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/scatterlist.h>
10 #include <linux/blk-cgroup.h>
12 #include <trace/events/block.h>
15 #include "blk-rq-qos.h"
17 static inline bool bio_will_gap(struct request_queue *q,
18 struct request *prev_rq, struct bio *prev, struct bio *next)
20 struct bio_vec pb, nb;
22 if (!bio_has_data(prev) || !queue_virt_boundary(q))
26 * Don't merge if the 1st bio starts with non-zero offset, otherwise it
27 * is quite difficult to respect the sg gap limit. We work hard to
28 * merge a huge number of small single bios in case of mkfs.
31 bio_get_first_bvec(prev_rq->bio, &pb);
33 bio_get_first_bvec(prev, &pb);
34 if (pb.bv_offset & queue_virt_boundary(q))
38 * We don't need to worry about the situation that the merged segment
39 * ends in unaligned virt boundary:
41 * - if 'pb' ends aligned, the merged segment ends aligned
42 * - if 'pb' ends unaligned, the next bio must include
43 * one single bvec of 'nb', otherwise the 'nb' can't
46 bio_get_last_bvec(prev, &pb);
47 bio_get_first_bvec(next, &nb);
48 if (biovec_phys_mergeable(q, &pb, &nb))
50 return __bvec_gap_to_prev(q, &pb, nb.bv_offset);
53 static inline bool req_gap_back_merge(struct request *req, struct bio *bio)
55 return bio_will_gap(req->q, req, req->biotail, bio);
58 static inline bool req_gap_front_merge(struct request *req, struct bio *bio)
60 return bio_will_gap(req->q, NULL, bio, req->bio);
63 static struct bio *blk_bio_discard_split(struct request_queue *q,
68 unsigned int max_discard_sectors, granularity;
71 unsigned split_sectors;
75 /* Zero-sector (unknown) and one-sector granularities are the same. */
76 granularity = max(q->limits.discard_granularity >> 9, 1U);
78 max_discard_sectors = min(q->limits.max_discard_sectors,
79 bio_allowed_max_sectors(q));
80 max_discard_sectors -= max_discard_sectors % granularity;
82 if (unlikely(!max_discard_sectors)) {
87 if (bio_sectors(bio) <= max_discard_sectors)
90 split_sectors = max_discard_sectors;
93 * If the next starting sector would be misaligned, stop the discard at
94 * the previous aligned sector.
96 alignment = (q->limits.discard_alignment >> 9) % granularity;
98 tmp = bio->bi_iter.bi_sector + split_sectors - alignment;
99 tmp = sector_div(tmp, granularity);
101 if (split_sectors > tmp)
102 split_sectors -= tmp;
104 return bio_split(bio, split_sectors, GFP_NOIO, bs);
107 static struct bio *blk_bio_write_zeroes_split(struct request_queue *q,
108 struct bio *bio, struct bio_set *bs, unsigned *nsegs)
112 if (!q->limits.max_write_zeroes_sectors)
115 if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors)
118 return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs);
121 static struct bio *blk_bio_write_same_split(struct request_queue *q,
128 if (!q->limits.max_write_same_sectors)
131 if (bio_sectors(bio) <= q->limits.max_write_same_sectors)
134 return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs);
138 * Return the maximum number of sectors from the start of a bio that may be
139 * submitted as a single request to a block device. If enough sectors remain,
140 * align the end to the physical block size. Otherwise align the end to the
141 * logical block size. This approach minimizes the number of non-aligned
142 * requests that are submitted to a block device if the start of a bio is not
143 * aligned to a physical block boundary.
