left, PAGE_SIZE - off);
1335                 get_page(*p++ = page);
1336         }
1337         if (!npages)
1338                 return -EFAULT;
1339         return maxsize - left;
1340 }
1341
1342 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1343                                           pgoff_t index, unsigned int nr_pages)
1344 {
1345         XA_STATE(xas, xa, index);
1346         struct page *page;
1347         unsigned int ret = 0;
1348
1349         rcu_read_lock();
1350         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1351                 if (xas_retry(&xas, page))
1352                         continue;
1353
1354                 /* Has the page moved or been split? */
1355                 if (unlikely(page != xas_reload(&xas))) {
1356                         xas_reset(&xas);
1357                         continue;
1358                 }
1359
1360                 pages[ret] = find_subpage(page, xas.xa_index);
1361                 get_page(pages[ret]);
1362                 if (++ret == nr_pages)
1363                         break;
1364         }
1365         rcu_read_unlock();
1366         return ret;
1367 }
1368
1369 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1370                                      struct page ***pages, size_t maxsize,
1371                                      unsigned maxpages, size_t *_start_offset)
1372 {
1373         unsigned nr, offset, count;
1374         pgoff_t index;
1375         loff_t pos;
1376
1377         pos = i->xarray_start + i->iov_offset;
1378         index = pos >> PAGE_SHIFT;
1379         offset = pos & ~PAGE_MASK;
1380         *_start_offset = offset;
1381
1382         count = want_pages_array(pages, maxsize, offset, maxpages);
1383         if (!count)
1384                 return -ENOMEM;
1385         nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
1386         if (nr == 0)
1387                 return 0;
1388
1389         maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1390         i->iov_offset += maxsize;
1391         i->count -= maxsize;
1392         return maxsize;
1393 }
1394
1395 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1396 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1397 {
1398         size_t skip;
1399         long k;
1400
1401         if (iter_is_ubuf(i))
1402                 return (unsigned long)i->ubuf + i->iov_offset;
1403
1404         for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1405                 size_t len = i->iov[k].iov_len - skip;
1406
1407                 if (unlikely(!len))
1408                         continue;
1409                 if (*size > len)
1410                         *size = len;
1411                 return (unsigned long)i->iov[k].iov_base + skip;
1412         }
1413         BUG(); // if it had been empty, we wouldn't get called
1414 }
1415
1416 /* must be done on non-empty ITER_BVEC one */
1417 static struct page *first_bvec_segment(const struct iov_iter *i,
1418                                        size_t *size, size_t *start)
1419 {
1420         struct page *page;
1421         size_t skip = i->iov_offset, len;
1422
1423         len = i->bvec->bv_len - skip;
1424         if (*size > len)
1425                 *size = len;
1426         skip += i->bvec->bv_offset;
1427         page = i->bvec->bv_page + skip / PAGE_SIZE;
1428         *start = skip % PAGE_SIZE;
1429         return page;
1430 }
1431
1432 static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
1433                    struct page ***pages, size_t maxsize,
1434                    unsigned int maxpages, size_t *start,
1435                    unsigned int gup_flags)
1436 {
1437         unsigned int n;
1438
1439         if (maxsize > i->count)
1440                 maxsize = i->count;
1441         if (!maxsize)
1442                 return 0;
1443         if (maxsize > MAX_RW_COUNT)
1444                 maxsize = MAX_RW_COUNT;
1445
1446         if (likely(user_backed_iter(i))) {
1447                 unsigned long addr;
1448                 int res;
1449
1450                 if (iov_iter_rw(i) != WRITE)
1451                         gup_flags |= FOLL_WRITE;
1452                 if (i->nofault)
1453                         gup_flags |= FOLL_NOFAULT;
1454
1455                 addr = first_iovec_segment(i, &maxsize);
1456                 *start = addr % PAGE_SIZE;
1457                 addr &= PAGE_MASK;
1458                 n = want_pages_array(pages, maxsize, *start, maxpages);
1459                 if (!n)
1460                         return -ENOMEM;
1461                 res = get_user_pages_fast(addr, n, gup_flags, *pages);
1462                 if (unlikely(res <= 0))
1463                         return res;
1464                 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1465                 iov_iter_advance(i, maxsize);
1466                 return maxsize;
1467         }
1468         if (iov_iter_is_bvec(i)) {
1469                 struct page **p;
1470                 struct page *page;
1471
1472                 page = first_bvec_segment(i, &maxsize, start);
1473                 n = want_pages_array(pages, maxsize, *start, maxpages);
1474                 if (!n)
1475                         return -ENOMEM;
1476                 p = *pages;
1477                 for (int k = 0; k < n; k++)
1478                         get_page(p[k] = page + k);
