]>
git.proxmox.com Git - mirror_ubuntu-hirsute-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
);
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
,
282 return bio_split(bio
, sectors
, GFP_NOIO
, bs
);
286 * __blk_queue_split - split a bio and submit the second half
287 * @bio: [in, out] bio to be split
288 * @nr_segs: [out] number of segments in the first bio
290 * Split a bio into two bios, chain the two bios, submit the second half and
291 * store a pointer to the first half in *@bio. If the second bio is still too
292 * big it will be split by a recursive call to this function. Since this
293 * function may allocate a new bio from @bio->bi_disk->queue->bio_split, it is
294 * the responsibility of the caller to ensure that
295 * @bio->bi_disk->queue->bio_split is only released after processing of the
296 * split bio has finished.
298 void __blk_queue_split(struct bio
**bio
, unsigned int *nr_segs
)
300 struct request_queue
*q
= (*bio
)->bi_disk
->queue
;
301 struct bio
*split
= NULL
;
303 switch (bio_op(*bio
)) {
305 case REQ_OP_SECURE_ERASE
:
306 split
= blk_bio_discard_split(q
, *bio
, &q
->bio_split
, nr_segs
);
308 case REQ_OP_WRITE_ZEROES
:
309 split
= blk_bio_write_zeroes_split(q
, *bio
, &q
->bio_split
,
312 case REQ_OP_WRITE_SAME
:
313 split
= blk_bio_write_same_split(q
, *bio
, &q
->bio_split
,
318 * All drivers must accept single-segments bios that are <=
319 * PAGE_SIZE. This is a quick and dirty check that relies on
320 * the fact that bi_io_vec[0] is always valid if a bio has data.
321 * The check might lead to occasional false negatives when bios
322 * are cloned, but compared to the performance impact of cloned
323 * bios themselves the loop below doesn't matter anyway.
325 if (!q
->limits
.chunk_sectors
&&
326 (*bio
)->bi_vcnt
== 1 &&
327 ((*bio
)->bi_io_vec
[0].bv_len
+
328 (*bio
)->bi_io_vec
[0].bv_offset
) <= PAGE_SIZE
) {
332 split
= blk_bio_segment_split(q
, *bio
, &q
->bio_split
, nr_segs
);
337 /* there isn't chance to merge the splitted bio */
338 split
->bi_opf
|= REQ_NOMERGE
;
340 bio_chain(split
, *bio
);
341 trace_block_split(q
, split
, (*bio
)->bi_iter
.bi_sector
);
342 submit_bio_noacct(*bio
);
348 * blk_queue_split - split a bio and submit the second half
349 * @bio: [in, out] bio to be split
351 * Split a bio into two bios, chains the two bios, submit the second half and
352 * store a pointer to the first half in *@bio. Since this function may allocate
353 * a new bio from @bio->bi_disk->queue->bio_split, it is the responsibility of
354 * the caller to ensure that @bio->bi_disk->queue->bio_split is only released
355 * after processing of the split bio has finished.
357 void blk_queue_split(struct bio
**bio
)
359 unsigned int nr_segs
;
361 __blk_queue_split(bio
, &nr_segs
);
363 EXPORT_SYMBOL(blk_queue_split
);
365 unsigned int blk_recalc_rq_segments(struct request
*rq
)
367 unsigned int nr_phys_segs
= 0;
368 unsigned int nr_sectors
= 0;
369 struct req_iterator iter
;
375 switch (bio_op(rq
->bio
)) {
377 case REQ_OP_SECURE_ERASE
:
378 case REQ_OP_WRITE_ZEROES
:
380 case REQ_OP_WRITE_SAME
:
384 rq_for_each_bvec(bv
, rq
, iter
)
385 bvec_split_segs(rq
->q
, &bv
, &nr_phys_segs
, &nr_sectors
,
390 static inline struct scatterlist
*blk_next_sg(struct scatterlist
**sg
,
391 struct scatterlist
*sglist
)
397 * If the driver previously mapped a shorter list, we could see a
398 * termination bit prematurely unless it fully inits the sg table
399 * on each mapping. We KNOW that there must be more entries here
400 * or the driver would be buggy, so force clear the termination bit
401 * to avoid doing a full sg_init_table() in drivers for each command.
