]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - block/blk-merge.c
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>
11 #include <trace/events/block.h>
14 #include "blk-rq-qos.h"
16 static inline bool bio_will_gap(struct request_queue
*q
,
17 struct request
*prev_rq
, struct bio
*prev
, struct bio
*next
)
19 struct bio_vec pb
, nb
;
21 if (!bio_has_data(prev
) || !queue_virt_boundary(q
))
25 * Don't merge if the 1st bio starts with non-zero offset, otherwise it
26 * is quite difficult to respect the sg gap limit. We work hard to
27 * merge a huge number of small single bios in case of mkfs.
30 bio_get_first_bvec(prev_rq
->bio
, &pb
);
32 bio_get_first_bvec(prev
, &pb
);
33 if (pb
.bv_offset
& queue_virt_boundary(q
))
37 * We don't need to worry about the situation that the merged segment
38 * ends in unaligned virt boundary:
40 * - if 'pb' ends aligned, the merged segment ends aligned
41 * - if 'pb' ends unaligned, the next bio must include
42 * one single bvec of 'nb', otherwise the 'nb' can't
45 bio_get_last_bvec(prev
, &pb
);
46 bio_get_first_bvec(next
, &nb
);
47 if (biovec_phys_mergeable(q
, &pb
, &nb
))
49 return __bvec_gap_to_prev(q
, &pb
, nb
.bv_offset
);
52 static inline bool req_gap_back_merge(struct request
*req
, struct bio
*bio
)
54 return bio_will_gap(req
->q
, req
, req
->biotail
, bio
);
57 static inline bool req_gap_front_merge(struct request
*req
, struct bio
*bio
)
59 return bio_will_gap(req
->q
, NULL
, bio
, req
->bio
);
62 static struct bio
*blk_bio_discard_split(struct request_queue
*q
,
67 unsigned int max_discard_sectors
, granularity
;
70 unsigned split_sectors
;
74 /* Zero-sector (unknown) and one-sector granularities are the same. */
75 granularity
= max(q
->limits
.discard_granularity
>> 9, 1U);
77 max_discard_sectors
= min(q
->limits
.max_discard_sectors
,
78 bio_allowed_max_sectors(q
));
79 max_discard_sectors
-= max_discard_sectors
% granularity
;
81 if (unlikely(!max_discard_sectors
)) {
86 if (bio_sectors(bio
) <= max_discard_sectors
)
89 split_sectors
= max_discard_sectors
;
92 * If the next starting sector would be misaligned, stop the discard at
93 * the previous aligned sector.
95 alignment
= (q
->limits
.discard_alignment
>> 9) % granularity
;
97 tmp
= bio
->bi_iter
.bi_sector
+ split_sectors
- alignment
;
98 tmp
= sector_div(tmp
, granularity
);
100 if (split_sectors
> tmp
)
101 split_sectors
-= tmp
;
103 return bio_split(bio
, split_sectors
, GFP_NOIO
, bs
);
106 static struct bio
*blk_bio_write_zeroes_split(struct request_queue
*q
,
107 struct bio
*bio
, struct bio_set
*bs
, unsigned *nsegs
)
111 if (!q
->limits
.max_write_zeroes_sectors
)
114 if (bio_sectors(bio
) <= q
->limits
.max_write_zeroes_sectors
)
117 return bio_split(bio
, q
->limits
.max_write_zeroes_sectors
, GFP_NOIO
, bs
);
120 static struct bio
*blk_bio_write_same_split(struct request_queue
*q
,
127 if (!q
->limits
.max_write_same_sectors
)
130 if (bio_sectors(bio
) <= q
->limits
.max_write_same_sectors
)
133 return bio_split(bio
, q
->limits
.max_write_same_sectors
, GFP_NOIO
, bs
);
137 * Return the maximum number of sectors from the start of a bio that may be
138 * submitted as a single request to a block device. If enough sectors remain,
139 * align the end to the physical block size. Otherwise align the end to the
140 * logical block size. This approach minimizes the number of non-aligned
141 * requests that are submitted to a block device if the start of a bio is not
142 * aligned to a physical block boundary.
