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);
355 /* there isn't chance to merge the splitted bio */
356 split->bi_opf |= REQ_NOMERGE;
358 bio_chain(split, *bio);
359 trace_block_split(split, (*bio)->bi_iter.bi_sector);
360 submit_bio_noacct(*bio);
363 blk_throtl_charge_bio_split(*bio);
368 * blk_queue_split - split a bio and submit the second half
369 * @bio: [in, out] bio to be split
371 * Split a bio into two bios, chains the two bios, submit the second half and
372 * store a pointer to the first half in *@bio. Since this function may allocate
373 * a new bio from q->bio_split, it is the responsibility of the caller to ensure
374 * that q->bio_split is only released after processing of the split bio has
377 void blk_queue_split(struct bio **bio)
379 unsigned int nr_segs;
381 __blk_queue_split(bio, &nr_segs);
383 EXPORT_SYMBOL(blk_queue_split);
385 unsigned int blk_recalc_rq_segments(struct request *rq)
387 unsigned int nr_phys_segs = 0;
388 unsigned int nr_sectors = 0;
389 struct req_iterator iter;
395 switch (bio_op(rq->bio)) {
397 case REQ_OP_SECURE_ERASE:
398 if (queue_max_discard_segments(rq->q) > 1) {
399 struct bio *bio = rq->bio;
406 case REQ_OP_WRITE_ZEROES:
408 case REQ_OP_WRITE_SAME:
412 rq_for_each_bvec(bv, rq, iter)
413 bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors,
418 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
419 struct scatterlist *sglist)
425 * If the driver previously mapped a shorter list, we could see a
426 * termination bit prematurely unless it fully inits the sg table
427 * on each mapping. We KNOW that there must be more entries here
428 * or the driver would be buggy, so force clear the termination bit
429 * to avoid doing a full sg_init_table() in drivers for each command.
435 static unsigned blk_bvec_map_sg(struct request_queue *q,
436 struct bio_vec *bvec, struct scatterlist *sglist,
437 struct scatterlist **sg)
439 unsigned nbytes = bvec->bv_len;
440 unsigned nsegs = 0, total = 0;
443 unsigned offset = bvec->bv_offset + total;
444 unsigned len = min(get_max_segment_size(q, bvec->bv_page,
446 struct page *page = bvec->bv_page;
449 * Unfortunately a fair number of drivers barf on scatterlists
450 * that have an offset larger than PAGE_SIZE, despite other
451 * subsystems dealing with that invariant just fine. For now
452 * stick to the legacy format where we never present those from
453 * the block layer, but the code below should be removed once
454 * these offenders (mostly MMC/SD drivers) are fixed.
456 page += (offset >> PAGE_SHIFT);
457 offset &= ~PAGE_MASK;
459 *sg = blk_next_sg(sg, sglist);
460 sg_set_page(*sg, page, len, offset);
470 static inline int __blk_bvec_map_sg(struct bio_vec bv,
471 struct scatterlist *sglist, struct scatterlist **sg)
473 *sg = blk_next_sg(sg, sglist);
474 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
478 /* only try to merge bvecs into one sg if they are from two bios */
480 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
481 struct bio_vec *bvprv, struct scatterlist **sg)
484 int nbytes = bvec->bv_len;
489 if ((*sg)->length + nbytes > queue_max_segment_size(q))
492 if (!biovec_phys_mergeable(q, bvprv, bvec))
495 (*sg)->length += nbytes;
500 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
501 struct scatterlist *sglist,
502 struct scatterlist **sg)
504 struct bio_vec bvec, bvprv = { NULL };
505 struct bvec_iter iter;
507 bool new_bio = false;
510 bio_for_each_bvec(bvec, bio, iter) {
512 * Only try to merge bvecs from two bios given we
513 * have done bio internal merge when adding pages
517 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
520 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
521 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
523 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
527 if (likely(bio->bi_iter.bi_size)) {
537 * map a request to scatterlist, return number of sg entries setup. Caller
538 * must make sure sg can hold rq->nr_phys_segments entries
540 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
541 struct scatterlist *sglist, struct scatterlist **last_sg)
545 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
546 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
547 else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME)
548 nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, last_sg);
550 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
553 sg_mark_end(*last_sg);
556 * Something must have been wrong if the figured number of
557 * segment is bigger than number of req's physical segments
559 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
563 EXPORT_SYMBOL(__blk_rq_map_sg);
