1 // SPDX-License-Identifier: GPL-2.0
3 * Functions related to segment and merge handling
5 #include <linux/kernel.h>
6 #include <linux/module.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>
14 #include <trace/events/block.h>
17 #include "blk-mq-sched.h"
18 #include "blk-rq-qos.h"
19 #include "blk-throttle.h"
21 static inline void bio_get_first_bvec(struct bio *bio, struct bio_vec *bv)
23 *bv = mp_bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
26 static inline void bio_get_last_bvec(struct bio *bio, struct bio_vec *bv)
28 struct bvec_iter iter = bio->bi_iter;
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 */
35 bio_advance_iter(bio, &iter, iter.bi_size);
37 if (!iter.bi_bvec_done)
38 idx = iter.bi_idx - 1;
39 else /* in the middle of bvec */
42 *bv = bio->bi_io_vec[idx];
45 * iter.bi_bvec_done records actual length of the last bvec
46 * if this bio ends in the middle of one io vector
48 if (iter.bi_bvec_done)
49 bv->bv_len = iter.bi_bvec_done;
52 static inline bool bio_will_gap(struct request_queue *q,
53 struct request *prev_rq, struct bio *prev, struct bio *next)
55 struct bio_vec pb, nb;
57 if (!bio_has_data(prev) || !queue_virt_boundary(q))
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.
66 bio_get_first_bvec(prev_rq->bio, &pb);
68 bio_get_first_bvec(prev, &pb);
69 if (pb.bv_offset & queue_virt_boundary(q))
73 * We don't need to worry about the situation that the merged segment
74 * ends in unaligned virt boundary:
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
81 bio_get_last_bvec(prev, &pb);
82 bio_get_first_bvec(next, &nb);
83 if (biovec_phys_mergeable(q, &pb, &nb))
85 return __bvec_gap_to_prev(q, &pb, nb.bv_offset);
88 static inline bool req_gap_back_merge(struct request *req, struct bio *bio)
90 return bio_will_gap(req->q, req, req->biotail, bio);
93 static inline bool req_gap_front_merge(struct request *req, struct bio *bio)
95 return bio_will_gap(req->q, NULL, bio, req->bio);
98 static struct bio *blk_bio_discard_split(struct request_queue *q,
103 unsigned int max_discard_sectors, granularity;
106 unsigned split_sectors;
110 /* Zero-sector (unknown) and one-sector granularities are the same. */
111 granularity = max(q->limits.discard_granularity >> 9, 1U);
113 max_discard_sectors = min(q->limits.max_discard_sectors,
114 bio_allowed_max_sectors(q));
115 max_discard_sectors -= max_discard_sectors % granularity;
117 if (unlikely(!max_discard_sectors)) {
122 if (bio_sectors(bio) <= max_discard_sectors)
125 split_sectors = max_discard_sectors;
128 * If the next starting sector would be misaligned, stop the discard at
129 * the previous aligned sector.
131 alignment = (q->limits.discard_alignment >> 9) % granularity;
133 tmp = bio->bi_iter.bi_sector + split_sectors - alignment;
134 tmp = sector_div(tmp, granularity);
136 if (split_sectors > tmp)
137 split_sectors -= tmp;
139 return bio_split(bio, split_sectors, GFP_NOIO, bs);
142 static struct bio *blk_bio_write_zeroes_split(struct request_queue *q,
143 struct bio *bio, struct bio_set *bs, unsigned *nsegs)
147 if (!q->limits.max_write_zeroes_sectors)
150 if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors)
153 return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs);
157 * Return the maximum number of sectors from the start of a bio that may be
158 * submitted as a single request to a block device. If enough sectors remain,
159 * align the end to the physical block size. Otherwise align the end to the
160 * logical block size. This approach minimizes the number of non-aligned
161 * requests that are submitted to a block device if the start of a bio is not
162 * aligned to a physical block boundary.
164 static inline unsigned get_max_io_size(struct request_queue *q,
167 unsigned pbs = queue_physical_block_size(q) >> SECTOR_SHIFT;
168 unsigned lbs = queue_logical_block_size(q) >> SECTOR_SHIFT;
169 unsigned max_sectors = queue_max_sectors(q), start, end;
171 if (q->limits.chunk_sectors) {
172 max_sectors = min(max_sectors,
173 blk_chunk_sectors_left(bio->bi_iter.bi_sector,
174 q->limits.chunk_sectors));
177 start = bio->bi_iter.bi_sector & (pbs - 1);
178 end = (start + max_sectors) & ~(pbs - 1);
181 return max_sectors & ~(lbs - 1);
184 static inline unsigned get_max_segment_size(const struct request_queue *q,
185 struct page *start_page,
186 unsigned long offset)
188 unsigned long mask = queue_segment_boundary(q);
190 offset = mask & (page_to_phys(start_page) + offset);
193 * overflow may be triggered in case of zero page physical address
194 * on 32bit arch, use queue's max segment size when that happens.