144 static inline unsigned get_max_io_size(struct request_queue
*q
,
147 unsigned sectors
= blk_max_size_offset(q
, bio
->bi_iter
.bi_sector
, 0);
148 unsigned max_sectors
= sectors
;
149 unsigned pbs
= queue_physical_block_size(q
) >> SECTOR_SHIFT
;
150 unsigned lbs
= queue_logical_block_size(q
) >> SECTOR_SHIFT
;
151 unsigned start_offset
= bio
->bi_iter
.bi_sector
& (pbs
- 1);
153 max_sectors
+= start_offset
;
154 max_sectors
&= ~(pbs
- 1);
155 if (max_sectors
> start_offset
)
156 return max_sectors
- start_offset
;
158 return sectors
& ~(lbs
- 1);
161 static inline unsigned get_max_segment_size(const struct request_queue
*q
,
162 struct page
*start_page
,
163 unsigned long offset
)
165 unsigned long mask
= queue_segment_boundary(q
);
167 offset
= mask
& (page_to_phys(start_page
) + offset
);
170 * overflow may be triggered in case of zero page physical address
171 * on 32bit arch, use queue's max segment size when that happens.
173 return min_not_zero(mask
- offset
+ 1,
174 (unsigned long)queue_max_segment_size(q
));
178 * bvec_split_segs - verify whether or not a bvec should be split in the middle
179 * @q: [in] request queue associated with the bio associated with @bv
180 * @bv: [in] bvec to examine
181 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
182 * by the number of segments from @bv that may be appended to that
183 * bio without exceeding @max_segs
184 * @sectors: [in,out] Number of sectors in the bio being built. Incremented
185 * by the number of sectors from @bv that may be appended to that
186 * bio without exceeding @max_sectors
187 * @max_segs: [in] upper bound for *@nsegs
188 * @max_sectors: [in] upper bound for *@sectors
190 * When splitting a bio, it can happen that a bvec is encountered that is too
191 * big to fit in a single segment and hence that it has to be split in the
192 * middle. This function verifies whether or not that should happen. The value
193 * %true is returned if and only if appending the entire @bv to a bio with
194 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
197 static bool bvec_split_segs(const struct request_queue
*q
,
198 const struct bio_vec
*bv
, unsigned *nsegs
,
199 unsigned *sectors
, unsigned max_segs
,
200 unsigned max_sectors
)
202 unsigned max_len
= (min(max_sectors
, UINT_MAX
>> 9) - *sectors
) << 9;
203 unsigned len
= min(bv
->bv_len
, max_len
);
204 unsigned total_len
= 0;
205 unsigned seg_size
= 0;
207 while (len
&& *nsegs
< max_segs
) {
208 seg_size
= get_max_segment_size(q
, bv
->bv_page
,
209 bv
->bv_offset
+ total_len
);
210 seg_size
= min(seg_size
, len
);
213 total_len
+= seg_size
;
216 if ((bv
->bv_offset
+ total_len
) & queue_virt_boundary(q
))
220 *sectors
+= total_len
>> 9;
222 /* tell the caller to split the bvec if it is too big to fit */
223 return len
> 0 || bv
->bv_len
> max_len
;
227 * blk_bio_segment_split - split a bio in two bios
228 * @q: [in] request queue pointer
229 * @bio: [in] bio to be split
230 * @bs: [in] bio set to allocate the clone from
231 * @segs: [out] number of segments in the bio with the first half of the sectors
233 * Clone @bio, update the bi_iter of the clone to represent the first sectors
234 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
235 * following is guaranteed for the cloned bio:
236 * - That it has at most get_max_io_size(@q, @bio) sectors.
237 * - That it has at most queue_max_segments(@q) segments.
239 * Except for discard requests the cloned bio will point at the bi_io_vec of
240 * the original bio. It is the responsibility of the caller to ensure that the
241 * original bio is not freed before the cloned bio. The caller is also
242 * responsible for ensuring that @bs is only destroyed after processing of the
243 * split bio has finished.
245 static struct bio
*blk_bio_segment_split(struct request_queue
*q
,
250 struct bio_vec bv
, bvprv
, *bvprvp
= NULL
;
251 struct bvec_iter iter
;
252 unsigned nsegs
= 0, sectors
= 0;
253 const unsigned max_sectors
= get_max_io_size(q
, bio
);
254 const unsigned max_segs
= queue_max_segments(q
);
256 bio_for_each_bvec(bv
, bio
, iter
) {
258 * If the queue doesn't support SG gaps and adding this
259 * offset would create a gap, disallow it.