1479                 maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1480                 i->count -= maxsize;
1481                 i->iov_offset += maxsize;
1482                 if (i->iov_offset == i->bvec->bv_len) {
1483                         i->iov_offset = 0;
1484                         i->bvec++;
1485                         i->nr_segs--;
1486                 }
1487                 return maxsize;
1488         }
1489         if (iov_iter_is_pipe(i))
1490                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1491         if (iov_iter_is_xarray(i))
1492                 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1493         return -EFAULT;
1494 }
1495
1496 ssize_t iov_iter_get_pages(struct iov_iter *i,
1497                    struct page **pages, size_t maxsize, unsigned maxpages,
1498                    size_t *start, unsigned gup_flags)
1499 {
1500         if (!maxpages)
1501                 return 0;
1502         BUG_ON(!pages);
1503
1504         return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages,
1505                                           start, gup_flags);
1506 }
1507 EXPORT_SYMBOL_GPL(iov_iter_get_pages);
1508
1509 ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages,
1510                 size_t maxsize, unsigned maxpages, size_t *start)
1511 {
1512         return iov_iter_get_pages(i, pages, maxsize, maxpages, start, 0);
1513 }
1514 EXPORT_SYMBOL(iov_iter_get_pages2);
1515
1516 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1517                    struct page ***pages, size_t maxsize,
1518                    size_t *start, unsigned gup_flags)
1519 {
1520         ssize_t len;
1521
1522         *pages = NULL;
1523
1524         len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start,
1525                                          gup_flags);
1526         if (len <= 0) {
1527                 kvfree(*pages);
1528                 *pages = NULL;
1529         }
1530         return len;
1531 }
1532 EXPORT_SYMBOL_GPL(iov_iter_get_pages_alloc);
1533
1534 ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1535                 struct page ***pages, size_t maxsize, size_t *start)
1536 {
1537         return iov_iter_get_pages_alloc(i, pages, maxsize, start, 0);
1538 }
1539 EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1540
1541 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1542                                struct iov_iter *i)
1543 {
1544         __wsum sum, next;
1545         sum = *csum;
1546         if (WARN_ON_ONCE(!i->data_source))
1547                 return 0;
1548
1549         iterate_and_advance(i, bytes, base, len, off, ({
1550                 next = csum_and_copy_from_user(base, addr + off, len);
1551                 sum = csum_block_add(sum, next, off);
1552                 next ? 0 : len;
1553         }), ({
1554                 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1555         })
1556         )
1557         *csum = sum;
1558         return bytes;
1559 }
1560 EXPORT_SYMBOL(csum_and_copy_from_iter);
1561
1562 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1563                              struct iov_iter *i)
1564 {
1565         struct csum_state *csstate = _csstate;
1566         __wsum sum, next;
1567
1568         if (WARN_ON_ONCE(i->data_source))
1569                 return 0;
1570         if (unlikely(iov_iter_is_discard(i))) {
1571                 // can't use csum_memcpy() for that one - data is not copied
1572                 csstate->csum = csum_block_add(csstate->csum,
1573                                                csum_partial(addr, bytes, 0),
1574                                                csstate->off);
1575                 csstate->off += bytes;
1576                 return bytes;
1577         }
1578
1579         sum = csum_shift(csstate->csum, csstate->off);
1580         if (unlikely(iov_iter_is_pipe(i)))
1581                 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1582         else iterate_and_advance(i, bytes, base, len, off, ({
1583                 next = csum_and_copy_to_user(addr + off, base, len);
1584                 sum = csum_block_add(sum, next, off);
1585                 next ? 0 : len;
1586         }), ({
1587                 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1588         })
1589         )
1590         csstate->csum = csum_shift(sum, csstate->off);
1591         csstate->off += bytes;
1592         return bytes;
1593 }
1594 EXPORT_SYMBOL(csum_and_copy_to_iter);
1595
1596 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1597                 struct iov_iter *i)
1598 {
1599 #ifdef CONFIG_CRYPTO_HASH
1600         struct ahash_request *hash = hashp;
1601         struct scatterlist sg;
1602         size_t copied;
1603
1604         copied = copy_to_iter(addr, bytes, i);
1605         sg_init_one(&sg, addr, copied);
1606         ahash_request_set_crypt(hash, &sg, NULL, copied);
1607         crypto_ahash_update(hash);
1608         return copied;
1609 #else
1610         return 0;
1611 #endif
1612 }
1613 EXPORT_SYMBOL(hash_and_copy_to_iter);
1614
1615 static int iov_npages(const struct iov_iter *i, int maxpages)
1616 {
1617         size_t skip = i->iov_offset, size = i->count;
1618         const struct iovec *p;
1619         int npages = 0;
1620
1621         for (p = i->iov; size; skip = 0, p++) {
