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);
156 static struct bio *blk_bio_write_same_split(struct request_queue *q,
163 if (!q->limits.max_write_same_sectors)
166 if (bio_sectors(bio) <= q->limits.max_write_same_sectors)
169 return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs);
173 * Return the maximum number of sectors from the start of a bio that may be
174 * submitted as a single request to a block device. If enough sectors remain,
175 * align the end to the physical block size. Otherwise align the end to the
176 * logical block size. This approach minimizes the number of non-aligned
177 * requests that are submitted to a block device if the start of a bio is not
178 * aligned to a physical block boundary.
180 static inline unsigned get_max_io_size(struct request_queue *q,
183 unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector, 0);
184 unsigned max_sectors = sectors;
185 unsigned pbs = queue_physical_block_size(q) >> SECTOR_SHIFT;
186 unsigned lbs = queue_logical_block_size(q) >> SECTOR_SHIFT;
187 unsigned start_offset = bio->bi_iter.bi_sector & (pbs - 1);
189 max_sectors += start_offset;
190 max_sectors &= ~(pbs - 1);
191 if (max_sectors > start_offset)
192 return max_sectors - start_offset;
194 return sectors & ~(lbs - 1);
197 static inline unsigned get_max_segment_size(const struct request_queue *q,
198 struct page *start_page,
199 unsigned long offset)
201 unsigned long mask = queue_segment_boundary(q);
203 offset = mask & (page_to_phys(start_page) + offset);
206 * overflow may be triggered in case of zero page physical address
207 * on 32bit arch, use queue's max segment size when that happens.
209 return min_not_zero(mask - offset + 1,
210 (unsigned long)queue_max_segment_size(q));
214 * bvec_split_segs - verify whether or not a bvec should be split in the middle
215 * @q: [in] request queue associated with the bio associated with @bv
216 * @bv: [in] bvec to examine
217 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
218 * by the number of segments from @bv that may be appended to that
219 * bio without exceeding @max_segs
220 * @sectors: [in,out] Number of sectors in the bio being built. Incremented
221 * by the number of sectors from @bv that may be appended to that
222 * bio without exceeding @max_sectors
223 * @max_segs: [in] upper bound for *@nsegs
224 * @max_sectors: [in] upper bound for *@sectors
226 * When splitting a bio, it can happen that a bvec is encountered that is too
227 * big to fit in a single segment and hence that it has to be split in the
228 * middle. This function verifies whether or not that should happen. The value
229 * %true is returned if and only if appending the entire @bv to a bio with
230 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
233 static bool bvec_split_segs(const struct request_queue *q,
234 const struct bio_vec *bv, unsigned *nsegs,
235 unsigned *sectors, unsigned max_segs,
236 unsigned max_sectors)
238 unsigned max_len = (min(max_sectors, UINT_MAX >> 9) - *sectors) << 9;
239 unsigned len = min(bv->bv_len, max_len);
240 unsigned total_len = 0;
241 unsigned seg_size = 0;
243 while (len && *nsegs < max_segs) {
244 seg_size = get_max_segment_size(q, bv->bv_page,
245 bv->bv_offset + total_len);
246 seg_size = min(seg_size, len);
249 total_len += seg_size;
252 if ((bv->bv_offset + total_len) & queue_virt_boundary(q))
256 *sectors += total_len >> 9;
258 /* tell the caller to split the bvec if it is too big to fit */
259 return len > 0 || bv->bv_len > max_len;
263 * blk_bio_segment_split - split a bio in two bios
264 * @q: [in] request queue pointer
265 * @bio: [in] bio to be split
266 * @bs: [in] bio set to allocate the clone from
267 * @segs: [out] number of segments in the bio with the first half of the sectors
269 * Clone @bio, update the bi_iter of the clone to represent the first sectors
270 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
271 * following is guaranteed for the cloned bio:
272 * - That it has at most get_max_io_size(@q, @bio) sectors.
273 * - That it has at most queue_max_segments(@q) segments.