261 if (bvprvp
&& bvec_gap_to_prev(q
, bvprvp
, bv
.bv_offset
))
264 if (nsegs
< max_segs
&&
265 sectors
+ (bv
.bv_len
>> 9) <= max_sectors
&&
266 bv
.bv_offset
+ bv
.bv_len
<= PAGE_SIZE
) {
268 sectors
+= bv
.bv_len
>> 9;
269 } else if (bvec_split_segs(q
, &bv
, &nsegs
, §ors
, max_segs
,
284 * Bio splitting may cause subtle trouble such as hang when doing sync
285 * iopoll in direct IO routine. Given performance gain of iopoll for
286 * big IO can be trival, disable iopoll when split needed.
288 bio_clear_hipri(bio
);
290 return bio_split(bio
, sectors
, GFP_NOIO
, bs
);
294 * __blk_queue_split - split a bio and submit the second half
295 * @bio: [in, out] bio to be split
296 * @nr_segs: [out] number of segments in the first bio
298 * Split a bio into two bios, chain the two bios, submit the second half and
299 * store a pointer to the first half in *@bio. If the second bio is still too
300 * big it will be split by a recursive call to this function. Since this
301 * function may allocate a new bio from q->bio_split, it is the responsibility
302 * of the caller to ensure that q->bio_split is only released after processing
303 * of the split bio has finished.
305 void __blk_queue_split(struct bio
**bio
, unsigned int *nr_segs
)
307 struct request_queue
*q
= (*bio
)->bi_bdev
->bd_disk
->queue
;
308 struct bio
*split
= NULL
;
310 switch (bio_op(*bio
)) {
312 case REQ_OP_SECURE_ERASE
:
313 split
= blk_bio_discard_split(q
, *bio
, &q
->bio_split
, nr_segs
);
315 case REQ_OP_WRITE_ZEROES
:
316 split
= blk_bio_write_zeroes_split(q
, *bio
, &q
->bio_split
,
319 case REQ_OP_WRITE_SAME
:
320 split
= blk_bio_write_same_split(q
, *bio
, &q
->bio_split
,
325 * All drivers must accept single-segments bios that are <=
326 * PAGE_SIZE. This is a quick and dirty check that relies on
327 * the fact that bi_io_vec[0] is always valid if a bio has data.
328 * The check might lead to occasional false negatives when bios
329 * are cloned, but compared to the performance impact of cloned
330 * bios themselves the loop below doesn't matter anyway.
332 if (!q
->limits
.chunk_sectors
&&
333 (*bio
)->bi_vcnt
== 1 &&
334 ((*bio
)->bi_io_vec
[0].bv_len
+
335 (*bio
)->bi_io_vec
[0].bv_offset
) <= PAGE_SIZE
) {
339 split
= blk_bio_segment_split(q
, *bio
, &q
->bio_split
, nr_segs
);
344 /* there isn't chance to merge the splitted bio */
345 split
->bi_opf
|= REQ_NOMERGE
;
347 bio_chain(split
, *bio
);
348 trace_block_split(split
, (*bio
)->bi_iter
.bi_sector
);
349 submit_bio_noacct(*bio
);
352 blk_throtl_charge_bio_split(*bio
);
357 * blk_queue_split - split a bio and submit the second half
358 * @bio: [in, out] bio to be split
360 * Split a bio into two bios, chains the two bios, submit the second half and
361 * store a pointer to the first half in *@bio. Since this function may allocate
362 * a new bio from q->bio_split, it is the responsibility of the caller to ensure
363 * that q->bio_split is only released after processing of the split bio has
366 void blk_queue_split(struct bio
**bio
)
368 unsigned int nr_segs
;
370 __blk_queue_split(bio
, &nr_segs
);
372 EXPORT_SYMBOL(blk_queue_split
);
374 unsigned int blk_recalc_rq_segments(struct request
*rq
)
376 unsigned int nr_phys_segs
= 0;
377 unsigned int nr_sectors
= 0;
378 struct req_iterator iter
;
384 switch (bio_op(rq
->bio
)) {
386 case REQ_OP_SECURE_ERASE
:
387 if (queue_max_discard_segments(rq
->q
) > 1) {
388 struct bio
*bio
= rq
->bio
;
395 case REQ_OP_WRITE_ZEROES
:
397 case REQ_OP_WRITE_SAME
:
401 rq_for_each_bvec(bv
, rq
, iter
)
402 bvec_split_segs(rq
->q
, &bv
, &nr_phys_segs
, &nr_sectors
,
407 static inline struct scatterlist
*blk_next_sg(struct scatterlist
**sg
,
408 struct scatterlist
*sglist
)
414 * If the driver previously mapped a shorter list, we could see a
415 * termination bit prematurely unless it fully inits the sg table
416 * on each mapping. We KNOW that there must be more entries here
417 * or the driver would be buggy, so force clear the termination bit
418 * to avoid doing a full sg_init_table() in drivers for each command.
