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/blk-integrity.h>
10 #include <linux/scatterlist.h>
11 #include <linux/part_stat.h>
12 #include <linux/blk-cgroup.h>
13
14 #include <trace/events/block.h>
15
16 #include "blk.h"
17 #include "blk-mq-sched.h"
18 #include "blk-rq-qos.h"
19 #include "blk-throttle.h"
20
bio_get_first_bvec(struct bio * bio,struct bio_vec * bv)21 static inline void bio_get_first_bvec(struct bio *bio, struct bio_vec *bv)
22 {
23 *bv = mp_bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
24 }
25
bio_get_last_bvec(struct bio * bio,struct bio_vec * bv)26 static inline void bio_get_last_bvec(struct bio *bio, struct bio_vec *bv)
27 {
28 struct bvec_iter iter = bio->bi_iter;
29 int idx;
30
31 bio_get_first_bvec(bio, bv);
32 if (bv->bv_len == bio->bi_iter.bi_size)
33 return; /* this bio only has a single bvec */
34
35 bio_advance_iter(bio, &iter, iter.bi_size);
36
37 if (!iter.bi_bvec_done)
38 idx = iter.bi_idx - 1;
39 else /* in the middle of bvec */
40 idx = iter.bi_idx;
41
42 *bv = bio->bi_io_vec[idx];
43
44 /*
45 * iter.bi_bvec_done records actual length of the last bvec
46 * if this bio ends in the middle of one io vector
47 */
48 if (iter.bi_bvec_done)
49 bv->bv_len = iter.bi_bvec_done;
50 }
51
bio_will_gap(struct request_queue * q,struct request * prev_rq,struct bio * prev,struct bio * next)52 static inline bool bio_will_gap(struct request_queue *q,
53 struct request *prev_rq, struct bio *prev, struct bio *next)
54 {
55 struct bio_vec pb, nb;
56
57 if (!bio_has_data(prev) || !queue_virt_boundary(q))
58 return false;
59
60 /*
61 * Don't merge if the 1st bio starts with non-zero offset, otherwise it
62 * is quite difficult to respect the sg gap limit. We work hard to
63 * merge a huge number of small single bios in case of mkfs.
64 */
65 if (prev_rq)
66 bio_get_first_bvec(prev_rq->bio, &pb);
67 else
68 bio_get_first_bvec(prev, &pb);
69 if (pb.bv_offset & queue_virt_boundary(q))
70 return true;
71
72 /*
73 * We don't need to worry about the situation that the merged segment
74 * ends in unaligned virt boundary:
75 *
76 * - if 'pb' ends aligned, the merged segment ends aligned
77 * - if 'pb' ends unaligned, the next bio must include
78 * one single bvec of 'nb', otherwise the 'nb' can't
79 * merge with 'pb'
80 */
81 bio_get_last_bvec(prev, &pb);
82 bio_get_first_bvec(next, &nb);
83 if (biovec_phys_mergeable(q, &pb, &nb))
84 return false;
85 return __bvec_gap_to_prev(&q->limits, &pb, nb.bv_offset);
86 }
87
req_gap_back_merge(struct request * req,struct bio * bio)88 static inline bool req_gap_back_merge(struct request *req, struct bio *bio)
89 {
90 return bio_will_gap(req->q, req, req->biotail, bio);
91 }
92
req_gap_front_merge(struct request * req,struct bio * bio)93 static inline bool req_gap_front_merge(struct request *req, struct bio *bio)
94 {
95 return bio_will_gap(req->q, NULL, bio, req->bio);
96 }
97
98 /*
99 * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
100 * is defined as 'unsigned int', meantime it has to be aligned to with the
101 * logical block size, which is the minimum accepted unit by hardware.
102 */
bio_allowed_max_sectors(struct queue_limits * lim)103 static unsigned int bio_allowed_max_sectors(struct queue_limits *lim)
104 {
105 return round_down(UINT_MAX, lim->logical_block_size) >> SECTOR_SHIFT;
106 }
107
bio_split_discard(struct bio * bio,struct queue_limits * lim,unsigned * nsegs,struct bio_set * bs)108 static struct bio *bio_split_discard(struct bio *bio, struct queue_limits *lim,
109 unsigned *nsegs, struct bio_set *bs)
110 {
111 unsigned int max_discard_sectors, granularity;
112 sector_t tmp;
113 unsigned split_sectors;
114
115 *nsegs = 1;
116
117 /* Zero-sector (unknown) and one-sector granularities are the same. */
118 granularity = max(lim->discard_granularity >> 9, 1U);
119
120 max_discard_sectors =
121 min(lim->max_discard_sectors, bio_allowed_max_sectors(lim));
122 max_discard_sectors -= max_discard_sectors % granularity;
123
124 if (unlikely(!max_discard_sectors)) {
125 /* XXX: warn */
126 return NULL;
127 }
128
129 if (bio_sectors(bio) <= max_discard_sectors)
130 return NULL;
131
132 split_sectors = max_discard_sectors;
133
134 /*
135 * If the next starting sector would be misaligned, stop the discard at
136 * the previous aligned sector.
