1
2
3
4
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/scatterlist.h>
10
11#include <trace/events/block.h>
12
13#include "blk.h"
14
15static struct bio *blk_bio_discard_split(struct request_queue *q,
16 struct bio *bio,
17 struct bio_set *bs,
18 unsigned *nsegs)
19{
20 unsigned int max_discard_sectors, granularity;
21 int alignment;
22 sector_t tmp;
23 unsigned split_sectors;
24
25 *nsegs = 1;
26
27
28 granularity = max(q->limits.discard_granularity >> 9, 1U);
29
30 max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9);
31 max_discard_sectors -= max_discard_sectors % granularity;
32
33 if (unlikely(!max_discard_sectors)) {
34
35 return NULL;
36 }
37
38 if (bio_sectors(bio) <= max_discard_sectors)
39 return NULL;
40
41 split_sectors = max_discard_sectors;
42
43
44
45
46
47 alignment = (q->limits.discard_alignment >> 9) % granularity;
48
49 tmp = bio->bi_iter.bi_sector + split_sectors - alignment;
50 tmp = sector_div(tmp, granularity);
51
52 if (split_sectors > tmp)
53 split_sectors -= tmp;
54
55 return bio_split(bio, split_sectors, GFP_NOIO, bs);
56}
57
58static struct bio *blk_bio_write_zeroes_split(struct request_queue *q,
59 struct bio *bio, struct bio_set *bs, unsigned *nsegs)
60{
61 *nsegs = 1;
62
63 if (!q->limits.max_write_zeroes_sectors)
64 return NULL;
65
66 if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors)
67 return NULL;
68
69 return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs);
70}
71
72static struct bio *blk_bio_write_same_split(struct request_queue *q,
73 struct bio *bio,
74 struct bio_set *bs,
75 unsigned *nsegs)
76{
77 *nsegs = 1;
78
79 if (!q->limits.max_write_same_sectors)
80 return NULL;
81
82 if (bio_sectors(bio) <= q->limits.max_write_same_sectors)
83 return NULL;
84
85 return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs);
86}
87
88static inline unsigned get_max_io_size(struct request_queue *q,
89 struct bio *bio)
90{
91 unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
92 unsigned mask = queue_logical_block_size(q) - 1;
93
94
95 sectors &= ~(mask >> 9);
96
97 return sectors;
98}
99
100static struct bio *blk_bio_segment_split(struct request_queue *q,
101 struct bio *bio,
102 struct bio_set *bs,
103 unsigned *segs)
104{
105 struct bio_vec bv, bvprv, *bvprvp = NULL;
106 struct bvec_iter iter;
107 unsigned seg_size = 0, nsegs = 0, sectors = 0;
108 unsigned front_seg_size = bio->bi_seg_front_size;
109 bool do_split = true;
110 struct bio *new = NULL;
111 const unsigned max_sectors = get_max_io_size(q, bio);
112
113 bio_for_each_segment(bv, bio, iter) {
114
115
116
117
118 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
119 goto split;
120
121 if (sectors + (bv.bv_len >> 9) > max_sectors) {
122
123
124
125
126 if (nsegs < queue_max_segments(q) &&
127 sectors < max_sectors) {
128 nsegs++;
129 sectors = max_sectors;
130 }
131 if (sectors)
132 goto split;
133
134 }
135
136 if (bvprvp && blk_queue_cluster(q)) {
137 if (seg_size + bv.bv_len > queue_max_segment_size(q))
138 goto new_segment;
139 if (!BIOVEC_PHYS_MERGEABLE(bvprvp, &bv))
140 goto new_segment;
141 if (!BIOVEC_SEG_BOUNDARY(q, bvprvp, &bv))
142 goto new_segment;
143
144 seg_size += bv.bv_len;
145 bvprv = bv;
146 bvprvp = &bvprv;
147 sectors += bv.bv_len >> 9;
148
149 if (nsegs == 1 && seg_size > front_seg_size)
150 front_seg_size = seg_size;
151 continue;
152 }
153new_segment:
154 if (nsegs == queue_max_segments(q))
155 goto split;
156
157 nsegs++;
158 bvprv = bv;
159 bvprvp = &bvprv;
160 seg_size = bv.bv_len;
161 sectors += bv.bv_len >> 9;
162
163 if (nsegs == 1 && seg_size > front_seg_size)
164 front_seg_size = seg_size;
165 }
166
