Loading...
1/*
2 * Functions related to segment and merge handling
3 */
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/bio.h>
7#include <linux/blkdev.h>
8#include <linux/scatterlist.h>
9
10#include "blk.h"
11
12static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
13 struct bio *bio)
14{
15 struct bio_vec bv, bvprv = { NULL };
16 int cluster, high, highprv = 1;
17 unsigned int seg_size, nr_phys_segs;
18 struct bio *fbio, *bbio;
19 struct bvec_iter iter;
20
21 if (!bio)
22 return 0;
23
24 /*
25 * This should probably be returning 0, but blk_add_request_payload()
26 * (Christoph!!!!)
27 */
28 if (bio->bi_rw & REQ_DISCARD)
29 return 1;
30
31 if (bio->bi_rw & REQ_WRITE_SAME)
32 return 1;
33
34 fbio = bio;
35 cluster = blk_queue_cluster(q);
36 seg_size = 0;
37 nr_phys_segs = 0;
38 for_each_bio(bio) {
39 bio_for_each_segment(bv, bio, iter) {
40 /*
41 * the trick here is making sure that a high page is
42 * never considered part of another segment, since that
43 * might change with the bounce page.
44 */
45 high = page_to_pfn(bv.bv_page) > queue_bounce_pfn(q);
46 if (!high && !highprv && cluster) {
47 if (seg_size + bv.bv_len
48 > queue_max_segment_size(q))
49 goto new_segment;
50 if (!BIOVEC_PHYS_MERGEABLE(&bvprv, &bv))
51 goto new_segment;
52 if (!BIOVEC_SEG_BOUNDARY(q, &bvprv, &bv))
53 goto new_segment;
54
55 seg_size += bv.bv_len;
56 bvprv = bv;
57 continue;
58 }
59new_segment:
60 if (nr_phys_segs == 1 && seg_size >
61 fbio->bi_seg_front_size)
62 fbio->bi_seg_front_size = seg_size;
63
64 nr_phys_segs++;
65 bvprv = bv;
66 seg_size = bv.bv_len;
67 highprv = high;
68 }
69 bbio = bio;
70 }
71
72 if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
73 fbio->bi_seg_front_size = seg_size;
74 if (seg_size > bbio->bi_seg_back_size)
75 bbio->bi_seg_back_size = seg_size;
76
77 return nr_phys_segs;
78}
79
80void blk_recalc_rq_segments(struct request *rq)
81{
82 rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio);
83}
84
85void blk_recount_segments(struct request_queue *q, struct bio *bio)
86{
87 struct bio *nxt = bio->bi_next;
88
89 bio->bi_next = NULL;
90 bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio);
91 bio->bi_next = nxt;
92 bio->bi_flags |= (1 << BIO_SEG_VALID);
93}
94EXPORT_SYMBOL(blk_recount_segments);
95
96static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
97 struct bio *nxt)
98{
99 struct bio_vec end_bv = { NULL }, nxt_bv;
100 struct bvec_iter iter;
101
102 if (!blk_queue_cluster(q))
103 return 0;
104
105 if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
106 queue_max_segment_size(q))
107 return 0;
108
109 if (!bio_has_data(bio))
110 return 1;
111
112 bio_for_each_segment(end_bv, bio, iter)
113 if (end_bv.bv_len == iter.bi_size)
114 break;
115
116 nxt_bv = bio_iovec(nxt);
117
118 if (!BIOVEC_PHYS_MERGEABLE(&end_bv, &nxt_bv))
119 return 0;
120
121 /*
122 * bio and nxt are contiguous in memory; check if the queue allows
123 * these two to be merged into one
124 */
125 if (BIOVEC_SEG_BOUNDARY(q, &end_bv, &nxt_bv))
126 return 1;
127
128 return 0;
129}
130
131static inline void
132__blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
133 struct scatterlist *sglist, struct bio_vec *bvprv,
134 struct scatterlist **sg, int *nsegs, int *cluster)
135{
136
137 int nbytes = bvec->bv_len;
138
139 if (*sg && *cluster) {
140 if ((*sg)->length + nbytes > queue_max_segment_size(q))
141 goto new_segment;
142
143 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
144 goto new_segment;
145 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
146 goto new_segment;
147
148 (*sg)->length += nbytes;
149 } else {
150new_segment:
151 if (!*sg)
152 *sg = sglist;
153 else {
154 /*
155 * If the driver previously mapped a shorter
156 * list, we could see a termination bit
157 * prematurely unless it fully inits the sg
158 * table on each mapping. We KNOW that there
159 * must be more entries here or the driver
160 * would be buggy, so force clear the
161 * termination bit to avoid doing a full
162 * sg_init_table() in drivers for each command.
163 */
164 sg_unmark_end(*sg);
165 *sg = sg_next(*sg);
166 }
167
168 sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
169 (*nsegs)++;
170 }
171 *bvprv = *bvec;
172}
173
174static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
175 struct scatterlist *sglist,
176 struct scatterlist **sg)
177{
178 struct bio_vec bvec, bvprv = { NULL };
179 struct bvec_iter iter;
180 int nsegs, cluster;
181
182 nsegs = 0;
183 cluster = blk_queue_cluster(q);
184
185 if (bio->bi_rw & REQ_DISCARD) {
186 /*
187 * This is a hack - drivers should be neither modifying the
188 * biovec, nor relying on bi_vcnt - but because of
189 * blk_add_request_payload(), a discard bio may or may not have
190 * a payload we need to set up here (thank you Christoph) and
191 * bi_vcnt is really the only way of telling if we need to.
