1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Functions related to segment and merge handling
4 */
5 #include <linux/kernel.h>
6 #include <linux/module.h>
7 #include <linux/bio.h>
8 #include <linux/blkdev.h>
9 #include <linux/scatterlist.h>
10
11 #include <trace/events/block.h>
12
13 #include "blk.h"
14 #include "blk-rq-qos.h"
15
bio_will_gap(struct request_queue * q,struct request * prev_rq,struct bio * prev,struct bio * next)16 static inline bool bio_will_gap(struct request_queue *q,
17 struct request *prev_rq, struct bio *prev, struct bio *next)
18 {
19 struct bio_vec pb, nb;
20
21 if (!bio_has_data(prev) || !queue_virt_boundary(q))
22 return false;
23
24 /*
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.
28 */
29 if (prev_rq)
30 bio_get_first_bvec(prev_rq->bio, &pb);
31 else
32 bio_get_first_bvec(prev, &pb);
33 if (pb.bv_offset & queue_virt_boundary(q))
34 return true;
35
36 /*
37 * We don't need to worry about the situation that the merged segment
38 * ends in unaligned virt boundary:
39 *
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
43 * merge with 'pb'
44 */
45 bio_get_last_bvec(prev, &pb);
46 bio_get_first_bvec(next, &nb);
47 if (biovec_phys_mergeable(q, &pb, &nb))
48 return false;
49 return __bvec_gap_to_prev(q, &pb, nb.bv_offset);
50 }
51
req_gap_back_merge(struct request * req,struct bio * bio)52 static inline bool req_gap_back_merge(struct request *req, struct bio *bio)
53 {
54 return bio_will_gap(req->q, req, req->biotail, bio);
55 }
56
req_gap_front_merge(struct request * req,struct bio * bio)57 static inline bool req_gap_front_merge(struct request *req, struct bio *bio)
58 {
59 return bio_will_gap(req->q, NULL, bio, req->bio);
60 }
61
blk_bio_discard_split(struct request_queue * q,struct bio * bio,struct bio_set * bs,unsigned * nsegs)62 static struct bio *blk_bio_discard_split(struct request_queue *q,
63 struct bio *bio,
64 struct bio_set *bs,
65 unsigned *nsegs)
66 {
67 unsigned int max_discard_sectors, granularity;
68 int alignment;
69 sector_t tmp;
70 unsigned split_sectors;
71
72 *nsegs = 1;
73
74 /* Zero-sector (unknown) and one-sector granularities are the same. */
75 granularity = max(q->limits.discard_granularity >> 9, 1U);
76
77 max_discard_sectors = min(q->limits.max_discard_sectors,
78 bio_allowed_max_sectors(q));
79 max_discard_sectors -= max_discard_sectors % granularity;
80
81 if (unlikely(!max_discard_sectors)) {
82 /* XXX: warn */
83 return NULL;
84 }
85
86 if (bio_sectors(bio) <= max_discard_sectors)
87 return NULL;
88
89 split_sectors = max_discard_sectors;
90
91 /*
92 * If the next starting sector would be misaligned, stop the discard at
93 * the previous aligned sector.
94 */
95 alignment = (q->limits.discard_alignment >> 9) % granularity;
96
97 tmp = bio->bi_iter.bi_sector + split_sectors - alignment;
98 tmp = sector_div(tmp, granularity);
99
100 if (split_sectors > tmp)
101 split_sectors -= tmp;
102
103 return bio_split(bio, split_sectors, GFP_NOIO, bs);
104 }
105
blk_bio_write_zeroes_split(struct request_queue * q,struct bio * bio,struct bio_set * bs,unsigned * nsegs)106 static struct bio *blk_bio_write_zeroes_split(struct request_queue *q,
107 struct bio *bio, struct bio_set *bs, unsigned *nsegs)
108 {
109 *nsegs = 0;
110
111 if (!q->limits.max_write_zeroes_sectors)
112 return NULL;
113
114 if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors)
115 return NULL;
116
117 return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs);
118 }
119
blk_bio_write_same_split(struct request_queue * q,struct bio * bio,struct bio_set * bs,unsigned * nsegs)120 static struct bio *blk_bio_write_same_split(struct request_queue *q,
121 struct bio *bio,
122 struct bio_set *bs,
123 unsigned *nsegs)
124 {
125 *nsegs = 1;
126
127 if (!q->limits.max_write_same_sectors)
128 return NULL;
129
130 if (bio_sectors(bio) <= q->limits.max_write_same_sectors)
131 return NULL;
132
133 return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs);
134 }
135
136 /*
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.