167 do_split = false;
168split:
169 *segs = nsegs;
170
171 if (do_split) {
172 new = bio_split(bio, sectors, GFP_NOIO, bs);
173 if (new)
174 bio = new;
175 }
176
177 bio->bi_seg_front_size = front_seg_size;
178 if (seg_size > bio->bi_seg_back_size)
179 bio->bi_seg_back_size = seg_size;
180
181 return do_split ? new : NULL;
182}
183
184void blk_queue_split(struct request_queue *q, struct bio **bio)
185{
186 struct bio *split, *res;
187 unsigned nsegs;
188
189 switch (bio_op(*bio)) {
190 case REQ_OP_DISCARD:
191 case REQ_OP_SECURE_ERASE:
192 split = blk_bio_discard_split(q, *bio, q->bio_split, &nsegs);
193 break;
194 case REQ_OP_WRITE_ZEROES:
195 split = blk_bio_write_zeroes_split(q, *bio, q->bio_split, &nsegs);
196 break;
197 case REQ_OP_WRITE_SAME:
198 split = blk_bio_write_same_split(q, *bio, q->bio_split, &nsegs);
199 break;
200 default:
201 split = blk_bio_segment_split(q, *bio, q->bio_split, &nsegs);
202 break;
203 }
204
205
206 res = split ? split : *bio;
207 res->bi_phys_segments = nsegs;
208 bio_set_flag(res, BIO_SEG_VALID);
209
210 if (split) {
211
212 split->bi_opf |= REQ_NOMERGE;
213
214 bio_chain(split, *bio);
215 trace_block_split(q, split, (*bio)->bi_iter.bi_sector);
216 generic_make_request(*bio);
217 *bio = split;
218 }
219}
220EXPORT_SYMBOL(blk_queue_split);
221
222static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
223 struct bio *bio,
224 bool no_sg_merge)
225{
226 struct bio_vec bv, bvprv = { NULL };
227 int cluster, prev = 0;
228 unsigned int seg_size, nr_phys_segs;
229 struct bio *fbio, *bbio;
230 struct bvec_iter iter;
231
232 if (!bio)
233 return 0;
234
235 switch (bio_op(bio)) {
236 case REQ_OP_DISCARD:
237 case REQ_OP_SECURE_ERASE:
238 case REQ_OP_WRITE_ZEROES:
239 return 0;
240 case REQ_OP_WRITE_SAME:
241 return 1;
242 }
243
244 fbio = bio;
245 cluster = blk_queue_cluster(q);
246 seg_size = 0;
247 nr_phys_segs = 0;
248 for_each_bio(bio) {
249 bio_for_each_segment(bv, bio, iter) {
250
251
252
253
254 if (no_sg_merge)
255 goto new_segment;
256
257 if (prev && cluster) {
258 if (seg_size + bv.bv_len
259 > queue_max_segment_size(q))
260 goto new_segment;
261 if (!BIOVEC_PHYS_MERGEABLE(&bvprv, &bv))
262 goto new_segment;
263 if (!BIOVEC_SEG_BOUNDARY(q, &bvprv, &bv))
264 goto new_segment;
265
266 seg_size += bv.bv_len;
267 bvprv = bv;
268 continue;
269 }
270new_segment:
271 if (nr_phys_segs == 1 && seg_size >
272 fbio->bi_seg_front_size)
273 fbio->bi_seg_front_size = seg_size;
274
275 nr_phys_segs++;
276 bvprv = bv;
277 prev = 1;
278 seg_size = bv.bv_len;
279 }
280 bbio = bio;
281 }
282
283 if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
284 fbio->bi_seg_front_size = seg_size;
285 if (seg_size > bbio->bi_seg_back_size)
286 bbio->bi_seg_back_size = seg_size;
287
288 return nr_phys_segs;
289}
290
291void blk_recalc_rq_segments(struct request *rq)
292{
293 bool no_sg_merge = !!test_bit(QUEUE_FLAG_NO_SG_MERGE,
294 &rq->q->queue_flags);
295
296 rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio,
297 no_sg_merge);
298}
299
300void blk_recount_segments(struct request_queue *q, struct bio *bio)
301{
302 unsigned short seg_cnt;
303
304
305 if (bio_flagged(bio, BIO_CLONED))
306 seg_cnt = bio_segments(bio);
307 else
308 seg_cnt = bio->bi_vcnt;
309
310 if (test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags) &&
311 (seg_cnt < queue_max_segments(q)))
312 bio->bi_phys_segments = seg_cnt;
313 else {
314 struct bio *nxt = bio->bi_next;
315
316 bio->bi_next = NULL;
317 bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio, false);
318 bio->bi_next = nxt;
319 }
320
321 bio_set_flag(bio, BIO_SEG_VALID);
322}
