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