145 static inline unsigned get_max_io_size(struct request_queue *q,
148 unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector, 0);
149 unsigned max_sectors = sectors;
150 unsigned pbs = queue_physical_block_size(q) >> SECTOR_SHIFT;
151 unsigned lbs = queue_logical_block_size(q) >> SECTOR_SHIFT;
152 unsigned start_offset = bio->bi_iter.bi_sector & (pbs - 1);
154 max_sectors += start_offset;
155 max_sectors &= ~(pbs - 1);
156 if (max_sectors > start_offset)
157 return max_sectors - start_offset;
159 return sectors & ~(lbs - 1);
162 static inline unsigned get_max_segment_size(const struct request_queue *q,
163 struct page *start_page,
164 unsigned long offset)
166 unsigned long mask = queue_segment_boundary(q);
168 offset = mask & (page_to_phys(start_page) + offset);
171 * overflow may be triggered in case of zero page physical address
172 * on 32bit arch, use queue's max segment size when that happens.
174 return min_not_zero(mask - offset + 1,
175 (unsigned long)queue_max_segment_size(q));
179 * bvec_split_segs - verify whether or not a bvec should be split in the middle
180 * @q: [in] request queue associated with the bio associated with @bv
181 * @bv: [in] bvec to examine
182 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
183 * by the number of segments from @bv that may be appended to that
184 * bio without exceeding @max_segs
185 * @sectors: [in,out] Number of sectors in the bio being built. Incremented
186 * by the number of sectors from @bv that may be appended to that
187 * bio without exceeding @max_sectors
188 * @max_segs: [in] upper bound for *@nsegs
189 * @max_sectors: [in] upper bound for *@sectors
191 * When splitting a bio, it can happen that a bvec is encountered that is too
192 * big to fit in a single segment and hence that it has to be split in the
193 * middle. This function verifies whether or not that should happen. The value
194 * %true is returned if and only if appending the entire @bv to a bio with
195 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
198 static bool bvec_split_segs(const struct request_queue *q,
199 const struct bio_vec *bv, unsigned *nsegs,
200 unsigned *sectors, unsigned max_segs,
201 unsigned max_sectors)
203 unsigned max_len = (min(max_sectors, UINT_MAX >> 9) - *sectors) << 9;
204 unsigned len = min(bv->bv_len, max_len);
205 unsigned total_len = 0;
206 unsigned seg_size = 0;
208 while (len && *nsegs < max_segs) {
209 seg_size = get_max_segment_size(q, bv->bv_page,
210 bv->bv_offset + total_len);
211 seg_size = min(seg_size, len);
214 total_len += seg_size;
217 if ((bv->bv_offset + total_len) & queue_virt_boundary(q))
221 *sectors += total_len >> 9;
223 /* tell the caller to split the bvec if it is too big to fit */
224 return len > 0 || bv->bv_len > max_len;
228 * blk_bio_segment_split - split a bio in two bios
229 * @q: [in] request queue pointer
230 * @bio: [in] bio to be split
231 * @bs: [in] bio set to allocate the clone from
232 * @segs: [out] number of segments in the bio with the first half of the sectors
234 * Clone @bio, update the bi_iter of the clone to represent the first sectors
235 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
236 * following is guaranteed for the cloned bio:
237 * - That it has at most get_max_io_size(@q, @bio) sectors.
238 * - That it has at most queue_max_segments(@q) segments.
240 * Except for discard requests the cloned bio will point at the bi_io_vec of
241 * the original bio. It is the responsibility of the caller to ensure that the
242 * original bio is not freed before the cloned bio. The caller is also
243 * responsible for ensuring that @bs is only destroyed after processing of the
244 * split bio has finished.
246 static struct bio *blk_bio_segment_split(struct request_queue *q,
251 struct bio_vec bv, bvprv, *bvprvp = NULL;
252 struct bvec_iter iter;
253 unsigned nsegs = 0, sectors = 0;
254 const unsigned max_sectors = get_max_io_size(q, bio);
255 const unsigned max_segs = queue_max_segments(q);
257 bio_for_each_bvec(bv, bio, iter) {
259 * If the queue doesn't support SG gaps and adding this
260 * offset would create a gap, disallow it.