323EXPORT_SYMBOL(blk_recount_segments);
324
325static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
326 struct bio *nxt)
327{
328 struct bio_vec end_bv = { NULL }, nxt_bv;
329
330 if (!blk_queue_cluster(q))
331 return 0;
332
333 if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
334 queue_max_segment_size(q))
335 return 0;
336
337 if (!bio_has_data(bio))
338 return 1;
339
340 bio_get_last_bvec(bio, &end_bv);
341 bio_get_first_bvec(nxt, &nxt_bv);
342
343 if (!BIOVEC_PHYS_MERGEABLE(&end_bv, &nxt_bv))
344 return 0;
345
346
347
348
349
350 if (BIOVEC_SEG_BOUNDARY(q, &end_bv, &nxt_bv))
351 return 1;
352
353 return 0;
354}
355
356static inline void
357__blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
358 struct scatterlist *sglist, struct bio_vec *bvprv,
359 struct scatterlist **sg, int *nsegs, int *cluster)
360{
361
362 int nbytes = bvec->bv_len;
363
364 if (*sg && *cluster) {
365 if ((*sg)->length + nbytes > queue_max_segment_size(q))
366 goto new_segment;
367
368 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
369 goto new_segment;
370 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
371 goto new_segment;
372
373 (*sg)->length += nbytes;
374 } else {
375new_segment:
376 if (!*sg)
377 *sg = sglist;
378 else {
379
380
381
382
383
384
385
386
387
388
389 sg_unmark_end(*sg);
390 *sg = sg_next(*sg);
391 }
392
393 sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
394 (*nsegs)++;
395 }
396 *bvprv = *bvec;
397}
398
399static inline int __blk_bvec_map_sg(struct request_queue *q, struct bio_vec bv,
400 struct scatterlist *sglist, struct scatterlist **sg)
401{
402 *sg = sglist;
403 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
404 return 1;
405}
406
407static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
408 struct scatterlist *sglist,
409 struct scatterlist **sg)
410{
411 struct bio_vec bvec, bvprv = { NULL };
412 struct bvec_iter iter;
413 int cluster = blk_queue_cluster(q), nsegs = 0;
414
415 for_each_bio(bio)
416 bio_for_each_segment(bvec, bio, iter)
417 __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg,
418 &nsegs, &cluster);
419
420 return nsegs;
421}
422
423
424
425
426
427int blk_rq_map_sg(struct request_queue *q, struct request *rq,
428 struct scatterlist *sglist)
429{
430 struct scatterlist *sg = NULL;
431 int nsegs = 0;
432
433 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
434 nsegs = __blk_bvec_map_sg(q, rq->special_vec, sglist, &sg);
435 else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME)
436 nsegs = __blk_bvec_map_sg(q, bio_iovec(rq->bio), sglist, &sg);
437 else if (rq->bio)
438 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg);
439
440 if (unlikely(rq->rq_flags & RQF_COPY_USER) &&
441 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
442 unsigned int pad_len =
443 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
444
445 sg->length += pad_len;
446 rq->extra_len += pad_len;
447 }
448
449 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
450 if (op_is_write(req_op(rq)))
451 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
452
453 sg_unmark_end(sg);
454 sg = sg_next(sg);
455 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
456 q->dma_drain_size,
457 ((unsigned long)q->dma_drain_buffer) &
458 (PAGE_SIZE - 1));
459 nsegs++;
460 rq->extra_len += q->dma_drain_size;
461 }
462
463 if (sg)
464 sg_mark_end(sg);
465
466
467
468
469
470 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
471
472 return nsegs;
473}
474EXPORT_SYMBOL(blk_rq_map_sg);
475
476static inline int ll_new_hw_segment(struct request_queue *q,
477 struct request *req,
478 struct bio *bio)
479{