1622                 unsigned offs = offset_in_page(p->iov_base + skip);
1623                 size_t len = min(p->iov_len - skip, size);
1624
1625                 if (len) {
1626                         size -= len;
1627                         npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1628                         if (unlikely(npages > maxpages))
1629                                 return maxpages;
1630                 }
1631         }
1632         return npages;
1633 }
1634
1635 static int bvec_npages(const struct iov_iter *i, int maxpages)
1636 {
1637         size_t skip = i->iov_offset, size = i->count;
1638         const struct bio_vec *p;
1639         int npages = 0;
1640
1641         for (p = i->bvec; size; skip = 0, p++) {
1642                 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1643                 size_t len = min(p->bv_len - skip, size);
1644
1645                 size -= len;
1646                 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1647                 if (unlikely(npages > maxpages))
1648                         return maxpages;
1649         }
1650         return npages;
1651 }
1652
1653 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1654 {
1655         if (unlikely(!i->count))
1656                 return 0;
1657         if (likely(iter_is_ubuf(i))) {
1658                 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1659                 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1660                 return min(npages, maxpages);
1661         }
1662         /* iovec and kvec have identical layouts */
1663         if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1664                 return iov_npages(i, maxpages);
1665         if (iov_iter_is_bvec(i))
1666                 return bvec_npages(i, maxpages);
1667         if (iov_iter_is_pipe(i)) {
1668                 int npages;
1669
1670                 if (!sanity(i))
1671                         return 0;
1672
1673                 pipe_npages(i, &npages);
1674                 return min(npages, maxpages);
1675         }
1676         if (iov_iter_is_xarray(i)) {
1677                 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1678                 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1679                 return min(npages, maxpages);
1680         }
1681         return 0;
1682 }
1683 EXPORT_SYMBOL(iov_iter_npages);
1684
1685 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1686 {
1687         *new = *old;
1688         if (unlikely(iov_iter_is_pipe(new))) {
1689                 WARN_ON(1);
1690                 return NULL;
1691         }
1692         if (iov_iter_is_bvec(new))
1693                 return new->bvec = kmemdup(new->bvec,
1694                                     new->nr_segs * sizeof(struct bio_vec),
1695                                     flags);
1696         else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1697                 /* iovec and kvec have identical layout */
1698                 return new->iov = kmemdup(new->iov,
1699                                    new->nr_segs * sizeof(struct iovec),
1700                                    flags);
1701         return NULL;
1702 }
1703 EXPORT_SYMBOL(dup_iter);
1704
1705 static int copy_compat_iovec_from_user(struct iovec *iov,
1706                 const struct iovec __user *uvec, unsigned long nr_segs)
1707 {
1708         const struct compat_iovec __user *uiov =
1709                 (const struct compat_iovec __user *)uvec;
1710         int ret = -EFAULT, i;
1711
1712         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1713                 return -EFAULT;
1714
1715         for (i = 0; i < nr_segs; i++) {
1716                 compat_uptr_t buf;
1717                 compat_ssize_t len;
1718
1719                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1720                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1721
1722                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1723                 if (len < 0) {
1724                         ret = -EINVAL;
1725                         goto uaccess_end;
1726                 }
1727                 iov[i].iov_base = compat_ptr(buf);
1728                 iov[i].iov_len = len;
1729         }
1730
1731         ret = 0;
1732 uaccess_end:
1733         user_access_end();
1734         return ret;
1735 }
1736
1737 static int copy_iovec_from_user(struct iovec *iov,
1738                 const struct iovec __user *uvec, unsigned long nr_segs)
1739 {
1740         unsigned long seg;
1741
1742         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1743                 return -EFAULT;
1744         for (seg = 0; seg < nr_segs; seg++) {
1745                 if ((ssize_t)iov[seg].iov_len < 0)
1746                         return -EINVAL;
1747         }
1748
1749         return 0;
1750 }
1751
1752 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1753                 unsigned long nr_segs, unsigned long fast_segs,
1754                 struct iovec *fast_iov, bool compat)
1755 {
1756         struct iovec *iov = fast_iov;
1757         int ret;
1758
1759         /*
1760          * SuS says "The readv() function *may* fail if the iovcnt argument was
1761          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1762          * traditionally returned zero for zero segments, so...
