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(const 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,
109 const struct queue_limits *lim,
110 unsigned *nsegs, struct bio_set *bs)
112 unsigned int max_discard_sectors, granularity;
114 unsigned split_sectors;
118 /* Zero-sector (unknown) and one-sector granularities are the same. */
119 granularity = max(lim->discard_granularity >> 9, 1U);
121 max_discard_sectors =
122 min(lim->max_discard_sectors, bio_allowed_max_sectors(lim));
123 max_discard_sectors -= max_discard_sectors % granularity;
125 if (unlikely(!max_discard_sectors)) {
130 if (bio_sectors(bio) <= max_discard_sectors)
133 split_sectors = max_discard_sectors;
136 * If the next starting sector would be misaligned, stop the discard at
137 * the previous aligned sector.
139 tmp = bio->bi_iter.bi_sector + split_sectors -
140 ((lim->discard_alignment >> 9) % granularity);
141 tmp = sector_div(tmp, granularity);
143 if (split_sectors > tmp)
144 split_sectors -= tmp;
146 return bio_split(bio, split_sectors, GFP_NOIO, bs);
149 static struct bio *bio_split_write_zeroes(struct bio *bio,
150 const struct queue_limits *lim,
151 unsigned *nsegs, struct bio_set *bs)
154 if (!lim->max_write_zeroes_sectors)
156 if (bio_sectors(bio) <= lim->max_write_zeroes_sectors)
158 return bio_split(bio, lim->max_write_zeroes_sectors, GFP_NOIO, bs);
162 * Return the maximum number of sectors from the start of a bio that may be
163 * submitted as a single request to a block device. If enough sectors remain,
164 * align the end to the physical block size. Otherwise align the end to the
165 * logical block size. This approach minimizes the number of non-aligned
166 * requests that are submitted to a block device if the start of a bio is not
167 * aligned to a physical block boundary.
169 static inline unsigned get_max_io_size(struct bio *bio,
170 const struct queue_limits *lim)
172 unsigned pbs = lim->physical_block_size >> SECTOR_SHIFT;
173 unsigned lbs = lim->logical_block_size >> SECTOR_SHIFT;
174 unsigned max_sectors = lim->max_sectors, start, end;
176 if (lim->chunk_sectors) {
177 max_sectors = min(max_sectors,
178 blk_chunk_sectors_left(bio->bi_iter.bi_sector,
179 lim->chunk_sectors));
182 start = bio->bi_iter.bi_sector & (pbs - 1);
183 end = (start + max_sectors) & ~(pbs - 1);
186 return max_sectors & ~(lbs - 1);
190 * get_max_segment_size() - maximum number of bytes to add as a single segment
191 * @lim: Request queue limits.
192 * @start_page: See below.
193 * @offset: Offset from @start_page where to add a segment.
195 * Returns the maximum number of bytes that can be added as a single segment.
197 static inline unsigned get_max_segment_size(const struct queue_limits *lim,
198 struct page *start_page, unsigned long offset)
200 unsigned long mask = lim->seg_boundary_mask;
202 offset = mask & (page_to_phys(start_page) + offset);
205 * Prevent an overflow if mask = ULONG_MAX and offset = 0 by adding 1
206 * after having calculated the minimum.
208 return min(mask - offset, (unsigned long)lim->max_segment_size - 1) + 1;
212 * bvec_split_segs - verify whether or not a bvec should be split in the middle
213 * @lim: [in] queue limits to split based on
214 * @bv: [in] bvec to examine
215 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
216 * by the number of segments from @bv that may be appended to that
217 * bio without exceeding @max_segs
218 * @bytes: [in,out] Number of bytes in the bio being built. Incremented
219 * by the number of bytes from @bv that may be appended to that
220 * bio without exceeding @max_bytes
221 * @max_segs: [in] upper bound for *@nsegs
222 * @max_bytes: [in] upper bound for *@bytes
224 * When splitting a bio, it can happen that a bvec is encountered that is too
225 * big to fit in a single segment and hence that it has to be split in the
226 * middle. This function verifies whether or not that should happen. The value
227 * %true is returned if and only if appending the entire @bv to a bio with
228 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
231 static bool bvec_split_segs(const struct queue_limits *lim,
232 const struct bio_vec *bv, unsigned *nsegs, unsigned *bytes,
233 unsigned max_segs, unsigned max_bytes)
235 unsigned max_len = min(max_bytes, UINT_MAX) - *bytes;
236 unsigned len = min(bv->bv_len, max_len);
237 unsigned total_len = 0;
238 unsigned seg_size = 0;
240 while (len && *nsegs < max_segs) {
241 seg_size = get_max_segment_size(lim, bv->bv_page,
242 bv->bv_offset + total_len);
243 seg_size = min(seg_size, len);
246 total_len += seg_size;
249 if ((bv->bv_offset + total_len) & lim->virt_boundary_mask)
255 /* tell the caller to split the bvec if it is too big to fit */
256 return len > 0 || bv->bv_len > max_len;
260 * bio_split_rw - split a bio in two bios
261 * @bio: [in] bio to be split
262 * @lim: [in] queue limits to split based on
263 * @segs: [out] number of segments in the bio with the first half of the sectors
264 * @bs: [in] bio set to allocate the clone from
265 * @max_bytes: [in] maximum number of bytes per bio
267 * Clone @bio, update the bi_iter of the clone to represent the first sectors
268 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
269 * following is guaranteed for the cloned bio:
270 * - That it has at most @max_bytes worth of data
271 * - That it has at most queue_max_segments(@q) segments.