480 int nr_phys_segs = bio_phys_segments(q, bio);
481
482 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
483 goto no_merge;
484
485 if (blk_integrity_merge_bio(q, req, bio) == false)
486 goto no_merge;
487
488
489
490
491
492 req->nr_phys_segments += nr_phys_segs;
493 return 1;
494
495no_merge:
496 req_set_nomerge(q, req);
497 return 0;
498}
499
500int ll_back_merge_fn(struct request_queue *q, struct request *req,
501 struct bio *bio)
502{
503 if (req_gap_back_merge(req, bio))
504 return 0;
505 if (blk_integrity_rq(req) &&
506 integrity_req_gap_back_merge(req, bio))
507 return 0;
508 if (blk_rq_sectors(req) + bio_sectors(bio) >
509 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
510 req_set_nomerge(q, req);
511 return 0;
512 }
513 if (!bio_flagged(req->biotail, BIO_SEG_VALID))
514 blk_recount_segments(q, req->biotail);
515 if (!bio_flagged(bio, BIO_SEG_VALID))
516 blk_recount_segments(q, bio);
517
518 return ll_new_hw_segment(q, req, bio);
519}
520
521int ll_front_merge_fn(struct request_queue *q, struct request *req,
522 struct bio *bio)
523{
524
525 if (req_gap_front_merge(req, bio))
526 return 0;
527 if (blk_integrity_rq(req) &&
528 integrity_req_gap_front_merge(req, bio))
529 return 0;
530 if (blk_rq_sectors(req) + bio_sectors(bio) >
531 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
532 req_set_nomerge(q, req);
533 return 0;
534 }
535 if (!bio_flagged(bio, BIO_SEG_VALID))
536 blk_recount_segments(q, bio);
537 if (!bio_flagged(req->bio, BIO_SEG_VALID))
538 blk_recount_segments(q, req->bio);
539
540 return ll_new_hw_segment(q, req, bio);
541}
542
543
544
545
546
547static bool req_no_special_merge(struct request *req)
548{
549 struct request_queue *q = req->q;
550
551 return !q->mq_ops && req->special;
552}
553
554static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
555 struct request *next)
556{
557 int total_phys_segments;
558 unsigned int seg_size =
559 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
560
561
562
563
564
565 if (req_no_special_merge(req) || req_no_special_merge(next))
566 return 0;
567
568 if (req_gap_back_merge(req, next->bio))
569 return 0;
570
571
572
573
574 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
575 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
576 return 0;
577
578 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
579 if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
580 if (req->nr_phys_segments == 1)
581 req->bio->bi_seg_front_size = seg_size;
582 if (next->nr_phys_segments == 1)
583 next->biotail->bi_seg_back_size = seg_size;
584 total_phys_segments--;
585 }
586
587 if (total_phys_segments > queue_max_segments(q))
588 return 0;
589
590 if (blk_integrity_merge_rq(q, req, next) == false)
591 return 0;
592
593
594 req->nr_phys_segments = total_phys_segments;
595 return 1;
596}
597
598
599
600
601
602
603
604
605
606
607void blk_rq_set_mixed_merge(struct request *rq)
608{
609 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
610 struct bio *bio;
611
612 if (rq->rq_flags & RQF_MIXED_MERGE)
613 return;
614
615
616
617
618
619
620 for (bio = rq->bio; bio; bio = bio->bi_next) {
621 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
622 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
623 bio->bi_opf |= ff;
624 }
625 rq->rq_flags |= RQF_MIXED_MERGE;
626}
627
628static void blk_account_io_merge(struct request *req)
629{
630 if (blk_do_io_stat(req)) {
631 struct hd_struct *part;
632 int cpu;
633
634 cpu = part_stat_lock();
635 part = req->part;
636
637 part_round_stats(req->q, cpu, part);
638 part_dec_in_flight(req->q, part, rq_data_dir(req));
639
640 hd_struct_put(part);