565 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
567 if (req_op(rq) == REQ_OP_DISCARD)
568 return queue_max_discard_segments(rq->q);
569 return queue_max_segments(rq->q);
572 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
573 unsigned int nr_phys_segs)
575 if (!blk_cgroup_mergeable(req, bio))
578 if (blk_integrity_merge_bio(req->q, req, bio) == false)
581 /* discard request merge won't add new segment */
582 if (req_op(req) == REQ_OP_DISCARD)
585 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
589 * This will form the start of a new hw segment. Bump both
592 req->nr_phys_segments += nr_phys_segs;
596 req_set_nomerge(req->q, req);
600 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
602 if (req_gap_back_merge(req, bio))
604 if (blk_integrity_rq(req) &&
605 integrity_req_gap_back_merge(req, bio))
607 if (!bio_crypt_ctx_back_mergeable(req, bio))
609 if (blk_rq_sectors(req) + bio_sectors(bio) >
610 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
611 req_set_nomerge(req->q, req);
615 return ll_new_hw_segment(req, bio, nr_segs);
618 static int ll_front_merge_fn(struct request *req, struct bio *bio,
619 unsigned int nr_segs)
621 if (req_gap_front_merge(req, bio))
623 if (blk_integrity_rq(req) &&
624 integrity_req_gap_front_merge(req, bio))
626 if (!bio_crypt_ctx_front_mergeable(req, bio))
628 if (blk_rq_sectors(req) + bio_sectors(bio) >
629 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
630 req_set_nomerge(req->q, req);
634 return ll_new_hw_segment(req, bio, nr_segs);
637 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
638 struct request *next)
640 unsigned short segments = blk_rq_nr_discard_segments(req);
642 if (segments >= queue_max_discard_segments(q))
644 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
645 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
648 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
651 req_set_nomerge(q, req);
655 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
656 struct request *next)
658 int total_phys_segments;
660 if (req_gap_back_merge(req, next->bio))
664 * Will it become too large?
666 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
667 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
670 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
671 if (total_phys_segments > blk_rq_get_max_segments(req))
674 if (!blk_cgroup_mergeable(req, next->bio))
677 if (blk_integrity_merge_rq(q, req, next) == false)
680 if (!bio_crypt_ctx_merge_rq(req, next))
684 req->nr_phys_segments = total_phys_segments;
689 * blk_rq_set_mixed_merge - mark a request as mixed merge
690 * @rq: request to mark as mixed merge
693 * @rq is about to be mixed merged. Make sure the attributes
694 * which can be mixed are set in each bio and mark @rq as mixed
697 void blk_rq_set_mixed_merge(struct request *rq)
699 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
702 if (rq->rq_flags & RQF_MIXED_MERGE)
706 * @rq will no longer represent mixable attributes for all the
707 * contained bios. It will just track those of the first one.
708 * Distributes the attributs to each bio.
710 for (bio = rq->bio; bio; bio = bio->bi_next) {
711 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
712 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
715 rq->rq_flags |= RQF_MIXED_MERGE;
718 static void blk_account_io_merge_request(struct request *req)
720 if (blk_do_io_stat(req)) {
722 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
727 static enum elv_merge blk_try_req_merge(struct request *req,
728 struct request *next)
730 if (blk_discard_mergable(req))
731 return ELEVATOR_DISCARD_MERGE;
732 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
733 return ELEVATOR_BACK_MERGE;
735 return ELEVATOR_NO_MERGE;
739 * For non-mq, this has to be called with the request spinlock acquired.
740 * For mq with scheduling, the appropriate queue wide lock should be held.
742 static struct request *attempt_merge(struct request_queue *q,
743 struct request *req, struct request *next)
745 if (!rq_mergeable(req) || !rq_mergeable(next))
748 if (req_op(req) != req_op(next))
751 if (rq_data_dir(req) != rq_data_dir(next)
752 || req->rq_disk != next->rq_disk)
755 if (req_op(req) == REQ_OP_WRITE_SAME &&
756 !blk_write_same_mergeable(req->bio, next->bio))
760 * Don't allow merge of different write hints, or for a hint with
763 if (req->write_hint != next->write_hint)
766 if (req->ioprio != next->ioprio)
770 * If we are allowed to merge, then append bio list
771 * from next to rq and release next. merge_requests_fn
772 * will have updated segment counts, update sector
773 * counts here. Handle DISCARDs separately, as they
774 * have separate settings.
