1/* 2 * Copyright (C) 2007 Jens Axboe <jens.axboe@oracle.com> 3 * 4 * Scatterlist handling helpers. 5 * 6 * This source code is licensed under the GNU General Public License, 7 * Version 2. See the file COPYING for more details. 8 */ 9#include <linux/export.h> 10#include <linux/slab.h> 11#include <linux/scatterlist.h> 12#include <linux/highmem.h> 13#include <linux/kmemleak.h> 14 15/** 16 * sg_next - return the next scatterlist entry in a list 17 * @sg: The current sg entry 18 * 19 * Description: 20 * Usually the next entry will be @sg@ + 1, but if this sg element is part 21 * of a chained scatterlist, it could jump to the start of a new 22 * scatterlist array. 23 * 24 **/ 25struct scatterlist *sg_next(struct scatterlist *sg) 26{ 27#ifdef CONFIG_DEBUG_SG 28 BUG_ON(sg->sg_magic != SG_MAGIC); 29#endif 30 if (sg_is_last(sg)) 31 return NULL; 32 33 sg++; 34 if (unlikely(sg_is_chain(sg))) 35 sg = sg_chain_ptr(sg); 36 37 return sg; 38} 39EXPORT_SYMBOL(sg_next); 40 41/** 42 * sg_nents - return total count of entries in scatterlist 43 * @sg: The scatterlist 44 * 45 * Description: 46 * Allows to know how many entries are in sg, taking into acount 47 * chaining as well 48 * 49 **/ 50int sg_nents(struct scatterlist *sg) 51{ 52 int nents; 53 for (nents = 0; sg; sg = sg_next(sg)) 54 nents++; 55 return nents; 56} 57EXPORT_SYMBOL(sg_nents); 58 59/** 60 * sg_nents_for_len - return total count of entries in scatterlist 61 * needed to satisfy the supplied length 62 * @sg: The scatterlist 63 * @len: The total required length 64 * 65 * Description: 66 * Determines the number of entries in sg that are required to meet 67 * the supplied length, taking into acount chaining as well 68 * 69 * Returns: 70 * the number of sg entries needed, negative error on failure 71 * 72 **/ 73int sg_nents_for_len(struct scatterlist *sg, u64 len) 74{ 75 int nents; 76 u64 total; 77 78 if (!len) 79 return 0; 80 81 for (nents = 0, total = 0; sg; sg = sg_next(sg)) { 82 nents++; 83 total += sg->length; 84 if (total >= len) 85 return nents; 86 } 87 88 return -EINVAL; 89} 90EXPORT_SYMBOL(sg_nents_for_len); 91 92/** 93 * sg_last - return the last scatterlist entry in a list 94 * @sgl: First entry in the scatterlist 95 * @nents: Number of entries in the scatterlist 96 * 97 * Description: 98 * Should only be used casually, it (currently) scans the entire list 99 * to get the last entry. 100 * 101 * Note that the @sgl@ pointer passed in need not be the first one, 102 * the important bit is that @nents@ denotes the number of entries that 103 * exist from @sgl@. 104 * 105 **/ 106struct scatterlist *sg_last(struct scatterlist *sgl, unsigned int nents) 107{ 108 struct scatterlist *sg, *ret = NULL; 109 unsigned int i; 110 111 for_each_sg(sgl, sg, nents, i) 112 ret = sg; 113 114#ifdef CONFIG_DEBUG_SG 115 BUG_ON(sgl[0].sg_magic != SG_MAGIC); 116 BUG_ON(!sg_is_last(ret)); 117#endif 118 return ret; 119} 120EXPORT_SYMBOL(sg_last); 121 122/** 123 * sg_init_table - Initialize SG table 124 * @sgl: The SG table 125 * @nents: Number of entries in table 126 * 127 * Notes: 128 * If this is part of a chained sg table, sg_mark_end() should be 129 * used only on the last table part. 130 * 131 **/ 132void sg_init_table(struct scatterlist *sgl, unsigned int nents) 133{ 134 memset(sgl, 0, sizeof(*sgl) * nents); 135#ifdef CONFIG_DEBUG_SG 136 { 137 unsigned int i; 138 for (i = 0; i < nents; i++) 139 sgl[i].sg_magic = SG_MAGIC; 140 } 141#endif 142 sg_mark_end(&sgl[nents - 1]); 143} 144EXPORT_SYMBOL(sg_init_table); 145 146/** 147 * sg_init_one - Initialize a single entry sg list 148 * @sg: SG entry 149 * @buf: Virtual address for IO 150 * @buflen: IO length 151 * 152 **/ 153void sg_init_one(struct scatterlist *sg, const void *buf, unsigned int buflen) 154{ 155 sg_init_table(sg, 1); 156 sg_set_buf(sg, buf, buflen); 157} 158EXPORT_SYMBOL(sg_init_one); 159 160/* 161 * The default behaviour of sg_alloc_table() is to use these kmalloc/kfree 162 * helpers. 