275 * Except for discard requests the cloned bio will point at the bi_io_vec of
276 * the original bio. It is the responsibility of the caller to ensure that the
277 * original bio is not freed before the cloned bio. The caller is also
278 * responsible for ensuring that @bs is only destroyed after processing of the
279 * split bio has finished.
281 static struct bio *blk_bio_segment_split(struct request_queue *q,
286 struct bio_vec bv, bvprv, *bvprvp = NULL;
287 struct bvec_iter iter;
288 unsigned nsegs = 0, sectors = 0;
289 const unsigned max_sectors = get_max_io_size(q, bio);
290 const unsigned max_segs = queue_max_segments(q);
292 bio_for_each_bvec(bv, bio, iter) {
294 * If the queue doesn't support SG gaps and adding this
295 * offset would create a gap, disallow it.
297 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
300 if (nsegs < max_segs &&
301 sectors + (bv.bv_len >> 9) <= max_sectors &&
302 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
304 sectors += bv.bv_len >> 9;
305 } else if (bvec_split_segs(q, &bv, &nsegs, §ors, max_segs,
320 * Bio splitting may cause subtle trouble such as hang when doing sync
321 * iopoll in direct IO routine. Given performance gain of iopoll for
322 * big IO can be trival, disable iopoll when split needed.
324 bio_clear_polled(bio);
325 return bio_split(bio, sectors, GFP_NOIO, bs);
329 * __blk_queue_split - split a bio and submit the second half
330 * @q: [in] request_queue new bio is being queued at
331 * @bio: [in, out] bio to be split
332 * @nr_segs: [out] number of segments in the first bio
334 * Split a bio into two bios, chain the two bios, submit the second half and
335 * store a pointer to the first half in *@bio. If the second bio is still too
336 * big it will be split by a recursive call to this function. Since this
337 * function may allocate a new bio from q->bio_split, it is the responsibility
338 * of the caller to ensure that q->bio_split is only released after processing
339 * of the split bio has finished.
341 void __blk_queue_split(struct request_queue *q, struct bio **bio,
342 unsigned int *nr_segs)
344 struct bio *split = NULL;
346 switch (bio_op(*bio)) {
348 case REQ_OP_SECURE_ERASE:
349 split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs);
351 case REQ_OP_WRITE_ZEROES:
352 split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split,
355 case REQ_OP_WRITE_SAME:
356 split = blk_bio_write_same_split(q, *bio, &q->bio_split,
360 split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs);
365 /* there isn't chance to merge the splitted bio */
366 split->bi_opf |= REQ_NOMERGE;
368 bio_chain(split, *bio);
369 trace_block_split(split, (*bio)->bi_iter.bi_sector);
370 submit_bio_noacct(*bio);
376 * blk_queue_split - split a bio and submit the second half
377 * @bio: [in, out] bio to be split
379 * Split a bio into two bios, chains the two bios, submit the second half and
380 * store a pointer to the first half in *@bio. Since this function may allocate
381 * a new bio from q->bio_split, it is the responsibility of the caller to ensure
382 * that q->bio_split is only released after processing of the split bio has
385 void blk_queue_split(struct bio **bio)
387 struct request_queue *q = bdev_get_queue((*bio)->bi_bdev);
388 unsigned int nr_segs;
390 if (blk_may_split(q, *bio))
391 __blk_queue_split(q, bio, &nr_segs);
393 EXPORT_SYMBOL(blk_queue_split);
395 unsigned int blk_recalc_rq_segments(struct request *rq)
397 unsigned int nr_phys_segs = 0;
398 unsigned int nr_sectors = 0;
399 struct req_iterator iter;
405 switch (bio_op(rq->bio)) {
407 case REQ_OP_SECURE_ERASE:
408 if (queue_max_discard_segments(rq->q) > 1) {
409 struct bio *bio = rq->bio;
416 case REQ_OP_WRITE_ZEROES:
418 case REQ_OP_WRITE_SAME:
422 rq_for_each_bvec(bv, rq, iter)
423 bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors,
428 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
429 struct scatterlist *sglist)
435 * If the driver previously mapped a shorter list, we could see a
436 * termination bit prematurely unless it fully inits the sg table
437 * on each mapping. We KNOW that there must be more entries here
438 * or the driver would be buggy, so force clear the termination bit
439 * to avoid doing a full sg_init_table() in drivers for each command.