192 */
193
194 if (bio->bi_vcnt)
195 goto single_segment;
196
197 return 0;
198 }
199
200 if (bio->bi_rw & REQ_WRITE_SAME) {
201single_segment:
202 *sg = sglist;
203 bvec = bio_iovec(bio);
204 sg_set_page(*sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset);
205 return 1;
206 }
207
208 for_each_bio(bio)
209 bio_for_each_segment(bvec, bio, iter)
210 __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg,
211 &nsegs, &cluster);
212
213 return nsegs;
214}
215
216/*
217 * map a request to scatterlist, return number of sg entries setup. Caller
218 * must make sure sg can hold rq->nr_phys_segments entries
219 */
220int blk_rq_map_sg(struct request_queue *q, struct request *rq,
221 struct scatterlist *sglist)
222{
223 struct scatterlist *sg = NULL;
224 int nsegs = 0;
225
226 if (rq->bio)
227 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg);
228
229 if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
230 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
231 unsigned int pad_len =
232 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
233
234 sg->length += pad_len;
235 rq->extra_len += pad_len;
236 }
237
238 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
239 if (rq->cmd_flags & REQ_WRITE)
240 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
241
242 sg->page_link &= ~0x02;
243 sg = sg_next(sg);
244 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
245 q->dma_drain_size,
246 ((unsigned long)q->dma_drain_buffer) &
247 (PAGE_SIZE - 1));
248 nsegs++;
249 rq->extra_len += q->dma_drain_size;
250 }
251
252 if (sg)
253 sg_mark_end(sg);
254
255 return nsegs;
256}
257EXPORT_SYMBOL(blk_rq_map_sg);
258
259/**
260 * blk_bio_map_sg - map a bio to a scatterlist
261 * @q: request_queue in question
262 * @bio: bio being mapped
263 * @sglist: scatterlist being mapped
264 *
265 * Note:
266 * Caller must make sure sg can hold bio->bi_phys_segments entries
267 *
268 * Will return the number of sg entries setup
269 */
270int blk_bio_map_sg(struct request_queue *q, struct bio *bio,
271 struct scatterlist *sglist)
272{
273 struct scatterlist *sg = NULL;
274 int nsegs;
275 struct bio *next = bio->bi_next;
276 bio->bi_next = NULL;
277
278 nsegs = __blk_bios_map_sg(q, bio, sglist, &sg);
279 bio->bi_next = next;
280 if (sg)
281 sg_mark_end(sg);
282
283 BUG_ON(bio->bi_phys_segments && nsegs > bio->bi_phys_segments);
284 return nsegs;
285}
286EXPORT_SYMBOL(blk_bio_map_sg);
287
288static inline int ll_new_hw_segment(struct request_queue *q,
289 struct request *req,
290 struct bio *bio)
291{
292 int nr_phys_segs = bio_phys_segments(q, bio);
293
294 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
295 goto no_merge;
296
297 if (bio_integrity(bio) && blk_integrity_merge_bio(q, req, bio))
298 goto no_merge;
299
300 /*
301 * This will form the start of a new hw segment. Bump both
302 * counters.
303 */
304 req->nr_phys_segments += nr_phys_segs;
305 return 1;
306
307no_merge:
308 req->cmd_flags |= REQ_NOMERGE;
309 if (req == q->last_merge)
310 q->last_merge = NULL;
311 return 0;
312}
313
314int ll_back_merge_fn(struct request_queue *q, struct request *req,
315 struct bio *bio)
316{
317 if (blk_rq_sectors(req) + bio_sectors(bio) >
318 blk_rq_get_max_sectors(req)) {
319 req->cmd_flags |= REQ_NOMERGE;
320 if (req == q->last_merge)
321 q->last_merge = NULL;
322 return 0;
323 }
324 if (!bio_flagged(req->biotail, BIO_SEG_VALID))
325 blk_recount_segments(q, req->biotail);
326 if (!bio_flagged(bio, BIO_SEG_VALID))
327 blk_recount_segments(q, bio);
328
329 return ll_new_hw_segment(q, req, bio);
330}
331
332int ll_front_merge_fn(struct request_queue *q, struct request *req,
333 struct bio *bio)
334{
335 if (blk_rq_sectors(req) + bio_sectors(bio) >
336 blk_rq_get_max_sectors(req)) {
337 req->cmd_flags |= REQ_NOMERGE;
338 if (req == q->last_merge)
339 q->last_merge = NULL;
340 return 0;
341 }
342 if (!bio_flagged(bio, BIO_SEG_VALID))
343 blk_recount_segments(q, bio);
344 if (!bio_flagged(req->bio, BIO_SEG_VALID))
345 blk_recount_segments(q, req->bio);
346
347 return ll_new_hw_segment(q, req, bio);
348}
349
350/*
351 * blk-mq uses req->special to carry normal driver per-request payload, it
352 * does not indicate a prepared command that we cannot merge with.
353 */
354static bool req_no_special_merge(struct request *req)
355{
356 struct request_queue *q = req->q;
357
358 return !q->mq_ops && req->special;
359}
360
361static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
362 struct request *next)
363{
364 int total_phys_segments;
365 unsigned int seg_size =
366 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
367
368 /*
369 * First check if the either of the requests are re-queued
370 * requests. Can't merge them if they are.
371 */
372 if (req_no_special_merge(req) || req_no_special_merge(next))
373 return 0;
374
375 /*
376 * Will it become too large?
377 */
378 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
379 blk_rq_get_max_sectors(req))
380 return 0;
381
382 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
383 if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
384 if (req->nr_phys_segments == 1)
385 req->bio->bi_seg_front_size = seg_size;
386 if (next->nr_phys_segments == 1)
387 next->biotail->bi_seg_back_size = seg_size;
388 total_phys_segments--;
389 }
390
391 if (total_phys_segments > queue_max_segments(q))
392 return 0;
393
394 if (blk_integrity_rq(req) && blk_integrity_merge_rq(q, req, next))
395 return 0;
396
397 /* Merge is OK... */
398 req->nr_phys_segments = total_phys_segments;
399 return 1;
400}
401
402/**
403 * blk_rq_set_mixed_merge - mark a request as mixed merge
404 * @rq: request to mark as mixed merge
405 *
406 * Description:
407 * @rq is about to be mixed merged. Make sure the attributes
408 * which can be mixed are set in each bio and mark @rq as mixed
409 * merged.
410 */
411void blk_rq_set_mixed_merge(struct request *rq)
412{
413 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
414 struct bio *bio;
415
416 if (rq->cmd_flags & REQ_MIXED_MERGE)
417 return;
418
419 /*
420 * @rq will no longer represent mixable attributes for all the
421 * contained bios. It will just track those of the first one.
422 * Distributes the attributs to each bio.
423 */
424 for (bio = rq->bio; bio; bio = bio->bi_next) {
425 WARN_ON_ONCE((bio->bi_rw & REQ_FAILFAST_MASK) &&
426 (bio->bi_rw & REQ_FAILFAST_MASK) != ff);
427 bio->bi_rw |= ff;
428 }
429 rq->cmd_flags |= REQ_MIXED_MERGE;
430}
431
432static void blk_account_io_merge(struct request *req)
433{
434 if (blk_do_io_stat(req)) {
435 struct hd_struct *part;
436 int cpu;
437
438 cpu = part_stat_lock();
439 part = req->part;
440
441 part_round_stats(cpu, part);
442 part_dec_in_flight(part, rq_data_dir(req));
443
444 hd_struct_put(part);
445 part_stat_unlock();
446 }
447}
448
449/*
450 * Has to be called with the request spinlock acquired
451 */
452static int attempt_merge(struct request_queue *q, struct request *req,
453 struct request *next)
454{
455 if (!rq_mergeable(req) || !rq_mergeable(next))
456 return 0;
457
458 if (!blk_check_merge_flags(req->cmd_flags, next->cmd_flags))
459 return 0;
460
461 /*
462 * not contiguous
463 */
464 if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
465 return 0;
466
467 if (rq_data_dir(req) != rq_data_dir(next)
468 || req->rq_disk != next->rq_disk
469 || req_no_special_merge(next))
470 return 0;
471
472 if (req->cmd_flags & REQ_WRITE_SAME &&
473 !blk_write_same_mergeable(req->bio, next->bio))
474 return 0;
475
476 /*
477 * If we are allowed to merge, then append bio list
478 * from next to rq and release next. merge_requests_fn
479 * will have updated segment counts, update sector
480 * counts here.
