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->limits, &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);
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.
103 static unsigned int bio_allowed_max_sectors(struct queue_limits *lim)
105 return round_down(UINT_MAX, lim->logical_block_size) >> SECTOR_SHIFT;
108 static struct bio *bio_split_discard(struct bio *bio, struct queue_limits *lim,
109 unsigned *nsegs, struct bio_set *bs)
111 unsigned int max_discard_sectors, granularity;
113 unsigned split_sectors;
117 /* Zero-sector (unknown) and one-sector granularities are the same. */
118 granularity = max(lim->discard_granularity >> 9, 1U);
120 max_discard_sectors =
121 min(lim->max_discard_sectors, bio_allowed_max_sectors(lim));
122 max_discard_sectors -= max_discard_sectors % granularity;
124 if (unlikely(!max_discard_sectors)) {
129 if (bio_sectors(bio) <= max_discard_sectors)
132 split_sectors = max_discard_sectors;
135 * If the next starting sector would be misaligned, stop the discard at
136 * the previous aligned sector.
138 tmp = bio->bi_iter.bi_sector + split_sectors -
139 ((lim->discard_alignment >> 9) % granularity);
140 tmp = sector_div(tmp, granularity);
142 if (split_sectors > tmp)
143 split_sectors -= tmp;
145 return bio_split(bio, split_sectors, GFP_NOIO, bs);
148 static struct bio *bio_split_write_zeroes(struct bio *bio,
149 struct queue_limits *lim, unsigned *nsegs, struct bio_set *bs)
152 if (!lim->max_write_zeroes_sectors)
154 if (bio_sectors(bio) <= lim->max_write_zeroes_sectors)
156 return bio_split(bio, lim->max_write_zeroes_sectors, GFP_NOIO, bs);
160 * Return the maximum number of sectors from the start of a bio that may be
161 * submitted as a single request to a block device. If enough sectors remain,
162 * align the end to the physical block size. Otherwise align the end to the
163 * logical block size. This approach minimizes the number of non-aligned
164 * requests that are submitted to a block device if the start of a bio is not
165 * aligned to a physical block boundary.
167 static inline unsigned get_max_io_size(struct bio *bio,
168 struct queue_limits *lim)
170 unsigned pbs = lim->physical_block_size >> SECTOR_SHIFT;
171 unsigned lbs = lim->logical_block_size >> SECTOR_SHIFT;
172 unsigned max_sectors = lim->max_sectors, start, end;
174 if (lim->chunk_sectors) {
175 max_sectors = min(max_sectors,
176 blk_chunk_sectors_left(bio->bi_iter.bi_sector,
177 lim->chunk_sectors));
180 start = bio->bi_iter.bi_sector & (pbs - 1);
181 end = (start + max_sectors) & ~(pbs - 1);
184 return max_sectors & ~(lbs - 1);
187 static inline unsigned get_max_segment_size(struct queue_limits *lim,
188 struct page *start_page, unsigned long offset)
190 unsigned long mask = lim->seg_boundary_mask;
192 offset = mask & (page_to_phys(start_page) + offset);
195 * overflow may be triggered in case of zero page physical address
196 * on 32bit arch, use queue's max segment size when that happens.