262 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
265 if (nsegs < max_segs &&
266 sectors + (bv.bv_len >> 9) <= max_sectors &&
267 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
269 sectors += bv.bv_len >> 9;
270 } else if (bvec_split_segs(q, &bv, &nsegs, §ors, max_segs,
283 * We can't sanely support splitting for a REQ_NOWAIT bio. End it
284 * with EAGAIN if splitting is required and return an error pointer.
286 if (bio->bi_opf & REQ_NOWAIT) {
287 bio->bi_status = BLK_STS_AGAIN;
289 return ERR_PTR(-EAGAIN);
295 * Bio splitting may cause subtle trouble such as hang when doing sync
296 * iopoll in direct IO routine. Given performance gain of iopoll for
297 * big IO can be trival, disable iopoll when split needed.
299 bio_clear_hipri(bio);
301 return bio_split(bio, sectors, GFP_NOIO, bs);
305 * __blk_queue_split - split a bio and submit the second half
306 * @bio: [in, out] bio to be split
307 * @nr_segs: [out] number of segments in the first bio
309 * Split a bio into two bios, chain the two bios, submit the second half and
310 * store a pointer to the first half in *@bio. If the second bio is still too
311 * big it will be split by a recursive call to this function. Since this
312 * function may allocate a new bio from q->bio_split, it is the responsibility
313 * of the caller to ensure that q->bio_split is only released after processing
314 * of the split bio has finished.
316 void __blk_queue_split(struct bio **bio, unsigned int *nr_segs)
318 struct request_queue *q = (*bio)->bi_bdev->bd_disk->queue;
319 struct bio *split = NULL;
321 switch (bio_op(*bio)) {
323 case REQ_OP_SECURE_ERASE:
324 split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs);
326 case REQ_OP_WRITE_ZEROES:
327 split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split,
330 case REQ_OP_WRITE_SAME:
331 split = blk_bio_write_same_split(q, *bio, &q->bio_split,
336 * All drivers must accept single-segments bios that are <=
337 * PAGE_SIZE. This is a quick and dirty check that relies on
338 * the fact that bi_io_vec[0] is always valid if a bio has data.
339 * The check might lead to occasional false negatives when bios
340 * are cloned, but compared to the performance impact of cloned
341 * bios themselves the loop below doesn't matter anyway.
343 if (!q->limits.chunk_sectors &&
344 (*bio)->bi_vcnt == 1 &&
345 ((*bio)->bi_io_vec[0].bv_len +
346 (*bio)->bi_io_vec[0].bv_offset) <= PAGE_SIZE) {
350 split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs);
357 /* there isn't chance to merge the split bio */
358 split->bi_opf |= REQ_NOMERGE;
360 bio_chain(split, *bio);
361 trace_block_split(split, (*bio)->bi_iter.bi_sector);
362 submit_bio_noacct(*bio);
365 blk_throtl_charge_bio_split(*bio);
370 * blk_queue_split - split a bio and submit the second half
371 * @bio: [in, out] bio to be split
373 * Split a bio into two bios, chains the two bios, submit the second half and
374 * store a pointer to the first half in *@bio. Since this function may allocate
375 * a new bio from q->bio_split, it is the responsibility of the caller to ensure
376 * that q->bio_split is only released after processing of the split bio has
379 void blk_queue_split(struct bio **bio)
381 unsigned int nr_segs;
383 __blk_queue_split(bio, &nr_segs);
385 EXPORT_SYMBOL(blk_queue_split);
387 unsigned int blk_recalc_rq_segments(struct request *rq)
389 unsigned int nr_phys_segs = 0;
390 unsigned int nr_sectors = 0;
391 struct req_iterator iter;
397 switch (bio_op(rq->bio)) {
399 case REQ_OP_SECURE_ERASE:
400 if (queue_max_discard_segments(rq->q) > 1) {
401 struct bio *bio = rq->bio;
408 case REQ_OP_WRITE_ZEROES:
410 case REQ_OP_WRITE_SAME:
414 rq_for_each_bvec(bv, rq, iter)
415 bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors,
420 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
421 struct scatterlist *sglist)
427 * If the driver previously mapped a shorter list, we could see a
428 * termination bit prematurely unless it fully inits the sg table
429 * on each mapping. We KNOW that there must be more entries here
430 * or the driver would be buggy, so force clear the termination bit
431 * to avoid doing a full sg_init_table() in drivers for each command.
