projects/linux·blob master · refs

1439 lines · 37.0 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2007 Jens Axboe <[email protected]>
4 *
5 * Scatterlist handling helpers.
6 */
7#include <linux/export.h>
8#include <linux/slab.h>
9#include <linux/scatterlist.h>
10#include <linux/highmem.h>
11#include <linux/kmemleak.h>
12#include <linux/bvec.h>
13#include <linux/uio.h>
14#include <linux/folio_queue.h>
15
16/**
17 * sg_nents - return total count of entries in scatterlist
18 * @sg: The scatterlist
19 *
20 * Description:
21 * Allows to know how many entries are in sg, taking into account
22 * chaining as well
23 *
24 **/
25int sg_nents(struct scatterlist *sg)
26{
27 int nents;
28 for (nents = 0; sg; sg = sg_next(sg))
29 nents++;
30 return nents;
31}
32EXPORT_SYMBOL(sg_nents);
33
34/**
35 * sg_nents_for_len - return total count of entries in scatterlist
36 * needed to satisfy the supplied length
37 * @sg: The scatterlist
38 * @len: The total required length
39 *
40 * Description:
41 * Determines the number of entries in sg that are required to meet
42 * the supplied length, taking into account chaining as well
43 *
44 * Returns:
45 * the number of sg entries needed, negative error on failure
46 *
47 **/
48int sg_nents_for_len(struct scatterlist *sg, u64 len)
49{
50 int nents;
51 u64 total;
52
53 if (!len)
54 return 0;
55
56 for (nents = 0, total = 0; sg; sg = sg_next(sg)) {
57 nents++;
58 total += sg->length;
59 if (total >= len)
60 return nents;
61 }
62
63 return -EINVAL;
64}
65EXPORT_SYMBOL(sg_nents_for_len);
66
67/**
68 * sg_nents_for_dma - return the count of DMA-capable entries in scatterlist
69 * @sgl: The scatterlist
70 * @sglen: The current number of entries
71 * @len: The maximum length of DMA-capable block
72 *
73 * Description:
74 * Determines the number of entries in @sgl which would be permitted in
75 * DMA-capable transfer if list had been split accordingly, taking into
76 * account chaining as well.
77 *
78 * Returns:
79 * the number of sgl entries needed
80 *
81 **/
82int sg_nents_for_dma(struct scatterlist *sgl, unsigned int sglen, size_t len)
83{
84 struct scatterlist *sg;
85 int i, nents = 0;
86
87 for_each_sg(sgl, sg, sglen, i)
88 nents += DIV_ROUND_UP(sg_dma_len(sg), len);
89 return nents;
90}
91EXPORT_SYMBOL(sg_nents_for_dma);
92
93/**
94 * sg_last - return the last scatterlist entry in a list
95 * @sgl: First entry in the scatterlist
96 * @nents: Number of entries in the scatterlist
97 *
98 * Description:
99 * Should only be used casually, it (currently) scans the entire list
100 * to get the last entry.
101 *
102 * Note that the @sgl pointer passed in need not be the first one,
103 * the important bit is that @nents denotes the number of entries that
104 * exist from @sgl.
105 *
106 **/
107struct scatterlist *sg_last(struct scatterlist *sgl, unsigned int nents)
108{
109 struct scatterlist *sg, *ret = NULL;
110 unsigned int i;
111
112 for_each_sg(sgl, sg, nents, i)
113 ret = sg;
114
115 BUG_ON(!sg_is_last(ret));
116 return ret;
117}
118EXPORT_SYMBOL(sg_last);
119
120/**
121 * sg_init_table - Initialize SG table
122 * @sgl: The SG table
123 * @nents: Number of entries in table
124 *
125 * Notes:
126 * If this is part of a chained sg table, sg_mark_end() should be
127 * used only on the last table part.
128 *
129 **/
130void sg_init_table(struct scatterlist *sgl, unsigned int nents)
131{
132 memset(sgl, 0, sizeof(*sgl) * nents);
133 sg_init_marker(sgl, nents);
134}
135EXPORT_SYMBOL(sg_init_table);
136
137/**
138 * sg_init_one - Initialize a single entry sg list
139 * @sg: SG entry
140 * @buf: Virtual address for IO
141 * @buflen: IO length
142 *
143 **/
144void sg_init_one(struct scatterlist *sg, const void *buf, unsigned int buflen)
145{
146 sg_init_table(sg, 1);
147 sg_set_buf(sg, buf, buflen);
148}
149EXPORT_SYMBOL(sg_init_one);
150
151/*
152 * The default behaviour of sg_alloc_table() is to use these kmalloc/kfree
153 * helpers.
154 */
155static struct scatterlist *sg_kmalloc(unsigned int nents, gfp_t gfp_mask)
156{
157 if (nents == SG_MAX_SINGLE_ALLOC) {
158 /*
159 * Kmemleak doesn't track page allocations as they are not
160 * commonly used (in a raw form) for kernel data structures.
161 * As we chain together a list of pages and then a normal
162 * kmalloc (tracked by kmemleak), in order to for that last
163 * allocation not to become decoupled (and thus a
164 * false-positive) we need to inform kmemleak of all the
165 * intermediate allocations.
166 */
167 void *ptr = (void *) __get_free_page(gfp_mask);
168 kmemleak_alloc(ptr, PAGE_SIZE, 1, gfp_mask);
169 return ptr;
170 } else
171 return kmalloc_objs(struct scatterlist, nents, gfp_mask);
172}
173
174static void sg_kfree(struct scatterlist *sg, unsigned int nents)
175{
176 if (nents == SG_MAX_SINGLE_ALLOC) {
177 kmemleak_free(sg);
178 free_page((unsigned long) sg);
179 } else
180 kfree(sg);
181}
182
183/**
184 * __sg_free_table - Free a previously mapped sg table
185 * @table: The sg table header to use
186 * @max_ents: The maximum number of entries per single scatterlist
187 * @nents_first_chunk: Number of entries int the (preallocated) first
188 * scatterlist chunk, 0 means no such preallocated first chunk
189 * @free_fn: Free function
190 * @num_ents: Number of entries in the table
191 *
192 * Description:
193 * Free an sg table previously allocated and setup with
194 * __sg_alloc_table(). The @max_ents value must be identical to
195 * that previously used with __sg_alloc_table().
