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