437 static unsigned blk_bvec_map_sg(struct request_queue *q,
438 struct bio_vec *bvec, struct scatterlist *sglist,
439 struct scatterlist **sg)
441 unsigned nbytes = bvec->bv_len;
442 unsigned nsegs = 0, total = 0;
445 unsigned offset = bvec->bv_offset + total;
446 unsigned len = min(get_max_segment_size(q, bvec->bv_page,
448 struct page *page = bvec->bv_page;
451 * Unfortunately a fair number of drivers barf on scatterlists
452 * that have an offset larger than PAGE_SIZE, despite other
453 * subsystems dealing with that invariant just fine. For now
454 * stick to the legacy format where we never present those from
455 * the block layer, but the code below should be removed once
456 * these offenders (mostly MMC/SD drivers) are fixed.
458 page += (offset >> PAGE_SHIFT);
459 offset &= ~PAGE_MASK;
461 *sg = blk_next_sg(sg, sglist);
462 sg_set_page(*sg, page, len, offset);
472 static inline int __blk_bvec_map_sg(struct bio_vec bv,
473 struct scatterlist *sglist, struct scatterlist **sg)
475 *sg = blk_next_sg(sg, sglist);
476 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
480 /* only try to merge bvecs into one sg if they are from two bios */
482 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
483 struct bio_vec *bvprv, struct scatterlist **sg)
486 int nbytes = bvec->bv_len;
491 if ((*sg)->length + nbytes > queue_max_segment_size(q))
494 if (!biovec_phys_mergeable(q, bvprv, bvec))
497 (*sg)->length += nbytes;
502 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
503 struct scatterlist *sglist,
504 struct scatterlist **sg)
506 struct bio_vec bvec, bvprv = { NULL };
507 struct bvec_iter iter;
509 bool new_bio = false;
512 bio_for_each_bvec(bvec, bio, iter) {
514 * Only try to merge bvecs from two bios given we
515 * have done bio internal merge when adding pages
519 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
522 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
523 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
525 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
529 if (likely(bio->bi_iter.bi_size)) {
539 * map a request to scatterlist, return number of sg entries setup. Caller
540 * must make sure sg can hold rq->nr_phys_segments entries
542 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
543 struct scatterlist *sglist, struct scatterlist **last_sg)
547 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
548 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
549 else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME)
550 nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, last_sg);
552 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
555 sg_mark_end(*last_sg);
558 * Something must have been wrong if the figured number of
559 * segment is bigger than number of req's physical segments
561 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
565 EXPORT_SYMBOL(__blk_rq_map_sg);
567 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
569 if (req_op(rq) == REQ_OP_DISCARD)
570 return queue_max_discard_segments(rq->q);
571 return queue_max_segments(rq->q);
574 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
575 unsigned int nr_phys_segs)
577 if (!blk_cgroup_mergeable(req, bio))
580 if (blk_integrity_merge_bio(req->q, req, bio) == false)
583 /* discard request merge won't add new segment */
584 if (req_op(req) == REQ_OP_DISCARD)
587 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
591 * This will form the start of a new hw segment. Bump both
594 req->nr_phys_segments += nr_phys_segs;
598 req_set_nomerge(req->q, req);
602 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
604 if (req_gap_back_merge(req, bio))
606 if (blk_integrity_rq(req) &&
607 integrity_req_gap_back_merge(req, bio))
609 if (!bio_crypt_ctx_back_mergeable(req, bio))
611 if (blk_rq_sectors(req) + bio_sectors(bio) >
612 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
613 req_set_nomerge(req->q, req);
617 return ll_new_hw_segment(req, bio, nr_segs);
620 static int ll_front_merge_fn(struct request *req, struct bio *bio,
621 unsigned int nr_segs)
623 if (req_gap_front_merge(req, bio))
625 if (blk_integrity_rq(req) &&
626 integrity_req_gap_front_merge(req, bio))
628 if (!bio_crypt_ctx_front_mergeable(req, bio))
630 if (blk_rq_sectors(req) + bio_sectors(bio) >
631 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
632 req_set_nomerge(req->q, req);
636 return ll_new_hw_segment(req, bio, nr_segs);
639 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
640 struct request *next)
642 unsigned short segments = blk_rq_nr_discard_segments(req);
644 if (segments >= queue_max_discard_segments(q))
646 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
647 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
650 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
653 req_set_nomerge(q, req);
657 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
658 struct request *next)
660 int total_phys_segments;
662 if (req_gap_back_merge(req, next->bio))
666 * Will it become too large?