163 */ 164static struct scatterlist *sg_kmalloc(unsigned int nents, gfp_t gfp_mask) 165{ 166 if (nents == SG_MAX_SINGLE_ALLOC) { 167 /* 168 * Kmemleak doesn't track page allocations as they are not 169 * commonly used (in a raw form) for kernel data structures. 170 * As we chain together a list of pages and then a normal 171 * kmalloc (tracked by kmemleak), in order to for that last 172 * allocation not to become decoupled (and thus a 173 * false-positive) we need to inform kmemleak of all the 174 * intermediate allocations. 175 */ 176 void *ptr = (void *) __get_free_page(gfp_mask); 177 kmemleak_alloc(ptr, PAGE_SIZE, 1, gfp_mask); 178 return ptr; 179 } else 180 return kmalloc(nents * sizeof(struct scatterlist), gfp_mask); 181} 182 183static void sg_kfree(struct scatterlist *sg, unsigned int nents) 184{ 185 if (nents == SG_MAX_SINGLE_ALLOC) { 186 kmemleak_free(sg); 187 free_page((unsigned long) sg); 188 } else 189 kfree(sg); 190} 191 192/** 193 * __sg_free_table - Free a previously mapped sg table 194 * @table: The sg table header to use 195 * @max_ents: The maximum number of entries per single scatterlist 196 * @skip_first_chunk: don't free the (preallocated) first scatterlist chunk 197 * @free_fn: Free function 198 * 199 * Description: 200 * Free an sg table previously allocated and setup with 201 * __sg_alloc_table(). The @max_ents value must be identical to 202 * that previously used with __sg_alloc_table(). 203 * 204 **/ 205void __sg_free_table(struct sg_table *table, unsigned int max_ents, 206 bool skip_first_chunk, sg_free_fn *free_fn) 207{ 208 struct scatterlist *sgl, *next; 209 210 if (unlikely(!table->sgl)) 211 return; 212 213 sgl = table->sgl; 214 while (table->orig_nents) { 215 unsigned int alloc_size = table->orig_nents; 216 unsigned int sg_size; 217 218 /* 219 * If we have more than max_ents segments left, 220 * then assign 'next' to the sg table after the current one. 221 * sg_size is then one less than alloc size, since the last 222 * element is the chain pointer. 223 */ 224 if (alloc_size > max_ents) { 225 next = sg_chain_ptr(&sgl[max_ents - 1]); 226 alloc_size = max_ents; 227 sg_size = alloc_size - 1; 228 } else { 229 sg_size = alloc_size; 230 next = NULL; 231 } 232 233 table->orig_nents -= sg_size; 234 if (skip_first_chunk) 235 skip_first_chunk = false; 236 else 237 free_fn(sgl, alloc_size); 238 sgl = next; 239 } 240 241 table->sgl = NULL; 242} 243EXPORT_SYMBOL(__sg_free_table); 244 245/** 246 * sg_free_table - Free a previously allocated sg table 247 * @table: The mapped sg table header 248 * 249 **/ 250void sg_free_table(struct sg_table *table) 251{ 252 __sg_free_table(table, SG_MAX_SINGLE_ALLOC, false, sg_kfree); 253} 254EXPORT_SYMBOL(sg_free_table); 255 256/** 257 * __sg_alloc_table - Allocate and initialize an sg table with given allocator 258 * @table: The sg table header to use 259 * @nents: Number of entries in sg list 260 * @max_ents: The maximum number of entries the allocator returns per call 261 * @gfp_mask: GFP allocation mask 262 * @alloc_fn: Allocator to use 263 * 264 * Description: 265 * This function returns a @table @nents long. The allocator is 266 * defined to return scatterlist chunks of maximum size @max_ents. 267 * Thus if @nents is bigger than @max_ents, the scatterlists will be 268 * chained in units of @max_ents. 269 * 270 * Notes: 271 * If this function returns non-0 (eg failure), the caller must call 272 * __sg_free_table() to cleanup any leftover allocations. 