196 return min_not_zero(mask - offset + 1,
197 (unsigned long)queue_max_segment_size(q));
201 * bvec_split_segs - verify whether or not a bvec should be split in the middle
202 * @q: [in] request queue associated with the bio associated with @bv
203 * @bv: [in] bvec to examine
204 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
205 * by the number of segments from @bv that may be appended to that
206 * bio without exceeding @max_segs
207 * @bytes: [in,out] Number of bytes in the bio being built. Incremented
208 * by the number of bytes from @bv that may be appended to that
209 * bio without exceeding @max_bytes
210 * @max_segs: [in] upper bound for *@nsegs
211 * @max_bytes: [in] upper bound for *@bytes
213 * When splitting a bio, it can happen that a bvec is encountered that is too
214 * big to fit in a single segment and hence that it has to be split in the
215 * middle. This function verifies whether or not that should happen. The value
216 * %true is returned if and only if appending the entire @bv to a bio with
217 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
220 static bool bvec_split_segs(const struct request_queue *q,
221 const struct bio_vec *bv, unsigned *nsegs,
222 unsigned *bytes, unsigned max_segs,
225 unsigned max_len = min(max_bytes, UINT_MAX) - *bytes;
226 unsigned len = min(bv->bv_len, max_len);
227 unsigned total_len = 0;
228 unsigned seg_size = 0;
230 while (len && *nsegs < max_segs) {
231 seg_size = get_max_segment_size(q, bv->bv_page,
232 bv->bv_offset + total_len);
233 seg_size = min(seg_size, len);
236 total_len += seg_size;
239 if ((bv->bv_offset + total_len) & queue_virt_boundary(q))
245 /* tell the caller to split the bvec if it is too big to fit */
246 return len > 0 || bv->bv_len > max_len;
250 * blk_bio_segment_split - split a bio in two bios
251 * @q: [in] request queue pointer
252 * @bio: [in] bio to be split
253 * @bs: [in] bio set to allocate the clone from
254 * @segs: [out] number of segments in the bio with the first half of the sectors
256 * Clone @bio, update the bi_iter of the clone to represent the first sectors
257 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
258 * following is guaranteed for the cloned bio:
259 * - That it has at most get_max_io_size(@q, @bio) sectors.
260 * - That it has at most queue_max_segments(@q) segments.
262 * Except for discard requests the cloned bio will point at the bi_io_vec of
263 * the original bio. It is the responsibility of the caller to ensure that the
264 * original bio is not freed before the cloned bio. The caller is also
265 * responsible for ensuring that @bs is only destroyed after processing of the
266 * split bio has finished.
268 static struct bio *blk_bio_segment_split(struct request_queue *q,
273 struct bio_vec bv, bvprv, *bvprvp = NULL;
274 struct bvec_iter iter;
275 unsigned nsegs = 0, bytes = 0;
276 const unsigned max_bytes = get_max_io_size(q, bio) << 9;
277 const unsigned max_segs = queue_max_segments(q);
279 bio_for_each_bvec(bv, bio, iter) {
281 * If the queue doesn't support SG gaps and adding this
282 * offset would create a gap, disallow it.
284 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
287 if (nsegs < max_segs &&
288 bytes + bv.bv_len <= max_bytes &&
289 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
292 } else if (bvec_split_segs(q, &bv, &nsegs, &bytes, max_segs,
307 * Individual bvecs might not be logical block aligned. Round down the
308 * split size so that each bio is properly block size aligned, even if
309 * we do not use the full hardware limits.
311 bytes = ALIGN_DOWN(bytes, queue_logical_block_size(q));
314 * Bio splitting may cause subtle trouble such as hang when doing sync
315 * iopoll in direct IO routine. Given performance gain of iopoll for
316 * big IO can be trival, disable iopoll when split needed.