407 static unsigned blk_bvec_map_sg(struct request_queue
*q
,
408 struct bio_vec
*bvec
, struct scatterlist
*sglist
,
409 struct scatterlist
**sg
)
411 unsigned nbytes
= bvec
->bv_len
;
412 unsigned nsegs
= 0, total
= 0;
415 unsigned offset
= bvec
->bv_offset
+ total
;
416 unsigned len
= min(get_max_segment_size(q
, bvec
->bv_page
,
418 struct page
*page
= bvec
->bv_page
;
421 * Unfortunately a fair number of drivers barf on scatterlists
422 * that have an offset larger than PAGE_SIZE, despite other
423 * subsystems dealing with that invariant just fine. For now
424 * stick to the legacy format where we never present those from
425 * the block layer, but the code below should be removed once
426 * these offenders (mostly MMC/SD drivers) are fixed.
428 page
+= (offset
>> PAGE_SHIFT
);
429 offset
&= ~PAGE_MASK
;
431 *sg
= blk_next_sg(sg
, sglist
);
432 sg_set_page(*sg
, page
, len
, offset
);
442 static inline int __blk_bvec_map_sg(struct bio_vec bv
,
443 struct scatterlist
*sglist
, struct scatterlist
**sg
)
445 *sg
= blk_next_sg(sg
, sglist
);
446 sg_set_page(*sg
, bv
.bv_page
, bv
.bv_len
, bv
.bv_offset
);
450 /* only try to merge bvecs into one sg if they are from two bios */
452 __blk_segment_map_sg_merge(struct request_queue
*q
, struct bio_vec
*bvec
,
453 struct bio_vec
*bvprv
, struct scatterlist
**sg
)
456 int nbytes
= bvec
->bv_len
;
461 if ((*sg
)->length
+ nbytes
> queue_max_segment_size(q
))
464 if (!biovec_phys_mergeable(q
, bvprv
, bvec
))
467 (*sg
)->length
+= nbytes
;
472 static int __blk_bios_map_sg(struct request_queue
*q
, struct bio
*bio
,
473 struct scatterlist
*sglist
,
474 struct scatterlist
**sg
)
476 struct bio_vec bvec
, bvprv
= { NULL
};
477 struct bvec_iter iter
;
479 bool new_bio
= false;
482 bio_for_each_bvec(bvec
, bio
, iter
) {
484 * Only try to merge bvecs from two bios given we
485 * have done bio internal merge when adding pages
489 __blk_segment_map_sg_merge(q
, &bvec
, &bvprv
, sg
))
492 if (bvec
.bv_offset
+ bvec
.bv_len
<= PAGE_SIZE
)
493 nsegs
+= __blk_bvec_map_sg(bvec
, sglist
, sg
);
495 nsegs
+= blk_bvec_map_sg(q
, &bvec
, sglist
, sg
);
499 if (likely(bio
->bi_iter
.bi_size
)) {
509 * map a request to scatterlist, return number of sg entries setup. Caller
510 * must make sure sg can hold rq->nr_phys_segments entries
512 int __blk_rq_map_sg(struct request_queue
*q
, struct request
*rq
,
513 struct scatterlist
*sglist
, struct scatterlist
**last_sg
)
517 if (rq
->rq_flags
& RQF_SPECIAL_PAYLOAD
)
518 nsegs
= __blk_bvec_map_sg(rq
->special_vec
, sglist
, last_sg
);
519 else if (rq
->bio
&& bio_op(rq
->bio
) == REQ_OP_WRITE_SAME
)
520 nsegs
= __blk_bvec_map_sg(bio_iovec(rq
->bio
), sglist
, last_sg
);
522 nsegs
= __blk_bios_map_sg(q
, rq
->bio
, sglist
, last_sg
);
525 sg_mark_end(*last_sg
);
528 * Something must have been wrong if the figured number of
529 * segment is bigger than number of req's physical segments
531 WARN_ON(nsegs
> blk_rq_nr_phys_segments(rq
));
535 EXPORT_SYMBOL(__blk_rq_map_sg
);
537 static inline unsigned int blk_rq_get_max_segments(struct request
*rq
)
539 if (req_op(rq
) == REQ_OP_DISCARD
)
540 return queue_max_discard_segments(rq
->q
);
541 return queue_max_segments(rq
->q
);
544 static inline int ll_new_hw_segment(struct request
*req
, struct bio
*bio
,
545 unsigned int nr_phys_segs
)
547 if (req
->nr_phys_segments
+ nr_phys_segs
> blk_rq_get_max_segments(req