198 return min_not_zero(mask - offset + 1,
199 (unsigned long)lim->max_segment_size);
203 * bvec_split_segs - verify whether or not a bvec should be split in the middle
204 * @lim: [in] queue limits to split based on
205 * @bv: [in] bvec to examine
206 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
207 * by the number of segments from @bv that may be appended to that
208 * bio without exceeding @max_segs
209 * @bytes: [in,out] Number of bytes in the bio being built. Incremented
210 * by the number of bytes from @bv that may be appended to that
211 * bio without exceeding @max_bytes
212 * @max_segs: [in] upper bound for *@nsegs
213 * @max_bytes: [in] upper bound for *@bytes
215 * When splitting a bio, it can happen that a bvec is encountered that is too
216 * big to fit in a single segment and hence that it has to be split in the
217 * middle. This function verifies whether or not that should happen. The value
218 * %true is returned if and only if appending the entire @bv to a bio with
219 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
222 static bool bvec_split_segs(struct queue_limits *lim, const struct bio_vec *bv,
223 unsigned *nsegs, unsigned *bytes, unsigned max_segs,
226 unsigned max_len = min(max_bytes, UINT_MAX) - *bytes;
227 unsigned len = min(bv->bv_len, max_len);
228 unsigned total_len = 0;
229 unsigned seg_size = 0;
231 while (len && *nsegs < max_segs) {
232 seg_size = get_max_segment_size(lim, bv->bv_page,
233 bv->bv_offset + total_len);
234 seg_size = min(seg_size, len);
237 total_len += seg_size;
240 if ((bv->bv_offset + total_len) & lim->virt_boundary_mask)
246 /* tell the caller to split the bvec if it is too big to fit */
247 return len > 0 || bv->bv_len > max_len;
251 * bio_split_rw - split a bio in two bios
252 * @bio: [in] bio to be split
253 * @lim: [in] queue limits to split based on
254 * @segs: [out] number of segments in the bio with the first half of the sectors
255 * @bs: [in] bio set to allocate the clone from
256 * @max_bytes: [in] maximum number of bytes per bio
258 * Clone @bio, update the bi_iter of the clone to represent the first sectors
259 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
260 * following is guaranteed for the cloned bio:
261 * - That it has at most @max_bytes worth of data
262 * - That it has at most queue_max_segments(@q) segments.
264 * Except for discard requests the cloned bio will point at the bi_io_vec of
265 * the original bio. It is the responsibility of the caller to ensure that the
266 * original bio is not freed before the cloned bio. The caller is also
267 * responsible for ensuring that @bs is only destroyed after processing of the
268 * split bio has finished.
270 static struct bio *bio_split_rw(struct bio *bio, struct queue_limits *lim,
271 unsigned *segs, struct bio_set *bs, unsigned max_bytes)
273 struct bio_vec bv, bvprv, *bvprvp = NULL;
274 struct bvec_iter iter;
275 unsigned nsegs = 0, bytes = 0;
277 bio_for_each_bvec(bv, bio, iter) {
279 * If the queue doesn't support SG gaps and adding this
280 * offset would create a gap, disallow it.
282 if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv.bv_offset))
285 if (nsegs < lim->max_segments &&
286 bytes + bv.bv_len <= max_bytes &&
287 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
291 if (bvec_split_segs(lim, &bv, &nsegs, &bytes,
292 lim->max_segments, max_bytes))
304 * We can't sanely support splitting for a REQ_NOWAIT bio. End it
305 * with EAGAIN if splitting is required and return an error pointer.
307 if (bio->bi_opf & REQ_NOWAIT) {
308 bio->bi_status = BLK_STS_AGAIN;
310 return ERR_PTR(-EAGAIN);
316 * Individual bvecs might not be logical block aligned. Round down the
317 * split size so that each bio is properly block size aligned, even if
318 * we do not use the full hardware limits.
320 bytes = ALIGN_DOWN(bytes, lim->logical_block_size);
323 * Bio splitting may cause subtle trouble such as hang when doing sync
324 * iopoll in direct IO routine. Given performance gain of iopoll for
325 * big IO can be trival, disable iopoll when split needed.
327 bio_clear_polled(bio);
328 return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs);
332 * __bio_split_to_limits - split a bio to fit the queue limits
333 * @bio: bio to be split
334 * @lim: queue limits to split based on
335 * @nr_segs: returns the number of segments in the returned bio
337 * Check if @bio needs splitting based on the queue limits, and if so split off
338 * a bio fitting the limits from the beginning of @bio and return it. @bio is
339 * shortened to the remainder and re-submitted.
341 * The split bio is allocated from @q->bio_split, which is provided by the
344 struct bio *__bio_split_to_limits(struct bio *bio, struct queue_limits *lim,
345 unsigned int *nr_segs)
347 struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split;
350 switch (bio_op(bio)) {
352 case REQ_OP_SECURE_ERASE:
353 split = bio_split_discard(bio, lim, nr_segs, bs);
355 case REQ_OP_WRITE_ZEROES:
356 split = bio_split_write_zeroes(bio, lim, nr_segs, bs);
359 split = bio_split_rw(bio, lim, nr_segs, bs,
360 get_max_io_size(bio, lim) << SECTOR_SHIFT);
367 /* there isn't chance to merge the split bio */
368 split->bi_opf |= REQ_NOMERGE;
370 blkcg_bio_issue_init(split);
371 bio_chain(split, bio);
372 trace_block_split(split, bio->bi_iter.bi_sector);
373 submit_bio_noacct(bio);
380 * bio_split_to_limits - split a bio to fit the queue limits
381 * @bio: bio to be split
383 * Check if @bio needs splitting based on the queue limits of @bio->bi_bdev, and
384 * if so split off a bio fitting the limits from the beginning of @bio and
385 * return it. @bio is shortened to the remainder and re-submitted.