318 bio_clear_polled(bio);
319 return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs);
323 * __blk_queue_split - split a bio and submit the second half
324 * @q: [in] request_queue new bio is being queued at
325 * @bio: [in, out] bio to be split
326 * @nr_segs: [out] number of segments in the first bio
328 * Split a bio into two bios, chain the two bios, submit the second half and
329 * store a pointer to the first half in *@bio. If the second bio is still too
330 * big it will be split by a recursive call to this function. Since this
331 * function may allocate a new bio from q->bio_split, it is the responsibility
332 * of the caller to ensure that q->bio_split is only released after processing
333 * of the split bio has finished.
335 void __blk_queue_split(struct request_queue *q, struct bio **bio,
336 unsigned int *nr_segs)
338 struct bio *split = NULL;
340 switch (bio_op(*bio)) {
342 case REQ_OP_SECURE_ERASE:
343 split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs);
345 case REQ_OP_WRITE_ZEROES:
346 split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split,
350 split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs);
355 /* there isn't chance to merge the splitted bio */
356 split->bi_opf |= REQ_NOMERGE;
358 blkcg_bio_issue_init(split);
359 bio_chain(split, *bio);
360 trace_block_split(split, (*bio)->bi_iter.bi_sector);
361 submit_bio_noacct(*bio);
367 * blk_queue_split - split a bio and submit the second half
368 * @bio: [in, out] bio to be split
370 * Split a bio into two bios, chains the two bios, submit the second half and
371 * store a pointer to the first half in *@bio. Since this function may allocate
372 * a new bio from q->bio_split, it is the responsibility of the caller to ensure
373 * that q->bio_split is only released after processing of the split bio has
376 void blk_queue_split(struct bio **bio)
378 struct request_queue *q = bdev_get_queue((*bio)->bi_bdev);
379 unsigned int nr_segs;
381 if (blk_may_split(q, *bio))
382 __blk_queue_split(q, bio, &nr_segs);
384 EXPORT_SYMBOL(blk_queue_split);
386 unsigned int blk_recalc_rq_segments(struct request *rq)
388 unsigned int nr_phys_segs = 0;
389 unsigned int bytes = 0;
390 struct req_iterator iter;
396 switch (bio_op(rq->bio)) {
398 case REQ_OP_SECURE_ERASE:
399 if (queue_max_discard_segments(rq->q) > 1) {
400 struct bio *bio = rq->bio;
407 case REQ_OP_WRITE_ZEROES:
413 rq_for_each_bvec(bv, rq, iter)
414 bvec_split_segs(rq->q, &bv, &nr_phys_segs, &bytes,
419 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
420 struct scatterlist *sglist)
426 * If the driver previously mapped a shorter list, we could see a
427 * termination bit prematurely unless it fully inits the sg table
428 * on each mapping. We KNOW that there must be more entries here
429 * or the driver would be buggy, so force clear the termination bit
430 * to avoid doing a full sg_init_table() in drivers for each command.
436 static unsigned blk_bvec_map_sg(struct request_queue *q,
437 struct bio_vec *bvec, struct scatterlist *sglist,
438 struct scatterlist **sg)
440 unsigned nbytes = bvec->bv_len;
441 unsigned nsegs = 0, total = 0;
444 unsigned offset = bvec->bv_offset + total;
445 unsigned len = min(get_max_segment_size(q, bvec->bv_page,
447 struct page *page = bvec->bv_page;
450 * Unfortunately a fair number of drivers barf on scatterlists
451 * that have an offset larger than PAGE_SIZE, despite other
452 * subsystems dealing with that invariant just fine. For now
453 * stick to the legacy format where we never present those from
454 * the block layer, but the code below should be removed once
455 * these offenders (mostly MMC/SD drivers) are fixed.