))
550 if (blk_integrity_merge_bio(req
->q
, req
, bio
) == false)
554 * This will form the start of a new hw segment. Bump both
557 req
->nr_phys_segments
+= nr_phys_segs
;
561 req_set_nomerge(req
->q
, req
);
565 int ll_back_merge_fn(struct request
*req
, struct bio
*bio
, unsigned int nr_segs
)
567 if (req_gap_back_merge(req
, bio
))
569 if (blk_integrity_rq(req
) &&
570 integrity_req_gap_back_merge(req
, bio
))
572 if (!bio_crypt_ctx_back_mergeable(req
, bio
))
574 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
575 blk_rq_get_max_sectors(req
, blk_rq_pos(req
))) {
576 req_set_nomerge(req
->q
, req
);
580 return ll_new_hw_segment(req
, bio
, nr_segs
);
583 static int ll_front_merge_fn(struct request
*req
, struct bio
*bio
,
584 unsigned int nr_segs
)
586 if (req_gap_front_merge(req
, bio
))
588 if (blk_integrity_rq(req
) &&
589 integrity_req_gap_front_merge(req
, bio
))
591 if (!bio_crypt_ctx_front_mergeable(req
, bio
))
593 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
594 blk_rq_get_max_sectors(req
, bio
->bi_iter
.bi_sector
)) {
595 req_set_nomerge(req
->q
, req
);
599 return ll_new_hw_segment(req
, bio
, nr_segs
);
602 static bool req_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
603 struct request
*next
)
605 unsigned short segments
= blk_rq_nr_discard_segments(req
);
607 if (segments
>= queue_max_discard_segments(q
))
609 if (blk_rq_sectors(req
) + bio_sectors(next
->bio
) >
610 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
613 req
->nr_phys_segments
= segments
+ blk_rq_nr_discard_segments(next
);
616 req_set_nomerge(q
, req
);
620 static int ll_merge_requests_fn(struct request_queue
*q
, struct request
*req
,
621 struct request
*next
)
623 int total_phys_segments
;
625 if (req_gap_back_merge(req
, next
->bio
))
629 * Will it become too large?
631 if ((blk_rq_sectors(req
) + blk_rq_sectors(next
)) >
632 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
635 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
636 if (total_phys_segments
> blk_rq_get_max_segments(req
))
639 if (blk_integrity_merge_rq(q
, req
, next
) == false)
642 if (!bio_crypt_ctx_merge_rq(req
, next
))
646 req
->nr_phys_segments
= total_phys_segments
;
651 * blk_rq_set_mixed_merge - mark a request as mixed merge
652 * @rq: request to mark as mixed merge
655 * @rq is about to be mixed merged. Make sure the attributes
656 * which can be mixed are set in each bio and mark @rq as mixed
659 void blk_rq_set_mixed_merge(struct request
*rq
)
661 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
664 if (rq
->rq_flags
& RQF_MIXED_MERGE
)
668 * @rq will no longer represent mixable attributes for all the
669 * contained bios. It will just track those of the first one.
670 * Distributes the attributs to each bio.
672 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
673 WARN_ON_ONCE((bio
->bi_opf
& REQ_FAILFAST_MASK
) &&
674 (bio
->bi_opf
& REQ_FAILFAST_MASK
) != ff
);
677 rq
->rq_flags
|= RQF_MIXED_MERGE
;
680 static void blk_account_io_merge_request(struct request
*req
)
682 if (blk_do_io_stat(req
)) {
684 part_stat_inc(req
->part
, merges
[op_stat_group(req_op(req
))]);
687 hd_struct_put(req
->part
);
692 * Two cases of handling DISCARD merge:
693 * If max_discard_segments > 1, the driver takes every bio
694 * as a range and send them to controller together. The ranges
695 * needn't to be contiguous.