481 */
482 if (!ll_merge_requests_fn(q, req, next))
483 return 0;
484
485 /*
486 * If failfast settings disagree or any of the two is already
487 * a mixed merge, mark both as mixed before proceeding. This
488 * makes sure that all involved bios have mixable attributes
489 * set properly.
490 */
491 if ((req->cmd_flags | next->cmd_flags) & REQ_MIXED_MERGE ||
492 (req->cmd_flags & REQ_FAILFAST_MASK) !=
493 (next->cmd_flags & REQ_FAILFAST_MASK)) {
494 blk_rq_set_mixed_merge(req);
495 blk_rq_set_mixed_merge(next);
496 }
497
498 /*
499 * At this point we have either done a back merge
500 * or front merge. We need the smaller start_time of
501 * the merged requests to be the current request
502 * for accounting purposes.
503 */
504 if (time_after(req->start_time, next->start_time))
505 req->start_time = next->start_time;
506
507 req->biotail->bi_next = next->bio;
508 req->biotail = next->biotail;
509
510 req->__data_len += blk_rq_bytes(next);
511
512 elv_merge_requests(q, req, next);
513
514 /*
515 * 'next' is going away, so update stats accordingly
516 */
517 blk_account_io_merge(next);
518
519 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
520 if (blk_rq_cpu_valid(next))
521 req->cpu = next->cpu;
522
523 /* owner-ship of bio passed from next to req */
524 next->bio = NULL;
525 __blk_put_request(q, next);
526 return 1;
527}
528
529int attempt_back_merge(struct request_queue *q, struct request *rq)
530{
531 struct request *next = elv_latter_request(q, rq);
532
533 if (next)
534 return attempt_merge(q, rq, next);
535
536 return 0;
537}
538
539int attempt_front_merge(struct request_queue *q, struct request *rq)
540{
541 struct request *prev = elv_former_request(q, rq);
542
543 if (prev)
544 return attempt_merge(q, prev, rq);
545
546 return 0;
547}
548
549int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
550 struct request *next)
551{
552 return attempt_merge(q, rq, next);
553}
554
555bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
556{
557 if (!rq_mergeable(rq) || !bio_mergeable(bio))
558 return false;
559
560 if (!blk_check_merge_flags(rq->cmd_flags, bio->bi_rw))
561 return false;
562
563 /* different data direction or already started, don't merge */
564 if (bio_data_dir(bio) != rq_data_dir(rq))
565 return false;
566
567 /* must be same device and not a special request */
568 if (rq->rq_disk != bio->bi_bdev->bd_disk || req_no_special_merge(rq))
569 return false;
570
571 /* only merge integrity protected bio into ditto rq */
572 if (bio_integrity(bio) != blk_integrity_rq(rq))
573 return false;
574
575 /* must be using the same buffer */
576 if (rq->cmd_flags & REQ_WRITE_SAME &&
577 !blk_write_same_mergeable(rq->bio, bio))
578 return false;
579
580 return true;
581}
582
583int blk_try_merge(struct request *rq, struct bio *bio)
584{
585 if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
586 return ELEVATOR_BACK_MERGE;
587 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
588 return ELEVATOR_FRONT_MERGE;
589 return ELEVATOR_NO_MERGE;
590}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to segment and merge handling
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/blk-integrity.h>
10#include <linux/scatterlist.h>
11#include <linux/part_stat.h>
12#include <linux/blk-cgroup.h>
13
14#include <trace/events/block.h>
15
16#include "blk.h"
17#include "blk-mq-sched.h"
18#include "blk-rq-qos.h"
19#include "blk-throttle.h"
20
21static inline void bio_get_first_bvec(struct bio *bio, struct bio_vec *bv)
22{
23 *bv = mp_bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
24}
25
26static inline void bio_get_last_bvec(struct bio *bio, struct bio_vec *bv)
27{
28 struct bvec_iter iter = bio->bi_iter;
29 int idx;
30
31 bio_get_first_bvec(bio, bv);
32 if (bv->bv_len == bio->bi_iter.bi_size)
33 return; /* this bio only has a single bvec */
34
35 bio_advance_iter(bio, &iter, iter.bi_size);
36
37 if (!iter.bi_bvec_done)
38 idx = iter.bi_idx - 1;
39 else /* in the middle of bvec */
40 idx = iter.bi_idx;
41
42 *bv = bio->bi_io_vec[idx];
43
44 /*
45 * iter.bi_bvec_done records actual length of the last bvec
46 * if this bio ends in the middle of one io vector
47 */
48 if (iter.bi_bvec_done)
49 bv->bv_len = iter.bi_bvec_done;
50}
51
52static inline bool bio_will_gap(struct request_queue *q,
53 struct request *prev_rq, struct bio *prev, struct bio *next)
54{
55 struct bio_vec pb, nb;
56
57 if (!bio_has_data(prev) || !queue_virt_boundary(q))
58 return false;
59
60 /*
61 * Don't merge if the 1st bio starts with non-zero offset, otherwise it
62 * is quite difficult to respect the sg gap limit. We work hard to
63 * merge a huge number of small single bios in case of mkfs.
64 */
65 if (prev_rq)
66 bio_get_first_bvec(prev_rq->bio, &pb);
67 else
68 bio_get_first_bvec(prev, &pb);
69 if (pb.bv_offset & queue_virt_boundary(q))
70 return true;
71
72 /*
73 * We don't need to worry about the situation that the merged segment
74 * ends in unaligned virt boundary:
75 *
76 * - if 'pb' ends aligned, the merged segment ends aligned
77 * - if 'pb' ends unaligned, the next bio must include
78 * one single bvec of 'nb', otherwise the 'nb' can't
79 * merge with 'pb'
80 */
81 bio_get_last_bvec(prev, &pb);
82 bio_get_first_bvec(next, &nb);
83 if (biovec_phys_mergeable(q, &pb, &nb))
84 return false;
85 return __bvec_gap_to_prev(&q->limits, &pb, nb.bv_offset);
86}
87
88static inline bool req_gap_back_merge(struct request *req, struct bio *bio)
89{
90 return bio_will_gap(req->q, req, req->biotail, bio);
91}
92
93static inline bool req_gap_front_merge(struct request *req, struct bio *bio)
94{
95 return bio_will_gap(req->q, NULL, bio, req->bio);
96}
97
98/*
99 * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
100 * is defined as 'unsigned int', meantime it has to be aligned to with the
101 * logical block size, which is the minimum accepted unit by hardware.