387 * The split bio is allocated from @q->bio_split, which is provided by the
390 struct bio *bio_split_to_limits(struct bio *bio)
392 struct queue_limits *lim = &bdev_get_queue(bio->bi_bdev)->limits;
393 unsigned int nr_segs;
395 if (bio_may_exceed_limits(bio, lim))
396 return __bio_split_to_limits(bio, lim, &nr_segs);
399 EXPORT_SYMBOL(bio_split_to_limits);
401 unsigned int blk_recalc_rq_segments(struct request *rq)
403 unsigned int nr_phys_segs = 0;
404 unsigned int bytes = 0;
405 struct req_iterator iter;
411 switch (bio_op(rq->bio)) {
413 case REQ_OP_SECURE_ERASE:
414 if (queue_max_discard_segments(rq->q) > 1) {
415 struct bio *bio = rq->bio;
422 case REQ_OP_WRITE_ZEROES:
428 rq_for_each_bvec(bv, rq, iter)
429 bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes,
434 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
435 struct scatterlist *sglist)
441 * If the driver previously mapped a shorter list, we could see a
442 * termination bit prematurely unless it fully inits the sg table
443 * on each mapping. We KNOW that there must be more entries here
444 * or the driver would be buggy, so force clear the termination bit
445 * to avoid doing a full sg_init_table() in drivers for each command.
451 static unsigned blk_bvec_map_sg(struct request_queue *q,
452 struct bio_vec *bvec, struct scatterlist *sglist,
453 struct scatterlist **sg)
455 unsigned nbytes = bvec->bv_len;
456 unsigned nsegs = 0, total = 0;
459 unsigned offset = bvec->bv_offset + total;
460 unsigned len = min(get_max_segment_size(&q->limits,
461 bvec->bv_page, offset), nbytes);
462 struct page *page = bvec->bv_page;
465 * Unfortunately a fair number of drivers barf on scatterlists
466 * that have an offset larger than PAGE_SIZE, despite other
467 * subsystems dealing with that invariant just fine. For now
468 * stick to the legacy format where we never present those from
469 * the block layer, but the code below should be removed once
470 * these offenders (mostly MMC/SD drivers) are fixed.
472 page += (offset >> PAGE_SHIFT);
473 offset &= ~PAGE_MASK;
475 *sg = blk_next_sg(sg, sglist);
476 sg_set_page(*sg, page, len, offset);
486 static inline int __blk_bvec_map_sg(struct bio_vec bv,
487 struct scatterlist *sglist, struct scatterlist **sg)
489 *sg = blk_next_sg(sg, sglist);
490 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
494 /* only try to merge bvecs into one sg if they are from two bios */
496 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
497 struct bio_vec *bvprv, struct scatterlist **sg)
500 int nbytes = bvec->bv_len;
505 if ((*sg)->length + nbytes > queue_max_segment_size(q))
508 if (!biovec_phys_mergeable(q, bvprv, bvec))
511 (*sg)->length += nbytes;
516 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
517 struct scatterlist *sglist,
518 struct scatterlist **sg)
520 struct bio_vec bvec, bvprv = { NULL };
521 struct bvec_iter iter;
523 bool new_bio = false;
526 bio_for_each_bvec(bvec, bio, iter) {
528 * Only try to merge bvecs from two bios given we
529 * have done bio internal merge when adding pages
533 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
536 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
537 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
539 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
543 if (likely(bio->bi_iter.bi_size)) {
553 * map a request to scatterlist, return number of sg entries setup. Caller
554 * must make sure sg can hold rq->nr_phys_segments entries
556 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
557 struct scatterlist *sglist, struct scatterlist **last_sg)
561 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
562 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
564 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
567 sg_mark_end(*last_sg);
570 * Something must have been wrong if the figured number of
571 * segment is bigger than number of req's physical segments
573 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
577 EXPORT_SYMBOL(__blk_rq_map_sg);
579 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
581 if (req_op(rq) == REQ_OP_DISCARD)
582 return queue_max_discard_segments(rq->q);