424 static unsigned blk_bvec_map_sg(struct request_queue
*q
,
425 struct bio_vec
*bvec
, struct scatterlist
*sglist
,
426 struct scatterlist
**sg
)
428 unsigned nbytes
= bvec
->bv_len
;
429 unsigned nsegs
= 0, total
= 0;
432 unsigned offset
= bvec
->bv_offset
+ total
;
433 unsigned len
= min(get_max_segment_size(q
, bvec
->bv_page
,
435 struct page
*page
= bvec
->bv_page
;
438 * Unfortunately a fair number of drivers barf on scatterlists
439 * that have an offset larger than PAGE_SIZE, despite other
440 * subsystems dealing with that invariant just fine. For now
441 * stick to the legacy format where we never present those from
442 * the block layer, but the code below should be removed once
443 * these offenders (mostly MMC/SD drivers) are fixed.
445 page
+= (offset
>> PAGE_SHIFT
);
446 offset
&= ~PAGE_MASK
;
448 *sg
= blk_next_sg(sg
, sglist
);
449 sg_set_page(*sg
, page
, len
, offset
);
459 static inline int __blk_bvec_map_sg(struct bio_vec bv
,
460 struct scatterlist
*sglist
, struct scatterlist
**sg
)
462 *sg
= blk_next_sg(sg
, sglist
);
463 sg_set_page(*sg
, bv
.bv_page
, bv
.bv_len
, bv
.bv_offset
);
467 /* only try to merge bvecs into one sg if they are from two bios */
469 __blk_segment_map_sg_merge(struct request_queue
*q
, struct bio_vec
*bvec
,
470 struct bio_vec
*bvprv
, struct scatterlist
**sg
)
473 int nbytes
= bvec
->bv_len
;
478 if ((*sg
)->length
+ nbytes
> queue_max_segment_size(q
))
481 if (!biovec_phys_mergeable(q
, bvprv
, bvec
))
484 (*sg
)->length
+= nbytes
;
489 static int __blk_bios_map_sg(struct request_queue
*q
, struct bio
*bio
,
490 struct scatterlist
*sglist
,
491 struct scatterlist
**sg
)
493 struct bio_vec bvec
, bvprv
= { NULL
};
494 struct bvec_iter iter
;
496 bool new_bio
= false;
499 bio_for_each_bvec(bvec
, bio
, iter
) {
501 * Only try to merge bvecs from two bios given we
502 * have done bio internal merge when adding pages
506 __blk_segment_map_sg_merge(q
, &bvec
, &bvprv
, sg
))
509 if (bvec
.bv_offset
+ bvec
.bv_len
<= PAGE_SIZE
)
510 nsegs
+= __blk_bvec_map_sg(bvec
, sglist
, sg
);
512 nsegs
+= blk_bvec_map_sg(q
, &bvec
, sglist
, sg
);
516 if (likely(bio
->bi_iter
.bi_size
)) {
526 * map a request to scatterlist, return number of sg entries setup. Caller
527 * must make sure sg can hold rq->nr_phys_segments entries
529 int __blk_rq_map_sg(struct request_queue
*q
, struct request
*rq
,
530 struct scatterlist
*sglist
, struct scatterlist
**last_sg
)
534 if (rq
->rq_flags
& RQF_SPECIAL_PAYLOAD
)
535 nsegs
= __blk_bvec_map_sg(rq
->special_vec
, sglist
, last_sg
);
536 else if (rq
->bio
&& bio_op(rq
->bio
) == REQ_OP_WRITE_SAME
)
537 nsegs
= __blk_bvec_map_sg(bio_iovec(rq
->bio
), sglist
, last_sg
);
539 nsegs
= __blk_bios_map_sg(q
, rq
->bio
, sglist
, last_sg
);
542 sg_mark_end(*last_sg
);
545 * Something must have been wrong if the figured number of
546 * segment is bigger than number of req's physical segments
548 WARN_ON(nsegs
> blk_rq_nr_phys_segments(rq
));
552 EXPORT_SYMBOL(__blk_rq_map_sg
);
554 static inline unsigned int blk_rq_get_max_segments(struct request
*rq
)
556 if (req_op(rq
) == REQ_OP_DISCARD
)