196 *
197 **/
198void __sg_free_table(struct sg_table *table, unsigned int max_ents,
199 unsigned int nents_first_chunk, sg_free_fn *free_fn,
200 unsigned int num_ents)
201{
202 struct scatterlist *sgl, *next;
203 unsigned curr_max_ents = nents_first_chunk ?: max_ents;
204
205 if (unlikely(!table->sgl))
206 return;
207
208 sgl = table->sgl;
209 while (num_ents) {
210 unsigned int alloc_size = num_ents;
211 unsigned int sg_size;
212
213 /*
214 * If we have more than max_ents segments left,
215 * then assign 'next' to the sg table after the current one.
216 * sg_size is then one less than alloc size, since the last
217 * element is the chain pointer.
218 */
219 if (alloc_size > curr_max_ents) {
220 next = sg_chain_ptr(&sgl[curr_max_ents - 1]);
221 alloc_size = curr_max_ents;
222 sg_size = alloc_size - 1;
223 } else {
224 sg_size = alloc_size;
225 next = NULL;
226 }
227
228 num_ents -= sg_size;
229 if (nents_first_chunk)
230 nents_first_chunk = 0;
231 else
232 free_fn(sgl, alloc_size);
233 sgl = next;
234 curr_max_ents = max_ents;
235 }
236
237 table->sgl = NULL;
238}
239EXPORT_SYMBOL(__sg_free_table);
240
241/**
242 * sg_free_append_table - Free a previously allocated append sg table.
243 * @table: The mapped sg append table header
244 *
245 **/
246void sg_free_append_table(struct sg_append_table *table)
247{
248 __sg_free_table(&table->sgt, SG_MAX_SINGLE_ALLOC, 0, sg_kfree,
249 table->total_nents);
250}
251EXPORT_SYMBOL(sg_free_append_table);
252
253
254/**
255 * sg_free_table - Free a previously allocated sg table
256 * @table: The mapped sg table header
257 *
258 **/
259void sg_free_table(struct sg_table *table)
260{
261 __sg_free_table(table, SG_MAX_SINGLE_ALLOC, 0, sg_kfree,
262 table->orig_nents);
263}
264EXPORT_SYMBOL(sg_free_table);
265
266/**
267 * __sg_alloc_table - Allocate and initialize an sg table with given allocator
268 * @table: The sg table header to use
269 * @nents: Number of entries in sg list
270 * @max_ents: The maximum number of entries the allocator returns per call
271 * @first_chunk: first SGL if preallocated (may be %NULL)
272 * @nents_first_chunk: Number of entries in the (preallocated) first
273 * scatterlist chunk, 0 means no such preallocated chunk provided by user
274 * @gfp_mask: GFP allocation mask
275 * @alloc_fn: Allocator to use
276 *
277 * Description:
278 * This function returns a @table @nents long. The allocator is
279 * defined to return scatterlist chunks of maximum size @max_ents.
280 * Thus if @nents is bigger than @max_ents, the scatterlists will be
281 * chained in units of @max_ents.
282 *
283 * Notes:
284 * If this function returns non-0 (eg failure), the caller must call
285 * __sg_free_table() to cleanup any leftover allocations.
286 *
287 **/
288int __sg_alloc_table(struct sg_table *table, unsigned int nents,
289 unsigned int max_ents, struct scatterlist *first_chunk,
290 unsigned int nents_first_chunk, gfp_t gfp_mask,
291 sg_alloc_fn *alloc_fn)
292{
293 struct scatterlist *sg, *prv;
294 unsigned int left;
295 unsigned curr_max_ents = nents_first_chunk ?: max_ents;
296 unsigned prv_max_ents;
297
298 memset(table, 0, sizeof(*table));
299
300 if (nents == 0)
301 return -EINVAL;
302#ifdef CONFIG_ARCH_NO_SG_CHAIN
303 if (WARN_ON_ONCE(nents > max_ents))
304 return -EINVAL;
305#endif
306
307 left = nents;
308 prv = NULL;
309 do {
310 unsigned int sg_size, alloc_size = left;
311
312 if (alloc_size > curr_max_ents) {
313 alloc_size = curr_max_ents;
314 sg_size = alloc_size - 1;
315 } else
316 sg_size = alloc_size;
317
318 left -= sg_size;
319
320 if (first_chunk) {
321 sg = first_chunk;
322 first_chunk = NULL;
323 } else {
324 sg = alloc_fn(alloc_size, gfp_mask);
325 }
326 if (unlikely(!sg)) {
327 /*
328 * Adjust entry count to reflect that the last
329 * entry of the previous table won't be used for
330 * linkage. Without this, sg_kfree() may get
331 * confused.
332 */
333 if (prv)
334 table->nents = ++table->orig_nents;
335
336 return -ENOMEM;
337 }
338
339 sg_init_table(sg, alloc_size);
340 table->nents = table->orig_nents += sg_size;
341
342 /*
343 * If this is the first mapping, assign the sg table header.
344 * If this is not the first mapping, chain previous part.
345 */
346 if (prv)
347 sg_chain(prv, prv_max_ents, sg);
348 else
349 table->sgl = sg;
350
351 /*
352 * If no more entries after this one, mark the end
353 */
354 if (!left)
355 sg_mark_end(&sg[sg_size - 1]);
356
357 prv = sg;
358 prv_max_ents = curr_max_ents;
359 curr_max_ents = max_ents;
360 } while (left);
361
362 return 0;
363}
364EXPORT_SYMBOL(__sg_alloc_table);
365
366/**
367 * sg_alloc_table - Allocate and initialize an sg table
368 * @table: The sg table header to use
369 * @nents: Number of entries in sg list
370 * @gfp_mask: GFP allocation mask
371 *
372 * Description:
373 * Allocate and initialize an sg table. If @nents is larger than
374 * SG_MAX_SINGLE_ALLOC a chained sg table will be setup.