273 * 274 **/ 275int __sg_alloc_table(struct sg_table *table, unsigned int nents, 276 unsigned int max_ents, struct scatterlist *first_chunk, 277 gfp_t gfp_mask, sg_alloc_fn *alloc_fn) 278{ 279 struct scatterlist *sg, *prv; 280 unsigned int left; 281 282 memset(table, 0, sizeof(*table)); 283 284 if (nents == 0) 285 return -EINVAL; 286#ifndef CONFIG_ARCH_HAS_SG_CHAIN 287 if (WARN_ON_ONCE(nents > max_ents)) 288 return -EINVAL; 289#endif 290 291 left = nents; 292 prv = NULL; 293 do { 294 unsigned int sg_size, alloc_size = left; 295 296 if (alloc_size > max_ents) { 297 alloc_size = max_ents; 298 sg_size = alloc_size - 1; 299 } else 300 sg_size = alloc_size; 301 302 left -= sg_size; 303 304 if (first_chunk) { 305 sg = first_chunk; 306 first_chunk = NULL; 307 } else { 308 sg = alloc_fn(alloc_size, gfp_mask); 309 } 310 if (unlikely(!sg)) { 311 /* 312 * Adjust entry count to reflect that the last 313 * entry of the previous table won't be used for 314 * linkage. Without this, sg_kfree() may get 315 * confused. 316 */ 317 if (prv) 318 table->nents = ++table->orig_nents; 319 320 return -ENOMEM; 321 } 322 323 sg_init_table(sg, alloc_size); 324 table->nents = table->orig_nents += sg_size; 325 326 /* 327 * If this is the first mapping, assign the sg table header. 328 * If this is not the first mapping, chain previous part. 329 */ 330 if (prv) 331 sg_chain(prv, max_ents, sg); 332 else 333 table->sgl = sg; 334 335 /* 336 * If no more entries after this one, mark the end 337 */ 338 if (!left) 339 sg_mark_end(&sg[sg_size - 1]); 340 341 prv = sg; 342 } while (left); 343 344 return 0; 345} 346EXPORT_SYMBOL(__sg_alloc_table); 347 348/** 349 * sg_alloc_table - Allocate and initialize an sg table 350 * @table: The sg table header to use 351 * @nents: Number of entries in sg list 352 * @gfp_mask: GFP allocation mask 353 * 354 * Description: 355 * Allocate and initialize an sg table. If @nents@ is larger than 356 * SG_MAX_SINGLE_ALLOC a chained sg table will be setup. 357 * 358 **/ 359int sg_alloc_table(struct sg_table *table, unsigned int nents, gfp_t gfp_mask) 360{ 361 int ret; 362 363 ret = __sg_alloc_table(table, nents, SG_MAX_SINGLE_ALLOC, 364 NULL, gfp_mask, sg_kmalloc); 365 if (unlikely(ret)) 366 __sg_free_table(table, SG_MAX_SINGLE_ALLOC, false, sg_kfree); 367 368 return ret; 369} 370EXPORT_SYMBOL(sg_alloc_table); 371 372/** 373 * sg_alloc_table_from_pages - Allocate and initialize an sg table from 374 * an array of pages 375 * @sgt: The sg table header to use 376 * @pages: Pointer to an array of page pointers 377 * @n_pages: Number of pages in the pages array 378 * @offset: Offset from start of the first page to the start of a buffer 379 * @size: Number of valid bytes in the buffer (after offset) 380 * @gfp_mask: GFP allocation mask 381 * 382 * Description: 383 * Allocate and initialize an sg table from a list of pages. Contiguous 384 * ranges of the pages are squashed into a single scatterlist node. A user 385 * may provide an offset at a start and a size of valid data in a buffer 386 * specified by the page array. The returned sg table is released by 387 * sg_free_table. 