137 */
138 tmp = bio->bi_iter.bi_sector + split_sectors -
139 ((lim->discard_alignment >> 9) % granularity);
140 tmp = sector_div(tmp, granularity);
141
142 if (split_sectors > tmp)
143 split_sectors -= tmp;
144
145 return bio_split(bio, split_sectors, GFP_NOIO, bs);
146 }
147
bio_split_write_zeroes(struct bio * bio,struct queue_limits * lim,unsigned * nsegs,struct bio_set * bs)148 static struct bio *bio_split_write_zeroes(struct bio *bio,
149 struct queue_limits *lim, unsigned *nsegs, struct bio_set *bs)
150 {
151 *nsegs = 0;
152 if (!lim->max_write_zeroes_sectors)
153 return NULL;
154 if (bio_sectors(bio) <= lim->max_write_zeroes_sectors)
155 return NULL;
156 return bio_split(bio, lim->max_write_zeroes_sectors, GFP_NOIO, bs);
157 }
158
159 /*
160 * Return the maximum number of sectors from the start of a bio that may be
161 * submitted as a single request to a block device. If enough sectors remain,
162 * align the end to the physical block size. Otherwise align the end to the
163 * logical block size. This approach minimizes the number of non-aligned
164 * requests that are submitted to a block device if the start of a bio is not
165 * aligned to a physical block boundary.
166 */
get_max_io_size(struct bio * bio,struct queue_limits * lim)167 static inline unsigned get_max_io_size(struct bio *bio,
168 struct queue_limits *lim)
169 {
170 unsigned pbs = lim->physical_block_size >> SECTOR_SHIFT;
171 unsigned lbs = lim->logical_block_size >> SECTOR_SHIFT;
172 unsigned max_sectors = lim->max_sectors, start, end;
173
174 if (lim->chunk_sectors) {
175 max_sectors = min(max_sectors,
176 blk_chunk_sectors_left(bio->bi_iter.bi_sector,
177 lim->chunk_sectors));
178 }
179
180 start = bio->bi_iter.bi_sector & (pbs - 1);
181 end = (start + max_sectors) & ~(pbs - 1);
182 if (end > start)
183 return end - start;
184 return max_sectors & ~(lbs - 1);
185 }
186
get_max_segment_size(struct queue_limits * lim,struct page * start_page,unsigned long offset)187 static inline unsigned get_max_segment_size(struct queue_limits *lim,
188 struct page *start_page, unsigned long offset)
189 {
190 unsigned long mask = lim->seg_boundary_mask;
191
192 offset = mask & (page_to_phys(start_page) + offset);
193
194 /*
195 * overflow may be triggered in case of zero page physical address
196 * on 32bit arch, use queue's max segment size when that happens.
197 */
198 return min_not_zero(mask - offset + 1,
199 (unsigned long)lim->max_segment_size);
200 }
201
202 /**
203 * bvec_split_segs - verify whether or not a bvec should be split in the middle
204 * @lim: [in] queue limits to split based on
205 * @bv: [in] bvec to examine
206 * @nsegs: [in,out] Number of segments in the bio being built. Incremented
207 * by the number of segments from @bv that may be appended to that
208 * bio without exceeding @max_segs
209 * @bytes: [in,out] Number of bytes in the bio being built. Incremented
210 * by the number of bytes from @bv that may be appended to that
211 * bio without exceeding @max_bytes
212 * @max_segs: [in] upper bound for *@nsegs
213 * @max_bytes: [in] upper bound for *@bytes
214 *
215 * When splitting a bio, it can happen that a bvec is encountered that is too
216 * big to fit in a single segment and hence that it has to be split in the
217 * middle. This function verifies whether or not that should happen. The value
218 * %true is returned if and only if appending the entire @bv to a bio with
219 * *@nsegs segments and *@sectors sectors would make that bio unacceptable for
220 * the block driver.
221 */
bvec_split_segs(struct queue_limits * lim,const struct bio_vec * bv,unsigned * nsegs,unsigned * bytes,unsigned max_segs,unsigned max_bytes)222 static bool bvec_split_segs(struct queue_limits *lim, const struct bio_vec *bv,
223 unsigned *nsegs, unsigned *bytes, unsigned max_segs,
224 unsigned max_bytes)
225 {
226 unsigned max_len = min(max_bytes, UINT_MAX) - *bytes;
227 unsigned len = min(bv->bv_len, max_len);
228 unsigned total_len = 0;
229 unsigned seg_size = 0;
230
231 while (len && *nsegs < max_segs) {
232 seg_size = get_max_segment_size(lim, bv->bv_page,
233 bv->bv_offset + total_len);
234 seg_size = min(seg_size, len);
235
236 (*nsegs)++;
237 total_len += seg_size;
238 len -= seg_size;
239
240 if ((bv->bv_offset + total_len) & lim->virt_boundary_mask)
241 break;
242 }
243
244 *bytes += total_len;
245
246 /* tell the caller to split the bvec if it is too big to fit */
247 return len > 0 || bv->bv_len > max_len;
248 }
249
250 /**
251 * bio_split_rw - split a bio in two bios
252 * @bio: [in] bio to be split
253 * @lim: [in] queue limits to split based on
254 * @segs: [out] number of segments in the bio with the first half of the sectors
255 * @bs: [in] bio set to allocate the clone from
256 * @max_bytes: [in] maximum number of bytes per bio
257 *
258 * Clone @bio, update the bi_iter of the clone to represent the first sectors
259 * of @bio and update @bio->bi_iter to represent the remaining sectors. The
260 * following is guaranteed for the cloned bio:
261 * - That it has at most @max_bytes worth of data
262 * - That it has at most queue_max_segments(@q) segments.