375 *
376 **/
377int sg_alloc_table(struct sg_table *table, unsigned int nents, gfp_t gfp_mask)
378{
379 int ret;
380
381 ret = __sg_alloc_table(table, nents, SG_MAX_SINGLE_ALLOC,
382 NULL, 0, gfp_mask, sg_kmalloc);
383 if (unlikely(ret))
384 sg_free_table(table);
385 return ret;
386}
387EXPORT_SYMBOL(sg_alloc_table);
388
389static struct scatterlist *get_next_sg(struct sg_append_table *table,
390 struct scatterlist *cur,
391 unsigned long needed_sges,
392 gfp_t gfp_mask)
393{
394 struct scatterlist *new_sg, *next_sg;
395 unsigned int alloc_size;
396
397 if (cur) {
398 next_sg = sg_next(cur);
399 /* Check if last entry should be keeped for chainning */
400 if (!sg_is_last(next_sg) || needed_sges == 1)
401 return next_sg;
402 }
403
404 alloc_size = min_t(unsigned long, needed_sges, SG_MAX_SINGLE_ALLOC);
405 new_sg = sg_kmalloc(alloc_size, gfp_mask);
406 if (!new_sg)
407 return ERR_PTR(-ENOMEM);
408 sg_init_table(new_sg, alloc_size);
409 if (cur) {
410 table->total_nents += alloc_size - 1;
411 __sg_chain(next_sg, new_sg);
412 } else {
413 table->sgt.sgl = new_sg;
414 table->total_nents = alloc_size;
415 }
416 return new_sg;
417}
418
419static bool pages_are_mergeable(struct page *a, struct page *b)
420{
421 if (page_to_pfn(a) != page_to_pfn(b) + 1)
422 return false;
423 if (!zone_device_pages_have_same_pgmap(a, b))
424 return false;
425 return true;
426}
427
428/**
429 * sg_alloc_append_table_from_pages - Allocate and initialize an append sg
430 * table from an array of pages
431 * @sgt_append: The sg append table to use
432 * @pages: Pointer to an array of page pointers
433 * @n_pages: Number of pages in the pages array
434 * @offset: Offset from start of the first page to the start of a buffer
435 * @size: Number of valid bytes in the buffer (after offset)
436 * @max_segment: Maximum size of a scatterlist element in bytes
437 * @left_pages: Left pages caller have to set after this call
438 * @gfp_mask: GFP allocation mask
439 *
440 * Description:
441 * In the first call it allocate and initialize an sg table from a list of
442 * pages, else reuse the scatterlist from sgt_append. Contiguous ranges of
443 * the pages are squashed into a single scatterlist entry up to the maximum
444 * size specified in @max_segment. A user may provide an offset at a start
445 * and a size of valid data in a buffer specified by the page array. The
446 * returned sg table is released by sg_free_append_table
447 *
448 * Returns:
449 * 0 on success, negative error on failure
450 *
451 * Notes:
452 * If this function returns non-0 (eg failure), the caller must call
453 * sg_free_append_table() to cleanup any leftover allocations.
454 *
455 * In the fist call, sgt_append must by initialized.
456 */
457int sg_alloc_append_table_from_pages(struct sg_append_table *sgt_append,
458 struct page **pages, unsigned int n_pages, unsigned int offset,
459 unsigned long size, unsigned int max_segment,
460 unsigned int left_pages, gfp_t gfp_mask)
461{
462 unsigned int chunks, cur_page, seg_len, i, prv_len = 0;
463 unsigned int added_nents = 0;
464 struct scatterlist *s = sgt_append->prv;
465 struct page *last_pg;
466
467 /*
468 * The algorithm below requires max_segment to be aligned to PAGE_SIZE
469 * otherwise it can overshoot.
470 */
471 max_segment = ALIGN_DOWN(max_segment, PAGE_SIZE);
472 if (WARN_ON(max_segment < PAGE_SIZE))
473 return -EINVAL;
474
475 if (IS_ENABLED(CONFIG_ARCH_NO_SG_CHAIN) && sgt_append->prv)
476 return -EOPNOTSUPP;
477
478 if (sgt_append->prv) {
479 unsigned long next_pfn;
480
481 if (WARN_ON(offset))
482 return -EINVAL;
483
484 /* Merge contiguous pages into the last SG */
485 prv_len = sgt_append->prv->length;
486 next_pfn = (sg_phys(sgt_append->prv) + prv_len) / PAGE_SIZE;
487 if (page_to_pfn(pages[0]) == next_pfn) {
488 last_pg = pfn_to_page(next_pfn - 1);
489 while (n_pages && pages_are_mergeable(pages[0], last_pg)) {
490 if (sgt_append->prv->length + PAGE_SIZE > max_segment)
491 break;
492 sgt_append->prv->length += PAGE_SIZE;
493 last_pg = pages[0];
494 pages++;
495 n_pages--;
496 }
497 if (!n_pages)
498 goto out;
499 }
500 }
501
502 /* compute number of contiguous chunks */
503 chunks = 1;
504 seg_len = 0;
505 for (i = 1; i < n_pages; i++) {
506 seg_len += PAGE_SIZE;
507 if (seg_len >= max_segment ||
508 !pages_are_mergeable(pages[i], pages[i - 1])) {
509 chunks++;
510 seg_len = 0;
511 }
512 }
513
514 /* merging chunks and putting them into the scatterlist */
515 cur_page = 0;
516 for (i = 0; i < chunks; i++) {
517 unsigned int j, chunk_size;
518
519 /* look for the end of the current chunk */
520 seg_len = 0;
521 for (j = cur_page + 1; j < n_pages; j++) {
522 seg_len += PAGE_SIZE;
523 if (seg_len >= max_segment ||
524 !pages_are_mergeable(pages[j], pages[j - 1]))
525 break;
526 }
527
528 /* Pass how many chunks might be left */
529 s = get_next_sg(sgt_append, s, chunks - i + left_pages,
530 gfp_mask);
531 if (IS_ERR(s)) {
532 /*
533 * Adjust entry length to be as before function was
534 * called.
535 */
536 if (sgt_append->prv)
537 sgt_append->prv->length = prv_len;
538 return PTR_ERR(s);
539 }
540 chunk_size = ((j - cur_page) << PAGE_SHIFT) - offset;
541 sg_set_page(s, pages[cur_page],
542 min_t(unsigned long, size, chunk_size), offset);
543 added_nents++;
544 size -= chunk_size;
545 offset = 0;
546 cur_page = j;
547 }
548 sgt_append->sgt.nents += added_nents;
549 sgt_append->sgt.orig_nents = sgt_append->sgt.nents;
550 sgt_append->prv = s;
551out:
552 if (!left_pages)
553 sg_mark_end(s);
554 return 0;
555}
556EXPORT_SYMBOL(sg_alloc_append_table_from_pages);
557
558/**
559 * sg_alloc_table_from_pages_segment - Allocate and initialize an sg table from
560 * an array of pages and given maximum
561 * segment.