273 * Except for discard requests the cloned bio will point at the bi_io_vec of
274 * the original bio. It is the responsibility of the caller to ensure that the
275 * original bio is not freed before the cloned bio. The caller is also
276 * responsible for ensuring that @bs is only destroyed after processing of the
277 * split bio has finished.
279 static struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim,
280 unsigned *segs, struct bio_set *bs, unsigned max_bytes)
282 struct bio_vec bv, bvprv, *bvprvp = NULL;
283 struct bvec_iter iter;
284 unsigned nsegs = 0, bytes = 0;
286 bio_for_each_bvec(bv, bio, iter) {
288 * If the queue doesn't support SG gaps and adding this
289 * offset would create a gap, disallow it.
291 if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv.bv_offset))
294 if (nsegs < lim->max_segments &&
295 bytes + bv.bv_len <= max_bytes &&
296 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
300 if (bvec_split_segs(lim, &bv, &nsegs, &bytes,
301 lim->max_segments, max_bytes))
313 * We can't sanely support splitting for a REQ_NOWAIT bio. End it
314 * with EAGAIN if splitting is required and return an error pointer.
316 if (bio->bi_opf & REQ_NOWAIT) {
317 bio->bi_status = BLK_STS_AGAIN;
319 return ERR_PTR(-EAGAIN);
325 * Individual bvecs might not be logical block aligned. Round down the
326 * split size so that each bio is properly block size aligned, even if
327 * we do not use the full hardware limits.
329 bytes = ALIGN_DOWN(bytes, lim->logical_block_size);
332 * Bio splitting may cause subtle trouble such as hang when doing sync
333 * iopoll in direct IO routine. Given performance gain of iopoll for
334 * big IO can be trival, disable iopoll when split needed.
336 bio_clear_polled(bio);
337 return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs);
341 * __bio_split_to_limits - split a bio to fit the queue limits
342 * @bio: bio to be split
343 * @lim: queue limits to split based on
344 * @nr_segs: returns the number of segments in the returned bio
346 * Check if @bio needs splitting based on the queue limits, and if so split off
347 * a bio fitting the limits from the beginning of @bio and return it. @bio is
348 * shortened to the remainder and re-submitted.
350 * The split bio is allocated from @q->bio_split, which is provided by the
353 struct bio *__bio_split_to_limits(struct bio *bio,
354 const struct queue_limits *lim,
355 unsigned int *nr_segs)
357 struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split;
360 switch (bio_op(bio)) {
362 case REQ_OP_SECURE_ERASE:
363 split = bio_split_discard(bio, lim, nr_segs, bs);
365 case REQ_OP_WRITE_ZEROES:
366 split = bio_split_write_zeroes(bio, lim, nr_segs, bs);
369 split = bio_split_rw(bio, lim, nr_segs, bs,
370 get_max_io_size(bio, lim) << SECTOR_SHIFT);
377 /* there isn't chance to merge the split bio */
378 split->bi_opf |= REQ_NOMERGE;
380 blkcg_bio_issue_init(split);
381 bio_chain(split, bio);
382 trace_block_split(split, bio->bi_iter.bi_sector);
383 submit_bio_noacct(bio);
390 * bio_split_to_limits - split a bio to fit the queue limits
391 * @bio: bio to be split
393 * Check if @bio needs splitting based on the queue limits of @bio->bi_bdev, and
394 * if so split off a bio fitting the limits from the beginning of @bio and
395 * return it. @bio is shortened to the remainder and re-submitted.