668 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
669 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
672 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
673 if (total_phys_segments > blk_rq_get_max_segments(req))
676 if (!blk_cgroup_mergeable(req, next->bio))
679 if (blk_integrity_merge_rq(q, req, next) == false)
682 if (!bio_crypt_ctx_merge_rq(req, next))
686 req->nr_phys_segments = total_phys_segments;
691 * blk_rq_set_mixed_merge - mark a request as mixed merge
692 * @rq: request to mark as mixed merge
695 * @rq is about to be mixed merged. Make sure the attributes
696 * which can be mixed are set in each bio and mark @rq as mixed
699 void blk_rq_set_mixed_merge(struct request *rq)
701 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
704 if (rq->rq_flags & RQF_MIXED_MERGE)
708 * @rq will no longer represent mixable attributes for all the
709 * contained bios. It will just track those of the first one.
710 * Distributes the attributs to each bio.
712 for (bio = rq->bio; bio; bio = bio->bi_next) {
713 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
714 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
717 rq->rq_flags |= RQF_MIXED_MERGE;
720 static void blk_account_io_merge_request(struct request *req)
722 if (blk_do_io_stat(req)) {
724 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
729 static enum elv_merge blk_try_req_merge(struct request *req,
730 struct request *next)
732 if (blk_discard_mergable(req))
733 return ELEVATOR_DISCARD_MERGE;
734 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
735 return ELEVATOR_BACK_MERGE;
737 return ELEVATOR_NO_MERGE;
741 * For non-mq, this has to be called with the request spinlock acquired.
742 * For mq with scheduling, the appropriate queue wide lock should be held.
744 static struct request *attempt_merge(struct request_queue *q,
745 struct request *req, struct request *next)
747 if (!rq_mergeable(req) || !rq_mergeable(next))
750 if (req_op(req) != req_op(next))
753 if (rq_data_dir(req) != rq_data_dir(next)
754 || req->rq_disk != next->rq_disk)
757 if (req_op(req) == REQ_OP_WRITE_SAME &&
758 !blk_write_same_mergeable(req->bio, next->bio))
762 * Don't allow merge of different write hints, or for a hint with
765 if (req->write_hint != next->write_hint)
768 if (req->ioprio != next->ioprio)
772 * If we are allowed to merge, then append bio list
773 * from next to rq and release next. merge_requests_fn
774 * will have updated segment counts, update sector
775 * counts here. Handle DISCARDs separately, as they
776 * have separate settings.
779 switch (blk_try_req_merge(req, next)) {
780 case ELEVATOR_DISCARD_MERGE:
781 if (!req_attempt_discard_merge(q, req, next))
784 case ELEVATOR_BACK_MERGE:
785 if (!ll_merge_requests_fn(q, req, next))
793 * If failfast settings disagree or any of the two is already
794 * a mixed merge, mark both as mixed before proceeding. This
795 * makes sure that all involved bios have mixable attributes
798 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
799 (req->cmd_flags & REQ_FAILFAST_MASK) !=
800 (next->cmd_flags & REQ_FAILFAST_MASK)) {
801 blk_rq_set_mixed_merge(req);
802 blk_rq_set_mixed_merge(next);
806 * At this point we have either done a back merge or front merge. We
807 * need the smaller start_time_ns of the merged requests to be the
808 * current request for accounting purposes.