562 * @sgt: The sg table header to use
563 * @pages: Pointer to an array of page pointers
564 * @n_pages: Number of pages in the pages array
565 * @offset: Offset from start of the first page to the start of a buffer
566 * @size: Number of valid bytes in the buffer (after offset)
567 * @max_segment: Maximum size of a scatterlist element in bytes
568 * @gfp_mask: GFP allocation mask
569 *
570 * Description:
571 * Allocate and initialize an sg table from a list of pages. Contiguous
572 * ranges of the pages are squashed into a single scatterlist node up to the
573 * maximum size specified in @max_segment. A user may provide an offset at a
574 * start and a size of valid data in a buffer specified by the page array.
575 *
576 * The returned sg table is released by sg_free_table.
577 *
578 * Returns:
579 * 0 on success, negative error on failure
580 */
581int sg_alloc_table_from_pages_segment(struct sg_table *sgt, struct page **pages,
582 unsigned int n_pages, unsigned int offset,
583 unsigned long size, unsigned int max_segment,
584 gfp_t gfp_mask)
585{
586 struct sg_append_table append = {};
587 int err;
588
589 err = sg_alloc_append_table_from_pages(&append, pages, n_pages, offset,
590 size, max_segment, 0, gfp_mask);
591 if (err) {
592 sg_free_append_table(&append);
593 return err;
594 }
595 memcpy(sgt, &append.sgt, sizeof(*sgt));
596 WARN_ON(append.total_nents != sgt->orig_nents);
597 return 0;
598}
599EXPORT_SYMBOL(sg_alloc_table_from_pages_segment);
600
601#ifdef CONFIG_SGL_ALLOC
602
603/**
604 * sgl_alloc_order - allocate a scatterlist and its pages
605 * @length: Length in bytes of the scatterlist. Must be at least one
606 * @order: Second argument for alloc_pages()
607 * @chainable: Whether or not to allocate an extra element in the scatterlist
608 * for scatterlist chaining purposes
609 * @gfp: Memory allocation flags
610 * @nent_p: [out] Number of entries in the scatterlist that have pages
611 *
612 * Returns: A pointer to an initialized scatterlist or %NULL upon failure.
613 */
614struct scatterlist *sgl_alloc_order(unsigned long long length,
615 unsigned int order, bool chainable,
616 gfp_t gfp, unsigned int *nent_p)
617{
618 struct scatterlist *sgl, *sg;
619 struct page *page;
620 unsigned int nent, nalloc;
621 u32 elem_len;
622
623 nent = round_up(length, PAGE_SIZE << order) >> (PAGE_SHIFT + order);
624 /* Check for integer overflow */
625 if (length > (nent << (PAGE_SHIFT + order)))
626 return NULL;
627 nalloc = nent;
628 if (chainable) {
629 /* Check for integer overflow */
630 if (nalloc + 1 < nalloc)
631 return NULL;
632 nalloc++;
633 }
634 sgl = kmalloc_objs(struct scatterlist, nalloc, gfp & ~GFP_DMA);
635 if (!sgl)
636 return NULL;
637
638 sg_init_table(sgl, nalloc);
639 sg = sgl;
640 while (length) {
641 elem_len = min_t(u64, length, PAGE_SIZE << order);
642 page = alloc_pages(gfp, order);
643 if (!page) {
644 sgl_free_order(sgl, order);
645 return NULL;
646 }
647
648 sg_set_page(sg, page, elem_len, 0);
649 length -= elem_len;
650 sg = sg_next(sg);
651 }
652 WARN_ONCE(length, "length = %lld\n", length);
653 if (nent_p)
654 *nent_p = nent;
655 return sgl;
656}
657EXPORT_SYMBOL(sgl_alloc_order);
658
659/**
660 * sgl_alloc - allocate a scatterlist and its pages
661 * @length: Length in bytes of the scatterlist
662 * @gfp: Memory allocation flags
663 * @nent_p: [out] Number of entries in the scatterlist
664 *
665 * Returns: A pointer to an initialized scatterlist or %NULL upon failure.
666 */
667struct scatterlist *sgl_alloc(unsigned long long length, gfp_t gfp,
668 unsigned int *nent_p)
669{
670 return sgl_alloc_order(length, 0, false, gfp, nent_p);
671}
672EXPORT_SYMBOL(sgl_alloc);
673
674/**
675 * sgl_free_n_order - free a scatterlist and its pages
676 * @sgl: Scatterlist with one or more elements
677 * @nents: Maximum number of elements to free
678 * @order: Second argument for __free_pages()
679 *
680 * Notes:
681 * - If several scatterlists have been chained and each chain element is
682 * freed separately then it's essential to set nents correctly to avoid that a
683 * page would get freed twice.
684 * - All pages in a chained scatterlist can be freed at once by setting @nents
685 * to a high number.