102 */
103static unsigned int bio_allowed_max_sectors(const struct queue_limits *lim)
104{
105 return round_down(UINT_MAX, lim->logical_block_size) >> SECTOR_SHIFT;
106}
107
108static struct bio *bio_split_discard(struct bio *bio,
109 const struct queue_limits *lim,
110 unsigned *nsegs, struct bio_set *bs)
111{
112 unsigned int max_discard_sectors, granularity;
113 sector_t tmp;
114 unsigned split_sectors;
115
116 *nsegs = 1;
117
118 granularity = max(lim->discard_granularity >> 9, 1U);
119
120 max_discard_sectors =
121 min(lim->max_discard_sectors, bio_allowed_max_sectors(lim));
122 max_discard_sectors -= max_discard_sectors % granularity;
123 if (unlikely(!max_discard_sectors))
124 return NULL;
125
126 if (bio_sectors(bio) <= max_discard_sectors)
127 return NULL;
128
129 split_sectors = max_discard_sectors;
130
131 /*
132 * If the next starting sector would be misaligned, stop the discard at
133 * the previous aligned sector.
134 */
135 tmp = bio->bi_iter.bi_sector + split_sectors -
136 ((lim->discard_alignment >> 9) % granularity);
137 tmp = sector_div(tmp, granularity);
138
139 if (split_sectors > tmp)
140 split_sectors -= tmp;
141
142 return bio_split(bio, split_sectors, GFP_NOIO, bs);
143}
144
145static struct bio *bio_split_write_zeroes(struct bio *bio,
146 const struct queue_limits *lim,
147 unsigned *nsegs, struct bio_set *bs)
148{
149 *nsegs = 0;
150 if (!lim->max_write_zeroes_sectors)
151 return NULL;
152 if (bio_sectors(bio) <= lim->max_write_zeroes_sectors)
153 return NULL;
154 return bio_split(bio, lim->max_write_zeroes_sectors, GFP_NOIO, bs);
155}
156
157/*
158 * Return the maximum number of sectors from the start of a bio that may be
159 * submitted as a single request to a block device. If enough sectors remain,
160 * align the end to the physical block size. Otherwise align the end to the
161 * logical block size. This approach minimizes the number of non-aligned
162 * requests that are submitted to a block device if the start of a bio is not
163 * aligned to a physical block boundary.
164 */
165static inline unsigned get_max_io_size(struct bio *bio,
166 const struct queue_limits *lim)
167{
168 unsigned pbs = lim->physical_block_size >> SECTOR_SHIFT;
169 unsigned lbs = lim->logical_block_size >> SECTOR_SHIFT;
170 unsigned max_sectors = lim->max_sectors, start, end;
171
172 if (lim->chunk_sectors) {
173 max_sectors = min(max_sectors,
174 blk_chunk_sectors_left(bio->bi_iter.bi_sector,
175 lim->chunk_sectors));
176 }
177
178 start = bio->bi_iter.bi_sector & (pbs - 1);
179 end = (start + max_sectors) & ~(pbs - 1);
180 if (end > start)
181 return end - start;
182 return max_sectors & ~(lbs - 1);
183}
184
185/**
186 * get_max_segment_size() - maximum number of bytes to add as a single segment
187 * @lim: Request queue limits.
188 * @start_page: See below.
189 * @offset: Offset from @start_page where to add a segment.
190 *
191 * Returns the maximum number of bytes that can be added as a single segment.
192 */
193static inline unsigned get_max_segment_size(const struct queue_limits *lim,
194 struct page *start_page, unsigned long offset)
195{
196 unsigned long mask = lim->seg_boundary_mask;
197
198 offset = mask & (page_to_phys(start_page) + offset);
199
200 /*
201 * Prevent an overflow if mask = ULONG_MAX and offset = 0 by adding 1
202 * after having calculated the minimum.
203 */
204 return min(mask - offset, (unsigned long)lim->max_segment_size - 1) + 1;
205}
206
207/**
208 * bvec_split_segs - verify whether or not a bvec should be split in the middle
209 * @lim: [in] queue limits to split based on
210 * @bv: [in] bvec to examine
211 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
212 * by the number of segments from @bv that may be appended to that
213 * bio without exceeding @max_segs
214 * @bytes: [in,out] Number of bytes in the bio being built. Incremented
215 * by the number of bytes from @bv that may be appended to that
216 * bio without exceeding @max_bytes
217 * @max_segs: [in] upper bound for *@nsegs
218 * @max_bytes: [in] upper bound for *@bytes
219 *
220 * When splitting a bio, it can happen that a bvec is encountered that is too
221 * big to fit in a single segment and hence that it has to be split in the
222 * middle. This function verifies whether or not that should happen. The value
223 * %true is returned if and only if appending the entire @bv to a bio with
224 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
225 * the block driver.
226 */
227static bool bvec_split_segs(const struct queue_limits *lim,
228 const struct bio_vec *bv, unsigned *nsegs, unsigned *bytes,
229 unsigned max_segs, unsigned max_bytes)
230{
231 unsigned max_len = min(max_bytes, UINT_MAX) - *bytes;
232 unsigned len = min(bv->bv_len, max_len);
233 unsigned total_len = 0;
234 unsigned seg_size = 0;
235
236 while (len && *nsegs < max_segs) {
237 seg_size = get_max_segment_size(lim, bv->bv_page,
238 bv->bv_offset + total_len);
239 seg_size = min(seg_size, len);
240
241 (*nsegs)++;
242 total_len += seg_size;
243 len -= seg_size;
244
245 if ((bv->bv_offset + total_len) & lim->virt_boundary_mask)
246 break;
247 }
248
249 *bytes += total_len;
250
251 /* tell the caller to split the bvec if it is too big to fit */
252 return len > 0 || bv->bv_len > max_len;
253}
254
255/**
256 * bio_split_rw - split a bio in two bios
257 * @bio: [in] bio to be split
258 * @lim: [in] queue limits to split based on
259 * @segs: [out] number of segments in the bio with the first half of the sectors
260 * @bs: [in] bio set to allocate the clone from
261 * @max_bytes: [in] maximum number of bytes per bio
262 *
263 * Clone @bio, update the bi_iter of the clone to represent the first sectors
264 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
265 * following is guaranteed for the cloned bio:
266 * - That it has at most @max_bytes worth of data
267 * - That it has at most queue_max_segments(@q) segments.
268 *
269 * Except for discard requests the cloned bio will point at the bi_io_vec of
270 * the original bio. It is the responsibility of the caller to ensure that the
271 * original bio is not freed before the cloned bio. The caller is also
272 * responsible for ensuring that @bs is only destroyed after processing of the
273 * split bio has finished.