388 * 389 * Returns: 390 * 0 on success, negative error on failure 391 */ 392int sg_alloc_table_from_pages(struct sg_table *sgt, 393 struct page **pages, unsigned int n_pages, 394 unsigned long offset, unsigned long size, 395 gfp_t gfp_mask) 396{ 397 unsigned int chunks; 398 unsigned int i; 399 unsigned int cur_page; 400 int ret; 401 struct scatterlist *s; 402 403 /* compute number of contiguous chunks */ 404 chunks = 1; 405 for (i = 1; i < n_pages; ++i) 406 if (page_to_pfn(pages[i]) != page_to_pfn(pages[i - 1]) + 1) 407 ++chunks; 408 409 ret = sg_alloc_table(sgt, chunks, gfp_mask); 410 if (unlikely(ret)) 411 return ret; 412 413 /* merging chunks and putting them into the scatterlist */ 414 cur_page = 0; 415 for_each_sg(sgt->sgl, s, sgt->orig_nents, i) { 416 unsigned long chunk_size; 417 unsigned int j; 418 419 /* look for the end of the current chunk */ 420 for (j = cur_page + 1; j < n_pages; ++j) 421 if (page_to_pfn(pages[j]) != 422 page_to_pfn(pages[j - 1]) + 1) 423 break; 424 425 chunk_size = ((j - cur_page) << PAGE_SHIFT) - offset; 426 sg_set_page(s, pages[cur_page], min(size, chunk_size), offset); 427 size -= chunk_size; 428 offset = 0; 429 cur_page = j; 430 } 431 432 return 0; 433} 434EXPORT_SYMBOL(sg_alloc_table_from_pages); 435 436void __sg_page_iter_start(struct sg_page_iter *piter, 437 struct scatterlist *sglist, unsigned int nents, 438 unsigned long pgoffset) 439{ 440 piter->__pg_advance = 0; 441 piter->__nents = nents; 442 443 piter->sg = sglist; 444 piter->sg_pgoffset = pgoffset; 445} 446EXPORT_SYMBOL(__sg_page_iter_start); 447 448static int sg_page_count(struct scatterlist *sg) 449{ 450 return PAGE_ALIGN(sg->offset + sg->length) >> PAGE_SHIFT; 451} 452 453bool __sg_page_iter_next(struct sg_page_iter *piter) 454{ 455 if (!piter->__nents || !piter->sg) 456 return false; 457 458 piter->sg_pgoffset += piter->__pg_advance; 459 piter->__pg_advance = 1; 460 461 while (piter->sg_pgoffset >= sg_page_count(piter->sg)) { 462 piter->sg_pgoffset -= sg_page_count(piter->sg); 463 piter->sg = sg_next(piter->sg); 464 if (!--piter->__nents || !piter->sg) 465 return false; 466 } 467 468 return true; 469} 470EXPORT_SYMBOL(__sg_page_iter_next); 471 472/** 473 * sg_miter_start - start mapping iteration over a sg list 474 * @miter: sg mapping iter to be started 475 * @sgl: sg list to iterate over 476 * @nents: number of sg entries 477 * 478 * Description: 479 * Starts mapping iterator @miter. 480 * 481 * Context: 482 * Don't care. 483 */ 484void sg_miter_start(struct sg_mapping_iter *miter, struct scatterlist *sgl, 485 unsigned int nents, unsigned int flags) 486{ 487 memset(miter, 0, sizeof(struct sg_mapping_iter)); 488 489 __sg_page_iter_start(&miter->piter, sgl, nents, 0); 490 WARN_ON(!(flags & (SG_MITER_TO_SG | SG_MITER_FROM_SG))); 491 miter->__flags = flags; 492} 493EXPORT_SYMBOL(sg_miter_start); 494 495static bool sg_miter_get_next_page(struct sg_mapping_iter *miter) 496{ 497 if (!miter->__remaining) { 498 struct scatterlist *sg; 499 unsigned long pgoffset; 500 501 if (!__sg_page_iter_next(&miter->piter)) 502 return false; 503 504 sg = miter->piter.sg; 505 pgoffset = miter->piter.sg_pgoffset; 506 507 miter->__offset = pgoffset ? 