686 */
687void sgl_free_n_order(struct scatterlist *sgl, int nents, int order)
688{
689 struct scatterlist *sg;
690 struct page *page;
691 int i;
692
693 for_each_sg(sgl, sg, nents, i) {
694 if (!sg)
695 break;
696 page = sg_page(sg);
697 if (page)
698 __free_pages(page, order);
699 }
700 kfree(sgl);
701}
702EXPORT_SYMBOL(sgl_free_n_order);
703
704/**
705 * sgl_free_order - free a scatterlist and its pages
706 * @sgl: Scatterlist with one or more elements
707 * @order: Second argument for __free_pages()
708 */
709void sgl_free_order(struct scatterlist *sgl, int order)
710{
711 sgl_free_n_order(sgl, INT_MAX, order);
712}
713EXPORT_SYMBOL(sgl_free_order);
714
715/**
716 * sgl_free - free a scatterlist and its pages
717 * @sgl: Scatterlist with one or more elements
718 */
719void sgl_free(struct scatterlist *sgl)
720{
721 sgl_free_order(sgl, 0);
722}
723EXPORT_SYMBOL(sgl_free);
724
725#endif /* CONFIG_SGL_ALLOC */
726
727void __sg_page_iter_start(struct sg_page_iter *piter,
728 struct scatterlist *sglist, unsigned int nents,
729 unsigned long pgoffset)
730{
731 piter->__pg_advance = 0;
732 piter->__nents = nents;
733
734 piter->sg = sglist;
735 piter->sg_pgoffset = pgoffset;
736}
737EXPORT_SYMBOL(__sg_page_iter_start);
738
739static int sg_page_count(struct scatterlist *sg)
740{
741 return PAGE_ALIGN(sg->offset + sg->length) >> PAGE_SHIFT;
742}
743
744bool __sg_page_iter_next(struct sg_page_iter *piter)
745{
746 if (!piter->__nents || !piter->sg)
747 return false;
748
749 piter->sg_pgoffset += piter->__pg_advance;
750 piter->__pg_advance = 1;
751
752 while (piter->sg_pgoffset >= sg_page_count(piter->sg)) {
753 piter->sg_pgoffset -= sg_page_count(piter->sg);
754 piter->sg = sg_next(piter->sg);
755 if (!--piter->__nents || !piter->sg)
756 return false;
757 }
758
759 return true;
760}
761EXPORT_SYMBOL(__sg_page_iter_next);
762
763static int sg_dma_page_count(struct scatterlist *sg)
764{
765 return PAGE_ALIGN(sg->offset + sg_dma_len(sg)) >> PAGE_SHIFT;
766}
767
768bool __sg_page_iter_dma_next(struct sg_dma_page_iter *dma_iter)
769{
770 struct sg_page_iter *piter = &dma_iter->base;
771
772 if (!piter->__nents || !piter->sg)
773 return false;
774
775 piter->sg_pgoffset += piter->__pg_advance;
776 piter->__pg_advance = 1;
777
778 while (piter->sg_pgoffset >= sg_dma_page_count(piter->sg)) {
779 piter->sg_pgoffset -= sg_dma_page_count(piter->sg);
780 piter->sg = sg_next(piter->sg);
781 if (!--piter->__nents || !piter->sg)
782 return false;
783 }
784
785 return true;
786}
787EXPORT_SYMBOL(__sg_page_iter_dma_next);
788
789/**
790 * sg_miter_start - start mapping iteration over a sg list
791 * @miter: sg mapping iter to be started
792 * @sgl: sg list to iterate over
793 * @nents: number of sg entries
794 * @flags: sg iterator flags
795 *
796 * Description:
797 * Starts mapping iterator @miter.
798 *
799 * Context:
800 * Don't care.
801 */
802void sg_miter_start(struct sg_mapping_iter *miter, struct scatterlist *sgl,
803 unsigned int nents, unsigned int flags)
804{
805 memset(miter, 0, sizeof(struct sg_mapping_iter));
806
807 __sg_page_iter_start(&miter->piter, sgl, nents, 0);
808 WARN_ON(!(flags & (SG_MITER_TO_SG | SG_MITER_FROM_SG)));
809 miter->__flags = flags;
810}
811EXPORT_SYMBOL(sg_miter_start);
812
813static bool sg_miter_get_next_page(struct sg_mapping_iter *miter)
814{
815 if (!miter->__remaining) {
816 struct scatterlist *sg;
817
818 if (!__sg_page_iter_next(&miter->piter))
819 return false;
820
821 sg = miter->piter.sg;
822
823 miter->__offset = miter->piter.sg_pgoffset ? 0 : sg->offset;
824 miter->piter.sg_pgoffset += miter->__offset >> PAGE_SHIFT;
825 miter->__offset &= PAGE_SIZE - 1;
826 miter->__remaining = sg->offset + sg->length -
827 (miter->piter.sg_pgoffset << PAGE_SHIFT) -
828 miter->__offset;
829 miter->__remaining = min_t(unsigned long, miter->__remaining,
830 PAGE_SIZE - miter->__offset);
831 }
832
833 return true;
834}
835
836/**
837 * sg_miter_skip - reposition mapping iterator
838 * @miter: sg mapping iter to be skipped
839 * @offset: number of bytes to plus the current location
840 *
841 * Description:
842 * Sets the offset of @miter to its current location plus @offset bytes.
843 * If mapping iterator @miter has been proceeded by sg_miter_next(), this
844 * stops @miter.
845 *
846 * Context:
847 * Don't care.
848 *
849 * Returns:
850 * true if @miter contains the valid mapping. false if end of sg
851 * list is reached.
852 */
853bool sg_miter_skip(struct sg_mapping_iter *miter, off_t offset)
854{
855 sg_miter_stop(miter);
856
857 while (offset) {
858 off_t consumed;
859
860 if (!sg_miter_get_next_page(miter))
861 return false;
862
863 consumed = min_t(off_t, offset, miter->__remaining);
864 miter->__offset += consumed;
865 miter->__remaining -= consumed;
866 offset -= consumed;
867 }
868
869 return true;
870}
871EXPORT_SYMBOL(sg_miter_skip);
872
873/**
874 * sg_miter_next - proceed mapping iterator to the next mapping
875 * @miter: sg mapping iter to proceed
876 *
877 * Description:
878 * Proceeds @miter to the next mapping. @miter should have been started
879 * using sg_miter_start(). On successful return, @miter->page,
880 * @miter->addr and @miter->length point to the current mapping.
881 *
882 * Context:
883 * May sleep if !SG_MITER_ATOMIC && !SG_MITER_LOCAL.
884 *
885 * Returns:
886 * true if @miter contains the next mapping. false if end of sg
887 * list is reached.
888 */
889bool sg_miter_next(struct sg_mapping_iter *miter)
890{
891 sg_miter_stop(miter);
892
893 /*
894 * Get to the next page if necessary.
895 * __remaining, __offset is adjusted by sg_miter_stop
896 */
897 if (!sg_miter_get_next_page(miter))
898 return false;
899
900 miter->page = sg_page_iter_page(&miter->piter);
901 miter->consumed = miter->length = miter->__remaining;
902
903 if (miter->__flags & SG_MITER_ATOMIC)
904 miter->addr = kmap_atomic(miter->page) + miter->__offset;
905 else if (miter->__flags & SG_MITER_LOCAL)
906 miter->addr = kmap_local_page(miter->page) + miter->__offset;
907 else
908 miter->addr = kmap(miter->page) + miter->__offset;
909
910 return true;
911}
912EXPORT_SYMBOL(sg_miter_next);
913
914/**
915 * sg_miter_stop - stop mapping iteration
916 * @miter: sg mapping iter to be stopped
917 *
918 * Description:
919 * Stops mapping iterator @miter. @miter should have been started
920 * using sg_miter_start(). A stopped iteration can be resumed by
921 * calling sg_miter_next() on it. This is useful when resources (kmap)
922 * need to be released during iteration.