445 static unsigned blk_bvec_map_sg(struct request_queue *q,
446 struct bio_vec *bvec, struct scatterlist *sglist,
447 struct scatterlist **sg)
449 unsigned nbytes = bvec->bv_len;
450 unsigned nsegs = 0, total = 0;
453 unsigned offset = bvec->bv_offset + total;
454 unsigned len = min(get_max_segment_size(q, bvec->bv_page,
456 struct page *page = bvec->bv_page;
459 * Unfortunately a fair number of drivers barf on scatterlists
460 * that have an offset larger than PAGE_SIZE, despite other
461 * subsystems dealing with that invariant just fine. For now
462 * stick to the legacy format where we never present those from
463 * the block layer, but the code below should be removed once
464 * these offenders (mostly MMC/SD drivers) are fixed.
466 page += (offset >> PAGE_SHIFT);
467 offset &= ~PAGE_MASK;
469 *sg = blk_next_sg(sg, sglist);
470 sg_set_page(*sg, page, len, offset);
480 static inline int __blk_bvec_map_sg(struct bio_vec bv,
481 struct scatterlist *sglist, struct scatterlist **sg)
483 *sg = blk_next_sg(sg, sglist);
484 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
488 /* only try to merge bvecs into one sg if they are from two bios */
490 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
491 struct bio_vec *bvprv, struct scatterlist **sg)
494 int nbytes = bvec->bv_len;
499 if ((*sg)->length + nbytes > queue_max_segment_size(q))
502 if (!biovec_phys_mergeable(q, bvprv, bvec))
505 (*sg)->length += nbytes;
510 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
511 struct scatterlist *sglist,
512 struct scatterlist **sg)
514 struct bio_vec bvec, bvprv = { NULL };
515 struct bvec_iter iter;
517 bool new_bio = false;
520 bio_for_each_bvec(bvec, bio, iter) {
522 * Only try to merge bvecs from two bios given we
523 * have done bio internal merge when adding pages
527 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
530 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
531 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
533 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
537 if (likely(bio->bi_iter.bi_size)) {
547 * map a request to scatterlist, return number of sg entries setup. Caller
548 * must make sure sg can hold rq->nr_phys_segments entries
550 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
551 struct scatterlist *sglist, struct scatterlist **last_sg)
555 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
556 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
557 else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME)
558 nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, last_sg);
560 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
563 sg_mark_end(*last_sg);
566 * Something must have been wrong if the figured number of
567 * segment is bigger than number of req's physical segments
569 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
573 EXPORT_SYMBOL(__blk_rq_map_sg);
575 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
577 if (req_op(rq) == REQ_OP_DISCARD)
578 return queue_max_discard_segments(rq->q);
579 return queue_max_segments(rq->q);
582 static inline unsigned int blk_rq_get_max_sectors(struct request *rq,
585 struct request_queue *q = rq->q;
587 if (blk_rq_is_passthrough(rq))
588 return q->limits.max_hw_sectors;
590 if (!q->limits.chunk_sectors ||
591 req_op(rq) == REQ_OP_DISCARD ||
592 req_op(rq) == REQ_OP_SECURE_ERASE)
593 return blk_queue_get_max_sectors(q, req_op(rq));
595 return min(blk_max_size_offset(q, offset, 0),
596 blk_queue_get_max_sectors(q, req_op(rq)));
599 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
600 unsigned int nr_phys_segs)
602 if (!blk_cgroup_mergeable(req, bio))
605 if (blk_integrity_merge_bio(req->q, req, bio) == false)
608 /* discard request merge won't add new segment */
609 if (req_op(req) == REQ_OP_DISCARD)
612 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
616 * This will form the start of a new hw segment. Bump both
619 req->nr_phys_segments += nr_phys_segs;
623 req_set_nomerge(req->q, req);
627 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