557 return queue_max_discard_segments(rq
->q
);
558 return queue_max_segments(rq
->q
);
561 static inline int ll_new_hw_segment(struct request
*req
, struct bio
*bio
,
562 unsigned int nr_phys_segs
)
564 if (blk_integrity_merge_bio(req
->q
, req
, bio
) == false)
567 /* discard request merge won't add new segment */
568 if (req_op(req
) == REQ_OP_DISCARD
)
571 if (req
->nr_phys_segments
+ nr_phys_segs
> blk_rq_get_max_segments(req
))
575 * This will form the start of a new hw segment. Bump both
578 req
->nr_phys_segments
+= nr_phys_segs
;
582 req_set_nomerge(req
->q
, req
);
586 int ll_back_merge_fn(struct request
*req
, struct bio
*bio
, unsigned int nr_segs
)
588 if (req_gap_back_merge(req
, bio
))
590 if (blk_integrity_rq(req
) &&
591 integrity_req_gap_back_merge(req
, bio
))
593 if (!bio_crypt_ctx_back_mergeable(req
, bio
))
595 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
596 blk_rq_get_max_sectors(req
, blk_rq_pos(req
))) {
597 req_set_nomerge(req
->q
, req
);
601 return ll_new_hw_segment(req
, bio
, nr_segs
);
604 static int ll_front_merge_fn(struct request
*req
, struct bio
*bio
,
605 unsigned int nr_segs
)
607 if (req_gap_front_merge(req
, bio
))
609 if (blk_integrity_rq(req
) &&
610 integrity_req_gap_front_merge(req
, bio
))
612 if (!bio_crypt_ctx_front_mergeable(req
, bio
))
614 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
615 blk_rq_get_max_sectors(req
, bio
->bi_iter
.bi_sector
)) {
616 req_set_nomerge(req
->q
, req
);
620 return ll_new_hw_segment(req
, bio
, nr_segs
);
623 static bool req_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
624 struct request
*next
)
626 unsigned short segments
= blk_rq_nr_discard_segments(req
);
628 if (segments
>= queue_max_discard_segments(q
))
630 if (blk_rq_sectors(req
) + bio_sectors(next
->bio
) >
631 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
634 req
->nr_phys_segments
= segments
+ blk_rq_nr_discard_segments(next
);
637 req_set_nomerge(q
, req
);
641 static int ll_merge_requests_fn(struct request_queue
*q
, struct request
*req
,
642 struct request
*next
)
644 int total_phys_segments
;
646 if (req_gap_back_merge(req
, next
->bio
))
650 * Will it become too large?
652 if ((blk_rq_sectors(req
) + blk_rq_sectors(next
)) >
653 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
656 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
657 if (total_phys_segments
> blk_rq_get_max_segments(req
))
660 if (blk_integrity_merge_rq(q
, req
, next
) == false)
663 if (!bio_crypt_ctx_merge_rq(req
, next
))
667 req
->nr_phys_segments
= total_phys_segments
;
672 * blk_rq_set_mixed_merge - mark a request as mixed merge
673 * @rq: request to mark as mixed merge
676 * @rq is about to be mixed merged. Make sure the attributes
677 * which can be mixed are set in each bio and mark @rq as mixed
680 void blk_rq_set_mixed_merge(struct request
*rq
)
682 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
685 if (rq
->rq_flags
& RQF_MIXED_MERGE
)
689 * @rq will no longer represent mixable attributes for all the
690 * contained bios. It will just track those of the first one.
691 * Distributes the attributs to each bio.