274 */
275struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim,
276 unsigned *segs, struct bio_set *bs, unsigned max_bytes)
277{
278 struct bio_vec bv, bvprv, *bvprvp = NULL;
279 struct bvec_iter iter;
280 unsigned nsegs = 0, bytes = 0;
281
282 bio_for_each_bvec(bv, bio, iter) {
283 /*
284 * If the queue doesn't support SG gaps and adding this
285 * offset would create a gap, disallow it.
286 */
287 if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv.bv_offset))
288 goto split;
289
290 if (nsegs < lim->max_segments &&
291 bytes + bv.bv_len <= max_bytes &&
292 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
293 nsegs++;
294 bytes += bv.bv_len;
295 } else {
296 if (bvec_split_segs(lim, &bv, &nsegs, &bytes,
297 lim->max_segments, max_bytes))
298 goto split;
299 }
300
301 bvprv = bv;
302 bvprvp = &bvprv;
303 }
304
305 *segs = nsegs;
306 return NULL;
307split:
308 /*
309 * We can't sanely support splitting for a REQ_NOWAIT bio. End it
310 * with EAGAIN if splitting is required and return an error pointer.
311 */
312 if (bio->bi_opf & REQ_NOWAIT) {
313 bio->bi_status = BLK_STS_AGAIN;
314 bio_endio(bio);
315 return ERR_PTR(-EAGAIN);
316 }
317
318 *segs = nsegs;
319
320 /*
321 * Individual bvecs might not be logical block aligned. Round down the
322 * split size so that each bio is properly block size aligned, even if
323 * we do not use the full hardware limits.
324 */
325 bytes = ALIGN_DOWN(bytes, lim->logical_block_size);
326
327 /*
328 * Bio splitting may cause subtle trouble such as hang when doing sync
329 * iopoll in direct IO routine. Given performance gain of iopoll for
330 * big IO can be trival, disable iopoll when split needed.
331 */
332 bio_clear_polled(bio);
333 return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs);
334}
335EXPORT_SYMBOL_GPL(bio_split_rw);
336
337/**
338 * __bio_split_to_limits - split a bio to fit the queue limits
339 * @bio: bio to be split
340 * @lim: queue limits to split based on
341 * @nr_segs: returns the number of segments in the returned bio
342 *
343 * Check if @bio needs splitting based on the queue limits, and if so split off
344 * a bio fitting the limits from the beginning of @bio and return it. @bio is
345 * shortened to the remainder and re-submitted.
346 *
347 * The split bio is allocated from @q->bio_split, which is provided by the
348 * block layer.
349 */
350struct bio *__bio_split_to_limits(struct bio *bio,
351 const struct queue_limits *lim,
352 unsigned int *nr_segs)
353{
354 struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split;
355 struct bio *split;
356
357 switch (bio_op(bio)) {
358 case REQ_OP_DISCARD:
359 case REQ_OP_SECURE_ERASE:
360 split = bio_split_discard(bio, lim, nr_segs, bs);
361 break;
362 case REQ_OP_WRITE_ZEROES:
363 split = bio_split_write_zeroes(bio, lim, nr_segs, bs);
364 break;
365 default:
366 split = bio_split_rw(bio, lim, nr_segs, bs,
367 get_max_io_size(bio, lim) << SECTOR_SHIFT);
368 if (IS_ERR(split))
369 return NULL;
370 break;
371 }
372
373 if (split) {
374 /* there isn't chance to merge the split bio */
375 split->bi_opf |= REQ_NOMERGE;
376
377 blkcg_bio_issue_init(split);
378 bio_chain(split, bio);
379 trace_block_split(split, bio->bi_iter.bi_sector);
380 submit_bio_noacct(bio);
381 return split;
382 }
383 return bio;
384}
385
386/**
387 * bio_split_to_limits - split a bio to fit the queue limits
388 * @bio: bio to be split
389 *
390 * Check if @bio needs splitting based on the queue limits of @bio->bi_bdev, and
391 * if so split off a bio fitting the limits from the beginning of @bio and
392 * return it. @bio is shortened to the remainder and re-submitted.
393 *
394 * The split bio is allocated from @q->bio_split, which is provided by the
395 * block layer.
396 */
397struct bio *bio_split_to_limits(struct bio *bio)
398{
399 const struct queue_limits *lim = &bdev_get_queue(bio->bi_bdev)->limits;
400 unsigned int nr_segs;
401
402 if (bio_may_exceed_limits(bio, lim))
403 return __bio_split_to_limits(bio, lim, &nr_segs);
404 return bio;
405}
406EXPORT_SYMBOL(bio_split_to_limits);
407
408unsigned int blk_recalc_rq_segments(struct request *rq)
409{
410 unsigned int nr_phys_segs = 0;
411 unsigned int bytes = 0;
412 struct req_iterator iter;
413 struct bio_vec bv;
414
415 if (!rq->bio)
416 return 0;
417
418 switch (bio_op(rq->bio)) {
419 case REQ_OP_DISCARD:
420 case REQ_OP_SECURE_ERASE:
421 if (queue_max_discard_segments(rq->q) > 1) {
422 struct bio *bio = rq->bio;
423
424 for_each_bio(bio)
425 nr_phys_segs++;
426 return nr_phys_segs;
427 }
428 return 1;
429 case REQ_OP_WRITE_ZEROES:
430 return 0;
431 default:
432 break;
433 }
434
435 rq_for_each_bvec(bv, rq, iter)
436 bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes,
437 UINT_MAX, UINT_MAX);
438 return nr_phys_segs;
439}
440
441static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
442 struct scatterlist *sglist)
443{
444 if (!*sg)
445 return sglist;
446
447 /*
448 * If the driver previously mapped a shorter list, we could see a
449 * termination bit prematurely unless it fully inits the sg table
450 * on each mapping. We KNOW that there must be more entries here
451 * or the driver would be buggy, so force clear the termination bit
452 * to avoid doing a full sg_init_table() in drivers for each command.
453 */
454 sg_unmark_end(*sg);
455 return sg_next(*sg);
456}
457
458static unsigned blk_bvec_map_sg(struct request_queue *q,
459 struct bio_vec *bvec, struct scatterlist *sglist,
460 struct scatterlist **sg)
461{
462 unsigned nbytes = bvec->bv_len;
463 unsigned nsegs = 0, total = 0;
464
465 while (nbytes > 0) {
466 unsigned offset = bvec->bv_offset + total;
467 unsigned len = min(get_max_segment_size(&q->limits,
468 bvec->bv_page, offset), nbytes);
469 struct page *page = bvec->bv_page;
470
471 /*
472 * Unfortunately a fair number of drivers barf on scatterlists
473 * that have an offset larger than PAGE_SIZE, despite other
474 * subsystems dealing with that invariant just fine. For now
475 * stick to the legacy format where we never present those from
476 * the block layer, but the code below should be removed once
477 * these offenders (mostly MMC/SD drivers) are fixed.