263 *
264 * Except for discard requests the cloned bio will point at the bi_io_vec of
265 * the original bio. It is the responsibility of the caller to ensure that the
266 * original bio is not freed before the cloned bio. The caller is also
267 * responsible for ensuring that @bs is only destroyed after processing of the
268 * split bio has finished.
269 */
bio_split_rw(struct bio * bio,struct queue_limits * lim,unsigned * segs,struct bio_set * bs,unsigned max_bytes)270 static struct bio *bio_split_rw(struct bio *bio, struct queue_limits *lim,
271 unsigned *segs, struct bio_set *bs, unsigned max_bytes)
272 {
273 struct bio_vec bv, bvprv, *bvprvp = NULL;
274 struct bvec_iter iter;
275 unsigned nsegs = 0, bytes = 0;
276
277 bio_for_each_bvec(bv, bio, iter) {
278 /*
279 * If the queue doesn't support SG gaps and adding this
280 * offset would create a gap, disallow it.
281 */
282 if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv.bv_offset))
283 goto split;
284
285 if (nsegs < lim->max_segments &&
286 bytes + bv.bv_len <= max_bytes &&
287 bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
288 nsegs++;
289 bytes += bv.bv_len;
290 } else {
291 if (bvec_split_segs(lim, &bv, &nsegs, &bytes,
292 lim->max_segments, max_bytes))
293 goto split;
294 }
295
296 bvprv = bv;
297 bvprvp = &bvprv;
298 }
299
300 *segs = nsegs;
301 return NULL;
302 split:
303 /*
304 * We can't sanely support splitting for a REQ_NOWAIT bio. End it
305 * with EAGAIN if splitting is required and return an error pointer.
306 */
307 if (bio->bi_opf & REQ_NOWAIT) {
308 bio->bi_status = BLK_STS_AGAIN;
309 bio_endio(bio);
310 return ERR_PTR(-EAGAIN);
311 }
312
313 *segs = nsegs;
314
315 /*
316 * Individual bvecs might not be logical block aligned. Round down the
317 * split size so that each bio is properly block size aligned, even if
318 * we do not use the full hardware limits.
319 */
320 bytes = ALIGN_DOWN(bytes, lim->logical_block_size);
321
322 /*
323 * Bio splitting may cause subtle trouble such as hang when doing sync
324 * iopoll in direct IO routine. Given performance gain of iopoll for
325 * big IO can be trival, disable iopoll when split needed.
326 */
327 bio_clear_polled(bio);
328 return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs);
329 }
330
331 /**
332 * __bio_split_to_limits - split a bio to fit the queue limits
333 * @bio: bio to be split
334 * @lim: queue limits to split based on
335 * @nr_segs: returns the number of segments in the returned bio
336 *
337 * Check if @bio needs splitting based on the queue limits, and if so split off
338 * a bio fitting the limits from the beginning of @bio and return it. @bio is
339 * shortened to the remainder and re-submitted.
340 *
341 * The split bio is allocated from @q->bio_split, which is provided by the
342 * block layer.
343 */
__bio_split_to_limits(struct bio * bio,struct queue_limits * lim,unsigned int * nr_segs)344 struct bio *__bio_split_to_limits(struct bio *bio, struct queue_limits *lim,
345 unsigned int *nr_segs)
346 {
347 struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split;
348 struct bio *split;
349
350 switch (bio_op(bio)) {
351 case REQ_OP_DISCARD:
352 case REQ_OP_SECURE_ERASE:
353 split = bio_split_discard(bio, lim, nr_segs, bs);
354 break;
355 case REQ_OP_WRITE_ZEROES:
356 split = bio_split_write_zeroes(bio, lim, nr_segs, bs);
357 break;
358 default:
359 split = bio_split_rw(bio, lim, nr_segs, bs,
360 get_max_io_size(bio, lim) << SECTOR_SHIFT);
361 if (IS_ERR(split))
362 return NULL;
363 break;
364 }
365
366 if (split) {
367 /* there isn't chance to merge the split bio */
368 split->bi_opf |= REQ_NOMERGE;
369
370 blkcg_bio_issue_init(split);
371 bio_chain(split, bio);
372 trace_block_split(split, bio->bi_iter.bi_sector);
373 submit_bio_noacct(bio);
374 return split;
375 }
376 return bio;
377 }
378
379 /**
380 * bio_split_to_limits - split a bio to fit the queue limits
381 * @bio: bio to be split
382 *
383 * Check if @bio needs splitting based on the queue limits of @bio->bi_bdev, and
384 * if so split off a bio fitting the limits from the beginning of @bio and
385 * return it. @bio is shortened to the remainder and re-submitted.