693 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
694 WARN_ON_ONCE((bio
->bi_opf
& REQ_FAILFAST_MASK
) &&
695 (bio
->bi_opf
& REQ_FAILFAST_MASK
) != ff
);
698 rq
->rq_flags
|= RQF_MIXED_MERGE
;
701 static void blk_account_io_merge_request(struct request
*req
)
703 if (blk_do_io_stat(req
)) {
705 part_stat_inc(req
->part
, merges
[op_stat_group(req_op(req
))]);
710 static enum elv_merge
blk_try_req_merge(struct request
*req
,
711 struct request
*next
)
713 if (blk_discard_mergable(req
))
714 return ELEVATOR_DISCARD_MERGE
;
715 else if (blk_rq_pos(req
) + blk_rq_sectors(req
) == blk_rq_pos(next
))
716 return ELEVATOR_BACK_MERGE
;
718 return ELEVATOR_NO_MERGE
;
722 * For non-mq, this has to be called with the request spinlock acquired.
723 * For mq with scheduling, the appropriate queue wide lock should be held.
725 static struct request
*attempt_merge(struct request_queue
*q
,
726 struct request
*req
, struct request
*next
)
728 if (!rq_mergeable(req
) || !rq_mergeable(next
))
731 if (req_op(req
) != req_op(next
))
734 if (rq_data_dir(req
) != rq_data_dir(next
)
735 || req
->rq_disk
!= next
->rq_disk
)
738 if (req_op(req
) == REQ_OP_WRITE_SAME
&&
739 !blk_write_same_mergeable(req
->bio
, next
->bio
))
743 * Don't allow merge of different write hints, or for a hint with
746 if (req
->write_hint
!= next
->write_hint
)
749 if (req
->ioprio
!= next
->ioprio
)
753 * If we are allowed to merge, then append bio list
754 * from next to rq and release next. merge_requests_fn
755 * will have updated segment counts, update sector
756 * counts here. Handle DISCARDs separately, as they
757 * have separate settings.
760 switch (blk_try_req_merge(req
, next
)) {
761 case ELEVATOR_DISCARD_MERGE
:
762 if (!req_attempt_discard_merge(q
, req
, next
))
765 case ELEVATOR_BACK_MERGE
:
766 if (!ll_merge_requests_fn(q
, req
, next
))
774 * If failfast settings disagree or any of the two is already
775 * a mixed merge, mark both as mixed before proceeding. This
776 * makes sure that all involved bios have mixable attributes
779 if (((req
->rq_flags
| next
->rq_flags
) & RQF_MIXED_MERGE
) ||
780 (req
->cmd_flags
& REQ_FAILFAST_MASK
) !=
781 (next
->cmd_flags
& REQ_FAILFAST_MASK
)) {
782 blk_rq_set_mixed_merge(req
);
783 blk_rq_set_mixed_merge(next
);
787 * At this point we have either done a back merge or front merge. We
788 * need the smaller start_time_ns of the merged requests to be the
789 * current request for accounting purposes.
791 if (next
->start_time_ns
< req
->start_time_ns
)
792 req
->start_time_ns
= next
->start_time_ns
;
794 req
->biotail
->bi_next
= next
->bio
;
795 req
->biotail
= next
->biotail
;
797 req
->__data_len
+= blk_rq_bytes(next
);
799 if (!blk_discard_mergable(req
))
800 elv_merge_requests(q
, req
, next
);
803 * 'next' is going away, so update stats accordingly
805 blk_account_io_merge_request(next
);
807 trace_block_rq_merge(next
);
810 * ownership of bio passed from next to req, return 'next' for
817 static struct request
*attempt_back_merge(struct request_queue
*q
,
820 struct request
*next
= elv_latter_request(q
, rq
);
823 return attempt_merge(q
, rq
, next
);
828 static struct request
*attempt_front_merge(struct request_queue
*q
,
831 struct request
*prev
= elv_former_request(q
, rq
);
834 return attempt_merge(q
, prev
, rq
);
840 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
841 * otherwise. The caller is responsible for freeing 'next' if the merge
844 bool blk_attempt_req_merge(struct request_queue
*q
, struct request
*rq
,