397 * The split bio is allocated from @q->bio_split, which is provided by the
400 struct bio *bio_split_to_limits(struct bio *bio)
402 const struct queue_limits *lim = &bdev_get_queue(bio->bi_bdev)->limits;
403 unsigned int nr_segs;
405 if (bio_may_exceed_limits(bio, lim))
406 return __bio_split_to_limits(bio, lim, &nr_segs);
409 EXPORT_SYMBOL(bio_split_to_limits);
411 unsigned int blk_recalc_rq_segments(struct request *rq)
413 unsigned int nr_phys_segs = 0;
414 unsigned int bytes = 0;
415 struct req_iterator iter;
421 switch (bio_op(rq->bio)) {
423 case REQ_OP_SECURE_ERASE:
424 if (queue_max_discard_segments(rq->q) > 1) {
425 struct bio *bio = rq->bio;
432 case REQ_OP_WRITE_ZEROES:
438 rq_for_each_bvec(bv, rq, iter)
439 bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes,
444 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
445 struct scatterlist *sglist)
451 * If the driver previously mapped a shorter list, we could see a
452 * termination bit prematurely unless it fully inits the sg table
453 * on each mapping. We KNOW that there must be more entries here
454 * or the driver would be buggy, so force clear the termination bit
455 * to avoid doing a full sg_init_table() in drivers for each command.
461 static unsigned blk_bvec_map_sg(struct request_queue *q,
462 struct bio_vec *bvec, struct scatterlist *sglist,
463 struct scatterlist **sg)
465 unsigned nbytes = bvec->bv_len;
466 unsigned nsegs = 0, total = 0;
469 unsigned offset = bvec->bv_offset + total;
470 unsigned len = min(get_max_segment_size(&q->limits,
471 bvec->bv_page, offset), nbytes);
472 struct page *page = bvec->bv_page;
475 * Unfortunately a fair number of drivers barf on scatterlists
476 * that have an offset larger than PAGE_SIZE, despite other
477 * subsystems dealing with that invariant just fine. For now
478 * stick to the legacy format where we never present those from
479 * the block layer, but the code below should be removed once
480 * these offenders (mostly MMC/SD drivers) are fixed.
482 page += (offset >> PAGE_SHIFT);
483 offset &= ~PAGE_MASK;
485 *sg = blk_next_sg(sg, sglist);
486 sg_set_page(*sg, page, len, offset);
496 static inline int __blk_bvec_map_sg(struct bio_vec bv,
497 struct scatterlist *sglist, struct scatterlist **sg)
499 *sg = blk_next_sg(sg, sglist);
500 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
504 /* only try to merge bvecs into one sg if they are from two bios */
506 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
507 struct bio_vec *bvprv, struct scatterlist **sg)
510 int nbytes = bvec->bv_len;
515 if ((*sg)->length + nbytes > queue_max_segment_size(q))
518 if (!biovec_phys_mergeable(q, bvprv, bvec))
521 (*sg)->length += nbytes;
526 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
527 struct scatterlist *sglist,
528 struct scatterlist **sg)
530 struct bio_vec bvec, bvprv = { NULL };
531 struct bvec_iter iter;
533 bool new_bio = false;
536 bio_for_each_bvec(bvec, bio, iter) {
538 * Only try to merge bvecs from two bios given we
539 * have done bio internal merge when adding pages
543 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
546 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
547 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
549 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
553 if (likely(bio->bi_iter.bi_size)) {
563 * map a request to scatterlist, return number of sg entries setup. Caller
564 * must make sure sg can hold rq->nr_phys_segments entries
566 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
567 struct scatterlist *sglist, struct scatterlist **last_sg)
571 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
572 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
574 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
577 sg_mark_end(*last_sg);
580 * Something must have been wrong if the figured number of
581 * segment is bigger than number of req's physical segments
583 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
587 EXPORT_SYMBOL(__blk_rq_map_sg);
589 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
591 if (req_op(rq) == REQ_OP_DISCARD)
592 return queue_max_discard_segments(rq->q);
593 return queue_max_segments(rq->q);
596 static inline unsigned int blk_rq_get_max_sectors(struct request *rq,
599 struct request_queue *q = rq->q;
600 unsigned int max_sectors;
602 if (blk_rq_is_passthrough(rq))
603 return q->limits.max_hw_sectors;
605 max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