777 switch (blk_try_req_merge(req, next)) {
778 case ELEVATOR_DISCARD_MERGE:
779 if (!req_attempt_discard_merge(q, req, next))
782 case ELEVATOR_BACK_MERGE:
783 if (!ll_merge_requests_fn(q, req, next))
791 * If failfast settings disagree or any of the two is already
792 * a mixed merge, mark both as mixed before proceeding. This
793 * makes sure that all involved bios have mixable attributes
796 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
797 (req->cmd_flags & REQ_FAILFAST_MASK) !=
798 (next->cmd_flags & REQ_FAILFAST_MASK)) {
799 blk_rq_set_mixed_merge(req);
800 blk_rq_set_mixed_merge(next);
804 * At this point we have either done a back merge or front merge. We
805 * need the smaller start_time_ns of the merged requests to be the
806 * current request for accounting purposes.
808 if (next->start_time_ns < req->start_time_ns)
809 req->start_time_ns = next->start_time_ns;
811 req->biotail->bi_next = next->bio;
812 req->biotail = next->biotail;
814 req->__data_len += blk_rq_bytes(next);
816 if (!blk_discard_mergable(req))
817 elv_merge_requests(q, req, next);
820 * 'next' is going away, so update stats accordingly
822 blk_account_io_merge_request(next);
824 trace_block_rq_merge(next);
827 * ownership of bio passed from next to req, return 'next' for
834 static struct request *attempt_back_merge(struct request_queue *q,
837 struct request *next = elv_latter_request(q, rq);
840 return attempt_merge(q, rq, next);
845 static struct request *attempt_front_merge(struct request_queue *q,
848 struct request *prev = elv_former_request(q, rq);
851 return attempt_merge(q, prev, rq);
857 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
858 * otherwise. The caller is responsible for freeing 'next' if the merge
861 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
862 struct request *next)
864 return attempt_merge(q, rq, next);
867 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
869 if (!rq_mergeable(rq) || !bio_mergeable(bio))
872 if (req_op(rq) != bio_op(bio))
875 /* different data direction or already started, don't merge */
876 if (bio_data_dir(bio) != rq_data_dir(rq))
879 /* must be same device */
880 if (rq->rq_disk != bio->bi_bdev->bd_disk)
883 /* don't merge across cgroup boundaries */
884 if (!blk_cgroup_mergeable(rq, bio))
887 /* only merge integrity protected bio into ditto rq */
888 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
891 /* Only merge if the crypt contexts are compatible */
892 if (!bio_crypt_rq_ctx_compatible(rq, bio))
895 /* must be using the same buffer */
896 if (req_op(rq) == REQ_OP_WRITE_SAME &&
897 !blk_write_same_mergeable(rq->bio, bio))
901 * Don't allow merge of different write hints, or for a hint with
904 if (rq->write_hint != bio->bi_write_hint)
907 if (rq->ioprio != bio_prio(bio))
913 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
915 if (blk_discard_mergable(rq))
916 return ELEVATOR_DISCARD_MERGE;
917 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
918 return ELEVATOR_BACK_MERGE;
919 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
920 return ELEVATOR_FRONT_MERGE;
921 return ELEVATOR_NO_MERGE;
924 static void blk_account_io_merge_bio(struct request *req)
926 if (!blk_do_io_stat(req))
930 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
934 enum bio_merge_status {
940 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
941 struct bio *bio, unsigned int nr_segs)
943 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
945 if (!ll_back_merge_fn(req, bio, nr_segs))
946 return BIO_MERGE_FAILED;
948 trace_block_bio_backmerge(bio);
949 rq_qos_merge(req->q, req, bio);
951 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
952 blk_rq_set_mixed_merge(req);
954 req->biotail->bi_next = bio;
956 req->__data_len += bio->bi_iter.bi_size;
958 bio_crypt_free_ctx(bio);
960 blk_account_io_merge_bio(req);
964 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
965 struct bio *bio, unsigned int nr_segs)
967 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
969 if (!ll_front_merge_fn(req, bio, nr_segs))
970 return BIO_MERGE_FAILED;
972 trace_block_bio_frontmerge(bio);
973 rq_qos_merge(req->q, req, bio);
975 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
976 blk_rq_set_mixed_merge(req);
978 bio->bi_next = req->bio;