0 : sg->offset; 508 miter->__remaining = sg->offset + sg->length - 509 (pgoffset << PAGE_SHIFT) - miter->__offset; 510 miter->__remaining = min_t(unsigned long, miter->__remaining, 511 PAGE_SIZE - miter->__offset); 512 } 513 514 return true; 515} 516 517/** 518 * sg_miter_skip - reposition mapping iterator 519 * @miter: sg mapping iter to be skipped 520 * @offset: number of bytes to plus the current location 521 * 522 * Description: 523 * Sets the offset of @miter to its current location plus @offset bytes. 524 * If mapping iterator @miter has been proceeded by sg_miter_next(), this 525 * stops @miter. 526 * 527 * Context: 528 * Don't care if @miter is stopped, or not proceeded yet. 529 * Otherwise, preemption disabled if the SG_MITER_ATOMIC is set. 530 * 531 * Returns: 532 * true if @miter contains the valid mapping. false if end of sg 533 * list is reached. 534 */ 535bool sg_miter_skip(struct sg_mapping_iter *miter, off_t offset) 536{ 537 sg_miter_stop(miter); 538 539 while (offset) { 540 off_t consumed; 541 542 if (!sg_miter_get_next_page(miter)) 543 return false; 544 545 consumed = min_t(off_t, offset, miter->__remaining); 546 miter->__offset += consumed; 547 miter->__remaining -= consumed; 548 offset -= consumed; 549 } 550 551 return true; 552} 553EXPORT_SYMBOL(sg_miter_skip); 554 555/** 556 * sg_miter_next - proceed mapping iterator to the next mapping 557 * @miter: sg mapping iter to proceed 558 * 559 * Description: 560 * Proceeds @miter to the next mapping. @miter should have been started 561 * using sg_miter_start(). On successful return, @miter->page, 562 * @miter->addr and @miter->length point to the current mapping. 563 * 564 * Context: 565 * Preemption disabled if SG_MITER_ATOMIC. Preemption must stay disabled 566 * till @miter is stopped. May sleep if !SG_MITER_ATOMIC. 567 * 568 * Returns: 569 * true if @miter contains the next mapping. false if end of sg 570 * list is reached. 571 */ 572bool sg_miter_next(struct sg_mapping_iter *miter) 573{ 574 sg_miter_stop(miter); 575 576 /* 577 * Get to the next page if necessary. 578 * __remaining, __offset is adjusted by sg_miter_stop 579 */ 580 if (!sg_miter_get_next_page(miter)) 581 return false; 582 583 miter->page = sg_page_iter_page(&miter->piter); 584 miter->consumed = miter->length = miter->__remaining; 585 586 if (miter->__flags & SG_MITER_ATOMIC) 587 miter->addr = kmap_atomic(miter->page) + miter->__offset; 588 else 589 miter->addr = kmap(miter->page) + miter->__offset; 590 591 return true; 592} 593EXPORT_SYMBOL(sg_miter_next); 594 595/** 596 * sg_miter_stop - stop mapping iteration 597 * @miter: sg mapping iter to be stopped 598 * 599 * Description: 600 * Stops mapping iterator @miter. @miter should have been started 601 * started using sg_miter_start(). A stopped iteration can be 602 * resumed by calling sg_miter_next() on it. This is useful when 603 * resources (kmap) need to be released during iteration. 604 * 605 * Context: 606 * Preemption disabled if the SG_MITER_ATOMIC is set. Don't care 607 * otherwise. 608 */ 609void sg_miter_stop(struct sg_mapping_iter *miter) 610{ 611 WARN_ON(miter->consumed > miter->length); 612 613 /* drop resources from the last iteration */ 614 if (miter->addr) { 615 miter->__offset += miter->consumed; 616 miter->__remaining -= miter->consumed; 617 618 if ((miter->__flags & SG_MITER_TO_SG) && 619 !PageSlab(miter->page)) 620 flush_kernel_dcache_page(miter->page); 621 622 if (miter->__flags & SG_MITER_ATOMIC) { 623 WARN_ON_ONCE(preemptible()); 624 kunmap_atomic(miter->addr); 625 } else 626 kunmap(miter->page); 627 628 miter->page = NULL; 629 miter->addr = NULL; 630 miter->length = 0; 631 miter->consumed = 0; 632 } 633} 634EXPORT_SYMBOL(sg_miter_stop); 635 636/** 637 * sg_copy_buffer - Copy data between a linear buffer and an SG list 638 * @sgl: The SG list 639 * @nents: Number of SG entries 640 * @buf: Where to copy from 641 * @buflen: The number of bytes to copy 642 * @skip: Number of bytes to skip before copying 643 * @to_buffer: transfer direction (true == from an sg list to a 644 * buffer, false == from a buffer to an sg list 645 * 646 * Returns the number of copied bytes. 