606 if (!q->limits.chunk_sectors ||
607 req_op(rq) == REQ_OP_DISCARD ||
608 req_op(rq) == REQ_OP_SECURE_ERASE)
610 return min(max_sectors,
611 blk_chunk_sectors_left(offset, q->limits.chunk_sectors));
614 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
615 unsigned int nr_phys_segs)
617 if (!blk_cgroup_mergeable(req, bio))
620 if (blk_integrity_merge_bio(req->q, req, bio) == false)
623 /* discard request merge won't add new segment */
624 if (req_op(req) == REQ_OP_DISCARD)
627 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
631 * This will form the start of a new hw segment. Bump both
634 req->nr_phys_segments += nr_phys_segs;
638 req_set_nomerge(req->q, req);
642 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
644 if (req_gap_back_merge(req, bio))
646 if (blk_integrity_rq(req) &&
647 integrity_req_gap_back_merge(req, bio))
649 if (!bio_crypt_ctx_back_mergeable(req, bio))
651 if (blk_rq_sectors(req) + bio_sectors(bio) >
652 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
653 req_set_nomerge(req->q, req);
657 return ll_new_hw_segment(req, bio, nr_segs);
660 static int ll_front_merge_fn(struct request *req, struct bio *bio,
661 unsigned int nr_segs)
663 if (req_gap_front_merge(req, bio))
665 if (blk_integrity_rq(req) &&
666 integrity_req_gap_front_merge(req, bio))
668 if (!bio_crypt_ctx_front_mergeable(req, bio))
670 if (blk_rq_sectors(req) + bio_sectors(bio) >
671 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
672 req_set_nomerge(req->q, req);
676 return ll_new_hw_segment(req, bio, nr_segs);
679 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
680 struct request *next)
682 unsigned short segments = blk_rq_nr_discard_segments(req);
684 if (segments >= queue_max_discard_segments(q))
686 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
687 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
690 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
693 req_set_nomerge(q, req);
697 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
698 struct request *next)
700 int total_phys_segments;
702 if (req_gap_back_merge(req, next->bio))
706 * Will it become too large?
708 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
709 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
712 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
713 if (total_phys_segments > blk_rq_get_max_segments(req))
716 if (!blk_cgroup_mergeable(req, next->bio))
719 if (blk_integrity_merge_rq(q, req, next) == false)
722 if (!bio_crypt_ctx_merge_rq(req, next))
726 req->nr_phys_segments = total_phys_segments;
731 * blk_rq_set_mixed_merge - mark a request as mixed merge
732 * @rq: request to mark as mixed merge
735 * @rq is about to be mixed merged. Make sure the attributes
736 * which can be mixed are set in each bio and mark @rq as mixed
739 void blk_rq_set_mixed_merge(struct request *rq)
741 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
744 if (rq->rq_flags & RQF_MIXED_MERGE)
748 * @rq will no longer represent mixable attributes for all the
749 * contained bios. It will just track those of the first one.
750 * Distributes the attributs to each bio.
752 for (bio = rq->bio; bio; bio = bio->bi_next) {
753 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
754 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
757 rq->rq_flags |= RQF_MIXED_MERGE;
760 static inline blk_opf_t bio_failfast(const struct bio *bio)
762 if (bio->bi_opf & REQ_RAHEAD)
763 return REQ_FAILFAST_MASK;
765 return bio->bi_opf & REQ_FAILFAST_MASK;
769 * After we are marked as MIXED_MERGE, any new RA bio has to be updated
770 * as failfast, and request's failfast has to be updated in case of
773 static inline void blk_update_mixed_merge(struct request *req,
774 struct bio *bio, bool front_merge)
776 if (req->rq_flags & RQF_MIXED_MERGE) {
777 if (bio->bi_opf & REQ_RAHEAD)
778 bio->bi_opf |= REQ_FAILFAST_MASK;
781 req->cmd_flags &= ~REQ_FAILFAST_MASK;
782 req->cmd_flags |= bio->bi_opf & REQ_FAILFAST_MASK;
787 static void blk_account_io_merge_request(struct request *req)
789 if (blk_do_io_stat(req)) {
791 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
796 static enum elv_merge blk_try_req_merge(struct request *req,
797 struct request *next)
799 if (blk_discard_mergable(req))
800 return ELEVATOR_DISCARD_MERGE;
801 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
802 return ELEVATOR_BACK_MERGE;
804 return ELEVATOR_NO_MERGE;
808 * For non-mq, this has to be called with the request spinlock acquired.