810 if (next->start_time_ns < req->start_time_ns)
811 req->start_time_ns = next->start_time_ns;
813 req->biotail->bi_next = next->bio;
814 req->biotail = next->biotail;
816 req->__data_len += blk_rq_bytes(next);
818 if (!blk_discard_mergable(req))
819 elv_merge_requests(q, req, next);
821 blk_crypto_rq_put_keyslot(next);
824 * 'next' is going away, so update stats accordingly
826 blk_account_io_merge_request(next);
828 trace_block_rq_merge(next);
831 * ownership of bio passed from next to req, return 'next' for
838 static struct request *attempt_back_merge(struct request_queue *q,
841 struct request *next = elv_latter_request(q, rq);
844 return attempt_merge(q, rq, next);
849 static struct request *attempt_front_merge(struct request_queue *q,
852 struct request *prev = elv_former_request(q, rq);
855 return attempt_merge(q, prev, rq);
861 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
862 * otherwise. The caller is responsible for freeing 'next' if the merge
865 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
866 struct request *next)
868 return attempt_merge(q, rq, next);
871 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
873 if (!rq_mergeable(rq) || !bio_mergeable(bio))
876 if (req_op(rq) != bio_op(bio))
879 /* different data direction or already started, don't merge */
880 if (bio_data_dir(bio) != rq_data_dir(rq))
883 /* must be same device */
884 if (rq->rq_disk != bio->bi_bdev->bd_disk)
887 /* don't merge across cgroup boundaries */
888 if (!blk_cgroup_mergeable(rq, bio))
891 /* only merge integrity protected bio into ditto rq */
892 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
895 /* Only merge if the crypt contexts are compatible */
896 if (!bio_crypt_rq_ctx_compatible(rq, bio))
899 /* must be using the same buffer */
900 if (req_op(rq) == REQ_OP_WRITE_SAME &&
901 !blk_write_same_mergeable(rq->bio, bio))
905 * Don't allow merge of different write hints, or for a hint with
908 if (rq->write_hint != bio->bi_write_hint)
911 if (rq->ioprio != bio_prio(bio))
917 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
919 if (blk_discard_mergable(rq))
920 return ELEVATOR_DISCARD_MERGE;
921 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
922 return ELEVATOR_BACK_MERGE;
923 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
924 return ELEVATOR_FRONT_MERGE;
925 return ELEVATOR_NO_MERGE;
928 static void blk_account_io_merge_bio(struct request *req)
930 if (!blk_do_io_stat(req))
934 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
938 enum bio_merge_status {
944 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
945 struct bio *bio, unsigned int nr_segs)
947 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
949 if (!ll_back_merge_fn(req, bio, nr_segs))
950 return BIO_MERGE_FAILED;
952 trace_block_bio_backmerge(bio);
953 rq_qos_merge(req->q, req, bio);
955 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
956 blk_rq_set_mixed_merge(req);
958 req->biotail->bi_next = bio;
960 req->__data_len += bio->bi_iter.bi_size;
962 bio_crypt_free_ctx(bio);
964 blk_account_io_merge_bio(req);
968 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
969 struct bio *bio, unsigned int nr_segs)
971 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
973 if (!ll_front_merge_fn(req, bio, nr_segs))
974 return BIO_MERGE_FAILED;
976 trace_block_bio_frontmerge(bio);
977 rq_qos_merge(req->q, req, bio);
979 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
980 blk_rq_set_mixed_merge(req);
982 bio->bi_next = req->bio;
985 req->__sector = bio->bi_iter.bi_sector;
986 req->__data_len += bio->bi_iter.bi_size;
988 bio_crypt_do_front_merge(req, bio);
990 blk_account_io_merge_bio(req);
994 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
995 struct request *req, struct bio *bio)
997 unsigned short segments = blk_rq_nr_discard_segments(req);
999 if (segments >= queue_max_discard_segments(q))