478 */
479 page += (offset >> PAGE_SHIFT);
480 offset &= ~PAGE_MASK;
481
482 *sg = blk_next_sg(sg, sglist);
483 sg_set_page(*sg, page, len, offset);
484
485 total += len;
486 nbytes -= len;
487 nsegs++;
488 }
489
490 return nsegs;
491}
492
493static inline int __blk_bvec_map_sg(struct bio_vec bv,
494 struct scatterlist *sglist, struct scatterlist **sg)
495{
496 *sg = blk_next_sg(sg, sglist);
497 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
498 return 1;
499}
500
501/* only try to merge bvecs into one sg if they are from two bios */
502static inline bool
503__blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
504 struct bio_vec *bvprv, struct scatterlist **sg)
505{
506
507 int nbytes = bvec->bv_len;
508
509 if (!*sg)
510 return false;
511
512 if ((*sg)->length + nbytes > queue_max_segment_size(q))
513 return false;
514
515 if (!biovec_phys_mergeable(q, bvprv, bvec))
516 return false;
517
518 (*sg)->length += nbytes;
519
520 return true;
521}
522
523static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
524 struct scatterlist *sglist,
525 struct scatterlist **sg)
526{
527 struct bio_vec bvec, bvprv = { NULL };
528 struct bvec_iter iter;
529 int nsegs = 0;
530 bool new_bio = false;
531
532 for_each_bio(bio) {
533 bio_for_each_bvec(bvec, bio, iter) {
534 /*
535 * Only try to merge bvecs from two bios given we
536 * have done bio internal merge when adding pages
537 * to bio
538 */
539 if (new_bio &&
540 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
541 goto next_bvec;
542
543 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
544 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
545 else
546 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
547 next_bvec:
548 new_bio = false;
549 }
550 if (likely(bio->bi_iter.bi_size)) {
551 bvprv = bvec;
552 new_bio = true;
553 }
554 }
555
556 return nsegs;
557}
558
559/*
560 * map a request to scatterlist, return number of sg entries setup. Caller
561 * must make sure sg can hold rq->nr_phys_segments entries
562 */
563int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
564 struct scatterlist *sglist, struct scatterlist **last_sg)
565{
566 int nsegs = 0;
567
568 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
569 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
570 else if (rq->bio)
571 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
572
573 if (*last_sg)
574 sg_mark_end(*last_sg);
575
576 /*
577 * Something must have been wrong if the figured number of
578 * segment is bigger than number of req's physical segments
579 */
580 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
581
582 return nsegs;
583}
584EXPORT_SYMBOL(__blk_rq_map_sg);
585
586static inline unsigned int blk_rq_get_max_sectors(struct request *rq,
587 sector_t offset)
588{
589 struct request_queue *q = rq->q;
590 unsigned int max_sectors;
591
592 if (blk_rq_is_passthrough(rq))
593 return q->limits.max_hw_sectors;
594
595 max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
596 if (!q->limits.chunk_sectors ||
597 req_op(rq) == REQ_OP_DISCARD ||
598 req_op(rq) == REQ_OP_SECURE_ERASE)
599 return max_sectors;
600 return min(max_sectors,
601 blk_chunk_sectors_left(offset, q->limits.chunk_sectors));
602}
603
604static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
605 unsigned int nr_phys_segs)
606{
607 if (!blk_cgroup_mergeable(req, bio))
608 goto no_merge;
609
610 if (blk_integrity_merge_bio(req->q, req, bio) == false)
611 goto no_merge;
612
613 /* discard request merge won't add new segment */
614 if (req_op(req) == REQ_OP_DISCARD)
615 return 1;
616
617 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
618 goto no_merge;
619
620 /*
621 * This will form the start of a new hw segment. Bump both
622 * counters.
623 */
624 req->nr_phys_segments += nr_phys_segs;
625 return 1;
626
627no_merge:
628 req_set_nomerge(req->q, req);
629 return 0;
630}
631
632int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
633{
634 if (req_gap_back_merge(req, bio))
635 return 0;
636 if (blk_integrity_rq(req) &&
637 integrity_req_gap_back_merge(req, bio))
638 return 0;
639 if (!bio_crypt_ctx_back_mergeable(req, bio))
640 return 0;
641 if (blk_rq_sectors(req) + bio_sectors(bio) >
642 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
643 req_set_nomerge(req->q, req);
644 return 0;
645 }
646
647 return ll_new_hw_segment(req, bio, nr_segs);
648}
649
650static int ll_front_merge_fn(struct request *req, struct bio *bio,
651 unsigned int nr_segs)
652{
653 if (req_gap_front_merge(req, bio))
654 return 0;
655 if (blk_integrity_rq(req) &&
656 integrity_req_gap_front_merge(req, bio))
657 return 0;
658 if (!bio_crypt_ctx_front_mergeable(req, bio))
659 return 0;
660 if (blk_rq_sectors(req) + bio_sectors(bio) >
661 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
662 req_set_nomerge(req->q, req);
663 return 0;
664 }
665
666 return ll_new_hw_segment(req, bio, nr_segs);
667}
668
669static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
670 struct request *next)
671{
672 unsigned short segments = blk_rq_nr_discard_segments(req);
673
674 if (segments >= queue_max_discard_segments(q))
675 goto no_merge;
676 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
677 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
678 goto no_merge;
679
680 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
681 return true;
682no_merge:
683 req_set_nomerge(q, req);
684 return false;
685}
686
687static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
688 struct request *next)
689{
690 int total_phys_segments;
691
692 if (req_gap_back_merge(req, next->bio))
693 return 0;
694
695 /*
696 * Will it become too large?
697 */
698 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
699 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
700 return 0;
701
702 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
703 if (total_phys_segments > blk_rq_get_max_segments(req))
704 return 0;
705
706 if (!blk_cgroup_mergeable(req, next->bio))
707 return 0;
708
709 if (blk_integrity_merge_rq(q, req, next) == false)
710 return 0;
711
712 if (!bio_crypt_ctx_merge_rq(req, next))
713 return 0;
714
715 /* Merge is OK... */
716 req->nr_phys_segments = total_phys_segments;
717 return 1;
718}
719
720/**
721 * blk_rq_set_mixed_merge - mark a request as mixed merge
722 * @rq: request to mark as mixed merge
723 *
724 * Description:
725 * @rq is about to be mixed merged. Make sure the attributes
726 * which can be mixed are set in each bio and mark @rq as mixed
727 * merged.
728 */
729static void blk_rq_set_mixed_merge(struct request *rq)
730{
731 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
732 struct bio *bio;
733
734 if (rq->rq_flags & RQF_MIXED_MERGE)
735 return;
736
737 /*
738 * @rq will no longer represent mixable attributes for all the
739 * contained bios. It will just track those of the first one.