143 */
get_max_io_size(struct request_queue * q,struct bio * bio)144 static inline unsigned get_max_io_size(struct request_queue *q,
145 struct bio *bio)
146 {
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);
152
153 max_sectors += start_offset;
154 max_sectors &= ~(pbs - 1);
155 if (max_sectors > start_offset)
156 return max_sectors - start_offset;
157
158 return sectors & ~(lbs - 1);
159 }
160
get_max_segment_size(const struct request_queue * q,struct page * start_page,unsigned long offset)161 static inline unsigned get_max_segment_size(const struct request_queue *q,
162 struct page *start_page,
163 unsigned long offset)
164 {
165 unsigned long mask = queue_segment_boundary(q);
166
167 offset = mask & (page_to_phys(start_page) + offset);
168
169 /*
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.
172 */
173 return min_not_zero(mask - offset + 1,
174 (unsigned long)queue_max_segment_size(q));
175 }
176
177 /**
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
189 *
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
195 * the block driver.
196 */
bvec_split_segs(const struct request_queue * q,const struct bio_vec * bv,unsigned * nsegs,unsigned * sectors,unsigned max_segs,unsigned max_sectors)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)
201 {
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;
206
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);
211
212 (*nsegs)++;
213 total_len += seg_size;
214 len -= seg_size;
215
216 if ((bv->bv_offset + total_len) & queue_virt_boundary(q))
217 break;
218 }
219
220 *sectors += total_len >> 9;
221
222 /* tell the caller to split the bvec if it is too big to fit */
223 return len > 0 || bv->bv_len > max_len;
224 }
225
226 /**
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
232 *
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.
238 *
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.
244 */
blk_bio_segment_split(struct request_queue * q,struct bio * bio,struct bio_set * bs,unsigned * segs)245 static struct bio *blk_bio_segment_split(struct request_queue *q,
246 struct bio *bio,
247 struct bio_set *bs,
248 unsigned *segs)
249 {
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);
255
256 bio_for_each_bvec(bv, bio, iter) {
257 /*
258 * If the queue doesn't support SG gaps and adding this
259 * offset would create a gap, disallow it.
260 */
261 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
262 goto split;
263
264 if (nsegs < max_segs &&
265 sectors + (bv.bv_len >> 9) <= max_sectors &&
266 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
267 nsegs++;
268 sectors += bv.bv_len >> 9;
269 } else if (bvec_split_segs(q, &bv, &nsegs, §ors, max_segs,
270 max_sectors)) {
271 goto split;
272 }
273
274 bvprv = bv;
275 bvprvp = &bvprv;
276 }
277
278 *segs = nsegs;
279 return NULL;
280 split:
281 *segs = nsegs;
282 return bio_split(bio, sectors, GFP_NOIO, bs);
283 }
284
285 /**
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
289 *
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.
297 */
__blk_queue_split(struct bio ** bio,unsigned int * nr_segs)298 void __blk_queue_split(struct bio **bio, unsigned int *nr_segs)
299 {
300 struct request_queue *q = (*bio)->bi_disk->queue;
301 struct bio *split = NULL;
302
303 switch (bio_op(*bio)) {
304 case REQ_OP_DISCARD:
305 case REQ_OP_SECURE_ERASE:
306 split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs);
307 break;
308 case REQ_OP_WRITE_ZEROES:
309 split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split,
310 nr_segs);
311 break;
312 case REQ_OP_WRITE_SAME:
313 split = blk_bio_write_same_split(q, *bio, &q->bio_split,
314 nr_segs);
315 break;
316 default:
317 /*
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.