696 * Otherwise, the bios/requests will be handled as same as
697 * others which should be contiguous.
699 static inline bool blk_discard_mergable(struct request
*req
)
701 if (req_op(req
) == REQ_OP_DISCARD
&&
702 queue_max_discard_segments(req
->q
) > 1)
707 static enum elv_merge
blk_try_req_merge(struct request
*req
,
708 struct request
*next
)
710 if (blk_discard_mergable(req
))
711 return ELEVATOR_DISCARD_MERGE
;
712 else if (blk_rq_pos(req
) + blk_rq_sectors(req
) == blk_rq_pos(next
))
713 return ELEVATOR_BACK_MERGE
;
715 return ELEVATOR_NO_MERGE
;
719 * For non-mq, this has to be called with the request spinlock acquired.
720 * For mq with scheduling, the appropriate queue wide lock should be held.
722 static struct request
*attempt_merge(struct request_queue
*q
,
723 struct request
*req
, struct request
*next
)
725 if (!rq_mergeable(req
) || !rq_mergeable(next
))
728 if (req_op(req
) != req_op(next
))
731 if (rq_data_dir(req
) != rq_data_dir(next
)
732 || req
->rq_disk
!= next
->rq_disk
)
735 if (req_op(req
) == REQ_OP_WRITE_SAME
&&
736 !blk_write_same_mergeable(req
->bio
, next
->bio
))
740 * Don't allow merge of different write hints, or for a hint with
743 if (req
->write_hint
!= next
->write_hint
)
746 if (req
->ioprio
!= next
->ioprio
)
750 * If we are allowed to merge, then append bio list
751 * from next to rq and release next. merge_requests_fn
752 * will have updated segment counts, update sector
753 * counts here. Handle DISCARDs separately, as they
754 * have separate settings.
757 switch (blk_try_req_merge(req
, next
)) {
758 case ELEVATOR_DISCARD_MERGE
:
759 if (!req_attempt_discard_merge(q
, req
, next
))
762 case ELEVATOR_BACK_MERGE
:
763 if (!ll_merge_requests_fn(q
, req
, next
))
771 * If failfast settings disagree or any of the two is already
772 * a mixed merge, mark both as mixed before proceeding. This
773 * makes sure that all involved bios have mixable attributes
776 if (((req
->rq_flags
| next
->rq_flags
) & RQF_MIXED_MERGE
) ||
777 (req
->cmd_flags
& REQ_FAILFAST_MASK
) !=
778 (next
->cmd_flags
& REQ_FAILFAST_MASK
)) {
779 blk_rq_set_mixed_merge(req
);
780 blk_rq_set_mixed_merge(next
);
784 * At this point we have either done a back merge or front merge. We
785 * need the smaller start_time_ns of the merged requests to be the
786 * current request for accounting purposes.
788 if (next
->start_time_ns
< req
->start_time_ns
)
789 req
->start_time_ns
= next
->start_time_ns
;
791 req
->biotail
->bi_next
= next
->bio
;
792 req
->biotail
= next
->biotail
;
794 req
->__data_len
+= blk_rq_bytes(next
);
796 if (!blk_discard_mergable(req
))
797 elv_merge_requests(q
, req
, next
);
800 * 'next' is going away, so update stats accordingly
802 blk_account_io_merge_request(next
);
804 trace_block_rq_merge(q
, next
);
807 * ownership of bio passed from next to req, return 'next' for
814 static struct request
*attempt_back_merge(struct request_queue
*q
,
817 struct request
*next
= elv_latter_request(q
, rq
);
820 return attempt_merge(q
, rq
, next
);
825 static struct request
*attempt_front_merge(struct request_queue
*q
,
828 struct request
*prev
= elv_former_request(q
, rq
);
831 return attempt_merge(q
, prev
, rq
);
836 int blk_attempt_req_merge(struct request_queue
*q
, struct request
*rq
,
837 struct request
*next
)
839 struct request
*free
;
841 free
= attempt_merge(q
, rq
, next
);
843 blk_put_request(free
);
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_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(req
->q
, req
, 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(req
->q
, req
, 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
);