845 struct request
*next
)
847 return attempt_merge(q
, rq
, next
);
850 bool blk_rq_merge_ok(struct request
*rq
, struct bio
*bio
)
852 if (!rq_mergeable(rq
) || !bio_mergeable(bio
))
855 if (req_op(rq
) != bio_op(bio
))
858 /* different data direction or already started, don't merge */
859 if (bio_data_dir(bio
) != rq_data_dir(rq
))
862 /* must be same device */
863 if (rq
->rq_disk
!= bio
->bi_bdev
->bd_disk
)
866 /* only merge integrity protected bio into ditto rq */
867 if (blk_integrity_merge_bio(rq
->q
, rq
, bio
) == false)
870 /* Only merge if the crypt contexts are compatible */
871 if (!bio_crypt_rq_ctx_compatible(rq
, bio
))
874 /* must be using the same buffer */
875 if (req_op(rq
) == REQ_OP_WRITE_SAME
&&
876 !blk_write_same_mergeable(rq
->bio
, bio
))
880 * Don't allow merge of different write hints, or for a hint with
883 if (rq
->write_hint
!= bio
->bi_write_hint
)
886 if (rq
->ioprio
!= bio_prio(bio
))
892 enum elv_merge
blk_try_merge(struct request
*rq
, struct bio
*bio
)
894 if (blk_discard_mergable(rq
))
895 return ELEVATOR_DISCARD_MERGE
;
896 else if (blk_rq_pos(rq
) + blk_rq_sectors(rq
) == bio
->bi_iter
.bi_sector
)
897 return ELEVATOR_BACK_MERGE
;
898 else if (blk_rq_pos(rq
) - bio_sectors(bio
) == bio
->bi_iter
.bi_sector
)
899 return ELEVATOR_FRONT_MERGE
;
900 return ELEVATOR_NO_MERGE
;
903 static void blk_account_io_merge_bio(struct request
*req
)
905 if (!blk_do_io_stat(req
))
909 part_stat_inc(req
->part
, merges
[op_stat_group(req_op(req
))]);
913 enum bio_merge_status
{
919 static enum bio_merge_status
bio_attempt_back_merge(struct request
*req
,
920 struct bio
*bio
, unsigned int nr_segs
)
922 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
924 if (!ll_back_merge_fn(req
, bio
, nr_segs
))
925 return BIO_MERGE_FAILED
;
927 trace_block_bio_backmerge(bio
);
928 rq_qos_merge(req
->q
, req
, bio
);
930 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
931 blk_rq_set_mixed_merge(req
);
933 req
->biotail
->bi_next
= bio
;
935 req
->__data_len
+= bio
->bi_iter
.bi_size
;
937 bio_crypt_free_ctx(bio
);
939 blk_account_io_merge_bio(req
);
943 static enum bio_merge_status
bio_attempt_front_merge(struct request
*req
,
944 struct bio
*bio
, unsigned int nr_segs
)
946 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
948 if (!ll_front_merge_fn(req
, bio
, nr_segs
))
949 return BIO_MERGE_FAILED
;
951 trace_block_bio_frontmerge(bio
);
952 rq_qos_merge(req
->q
, req
, bio
);
954 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
955 blk_rq_set_mixed_merge(req
);
957 bio
->bi_next
= req
->bio
;
960 req
->__sector
= bio
->bi_iter
.bi_sector
;
961 req
->__data_len
+= bio
->bi_iter
.bi_size
;
963 bio_crypt_do_front_merge(req
, bio
);
965 blk_account_io_merge_bio(req
);
969 static enum bio_merge_status
bio_attempt_discard_merge(struct request_queue
*q
,
970 struct request
*req
, struct bio
*bio
)
972 unsigned short segments
= blk_rq_nr_discard_segments(req
);
974 if (segments
>= queue_max_discard_segments(q
))
976 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
977 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
980 rq_qos_merge(q
, req
, bio
);
982 req
->biotail
->bi_next
= bio
;
984 req
->__data_len
+= bio
->bi_iter
.bi_size
;
985 req
->nr_phys_segments
= segments
+ 1;
987 blk_account_io_merge_bio(req
);
990 req_set_nomerge(q
, req
);
991 return BIO_MERGE_FAILED
;
994 static enum bio_merge_status
blk_attempt_bio_merge(struct request_queue
*q
,
997 unsigned int nr_segs
,
998 bool sched_allow_merge
)
1000 if (!blk_rq_merge_ok(rq