324 */
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) {
329 *nr_segs = 1;
330 break;
331 }
332 split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs);
333 break;
334 }
335
336 if (split) {
337 /* there isn't chance to merge the splitted bio */
338 split->bi_opf |= REQ_NOMERGE;
339
340 bio_chain(split, *bio);
341 trace_block_split(q, split, (*bio)->bi_iter.bi_sector);
342 submit_bio_noacct(*bio);
343 *bio = split;
344
345 blk_throtl_charge_bio_split(*bio);
346 }
347 }
348
349 /**
350 * blk_queue_split - split a bio and submit the second half
351 * @bio: [in, out] bio to be split
352 *
353 * Split a bio into two bios, chains the two bios, submit the second half and
354 * store a pointer to the first half in *@bio. Since this function may allocate
355 * a new bio from @bio->bi_disk->queue->bio_split, it is the responsibility of
356 * the caller to ensure that @bio->bi_disk->queue->bio_split is only released
357 * after processing of the split bio has finished.
358 */
blk_queue_split(struct bio ** bio)359 void blk_queue_split(struct bio **bio)
360 {
361 unsigned int nr_segs;
362
363 __blk_queue_split(bio, &nr_segs);
364 }
365 EXPORT_SYMBOL(blk_queue_split);
366
blk_recalc_rq_segments(struct request * rq)367 unsigned int blk_recalc_rq_segments(struct request *rq)
368 {
369 unsigned int nr_phys_segs = 0;
370 unsigned int nr_sectors = 0;
371 struct req_iterator iter;
372 struct bio_vec bv;
373
374 if (!rq->bio)
375 return 0;
376
377 switch (bio_op(rq->bio)) {
378 case REQ_OP_DISCARD:
379 case REQ_OP_SECURE_ERASE:
380 if (queue_max_discard_segments(rq->q) > 1) {
381 struct bio *bio = rq->bio;
382
383 for_each_bio(bio)
384 nr_phys_segs++;
385 return nr_phys_segs;
386 }
387 return 1;
388 case REQ_OP_WRITE_ZEROES:
389 return 0;
390 case REQ_OP_WRITE_SAME:
391 return 1;
392 }
393
394 rq_for_each_bvec(bv, rq, iter)
395 bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors,
396 UINT_MAX, UINT_MAX);
397 return nr_phys_segs;
398 }
399
blk_next_sg(struct scatterlist ** sg,struct scatterlist * sglist)400 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
401 struct scatterlist *sglist)
402 {
403 if (!*sg)
404 return sglist;
405
406 /*
407 * If the driver previously mapped a shorter list, we could see a
408 * termination bit prematurely unless it fully inits the sg table
409 * on each mapping. We KNOW that there must be more entries here
410 * or the driver would be buggy, so force clear the termination bit
411 * to avoid doing a full sg_init_table() in drivers for each command.
412 */
413 sg_unmark_end(*sg);
414 return sg_next(*sg);
415 }
416
blk_bvec_map_sg(struct request_queue * q,struct bio_vec * bvec,struct scatterlist * sglist,struct scatterlist ** sg)417 static unsigned blk_bvec_map_sg(struct request_queue *q,
418 struct bio_vec *bvec, struct scatterlist *sglist,
419 struct scatterlist **sg)
420 {
421 unsigned nbytes = bvec->bv_len;
422 unsigned nsegs = 0, total = 0;
423
424 while (nbytes > 0) {
425 unsigned offset = bvec->bv_offset + total;
426 unsigned len = min(get_max_segment_size(q, bvec->bv_page,
427 offset), nbytes);
428 struct page *page = bvec->bv_page;
429
430 /*
431 * Unfortunately a fair number of drivers barf on scatterlists
432 * that have an offset larger than PAGE_SIZE, despite other
433 * subsystems dealing with that invariant just fine. For now
434 * stick to the legacy format where we never present those from
435 * the block layer, but the code below should be removed once
436 * these offenders (mostly MMC/SD drivers) are fixed.