629 if (req_gap_back_merge(req, bio))
631 if (blk_integrity_rq(req) &&
632 integrity_req_gap_back_merge(req, bio))
634 if (!bio_crypt_ctx_back_mergeable(req, bio))
636 if (blk_rq_sectors(req) + bio_sectors(bio) >
637 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
638 req_set_nomerge(req->q, req);
642 return ll_new_hw_segment(req, bio, nr_segs);
645 static int ll_front_merge_fn(struct request *req, struct bio *bio,
646 unsigned int nr_segs)
648 if (req_gap_front_merge(req, bio))
650 if (blk_integrity_rq(req) &&
651 integrity_req_gap_front_merge(req, bio))
653 if (!bio_crypt_ctx_front_mergeable(req, bio))
655 if (blk_rq_sectors(req) + bio_sectors(bio) >
656 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
657 req_set_nomerge(req->q, req);
661 return ll_new_hw_segment(req, bio, nr_segs);
664 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
665 struct request *next)
667 unsigned short segments = blk_rq_nr_discard_segments(req);
669 if (segments >= queue_max_discard_segments(q))
671 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
672 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
675 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
678 req_set_nomerge(q, req);
682 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
683 struct request *next)
685 int total_phys_segments;
687 if (req_gap_back_merge(req, next->bio))
691 * Will it become too large?
693 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
694 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
697 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
698 if (total_phys_segments > blk_rq_get_max_segments(req))
701 if (!blk_cgroup_mergeable(req, next->bio))
704 if (blk_integrity_merge_rq(q, req, next) == false)
707 if (!bio_crypt_ctx_merge_rq(req, next))
711 req->nr_phys_segments = total_phys_segments;
716 * blk_rq_set_mixed_merge - mark a request as mixed merge
717 * @rq: request to mark as mixed merge
720 * @rq is about to be mixed merged. Make sure the attributes
721 * which can be mixed are set in each bio and mark @rq as mixed
724 void blk_rq_set_mixed_merge(struct request *rq)
726 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
729 if (rq->rq_flags & RQF_MIXED_MERGE)
733 * @rq will no longer represent mixable attributes for all the
734 * contained bios. It will just track those of the first one.
735 * Distributes the attributs to each bio.
737 for (bio = rq->bio; bio; bio = bio->bi_next) {
738 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
739 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
742 rq->rq_flags |= RQF_MIXED_MERGE;
745 static void blk_account_io_merge_request(struct request *req)
747 if (blk_do_io_stat(req)) {
749 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
754 static enum elv_merge blk_try_req_merge(struct request *req,
755 struct request *next)
757 if (blk_discard_mergable(req))
758 return ELEVATOR_DISCARD_MERGE;
759 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
760 return ELEVATOR_BACK_MERGE;
762 return ELEVATOR_NO_MERGE;
765 static inline bool blk_write_same_mergeable(struct bio *a, struct bio *b)
767 if (bio_page(a) == bio_page(b) && bio_offset(a) == bio_offset(b))
773 * For non-mq, this has to be called with the request spinlock acquired.
774 * For mq with scheduling, the appropriate queue wide lock should be held.
776 static struct request *attempt_merge(struct request_queue *q,
777 struct request *req, struct request *next)
779 if (!rq_mergeable(req) || !rq_mergeable(next))
782 if (req_op(req) != req_op(next))
785 if (rq_data_dir(req) != rq_data_dir(next))
788 if (req_op(req) == REQ_OP_WRITE_SAME &&
789 !blk_write_same_mergeable(req->bio, next->bio))
793 * Don't allow merge of different write hints, or for a hint with
796 if (req->write_hint != next->write_hint)
799 if (req->ioprio != next->ioprio)
803 * If we are allowed to merge, then append bio list
804 * from next to rq and release next. merge_requests_fn
805 * will have updated segment counts, update sector
806 * counts here. Handle DISCARDs separately, as they
807 * have separate settings.