641 part_stat_unlock();
642 }
643}
644
645
646
647
648
649static struct request *attempt_merge(struct request_queue *q,
650 struct request *req, struct request *next)
651{
652 if (!q->mq_ops)
653 lockdep_assert_held(q->queue_lock);
654
655 if (!rq_mergeable(req) || !rq_mergeable(next))
656 return NULL;
657
658 if (req_op(req) != req_op(next))
659 return NULL;
660
661
662
663
664 if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
665 return NULL;
666
667 if (rq_data_dir(req) != rq_data_dir(next)
668 || req->rq_disk != next->rq_disk
669 || req_no_special_merge(next))
670 return NULL;
671
672 if (req_op(req) == REQ_OP_WRITE_SAME &&
673 !blk_write_same_mergeable(req->bio, next->bio))
674 return NULL;
675
676
677
678
679
680 if (req->write_hint != next->write_hint)
681 return NULL;
682
683
684
685
686
687
688
689 if (!ll_merge_requests_fn(q, req, next))
690 return NULL;
691
692
693
694
695
696
697
698 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
699 (req->cmd_flags & REQ_FAILFAST_MASK) !=
700 (next->cmd_flags & REQ_FAILFAST_MASK)) {
701 blk_rq_set_mixed_merge(req);
702 blk_rq_set_mixed_merge(next);
703 }
704
705
706
707
708
709
710
711 if (time_after(req->start_time, next->start_time))
712 req->start_time = next->start_time;
713
714 req->biotail->bi_next = next->bio;
715 req->biotail = next->biotail;
716
717 req->__data_len += blk_rq_bytes(next);
718
719 elv_merge_requests(q, req, next);
720
721
722
723
724 blk_account_io_merge(next);
725
726 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
727 if (blk_rq_cpu_valid(next))
728 req->cpu = next->cpu;
729
730
731
732
733
734 next->bio = NULL;
735 return next;
736}
737
738struct request *attempt_back_merge(struct request_queue *q, struct request *rq)
739{
740 struct request *next = elv_latter_request(q, rq);
741
742 if (next)
743 return attempt_merge(q, rq, next);
744
745 return NULL;
746}
747
748struct request *attempt_front_merge(struct request_queue *q, struct request *rq)
749{
750 struct request *prev = elv_former_request(q, rq);
751
752 if (prev)
753 return attempt_merge(q, prev, rq);
754
755 return NULL;
756}
757
758int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
759 struct request *next)
760{
761 struct elevator_queue *e = q->elevator;
762 struct request *free;
763
764 if (!e->uses_mq && e->type->ops.sq.elevator_allow_rq_merge_fn)
765 if (!e->type->ops.sq.elevator_allow_rq_merge_fn(q, rq, next))
766 return 0;
767
768 free = attempt_merge(q, rq, next);
769 if (free) {
770 __blk_put_request(q, free);
771 return 1;
772 }
773
774 return 0;
775}
776
777bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
778{
779 if (!rq_mergeable(rq) || !bio_mergeable(bio))
780 return false;
781
782 if (req_op(rq) != bio_op(bio))
783 return false;
784
785
786 if (bio_data_dir(bio) != rq_data_dir(rq))
787 return false;
788
789
790 if (rq->rq_disk != bio->bi_disk || req_no_special_merge(rq))
791 return false;
792
793
794 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
795 return false;
796
797
798 if (req_op(rq) == REQ_OP_WRITE_SAME &&
799 !blk_write_same_mergeable(rq->bio, bio))
800 return false;
801
802
803
804
805
806 if (rq->write_hint != bio->bi_write_hint)
807 return false;
808
809 return true;
810}
811
812enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
813{
814 if (req_op(rq) == REQ_OP_DISCARD &&
815 queue_max_discard_segments(rq->q) > 1)
816 return ELEVATOR_DISCARD_MERGE;
817 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
818 return ELEVATOR_BACK_MERGE;
819 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
820 return ELEVATOR_FRONT_MERGE;
821 return ELEVATOR_NO_MERGE;
822}
823