437 */
438 page += (offset >> PAGE_SHIFT);
439 offset &= ~PAGE_MASK;
440
441 *sg = blk_next_sg(sg, sglist);
442 sg_set_page(*sg, page, len, offset);
443
444 total += len;
445 nbytes -= len;
446 nsegs++;
447 }
448
449 return nsegs;
450 }
451
__blk_bvec_map_sg(struct bio_vec bv,struct scatterlist * sglist,struct scatterlist ** sg)452 static inline int __blk_bvec_map_sg(struct bio_vec bv,
453 struct scatterlist *sglist, struct scatterlist **sg)
454 {
455 *sg = blk_next_sg(sg, sglist);
456 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
457 return 1;
458 }
459
460 /* only try to merge bvecs into one sg if they are from two bios */
461 static inline bool
__blk_segment_map_sg_merge(struct request_queue * q,struct bio_vec * bvec,struct bio_vec * bvprv,struct scatterlist ** sg)462 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
463 struct bio_vec *bvprv, struct scatterlist **sg)
464 {
465
466 int nbytes = bvec->bv_len;
467
468 if (!*sg)
469 return false;
470
471 if ((*sg)->length + nbytes > queue_max_segment_size(q))
472 return false;
473
474 if (!biovec_phys_mergeable(q, bvprv, bvec))
475 return false;
476
477 (*sg)->length += nbytes;
478
479 return true;
480 }
481
__blk_bios_map_sg(struct request_queue * q,struct bio * bio,struct scatterlist * sglist,struct scatterlist ** sg)482 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
483 struct scatterlist *sglist,
484 struct scatterlist **sg)
485 {
486 struct bio_vec bvec, bvprv = { NULL };
487 struct bvec_iter iter;
488 int nsegs = 0;
489 bool new_bio = false;
490
491 for_each_bio(bio) {
492 bio_for_each_bvec(bvec, bio, iter) {
493 /*
494 * Only try to merge bvecs from two bios given we
495 * have done bio internal merge when adding pages
496 * to bio
497 */
498 if (new_bio &&
499 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
500 goto next_bvec;
501
502 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
503 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
504 else
505 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
506 next_bvec:
507 new_bio = false;
508 }
509 if (likely(bio->bi_iter.bi_size)) {
510 bvprv = bvec;
511 new_bio = true;
512 }
513 }
514
515 return nsegs;
516 }
517
518 /*
519 * map a request to scatterlist, return number of sg entries setup. Caller
520 * must make sure sg can hold rq->nr_phys_segments entries
521 */
__blk_rq_map_sg(struct request_queue * q,struct request * rq,struct scatterlist * sglist,struct scatterlist ** last_sg)522 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
523 struct scatterlist *sglist, struct scatterlist **last_sg)
524 {
525 int nsegs = 0;
526
527 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
528 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
529 else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME)
530 nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, last_sg);
531 else if (rq->bio)
532 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
533
534 if (*last_sg)
535 sg_mark_end(*last_sg);
536
537 /*
538 * Something must have been wrong if the figured number of
539 * segment is bigger than number of req's physical segments
540 */
541 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
542
543 return nsegs;
544 }
545 EXPORT_SYMBOL(__blk_rq_map_sg);
546
blk_rq_get_max_segments(struct request * rq)547 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
548 {
549 if (req_op(rq) == REQ_OP_DISCARD)
550 return queue_max_discard_segments(rq->q);
551 return queue_max_segments(rq->q);
552 }
553
ll_new_hw_segment(struct request * req,struct bio * bio,unsigned int nr_phys_segs)554 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
555 unsigned int nr_phys_segs)
556 {
557 if (blk_integrity_merge_bio(req->q, req, bio) == false)
558 goto no_merge;
559
560 /* discard request merge won't add new segment */
561 if (req_op(req) == REQ_OP_DISCARD)
562 return 1;
563
564 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
565 goto no_merge;
566
567 /*
568 * This will form the start of a new hw segment. Bump both
569 * counters.