386 *
387 * The split bio is allocated from @q->bio_split, which is provided by the
388 * block layer.
389 */
bio_split_to_limits(struct bio * bio)390 struct bio *bio_split_to_limits(struct bio *bio)
391 {
392 struct queue_limits *lim = &bdev_get_queue(bio->bi_bdev)->limits;
393 unsigned int nr_segs;
394
395 if (bio_may_exceed_limits(bio, lim))
396 return __bio_split_to_limits(bio, lim, &nr_segs);
397 return bio;
398 }
399 EXPORT_SYMBOL(bio_split_to_limits);
400
blk_recalc_rq_segments(struct request * rq)401 unsigned int blk_recalc_rq_segments(struct request *rq)
402 {
403 unsigned int nr_phys_segs = 0;
404 unsigned int bytes = 0;
405 struct req_iterator iter;
406 struct bio_vec bv;
407
408 if (!rq->bio)
409 return 0;
410
411 switch (bio_op(rq->bio)) {
412 case REQ_OP_DISCARD:
413 case REQ_OP_SECURE_ERASE:
414 if (queue_max_discard_segments(rq->q) > 1) {
415 struct bio *bio = rq->bio;
416
417 for_each_bio(bio)
418 nr_phys_segs++;
419 return nr_phys_segs;
420 }
421 return 1;
422 case REQ_OP_WRITE_ZEROES:
423 return 0;
424 default:
425 break;
426 }
427
428 rq_for_each_bvec(bv, rq, iter)
429 bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes,
430 UINT_MAX, UINT_MAX);
431 return nr_phys_segs;
432 }
433
blk_next_sg(struct scatterlist ** sg,struct scatterlist * sglist)434 static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
435 struct scatterlist *sglist)
436 {
437 if (!*sg)
438 return sglist;
439
440 /*
441 * If the driver previously mapped a shorter list, we could see a
442 * termination bit prematurely unless it fully inits the sg table
443 * on each mapping. We KNOW that there must be more entries here
444 * or the driver would be buggy, so force clear the termination bit
445 * to avoid doing a full sg_init_table() in drivers for each command.
446 */
447 sg_unmark_end(*sg);
448 return sg_next(*sg);
449 }
450
blk_bvec_map_sg(struct request_queue * q,struct bio_vec * bvec,struct scatterlist * sglist,struct scatterlist ** sg)451 static unsigned blk_bvec_map_sg(struct request_queue *q,
452 struct bio_vec *bvec, struct scatterlist *sglist,
453 struct scatterlist **sg)
454 {
455 unsigned nbytes = bvec->bv_len;
456 unsigned nsegs = 0, total = 0;
457
458 while (nbytes > 0) {
459 unsigned offset = bvec->bv_offset + total;
460 unsigned len = min(get_max_segment_size(&q->limits,
461 bvec->bv_page, offset), nbytes);
462 struct page *page = bvec->bv_page;
463
464 /*
465 * Unfortunately a fair number of drivers barf on scatterlists
466 * that have an offset larger than PAGE_SIZE, despite other
467 * subsystems dealing with that invariant just fine. For now
468 * stick to the legacy format where we never present those from
469 * the block layer, but the code below should be removed once
470 * these offenders (mostly MMC/SD drivers) are fixed.
471 */
472 page += (offset >> PAGE_SHIFT);
473 offset &= ~PAGE_MASK;
474
475 *sg = blk_next_sg(sg, sglist);
476 sg_set_page(*sg, page, len, offset);
477
478 total += len;
479 nbytes -= len;
480 nsegs++;
481 }
482
483 return nsegs;
484 }
485
__blk_bvec_map_sg(struct bio_vec bv,struct scatterlist * sglist,struct scatterlist ** sg)486 static inline int __blk_bvec_map_sg(struct bio_vec bv,
487 struct scatterlist *sglist, struct scatterlist **sg)
488 {
489 *sg = blk_next_sg(sg, sglist);
490 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
491 return 1;
492 }
493
494 /* only try to merge bvecs into one sg if they are from two bios */
495 static inline bool
__blk_segment_map_sg_merge(struct request_queue * q,struct bio_vec * bvec,struct bio_vec * bvprv,struct scatterlist ** sg)496 __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
497 struct bio_vec *bvprv, struct scatterlist **sg)
498 {
499
500 int nbytes = bvec->bv_len;
501
502 if (!*sg)
503 return false;
504
505 if ((*sg)->length + nbytes > queue_max_segment_size(q))
506 return false;
507
508 if (!biovec_phys_mergeable(q, bvprv, bvec))
509 return false;
510
511 (*sg)->length += nbytes;
512
513 return true;
514 }
515
__blk_bios_map_sg(struct request_queue * q,struct bio * bio,struct scatterlist * sglist,struct scatterlist ** sg)516 static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
517 struct scatterlist *sglist,
518 struct scatterlist **sg)
519 {
520 struct bio_vec bvec, bvprv = { NULL };
521 struct bvec_iter iter;
522 int nsegs = 0;
523 bool new_bio = false;
524
525 for_each_bio(bio) {
526 bio_for_each_bvec(bvec, bio, iter) {
527 /*
528 * Only try to merge bvecs from two bios given we
529 * have done bio internal merge when adding pages
530 * to bio
531 */
532 if (new_bio &&
533 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
534 goto next_bvec;
535
536 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
537 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
538 else
539 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