810 switch (blk_try_req_merge(req, next)) {
811 case ELEVATOR_DISCARD_MERGE:
812 if (!req_attempt_discard_merge(q, req, next))
815 case ELEVATOR_BACK_MERGE:
816 if (!ll_merge_requests_fn(q, req, next))
824 * If failfast settings disagree or any of the two is already
825 * a mixed merge, mark both as mixed before proceeding. This
826 * makes sure that all involved bios have mixable attributes
829 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
830 (req->cmd_flags & REQ_FAILFAST_MASK) !=
831 (next->cmd_flags & REQ_FAILFAST_MASK)) {
832 blk_rq_set_mixed_merge(req);
833 blk_rq_set_mixed_merge(next);
837 * At this point we have either done a back merge or front merge. We
838 * need the smaller start_time_ns of the merged requests to be the
839 * current request for accounting purposes.
841 if (next->start_time_ns < req->start_time_ns)
842 req->start_time_ns = next->start_time_ns;
844 req->biotail->bi_next = next->bio;
845 req->biotail = next->biotail;
847 req->__data_len += blk_rq_bytes(next);
849 if (!blk_discard_mergable(req))
850 elv_merge_requests(q, req, next);
853 * 'next' is going away, so update stats accordingly
855 blk_account_io_merge_request(next);
857 trace_block_rq_merge(next);
860 * ownership of bio passed from next to req, return 'next' for
867 static struct request *attempt_back_merge(struct request_queue *q,
870 struct request *next = elv_latter_request(q, rq);
873 return attempt_merge(q, rq, next);
878 static struct request *attempt_front_merge(struct request_queue *q,
881 struct request *prev = elv_former_request(q, rq);
884 return attempt_merge(q, prev, rq);
890 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
891 * otherwise. The caller is responsible for freeing 'next' if the merge
894 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
895 struct request *next)
897 return attempt_merge(q, rq, next);
900 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
902 if (!rq_mergeable(rq) || !bio_mergeable(bio))
905 if (req_op(rq) != bio_op(bio))
908 /* different data direction or already started, don't merge */
909 if (bio_data_dir(bio) != rq_data_dir(rq))
912 /* don't merge across cgroup boundaries */
913 if (!blk_cgroup_mergeable(rq, bio))
916 /* only merge integrity protected bio into ditto rq */
917 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
920 /* Only merge if the crypt contexts are compatible */
921 if (!bio_crypt_rq_ctx_compatible(rq, bio))
924 /* must be using the same buffer */
925 if (req_op(rq) == REQ_OP_WRITE_SAME &&
926 !blk_write_same_mergeable(rq->bio, bio))
930 * Don't allow merge of different write hints, or for a hint with
933 if (rq->write_hint != bio->bi_write_hint)
936 if (rq->ioprio != bio_prio(bio))
942 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
944 if (blk_discard_mergable(rq))
945 return ELEVATOR_DISCARD_MERGE;
946 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
947 return ELEVATOR_BACK_MERGE;
948 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
949 return ELEVATOR_FRONT_MERGE;
950 return ELEVATOR_NO_MERGE;
953 static void blk_account_io_merge_bio(struct request *req)
955 if (!blk_do_io_stat(req))
959 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
963 enum bio_merge_status {
969 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
970 struct bio *bio, unsigned int nr_segs)
972 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
974 if (!ll_back_merge_fn(req, bio, nr_segs))
975 return BIO_MERGE_FAILED;
977 trace_block_bio_backmerge(bio);
978 rq_qos_merge(req->q, req, bio);
980 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
981 blk_rq_set_mixed_merge(req);
983 req->biotail->bi_next = bio;
985 req->__data_len += bio->bi_iter.bi_size;
987 bio_crypt_free_ctx(bio);
989 blk_account_io_merge_bio(req);
993 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
994 struct bio *bio, unsigned int nr_segs)
996 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
998 if (!ll_front_merge_fn(req, bio, nr_segs))
999 return BIO_MERGE_FAILED;
1001 trace_block_bio_frontmerge(bio);
1002 rq_qos_merge(req->q, req, bio);