647 * 648 **/ 649size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf, 650 size_t buflen, off_t skip, bool to_buffer) 651{ 652 unsigned int offset = 0; 653 struct sg_mapping_iter miter; 654 unsigned long flags; 655 unsigned int sg_flags = SG_MITER_ATOMIC; 656 657 if (to_buffer) 658 sg_flags |= SG_MITER_FROM_SG; 659 else 660 sg_flags |= SG_MITER_TO_SG; 661 662 sg_miter_start(&miter, sgl, nents, sg_flags); 663 664 if (!sg_miter_skip(&miter, skip)) 665 return false; 666 667 local_irq_save(flags); 668 669 while (sg_miter_next(&miter) && offset < buflen) { 670 unsigned int len; 671 672 len = min(miter.length, buflen - offset); 673 674 if (to_buffer) 675 memcpy(buf + offset, miter.addr, len); 676 else 677 memcpy(miter.addr, buf + offset, len); 678 679 offset += len; 680 } 681 682 sg_miter_stop(&miter); 683 684 local_irq_restore(flags); 685 return offset; 686} 687EXPORT_SYMBOL(sg_copy_buffer); 688 689/** 690 * sg_copy_from_buffer - Copy from a linear buffer to an SG list 691 * @sgl: The SG list 692 * @nents: Number of SG entries 693 * @buf: Where to copy from 694 * @buflen: The number of bytes to copy 695 * 696 * Returns the number of copied bytes. 697 * 698 **/ 699size_t sg_copy_from_buffer(struct scatterlist *sgl, unsigned int nents, 700 const void *buf, size_t buflen) 701{ 702 return sg_copy_buffer(sgl, nents, (void *)buf, buflen, 0, false); 703} 704EXPORT_SYMBOL(sg_copy_from_buffer); 705 706/** 707 * sg_copy_to_buffer - Copy from an SG list to a linear buffer 708 * @sgl: The SG list 709 * @nents: Number of SG entries 710 * @buf: Where to copy to 711 * @buflen: The number of bytes to copy 712 * 713 * Returns the number of copied bytes. 714 * 715 **/ 716size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents, 717 void *buf, size_t buflen) 718{ 719 return sg_copy_buffer(sgl, nents, buf, buflen, 0, true); 720} 721EXPORT_SYMBOL(sg_copy_to_buffer); 722 723/** 724 * sg_pcopy_from_buffer - Copy from a linear buffer to an SG list 725 * @sgl: The SG list 726 * @nents: Number of SG entries 727 * @buf: Where to copy from 728 * @buflen: The number of bytes to copy 729 * @skip: Number of bytes to skip before copying 730 * 731 * Returns the number of copied bytes. 732 * 733 **/ 734size_t sg_pcopy_from_buffer(struct scatterlist *sgl, unsigned int nents, 735 const void *buf, size_t buflen, off_t skip) 736{ 737 return sg_copy_buffer(sgl, nents, (void *)buf, buflen, skip, false); 738} 739EXPORT_SYMBOL(sg_pcopy_from_buffer); 740 741/** 742 * sg_pcopy_to_buffer - Copy from an SG list to a linear buffer 743 * @sgl: The SG list 744 * @nents: Number of SG entries 745 * @buf: Where to copy to 746 * @buflen: The number of bytes to copy 747 * @skip: Number of bytes to skip before copying 748 * 749 * Returns the number of copied bytes. 750 * 751 **/ 752size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents, 753 void *buf, size_t buflen, off_t skip) 754{ 755 return sg_copy_buffer(sgl, nents, buf, buflen, skip, true); 756} 757EXPORT_SYMBOL(sg_pcopy_to_buffer); 758