740 * Distributes the attributs to each bio.
741 */
742 for (bio = rq->bio; bio; bio = bio->bi_next) {
743 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
744 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
745 bio->bi_opf |= ff;
746 }
747 rq->rq_flags |= RQF_MIXED_MERGE;
748}
749
750static inline blk_opf_t bio_failfast(const struct bio *bio)
751{
752 if (bio->bi_opf & REQ_RAHEAD)
753 return REQ_FAILFAST_MASK;
754
755 return bio->bi_opf & REQ_FAILFAST_MASK;
756}
757
758/*
759 * After we are marked as MIXED_MERGE, any new RA bio has to be updated
760 * as failfast, and request's failfast has to be updated in case of
761 * front merge.
762 */
763static inline void blk_update_mixed_merge(struct request *req,
764 struct bio *bio, bool front_merge)
765{
766 if (req->rq_flags & RQF_MIXED_MERGE) {
767 if (bio->bi_opf & REQ_RAHEAD)
768 bio->bi_opf |= REQ_FAILFAST_MASK;
769
770 if (front_merge) {
771 req->cmd_flags &= ~REQ_FAILFAST_MASK;
772 req->cmd_flags |= bio->bi_opf & REQ_FAILFAST_MASK;
773 }
774 }
775}
776
777static void blk_account_io_merge_request(struct request *req)
778{
779 if (blk_do_io_stat(req)) {
780 part_stat_lock();
781 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
782 part_stat_local_dec(req->part,
783 in_flight[op_is_write(req_op(req))]);
784 part_stat_unlock();
785 }
786}
787
788static enum elv_merge blk_try_req_merge(struct request *req,
789 struct request *next)
790{
791 if (blk_discard_mergable(req))
792 return ELEVATOR_DISCARD_MERGE;
793 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
794 return ELEVATOR_BACK_MERGE;
795
796 return ELEVATOR_NO_MERGE;
797}
798
799/*
800 * For non-mq, this has to be called with the request spinlock acquired.
801 * For mq with scheduling, the appropriate queue wide lock should be held.
802 */
803static struct request *attempt_merge(struct request_queue *q,
804 struct request *req, struct request *next)
805{
806 if (!rq_mergeable(req) || !rq_mergeable(next))
807 return NULL;
808
809 if (req_op(req) != req_op(next))
810 return NULL;
811
812 if (rq_data_dir(req) != rq_data_dir(next))
813 return NULL;
814
815 /* Don't merge requests with different write hints. */
816 if (req->write_hint != next->write_hint)
817 return NULL;
818
819 if (req->ioprio != next->ioprio)
820 return NULL;
821
822 /*
823 * If we are allowed to merge, then append bio list
824 * from next to rq and release next. merge_requests_fn
825 * will have updated segment counts, update sector
826 * counts here. Handle DISCARDs separately, as they
827 * have separate settings.
828 */
829
830 switch (blk_try_req_merge(req, next)) {
831 case ELEVATOR_DISCARD_MERGE:
832 if (!req_attempt_discard_merge(q, req, next))
833 return NULL;
834 break;
835 case ELEVATOR_BACK_MERGE:
836 if (!ll_merge_requests_fn(q, req, next))
837 return NULL;
838 break;
839 default:
840 return NULL;
841 }
842
843 /*
844 * If failfast settings disagree or any of the two is already
845 * a mixed merge, mark both as mixed before proceeding. This
846 * makes sure that all involved bios have mixable attributes
847 * set properly.
848 */
849 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
850 (req->cmd_flags & REQ_FAILFAST_MASK) !=
851 (next->cmd_flags & REQ_FAILFAST_MASK)) {
852 blk_rq_set_mixed_merge(req);
853 blk_rq_set_mixed_merge(next);
854 }
855
856 /*
857 * At this point we have either done a back merge or front merge. We
858 * need the smaller start_time_ns of the merged requests to be the
859 * current request for accounting purposes.
860 */
861 if (next->start_time_ns < req->start_time_ns)
862 req->start_time_ns = next->start_time_ns;
863
864 req->biotail->bi_next = next->bio;
865 req->biotail = next->biotail;
866
867 req->__data_len += blk_rq_bytes(next);
868
869 if (!blk_discard_mergable(req))
870 elv_merge_requests(q, req, next);
871
872 blk_crypto_rq_put_keyslot(next);
873
874 /*
875 * 'next' is going away, so update stats accordingly
876 */
877 blk_account_io_merge_request(next);
878
879 trace_block_rq_merge(next);
880
881 /*
882 * ownership of bio passed from next to req, return 'next' for
883 * the caller to free
884 */
885 next->bio = NULL;
886 return next;
887}
888
889static struct request *attempt_back_merge(struct request_queue *q,
890 struct request *rq)
891{
892 struct request *next = elv_latter_request(q, rq);
893
894 if (next)
895 return attempt_merge(q, rq, next);
896
897 return NULL;
898}
899
900static struct request *attempt_front_merge(struct request_queue *q,
901 struct request *rq)
902{
903 struct request *prev = elv_former_request(q, rq);
904
905 if (prev)
906 return attempt_merge(q, prev, rq);
907
908 return NULL;
909}
910
911/*
912 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
913 * otherwise. The caller is responsible for freeing 'next' if the merge
914 * happened.