540 next_bvec:
541 new_bio = false;
542 }
543 if (likely(bio->bi_iter.bi_size)) {
544 bvprv = bvec;
545 new_bio = true;
546 }
547 }
548
549 return nsegs;
550 }
551
552 /*
553 * map a request to scatterlist, return number of sg entries setup. Caller
554 * must make sure sg can hold rq->nr_phys_segments entries
555 */
__blk_rq_map_sg(struct request_queue * q,struct request * rq,struct scatterlist * sglist,struct scatterlist ** last_sg)556 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
557 struct scatterlist *sglist, struct scatterlist **last_sg)
558 {
559 int nsegs = 0;
560
561 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
562 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg);
563 else if (rq->bio)
564 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg);
565
566 if (*last_sg)
567 sg_mark_end(*last_sg);
568
569 /*
570 * Something must have been wrong if the figured number of
571 * segment is bigger than number of req's physical segments
572 */
573 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
574
575 return nsegs;
576 }
577 EXPORT_SYMBOL(__blk_rq_map_sg);
578
blk_rq_get_max_segments(struct request * rq)579 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
580 {
581 if (req_op(rq) == REQ_OP_DISCARD)
582 return queue_max_discard_segments(rq->q);
583 return queue_max_segments(rq->q);
584 }
585
blk_rq_get_max_sectors(struct request * rq,sector_t offset)586 static inline unsigned int blk_rq_get_max_sectors(struct request *rq,
587 sector_t offset)
588 {
589 struct request_queue *q = rq->q;
590 unsigned int max_sectors;
591
592 if (blk_rq_is_passthrough(rq))
593 return q->limits.max_hw_sectors;
594
595 max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
596 if (!q->limits.chunk_sectors ||
597 req_op(rq) == REQ_OP_DISCARD ||
598 req_op(rq) == REQ_OP_SECURE_ERASE)
599 return max_sectors;
600 return min(max_sectors,
601 blk_chunk_sectors_left(offset, q->limits.chunk_sectors));
602 }
603
ll_new_hw_segment(struct request * req,struct bio * bio,unsigned int nr_phys_segs)604 static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
605 unsigned int nr_phys_segs)
606 {
607 if (!blk_cgroup_mergeable(req, bio))
608 goto no_merge;
609
610 if (blk_integrity_merge_bio(req->q, req, bio) == false)
611 goto no_merge;
612
613 /* discard request merge won't add new segment */
614 if (req_op(req) == REQ_OP_DISCARD)
615 return 1;
616
617 if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req))
618 goto no_merge;
619
620 /*
621 * This will form the start of a new hw segment. Bump both
622 * counters.
623 */
624 req->nr_phys_segments += nr_phys_segs;
625 return 1;
626
627 no_merge:
628 req_set_nomerge(req->q, req);
629 return 0;
630 }
631
ll_back_merge_fn(struct request * req,struct bio * bio,unsigned int nr_segs)632 int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
633 {
634 if (req_gap_back_merge(req, bio))
635 return 0;
636 if (blk_integrity_rq(req) &&
637 integrity_req_gap_back_merge(req, bio))
638 return 0;
639 if (!bio_crypt_ctx_back_mergeable(req, bio))
640 return 0;
641 if (blk_rq_sectors(req) + bio_sectors(bio) >
642 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
643 req_set_nomerge(req->q, req);
644 return 0;
645 }
646
647 return ll_new_hw_segment(req, bio, nr_segs);
648 }
649
ll_front_merge_fn(struct request * req,struct bio * bio,unsigned int nr_segs)650 static int ll_front_merge_fn(struct request *req, struct bio *bio,
651 unsigned int nr_segs)
652 {
653 if (req_gap_front_merge(req, bio))
654 return 0;
655 if (blk_integrity_rq(req) &&
656 integrity_req_gap_front_merge(req, bio))
657 return 0;
658 if (!bio_crypt_ctx_front_mergeable(req, bio))
659 return 0;
660 if (blk_rq_sectors(req) + bio_sectors(bio) >
661 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
662 req_set_nomerge(req->q, req);
663 return 0;
664 }
665
666 return ll_new_hw_segment(req, bio, nr_segs);
667 }
668
req_attempt_discard_merge(struct request_queue * q,struct request * req,struct request * next)669 static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
670 struct request *next)
671 {
672 unsigned short segments = blk_rq_nr_discard_segments(req);
673
674 if (segments >= queue_max_discard_segments(q))
675 goto no_merge;
676 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
677 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
678 goto no_merge;
679
680 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
681 return true;
682 no_merge:
683 req_set_nomerge(q, req);
684 return false;
685 }
686
ll_merge_requests_fn(struct request_queue * q,struct request * req,struct request * next)687 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
688 struct request *next)
689 {
690 int total_phys_segments;
691
692 if (req_gap_back_merge(req, next->bio))
693 return 0;
694
695 /*
696 * Will it become too large?