570 */
571 req->nr_phys_segments += nr_phys_segs;
572 return 1;
573
574 no_merge:
575 req_set_nomerge(req->q, req);
576 return 0;
577 }
578
ll_back_merge_fn(struct request * req,struct bio * bio,unsigned int nr_segs)579 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
580 {
581 if (req_gap_back_merge(req, bio))
582 return 0;
583 if (blk_integrity_rq(req) &&
584 integrity_req_gap_back_merge(req, bio))
585 return 0;
586 if (!bio_crypt_ctx_back_mergeable(req, bio))
587 return 0;
588 if (blk_rq_sectors(req) + bio_sectors(bio) >
589 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
590 req_set_nomerge(req->q, req);
591 return 0;
592 }
593
594 return ll_new_hw_segment(req, bio, nr_segs);
595 }
596
ll_front_merge_fn(struct request * req,struct bio * bio,unsigned int nr_segs)597 static int ll_front_merge_fn(struct request *req, struct bio *bio,
598 unsigned int nr_segs)
599 {
600 if (req_gap_front_merge(req, bio))
601 return 0;
602 if (blk_integrity_rq(req) &&
603 integrity_req_gap_front_merge(req, bio))
604 return 0;
605 if (!bio_crypt_ctx_front_mergeable(req, bio))
606 return 0;
607 if (blk_rq_sectors(req) + bio_sectors(bio) >
608 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
609 req_set_nomerge(req->q, req);
610 return 0;
611 }
612
613 return ll_new_hw_segment(req, bio, nr_segs);
614 }
615
req_attempt_discard_merge(struct request_queue * q,struct request * req,struct request * next)616 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
617 struct request *next)
618 {
619 unsigned short segments = blk_rq_nr_discard_segments(req);
620
621 if (segments >= queue_max_discard_segments(q))
622 goto no_merge;
623 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
624 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
625 goto no_merge;
626
627 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
628 return true;
629 no_merge:
630 req_set_nomerge(q, req);
631 return false;
632 }
633
ll_merge_requests_fn(struct request_queue * q,struct request * req,struct request * next)634 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
635 struct request *next)
636 {
637 int total_phys_segments;
638
639 if (req_gap_back_merge(req, next->bio))
640 return 0;
641
642 /*
643 * Will it become too large?
644 */
645 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
646 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
647 return 0;
648
649 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
650 if (total_phys_segments > blk_rq_get_max_segments(req))
651 return 0;
652
653 if (blk_integrity_merge_rq(q, req, next) == false)
654 return 0;
655
656 if (!bio_crypt_ctx_merge_rq(req, next))
657 return 0;
658
659 /* Merge is OK... */
660 req->nr_phys_segments = total_phys_segments;
661 return 1;
662 }
663
664 /**
665 * blk_rq_set_mixed_merge - mark a request as mixed merge
666 * @rq: request to mark as mixed merge
667 *
668 * Description:
669 * @rq is about to be mixed merged. Make sure the attributes
670 * which can be mixed are set in each bio and mark @rq as mixed
671 * merged.
672 */
blk_rq_set_mixed_merge(struct request * rq)673 void blk_rq_set_mixed_merge(struct request *rq)
674 {
675 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
676 struct bio *bio;
677
678 if (rq->rq_flags & RQF_MIXED_MERGE)
679 return;
680
681 /*
682 * @rq will no longer represent mixable attributes for all the
683 * contained bios. It will just track those of the first one.
684 * Distributes the attributs to each bio.
685 */
686 for (bio = rq->bio; bio; bio = bio->bi_next) {
687 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
688 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
689 bio->bi_opf |= ff;
690 }
691 rq->rq_flags |= RQF_MIXED_MERGE;
692 }
693
blk_account_io_merge_request(struct request * req)694 static void blk_account_io_merge_request(struct request *req)
695 {
696 if (blk_do_io_stat(req)) {
697 part_stat_lock();
698 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
699 part_stat_unlock();
700
701 hd_struct_put(req->part);
702 }
703 }
704
blk_try_req_merge(struct request * req,struct request * next)705 static enum elv_merge blk_try_req_merge(struct request *req,
706 struct request *next)
707 {
708 if (blk_discard_mergable(req))
709 return ELEVATOR_DISCARD_MERGE;
710 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
711 return ELEVATOR_BACK_MERGE;
712
713 return ELEVATOR_NO_MERGE;
714 }
715
716 /*
717 * For non-mq, this has to be called with the request spinlock acquired.
718 * For mq with scheduling, the appropriate queue wide lock should be held.
719 */
attempt_merge(struct request_queue * q,struct request * req,struct request * next)720 static struct request *attempt_merge(struct request_queue *q,
721 struct request *req, struct request *next)
722 {
723 if (!rq_mergeable(req) || !rq_mergeable(next))
724 return NULL;
725
726 if (req_op(req) != req_op(next))
727 return NULL;
728
729 if (rq_data_dir(req) != rq_data_dir(next)
730 || req->rq_disk != next->rq_disk)
731 return NULL;
732
733 if (req_op(req) == REQ_OP_WRITE_SAME &&
734 !blk_write_same_mergeable(req->bio, next->bio))
735 return NULL;
736
737 /*
738 * Don't allow merge of different write hints, or for a hint with
739 * non-hint IO.