923 *
924 * Context:
925 * Don't care otherwise.
926 */
927void sg_miter_stop(struct sg_mapping_iter *miter)
928{
929 WARN_ON(miter->consumed > miter->length);
930
931 /* drop resources from the last iteration */
932 if (miter->addr) {
933 miter->__offset += miter->consumed;
934 miter->__remaining -= miter->consumed;
935
936 if (miter->__flags & SG_MITER_TO_SG)
937 flush_dcache_page(miter->page);
938
939 if (miter->__flags & SG_MITER_ATOMIC) {
940 WARN_ON_ONCE(!pagefault_disabled());
941 kunmap_atomic(miter->addr);
942 } else if (miter->__flags & SG_MITER_LOCAL)
943 kunmap_local(miter->addr);
944 else
945 kunmap(miter->page);
946
947 miter->page = NULL;
948 miter->addr = NULL;
949 miter->length = 0;
950 miter->consumed = 0;
951 }
952}
953EXPORT_SYMBOL(sg_miter_stop);
954
955/**
956 * sg_copy_buffer - Copy data between a linear buffer and an SG list
957 * @sgl: The SG list
958 * @nents: Number of SG entries
959 * @buf: Where to copy from
960 * @buflen: The number of bytes to copy
961 * @skip: Number of bytes to skip before copying
962 * @to_buffer: transfer direction (true == from an sg list to a
963 * buffer, false == from a buffer to an sg list)
964 *
965 * Returns the number of copied bytes.
966 *
967 **/
968size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf,
969 size_t buflen, off_t skip, bool to_buffer)
970{
971 unsigned int offset = 0;
972 struct sg_mapping_iter miter;
973 unsigned int sg_flags = SG_MITER_LOCAL;
974
975 if (to_buffer)
976 sg_flags |= SG_MITER_FROM_SG;
977 else
978 sg_flags |= SG_MITER_TO_SG;
979
980 sg_miter_start(&miter, sgl, nents, sg_flags);
981
982 if (!sg_miter_skip(&miter, skip))
983 return 0;
984
985 while ((offset < buflen) && sg_miter_next(&miter)) {
986 unsigned int len;
987
988 len = min(miter.length, buflen - offset);
989
990 if (to_buffer)
991 memcpy(buf + offset, miter.addr, len);
992 else
993 memcpy(miter.addr, buf + offset, len);
994
995 offset += len;
996 }
997
998 sg_miter_stop(&miter);
999
1000 return offset;
1001}
1002EXPORT_SYMBOL(sg_copy_buffer);
1003
1004/**
1005 * sg_copy_from_buffer - Copy from a linear buffer to an SG list
1006 * @sgl: The SG list
1007 * @nents: Number of SG entries
1008 * @buf: Where to copy from
1009 * @buflen: The number of bytes to copy
1010 *
1011 * Returns the number of copied bytes.
1012 *
1013 **/
1014size_t sg_copy_from_buffer(struct scatterlist *sgl, unsigned int nents,
1015 const void *buf, size_t buflen)
1016{
1017 return sg_copy_buffer(sgl, nents, (void *)buf, buflen, 0, false);
1018}
1019EXPORT_SYMBOL(sg_copy_from_buffer);
1020
1021/**
1022 * sg_copy_to_buffer - Copy from an SG list to a linear buffer
1023 * @sgl: The SG list
1024 * @nents: Number of SG entries
1025 * @buf: Where to copy to
1026 * @buflen: The number of bytes to copy
1027 *
1028 * Returns the number of copied bytes.
1029 *
1030 **/
1031size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
1032 void *buf, size_t buflen)
1033{
1034 return sg_copy_buffer(sgl, nents, buf, buflen, 0, true);
1035}
1036EXPORT_SYMBOL(sg_copy_to_buffer);
1037
1038/**
1039 * sg_pcopy_from_buffer - Copy from a linear buffer to an SG list
1040 * @sgl: The SG list
1041 * @nents: Number of SG entries
1042 * @buf: Where to copy from
1043 * @buflen: The number of bytes to copy
1044 * @skip: Number of bytes to skip before copying
1045 *
1046 * Returns the number of copied bytes.
1047 *
1048 **/
1049size_t sg_pcopy_from_buffer(struct scatterlist *sgl, unsigned int nents,
1050 const void *buf, size_t buflen, off_t skip)
1051{
1052 return sg_copy_buffer(sgl, nents, (void *)buf, buflen, skip, false);
1053}
1054EXPORT_SYMBOL(sg_pcopy_from_buffer);
1055
1056/**
1057 * sg_pcopy_to_buffer - Copy from an SG list to a linear buffer
1058 * @sgl: The SG list
1059 * @nents: Number of SG entries
1060 * @buf: Where to copy to
1061 * @buflen: The number of bytes to copy
1062 * @skip: Number of bytes to skip before copying
1063 *
1064 * Returns the number of copied bytes.
1065 *
1066 **/
1067size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents,
1068 void *buf, size_t buflen, off_t skip)
1069{
1070 return sg_copy_buffer(sgl, nents, buf, buflen, skip, true);
1071}
1072EXPORT_SYMBOL(sg_pcopy_to_buffer);
1073
1074/**
1075 * sg_zero_buffer - Zero-out a part of a SG list
1076 * @sgl: The SG list
1077 * @nents: Number of SG entries
1078 * @buflen: The number of bytes to zero out
1079 * @skip: Number of bytes to skip before zeroing
1080 *
1081 * Returns the number of bytes zeroed.