457 page += (offset >> PAGE_SHIFT);
458 offset &= ~PAGE_MASK;
460 *sg = blk_next_sg(sg, sglist);
461 sg_set_page(*sg, page, len, offset);
471 static inline int __blk_bvec_map_sg(struct bio_vec bv,
472 struct scatterlist *sglist, struct scatterlist **sg)
474 *sg = blk_next_sg(sg, sglist);
475 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
479 /* only try to merge bvecs into one sg if they are from two bios */
481 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
482 struct bio_vec *bvprv, struct scatterlist **sg)
485 int nbytes = bvec->bv_len;
490 if ((*sg)->length + nbytes > queue_max_segment_size(q))
493 if (!biovec_phys_mergeable(q, bvprv, bvec))
496 (*sg)->length += nbytes;
501 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
502 struct scatterlist *sglist,
503 struct scatterlist **sg)
505 struct bio_vec bvec, bvprv = { NULL };
506 struct bvec_iter iter;
508 bool new_bio = false;
511 bio_for_each_bvec(bvec, bio, iter) {
513 * Only try to merge bvecs from two bios given we
514 * have done bio internal merge when adding pages
518 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
521 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
522 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
524 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
528 if (likely(bio->bi_iter.bi_size)) {
538 * map a request to scatterlist, return number of sg entries setup. Caller
539 * must make sure sg can hold rq->nr_phys_segments entries
541 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
542 struct scatterlist *sglist, struct scatterlist **last_sg)
546 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
547 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
549 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
552 sg_mark_end(*last_sg);
555 * Something must have been wrong if the figured number of
556 * segment is bigger than number of req's physical segments
558 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
562 EXPORT_SYMBOL(__blk_rq_map_sg);
564 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
566 if (req_op(rq) == REQ_OP_DISCARD)
567 return queue_max_discard_segments(rq->q);
568 return queue_max_segments(rq->q);
571 static inline unsigned int blk_rq_get_max_sectors(struct request *rq,
574 struct request_queue *q = rq->q;
575 unsigned int max_sectors;
577 if (blk_rq_is_passthrough(rq))
578 return q->limits.max_hw_sectors;
580 max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
581 if (!q->limits.chunk_sectors ||
582 req_op(rq) == REQ_OP_DISCARD ||
583 req_op(rq) == REQ_OP_SECURE_ERASE)
585 return min(max_sectors,
586 blk_chunk_sectors_left(offset, q->limits.chunk_sectors));
589 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
590 unsigned int nr_phys_segs)
592 if (!blk_cgroup_mergeable(req, bio))
595 if (blk_integrity_merge_bio(req->q, req, bio) == false)
598 /* discard request merge won't add new segment */
599 if (req_op(req) == REQ_OP_DISCARD)
602 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
606 * This will form the start of a new hw segment. Bump both
609 req->nr_phys_segments += nr_phys_segs;
613 req_set_nomerge(req->q, req);
617 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
619 if (req_gap_back_merge(req, bio))
621 if (blk_integrity_rq(req) &&
622 integrity_req_gap_back_merge(req, bio))
624 if (!bio_crypt_ctx_back_mergeable(req, bio))
626 if (blk_rq_sectors(req) + bio_sectors(bio) >
627 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
628 req_set_nomerge(req->q, req);
632 return ll_new_hw_segment(req, bio, nr_segs);
635 static int ll_front_merge_fn(struct request *req, struct bio *bio,
636 unsigned int nr_segs)
638 if (req_gap_front_merge(req, bio))
640 if (blk_integrity_rq(req) &&
641 integrity_req_gap_front_merge(req, bio))
643 if (!bio_crypt_ctx_front_mergeable(req, bio))
645 if (blk_rq_sectors(req) + bio_sectors(bio) >
646 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
647 req_set_nomerge(req->q, req);
651 return ll_new_hw_segment(req, bio, nr_segs);
654 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
655 struct request *next)
657 unsigned short segments = blk_rq_nr_discard_segments(req);
659 if (segments >= queue_max_discard_segments(q))
661 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
662 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
665 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
668 req_set_nomerge(q, req);
672 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
673 struct request *next)
675 int total_phys_segments;
677 if (req_gap_back_merge(req, next->bio))
681 * Will it become too large?
683 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
684 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
687 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
688 if (total_phys_segments > blk_rq_get_max_segments(req))
691 if (!blk_cgroup_mergeable(req, next->bio))
694 if (blk_integrity_merge_rq(q, req, next) == false)
697 if (!bio_crypt_ctx_merge_rq(req, next))
701 req->nr_phys_segments = total_phys_segments;
706 * blk_rq_set_mixed_merge - mark a request as mixed merge
707 * @rq: request to mark as mixed merge
710 * @rq is about to be mixed merged. Make sure the attributes
711 * which can be mixed are set in each bio and mark @rq as mixed
714 void blk_rq_set_mixed_merge(struct request *rq)
716 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
719 if (rq->rq_flags & RQF_MIXED_MERGE)
723 * @rq will no longer represent mixable attributes for all the
724 * contained bios. It will just track those of the first one.
725 * Distributes the attributs to each bio.