, bio
))
1001 return BIO_MERGE_NONE
;
1003 switch (blk_try_merge(rq
, bio
)) {
1004 case ELEVATOR_BACK_MERGE
:
1005 if (!sched_allow_merge
|| blk_mq_sched_allow_merge(q
, rq
, bio
))
1006 return bio_attempt_back_merge(rq
, bio
, nr_segs
);
1008 case ELEVATOR_FRONT_MERGE
:
1009 if (!sched_allow_merge
|| blk_mq_sched_allow_merge(q
, rq
, bio
))
1010 return bio_attempt_front_merge(rq
, bio
, nr_segs
);
1012 case ELEVATOR_DISCARD_MERGE
:
1013 return bio_attempt_discard_merge(q
, rq
, bio
);
1015 return BIO_MERGE_NONE
;
1018 return BIO_MERGE_FAILED
;
1022 * blk_attempt_plug_merge - try to merge with %current's plugged list
1023 * @q: request_queue new bio is being queued at
1024 * @bio: new bio being queued
1025 * @nr_segs: number of segments in @bio
1026 * @same_queue_rq: pointer to &struct request that gets filled in when
1027 * another request associated with @q is found on the plug list
1028 * (optional, may be %NULL)
1030 * Determine whether @bio being queued on @q can be merged with a request
1031 * on %current's plugged list. Returns %true if merge was successful,
1034 * Plugging coalesces IOs from the same issuer for the same purpose without
1035 * going through @q->queue_lock. As such it's more of an issuing mechanism
1036 * than scheduling, and the request, while may have elvpriv data, is not
1037 * added on the elevator at this point. In addition, we don't have
1038 * reliable access to the elevator outside queue lock. Only check basic
1039 * merging parameters without querying the elevator.
1041 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1043 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1044 unsigned int nr_segs
, struct request
**same_queue_rq
)
1046 struct blk_plug
*plug
;
1048 struct list_head
*plug_list
;
1050 plug
= blk_mq_plug(q
, bio
);
1054 plug_list
= &plug
->mq_list
;
1056 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1057 if (rq
->q
== q
&& same_queue_rq
) {
1059 * Only blk-mq multiple hardware queues case checks the
1060 * rq in the same queue, there should be only one such
1063 *same_queue_rq
= rq
;
1069 if (blk_attempt_bio_merge(q
, rq
, bio
, nr_segs
, false) ==
1078 * Iterate list of requests and see if we can merge this bio with any
1081 bool blk_bio_list_merge(struct request_queue
*q
, struct list_head
*list
,
1082 struct bio
*bio
, unsigned int nr_segs
)
1087 list_for_each_entry_reverse(rq
, list
, queuelist
) {
1091 switch (blk_attempt_bio_merge(q
, rq
, bio
, nr_segs
, true)) {
1092 case BIO_MERGE_NONE
:
1096 case BIO_MERGE_FAILED
:
1104 EXPORT_SYMBOL_GPL(blk_bio_list_merge
);
1106 bool blk_mq_sched_try_merge(struct request_queue
*q
, struct bio
*bio
,
1107 unsigned int nr_segs
, struct request
**merged_request
)
1111 switch (elv_merge(q
, &rq
, bio
)) {
1112 case ELEVATOR_BACK_MERGE
:
1113 if (!blk_mq_sched_allow_merge(q
, rq
, bio
))
1115 if (bio_attempt_back_merge(rq
, bio
, nr_segs
) != BIO_MERGE_OK
)
1117 *merged_request
= attempt_back_merge(q
, rq
);
1118 if (!*merged_request
)
1119 elv_merged_request(q
, rq
, ELEVATOR_BACK_MERGE
);
1121 case ELEVATOR_FRONT_MERGE
:
1122 if (!blk_mq_sched_allow_merge(q
, rq
, bio
))
1124 if (bio_attempt_front_merge(rq
, bio
, nr_segs
) != BIO_MERGE_OK
)
1126 *merged_request
= attempt_front_merge(q
, rq
);
1127 if (!*merged_request
)
1128 elv_merged_request(q
, rq
, ELEVATOR_FRONT_MERGE
);
1130 case ELEVATOR_DISCARD_MERGE
:
1131 return bio_attempt_discard_merge(q
, rq
, bio
) == BIO_MERGE_OK
;
1136 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge
);