1763          */
1764         if (nr_segs == 0)
1765                 return iov;
1766         if (nr_segs > UIO_MAXIOV)
1767                 return ERR_PTR(-EINVAL);
1768         if (nr_segs > fast_segs) {
1769                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1770                 if (!iov)
1771                         return ERR_PTR(-ENOMEM);
1772         }
1773
1774         if (compat)
1775                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1776         else
1777                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1778         if (ret) {
1779                 if (iov != fast_iov)
1780                         kfree(iov);
1781                 return ERR_PTR(ret);
1782         }
1783
1784         return iov;
1785 }
1786
1787 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1788                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1789                  struct iov_iter *i, bool compat)
1790 {
1791         ssize_t total_len = 0;
1792         unsigned long seg;
1793         struct iovec *iov;
1794
1795         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1796         if (IS_ERR(iov)) {
1797                 *iovp = NULL;
1798                 return PTR_ERR(iov);
1799         }
1800
1801         /*
1802          * According to the Single Unix Specification we should return EINVAL if
1803          * an element length is < 0 when cast to ssize_t or if the total length
1804          * would overflow the ssize_t return value of the system call.
1805          *
1806          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1807          * overflow case.
1808          */
1809         for (seg = 0; seg < nr_segs; seg++) {
1810                 ssize_t len = (ssize_t)iov[seg].iov_len;
1811
1812                 if (!access_ok(iov[seg].iov_base, len)) {
1813                         if (iov != *iovp)
1814                                 kfree(iov);
1815                         *iovp = NULL;
1816                         return -EFAULT;
1817                 }
1818
1819                 if (len > MAX_RW_COUNT - total_len) {
1820                         len = MAX_RW_COUNT - total_len;
1821                         iov[seg].iov_len = len;
1822                 }
1823                 total_len += len;
1824         }
1825
1826         iov_iter_init(i, type, iov, nr_segs, total_len);
1827         if (iov == *iovp)
1828                 *iovp = NULL;
1829         else
1830                 *iovp = iov;
1831         return total_len;
1832 }
1833
1834 /**
1835  * import_iovec() - Copy an array of &struct iovec from userspace
1836  *     into the kernel, check that it is valid, and initialize a new
1837  *     &struct iov_iter iterator to access it.
1838  *
1839  * @type: One of %READ or %WRITE.
1840  * @uvec: Pointer to the userspace array.
1841  * @nr_segs: Number of elements in userspace array.
1842  * @fast_segs: Number of elements in @iov.
1843  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1844  *     on-stack) kernel array.
1845  * @i: Pointer to iterator that will be initialized on success.
1846  *
1847  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1848  * then this function places %NULL in *@iov on return. Otherwise, a new
1849  * array will be allocated and the result placed in *@iov. This means that
1850  * the caller may call kfree() on *@iov regardless of whether the small
1851  * on-stack array was used or not (and regardless of whether this function
1852  * returns an error or not).
1853  *
1854  * Return: Negative error code on error, bytes imported on success
1855  */
1856 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1857                  unsigned nr_segs, unsigned fast_segs,
1858                  struct iovec **iovp, struct iov_iter *i)
1859 {
1860         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1861                               in_compat_syscall());
1862 }
1863 EXPORT_SYMBOL(import_iovec);
1864
1865 int import_single_range(int rw, void __user *buf, size_t len,
1866                  struct iovec *iov, struct iov_iter *i)
1867 {
1868         if (len > MAX_RW_COUNT)
1869                 len = MAX_RW_COUNT;
1870         if (unlikely(!access_ok(buf, len)))
1871                 return -EFAULT;
1872
1873         iov->iov_base = buf;
1874         iov->iov_len = len;
1875         iov_iter_init(i, rw, iov, 1, len);
1876         return 0;
1877 }
1878 EXPORT_SYMBOL(import_single_range);
1879
1880 /**
1881  * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1882  *     iov_iter_save_state() was called.
1883  *
1884  * @i: &struct iov_iter to restore
1885  * @state: state to restore from
1886  *
1887  * Used after iov_iter_save_state() to bring restore @i, if operations may
1888  * have advanced it.
1889  *
1890  * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1891  */
1892 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1893 {
1894         if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
1895                          !iov_iter_is_kvec(i) && !iter_is_ubuf(i))
1896                 return;
1897         i->iov_offset = state->iov_offset;
1898         i->count = state->count;
1899         if (iter_is_ubuf(i))
1900                 return;
1901         /*
1902          * For the *vec iters, nr_segs + iov is constant - if we increment
1903          * the vec, then we also decrement the nr_segs count. Hence we don't
1904          * need to track both of these, just one is enough and we can deduct
1905          * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1906          * size, so we can just increment the iov pointer as they are unionzed.
1907          * ITER_BVEC _may_ be the same size on some archs, but on others it is
1908          * not. Be safe and handle it separately.
1909          */
1910         BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1911         if (iov_iter_is_bvec(i))
1912                 i->bvec -= state->nr_segs - i->nr_segs;
1913         else
1914                 i->iov -= state->nr_segs - i->nr_segs;
1915         i->nr_segs = state->nr_segs;
1916 }