915 */
916bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
917 struct request *next)
918{
919 return attempt_merge(q, rq, next);
920}
921
922bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
923{
924 if (!rq_mergeable(rq) || !bio_mergeable(bio))
925 return false;
926
927 if (req_op(rq) != bio_op(bio))
928 return false;
929
930 /* different data direction or already started, don't merge */
931 if (bio_data_dir(bio) != rq_data_dir(rq))
932 return false;
933
934 /* don't merge across cgroup boundaries */
935 if (!blk_cgroup_mergeable(rq, bio))
936 return false;
937
938 /* only merge integrity protected bio into ditto rq */
939 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
940 return false;
941
942 /* Only merge if the crypt contexts are compatible */
943 if (!bio_crypt_rq_ctx_compatible(rq, bio))
944 return false;
945
946 /* Don't merge requests with different write hints. */
947 if (rq->write_hint != bio->bi_write_hint)
948 return false;
949
950 if (rq->ioprio != bio_prio(bio))
951 return false;
952
953 return true;
954}
955
956enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
957{
958 if (blk_discard_mergable(rq))
959 return ELEVATOR_DISCARD_MERGE;
960 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
961 return ELEVATOR_BACK_MERGE;
962 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
963 return ELEVATOR_FRONT_MERGE;
964 return ELEVATOR_NO_MERGE;
965}
966
967static void blk_account_io_merge_bio(struct request *req)
968{
969 if (!blk_do_io_stat(req))
970 return;
971
972 part_stat_lock();
973 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
974 part_stat_unlock();
975}
976
977enum bio_merge_status {
978 BIO_MERGE_OK,
979 BIO_MERGE_NONE,
980 BIO_MERGE_FAILED,
981};
982
983static enum bio_merge_status bio_attempt_back_merge(struct request *req,
984 struct bio *bio, unsigned int nr_segs)
985{
986 const blk_opf_t ff = bio_failfast(bio);
987
988 if (!ll_back_merge_fn(req, bio, nr_segs))
989 return BIO_MERGE_FAILED;
990
991 trace_block_bio_backmerge(bio);
992 rq_qos_merge(req->q, req, bio);
993
994 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
995 blk_rq_set_mixed_merge(req);
996
997 blk_update_mixed_merge(req, bio, false);
998
999 req->biotail->bi_next = bio;
1000 req->biotail = bio;
1001 req->__data_len += bio->bi_iter.bi_size;
1002
1003 bio_crypt_free_ctx(bio);
1004
1005 blk_account_io_merge_bio(req);
1006 return BIO_MERGE_OK;
1007}
1008
1009static enum bio_merge_status bio_attempt_front_merge(struct request *req,
1010 struct bio *bio, unsigned int nr_segs)
1011{
1012 const blk_opf_t ff = bio_failfast(bio);
1013
1014 if (!ll_front_merge_fn(req, bio, nr_segs))
1015 return BIO_MERGE_FAILED;
1016
1017 trace_block_bio_frontmerge(bio);
1018 rq_qos_merge(req->q, req, bio);
1019
1020 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1021 blk_rq_set_mixed_merge(req);
1022
1023 blk_update_mixed_merge(req, bio, true);
1024
1025 bio->bi_next = req->bio;
1026 req->bio = bio;
1027
1028 req->__sector = bio->bi_iter.bi_sector;
1029 req->__data_len += bio->bi_iter.bi_size;
1030
1031 bio_crypt_do_front_merge(req, bio);
1032
1033 blk_account_io_merge_bio(req);
1034 return BIO_MERGE_OK;
1035}
1036
1037static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
1038 struct request *req, struct bio *bio)
1039{
1040 unsigned short segments = blk_rq_nr_discard_segments(req);
1041
1042 if (segments >= queue_max_discard_segments(q))
1043 goto no_merge;
1044 if (blk_rq_sectors(req) + bio_sectors(bio) >
1045 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1046 goto no_merge;
1047
1048 rq_qos_merge(q, req, bio);
1049
1050 req->biotail->bi_next = bio;
1051 req->biotail = bio;
1052 req->__data_len += bio->bi_iter.bi_size;
1053 req->nr_phys_segments = segments + 1;
1054
1055 blk_account_io_merge_bio(req);
1056 return BIO_MERGE_OK;
1057no_merge:
1058 req_set_nomerge(q, req);
1059 return BIO_MERGE_FAILED;
1060}
1061
1062static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1063 struct request *rq,
1064 struct bio *bio,
1065 unsigned int nr_segs,
1066 bool sched_allow_merge)
1067{
1068 if (!blk_rq_merge_ok(rq, bio))
1069 return BIO_MERGE_NONE;
1070
1071 switch (blk_try_merge(rq, bio)) {
1072 case ELEVATOR_BACK_MERGE:
1073 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1074 return bio_attempt_back_merge(rq, bio, nr_segs);
1075 break;
1076 case ELEVATOR_FRONT_MERGE:
1077 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1078 return bio_attempt_front_merge(rq, bio, nr_segs);
1079 break;
1080 case ELEVATOR_DISCARD_MERGE:
1081 return bio_attempt_discard_merge(q, rq, bio);
1082 default:
1083 return BIO_MERGE_NONE;
1084 }
1085
1086 return BIO_MERGE_FAILED;
1087}
1088
1089/**
1090 * blk_attempt_plug_merge - try to merge with %current's plugged list
1091 * @q: request_queue new bio is being queued at
1092 * @bio: new bio being queued
1093 * @nr_segs: number of segments in @bio
1094 * from the passed in @q already in the plug list
1095 *
1096 * Determine whether @bio being queued on @q can be merged with the previous
1097 * request on %current's plugged list. Returns %true if merge was successful,
1098 * otherwise %false.
1099 *
1100 * Plugging coalesces IOs from the same issuer for the same purpose without
1101 * going through @q->queue_lock. As such it's more of an issuing mechanism
1102 * than scheduling, and the request, while may have elvpriv data, is not
1103 * added on the elevator at this point. In addition, we don't have
1104 * reliable access to the elevator outside queue lock. Only check basic
1105 * merging parameters without querying the elevator.
1106 *
1107 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1108 */
1109bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1110 unsigned int nr_segs)
1111{
1112 struct blk_plug *plug;
1113 struct request *rq;
1114
1115 plug = blk_mq_plug(bio);
1116 if (!plug || rq_list_empty(plug->mq_list))
1117 return false;
1118
1119 rq_list_for_each(&plug->mq_list, rq) {
1120 if (rq->q == q) {
1121 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1122 BIO_MERGE_OK)
1123 return true;
1124 break;
1125 }
1126
1127 /*
1128 * Only keep iterating plug list for merges if we have multiple
1129 * queues
1130 */
1131 if (!plug->multiple_queues)
1132 break;
1133 }
1134 return false;
1135}
1136
1137/*
1138 * Iterate list of requests and see if we can merge this bio with any
1139 * of them.
1140 */
1141bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1142 struct bio *bio, unsigned int nr_segs)
1143{
1144 struct request *rq;
1145 int checked = 8;
1146
1147 list_for_each_entry_reverse(rq, list, queuelist) {
1148 if (!checked--)
1149 break;
1150
1151 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1152 case BIO_MERGE_NONE:
1153 continue;
1154 case BIO_MERGE_OK:
1155 return true;
1156 case BIO_MERGE_FAILED:
1157 return false;
1158 }
1159
1160 }
1161
1162 return false;
1163}
1164EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1165
1166bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1167 unsigned int nr_segs, struct request **merged_request)
1168{
1169 struct request *rq;
1170
1171 switch (elv_merge(q, &rq, bio)) {
1172 case ELEVATOR_BACK_MERGE:
1173 if (!blk_mq_sched_allow_merge(q, rq, bio))
1174 return false;
1175 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1176 return false;
1177 *merged_request = attempt_back_merge(q, rq);
1178 if (!*merged_request)
1179 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1180 return true;
1181 case ELEVATOR_FRONT_MERGE:
1182 if (!blk_mq_sched_allow_merge(q, rq, bio))
1183 return false;
1184 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1185 return false;
1186 *merged_request = attempt_front_merge(q, rq);
1187 if (!*merged_request)
1188 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1189 return true;
1190 case ELEVATOR_DISCARD_MERGE:
1191 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1192 default:
1193 return false;
1194 }
1195}
1196EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);