740 */
741 if (req->write_hint != next->write_hint)
742 return NULL;
743
744 if (req->ioprio != next->ioprio)
745 return NULL;
746
747 /*
748 * If we are allowed to merge, then append bio list
749 * from next to rq and release next. merge_requests_fn
750 * will have updated segment counts, update sector
751 * counts here. Handle DISCARDs separately, as they
752 * have separate settings.
753 */
754
755 switch (blk_try_req_merge(req, next)) {
756 case ELEVATOR_DISCARD_MERGE:
757 if (!req_attempt_discard_merge(q, req, next))
758 return NULL;
759 break;
760 case ELEVATOR_BACK_MERGE:
761 if (!ll_merge_requests_fn(q, req, next))
762 return NULL;
763 break;
764 default:
765 return NULL;
766 }
767
768 /*
769 * If failfast settings disagree or any of the two is already
770 * a mixed merge, mark both as mixed before proceeding. This
771 * makes sure that all involved bios have mixable attributes
772 * set properly.
773 */
774 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
775 (req->cmd_flags & REQ_FAILFAST_MASK) !=
776 (next->cmd_flags & REQ_FAILFAST_MASK)) {
777 blk_rq_set_mixed_merge(req);
778 blk_rq_set_mixed_merge(next);
779 }
780
781 /*
782 * At this point we have either done a back merge or front merge. We
783 * need the smaller start_time_ns of the merged requests to be the
784 * current request for accounting purposes.
785 */
786 if (next->start_time_ns < req->start_time_ns)
787 req->start_time_ns = next->start_time_ns;
788
789 req->biotail->bi_next = next->bio;
790 req->biotail = next->biotail;
791
792 req->__data_len += blk_rq_bytes(next);
793
794 if (!blk_discard_mergable(req))
795 elv_merge_requests(q, req, next);
796
797 /*
798 * 'next' is going away, so update stats accordingly
799 */
800 blk_account_io_merge_request(next);
801
802 trace_block_rq_merge(q, next);
803
804 /*
805 * ownership of bio passed from next to req, return 'next' for
806 * the caller to free
807 */
808 next->bio = NULL;
809 return next;
810 }
811
attempt_back_merge(struct request_queue * q,struct request * rq)812 static struct request *attempt_back_merge(struct request_queue *q,
813 struct request *rq)
814 {
815 struct request *next = elv_latter_request(q, rq);
816
817 if (next)
818 return attempt_merge(q, rq, next);
819
820 return NULL;
821 }
822
attempt_front_merge(struct request_queue * q,struct request * rq)823 static struct request *attempt_front_merge(struct request_queue *q,
824 struct request *rq)
825 {
826 struct request *prev = elv_former_request(q, rq);
827
828 if (prev)
829 return attempt_merge(q, prev, rq);
830
831 return NULL;
832 }
833
834 /*
835 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
836 * otherwise. The caller is responsible for freeing 'next' if the merge
837 * happened.
838 */
blk_attempt_req_merge(struct request_queue * q,struct request * rq,struct request * next)839 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
840 struct request *next)
841 {
842 return attempt_merge(q, rq, next);
843 }
844
blk_rq_merge_ok(struct request * rq,struct bio * bio)845 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
846 {
847 if (!rq_mergeable(rq) || !bio_mergeable(bio))
848 return false;
849
850 if (req_op(rq) != bio_op(bio))
851 return false;
852
853 /* different data direction or already started, don't merge */
854 if (bio_data_dir(bio) != rq_data_dir(rq))
855 return false;
856
857 /* must be same device */
858 if (rq->rq_disk != bio->bi_disk)
859 return false;
860
861 /* only merge integrity protected bio into ditto rq */
862 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
863 return false;
864
865 /* Only merge if the crypt contexts are compatible */
866 if (!bio_crypt_rq_ctx_compatible(rq, bio))
867 return false;
868
869 /* must be using the same buffer */
870 if (req_op(rq) == REQ_OP_WRITE_SAME &&
871 !blk_write_same_mergeable(rq->bio, bio))
872 return false;
873
874 /*
875 * Don't allow merge of different write hints, or for a hint with
876 * non-hint IO.