809 * For mq with scheduling, the appropriate queue wide lock should be held.
811 static struct request *attempt_merge(struct request_queue *q,
812 struct request *req, struct request *next)
814 if (!rq_mergeable(req) || !rq_mergeable(next))
817 if (req_op(req) != req_op(next))
820 if (rq_data_dir(req) != rq_data_dir(next))
823 if (req->ioprio != next->ioprio)
827 * If we are allowed to merge, then append bio list
828 * from next to rq and release next. merge_requests_fn
829 * will have updated segment counts, update sector
830 * counts here. Handle DISCARDs separately, as they
831 * have separate settings.
834 switch (blk_try_req_merge(req, next)) {
835 case ELEVATOR_DISCARD_MERGE:
836 if (!req_attempt_discard_merge(q, req, next))
839 case ELEVATOR_BACK_MERGE:
840 if (!ll_merge_requests_fn(q, req, next))
848 * If failfast settings disagree or any of the two is already
849 * a mixed merge, mark both as mixed before proceeding. This
850 * makes sure that all involved bios have mixable attributes
853 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
854 (req->cmd_flags & REQ_FAILFAST_MASK) !=
855 (next->cmd_flags & REQ_FAILFAST_MASK)) {
856 blk_rq_set_mixed_merge(req);
857 blk_rq_set_mixed_merge(next);
861 * At this point we have either done a back merge or front merge. We
862 * need the smaller start_time_ns of the merged requests to be the
863 * current request for accounting purposes.
865 if (next->start_time_ns < req->start_time_ns)
866 req->start_time_ns = next->start_time_ns;
868 req->biotail->bi_next = next->bio;
869 req->biotail = next->biotail;
871 req->__data_len += blk_rq_bytes(next);
873 if (!blk_discard_mergable(req))
874 elv_merge_requests(q, req, next);
877 * 'next' is going away, so update stats accordingly
879 blk_account_io_merge_request(next);
881 trace_block_rq_merge(next);
884 * ownership of bio passed from next to req, return 'next' for
891 static struct request *attempt_back_merge(struct request_queue *q,
894 struct request *next = elv_latter_request(q, rq);
897 return attempt_merge(q, rq, next);
902 static struct request *attempt_front_merge(struct request_queue *q,
905 struct request *prev = elv_former_request(q, rq);
908 return attempt_merge(q, prev, rq);
914 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
915 * otherwise. The caller is responsible for freeing 'next' if the merge
918 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
919 struct request *next)
921 return attempt_merge(q, rq, next);
924 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
926 if (!rq_mergeable(rq) || !bio_mergeable(bio))
929 if (req_op(rq) != bio_op(bio))
932 /* different data direction or already started, don't merge */
933 if (bio_data_dir(bio) != rq_data_dir(rq))
936 /* don't merge across cgroup boundaries */
937 if (!blk_cgroup_mergeable(rq, bio))
940 /* only merge integrity protected bio into ditto rq */
941 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
944 /* Only merge if the crypt contexts are compatible */
945 if (!bio_crypt_rq_ctx_compatible(rq, bio))
948 if (rq->ioprio != bio_prio(bio))
954 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
956 if (blk_discard_mergable(rq))
957 return ELEVATOR_DISCARD_MERGE;
958 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
959 return ELEVATOR_BACK_MERGE;
960 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
961 return ELEVATOR_FRONT_MERGE;
962 return ELEVATOR_NO_MERGE;
965 static void blk_account_io_merge_bio(struct request *req)
967 if (!blk_do_io_stat(req))
971 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
975 enum bio_merge_status {
981 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
982 struct bio *bio, unsigned int nr_segs)
984 const blk_opf_t ff = bio_failfast(bio);
986 if (!ll_back_merge_fn(req, bio, nr_segs))
987 return BIO_MERGE_FAILED;
989 trace_block_bio_backmerge(bio);
990 rq_qos_merge(req->q, req, bio);
992 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
993 blk_rq_set_mixed_merge(req);
995 blk_update_mixed_merge(req, bio, false);
997 req->biotail->bi_next = bio;
999 req->__data_len += bio->bi_iter.bi_size;
1001 bio_crypt_free_ctx(bio);
1003 blk_account_io_merge_bio(req);
1004 return BIO_MERGE_OK;
1007 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
1008 struct bio *bio, unsigned int nr_segs)
1010 const blk_opf_t ff = bio_failfast(bio);