981 req->__sector = bio->bi_iter.bi_sector;
982 req->__data_len += bio->bi_iter.bi_size;
984 bio_crypt_do_front_merge(req, bio);
986 blk_account_io_merge_bio(req);
990 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
991 struct request *req, struct bio *bio)
993 unsigned short segments = blk_rq_nr_discard_segments(req);
995 if (segments >= queue_max_discard_segments(q))
997 if (blk_rq_sectors(req) + bio_sectors(bio) >
998 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1001 rq_qos_merge(q, req, bio);
1003 req->biotail->bi_next = bio;
1005 req->__data_len += bio->bi_iter.bi_size;
1006 req->nr_phys_segments = segments + 1;
1008 blk_account_io_merge_bio(req);
1009 return BIO_MERGE_OK;
1011 req_set_nomerge(q, req);
1012 return BIO_MERGE_FAILED;
1015 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1018 unsigned int nr_segs,
1019 bool sched_allow_merge)
1021 if (!blk_rq_merge_ok(rq, bio))
1022 return BIO_MERGE_NONE;
1024 switch (blk_try_merge(rq, bio)) {
1025 case ELEVATOR_BACK_MERGE:
1026 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1027 return bio_attempt_back_merge(rq, bio, nr_segs);
1029 case ELEVATOR_FRONT_MERGE:
1030 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1031 return bio_attempt_front_merge(rq, bio, nr_segs);
1033 case ELEVATOR_DISCARD_MERGE:
1034 return bio_attempt_discard_merge(q, rq, bio);
1036 return BIO_MERGE_NONE;
1039 return BIO_MERGE_FAILED;
1043 * blk_attempt_plug_merge - try to merge with %current's plugged list
1044 * @q: request_queue new bio is being queued at
1045 * @bio: new bio being queued
1046 * @nr_segs: number of segments in @bio
1047 * @same_queue_rq: pointer to &struct request that gets filled in when
1048 * another request associated with @q is found on the plug list
1049 * (optional, may be %NULL)
1051 * Determine whether @bio being queued on @q can be merged with a request
1052 * on %current's plugged list. Returns %true if merge was successful,
1055 * Plugging coalesces IOs from the same issuer for the same purpose without
1056 * going through @q->queue_lock. As such it's more of an issuing mechanism
1057 * than scheduling, and the request, while may have elvpriv data, is not
1058 * added on the elevator at this point. In addition, we don't have
1059 * reliable access to the elevator outside queue lock. Only check basic
1060 * merging parameters without querying the elevator.
1062 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1064 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1065 unsigned int nr_segs, struct request **same_queue_rq)
1067 struct blk_plug *plug;
1069 struct list_head *plug_list;
1071 plug = blk_mq_plug(q, bio);
1075 plug_list = &plug->mq_list;
1077 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1078 if (rq->q == q && same_queue_rq) {
1080 * Only blk-mq multiple hardware queues case checks the
1081 * rq in the same queue, there should be only one such
1084 *same_queue_rq = rq;
1090 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1099 * Iterate list of requests and see if we can merge this bio with any
1102 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1103 struct bio *bio, unsigned int nr_segs)
1108 list_for_each_entry_reverse(rq, list, queuelist) {
1112 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1113 case BIO_MERGE_NONE:
1117 case BIO_MERGE_FAILED:
1125 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1127 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1128 unsigned int nr_segs, struct request **merged_request)
1132 switch (elv_merge(q, &rq, bio)) {
1133 case ELEVATOR_BACK_MERGE:
1134 if (!blk_mq_sched_allow_merge(q, rq, bio))
1136 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1138 *merged_request = attempt_back_merge(q, rq);
1139 if (!*merged_request)
1140 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1142 case ELEVATOR_FRONT_MERGE:
1143 if (!blk_mq_sched_allow_merge(q, rq, bio))
1145 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1147 *merged_request = attempt_front_merge(q, rq);
1148 if (!*merged_request)
1149 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1151 case ELEVATOR_DISCARD_MERGE:
1152 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1157 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);