1082 **/
1083size_t sg_zero_buffer(struct scatterlist *sgl, unsigned int nents,
1084 size_t buflen, off_t skip)
1085{
1086 unsigned int offset = 0;
1087 struct sg_mapping_iter miter;
1088 unsigned int sg_flags = SG_MITER_LOCAL | SG_MITER_TO_SG;
1089
1090 sg_miter_start(&miter, sgl, nents, sg_flags);
1091
1092 if (!sg_miter_skip(&miter, skip))
1093 return false;
1094
1095 while (offset < buflen && sg_miter_next(&miter)) {
1096 unsigned int len;
1097
1098 len = min(miter.length, buflen - offset);
1099 memset(miter.addr, 0, len);
1100
1101 offset += len;
1102 }
1103
1104 sg_miter_stop(&miter);
1105 return offset;
1106}
1107EXPORT_SYMBOL(sg_zero_buffer);
1108
1109/*
1110 * Extract and pin a list of up to sg_max pages from UBUF- or IOVEC-class
1111 * iterators, and add them to the scatterlist.
1112 */
1113static ssize_t extract_user_to_sg(struct iov_iter *iter,
1114 ssize_t maxsize,
1115 struct sg_table *sgtable,
1116 unsigned int sg_max,
1117 iov_iter_extraction_t extraction_flags)
1118{
1119 struct scatterlist *sg = sgtable->sgl + sgtable->nents;
1120 struct page **pages;
1121 unsigned int npages;
1122 ssize_t ret = 0, res;
1123 size_t len, off;
1124
1125 /* We decant the page list into the tail of the scatterlist */
1126 pages = (void *)sg + array_size(sg_max, sizeof(struct scatterlist));
1127 pages -= sg_max;
1128
1129 do {
1130 res = iov_iter_extract_pages(iter, &pages, maxsize, sg_max,
1131 extraction_flags, &off);
1132 if (res <= 0)
1133 goto failed;
1134
1135 len = res;
1136 maxsize -= len;
1137 ret += len;
1138 npages = DIV_ROUND_UP(off + len, PAGE_SIZE);
1139 sg_max -= npages;
1140
1141 for (; npages > 0; npages--) {
1142 struct page *page = *pages;
1143 size_t seg = min_t(size_t, PAGE_SIZE - off, len);
1144
1145 *pages++ = NULL;
1146 sg_set_page(sg, page, seg, off);
1147 sgtable->nents++;
1148 sg++;
1149 len -= seg;
1150 off = 0;
1151 }
1152 } while (maxsize > 0 && sg_max > 0);
1153
1154 return ret;
1155
1156failed:
1157 while (sgtable->nents > sgtable->orig_nents)
1158 unpin_user_page(sg_page(&sgtable->sgl[--sgtable->nents]));
1159 return res;
1160}
1161
1162/*
1163 * Extract up to sg_max pages from a BVEC-type iterator and add them to the
1164 * scatterlist. The pages are not pinned.
1165 */
1166static ssize_t extract_bvec_to_sg(struct iov_iter *iter,
1167 ssize_t maxsize,
1168 struct sg_table *sgtable,
1169 unsigned int sg_max,
1170 iov_iter_extraction_t extraction_flags)
1171{
1172 const struct bio_vec *bv = iter->bvec;
1173 struct scatterlist *sg = sgtable->sgl + sgtable->nents;
1174 unsigned long start = iter->iov_offset;
1175 unsigned int i;
1176 ssize_t ret = 0;
1177
1178 for (i = 0; i < iter->nr_segs; i++) {
1179 size_t off, len;
1180
1181 len = bv[i].bv_len;
1182 if (start >= len) {
1183 start -= len;
1184 continue;
1185 }
1186
1187 len = min_t(size_t, maxsize, len - start);
1188 off = bv[i].bv_offset + start;
1189
1190 sg_set_page(sg, bv[i].bv_page, len, off);
1191 sgtable->nents++;
1192 sg++;
1193 sg_max--;
1194
1195 ret += len;
1196 maxsize -= len;
1197 if (maxsize <= 0 || sg_max == 0)
1198 break;
1199 start = 0;
1200 }
1201
1202 if (ret > 0)
1203 iov_iter_advance(iter, ret);
1204 return ret;
1205}
1206
1207/*
1208 * Extract up to sg_max pages from a KVEC-type iterator and add them to the
1209 * scatterlist. This can deal with vmalloc'd buffers as well as kmalloc'd or
1210 * static buffers. The pages are not pinned.
1211 */
1212static ssize_t extract_kvec_to_sg(struct iov_iter *iter,
1213 ssize_t maxsize,
1214 struct sg_table *sgtable,
1215 unsigned int sg_max,
1216 iov_iter_extraction_t extraction_flags)
1217{
1218 const struct kvec *kv = iter->kvec;
1219 struct scatterlist *sg = sgtable->sgl + sgtable->nents;
1220 unsigned long start = iter->iov_offset;
1221 unsigned int i;
1222 ssize_t ret = 0;
1223
1224 for (i = 0; i < iter->nr_segs; i++) {
1225 struct page *page;
1226 unsigned long kaddr;
1227 size_t off, len, seg;
1228
1229 len = kv[i].iov_len;
1230 if (start >= len) {
1231 start -= len;
1232 continue;
1233 }
1234
1235 kaddr = (unsigned long)kv[i].iov_base + start;
1236 off = kaddr & ~PAGE_MASK;
1237 len = min_t(size_t, maxsize, len - start);
1238 kaddr &= PAGE_MASK;
1239
1240 maxsize -= len;
1241 ret += len;
1242 do {
1243 seg = min_t(size_t, len, PAGE_SIZE - off);
1244 if (is_vmalloc_or_module_addr((void *)kaddr))
1245 page = vmalloc_to_page((void *)kaddr);
1246 else
1247 page = virt_to_page((void *)kaddr);
1248
1249 sg_set_page(sg, page, seg, off);
1250 sgtable->nents++;
1251 sg++;
1252 sg_max--;
1253
1254 len -= seg;
1255 kaddr += PAGE_SIZE;
1256 off = 0;
1257 } while (len > 0 && sg_max > 0);
1258 ret -= len;
1259
1260 if (maxsize <= 0 || sg_max == 0)
1261 break;
1262 start = 0;
1263 }
1264
1265 if (ret > 0)
1266 iov_iter_advance(iter, ret);
1267 return ret;
1268}
1269
1270/*
1271 * Extract up to sg_max folios from an FOLIOQ-type iterator and add them to
1272 * the scatterlist. The pages are not pinned.