583 return queue_max_segments(rq->q);
586 static inline unsigned int blk_rq_get_max_sectors(struct request *rq,
589 struct request_queue *q = rq->q;
590 unsigned int max_sectors;
592 if (blk_rq_is_passthrough(rq))
593 return q->limits.max_hw_sectors;
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)
600 return min(max_sectors,
601 blk_chunk_sectors_left(offset, q->limits.chunk_sectors));
604 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
605 unsigned int nr_phys_segs)
607 if (!blk_cgroup_mergeable(req, bio))
610 if (blk_integrity_merge_bio(req->q, req, bio) == false)
613 /* discard request merge won't add new segment */
614 if (req_op(req) == REQ_OP_DISCARD)
617 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
621 * This will form the start of a new hw segment. Bump both
624 req->nr_phys_segments += nr_phys_segs;
628 req_set_nomerge(req->q, req);
632 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
634 if (req_gap_back_merge(req, bio))
636 if (blk_integrity_rq(req) &&
637 integrity_req_gap_back_merge(req, bio))
639 if (!bio_crypt_ctx_back_mergeable(req, bio))
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);
647 return ll_new_hw_segment(req, bio, nr_segs);
650 static int ll_front_merge_fn(struct request *req, struct bio *bio,
651 unsigned int nr_segs)
653 if (req_gap_front_merge(req, bio))
655 if (blk_integrity_rq(req) &&
656 integrity_req_gap_front_merge(req, bio))
658 if (!bio_crypt_ctx_front_mergeable(req, bio))
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);
666 return ll_new_hw_segment(req, bio, nr_segs);
669 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
670 struct request *next)
672 unsigned short segments = blk_rq_nr_discard_segments(req);
674 if (segments >= queue_max_discard_segments(q))
676 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
677 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
680 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
683 req_set_nomerge(q, req);
687 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
688 struct request *next)
690 int total_phys_segments;
692 if (req_gap_back_merge(req, next->bio))
696 * Will it become too large?
698 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
699 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
702 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
703 if (total_phys_segments > blk_rq_get_max_segments(req))
706 if (!blk_cgroup_mergeable(req, next->bio))
709 if (blk_integrity_merge_rq(q, req, next) == false)
712 if (!bio_crypt_ctx_merge_rq(req, next))
716 req->nr_phys_segments = total_phys_segments;
721 * blk_rq_set_mixed_merge - mark a request as mixed merge
722 * @rq: request to mark as mixed merge
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
729 void blk_rq_set_mixed_merge(struct request *rq)
731 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
734 if (rq->rq_flags & RQF_MIXED_MERGE)
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.
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);
747 rq->rq_flags |= RQF_MIXED_MERGE;
750 static inline blk_opf_t bio_failfast(const struct bio *bio)
752 if (bio->bi_opf & REQ_RAHEAD)
753 return REQ_FAILFAST_MASK;
755 return bio->bi_opf & REQ_FAILFAST_MASK;
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
763 static inline void blk_update_mixed_merge(struct request *req,
764 struct bio *bio, bool front_merge)
766 if (req->rq_flags & RQF_MIXED_MERGE) {
767 if (bio->bi_opf & REQ_RAHEAD)
768 bio->bi_opf |= REQ_FAILFAST_MASK;
771 req->cmd_flags &= ~REQ_FAILFAST_MASK;
772 req->cmd_flags |= bio->bi_opf & REQ_FAILFAST_MASK;
777 static void blk_account_io_merge_request(struct request *req)
779 if (blk_do_io_stat(req)) {
781 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
786 static enum elv_merge blk_try_req_merge(struct request *req,
787 struct request *next)
789 if (blk_discard_mergable(req))
790 return ELEVATOR_DISCARD_MERGE;
791 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
792 return ELEVATOR_BACK_MERGE;
794 return ELEVATOR_NO_MERGE;
798 * For non-mq, this has to be called with the request spinlock acquired.
799 * For mq with scheduling, the appropriate queue wide lock should be held.