727 for (bio = rq->bio; bio; bio = bio->bi_next) {
728 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
729 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
732 rq->rq_flags |= RQF_MIXED_MERGE;
735 static void blk_account_io_merge_request(struct request *req)
737 if (blk_do_io_stat(req)) {
739 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
744 static enum elv_merge blk_try_req_merge(struct request *req,
745 struct request *next)
747 if (blk_discard_mergable(req))
748 return ELEVATOR_DISCARD_MERGE;
749 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
750 return ELEVATOR_BACK_MERGE;
752 return ELEVATOR_NO_MERGE;
756 * For non-mq, this has to be called with the request spinlock acquired.
757 * For mq with scheduling, the appropriate queue wide lock should be held.
759 static struct request *attempt_merge(struct request_queue *q,
760 struct request *req, struct request *next)
762 if (!rq_mergeable(req) || !rq_mergeable(next))
765 if (req_op(req) != req_op(next))
768 if (rq_data_dir(req) != rq_data_dir(next))
771 if (req->ioprio != next->ioprio)
775 * If we are allowed to merge, then append bio list
776 * from next to rq and release next. merge_requests_fn
777 * will have updated segment counts, update sector
778 * counts here. Handle DISCARDs separately, as they
779 * have separate settings.
782 switch (blk_try_req_merge(req, next)) {
783 case ELEVATOR_DISCARD_MERGE:
784 if (!req_attempt_discard_merge(q, req, next))
787 case ELEVATOR_BACK_MERGE:
788 if (!ll_merge_requests_fn(q, req, next))
796 * If failfast settings disagree or any of the two is already
797 * a mixed merge, mark both as mixed before proceeding. This
798 * makes sure that all involved bios have mixable attributes
801 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
802 (req->cmd_flags & REQ_FAILFAST_MASK) !=
803 (next->cmd_flags & REQ_FAILFAST_MASK)) {
804 blk_rq_set_mixed_merge(req);
805 blk_rq_set_mixed_merge(next);
809 * At this point we have either done a back merge or front merge. We
810 * need the smaller start_time_ns of the merged requests to be the
811 * current request for accounting purposes.
813 if (next->start_time_ns < req->start_time_ns)
814 req->start_time_ns = next->start_time_ns;
816 req->biotail->bi_next = next->bio;
817 req->biotail = next->biotail;
819 req->__data_len += blk_rq_bytes(next);
821 if (!blk_discard_mergable(req))
822 elv_merge_requests(q, req, next);
825 * 'next' is going away, so update stats accordingly
827 blk_account_io_merge_request(next);
829 trace_block_rq_merge(next);
832 * ownership of bio passed from next to req, return 'next' for
839 static struct request *attempt_back_merge(struct request_queue *q,
842 struct request *next = elv_latter_request(q, rq);
845 return attempt_merge(q, rq, next);
850 static struct request *attempt_front_merge(struct request_queue *q,
853 struct request *prev = elv_former_request(q, rq);
856 return attempt_merge(q, prev, rq);
862 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
863 * otherwise. The caller is responsible for freeing 'next' if the merge
866 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
867 struct request *next)
869 return attempt_merge(q, rq, next);
872 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
874 if (!rq_mergeable(rq) || !bio_mergeable(bio))
877 if (req_op(rq) != bio_op(bio))
880 /* different data direction or already started, don't merge */
881 if (bio_data_dir(bio) != rq_data_dir(rq))
884 /* don't merge across cgroup boundaries */
885 if (!blk_cgroup_mergeable(rq, bio))
888 /* only merge integrity protected bio into ditto rq */
889 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
892 /* Only merge if the crypt contexts are compatible */
893 if (!bio_crypt_rq_ctx_compatible(rq, bio))
896 if (rq->ioprio != bio_prio(bio))
902 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
904 if (blk_discard_mergable(rq))
905 return ELEVATOR_DISCARD_MERGE;
906 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
907 return ELEVATOR_BACK_MERGE;
908 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
909 return ELEVATOR_FRONT_MERGE;
910 return ELEVATOR_NO_MERGE;
913 static void blk_account_io_merge_bio(struct request *req)
915 if (!blk_do_io_stat(req))
919 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
923 enum bio_merge_status {
929 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
930 struct bio *bio, unsigned int nr_segs)
932 const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK;
934 if (!ll_back_merge_fn(req, bio, nr_segs))
935 return BIO_MERGE_FAILED;
937 trace_block_bio_backmerge(bio);
938 rq_qos_merge(req->q, req, bio);