697 */
698 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
699 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
700 return 0;
701
702 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
703 if (total_phys_segments > blk_rq_get_max_segments(req))
704 return 0;
705
706 if (!blk_cgroup_mergeable(req, next->bio))
707 return 0;
708
709 if (blk_integrity_merge_rq(q, req, next) == false)
710 return 0;
711
712 if (!bio_crypt_ctx_merge_rq(req, next))
713 return 0;
714
715 /* Merge is OK... */
716 req->nr_phys_segments = total_phys_segments;
717 return 1;
718 }
719
720 /**
721 * blk_rq_set_mixed_merge - mark a request as mixed merge
722 * @rq: request to mark as mixed merge
723 *
724 * Description:
725 * @rq is about to be mixed merged. Make sure the attributes
726 * which can be mixed are set in each bio and mark @rq as mixed
727 * merged.
728 */
blk_rq_set_mixed_merge(struct request * rq)729 void blk_rq_set_mixed_merge(struct request *rq)
730 {
731 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
732 struct bio *bio;
733
734 if (rq->rq_flags & RQF_MIXED_MERGE)
735 return;
736
737 /*
738 * @rq will no longer represent mixable attributes for all the
739 * contained bios. It will just track those of the first one.
740 * Distributes the attributs to each bio.
741 */
742 for (bio = rq->bio; bio; bio = bio->bi_next) {
743 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
744 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
745 bio->bi_opf |= ff;
746 }
747 rq->rq_flags |= RQF_MIXED_MERGE;
748 }
749
blk_account_io_merge_request(struct request * req)750 static void blk_account_io_merge_request(struct request *req)
751 {
752 if (blk_do_io_stat(req)) {
753 part_stat_lock();
754 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
755 part_stat_unlock();
756 }
757 }
758
blk_try_req_merge(struct request * req,struct request * next)759 static enum elv_merge blk_try_req_merge(struct request *req,
760 struct request *next)
761 {
762 if (blk_discard_mergable(req))
763 return ELEVATOR_DISCARD_MERGE;
764 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
765 return ELEVATOR_BACK_MERGE;
766
767 return ELEVATOR_NO_MERGE;
768 }
769
770 /*
771 * For non-mq, this has to be called with the request spinlock acquired.
772 * For mq with scheduling, the appropriate queue wide lock should be held.
773 */
attempt_merge(struct request_queue * q,struct request * req,struct request * next)774 static struct request *attempt_merge(struct request_queue *q,
775 struct request *req, struct request *next)
776 {
777 if (!rq_mergeable(req) || !rq_mergeable(next))
778 return NULL;
779
780 if (req_op(req) != req_op(next))
781 return NULL;
782
783 if (rq_data_dir(req) != rq_data_dir(next))
784 return NULL;
785
786 if (req->ioprio != next->ioprio)
787 return NULL;
788
789 /*
790 * If we are allowed to merge, then append bio list
791 * from next to rq and release next. merge_requests_fn
792 * will have updated segment counts, update sector
793 * counts here. Handle DISCARDs separately, as they
794 * have separate settings.
795 */
796
797 switch (blk_try_req_merge(req, next)) {
798 case ELEVATOR_DISCARD_MERGE:
799 if (!req_attempt_discard_merge(q, req, next))
800 return NULL;
801 break;
802 case ELEVATOR_BACK_MERGE:
803 if (!ll_merge_requests_fn(q, req, next))
804 return NULL;
805 break;
806 default:
807 return NULL;
808 }
809
810 /*
811 * If failfast settings disagree or any of the two is already
812 * a mixed merge, mark both as mixed before proceeding. This
813 * makes sure that all involved bios have mixable attributes
814 * set properly.
815 */
816 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
817 (req->cmd_flags & REQ_FAILFAST_MASK) !=
818 (next->cmd_flags & REQ_FAILFAST_MASK)) {
819 blk_rq_set_mixed_merge(req);
820 blk_rq_set_mixed_merge(next);
821 }
822
823 /*
824 * At this point we have either done a back merge or front merge. We
825 * need the smaller start_time_ns of the merged requests to be the
826 * current request for accounting purposes.