877 */
878 if (rq->write_hint != bio->bi_write_hint)
879 return false;
880
881 if (rq->ioprio != bio_prio(bio))
882 return false;
883
884 return true;
885 }
886
blk_try_merge(struct request * rq,struct bio * bio)887 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
888 {
889 if (blk_discard_mergable(rq))
890 return ELEVATOR_DISCARD_MERGE;
891 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
892 return ELEVATOR_BACK_MERGE;
893 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
894 return ELEVATOR_FRONT_MERGE;
895 return ELEVATOR_NO_MERGE;
896 }
897
blk_account_io_merge_bio(struct request * req)898 static void blk_account_io_merge_bio(struct request *req)
899 {
900 if (!blk_do_io_stat(req))
901 return;
902
903 part_stat_lock();
904 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
905 part_stat_unlock();
906 }
907
908 enum bio_merge_status {
909 BIO_MERGE_OK,
910 BIO_MERGE_NONE,
911 BIO_MERGE_FAILED,
912 };
913
bio_attempt_back_merge(struct request * req,struct bio * bio,unsigned int nr_segs)914 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
915 struct bio *bio, unsigned int nr_segs)
916 {
917 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
918
919 if (!ll_back_merge_fn(req, bio, nr_segs))
920 return BIO_MERGE_FAILED;
921
922 trace_block_bio_backmerge(req->q, req, bio);
923 rq_qos_merge(req->q, req, bio);
924
925 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
926 blk_rq_set_mixed_merge(req);
927
928 req->biotail->bi_next = bio;
929 req->biotail = bio;
930 req->__data_len += bio->bi_iter.bi_size;
931
932 bio_crypt_free_ctx(bio);
933
934 blk_account_io_merge_bio(req);
935 return BIO_MERGE_OK;
936 }
937
bio_attempt_front_merge(struct request * req,struct bio * bio,unsigned int nr_segs)938 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
939 struct bio *bio, unsigned int nr_segs)
940 {
941 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
942
943 if (!ll_front_merge_fn(req, bio, nr_segs))
944 return BIO_MERGE_FAILED;
945
946 trace_block_bio_frontmerge(req->q, req, bio);
947 rq_qos_merge(req->q, req, bio);
948
949 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
950 blk_rq_set_mixed_merge(req);
951
952 bio->bi_next = req->bio;
953 req->bio = bio;
954
955 req->__sector = bio->bi_iter.bi_sector;
956 req->__data_len += bio->bi_iter.bi_size;
957
958 bio_crypt_do_front_merge(req, bio);
959
960 blk_account_io_merge_bio(req);
961 return BIO_MERGE_OK;
962 }
963
bio_attempt_discard_merge(struct request_queue * q,struct request * req,struct bio * bio)964 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
965 struct request *req, struct bio *bio)
966 {
967 unsigned short segments = blk_rq_nr_discard_segments(req);
968
969 if (segments >= queue_max_discard_segments(q))
970 goto no_merge;
971 if (blk_rq_sectors(req) + bio_sectors(bio) >
972 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
973 goto no_merge;
974
975 rq_qos_merge(q, req, bio);
976
977 req->biotail->bi_next = bio;
978 req->biotail = bio;
979 req->__data_len += bio->bi_iter.bi_size;
980 req->nr_phys_segments = segments + 1;
981
982 blk_account_io_merge_bio(req);
983 return BIO_MERGE_OK;
984 no_merge:
985 req_set_nomerge(q, req);
986 return BIO_MERGE_FAILED;
987 }
988
blk_attempt_bio_merge(struct request_queue * q,struct request * rq,struct bio * bio,unsigned int nr_segs,bool sched_allow_merge)989 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
990 struct request *rq,
991 struct bio *bio,
992 unsigned int nr_segs,
993 bool sched_allow_merge)
994 {
995 if (!blk_rq_merge_ok(rq, bio))
996 return BIO_MERGE_NONE;
997
998 switch (blk_try_merge(rq, bio)) {
999 case ELEVATOR_BACK_MERGE:
1000 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1001 return bio_attempt_back_merge(rq, bio, nr_segs);
1002 break;
1003 case ELEVATOR_FRONT_MERGE:
1004 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1005 return bio_attempt_front_merge(rq, bio, nr_segs);
1006 break;
1007 case ELEVATOR_DISCARD_MERGE:
1008 return bio_attempt_discard_merge(q, rq, bio);
1009 default:
1010 return BIO_MERGE_NONE;
1011 }
1012
1013 return BIO_MERGE_FAILED;
1014 }
1015
1016 /**
1017 * blk_attempt_plug_merge - try to merge with %current's plugged list
1018 * @q: request_queue new bio is being queued at
1019 * @bio: new bio being queued
1020 * @nr_segs: number of segments in @bio
1021 * @same_queue_rq: pointer to &struct request that gets filled in when
1022 * another request associated with @q is found on the plug list
1023 * (optional, may be %NULL)
1024 *
1025 * Determine whether @bio being queued on @q can be merged with a request
1026 * on %current's plugged list. Returns %true if merge was successful,
1027 * otherwise %false.