1004 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1005 blk_rq_set_mixed_merge(req);
1007 bio->bi_next = req->bio;
1010 req->__sector = bio->bi_iter.bi_sector;
1011 req->__data_len += bio->bi_iter.bi_size;
1013 bio_crypt_do_front_merge(req, bio);
1015 blk_account_io_merge_bio(req);
1016 return BIO_MERGE_OK;
1019 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
1020 struct request *req, struct bio *bio)
1022 unsigned short segments = blk_rq_nr_discard_segments(req);
1024 if (segments >= queue_max_discard_segments(q))
1026 if (blk_rq_sectors(req) + bio_sectors(bio) >
1027 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1030 rq_qos_merge(q, req, bio);
1032 req->biotail->bi_next = bio;
1034 req->__data_len += bio->bi_iter.bi_size;
1035 req->nr_phys_segments = segments + 1;
1037 blk_account_io_merge_bio(req);
1038 return BIO_MERGE_OK;
1040 req_set_nomerge(q, req);
1041 return BIO_MERGE_FAILED;
1044 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1047 unsigned int nr_segs,
1048 bool sched_allow_merge)
1050 if (!blk_rq_merge_ok(rq, bio))
1051 return BIO_MERGE_NONE;
1053 switch (blk_try_merge(rq, bio)) {
1054 case ELEVATOR_BACK_MERGE:
1055 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1056 return bio_attempt_back_merge(rq, bio, nr_segs);
1058 case ELEVATOR_FRONT_MERGE:
1059 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1060 return bio_attempt_front_merge(rq, bio, nr_segs);
1062 case ELEVATOR_DISCARD_MERGE:
1063 return bio_attempt_discard_merge(q, rq, bio);
1065 return BIO_MERGE_NONE;
1068 return BIO_MERGE_FAILED;
1072 * blk_attempt_plug_merge - try to merge with %current's plugged list
1073 * @q: request_queue new bio is being queued at
1074 * @bio: new bio being queued
1075 * @nr_segs: number of segments in @bio
1076 * from the passed in @q already in the plug list
1078 * Determine whether @bio being queued on @q can be merged with the previous
1079 * request on %current's plugged list. Returns %true if merge was successful,
1082 * Plugging coalesces IOs from the same issuer for the same purpose without
1083 * going through @q->queue_lock. As such it's more of an issuing mechanism
1084 * than scheduling, and the request, while may have elvpriv data, is not
1085 * added on the elevator at this point. In addition, we don't have
1086 * reliable access to the elevator outside queue lock. Only check basic
1087 * merging parameters without querying the elevator.
1089 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1091 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1092 unsigned int nr_segs)
1094 struct blk_plug *plug;
1097 plug = blk_mq_plug(q, bio);
1098 if (!plug || rq_list_empty(plug->mq_list))
1101 rq_list_for_each(&plug->mq_list, rq) {
1103 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1110 * Only keep iterating plug list for merges if we have multiple
1113 if (!plug->multiple_queues)
1120 * Iterate list of requests and see if we can merge this bio with any
1123 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1124 struct bio *bio, unsigned int nr_segs)
1129 list_for_each_entry_reverse(rq, list, queuelist) {
1133 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1134 case BIO_MERGE_NONE:
1138 case BIO_MERGE_FAILED:
1146 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1148 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1149 unsigned int nr_segs, struct request **merged_request)
1153 switch (elv_merge(q, &rq, bio)) {
1154 case ELEVATOR_BACK_MERGE:
1155 if (!blk_mq_sched_allow_merge(q, rq, bio))
1157 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1159 *merged_request = attempt_back_merge(q, rq);
1160 if (!*merged_request)
1161 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1163 case ELEVATOR_FRONT_MERGE:
1164 if (!blk_mq_sched_allow_merge(q, rq, bio))
1166 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1168 *merged_request = attempt_front_merge(q, rq);
1169 if (!*merged_request)
1170 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1172 case ELEVATOR_DISCARD_MERGE:
1173 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1178 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);