940 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
941 blk_rq_set_mixed_merge(req);
943 req->biotail->bi_next = bio;
945 req->__data_len += bio->bi_iter.bi_size;
947 bio_crypt_free_ctx(bio);
949 blk_account_io_merge_bio(req);
953 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
954 struct bio *bio, unsigned int nr_segs)
956 const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK;
958 if (!ll_front_merge_fn(req, bio, nr_segs))
959 return BIO_MERGE_FAILED;
961 trace_block_bio_frontmerge(bio);
962 rq_qos_merge(req->q, req, bio);
964 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
965 blk_rq_set_mixed_merge(req);
967 bio->bi_next = req->bio;
970 req->__sector = bio->bi_iter.bi_sector;
971 req->__data_len += bio->bi_iter.bi_size;
973 bio_crypt_do_front_merge(req, bio);
975 blk_account_io_merge_bio(req);
979 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
980 struct request *req, struct bio *bio)
982 unsigned short segments = blk_rq_nr_discard_segments(req);
984 if (segments >= queue_max_discard_segments(q))
986 if (blk_rq_sectors(req) + bio_sectors(bio) >
987 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
990 rq_qos_merge(q, req, bio);
992 req->biotail->bi_next = bio;
994 req->__data_len += bio->bi_iter.bi_size;
995 req->nr_phys_segments = segments + 1;
997 blk_account_io_merge_bio(req);
1000 req_set_nomerge(q, req);
1001 return BIO_MERGE_FAILED;
1004 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1007 unsigned int nr_segs,
1008 bool sched_allow_merge)
1010 if (!blk_rq_merge_ok(rq, bio))
1011 return BIO_MERGE_NONE;
1013 switch (blk_try_merge(rq, bio)) {
1014 case ELEVATOR_BACK_MERGE:
1015 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1016 return bio_attempt_back_merge(rq, bio, nr_segs);
1018 case ELEVATOR_FRONT_MERGE:
1019 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1020 return bio_attempt_front_merge(rq, bio, nr_segs);
1022 case ELEVATOR_DISCARD_MERGE:
1023 return bio_attempt_discard_merge(q, rq, bio);
1025 return BIO_MERGE_NONE;
1028 return BIO_MERGE_FAILED;
1032 * blk_attempt_plug_merge - try to merge with %current's plugged list
1033 * @q: request_queue new bio is being queued at
1034 * @bio: new bio being queued
1035 * @nr_segs: number of segments in @bio
1036 * from the passed in @q already in the plug list
1038 * Determine whether @bio being queued on @q can be merged with the previous
1039 * request on %current's plugged list. Returns %true if merge was successful,
1042 * Plugging coalesces IOs from the same issuer for the same purpose without
1043 * going through @q->queue_lock. As such it's more of an issuing mechanism
1044 * than scheduling, and the request, while may have elvpriv data, is not
1045 * added on the elevator at this point. In addition, we don't have
1046 * reliable access to the elevator outside queue lock. Only check basic
1047 * merging parameters without querying the elevator.
1049 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1051 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1052 unsigned int nr_segs)
1054 struct blk_plug *plug;
1057 plug = blk_mq_plug(bio);
1058 if (!plug || rq_list_empty(plug->mq_list))
1061 rq_list_for_each(&plug->mq_list, rq) {
1063 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1070 * Only keep iterating plug list for merges if we have multiple
1073 if (!plug->multiple_queues)
1080 * Iterate list of requests and see if we can merge this bio with any
1083 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1084 struct bio *bio, unsigned int nr_segs)
1089 list_for_each_entry_reverse(rq, list, queuelist) {
1093 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1094 case BIO_MERGE_NONE:
1098 case BIO_MERGE_FAILED:
1106 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1108 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1109 unsigned int nr_segs, struct request **merged_request)
1113 switch (elv_merge(q, &rq, bio)) {
1114 case ELEVATOR_BACK_MERGE:
1115 if (!blk_mq_sched_allow_merge(q, rq, bio))
1117 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1119 *merged_request = attempt_back_merge(q, rq);
1120 if (!*merged_request)
1121 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1123 case ELEVATOR_FRONT_MERGE:
1124 if (!blk_mq_sched_allow_merge(q, rq, bio))
1126 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1128 *merged_request = attempt_front_merge(q, rq);
1129 if (!*merged_request)
1130 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1132 case ELEVATOR_DISCARD_MERGE:
1133 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1138 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);