801 static struct request *attempt_merge(struct request_queue *q,
802 struct request *req, struct request *next)
804 if (!rq_mergeable(req) || !rq_mergeable(next))
807 if (req_op(req) != req_op(next))
810 if (rq_data_dir(req) != rq_data_dir(next))
813 if (req->ioprio != next->ioprio)
817 * If we are allowed to merge, then append bio list
818 * from next to rq and release next. merge_requests_fn
819 * will have updated segment counts, update sector
820 * counts here. Handle DISCARDs separately, as they
821 * have separate settings.
824 switch (blk_try_req_merge(req, next)) {
825 case ELEVATOR_DISCARD_MERGE:
826 if (!req_attempt_discard_merge(q, req, next))
829 case ELEVATOR_BACK_MERGE:
830 if (!ll_merge_requests_fn(q, req, next))
838 * If failfast settings disagree or any of the two is already
839 * a mixed merge, mark both as mixed before proceeding. This
840 * makes sure that all involved bios have mixable attributes
843 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
844 (req->cmd_flags & REQ_FAILFAST_MASK) !=
845 (next->cmd_flags & REQ_FAILFAST_MASK)) {
846 blk_rq_set_mixed_merge(req);
847 blk_rq_set_mixed_merge(next);
851 * At this point we have either done a back merge or front merge. We
852 * need the smaller start_time_ns of the merged requests to be the
853 * current request for accounting purposes.
855 if (next->start_time_ns < req->start_time_ns)
856 req->start_time_ns = next->start_time_ns;
858 req->biotail->bi_next = next->bio;
859 req->biotail = next->biotail;
861 req->__data_len += blk_rq_bytes(next);
863 if (!blk_discard_mergable(req))
864 elv_merge_requests(q, req, next);
866 blk_crypto_rq_put_keyslot(next);
869 * 'next' is going away, so update stats accordingly
871 blk_account_io_merge_request(next);
873 trace_block_rq_merge(next);
876 * ownership of bio passed from next to req, return 'next' for
883 static struct request *attempt_back_merge(struct request_queue *q,
886 struct request *next = elv_latter_request(q, rq);
889 return attempt_merge(q, rq, next);
894 static struct request *attempt_front_merge(struct request_queue *q,
897 struct request *prev = elv_former_request(q, rq);
900 return attempt_merge(q, prev, rq);
906 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
907 * otherwise. The caller is responsible for freeing 'next' if the merge
910 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
911 struct request *next)
913 return attempt_merge(q, rq, next);
916 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
918 if (!rq_mergeable(rq) || !bio_mergeable(bio))
921 if (req_op(rq) != bio_op(bio))
924 /* different data direction or already started, don't merge */
925 if (bio_data_dir(bio) != rq_data_dir(rq))
928 /* don't merge across cgroup boundaries */
929 if (!blk_cgroup_mergeable(rq, bio))
932 /* only merge integrity protected bio into ditto rq */
933 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
936 /* Only merge if the crypt contexts are compatible */
937 if (!bio_crypt_rq_ctx_compatible(rq, bio))
940 if (rq->ioprio != bio_prio(bio))
946 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
948 if (blk_discard_mergable(rq))
949 return ELEVATOR_DISCARD_MERGE;
950 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
951 return ELEVATOR_BACK_MERGE;
952 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
953 return ELEVATOR_FRONT_MERGE;
954 return ELEVATOR_NO_MERGE;
957 static void blk_account_io_merge_bio(struct request *req)
959 if (!blk_do_io_stat(req))
963 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
967 enum bio_merge_status {
973 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
974 struct bio *bio, unsigned int nr_segs)
976 const blk_opf_t ff = bio_failfast(bio);
978 if (!ll_back_merge_fn(req, bio, nr_segs))
979 return BIO_MERGE_FAILED;
981 trace_block_bio_backmerge(bio);
982 rq_qos_merge(req->q, req, bio);
984 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
985 blk_rq_set_mixed_merge(req);
987 blk_update_mixed_merge(req, bio, false);
989 req->biotail->bi_next = bio;
991 req->__data_len += bio->bi_iter.bi_size;
993 bio_crypt_free_ctx(bio);
995 blk_account_io_merge_bio(req);
999 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
1000 struct bio *bio, unsigned int nr_segs)
1002 const blk_opf_t ff = bio_failfast(bio);
1004 if (!ll_front_merge_fn(req, bio, nr_segs))