1273 */
1274static ssize_t extract_folioq_to_sg(struct iov_iter *iter,
1275 ssize_t maxsize,
1276 struct sg_table *sgtable,
1277 unsigned int sg_max,
1278 iov_iter_extraction_t extraction_flags)
1279{
1280 const struct folio_queue *folioq = iter->folioq;
1281 struct scatterlist *sg = sgtable->sgl + sgtable->nents;
1282 unsigned int slot = iter->folioq_slot;
1283 ssize_t ret = 0;
1284 size_t offset = iter->iov_offset;
1285
1286 BUG_ON(!folioq);
1287
1288 if (slot >= folioq_nr_slots(folioq)) {
1289 folioq = folioq->next;
1290 if (WARN_ON_ONCE(!folioq))
1291 return 0;
1292 slot = 0;
1293 }
1294
1295 do {
1296 struct folio *folio = folioq_folio(folioq, slot);
1297 size_t fsize = folioq_folio_size(folioq, slot);
1298
1299 if (offset < fsize) {
1300 size_t part = umin(maxsize - ret, fsize - offset);
1301
1302 sg_set_page(sg, folio_page(folio, 0), part, offset);
1303 sgtable->nents++;
1304 sg++;
1305 sg_max--;
1306 offset += part;
1307 ret += part;
1308 }
1309
1310 if (offset >= fsize) {
1311 offset = 0;
1312 slot++;
1313 if (slot >= folioq_nr_slots(folioq)) {
1314 if (!folioq->next) {
1315 WARN_ON_ONCE(ret < iter->count);
1316 break;
1317 }
1318 folioq = folioq->next;
1319 slot = 0;
1320 }
1321 }
1322 } while (sg_max > 0 && ret < maxsize);
1323
1324 iter->folioq = folioq;
1325 iter->folioq_slot = slot;
1326 iter->iov_offset = offset;
1327 iter->count -= ret;
1328 return ret;
1329}
1330
1331/*
1332 * Extract up to sg_max folios from an XARRAY-type iterator and add them to
1333 * the scatterlist. The pages are not pinned.
1334 */
1335static ssize_t extract_xarray_to_sg(struct iov_iter *iter,
1336 ssize_t maxsize,
1337 struct sg_table *sgtable,
1338 unsigned int sg_max,
1339 iov_iter_extraction_t extraction_flags)
1340{
1341 struct scatterlist *sg = sgtable->sgl + sgtable->nents;
1342 struct xarray *xa = iter->xarray;
1343 struct folio *folio;
1344 loff_t start = iter->xarray_start + iter->iov_offset;
1345 pgoff_t index = start / PAGE_SIZE;
1346 ssize_t ret = 0;
1347 size_t offset, len;
1348 XA_STATE(xas, xa, index);
1349
1350 rcu_read_lock();
1351
1352 xas_for_each(&xas, folio, ULONG_MAX) {
1353 if (xas_retry(&xas, folio))
1354 continue;
1355 if (WARN_ON(xa_is_value(folio)))
1356 break;
1357 if (WARN_ON(folio_test_hugetlb(folio)))
1358 break;
1359
1360 offset = offset_in_folio(folio, start);
1361 len = min_t(size_t, maxsize, folio_size(folio) - offset);
1362
1363 sg_set_page(sg, folio_page(folio, 0), len, offset);
1364 sgtable->nents++;
1365 sg++;
1366 sg_max--;
1367
1368 maxsize -= len;
1369 start += len;
1370 ret += len;
1371 if (maxsize <= 0 || sg_max == 0)
1372 break;
1373 }
1374
1375 rcu_read_unlock();
1376 if (ret > 0)
1377 iov_iter_advance(iter, ret);
1378 return ret;
1379}
1380
1381/**
1382 * extract_iter_to_sg - Extract pages from an iterator and add to an sglist
1383 * @iter: The iterator to extract from
1384 * @maxsize: The amount of iterator to copy
1385 * @sgtable: The scatterlist table to fill in
1386 * @sg_max: Maximum number of elements in @sgtable that may be filled
1387 * @extraction_flags: Flags to qualify the request
1388 *
1389 * Extract the page fragments from the given amount of the source iterator and
1390 * add them to a scatterlist that refers to all of those bits, to a maximum
1391 * addition of @sg_max elements.
1392 *
1393 * The pages referred to by UBUF- and IOVEC-type iterators are extracted and
1394 * pinned; BVEC-, KVEC-, FOLIOQ- and XARRAY-type are extracted but aren't
1395 * pinned; DISCARD-type is not supported.
1396 *
1397 * No end mark is placed on the scatterlist; that's left to the caller.
1398 *
1399 * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA
1400 * be allowed on the pages extracted.
1401 *
1402 * If successful, @sgtable->nents is updated to include the number of elements
1403 * added and the number of bytes added is returned. @sgtable->orig_nents is
1404 * left unaltered.
1405 *
1406 * The iov_iter_extract_mode() function should be used to query how cleanup
1407 * should be performed.
1408 */
1409ssize_t extract_iter_to_sg(struct iov_iter *iter, size_t maxsize,
1410 struct sg_table *sgtable, unsigned int sg_max,
1411 iov_iter_extraction_t extraction_flags)
1412{
1413 if (maxsize == 0 || sg_max == 0)
1414 return 0;
1415
1416 switch (iov_iter_type(iter)) {
1417 case ITER_UBUF:
1418 case ITER_IOVEC:
1419 return extract_user_to_sg(iter, maxsize, sgtable, sg_max,
1420 extraction_flags);
1421 case ITER_BVEC:
1422 return extract_bvec_to_sg(iter, maxsize, sgtable, sg_max,
1423 extraction_flags);
1424 case ITER_KVEC:
1425 return extract_kvec_to_sg(iter, maxsize, sgtable, sg_max,
1426 extraction_flags);
1427 case ITER_FOLIOQ:
1428 return extract_folioq_to_sg(iter, maxsize, sgtable, sg_max,
1429 extraction_flags);
1430 case ITER_XARRAY:
1431 return extract_xarray_to_sg(iter, maxsize, sgtable, sg_max,
1432 extraction_flags);
1433 default:
1434 pr_err("%s(%u) unsupported\n", __func__, iov_iter_type(iter));
1435 WARN_ON_ONCE(1);
1436 return -EIO;
1437 }
1438}
1439EXPORT_SYMBOL_GPL(extract_iter_to_sg);