1001 if (blk_rq_sectors(req) + bio_sectors(bio) >
1002 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1005 rq_qos_merge(q, req, bio);
1007 req->biotail->bi_next = bio;
1009 req->__data_len += bio->bi_iter.bi_size;
1010 req->nr_phys_segments = segments + 1;
1012 blk_account_io_merge_bio(req);
1013 return BIO_MERGE_OK;
1015 req_set_nomerge(q, req);
1016 return BIO_MERGE_FAILED;
1019 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1022 unsigned int nr_segs,
1023 bool sched_allow_merge)
1025 if (!blk_rq_merge_ok(rq, bio))
1026 return BIO_MERGE_NONE;
1028 switch (blk_try_merge(rq, bio)) {
1029 case ELEVATOR_BACK_MERGE:
1030 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1031 return bio_attempt_back_merge(rq, bio, nr_segs);
1033 case ELEVATOR_FRONT_MERGE:
1034 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1035 return bio_attempt_front_merge(rq, bio, nr_segs);
1037 case ELEVATOR_DISCARD_MERGE:
1038 return bio_attempt_discard_merge(q, rq, bio);
1040 return BIO_MERGE_NONE;
1043 return BIO_MERGE_FAILED;
1047 * blk_attempt_plug_merge - try to merge with %current's plugged list
1048 * @q: request_queue new bio is being queued at
1049 * @bio: new bio being queued
1050 * @nr_segs: number of segments in @bio
1051 * @same_queue_rq: pointer to &struct request that gets filled in when
1052 * another request associated with @q is found on the plug list
1053 * (optional, may be %NULL)
1055 * Determine whether @bio being queued on @q can be merged with a request
1056 * on %current's plugged list. Returns %true if merge was successful,
1059 * Plugging coalesces IOs from the same issuer for the same purpose without
1060 * going through @q->queue_lock. As such it's more of an issuing mechanism
1061 * than scheduling, and the request, while may have elvpriv data, is not
1062 * added on the elevator at this point. In addition, we don't have
1063 * reliable access to the elevator outside queue lock. Only check basic
1064 * merging parameters without querying the elevator.
1066 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1068 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1069 unsigned int nr_segs, struct request **same_queue_rq)
1071 struct blk_plug *plug;
1073 struct list_head *plug_list;
1075 plug = blk_mq_plug(q, bio);
1079 plug_list = &plug->mq_list;
1081 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1082 if (rq->q == q && same_queue_rq) {
1084 * Only blk-mq multiple hardware queues case checks the
1085 * rq in the same queue, there should be only one such
1088 *same_queue_rq = rq;
1094 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1103 * Iterate list of requests and see if we can merge this bio with any
1106 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1107 struct bio *bio, unsigned int nr_segs)
1112 list_for_each_entry_reverse(rq, list, queuelist) {
1116 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1117 case BIO_MERGE_NONE:
1121 case BIO_MERGE_FAILED:
1129 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1131 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1132 unsigned int nr_segs, struct request **merged_request)
1136 switch (elv_merge(q, &rq, bio)) {
1137 case ELEVATOR_BACK_MERGE:
1138 if (!blk_mq_sched_allow_merge(q, rq, bio))
1140 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1142 *merged_request = attempt_back_merge(q, rq);
1143 if (!*merged_request)
1144 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1146 case ELEVATOR_FRONT_MERGE:
1147 if (!blk_mq_sched_allow_merge(q, rq, bio))
1149 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1151 *merged_request = attempt_front_merge(q, rq);
1152 if (!*merged_request)
1153 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1155 case ELEVATOR_DISCARD_MERGE:
1156 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1161 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);