827 */
828 if (next->start_time_ns < req->start_time_ns)
829 req->start_time_ns = next->start_time_ns;
830
831 req->biotail->bi_next = next->bio;
832 req->biotail = next->biotail;
833
834 req->__data_len += blk_rq_bytes(next);
835
836 if (!blk_discard_mergable(req))
837 elv_merge_requests(q, req, next);
838
839 /*
840 * 'next' is going away, so update stats accordingly
841 */
842 blk_account_io_merge_request(next);
843
844 trace_block_rq_merge(next);
845
846 /*
847 * ownership of bio passed from next to req, return 'next' for
848 * the caller to free
849 */
850 next->bio = NULL;
851 return next;
852 }
853
attempt_back_merge(struct request_queue * q,struct request * rq)854 static struct request *attempt_back_merge(struct request_queue *q,
855 struct request *rq)
856 {
857 struct request *next = elv_latter_request(q, rq);
858
859 if (next)
860 return attempt_merge(q, rq, next);
861
862 return NULL;
863 }
864
attempt_front_merge(struct request_queue * q,struct request * rq)865 static struct request *attempt_front_merge(struct request_queue *q,
866 struct request *rq)
867 {
868 struct request *prev = elv_former_request(q, rq);
869
870 if (prev)
871 return attempt_merge(q, prev, rq);
872
873 return NULL;
874 }
875
876 /*
877 * Try to merge 'next' into 'rq'. Return true if the merge happened, false
878 * otherwise. The caller is responsible for freeing 'next' if the merge
879 * happened.
880 */
blk_attempt_req_merge(struct request_queue * q,struct request * rq,struct request * next)881 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
882 struct request *next)
883 {
884 return attempt_merge(q, rq, next);
885 }
886
blk_rq_merge_ok(struct request * rq,struct bio * bio)887 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
888 {
889 if (!rq_mergeable(rq) || !bio_mergeable(bio))
890 return false;
891
892 if (req_op(rq) != bio_op(bio))
893 return false;
894
895 /* different data direction or already started, don't merge */
896 if (bio_data_dir(bio) != rq_data_dir(rq))
897 return false;
898
899 /* don't merge across cgroup boundaries */
900 if (!blk_cgroup_mergeable(rq, bio))
901 return false;
902
903 /* only merge integrity protected bio into ditto rq */
904 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
905 return false;
906
907 /* Only merge if the crypt contexts are compatible */
908 if (!bio_crypt_rq_ctx_compatible(rq, bio))
909 return false;
910
911 if (rq->ioprio != bio_prio(bio))
912 return false;
913
914 return true;
915 }
916
blk_try_merge(struct request * rq,struct bio * bio)917 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
918 {
919 if (blk_discard_mergable(rq))
920 return ELEVATOR_DISCARD_MERGE;
921 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
922 return ELEVATOR_BACK_MERGE;
923 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
924 return ELEVATOR_FRONT_MERGE;
925 return ELEVATOR_NO_MERGE;
926 }
927
blk_account_io_merge_bio(struct request * req)928 static void blk_account_io_merge_bio(struct request *req)
929 {
930 if (!blk_do_io_stat(req))
931 return;
932
933 part_stat_lock();
934 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
935 part_stat_unlock();
936 }
937
938 enum bio_merge_status {
939 BIO_MERGE_OK,
940 BIO_MERGE_NONE,
941 BIO_MERGE_FAILED,
942 };
943
bio_attempt_back_merge(struct request * req,struct bio * bio,unsigned int nr_segs)944 static enum bio_merge_status bio_attempt_back_merge(struct request *req,
945 struct bio *bio, unsigned int nr_segs)
946 {
947 const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK;
948
949 if (!ll_back_merge_fn(req, bio, nr_segs))
950 return BIO_MERGE_FAILED;
951
952 trace_block_bio_backmerge(bio);
953 rq_qos_merge(req->q, req, bio);
954
955 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
956 blk_rq_set_mixed_merge(req);
957
958 req->biotail->bi_next = bio;
959 req->biotail = bio;
960 req->__data_len += bio->bi_iter.bi_size;
961
962 bio_crypt_free_ctx(bio);
963
964 blk_account_io_merge_bio(req);
965 return BIO_MERGE_OK;
966 }
967
bio_attempt_front_merge(struct request * req,struct bio * bio,unsigned int nr_segs)968 static enum bio_merge_status bio_attempt_front_merge(struct request *req,
969 struct bio *bio, unsigned int nr_segs)
970 {
971 const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK;
972
973 if (!ll_front_merge_fn(req, bio, nr_segs))
974 return BIO_MERGE_FAILED;
975
976 trace_block_bio_frontmerge(bio);
977 rq_qos_merge(req->q, req, bio);
978
979 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
980 blk_rq_set_mixed_merge(req);
981
982 bio->bi_next = req->bio;
983 req->bio = bio;
984
985 req->__sector = bio->bi_iter.bi_sector;
986 req->__data_len += bio->bi_iter.bi_size;
987
988 bio_crypt_do_front_merge(req, bio);
989
990 blk_account_io_merge_bio(req);
991 return BIO_MERGE_OK;
992 }
993
bio_attempt_discard_merge(struct request_queue * q,struct request * req,struct bio * bio)994 static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q,
995 struct request *req, struct bio *bio)
996 {
997 unsigned short segments = blk_rq_nr_discard_segments(req);
998