1028 *
1029 * Plugging coalesces IOs from the same issuer for the same purpose without
1030 * going through @q->queue_lock. As such it's more of an issuing mechanism
1031 * than scheduling, and the request, while may have elvpriv data, is not
1032 * added on the elevator at this point. In addition, we don't have
1033 * reliable access to the elevator outside queue lock. Only check basic
1034 * merging parameters without querying the elevator.
1035 *
1036 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1037 */
blk_attempt_plug_merge(struct request_queue * q,struct bio * bio,unsigned int nr_segs,struct request ** same_queue_rq)1038 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1039 unsigned int nr_segs, struct request **same_queue_rq)
1040 {
1041 struct blk_plug *plug;
1042 struct request *rq;
1043 struct list_head *plug_list;
1044
1045 plug = blk_mq_plug(q, bio);
1046 if (!plug)
1047 return false;
1048
1049 plug_list = &plug->mq_list;
1050
1051 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1052 if (rq->q == q && same_queue_rq) {
1053 /*
1054 * Only blk-mq multiple hardware queues case checks the
1055 * rq in the same queue, there should be only one such
1056 * rq in a queue
1057 **/
1058 *same_queue_rq = rq;
1059 }
1060
1061 if (rq->q != q)
1062 continue;
1063
1064 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1065 BIO_MERGE_OK)
1066 return true;
1067 }
1068
1069 return false;
1070 }
1071
1072 /*
1073 * Iterate list of requests and see if we can merge this bio with any
1074 * of them.
1075 */
blk_bio_list_merge(struct request_queue * q,struct list_head * list,struct bio * bio,unsigned int nr_segs)1076 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1077 struct bio *bio, unsigned int nr_segs)
1078 {
1079 struct request *rq;
1080 int checked = 8;
1081
1082 list_for_each_entry_reverse(rq, list, queuelist) {
1083 if (!checked--)
1084 break;
1085
1086 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1087 case BIO_MERGE_NONE:
1088 continue;
1089 case BIO_MERGE_OK:
1090 return true;
1091 case BIO_MERGE_FAILED:
1092 return false;
1093 }
1094
1095 }
1096
1097 return false;
1098 }
1099 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1100
blk_mq_sched_try_merge(struct request_queue * q,struct bio * bio,unsigned int nr_segs,struct request ** merged_request)1101 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1102 unsigned int nr_segs, struct request **merged_request)
1103 {
1104 struct request *rq;
1105
1106 switch (elv_merge(q, &rq, bio)) {
1107 case ELEVATOR_BACK_MERGE:
1108 if (!blk_mq_sched_allow_merge(q, rq, bio))
1109 return false;
1110 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1111 return false;
1112 *merged_request = attempt_back_merge(q, rq);
1113 if (!*merged_request)
1114 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1115 return true;
1116 case ELEVATOR_FRONT_MERGE:
1117 if (!blk_mq_sched_allow_merge(q, rq, bio))
1118 return false;
1119 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1120 return false;
1121 *merged_request = attempt_front_merge(q, rq);
1122 if (!*merged_request)
1123 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1124 return true;
1125 case ELEVATOR_DISCARD_MERGE:
1126 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1127 default:
1128 return false;
1129 }
1130 }
1131 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);
1132