1012 if (!ll_front_merge_fn(req, bio, nr_segs))
1013 return BIO_MERGE_FAILED;
1015 trace_block_bio_frontmerge(bio);
1016 rq_qos_merge(req->q, req, bio);
1018 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1019 blk_rq_set_mixed_merge(req);
1021 blk_update_mixed_merge(req, bio, true);
1023 bio->bi_next = req->bio;
1026 req->__sector = bio->bi_iter.bi_sector;
1027 req->__data_len += bio->bi_iter.bi_size;
1029 bio_crypt_do_front_merge(req, bio);
1031 blk_account_io_merge_bio(req);
1032 return BIO_MERGE_OK;
1035 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
1036 struct request *req, struct bio *bio)
1038 unsigned short segments = blk_rq_nr_discard_segments(req);
1040 if (segments >= queue_max_discard_segments(q))
1042 if (blk_rq_sectors(req) + bio_sectors(bio) >
1043 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1046 rq_qos_merge(q, req, bio);
1048 req->biotail->bi_next = bio;
1050 req->__data_len += bio->bi_iter.bi_size;
1051 req->nr_phys_segments = segments + 1;
1053 blk_account_io_merge_bio(req);
1054 return BIO_MERGE_OK;
1056 req_set_nomerge(q, req);
1057 return BIO_MERGE_FAILED;
1060 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1063 unsigned int nr_segs,
1064 bool sched_allow_merge)
1066 if (!blk_rq_merge_ok(rq, bio))
1067 return BIO_MERGE_NONE;
1069 switch (blk_try_merge(rq, bio)) {
1070 case ELEVATOR_BACK_MERGE:
1071 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1072 return bio_attempt_back_merge(rq, bio, nr_segs);
1074 case ELEVATOR_FRONT_MERGE:
1075 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1076 return bio_attempt_front_merge(rq, bio, nr_segs);
1078 case ELEVATOR_DISCARD_MERGE:
1079 return bio_attempt_discard_merge(q, rq, bio);
1081 return BIO_MERGE_NONE;
1084 return BIO_MERGE_FAILED;
1088 * blk_attempt_plug_merge - try to merge with %current's plugged list
1089 * @q: request_queue new bio is being queued at
1090 * @bio: new bio being queued
1091 * @nr_segs: number of segments in @bio
1092 * from the passed in @q already in the plug list
1094 * Determine whether @bio being queued on @q can be merged with the previous
1095 * request on %current's plugged list. Returns %true if merge was successful,
1098 * Plugging coalesces IOs from the same issuer for the same purpose without
1099 * going through @q->queue_lock. As such it's more of an issuing mechanism
1100 * than scheduling, and the request, while may have elvpriv data, is not
1101 * added on the elevator at this point. In addition, we don't have
1102 * reliable access to the elevator outside queue lock. Only check basic
1103 * merging parameters without querying the elevator.
1105 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1107 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1108 unsigned int nr_segs)
1110 struct blk_plug *plug;
1113 plug = blk_mq_plug(bio);
1114 if (!plug || rq_list_empty(plug->mq_list))
1117 rq_list_for_each(&plug->mq_list, rq) {
1119 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1126 * Only keep iterating plug list for merges if we have multiple
1129 if (!plug->multiple_queues)
1136 * Iterate list of requests and see if we can merge this bio with any
1139 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1140 struct bio *bio, unsigned int nr_segs)
1145 list_for_each_entry_reverse(rq, list, queuelist) {
1149 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1150 case BIO_MERGE_NONE:
1154 case BIO_MERGE_FAILED:
1162 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1164 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1165 unsigned int nr_segs, struct request **merged_request)
1169 switch (elv_merge(q, &rq, bio)) {
1170 case ELEVATOR_BACK_MERGE:
1171 if (!blk_mq_sched_allow_merge(q, rq, bio))
1173 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1175 *merged_request = attempt_back_merge(q, rq);
1176 if (!*merged_request)
1177 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1179 case ELEVATOR_FRONT_MERGE:
1180 if (!blk_mq_sched_allow_merge(q, rq, bio))
1182 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1184 *merged_request = attempt_front_merge(q, rq);
1185 if (!*merged_request)
1186 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1188 case ELEVATOR_DISCARD_MERGE:
1189 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1194 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);