1005 return BIO_MERGE_FAILED;
1007 trace_block_bio_frontmerge(bio);
1008 rq_qos_merge(req->q, req, bio);
1010 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1011 blk_rq_set_mixed_merge(req);
1013 blk_update_mixed_merge(req, bio, true);
1015 bio->bi_next = req->bio;
1018 req->__sector = bio->bi_iter.bi_sector;
1019 req->__data_len += bio->bi_iter.bi_size;
1021 bio_crypt_do_front_merge(req, bio);
1023 blk_account_io_merge_bio(req);
1024 return BIO_MERGE_OK;
1027 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
1028 struct request *req, struct bio *bio)
1030 unsigned short segments = blk_rq_nr_discard_segments(req);
1032 if (segments >= queue_max_discard_segments(q))
1034 if (blk_rq_sectors(req) + bio_sectors(bio) >
1035 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1038 rq_qos_merge(q, req, bio);
1040 req->biotail->bi_next = bio;
1042 req->__data_len += bio->bi_iter.bi_size;
1043 req->nr_phys_segments = segments + 1;
1045 blk_account_io_merge_bio(req);
1046 return BIO_MERGE_OK;
1048 req_set_nomerge(q, req);
1049 return BIO_MERGE_FAILED;
1052 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1055 unsigned int nr_segs,
1056 bool sched_allow_merge)
1058 if (!blk_rq_merge_ok(rq, bio))
1059 return BIO_MERGE_NONE;
1061 switch (blk_try_merge(rq, bio)) {
1062 case ELEVATOR_BACK_MERGE:
1063 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1064 return bio_attempt_back_merge(rq, bio, nr_segs);
1066 case ELEVATOR_FRONT_MERGE:
1067 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1068 return bio_attempt_front_merge(rq, bio, nr_segs);
1070 case ELEVATOR_DISCARD_MERGE:
1071 return bio_attempt_discard_merge(q, rq, bio);
1073 return BIO_MERGE_NONE;
1076 return BIO_MERGE_FAILED;
1080 * blk_attempt_plug_merge - try to merge with %current's plugged list
1081 * @q: request_queue new bio is being queued at
1082 * @bio: new bio being queued
1083 * @nr_segs: number of segments in @bio
1084 * from the passed in @q already in the plug list
1086 * Determine whether @bio being queued on @q can be merged with the previous
1087 * request on %current's plugged list. Returns %true if merge was successful,
1090 * Plugging coalesces IOs from the same issuer for the same purpose without
1091 * going through @q->queue_lock. As such it's more of an issuing mechanism
1092 * than scheduling, and the request, while may have elvpriv data, is not
1093 * added on the elevator at this point. In addition, we don't have
1094 * reliable access to the elevator outside queue lock. Only check basic
1095 * merging parameters without querying the elevator.
1097 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1099 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1100 unsigned int nr_segs)
1102 struct blk_plug *plug;
1105 plug = blk_mq_plug(bio);
1106 if (!plug || rq_list_empty(plug->mq_list))
1109 rq_list_for_each(&plug->mq_list, rq) {
1111 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1118 * Only keep iterating plug list for merges if we have multiple
1121 if (!plug->multiple_queues)
1128 * Iterate list of requests and see if we can merge this bio with any
1131 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1132 struct bio *bio, unsigned int nr_segs)
1137 list_for_each_entry_reverse(rq, list, queuelist) {
1141 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1142 case BIO_MERGE_NONE:
1146 case BIO_MERGE_FAILED:
1154 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1156 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1157 unsigned int nr_segs, struct request **merged_request)
1161 switch (elv_merge(q, &rq, bio)) {
1162 case ELEVATOR_BACK_MERGE:
1163 if (!blk_mq_sched_allow_merge(q, rq, bio))
1165 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1167 *merged_request = attempt_back_merge(q, rq);
1168 if (!*merged_request)
1169 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1171 case ELEVATOR_FRONT_MERGE:
1172 if (!blk_mq_sched_allow_merge(q, rq, bio))
1174 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1176 *merged_request = attempt_front_merge(q, rq);
1177 if (!*merged_request)
1178 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1180 case ELEVATOR_DISCARD_MERGE:
1181 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1186 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);