999 if (segments >= queue_max_discard_segments(q))
1000 goto no_merge;
1001 if (blk_rq_sectors(req) + bio_sectors(bio) >
1002 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1003 goto no_merge;
1004
1005 rq_qos_merge(q, req, bio);
1006
1007 req->biotail->bi_next = bio;
1008 req->biotail = bio;
1009 req->__data_len += bio->bi_iter.bi_size;
1010 req->nr_phys_segments = segments + 1;
1011
1012 blk_account_io_merge_bio(req);
1013 return BIO_MERGE_OK;
1014 no_merge:
1015 req_set_nomerge(q, req);
1016 return BIO_MERGE_FAILED;
1017 }
1018
blk_attempt_bio_merge(struct request_queue * q,struct request * rq,struct bio * bio,unsigned int nr_segs,bool sched_allow_merge)1019 static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q,
1020 struct request *rq,
1021 struct bio *bio,
1022 unsigned int nr_segs,
1023 bool sched_allow_merge)
1024 {
1025 if (!blk_rq_merge_ok(rq, bio))
1026 return BIO_MERGE_NONE;
1027
1028 switch (blk_try_merge(rq, bio)) {
1029 case ELEVATOR_BACK_MERGE:
1030 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1031 return bio_attempt_back_merge(rq, bio, nr_segs);
1032 break;
1033 case ELEVATOR_FRONT_MERGE:
1034 if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio))
1035 return bio_attempt_front_merge(rq, bio, nr_segs);
1036 break;
1037 case ELEVATOR_DISCARD_MERGE:
1038 return bio_attempt_discard_merge(q, rq, bio);
1039 default:
1040 return BIO_MERGE_NONE;
1041 }
1042
1043 return BIO_MERGE_FAILED;
1044 }
1045
1046 /**
1047 * blk_attempt_plug_merge - try to merge with %current's plugged list
1048 * @q: request_queue new bio is being queued at
1049 * @bio: new bio being queued
1050 * @nr_segs: number of segments in @bio
1051 * from the passed in @q already in the plug list
1052 *
1053 * Determine whether @bio being queued on @q can be merged with the previous
1054 * request on %current's plugged list. Returns %true if merge was successful,
1055 * otherwise %false.
1056 *
1057 * Plugging coalesces IOs from the same issuer for the same purpose without
1058 * going through @q->queue_lock. As such it's more of an issuing mechanism
1059 * than scheduling, and the request, while may have elvpriv data, is not
1060 * added on the elevator at this point. In addition, we don't have
1061 * reliable access to the elevator outside queue lock. Only check basic
1062 * merging parameters without querying the elevator.
1063 *
1064 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1065 */
blk_attempt_plug_merge(struct request_queue * q,struct bio * bio,unsigned int nr_segs)1066 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1067 unsigned int nr_segs)
1068 {
1069 struct blk_plug *plug;
1070 struct request *rq;
1071
1072 plug = blk_mq_plug(bio);
1073 if (!plug || rq_list_empty(plug->mq_list))
1074 return false;
1075
1076 rq_list_for_each(&plug->mq_list, rq) {
1077 if (rq->q == q) {
1078 if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
1079 BIO_MERGE_OK)
1080 return true;
1081 break;
1082 }
1083
1084 /*
1085 * Only keep iterating plug list for merges if we have multiple
1086 * queues
1087 */
1088 if (!plug->multiple_queues)
1089 break;
1090 }
1091 return false;
1092 }
1093
1094 /*
1095 * Iterate list of requests and see if we can merge this bio with any
1096 * of them.
1097 */
blk_bio_list_merge(struct request_queue * q,struct list_head * list,struct bio * bio,unsigned int nr_segs)1098 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
1099 struct bio *bio, unsigned int nr_segs)
1100 {
1101 struct request *rq;
1102 int checked = 8;
1103
1104 list_for_each_entry_reverse(rq, list, queuelist) {
1105 if (!checked--)
1106 break;
1107
1108 switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) {
1109 case BIO_MERGE_NONE:
1110 continue;
1111 case BIO_MERGE_OK:
1112 return true;
1113 case BIO_MERGE_FAILED:
1114 return false;
1115 }
1116
1117 }
1118
1119 return false;
1120 }
1121 EXPORT_SYMBOL_GPL(blk_bio_list_merge);
1122
blk_mq_sched_try_merge(struct request_queue * q,struct bio * bio,unsigned int nr_segs,struct request ** merged_request)1123 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
1124 unsigned int nr_segs, struct request **merged_request)
1125 {
1126 struct request *rq;
1127
1128 switch (elv_merge(q, &rq, bio)) {
1129 case ELEVATOR_BACK_MERGE:
1130 if (!blk_mq_sched_allow_merge(q, rq, bio))
1131 return false;
1132 if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1133 return false;
1134 *merged_request = attempt_back_merge(q, rq);
1135 if (!*merged_request)
1136 elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
1137 return true;
1138 case ELEVATOR_FRONT_MERGE:
1139 if (!blk_mq_sched_allow_merge(q, rq, bio))
1140 return false;
1141 if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
1142 return false;
1143 *merged_request = attempt_front_merge(q, rq);
1144 if (!*merged_request)
1145 elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
1146 return true;
1147 case ELEVATOR_DISCARD_MERGE:
1148 return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
1149 default:
1150 return false;
1151 }
1152 }
1153 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);
1154